The Recycling Industry in the United States is a for-profit industry. They profit from taking recyclable material, refining it, and reselling it to companies at a cheaper price than producing the material from scratch.

Supply and Demand in Recycling

If you look at the demand side of the recycling industry, an array of multi-billion dollar companies like Coca-Cola and PepsiCo are incentivized to buy recycled goods and reduce their materials costs.

On the supply side, ~300 million tons of trash are generated annually in the United States. Estimates suggest that up to 75% of that is recyclable.

On paper, it seems clear that maximizing the amount of trash the US recycles is in everyone’s interest. One issue though, less than a third of the trash ends up recycled.

Areeb, co-founder of Glacier, breaks down the multi-layered reasoning behind why the Recycling industry cannot handle this volume of trash, and what Glacier is doing to address this.

Areeb Malik

Areeb Malik is the Co-Founder of Glacier, and he is on a personal mission to fight climate change and extract value from the $123B worth of recyclables that fill the landfills and oceans. Before founding Glacier, Areeb was a Software Engineer at Facebook, where he used Machine Learning and Computer Vision to build out new product features.

Links

• Glacier
• Download mp3
• Subscribe to Robohub using iTunes, RSS, or Spotify
• Support us on Patreon

Prof Brendan Englot, from Stevens Institute of Technology, discusses the challenges in perception and decision-making for underwater robots – especially in the field. He discusses ongoing research using the BlueROV platform and autonomous driving simulators.

Brendan Englot

Brendan Englot received his S.B., S.M., and Ph.D. degrees in mechanical engineering from the Massachusetts Institute of Technology in 2007, 2009, and 2012, respectively. He is currently an Associate Professor with the Department of Mechanical Engineering at Stevens Institute of Technology in Hoboken, New Jersey. At Stevens, he also serves as interim director of the Stevens Institute for Artificial Intelligence. He is interested in perception, planning, optimization, and control that enable mobile robots to achieve robust autonomy in complex physical environments, and his recent work has considered sensing tasks motivated by underwater surveillance and inspection applications, and path planning with multiple objectives, unreliable sensors, and imprecise maps.

Links

• Download mp3 (48.0 MB)
• Subscribe to Robohub using iTunes, RSS, or Spotify
• Support us on Patreon

Dr. Sihao Sun discusses his award-winning research in the area of controlling the flight of a drone when faced with a sudden rotor failure.

Typical research in this area addressed the case where one of the four rotors in a quadrotor suddenly, spontaneously stops working. This previous research does not take into full account real-life scenarios where rotor failure is common. This includes collisions with other drones, walls, birds, and operating in degraded GPS environments.

Dr. Sihao Sun

Dr. Sihao Sun is a postdoctoral research assistant at the Robotics and Perception Group (RPG) in University of Zurich directed by Prof. Davide Scaramuzza. Currently, he is working on control and perception for aerial robots (drones).

In December 2020, he received his PhD degree in Aerospace Engineering from the Control and Simulation Group of Delft University of Technology. His works on quadrotor fault-tolerant flight control have been featured by reputable media, such as IEEE Spectrum.

Links

• Download mp3 (16.9 MB)
• Subscribe to Robohub using iTunes, RSS, or Spotify
• Support us on Patreon

At ICRA 2022, Competitions are a core part of the conference. We shine a spotlight on influential competitions in Robotics. In this episode, Dr Liam Paull talks about the Duckietown Competition, where robots drive around Rubber Ducky passengers in an autonomous driving track.

Dr. Liam Paull

Liam Paull is an assistant professor at l’Université de Montréal and the head of the Montreal Robotics and Embodied AI Lab (REAL). His lab focuses on robotics problems including building representations of the world (such as for simultaneous localization and mapping), modeling of uncertainty, and building better workflows to teach robotic agents new tasks (such as through simulation or demonstration). Previous to this, Liam was a research scientist at CSAIL MIT where he led the TRI funded autonomous car project. He was also a postdoc in the marine robotics lab at MIT where he worked on SLAM for underwater robots. He obtained his PhD from the University of New Brunswick in 2013 where he worked on robust and adaptive planning for underwater vehicles. He is a co-founder and director of the Duckietown Foundation, which is dedicated to making engaging robotics learning experiences accessible to everyone. The Duckietown class was originally taught at MIT but now the platform is used at numerous institutions worldwide.

Links

• Download mp3 (20.1 MB)
• Subscribe to Robohub using iTunes, RSS, or Spotify
• Support us on Patreon

In this episode, Audrow Nash speaks to Tim Chung, Program Manager in the Tactical Technology Office at the Defense Advanced Research Projects Agency (DARPA), on the DARPA Subterranean (SubT) Challenge. The SubT Challenge is a robotics challenge that aims to develop innovative technologies that would augment operations underground. In this conversation, they talk about the motivation of the SubT Challenge, its systems (hardware) and virtual (simulation) challenges, how the resulting technology has been shared with the world, on benchmarking the robotics behaviors in simulation, on the challenge of scoping the SubT Challenge, and on building the final environment for the systems challenge.

Episode Links

• Download the episode
• The SubT Challenge’s Website
• Tim’s LinkedIn
• DARPAtv’s YouTube Channel
• SubT’s Repository on OSRF

Podcast info

• Podcast website
• Apple Podcasts
• Spotify
• Podcast RSS feed
• Full episodes
• Clips

Amit Goel, Director of Product Management for Autonomous Machines at NVIDIA, discusses the new collaboration between Open Robotics and NVIDIA. The collaboration will improve the way ROS and NVIDIA’s line of products such as Isaac SIM and the Jetson line of embedded boards operate together.

NVIDIA’s Isaac SIM lets developers build robust and scalable simulations. Dramatically reducing the costs of capturing real-world data and speeding up development time.

Their Jetson line of embedded boards is core to many robotics architectures, leveraging hardware-optimized chips for machine learning, computer vision, video processing, and more.

The improvements to ROS will allow robotics companies to better utilize the available computational power, while still developing on the robotics-centric platform familiar to many.

For the developers out there, check out the ISAAC ROS Github repository to implement this yourself.

Amit Goel

Amit Goel is Director of Product Management for Autonomous Machines at NVIDIA, where he leads the product development of NVIDIA Jetson, the most advanced platform for AI computing at the edge.

Amit has more than 15 years of experience in the technology industry working in both software and hardware design roles. Prior to joining NVIDIA in 2011, he worked as a senior software engineer at Synopsys, where he developed algorithms for statistical performance modeling of digital designs.

Amit holds a Bachelor of Engineering in electronics and communication from Delhi College of Engineering, a Master of Science in electrical engineering from Arizona State University, and an MBA from the University of California at Berkeley.

Links

• Download mp3 (29.8 MB)
• Subscribe to Robohub using iTunes, RSS, or Spotify
• Support us on Patreon

RE2 Sapien 6M Robot Arm

In this episode, Shihan Lu interviews Travis Schneider, Business Development Manager at RE2 Robotics, focusing on their work on mobile outdoor manipulation. Travis introduces several robotic products and services of RE2, including RE2 Sapien robotic arms and human-centered system integration, and how they can be used to help rather than replace workers. He also shares his thoughts about challenges in human-robot collaboration.

Travis Schneider
Travis Schneider, Business Development Manager at RE2 Robotics (RE2), works to foster commercial B2B partnerships with groups interested in leveraging RE2’s robotic technology for applications in a wide variety of markets, including aviation, construction, energy, and medical. With a background in Mechanical Engineering, Travis has an intimate knowledge of many of the fundamentals associated with robotic systems including electromechanical design, motion control, and associated software.

Links

• RE2 Robotics
• Download mp3 (12.7 MB)
• Subscribe to Robohub using iTunes, RSS, or Spotify
• Support us on Patreon

In this episode, Shihan Lu interviews Jivko Sinapov, Assistant Professor in the Computer Science Department at Tufts University, about his work on behavior-grounded multisensory perception and exploration in robotics. Dr. Sinapov discusses several perspectives on multisensory perception in robotics, including data collection, data fusion, and robot control and planning. He also shares his experience about using robotics for K-12 education.

Jivko Sinapov

Jivko Sinapov received his Ph.D. in Computer Science and Human-Computer Interaction from Iowa State University (ISU). While working toward his Ph.D. at ISU’s Developmental Robotics Lab, he developed novel methods for behavioral object exploration and multi-modal perception. He went on to be a clinical assistant professor with the Texas Institute for Discovery, Education, and Science at UT Austin and a postdoctoral associate working with Peter Stone at the Artificial Intelligence lab. Sinapov’s research interests include developmental robotics, computational perception, autonomous manipulation, and human-robot interaction.

Links

• Download mp3 (10.6 MB)
• Sinapov’s research
• Subscribe to Robohub using iTunes, RSS, or Spotify
• Support us on Patreon

In this episode, Lilly interviews Nathan Michael, Professor and Director of the Resilient Intelligent Systems Lab at the Robotics Institute within Carnegie Mellon University. They discuss introspection, adaptation, and evolvement in robotics. Michael speaks about topics in state estimation and distributed perception, and other challenges in control, perception, and cognition for both single and multi-robot systems.

Nathan Michael

Nathan Michael is an Associate Research Professor and the Director of the Resilient Intelligent Systems Lab at the Robotics Institute within Carnegie Mellon University (CMU). Professor Michael’s research seeks to enable the development of self-sufficient robots and teams of robots that can respond to disasters; robots which can perform the tasks of first responders in order to reduce the number of people placed in harm’s way. His research employs the principles of resilient intelligence and persistent knowledge to enable robots to innovate to solve complex problems, to overcome unanticipated challenges and to learn from their experiences. Applications of Professor Michael’s work include the use of autonomous robots to explore earthquake-damaged buildings and the deployment of unmanned aerial systems to inspect nuclear reactor cores and measure radiation levels.

Links

• Download mp3 (14.9 MB)
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS
• Support us on Patreon

In this episode, we hear from Luca Colasanto, Senior Robotic Scientist at Realtime Robotics, about real-time robot motion planning in dynamic and complex environments with human-robot collaboration. Realtime Robotics focuses on accelerating conventional motion planning through optimization of algorithms and hardware to allow safe use of robotic tools in work areas with humans. Luca spoke to our interviewer Kate about Realtime Robotic’s fast motion planning technology, including key aspects, such as perception, algorithms and custom hardware.

Luca Colasanto

Luca Colasanto is a Sr. Scientist at Realtime Robotics focusing on AI-based grasping and multi-robot optimization. Luca completed his PhD in Humanoid Robotics at Italian Institute of Technology, focusing on control systems for bipedal walking machines and compliant actuators. During the past ten years, Luca has worked with successful startups to design and prototype new robotics products, and with science and technology leaders, such as Robert Bosch LLC (CA), Institute for Human and Machine Cognition (FL), and École Polytechnique Fédérale de Lausanne (Switzerland), He is the listed inventor in two patent applications and the author of several scientific publications. Luca speaks four languages and is a keen traveler, having visited or lived in more than 20 countries.

Links

• Download mp3 (15.1 MB)
• Realtime Robotics (Videos, Twitter, LinkedIn)
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS
• Support us on Patreon

In this episode, Shihan Lu interviews Dr. Heather Culbertson, Assistant Professor in the Computer Science Department at the University of Southern California, about her work in haptics. Dr. Culbertson discusses the data-driven realistic texture modeling and rendering, haptic technologies in the social touch, the combination of haptics and robots, expectations and obstacles of haptics in the next 5 years.

Heather Culbertson

Heather Culbertson is a Gabilan Assistant Professor of Computer Science at the University of Southern California. Her research focuses on the design and control of haptic devices and rendering systems, human-robot interaction, and virtual reality. Particularly she is interested in creating haptic interactions that are natural and realistically mimic the touch sensations experienced during interactions with the physical world. Previously, she was a research scientist in the Department of Mechanical Engineering at Stanford University where she worked in the Collaborative Haptics and Robotics in Medicine (CHARM) Lab. She received her PhD in the Department of Mechanical Engineering and Applied Mechanics (MEAM) at the University of Pennsylvania in 2015 working in the Haptics Group, part of the General Robotics, Automation, Sensing and Perception (GRASP) Laboratory.

Links

• Download mp3 (14.8 MB)
• Culbertson’s research website
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS
• Support us on Patreon

In this episode, Audrow Nash interviews Bilge Mutlu, Associate Professor at the University of Wisconsin–Madison, about design-thinking in human-robot interaction. Professor Mutlu discusses design-thinking at a high-level, how design relates to science, and he speaks about the main areas of his work: the design space, the evaluation space, and how features are used within a context. He also gives advice on how to apply a design-oriented mindset.

Bilge Mutlu

Bilge Mutlu is an Associate Professor of Computer Science, Psychology, and Industrial Engineering at the University of Wisconsin–Madison. He directs the Wisconsin HCI Laboratory and organizes the WHCI+D Group. He received his PhD degree from Carnegie Mellon University‘s Human-Computer Interaction Institute.

Links

• Download mp3 (15.4 MB)
• Bilge’s personal website
• The HCI Lab’s website
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS
• Support us on Patreon

In this interview, Audrow Nash interviews Jaime Fernández Fisac, a PhD student at University of California, Berkeley, working with Professors Shankar Sastry, Claire Tomlin, and Anca Dragan. Fisac is interested in ensuring that autonomous systems such as self-driving cars, delivery drones, and home robots can operate and learn in the world—while satisfying safety constraints. Towards this goal, Fisac discusses different examples of his work with unmanned aerial vehicles and talks about safe robot learning in general; including, the curse of dimensionality and how it impacts control problems (including how some systems can be decomposed into simpler control problems), how simulation can be leveraged before trying learning on a physical robot, safe sets, and how a robot can modify its behavior based on how confident it is that its model is correct.

Below are two videos of work that was discussed during the interview.  The top video is on a framework for learning-based control, and the bottom video discusses adjusting the robot’s confidence about a human’s actions based on how predictably the human is behaving.

Jaime Fernández Fisac

Jaime Fernández Fisac is a final-year Ph.D. candidate in Electrical Engineering and Computer Sciences at the University of California, Berkeley. He received a B.S./M.S. degree in Electrical Engineering from the Universidad Politécnica de Madrid, Spain, in 2012, and a M.Sc. in Aeronautics from Cranfield University, U.K., in 2013. He is a recipient of the La Caixa Foundation fellowship. His research interests lie between control theory and artificial intelligence, with a focus on safety assurance for autonomous systems. He works to enable AI systems to reason explicitly about the gap between their models and the real world, so that they can safely interact with uncertain environments and human beings, even under inaccurate assumptions.

Links

• Download mp3 (28.0 MB)
• Jaime Fernández Fisac’s website
• InterACT Lab
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS
• Support us on Patreon

In this episode, Audrow Nash interviews Pauline Pound, Philippe Morere, and Yujung Liu about the work they presented at the 2018 International Conference on Intelligent Robots and Systems (IROS) in Madrid, Spain.

Pauline Pounds, Associate Professor at the University of Queensland, speaks about building robots that can endure children.  She discusses the tradeoffs of designing a robot that can survive children and cannot harm the children. Pounds talks about how the robot has performed with children so far, the hardware design, and her future direction with this work.

Philippe Morere, a PhD student at University of Sydney, speaks about reinforcement learning in partially observable environments.  Morere discusses the intuition behind this work and the problems that he solved using this approach.

Yujung Liu, a master’s student from the National Taiwan University, speaks about a painting robot.  He discusses how the portrait is constructed, including how colors are chosen, how the features are made cartoon-like, and his future direction.

Links

• Download mp3 (11.3 MB)
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS
• Support us on Patreon

In this interview, Audrow Nash interviews Kristoffer Richardsson, Michael Zillich, and Paulo Alvito.

Kristoffer Richardsson, Developer at Bitcraze in Sweden, speaks about small open-source unmanned aerial vehicle and different ways of localizing them.

Michael Zillich, CTO of Blue Danube Robotics in Austria, discusses a robotics platform for picking up toys, called Kenny. Zillich talks about Kenny’s design, including sensors and its robotic arm. He discusses how children can show the robot where to put specific toys, about how this large European project was coordinated, and about possibly commercializing this platform.

Paulo Alvito, CEO and CTO and Co-founder of IDMind in Portugal, discusses remote inspection robot for sewers. Alvito discusses what it means to be remotely autonomous, the design of the robot, and how these robots communicate data.

Links

• Download mp3 (17.3 MB)
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS
• Support us on Patreon

In this episode, Audrow Nash interviews Alexandros Kogkas, Katie Driggs-Campbell, and Martin Karlsson about the work they presented at the 2018 International Conference on Intelligent Robots and Systems (IROS) in Madrid, Spain.

Alexandros Kogkas is a PhD Candidate at the Imperial College London and he speaks about an eye tracking framework to understand where a person is looking.  This framework can be used to understand a person’s intentions, for example to hand a surgeon the correct tool or helping a person who is paraplegic.   Kogkas discusses how the framework works, possible applications, and his future plans for this framework.

Katie Driggs-Campbell is a Post Doctoral Researcher at Stanford’s Intelligent System Laboratory and is—soon to be—an Assistant Professor at University of Illinois Urbana-Champaign (UIUC). She speaks about making inferences about the world from human actions, specifically in the context of autonomous cars.  In the work she discusses, they use a model of a human driver that they use infer what is happening in the world, for example a human using a crosswalk. Driggs-Campbell talks about how they evaluate this work.

Martin Karlsson is a PhD student at Lund University in Sweden, and he speaks about a haptic interface to mirror robotic arms that requires no force sensing.  He discusses a feedback law that allows a mirroring of forces and his future work to deal with joint friction.

Links

• Download mp3 (12.0 MB)
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS
• Support us on Patreon

In this episode, Audrow Nash interview Magnus Egerstedt, Professor at the Georgia Institute of Technology, about a way for anyone interested in swarm robotics to test their ideas on hardware, called the Robotarium.  The Robotarium is a 725-square-foot lab at the the Georgia Institute of Technology that houses nearly 100 rolling and flying robots.  To test their ideas, people can write their own programs, upload them to the Robotarium, and then watch the machines carry out their commands.

In this interview, Egerstedt speaks about the kinds of robots used in the Robotarium, design decisions in making the Robotarium, the differences between doing research in simulation and on hardware, and about lessons learnt and the challenges of building the Robotarium.

Magnus Egerstedt

Dr. Magnus Egerstedt is the Steve W. Chaddick School Chair and Professor in the School of Electrical and Computer Engineering at the Georgia Institute of Technology. He holds secondary appointments in the Woodruff School of Mechanical Engineering, the School of Interactive Computing, and the Guggenheim School of Aerospace Engineering, and has previously served as the Executive Director for the Institute for Robotics and Intelligent Machines at Georgia Tech. He received the M.S. degree in Engineering Physics and the Ph.D. degree in Applied Mathematics from the Royal Institute of Technology, Stockholm, Sweden, the B.A. degree in Philosophy from Stockholm University, and was a Postdoctoral Scholar at Harvard University. Dr. Egerstedt conducts research in the areas of control theory and robotics, with particular focus on control and coordination of complex networks, such as multi-robot systems, mobile sensor networks, and cyber-physical systems. Magnus Egerstedt is a Fellow of the IEEE and has received a number of teaching and research awards, including the Ragazzini Award from the American Automatic Control Council, the Outstanding Doctoral Advisor Award and the HKN Outstanding Teacher Award from Georgia Tech, the Alumni of the Year Award from the Royal Institute of Technology, and the CAREER Award from the U.S. National Science Foundation.

Links

• Download mp3 (14.4 MB)
• GRITS Lab
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS
• Support us on Patreon

In this episode, Audrow Nash interviews Juxi Leitner, a Postdoctoral Research Fellow at QUT; and Nicholas Panitz, Ben Wilson, and James Brett, from CSIRO.

Leitner speaks about the Amazon Picking challenge, a challenge to advance the state of robotic grasping, and their robot which won the challenge in 2017. Their robot is similar to a cartesian 3D printer in form and uses either a suction cup or a pinch gripper for grabbing objects. Their robot has a depth camera and uses a digital scale to determine if an object has been picked up successfully. Leitner discusses what their team did differently from other teams that helped them win the competition.

Panitz, Wilson, and Brett speak about their hexapod robots. Their hexapods are for several purposes, such as environmental monitoring and remote inspection. They choose to use hexapods because they are statically stable. They discuss the design of their hexapods and how research works at Commonwealth Scientific and Industrial Research Organization, or CSIRO.

A video of the robot Leitner discusses, #Cartman:

An example of CSIRO’s hexapod robots for inspection:

Links

• Download mp3 (12.9 MB)
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS
• Support us on Patreon

In this interview, Audrow speaks with Andrea Bajcsy and Dylan P. Losey about a method that allows robots to infer a human’s objective through physical interaction. They discuss their approach, the challenges of learning complex tasks, and their experience collaborating between different universities.

Some examples of people working with the more typical impedance control (left) and Bajcsy and Losey’s learning method (right).

To learn more, see this post on Robohub from the Berkeley Artificial Intelligence Research (BAIR) Lab.

Andrea Bajcsy
Andrea Bajcsy is a Ph.D. student in Electrical Engineering and Computer Sciences at the University of California Berkeley. She received her B.S. degree in Computer Science at the University of Maryland and was awarded the NSF Graduate Research Fellowship in 2016. At Berkeley, she works in the Interactive Autonomy and Collaborative Technologies Laboratory researching physical human-robot interaction.

Dylan P. Losey

Dylan P. Losey received the B.S. degree in mechanical engineering from Vanderbilt University, Nashville, TN, USA, in 2014, and the M.S. degree in mechanical engineering from Rice University, Houston, TX, USA, in 2016.

He is currently working towards the Ph.D. degree in mechanical engineering at Rice University, where he has been a member of the Mechatronics and Haptic Interfaces Laboratory since 2014.  In addition, between May and August 2017, he was a visiting scholar in the Interactive Autonomy and Collaborative Technologies Laboratory at the University of California, Berkeley.  He researches physical human-robot interaction; in particular, how robots can learn from and adapt to human corrections.

Mr. Losey received an NSF Graduate Research Fellowship in 2014, and the 2016 IEEE/ASME Transactions on Mechatronics Best Paper Award as a first author.

Links

• InterACT Lab
• MAHL Lab
• Andrea Bajcsy’s website
• Dylan’s Losey’s website
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS
• Support us on Patreon
• Download MP3 (14.9 MB)

A Lucie micro drone takes off from a performer’s hand as part of a drone show. Photo: Verity Studios 2017

2017 was the year where indoor drone shows came into their own. Verity Studios’ Lucie drones alone completed more than 20,000 autonomous flights. A Synthetic Swarm of 99 Lucie micro drones started touring with Metallica (the tour is ongoing and was just announced the 5th highest grossing tour worldwide for 2017). Micro drones are now performing at Madison Square Garden as part of each New York Knicks home game — the first resident drone show in a full-scale arena setting. Since early 2017, a drone swarm has been performing weekly on a first cruise ship. And micro drones performed thousands of flights at Changi Airport Singapore as part of its 2017 Christmas show.

Technologically, indoor drone show systems are challenging. They are among the most sophisticated automation systems in existence, with dozens of autonomous robotic aircraft operating in a safety-critical environment. Indoor drone shows require sophisticated, distributed system control and communications architectures to split up and recombine sensing and computation between aircraft and their off-board infrastructure. Core challenges are not unlike those found in modern systems for manned aviation (e.g., combining auto-pilots, GPS, and air traffic control) and in creating tomorrow’s smart cities (e.g., combining semi-autonomous cars with intelligent traffic lights in a city).

These technological challenges are compounded by another: At least for permanent show installations, these systems need to be operated by non-experts. Two years ago, in one of the first major indoor drone shows, a swarm of micro drones flew over the audience at TED 2016. That system was operated by Verity Studios’ expert engineers. Creating a system that is easy enough to use, and reliable enough, to be operated by show staff is a huge technical challenge of its own. All of Verity’s 2017 shows mentioned above were fully client-operated, which speaks to the maturity that Verity’s drone show system has achieved.

Selection of Verity Studios’ indoor drone shows, from the drone swarm at TED 2016 to 20,000 autonomous indoor drone show flights in 2017 alone.

For my colleagues and me, it is these technological challenges, together with the visual impact of indoor drone shows, that makes these systems so much fun and hugely rewarding to work with.

Creative potential

Creatively, the capabilities of today’s indoor drone show systems barely scratch the surface of the technology’s potential. For centuries, show designers were restricted to static scenes. Curtains were required to hide scene changes from the audience, lest stage hands rushing to move set pieces destroy the magic created by a live show. The introduction of automation to seamlessly move backdrops and other stage elements, followed by the debut of automated lighting to smoothly pan and tilt traditional, stationary illumination were revolutionary.

Drones hold the potential for pushing automation further. The Lucies shown in the images above give a first inkling of the creative potential of flying lights that can be freely positioned in 3D space, appearing at will. Larger drones allow to extend that concept to nearly any object, including the creation of flying characters.

Safety

The most critical challenge for indoor drone show systems is safety. Indoor drone shows feature dozens of drones flying simultaneously and in tight formations, close to crowds of people, in a repeated fashion, in the high-pressure environment of a live show. For example, as part of the currently running New York Knicks drone show, 32 drones perform above 16 dancers, live in front of up to 20,000 people in New York’s Madison Square Garden arena, 44 times per season.

There are really only three ways to safely fly drones at live events.

The first way to achieve safety is the same that keeps commercial aviation safe: System redundancy. Using this approach, Verity Studios’ larger, Stage Flyer drones performed safely on Broadway, completing 398 shows and more than 7,000 autonomous flights, flying 8 times a week, in front of up to 2,000 people for a year, without safety nets. The Stage Flyer drones are designed around redundancy. At least two of each components are used (e.g., two batteries, two flight computers, and a duplicate of each sensor) or existing redundancies are exploited. For example, the Stage Flyer drones have only four propellers and motors, like any quadcopter. However, advanced algorithms that exploit the physics of flight allow these multi-rotor vehicles to fly with less than 4 propellers. The overall design allows these drones to continue to fly in spite of any individual component failure. For example, in one of the last Broadway shows, a Stage Flyer experienced a battery failure. The drone switched into its safety flight mode and landed, and the show continued with 7 instead of 8 drones. This approach to drone safety remains highly unusual — all drones available for purchase today have single points of failure.

Verity Studios drone show, 2017 Event Safety Summit, Rock Lititz. Photo: Event Safety Alliance 2017

The second approach to safety is physical separation. This is how safety is usually achieved for outdoor drone shows: Drones perform over a body of water or some roads are temporarily closed to create a large-enough area without people. For example, the Intel drone show at the Super Bowl was recorded far away from the NRG stadium. In fact, for the Super Bowl, safety went even a step further, also adding “temporal separation” to the physical separation (the drone show was actually pre-recorded days ahead of time, and viewers in the stadium and on TV were only shown a video recording). For indoor drone lightshows, physical separation can be achieved using safety nets.

The third approach to safely flying drones at live events is to make the drones so small that they have high inherent safety. Verity Studios’ Lucie micro drones weigh less than 1.8 ounces or 50 grams (including their flexible hull).

As the continuing string of safety incidents involving drones at live events attests, not everyone takes drone safety seriously. This is why my colleagues and I have worked with aviation experts and leading creatives to summarize best practices in an overview paper: Drone shows – Creative potential and best practices.

So, what’s in store for 2018? The appetite for indoor drone shows is huge, which is why Verity Studios is growing its team. And given the 2017 track record, there is a lot to look forward to — your favorite venue’s ceiling is the limit!

In this interview, Audrow Nash interviews Helen Huang, Joint Professor at the University of North Carolina at Chapel Hill and North Carolina State, about a method of tuning powered lower limb prostheses. Huang explains how powered prostheses are adjusted for each patient and how she is using supervised and reinforcement learning to tune prosthesis. Huang also discusses why she is not using the energetic cost of transport as a metric and the challenge of people adapting to a device while it learns from them.

Helen Huang

Helen Huang is a Professor in the Joint Department of Biomedical Engineering at the University of North Carolina at Chapel Hill (UNC) and North Carolina State University (NCSU) and the Director of the NCSU/UNC Closed-Loop Engineering for Advanced Rehabilitation (CLEAR) Core. Huang is interested in neural-machine interfaces for prostheses and exoskeletons, wearer-robot interaction, adaptive and optimal control of wearable robots, and human movement control. She has developed an EMG-based neural interface for robotic prosthetic legs. She was a recipient of the Delsys Prize for Innovation in Electromyography, the Mary E. Switzer Fellowship with the National Institute on Disability, Independent Living, and Rehabilitation Research (NIDILRR), and a NSF CAREER Award and was named NC State faculty scholar in 2015.

Links

• Huang’s research lab
• Download mp3 (19.9 MB)
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS
• Support us on Patreon

Image: ICRA 2017

In this episode, Audrow Nash interviews several companies at the International Conference for Robotics and Automation (ICRA). ICRA is the IEEE Robotics and Automation Society’s biggest conference and one of the leading international forums for robotics researchers to present their work.

Interviews:

Howard Michel, Chief Technology Officer of UBTECH Education, on a humanoid robot for STEM (Science, Technology, Engineering, and Math) education for a range of ages. (Begins at 1m 20s.)

Li Bingbing, Software Engineer and Cofounder of Transforma in Singapore, on a robot for painting tall buildings. (Begins at 10m 25s.)

Xianbao Chen, Associate Researcher at Shanghai Jiao Tong University, on a hexapod robot for large parts machining. (Begins at 17m 5s.)

Lester Teh Chee Onn, Environmental Engineer at Advisian, on a watercraft for environmental monitoring. (Begins at 27m.)

Links

• Download mp3 (18.8 MB)
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS
• Support us on Patreon

In this episode, Audrow Nash interviews Katsu Yamane, Senior Research Scientist at Disney, about robotics in Disney. Yamane discusses Disney’s history with robots, how Disney currently uses Robots, how designing robots at Disney is different than in academia or industry, a realistic robot simulator used by Disney’s animators, and on becoming a Disney Research “Imagineer.”

Katsu Yamane

Katsu received his PhD in mechanical engineering from University of Tokyo in 2002. Following postdoctoral work at Carnegie Mellon University from 2002 to 2003, he was a faculty member at University of Tokyo until he joined Disney Research, Pittsburgh, in October 2008. His main research area is humanoid robot control and motion synthesis, in particular methods involving human motion data and dynamic balancing. He is also interested in developing algorithms for creating character animation. He has always been fascinated by the way humans control their bodies, which led him to the research on biomechanical human modeling and simulation to understand human sensation and motor control.

Links

• Download mp3 (9.1MB)
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS

Adriana Schulz, an MIT PhD student in the Computer Science and Artificial Intelligence Laboratory, demonstrates the InstantCAD computer-aided-design-optimizing interface. Photo: Rachel Gordon/MIT CSAIL

Almost every object we use is developed with computer-aided design (CAD). Ironically, while CAD programs are good for creating designs, using them is actually very difficult and time-consuming if you’re trying to improve an existing design to make the most optimal product. Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and Columbia University are trying to make the process faster and easier: In a new paper, they’ve developed InstantCAD, a tool that lets designers interactively edit, improve, and optimize CAD models using a more streamlined and intuitive workflow.

“From more ergonomic desks to higher-performance cars, this is really about creating better products in less time,” says Department of Electrical Engineering and Computer Science PhD student and lead author Adriana Schulz, who will be presenting the paper at this month’s SIGGRAPH computer-graphics conference in Los Angeles. “We think this could be a real game changer for automakers and other companies that want to be able to test and improve complex designs in a matter of seconds to minutes, instead of hours to days.”

The paper was co-written by Associate Professor Wojciech Matusik, PhD student Jie Xu, and postdoc Bo Zhu of CSAIL, as well as Associate Professor Eitan Grinspun and Assistant Professor Changxi Zheng of Columbia University.

Traditional CAD systems are “parametric,” which means that when engineers design models, they can change properties like shape and size (“parameters”) based on different priorities. For example, when designing a wind turbine you might have to make trade-offs between how much airflow you can get versus how much energy it will generate.

InstantCAD enables designers to interactively edit, improve, and optimize CAD models using a more streamlined and intuitive workflow. Photo: Rachel Gordon/MIT CSAIL

However, it can be difficult to determine the absolute best design for what you want your object to do, because there are many different options for modifying the design. On top of that, the process is time-consuming because changing a single property means having to wait to regenerate the new design, run a simulation, see the result, and then figure out what to do next.

With InstantCAD, the process of improving and optimizing the design can be done in real-time, saving engineers days or weeks. After an object is designed in a commercial CAD program, it is sent to a cloud platform where multiple geometric evaluations and simulations are run at the same time.

With this precomputed data, you can instantly improve and optimize the design in two ways. With “interactive exploration,” a user interface provides real-time feedback on how design changes will affect performance, like how the shape of a plane wing impacts air pressure distribution. With “automatic optimization,” you simply tell the system to give you a design with specific characteristics, like a drone that’s as lightweight as possible while still being able to carry the maximum amount of weight.

The reason it’s hard to optimize an object’s design is because of the massive size of the design space (the number of possible design options).

“It’s too data-intensive to compute every single point, so we have to come up with a way to predict any point in this space from just a small number of sampled data points,” says Schulz. “This is called ‘interpolation,’ and our key technical contribution is a new algorithm we developed to take these samples and estimate points in the space.”

Matusik says InstantCAD could be particularly helpful for more intricate designs for objects like cars, planes, and robots, particularly for industries like car manufacturing that care a lot about squeezing every little bit of performance out of a product.

“Our system doesn’t just save you time for changing designs, but has the potential to dramatically improve the quality of the products themselves,” says Matusik. “The more complex your design gets, the more important this kind of a tool can be.”

Because of the system’s productivity boosts and CAD integration, Schulz is confident that it will have immediate applications for industry. Down the line, she hopes that InstantCAD can also help lower the barrier for entry for casual users.

“In a world where 3-D printing and industrial robotics are making manufacturing more accessible, we need systems that make the actual design process more accessible, too,” Schulz says. “With systems like this that make it easier to customize objects to meet your specific needs, we hope to be paving the way to a new age of personal manufacturing and DIY design.”

The project was supported by the National Science Foundation.

By Christoph Salge, Marie Curie Global Fellow, University of Hertfordshire

How do you stop a robot from hurting people? Many existing robots, such as those assembling cars in factories, shut down immediately when a human comes near. But this quick fix wouldn’t work for something like a self-driving car that might have to move to avoid a collision, or a care robot that might need to catch an old person if they fall. With robots set to become our servants, companions and co-workers, we need to deal with the increasingly complex situations this will create and the ethical and safety questions this will raise.

Science fiction already envisioned this problem and has suggested various potential solutions. The most famous was author Isaac Asimov’s Three Laws of Robotics, which are designed to prevent robots harming humans. But since 2005, my colleagues and I at the University of Hertfordshire, have been working on an idea that could be an alternative.

Instead of laws to restrict robot behaviour, we think robots should be empowered to maximise the possible ways they can act so they can pick the best solution for any given scenario. As we describe in a new paper in Frontiers, this principle could form the basis of a new set of universal guidelines for robots to keep humans as safe as possible.

The Three Laws

Asimov’s Three Laws are as follows:

• A robot may not injure a human being or, through inaction, allow a human being to come to harm.
• A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
• A robot must protect its own existence as long as such protection does not conflict with the First or Second Laws.

While these laws sound plausible, numerous arguments have demonstrated why they are inadequate. Asimov’s own stories are arguably a deconstruction of the laws, showing how they repeatedly fail in different situations. Most attempts to draft new guidelines follow a similar principle to create safe, compliant and robust robots.

One problem with any explicitly formulated robot guidelines is the need to translate them into a format that robots can work with. Understanding the full range of human language and the experience it represents is a very hard job for a robot. Broad behavioural goals, such as preventing harm to humans or protecting a robot’s existence, can mean different things in different contexts. Sticking to the rules might end up leaving a robot helpless to act as its creators might hope.

Our alternative concept, empowerment, stands for the opposite of helplessness. Being empowered means having the ability to affect a situation and being aware that you can. We have been developing ways to translate this social concept into a quantifiable and operational technical language. This would endow robots with the drive to keep their options open and act in a way that increases their influence on the world.

When we tried simulating how robots would use the empowerment principle in various scenarios, we found they would often act in surprisingly “natural” ways. It typically only requires them to model how the real world works but doesn’t need any specialised artificial intelligence programming designed to deal with the particular scenario.

But to keep people safe, the robots need to try to maintain or improve human empowerment as well as their own. This essentially means being protective and supportive. Opening a locked door for someone would increase their empowerment. Restraining them would result in a short-term loss of empowerment. And significantly hurting them could remove their empowerment altogether. At the same time, the robot has to try to maintain its own empowerment, for example by ensuring it has enough power to operate and it does not get stuck or damaged.

Robots could adapt to new situations

Using this general principle rather than predefined rules of behaviour would allow the robot to take account of the context and evaluate scenarios no one has previously envisaged. For example, instead of always following the rule “don’t push humans”, a robot would generally avoid pushing them but still be able to push them out of the way of a falling object. The human might still be harmed but less so than if the robot didn’t push them.

In the film I, Robot, based on several Asimov stories, robots create an oppressive state that is supposed to minimise the overall harm to humans by keeping them confined and “protected”. But our principle would avoid such a scenario because it would mean a loss of human empowerment.

While empowerment provides a new way of thinking about safe robot behaviour, we still have much work to do on scaling up its efficiency so it can easily be deployed on any robot and translate to good and safe behaviour in all respects. This poses a very difficult challenge. But we firmly believe empowerment can lead us towards a practical solution to the ongoing and highly debated problem of how to rein in robots’ behaviour, and how to keep robots -– in the most naive sense -– “ethical”.

This article was originally published on The Conversation. Read the original article.

The Robot Academy is a new learning resource from Professor Peter Corke and the Queensland University of Technology (QUT), the team behind the award-winning Introduction to Robotics and Robotic Vision courses. There are over 200 lessons available, all for free.

The lessons were created in 2015 for the Introduction to Robotics and Robotic Vision courses. We describe our approach to creating the original courses in the article, An Innovative Educational Change: Massive Open Online Courses in Robotics and Robotic Vision. The courses were designed for university undergraduate students but many lessons are suitable for anybody, as you can easily see the difficulty rating for each lesson. Below are lessons from inverse kinematics and robot motion.

You can watch the entire masterclass on the Robot Academy website.

Introduction

In this video lecture, we will learn about inverse kinematics, that is, how to compute the robot’s joint angles given the desired pose of their end-effector and knowledge about the dimensions of its links. We will also learn about how to generate paths that lead to a smooth coordinated motion of the end-effector.

Inverse kinematics for a 2-joint robot arm using geometry

In this lesson, we revisit the simple 2-link planar robot and determine the inverse kinematic function using simple geometry and trigonometry.

Inverse kinematics for a 2-joint robot arm using algebra

You can watch the entire masterclass on the Robot Academy website.

If you liked this article, you may also enjoy:

• Smart Grasping System available on ROS Development Studio
• The Robot Academy: Lessons in image formation and 3D vision
• The Robot Academy: The open online robotics education resource 
• Machine Learning with OpenAI Gym on ROS Development Studio
• SICK LMS full LIDAR teardown
• Envisioning the future of robotics
• How to mount an external removable hard drive on a robot

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

MATLAB© is a programming language and environment designed for scientific computing. It is one of the best languages for developing robot control algorithms and is widely used in the research community. While it is often thought of as an offline programming language, there are several ways to interface with it to control robotic hardware ‘in the loop’. As part of our own development we surveyed a number of different projects that accomplish this by using a message passing system and we compared the approaches they took. This post focuses on bindings for the following message passing frameworks: LCM, ROS, DDS, and ZeroMQ.

The main motivation for using MATLAB to prototype directly on real hardware is to dramatically accelerate the development cycle by reducing the time it takes to find out out whether an algorithm can withstand ubiquitous real-world problems like noisy and poorly-calibrated sensors, imperfect actuator controls, and unmodeled robot dynamics. Additionally, a workflow that requires researchers to port prototype code to another language before being able to test on real hardware can often lead to weeks or months being lost in chasing down new technical bugs introduceed by the port. Finally, programming in a language like C++ can pose a significant barrier to controls engineers who often have a strong electro-mechanical background but are not as strong in computer science or software engineering.

We have also noticed that over the past few years several other groups in the robotics community also experience these problems and have started to develop ways to control hardware directly from MATLAB.

---

The Need for External Languages

The main limitation when trying to use MATLAB to interface with hardware stems from the fact that its scripting language is fundamentally single threaded. It has been designed to allow non-programmers to do complex math operations without needing to worry about programming concepts like multi-threading or synchronization. However, this poses a problem for real-time control of hardware because all communication is forced to happen synchronously in the main thread. For example, if a control loop runs at 100Hz and it takes a message ~8ms for a round-trip, the main thread ends up wasting 80% of the available time budget waiting for a response without doing any actual work.

A second hurdle is that while MATLAB is very efficient in the execution of math operations, it is not particularly well suited for byte manipulation. This makes it difficult to develop code that can efficiently create and parse binary message formats that the target hardware can understand. Thus, after having the main thread spend its time waiting for and parsing the incoming data, there may not be any time left for performing interesting math operations.

Figure 1. Communications overhead in the main MATLAB thread

Pure MATLAB implementations can work for simple applications, such as interfacing with an Arduino to gather temperature data or blink an LED, but it is not feasible to control complex robotic systems (e.g. a humanoid) at high rates (e.g. 100Hz-1KHz). Fortunately, MATLAB does have the ability to interface with other programming languages that allow users to create background threads that can offload the communications aspect from the main thread.

Figure 2. Communications overhead offloaded to other threads

Out of the box MATLAB provides two interfaces to other languages: MEX for calling C/C++ code, and the Java Interface for calling Java code. There are some differences between the two, but at the end of the day the choice effectively comes down to personal preference. Both provide enough capabilities for developing sophisticated interfaces and have orders of magnitude better performance than required. There are additional interfaces to other languages, but those require additional setup steps.

---

Message Passing Frameworks

Message passing frameworks such as Robot Operating System (ROS) and Lightweight Communication and Marshalling (LCM) have been widely adopted in the robotics research community. At the core they typically consist of two parts: a way to exchange data between processes (e.g. UDP/TCP), as well as a defined binary format for encoding and decoding the messages. They allow systems to be built with distributed components (e.g. processes) that run on different computers, different operating systems, and different programming languages.

The resulting systems are very extensible and provide convenient ways for prototyping. For example, a component communicating with a physical robot can be exchanged with a simulator without affecting the rest of the system. Similarly, a new walking controller could be implemented in MATLAB and communicate with external processes (e.g. robot comms) through the exchange of messages. With ROS and LCM in particular, their flexibility, wide-spread adoption, and support for different languages make them a nice starting point for a MATLAB-hardware interface.

---

Lightweight Communication and Marshalling (LCM)

LCM was developed in 2006 at MIT for their entry to DARPA’s Urban Challenge. In recent years it has become a popular alternative to ROS-messaging, and it was as far as we know the first message passing framework for robotics that supported MATLAB as a core language.

The snippet below shows how the MATLAB code for sending a command message could look like. The code creates a struct-like message, sets desired values, and publishes it on an appropriate channel.

%% MATLAB code for sending an LCM message
% Setup
lc = lcm.lcm.LCM.getSingleton();

% Fill message
cmd = types.command();
cmd.position = [1 2 3];
cmd.velocity = [1 2 3];

% Publish
lc.publish('COMMAND_CHANNEL', cmd);

Interestingly, the backing implementation of these bindings was done in pure Java and did not contain any actual MATLAB code. The exposed interface consisted of two Java classes as well as auto-generated message types.

• The LCM class provides a way to publish messages and subscribe to channels
• The generated Java messages handle the binary encoding and exposed fields that MATLAB can access
• The MessageAggregator class provides a way to receive messages on a background thread and queue them for MATLAB.

Thus, even though the snippet looks similar to MATLAB code, all variables are actually Java objects. For example, the struct-like command type is a Java object that exposes public fields as shown in the snippet below. Users can access them the same way as fields of a standard MATLAB struct (or class properties) resulting in nice syntax. The types are automatically converted according to the type mapping.

/**
 * Java class that behaves like a MATLAB struct
 */
public final class command implements lcm.lcm.LCMEncodable
{
    public double[] position;
    public double[] velocity;
    // etc. ...
}

Receiving messages is done by subscribing an aggregator to one or more channels. The aggregator receives messages from a background thread and stores them in a queue that MATLAB can access in a synchronous manner using aggregator.getNextMessage(). Each message contains the raw bytes as well as some meta data for selecting an appropriate type for decoding.

%% MATLAB code for receiving an LCM message
% Setup
lc = lcm.lcm.LCM.getSingleton();
aggregator = lcm.lcm.MessageAggregator();
lc.subscribe('FEEDBACK_CHANNEL', aggregator);

% Continuously check for new messages
timeoutMs = 1000;
while true

    % Receive raw message
    msg = aggregator.getNextMessage(timeoutMs);

    % Ignore timeouts
    if ~isempty(msg)

        % Select message type based on channel name
        if strcmp('FEEDBACK_CHANNEL', char(msg.channel))

            % Decode raw bytes to a usable type
            fbk = types.feedback(msg.data);

            % Use data
            position = fbk.position;
            velocity = fbk.velocity;

        end

    end
end

The snippet below shows a simplified version of the backing Java code for the aggregator class. Since Java is limited to a single return argument, the getNextMessage call returns a Java type that contains the received bytes as well as meta data to identify the type, i.e., the source channel name.

/**
 * Java class for receiving messages in the background
 */
public class MessageAggregator implements LCMSubscriber {

    /**
     * Value type that combines multiple return arguments
     */
    public static class Message {

        final public byte[] data; // raw bytes
        final public String channel; // source channel name

        public Message(String channel_, byte[] data_) {
            data = data_;
            channel = channel_;
        }
    }

    /**
     * Method that gets called from MATLAB to receive new messages
     */
    public synchronized Message getNextMessage(long timeout_ms) {

		if (!messages.isEmpty()) {
		    return messages.removeFirst();
        }

        if (timeout_ms == 0) { // non-blocking
            return null;
        }

        // Wait for new message until timeout ...
    }

}

Note that the getNextMessage method requires a timeout argument. In general it is important for blocking Java methods to have a timeout in order to prevent the main thread from getting stuck permanently. Being in a Java call prohibits users from aborting the execution (ctrl-c), so timeouts should be reasonably short, i.e., in the low seconds. Otherwise this could cause the UI to become unresponsive and users may be forced to close MATLAB without being able to save their workspace. Passing in a timeout of zero serves as a non-blocking interface that immediately returns empty if no messages are available. This is often useful for working with multiple aggregators or for integrating asynchronous messages with unknown timing, such as user input.

Overall, we thought that this was a well thought out API and a great example for a minimum viable interface that works well in practice. By receiving messages on a background thread and by moving the encoding and decoding steps to the Java language, the main thread is able to spend most of its time on actually working with the data. Its minimalistic implementation is comparatively simple and we would recommend it as a starting point for developing similar interfaces.

Some minor points for improvement that we found were:

• The decoding step fbk = types.feedback(msg.data) forces two unnecessary translations due to msg.data being a byte[], which automatically gets converted to and from int8. This could result in a noticeable performance hit when receiving larger messages (e.g. images) and could be avoided by adding an overload that accepts a non-primitive type that does not get translated, e.g., fbk = types.feedback(msg).
• The Java classes did not implement Serializable, which could become bothersome when trying to save the workspace.
• We would prefer to select the decoding type during the subscription step, e.g., lc.subscribe(‘FEEDBACK_CHANNEL’, aggregator, ‘types.feedback’), rather than requiring users to instantiate the type manually. This would clean up the parsing code a bit and allow for a less confusing error message if types are missing.

---

Robot Operating System (ROS)

ROS is by far the most widespread messaging framework in the robotics research community and has been officially supported by Mathworks’ Robotics System Toolbox since 2014. While the Simulink code generation uses ROS C++, the MATLAB implementation is built on the less common RosJava.

The API was designed such that each topic requires dedicated publishers and subscribers, which is different from LCM where each subscriber may listen to multiple channels/topics. While this may result in potentially more subscribers, the specification of the expected type at initialization removes much of the boiler plate code necessary for dealing with message types.

%% MATLAB code for publishing a ROS message
% Setup Publisher
chatpub = rospublisher('/chatter', 'std_msgs/String');

% Fill message
msg = rosmessage(chatpub);
msg.Data = 'Some test string';

% Publish
chatpub.send(msg);

Subscribers support three different styles to access messages: blocking calls, non-blocking calls, and callbacks.

%% MATLAB code for receiving a ROS message
% Setup Subscriber
laser = rossubscriber('/scan');

% (1) Blocking receive
scan = laser.receive(1); % timeout [s]

% (2) Non-blocking latest message (may not be new)
scan = laser.LatestMessage;

% (3) Callback
callback = @(msg) disp(msg);
subscriber = rossubscriber('/scan', @callback);

Contrary to LCM, all objects that are visible to users are actually MATLAB classes. Even though the implementation is using Java underneath, all exposed functionality is wrapped in MATLAB classes that hide all Java calls. For example, each message type is associated with a generated wrapper class. The code below shows a simplified example of a wrapper for a message that has a Name property.

%% MATLAB code for wrapping a Java message type
classdef WrappedMessage

    properties (Access = protected)
        % The underlying Java message object (hidden from user)
        JavaMessage
    end

    methods

        function name = get.Name(obj)
            % value = msg.Name;
            name = char(obj.JavaMessage.getName);
        end

        function set.Name(obj, name)
            % msg.Name = value;
            validateattributes(name, {'char'}, {}, 'WrappedMessage', 'Name');
            obj.JavaMessage.setName(name); % Forward to Java method
        end

        function out = doSomething(obj)
            % msg.doSomething() and doSomething(msg)
            try
                out = obj.JavaMessage.doSomething(); % Forward to Java method
            catch javaException
                throw(WrappedException(javaException)); % Hide Java exception
            end
        end

    end
end

Due to the implementation being closed-source, we were only able to look at the public toolbox files as well as the compiled Java bytecode. As far as we could tell they built a small Java library that wrapped RosJava functionality in order to provide an interface that is easier to call from MATLAB. Most of the actual logic seemed to be implemented in MATLAB code, but we also found several calls to various Java libraries for problems that would have been difficult to implement in pure MATLAB, e.g., listing networking interfaces or doing in-memory decompression of images.

Overall, we found that the ROS support toolbox looked very nice and was a great example of how seamless external languages could be integrated with MATLAB. We also really liked that they offered a way to load log files (rosbags).

One concern we had was that there did not seem to be a simple non-blocking way to check for new messages, e.g., a hasNewMessage() method or functionality equivalent to LCM’s getNextMessage(0). We often found this useful for applications that combined data from multiple topics that arrived at different rates (e.g. sensor feedback and joystick input events). We checked whether this behavior could be emulated by specifying a very small timeout in the receive method (shown in the snippet below), but any value below 0.1s seemed to never successfully return.

---

Data Distribution Service (DDS)

In 2014 Mathworks also added a support package for DDS, which is the messaging middleware that ROS 2.0 is based on. It supports MATLAB and Simulink, as
well as code generation. Unfortunately, we did not have all the requirements to get it setup, and we could not find much information about the underlying implementation. After looking at some of the intro videos, we believe that the resulting code should look as follows.

%% MATLAB code for sending and receiving DDS messages
% Setup
DDS.import('ShapeType.idl','matlab');
dp = DDS.DomainParticipant

% Create message
myTopic = ShapeType;
myTopic.x = int32(23);
myTopic.y = int32(35);

% Send Message
dp.addWriter('ShapeType', 'Square');
dp.write(myTopic);

% Receive message
dp.addReader('ShapeType', 'Square');
readTopic = dp.read();

---

ZeroMQ

ZeroMQ is another asynchonous messaging library that is popular for building distributed systems. It only handles the messaging aspect, so users need to supply their own wire format. ZeroMQ-matlab is a MATLAB interface to ZeroMQ that was developed at UPenn between 2013-2015. We were not able to find much documentation, but as far as we could tell the resulting code should look similar to following snippet.

%% MATLAB code for sending and receiving ZeroMQ data
% Setup
subscriber = zmq( 'subscribe', 'tcp', '127.0.0.1', 43210 );
publisher = zmq( 'publish', 'tcp', 43210 );

% Publish data
bytes = uint8(rand(100,1));
nbytes = zmq( 'send', publisher, bytes );

% Receive data
receiver = zmq('poll', 1000); // polls for next message
[recv_data, has_more] = zmq( 'receive', receiver );

disp(char(recv_data));

It was implemented as a single MEX function that selects appropriate sub-functions based on a string argument. State was maintained by using socket IDs that were passed in by the user at every call. The code below shows a simplified snippet of the send action.

// Parsing the selected ZeroMQ action behind the MEX barrier
// Grab command String
if ( !(command = mxArrayToString(prhs[0])) )
	mexErrMsgTxt("Could not read command string. (1st argument)");

// Match command String with desired action (e.g. 'send')
if (strcasecmp(command, "send") == 0){
	// ... (argument validation)

	// retrieve arguments
	socket_id = *( (uint8_t*)mxGetData(prhs[1]) );
	size_t n_el = mxGetNumberOfElements(prhs[2]);
	size_t el_sz = mxGetElementSize(prhs[2]);
	size_t msglen = n_el*el_sz;

	// send data
	void* msg = (void*)mxGetData(prhs[2]);
	int nbytes = zmq_send( sockets[ socket_id ], msg, msglen, 0 );

	// ... check outcome and return
}
// ... other actions

---

Other Frameworks

Below is a list of APIs to other frameworks that we looked at but could not cover in more detail.

Project	Notes
RabbitMQ-Matlab-Client	Simple Java wrapper for RabbitMQ with callbacks into MATLAB
URBI (tutorial)	Seems to be deprecated

By Eran Klein, University of Washington and Katherine Pratt, University of Washington

In the 1995 film “Batman Forever,” the Riddler used 3-D television to secretly access viewers’ most personal thoughts in his hunt for Batman’s true identity. By 2011, the metrics company Nielsen had acquired Neurofocus and had created a “consumer neuroscience” division that uses integrated conscious and unconscious data to track customer decision-making habits. What was once a nefarious scheme in a Hollywood blockbuster seems poised to become a reality.

Recent announcements by Elon Musk and Facebook about brain-computer interface (BCI) technology are just the latest headlines in an ongoing science-fiction-becomes-reality story.

BCIs use brain signals to control objects in the outside world. They’re a potentially world-changing innovation – imagine being paralyzed but able to “reach” for something with a prosthetic arm just by thinking about it. But the revolutionary technology also raises concerns. Here at the University of Washington’s Center for Sensorimotor Neural Engineering (CSNE) we and our colleagues are researching BCI technology – and a crucial part of that includes working on issues such as neuroethics and neural security. Ethicists and engineers are working together to understand and quantify risks and develop ways to protect the public now.

Picking up on P300 signals

All BCI technology relies on being able to collect information from a brain that a device can then use or act on in some way. There are numerous places from which signals can be recorded, as well as infinite ways the data can be analyzed, so there are many possibilities for how a BCI can be used.

Some BCI researchers zero in on one particular kind of regularly occurring brain signal that alerts us to important changes in our environment. Neuroscientists call these signals “event-related potentials.” In the lab, they help us identify a reaction to a stimulus.

In particular, we capitalize on one of these specific signals, called the P300. It’s a positive peak of electricity that occurs toward the back of the head about 300 milliseconds after the stimulus is shown. The P300 alerts the rest of your brain to an “oddball” that stands out from the rest of what’s around you.

For example, you don’t stop and stare at each person’s face when you’re searching for your friend at the park. Instead, if we were recording your brain signals as you scanned the crowd, there would be a detectable P300 response when you saw someone who could be your friend. The P300 carries an unconscious message alerting you to something important that deserves attention. These signals are part of a still unknown brain pathway that aids in detection and focusing attention.

Reading your mind using P300s

P300s reliably occur any time you notice something rare or disjointed, like when you find the shirt you were looking for in your closet or your car in a parking lot. Researchers can use the P300 in an experimental setting to determine what is important or relevant to you. That’s led to the creation of devices like spellers that allow paralyzed individuals to type using their thoughts, one character at a time.

It also can be used to determine what you know, in what’s called a “guilty knowledge test.” In the lab, subjects are asked to choose an item to “steal” or hide, and are then shown many images repeatedly of both unrelated and related items. For instance, subjects choose between a watch and a necklace, and are then shown typical items from a jewelry box; a P300 appears when the subject is presented with the image of the item he took.

Everyone’s P300 is unique. In order to know what they’re looking for, researchers need “training” data. These are previously obtained brain signal recordings that researchers are confident contain P300s; they’re then used to calibrate the system. Since the test measures an unconscious neural signal that you don’t even know you have, can you fool it? Maybe, if you know that you’re being probed and what the stimuli are.

Techniques like these are still considered unreliable and unproven, and thus U.S. courts have resisted admitting P300 data as evidence.

Imagine that instead of using a P300 signal to solve the mystery of a “stolen” item in the lab, someone used this technology to extract information about what month you were born or which bank you use – without your telling them. Our research group has collected data suggesting this is possible. Just using an individual’s brain activity – specifically, their P300 response – we could determine a subject’s preferences for things like favorite coffee brand or favorite sports.

But we could do it only when subject-specific training data were available. What if we could figure out someone’s preferences without previous knowledge of their brain signal patterns? Without the need for training, users could simply put on a device and go, skipping the step of loading a personal training profile or spending time in calibration. Research on trained and untrained devices is the subject of continuing experiments at the University of Washington and elsewhere.

It’s when the technology is able to “read” someone’s mind who isn’t actively cooperating that ethical issues become particularly pressing. After all, we willingly trade bits of our privacy all the time – when we open our mouths to have conversations or use GPS devices that allow companies to collect data about us. But in these cases we consent to sharing what’s in our minds. The difference with next-generation P300 technology under development is that the protection consent gives us may get bypassed altogether.

What if it’s possible to decode what you’re thinking or planning without you even knowing? Will you feel violated? Will you feel a loss of control? Privacy implications may be wide-ranging. Maybe advertisers could know your preferred brands and send you personalized ads – which may be convenient or creepy. Or maybe malicious entities could determine where you bank and your account’s PIN – which would be alarming.

With great power comes great responsibility

The potential ability to determine individuals’ preferences and personal information using their own brain signals has spawned a number of difficult but pressing questions: Should we be able to keep our neural signals private? That is, should neural security be a human right? How do we adequately protect and store all the neural data being recorded for research, and soon for leisure? How do consumers know if any protective or anonymization measures are being made with their neural data? As of now, neural data collected for commercial uses are not subject to the same legal protections covering biomedical research or health care. Should neural data be treated differently?

These are the kinds of conundrums that are best addressed by neural engineers and ethicists working together. Putting ethicists in labs alongside engineers – as we have done at the CSNE – is one way to ensure that privacy and security risks of neurotechnology, as well as other ethically important issues, are an active part of the research process instead of an afterthought. For instance, Tim Brown, an ethicist at the CSNE, is “housed” within a neural engineering research lab, allowing him to have daily conversations with researchers about ethical concerns. He’s also easily able to interact with – and, in fact, interview – research subjects about their ethical concerns about brain research.

There are important ethical and legal lessons to be drawn about technology and privacy from other areas, such as genetics and neuromarketing. But there seems to be something important and different about reading neural data. They’re more intimately connected to the mind and who we take ourselves to be. As such, ethical issues raised by BCI demand special attention.

Working on ethics while tech’s in its infancy

As we wrestle with how to address these privacy and security issues, there are two features of current P300 technology that will buy us time.

First, most commercial devices available use dry electrodes, which rely solely on skin contact to conduct electrical signals. This technology is prone to a low signal-to-noise ratio, meaning that we can extract only relatively basic forms of information from users. The brain signals we record are known to be highly variable (even for the same person) due to things like electrode movement and the constantly changing nature of brain signals themselves. Second, electrodes are not always in ideal locations to record.

All together, this inherent lack of reliability means that BCI devices are not nearly as ubiquitous today as they may be in the future. As electrode hardware and signal processing continue to improve, it will be easier to continuously use devices like these, and make it easier to extract personal information from an unknowing individual as well. The safest advice would be to not use these devices at all.

The goal should be that the ethical standards and the technology will mature together to ensure future BCI users are confident their privacy is being protected as they use these kinds of devices. It’s a rare opportunity for scientists, engineers, ethicists and eventually regulators to work together to create even better products than were originally dreamed of in science fiction.

Credit: sk.ru

In this episode, Audrow Nash and Christina Brester conduct interviews at the 2016 International Association of Science Parks and Areas of Innovation conference in Moscow, Russia. They speak with Vadim Kotenev of Rehabot and Motorica about prosthetic hands and rehabilatative devices; and Vagan Martirosyan, CEO of TryFit, a company that uses robotic sensors to help people find shoes that fit them well.

An image of one of the rehabilitative devices from Rehabot.

The robotic platform that scans your feet from TryFit.

Links

• Download mp3 (13.3 MB)
• IASP 2016 conference website
• Skolkovo’s robotics page
• Rehabot’s website
• TryFit’s website
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS

Lecturer Steffen Pfiffner of University of Weingarten in Germany is teaching ROS to 26 students at the same time at a very fast pace. His students, all of them within the Master on Computer Science of University of Weingarten, use only a web browser. They connect to a web page containing the lessons, a ROS development environment and several ROS based simulated robots. Using the browser, Pfiffner and his colleague Benjamin Stähle, are able to teach how to program with ROS quickly and to many students. This is what Robot Ignite Academy is made for.

“With Ignite Academy our students can jump right into ROS without all the hardware and software setup problems. And the best: they can do this from everywhere,” says Pfiffner.

Robot Ignite Academy provides a web service which contains the teaching material in text and video format, the simulations of several ROS based robots that the students must learn to program, and the development environment required to build ROS programs and test them on the simulated robot.

Student’s point of view

Students bring their own laptops to the class and connect to the online platform. From that moment, their laptop becomes a ROS development machine, ready to develop programs for many simulated real robots.

The Academy provides the text, the videos and the examples that the student has to follow. Then, the student creates her own ROS program and makes the robot perform a specific action. The student develops the ROS programs as if she is in a typical ROS development computer.

The main advantage is that students can use a Windows, Linux or Mac machine to learn ROS. They don’t even have to install ROS in their computers. The only prerequisite of the laptop is to have a browser. So students do not mess with all the installation problems that frustrate them (and the teachers!), especially when they are starting.

After class, students can continue with their learning at home, library or even the beach if there is a wifi available! All their code, learning material and simulations are stored online so they can access them from anywhere, anytime using any computer.

Teacher’s point of view

The advantage of using the platform is not only for the students but also for the teachers. Teachers do not have to create the material and maintain it. They do not have to prepare the simulations or work on multiple different computers. They don’t even have to prepare the exams!! (which are already provided by the platform).

So what are the teachers for?

By making use of the provided material, the teacher can concentrate on guiding the students by explaining the most confusing parts, answer questions, suggest modifications according to the level of each student, and adapt the pace to the different types of students.

This new method of teaching ROS is exploding among the Universities and High Schools that want to provide the latest and most practical teachings to their students. The method, developed by Robot Ignite Academy, combines a new way of teaching based on practice and an online learning platform. Those two points combined make the teaching of ROS a smooth experience and can potentially see the students’ knowledge base skyrocket.

As user Walace Rosa indicates in his video comment about Robot Ignite Academy:

It is a game changer [in] teaching ROS!

The method is becoming very popular in the robotics circuits too, and many teachers are using it for younger students. For example, High School Mundet in Barcelona is using it to teach ROS to 15 years old students.

Additionally, the academy provides a free online certification exam with different levels of knowledge certification. Many Universities are using this exam to certify that their students did learn the material since the exam is quite demanding.

Some examples of past events

•  1 week ROS course in Barcelona for SMART-E project team members. This is a private course given by Robot Ignite Academy at Barcelona for 15 members of the SMART-E project that need to be up to speed with ROS fast. From 8th to 12nd of May 2017
•  1 day ROS course for the Col·legi d’Enginyers de Barcelona. The 17th of May 2017.
•  3 months course for University of La Salle in Barcelona within the Master on Automatics, Domotics and Robotics. From 10th of May to 29th of June 2017.
•  1 weekend ROS course for teenagers in Bilbao, Spain. The 20th and 21st of May 2017.
•  We can also organize a special event like these for you and your team.

Helpful ROS videos

In this episode, Audrow Nash and Christina Brester conduct interviews at the 2016 International Association of Science Parks and Areas of Innovation conference in Moscow, Russia. They speak with Roman Luchin, CEO of CyberTech Labs., about a robotics development platform called Trik. Trik is intended to be an intermediate step when learning about robotics between Lego Mindstorms and programming on an embedded platform. Trik allows users to program with a graphical interface by ordering blocks. These blocks contain code in several common programming languages (python, F#, Pascal, etc.) and the code can be modified directly.

This is the second of three interviews from the conference.

Roman Luchin

Roman Luchin is Founder of TRIK and CEO of CyberTech Labs in Russia. He started as a researcher in a laboratory of theoretical cybernetics in Saint Petersburg State University, where he developed multi-agent control systems. Faced with two challenges in Russian robotics, young professionals and outdated approaches, he decided to devote himself to the improvement of STEM education through robotics and the creation of new educational tools. As a result, he developed a unified cybernetic platform for education, research, and robotics projects called TRIK. There is a continuously growing community around TRIK, including universities, researchers, prototyping labs, and creative teams working to provide everyone with the opportunity to create modern intelligent robots that can “see”, “hear” and collaborate.

Links

• Download mp3 (16.5 MB)
• IASP 2016 conference website
• Skolkovo’s robotics page
• Trik’s homepage
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS

D’Arpino Perez discusses her work teaching Optimus to pick up a bottle. Image: Jason Dorfman, MIT CSAIL

Most robots are programmed using one of two methods: learning from demonstration, in which they watch a task being done and then replicate it, or via motion-planning techniques like optimization or sampling, which require a programmer to explicitly specify a task’s goals and constraints.

Both methods have drawbacks. Robots that learn from demonstration can’t easily transfer one skill they’ve learned to another situation and remain accurate. On the other hand, motion planning systems that use sampling or optimization can adapt to these changes, but are time-consuming, since they usually have to be hand-coded by expert programmers.

Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have recently developed a system that aims to bridge the two techniques: C-LEARN, which allows non-coders to teach robots a wider range of tasks simply by providing some information about how objects are typically manipulated and then showing the robot a single demo of the task. Importantly, this enables users to teach robots skills that can be automatically transferred to other robots with different “kinematics” (ways of moving) – a key time- and cost-saving measure for companies that want a range of robots to perform similar actions.

“By combining the intuitiveness of learning from demonstration with the precision of motion-planning algorithms, this approach can help robots do new types of tasks that they haven’t been able to learn before, like multi-step assembly using both of their arms,” says Claudia Pérez-D’Arpino, a PhD student who wrote a paper on C-LEARN with MIT professor Julie Shah.

The team tested the system on Optimus, a new two-armed robot designed for bomb disposal that they programmed to perform tasks like opening doors, transporting objects and extracting objects from containers.

(The robot is made of components from multiple manufacturers, including a manipulation system by RE2, a Husky unmanned ground vehicle by Clearpath, 3-finger grippers by Robotiq and the multisense sensor from Carnegie Robotics that is used in the Atlas robot.)

In simulations they showed that Optimus’ learned skills could be seamlessly transferred to Atlas, CSAIL’s six-foot-tall, 400-pound humanoid robot.

The paper was recently accepted to the IEEE International Conference on Robotics and Automation (ICRA), which takes place May 29-June 3 in Singapore.

How it works

With C-LEARN the user first gives the robot a knowledge base of information on how to reach and grasp different objects that have different constraints. (The C in C-LEARN stands for “constraints.”) The operator then uses a 3D interface to show the robot a single demonstration of the specific task, which is represented by a sequence of relevant moments known as “keyframes.” By matching these keyframes with the knowledge base, the robot can automatically suggest motion plans for the operator to approve or edit as needed.

“This approach is actually very similar to how humans learn in terms of seeing how something’s done and connecting it to what we already know about the world,” says Pérez-D’Arpino. “We can’t magically learn from a single demonstration, so we take new information and match it to previous knowledge about our environment.”

One challenge was that existing constraints that could be learned from demonstrations weren’t accurate enough to enable robots to precisely manipulate objects. To overcome that, the researchers developed constraints inspired by computer-aided design (CAD) programs that can tell the robot if its hands should be parallel or perpendicular to the objects it is interacting with.

The team also showed that the robot performed significantly better in collaboration with humans. While the robot successfully executed tasks 87.5 percent of the time on its own, it did so 100 percent of the time when it had a operator that could correct minor sensory inaccuracies.

“Having a knowledge base is fairly common, but what’s not common is integrating it with learning from demonstration,” says Dmitry Berenson, an assistant professor of computer science at the University of Michigan who was not involved in the research. “That’s very helpful, because if you are dealing with the same objects over and over again, you don’t want to then have to start from scratch to teach the robot every new task.”

The system is part of a larger wave of research focused on making LfD approaches more adaptive. If you’re a robot that has learned to take an object out of a tube from a demonstration, you might not be able to do it if there’s an obstacle in the way that requires you to move your arm differently. However, C-LEARN can do this, because it does not learn one specific way to perform the action.

“It’s good for the field that we’re moving away from directly imitating motion, towards actually trying to infer the principles behind the motion,” says Berenson. “By using these learned constraints in a motion planner, we can make systems that are far more flexible than those which just try to mimic what’s being demonstrated”

Shah says that advanced LfD methods could prove important in time-sensitive scenarios like bomb disposal and disaster response, where robots are currently tele-operated at the level of individual joint movements.

“Something as simple as picking up a box could take 20-30 minutes, which is significant for an emergency situation,” says Pérez-D’Arpino.

C-LEARN can’t yet handle certain advanced tasks, like avoiding collisions or planning for different step sequences for a given task. But the team is hopeful that incorporating more insights from human learning will give robots an even wider range of physical capabilities.

“Traditional programming of robots in real-world scenarios is difficult, tedious and requires a lot of domain knowledge,” says Shah. “It would be much more effective if we could train them more like how we train people: by giving them some basic knowledge and a single demonstration. This is an exciting step towards teaching robots to perform complex multi-arm and multi-step tasks necessary for assembly manufacturing and ship or aircraft maintenance.”

Click here to access the full paper.

Would you like to make a robot to grasp something, but you think that is impossible to you just because you can’t buy a robot arm? I’m here to tell that you can definitely achieve this without buying a real robot. Let’s see how:

How is that possible?

The Smart Grasping Sandbox built by Shadow Robotics is now available for everybody on the ROS Development Studio – a system that allows you to create and test your robot programs through simulations using only a web browser.

Wait, what is the Smart Grasping Sandbox?

The Smart Grasping Sandbox is a public simulation for the Shadow’s Smart Grasping System with theUR10 robot from Universal Robots. It allows you to make a robot to grasp something without having to learn everything related to Machine Learning, and being available on the ROS Development Studio, it allows you to test it without the hassle of installing all the requirements.

As Ugo Cupcic said:

I don’t want to have to specify every aspect of a problem — I’d rather the system learn the best way to approach a given problem itself.

Using the Development Environment for Grasping

In order to user the Smart Grasping Sandbox on ROS Development Studio—also known as RDS—just go to http://rds.theconstructsim.com and Sign In. Once logged in, go to the Public Simulations, select the Smart Grasping Sandbox simulation and press the red Launch this simulation button.

After pressing the Launch this simulation, a new screen will appears asking you to select the launch file. Just keep the default main.launch and press the red run button.

After a few seconds, the simulation will be loaded and you will be able interact with the simulation.

Once the simulation is loaded, you can see the instructions about how to control it on the left side.

In the center you have the simulation itself, and on the right side you can see the Integrated Development Environment and the Web Shell that allows you to modify the code and send Linuxcommands to the simulation respectively.

You can modify, the simulation, the control programs or even create your own manipulation control programs, make the grasping system learn using deep learning or else. This is up to you as developer!

You can also see what the robot sees using the online ROS 3D Visualizer (RViz) typing the following command on the Web Shell:

$ rosrun rviz rviz

After running Rviz, you press the red Open ROS Graphic Tools button on the bottom left side of the system and a new tab should appear with Rviz.

Conclusion

Here we have seen how to use a simulation to programme a robot to grasp something.

The simulation uses ROS as the middleware to control the robot.

If you want to learn ROS or master your ROS Skills, we recommend you to give a try to Robot Ignite Academy.

If you are a robot builder and want to have your own simulation available for everybody on RDS, or you want to have a course specifically for your robot, contact us through the email info@theconstructsim.com.

Today we are looking at how to program your NAO Robot for Human Interaction. Watch the video and follow the steps below to get interactive with your robot pal!

1.  Create an animation of movements
   • Right click, select “Create a new box”
   • Select “Timeline”

We want the robot to do 3 things, a five high, a hello, and a goodbye move:

• On the Timeline Edit Box
  • Name: High Five
  • Image: Click “Edit”
    • On the Edit box image
    • Click “Browse” and select a file
  • Click “OK”

Now, as you can remember, if we take it to the box, we can get access to the NAO’s Timeline. Mine now is connected to the PC. So what I’m going to do is I’m going to set some points of the NAO to do some movements. As I have said before on one of my tutorial I’m going to make it set all the joints to the whole body.

We want to make him look like he is doing a high five:

• Right click the arm
  • Click “Stiffen chain on/off”
  • Raise his arm up
• Set joints and keyframe
  • Select “Arms”
• Click “Play” on the Motion Controller

That’s a quite quick high five actually. What we want to do is to actually give him a pause when he got his arm in the air. So, the way we do that is to copy and paste the keyframe.

Grab a few already built in animation, just to show you how much it works:

• Go to Motions>Animations
  • Drag a “Hello” box to the workspace
  • Drag a “Wipe Forehead” box for a goodbye move to the workspace

We want to hear him speak, so:

• Go to Voice
  • Drag a “Speech Recognition” box to the workspace
• Drag a “Switch Case”
  • Connect the “Speech Recognition” box to the “Switch Case” box
  • On the first case, write “Hello” and connect to the “Hello” box
  • On the second case, write “High Five” and connect it to the “High Five” box
  • On the third case, write “Goodbye” and connect it to the “Wipe Forehand” box
• Set parameters of Speech Recognition
  • Word list: Hi; High Five; Goodbye;
  • Play with the threshold to see what works better
  • Click “OK”

So there we have a speech recognition, he will listen to what I am saying and then we will act a motion. Now, what we need to do is to make sure that the robot doesn’t repeat it, so you know, it doesn’t loop back. Before we do that, we are going to:

• Drag 3 “Say” box to the workspace so he can  speak back to us
  • On the “Switch case”, connect the “Hello” case on the “Say” box
    • Localized Text
      • Change the language to English(United States)
      • Write “Hi” on the textbox
  • Connect the “High Five” case on the “Say (1)” box
    • Localized Text
      • Change the language to English(United States)
      • Write “High Five” on the textbox
  • Connect the “Goodbye” case on the “Say(2)” box
    • Localized Text
      • Change the language to English(United States)
      • Write “Goodbye” on the textbox

So, here’s what we are saying, he recognizes it, says a word and does an animation, now what we want to do is we don’t want him to loop, as in he keeps hearing the same sound and maybe he starts to hear himself and repeat himself, so to stop that we need to connect the end of the motion boxes back to the “Speech Recognition” box. A simple output to the input speech.

I am going to make a little change actually, instead of “Hello”, we should put “Hi” here and I’m going to put “Hi” in there which is already done because they have to match otherwise if it doesn’t match he won’t recognize it. So if he hit “Load”

Philip: High Five!

Nao: High Five!

Philip: Goodbye!

Nao: Goodbye!

Philip: Hi!

Nao: Hello!

Brilliant! So that looks likes he’s working very well, so again if we says “Hi”, “High Five” or “Good bye” then he will greet you. If you walk in to room and you say “Hi” to a robot, obviously, he will do his animation and say a speech. Here you can use your imagination and try different things, different animations.

If you liked this article, you may also enjoy these from Robo-Phil:

• Python programming your NAO robot
• Hard at work: A review of the Laevo Exoskeleton

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

Imagine how easy it would be to learn skating, if only it doesn’t hurt everytime you fall. Unfortunately, we, humans,  don’t have that option. Robots, however, can now “learn” their skills on a simulation platform without being afraid of crashing into a wall. Yes, “it learns“! This is possible with the reinforcement learning algorithms provided by OpenAI Gym and the ROS Development Studio.

You can now train your robot to navigate through an environment filled with obstacles just based on the sensor inputs, with the help of OpenAI Gym. In April 2016, OpenAI introduced “Gym”, a platform for developing and comparing reinforcement learning algorithms. Reinforcement learning is an area of machine learning that allows an intelligent agent (for example, robot) to learn the best behaviors in an environment by trial-and-error. The agent takes actions in an environment so as to maximize its rewards. We have deployed the gym_gazebo package from Erle Robotics. in the ROS Development Studio. It enables the users to test their reinforcement learning for their robots in Gazebo.

How to Train your Robot

Robot training to navigate through an environment with obstacles

In this example we will be seeing how a turtlebot is able to learn navigating through an environment without hitting an obstacle. The turtlebot will use a reinforcement learning method known as Q-learning.

There are four environments already available for the user to test their simulations with. These environments can be launched using the respective launch files:

• GazeboCircuitTurtlebotLidar_v0.launch
• GazeboCircuit2TurtlebotLidar_v0.launch
• GazeboRoundTurtlebotLidar_v0.launch
• GazeboMazeTurtlebotLidar_v0.launch

The images of the different environments are given below:

The various environments already available:

Circuit	Circuit2
Round	Maze

The user is requested to try out the existing environment before developing their own environments for training the robot. An environment is where the robot’s possible actions and rewards are defined. For example, in the available environments, there are three possible actions for the Turtlebot robot:

– Forward ( with a reward of 5 points)
– Left ( with a reward of 1 point)
– Right ( with a reward of 1 point)

If it collides into the walls, then the training episode ends (with a penalty of 200 points). The turtlebot has to learn to navigate through the environment, based on the rewards obtained from different episodes. This can be achieved using the Q-learning algorithm. Let us see, how it can be done using the ROS Development Studio.

Using the ROS Development Studio for training the robot

First, we have to set the path in the jupyter-notebook as given below:

import sys
sys.path.append("/usr/local/lib/python2.7/dist-packages/")
sys.path.append("/home/ubuntu/gym-gazebo")
sys.path.append("/home/user/catkin_ws/src/gym_construct/src")
%matplotlib inline


The python scripts in the folder gym_construct/src/ help us simulate the reinforcement learning techniques for a Turtlebot. Currently, the number of episodes has been set to 20.
Feel free to increase the number of episodes in the python scripts (usually up to 5000) to actually train the robot to navigate the environment completely.Run the  only the script corresponding to the environment:

Run the python corresponding to the environment:

## Circuit-2 Environment --> Q-learning
%run /home/user/catkin_ws/src/gym_construct/src/circuit2_turtlebot_lidar_qlearn.py


The robot undergoes several training episodes and each of these episodes are rewarded based on the number of steps taken by the robot before hitting the environment. We will be able to see that the robot incrementally increases its rewards over time compared to its previous versions. With a very large number of episodes, the robot will learn to navigate the environment without hitting the obstacle.

Plotting the learning curve of the robot

The machine learning algorithm generates the output files in the output directory specified in the python script file. In order to plot the curve, we run the display_plot.py. But before that, don’t forget to restart the kernel and set the path once again.

As the number of episodes increases, we will see that the robot’s mean rewards also increases. The user can choose his own robot, environments, action and rewards for testing his reinforcement learning algorithms in OpenAI Gym and RDS. Watch the video below for more on this.

Video describing the procedure to run OpenAI Gym on RDS

I hope you were able to follow the tutorial.  So, that’s all folks. It’s now up to you to develop and test reinforcement learning algorithms in OpenAI Gym and RDS.

Have fun training your robot!

A few decades ago, touchscreens were impressive, yet clunky pieces of technology reserved for applications that did little more than show off that touchscreens were possible. Today’s touchscreens are commonplace and readily accepted as an easy way to interact with an operating system; they’re accurate, support multi-touch, are comfortable to use, and can even operate in multiple dimensions (pressing hard).

We may not have perfected touchscreens yet, but we’re getting close. Accordingly, engineers and researchers are already speculating about the next phase of UI development, especially for robotics control. So far, the leading candidate is gesture-based control—the use of physical gestures to relay commands.

The idea

The major limitation for touchscreens is the fact that they operate only in two dimensions; the third dimension introduced with force touch is extremely limited. Comparatively, hand gestures and physical movements can operate in three dimensions, and depending on how they’re designed, could feel more natural than swiping and tapping a smartphone screen.

Demand for three-dimensional gestural control is increasing with the onset of virtual reality (VR) and augmented reality (AR) technology; because the digital world we experience will be moving from two dimensions to three dimensions, the controls we use to manage those experiences will also need to change. With enough sophistication, these systems could provide better feedback to the users in control; rather than merely responding with visual feedback like movement or lights, users could feasibly be immersed with physical feedback like vibration or resistance.

Where we stand

Currently, one of the most advanced gestural systems is the Real-Time 3D Gesture Analysis for Natural Interaction with Smart Devices, a project headed by researchers in Linnaeus University in Sweden. However, simpler forms of gesture-based control are already available.

For example, there are projects that use Arduino to create a robot that can respond to four different simple hand gestures (plus a neutral position). Of course, iPhone technology also makes use of some simple “gestures,” such as shaking the phone to undo typing or rotating the phone to experience an immersive 360-degree view of an image or video.

The main obstacles

There are a few obstacles preventing gestures from being solidified as the next landmark in robotics control, however:

• Gesture complexity. The biggest obstacle described by researchers at Linnaeus University is the sheer complexity of physical gestures; to be effective, their recognition systems need to be able to gather and recognize thousands of tiny data points and interpret complex patterns to “understand” what movement is being attempted. Moreover, this big data-based interpretation needs to happen in real-time—especially for applications like live robotics control, or engagement in a VR-based video game. That demands not only an incredibly intelligent system, but a processor that can operate quickly.
• Accessibility. Gesture recognition systems would likely be developed to accommodate a “standard” human model, such as a “standard” right human hand. How would the system accommodate somebody whose hand was missing, or who is missing a few fingers? What about left-handed people? And people with Parkinson’s disease, or who are unable to operate their hands with precision control?
• Applications. For gestures to be accepted as a mainstream way to interact with robotics and virtual environments, robotics and virtual environments need to be available. Virtual reality (VR) technology has been advancing strongly for the past several years and is poised to take a big step forward in user adoption by 2020, but until those user adoption numbers are reliable, the demand for gestural systems is relatively low. Using hand movements and other physical gestures to control a two-dimensional screen, for example, would be largely ineffective.

On paper, gestures seem like the best form of control for the digital devices, robotics, and VR systems of the future. However, there are still many obstacles to overcome before we’re ready for large-scale adoption. Fortunately, researchers are ahead of the curve, already preparing the intelligent gesture-based recognition programs we’ll need when touchscreens become obsolete.

Northwestern University mechanical engineering professor Todd Murphey and his team are engineering robots that one might say could make robotic assistance as seamless as “humanly” possible. With support from the National Science Foundation (NSF), the team is using novel algorithmic tools, such as a drawing robot, to develop the algorithms, or rules of behavior, that would greatly enhance a robot’s ability to adapt to human unpredictability.

Murphey points out that in order for robots to help people, they have to have at least a basic understanding of the types of tasks people can do. Some tasks, like lifting and placing an object, are close to the types of tasks that robots already do. Other tasks, like drawing, are harder for robots, partly because there are so many ways to get the same image. As Murphey explains: “And so, drawing is a type of task that’s maybe not the same as that sort of precision manufacturing task that we’ve seen robots do historically.”

Murphey sees many possibilities for robots that work alongside humans in everyday tasks, but one application his lab is focused on currently is physical therapy. The goal is to develop robots that can help patients not only move through the paces of their physical therapy without hurting themselves but allow them to complete movement tasks by creating mechanical environments to make this feasible. “Algorithms developed here will eventually run on physical therapy robots designed to help people with tasks like balance and feeding themselves,” says Murphey.

His collaborator, Julius P. Dewald, runs the Northwestern School of Medicine’s Physical Therapy and Human Movement Sciences Department. Dewald has pioneered the use of robotics in stroke rehabilitation and sees great promise in Murphey’s new approaches.

In this episode, Audrow Nash interviews Bradley Knox, founder of bots_alive. Knox speaks about an add-on to a Hexbug, a six-legged robotic toy, that makes the bot behave more like a character. They discuss the novel way Knox uses machine learning to create a sense character. They also discuss the limitation of technology to emulate living creatures, and how the bots_alive robot was built within these limitations.

Brad Knox

Dr. Bradley Knox is the founder of bots_alive. He researched human-robot interaction, interactive machine learning, and artificial intelligence at the MIT Media Lab and at UT Austin. At MIT, he designed and taught Interactive Machine Learning. He has won two best paper awards at major robotics and AI conferences, was awarded best dissertation from UT Austin’s Computer Science Department, and was named to IEEE’s AI’s 10 to Watch in 2013.

Links

• bots_alive Kickstarter page
• Download mp3 (16.8 MB)
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS

If you take humans out of the driving seat, could traffic jams, accidents and high fuel bills become a thing of the past? As cars become more automated and connected, attention is turning to how to best choreograph the interaction between the tens or hundreds of automated vehicles that will one day share the same segment of Europe’s road network.

It is one of the most keenly studied fields in transport – how to make sure that automated cars get to their destinations safely and efficiently. But the prospect of having a multitude of vehicles taking decisions while interacting on Europe’s roads is leading researchers to design new traffic management systems suitable for an era of connected transport.

The idea is to ensure that traffic flows as smoothly and efficiently as possible, potentially avoiding the jams and delays caused by human behaviour.

‘Travelling distances and time gaps between vehicles are crucial,’ said Professor Markos Papageorgiou, head of the Dynamic Systems & Simulation Laboratory at the Technical University of Crete, Greece. ‘It is also important to consider things such as how vehicles decide which lane to drive in.’

Prof. Papageorgiou’s TRAMAN21 project, funded by the EU’s European Research Council, is studying ways to manage the behaviour of individual vehicles, as well as highway control systems.

For example, the researchers have been looking at how adaptive cruise control (ACC) could improve traffic flows. ACC is a ‘smart’ system that speeds up and slows down a car as necessary to keep up with the one in front. Highway control systems using ACC to adjust time gaps between cars could help to reduce congestion.

‘It may be possible to have a traffic control system that looks at the traffic situation and recommends or even orders ACC cars to adopt a shorter time gap from the car in front,’ Prof. Papageorgiou said.

‘So during a peak period, or if you are near a bottleneck, the system could work out a gap that helps you avoid the congestion and gives higher flow and higher capacity at the time and place where this is needed.’

Variable speed limits

TRAMAN21, which runs to 2018, has been running tests on a highway near Melbourne, Australia, and is currently using variable speed limits to actively intervene in traffic to improve flows.

An active traffic management system of this kind could even help when only relatively few vehicles on the highway have sophisticated automation. But he believes that self-driving vehicle systems must be robust enough to be able to communicate with each other even when there are no overall traffic control systems.

‘Schools of fish and flocks of birds do not have central controls, and the individuals base their movement on the information from their own senses and the behaviour of their neighbours,’ Prof. Papageorgiou said.

‘In theory this could also work in traffic flow, but there is a lot of work to be done if this is to be perfected. Nature has had a long head-start.’

One way of managing traffic flow is platooning – a way to schedule trucks to meet up and drive in convoy on the highway. Magnus Adolfson from Swedish truckmaker Scania AB, who coordinated the EU-funded COMPANION project, says that platooning – which has already been demonstrated on Europe’s roads – can also reduce fuel costs and accidents.

The three-year project tested different combinations of distances between trucks, speeds and unexpected disruptions or stoppages.

Fuel savings

In tests with three-vehicle platoons, researchers achieved fuel savings of 5 %. And by keeping radio contact with each other, the trucks can also reduce the risk of accidents.

‘About 90 percent of road accidents are caused by driver error, and this system, particularly by taking speed out of the driver’s control, can make it safer than driving with an actual driver,’ Adolfson said.

The COMPANION project also showed the benefits of close communication between vehicles to reduce the likelihood of braking too hard and causing traffic jams further back.

‘There is enough evidence to show that using such a system can have a noticeable impact, so it would be good to get it into production as soon as possible,’ Adolfson said. The researchers have extended their collaboration to working with the Swedish authorities on possible implementation.

Rutger Beekelaar, a project manager at Dutch-based research organisation TNO, says that researchers need to demonstrate how automated cars can work safely together in order to increase their popularity.

‘Collaboration is essential to ensure vehicles can work together,’ he said. ‘We believe that in the near future, there will be more and more automation in traffic, in cars and trucks. But automated driving is not widely accepted yet.’

To tackle this, Beekelaar led a group of researchers in the EU-funded i-GAME project, which developed technology that uses wireless communication that contributes to managing and controlling automated vehicles.

They demonstrated these systems in highway conditions in the 2016 Grand Cooperative Driving Challenge in Helmond, in the Netherlands, which put groups of real vehicles through their paces to demonstrate cooperation, how they safely negotiated an intersection crossing, and merged with another column of traffic.

Beekelaar says that their technology is now being used in other European research projects, but that researchers, auto manufacturers, policymakers, and road authorities still need to work together to develop protocols, systems and standardisation, along with extra efforts to address cyber security, ethics and particularly the issue of public acceptance.

You probably know the Sphero robot. It is a small robot with the shape of a ball. In case that you have one, you must know that it is possible to control it using ROS, by installing in your computer the Sphero ROS packages developed by Melonee Wise and connecting to the robot using the bluetooth of the computer.

Now, you can use the ROS Development Studio to create ROS control programs for that robot, testing as you go by using the integrated simulation.

The ROS Development Studio (RDS) provides off-the-shelf a simulation of Sphero with a maze environment. The simulation provides the same interface as the ROS module created by Melonee, so you can test your develop and test the programs on the environment, and once working properly, transfer it to the real robot.

We created the simulation to teach ROS to the students of the Robot Ignite Academy. They have to learn ROS enough to make the Sphero get out of the maze by using odometry and IMU.

Using the simulation

To use the Sphero simulation on RDS go to rds.theconstructsim.com and sign in. If you select the Public simulations, you will quickly identify the Sphero simulation.

Press the red Play button. A new screen will appear giving you details about the simulation and asking you which launch file you want to launch. The main.launch selected by default is the correct one, so just press Run.

After a few seconds the simulation will appear together with the development environment for creating the programs for Sphero and testing them.

On the left hand side you have a notebook containing information about the robot and how to program it with ROS. This notebook contains just some examples, but it can be completed and/or modified at your will. As you can see it is an iPython notebook and follows its standard. So it is up to you to modify it, add new information or else. Remember that any change you do to the notebook will be saved in a simulation in your private area of RDS, so you can come back later and launch it with your modifications.

You must know that the code included in the notebook is directly executable by selecting the cell of the code (do a single click on it) and pressing the small play button at the top of the notebook. This means that, once you press that button, the code will be executed and start controlling the Sphero simulated robot for a few time-steps (remember to have the simulation activated (Play button of the simulation activated) to see the robot move).

On the center area, you can see the IDE. It is the development environment for developing the code. You can browse there all the packages related to the simulation or any other packages that you may create.

On the right hand side, you can see the simulation and beneath it, the shell. The simulation shows the Sphero robot as well as the environment of the maze. On the shell, you can issue commands in the computer that contains the simulation of the robot. For instance, you can use the shell to launch the keyboard controller and move the Sphero around. Try typing the following:

• $ roslaunch sphero_gazebo keyboard_teleop.launch

Now you must be able to move the robot around the maze by pressing some keys of the keyboard (instructions provided on the screen).

You can also launch there Rviz, and then watch the robot, the frames and any other additional information you may want of the robot. Type the following:

• $ rosrun rviz rviz

Then press the Screen red icon located at the bottom-left of the screen (named the graphical tools). A new tab should appear, showing how the Rviz is loading. After a while, you can configure the Rviz to show the information you desire.

There are many ways you can configure the screen to provide more focus to what interests you the most.

To end this post, I would like to indicate that you can download the simulation to your computer at any time, by doing right-click on the directories and selecting Download. You can also clone the The Construct simulations repository to download it (among other simulations available).

If you liked this tutorial, you may also enjoy these:

• UgCS photogrammetry technique for UAV land surveying missions
• Programming for robotics: Introduction to ROS
• Python programming your NAO robot
• Robotics, maths, python: A fledgling computer scientist’s guide to inverse kinematics
• 3-D printing hydraulically-powered robots, no assembly required

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

Level 3 automation, where the car handles all aspects of driving with the driver on standby, is being tested in Sweden. Image courtesy of Volvo cars

Today’s self-driving car isn’t exactly autonomous – the driver has to be able to take over in a pinch, and therein lies the roadblock researchers are trying to overcome. Automated cars are hurtling towards us at breakneck speed, with all-electric Teslas already running limited autopilot systems on roads worldwide and Google trialling its own autonomous pod cars.

However, before we can reply to emails while being driven to work, we have to have a foolproof way to determine when drivers can safely take control and when it should be left to the car.

‘Even in a limited number of tests, we have found that humans are not always performing as required,’ explained Dr Riender Happee, from Delft University of Technology in the Netherlands, who is coordinating the EU-funded HFAuto project to examine the problem and potential solutions.

‘We are close to concluding that the technology always has to be ready to resolve the situation if the driver doesn’t take back control.’

But in these car-to-human transitions, how can a computer decide whether it should hand back control?

‘Eye tracking can indicate driver state and attention,’ said Dr Happee. ‘We’re still to prove the practical usability, but if the car detects the driver is not in an adequate state, the car can stop in the safety lane instead of giving back control.’

Next level

It’s all a question of the level of automation. According to the scale of US-based standards organisation SAE International, Level 1 automation already exists in the form of automated braking and self-parking.

Level 4 & 5 automation, where you punch in the destination and sit back for a nap, is still on the horizon.

But we’ll soon reach Level 3 automation, where drivers can hand over control in situations like motorway driving and let their attention wander, as long as they can safely intervene when the car asks them to.

HFAuto’s 13 PhD students have been researching this human-machine transition challenge since 2013.

Backed with Marie Skłodowska-Curie action funding, the students have travelled Europe for secondments, to examine carmakers’ latest prototypes, and to carry out simulator and on-road tests of transition takeovers.

Alongside further trials of their transition interface, HFAuto partner Volvo has already started testing 100 highly automated Level 3 cars on Swedish public roads.

Another European research group is approaching the problem with a self-driving system that uses external sensors together with cameras inside the cab to monitor the driver’s attentiveness and actions.

Blink

‘Looking at what’s happening in the scene outside of the cars is nothing without the perspective of what’s happening inside the car,’ explained Dr Oihana Otaegui, head of the Vicomtech-IK4 applied research centre in San Sebastián, Spain.

She coordinates the work as part of the EU-funded VI-DAS project. The idea is to avoid high-risk transitions by monitoring factors like a driver’s gaze, blinking frequency and head pose — and combining this with real-time on-road factors to calculate how much time a driver needs to take the wheel.

Its self-driving system uses external cameras as affordable sensors, collecting data for the underlying artificial intelligence system, which tries to understand road situations like a human would.

VI-DAS is also studying real accidents to discern challenging situations where humans fail and using this to help train the system to detect and avoid such situations.

The group aims to have its first interface prototype working by September, with iterated prototypes appearing at the end of 2018 and 2019.

Dr Otaegui says the system could have potential security sector uses given its focus on creating artificial intelligence perception in any given environment, and hopes it could lead to fully automated driving.

‘It could even go down the path of Levels 4 and 5, depending on how well we can teach our system to react — and it will indeed be improving all the time we are working on this automation.’

The question of transitions is so important because it has an impact on liability – who is responsible in the case of an accident.

It’s clear that Level 2 drivers can be held liable if they cause a fender bender, while carmakers will take the rap once Level 4 is deployed. However, with Level 3 transitions, liability remains a burning question.

HFAuto’s Dr Happee believes the solution lies in specialist insurance options that will emerge.

‘Insurance solutions are expected (to emerge) where a car can be bought with risk insurance covering your own errors, and those which can be blamed on carmakers,’ he said.

Yet it goes further than that. Should a car choose to hit pedestrians in the road, or swerve into the path of an oncoming lorry, killing its occupants?

‘One thing coming out of our discussions is that no one would buy a car which will sacrifice its owner for the lives of others,’ said Dr Happee. ‘So it comes down to making these as safe as possible.’

The five levels of automation:

1. Driver Assistance: the car can either steer or regulate speed on its own.
2. Partial Automation: the vehicle can handle both steering and speed selection on its own in specific controlled situations, such as on a motorway.
3. Conditional Automation: the vehicle can be instructed to handle all aspects of driving, but the driver needs to be on standby to intervene if needed.
4. High Automation: the vehicle can be instructed to handle all aspects of driving, even if the driver is not available to intervene.
5. Level 5 – Full Automation: the vehicle handles all aspects of driving, all the time.

This handy video-tutorial course gives an introduction to the Robot Operating System (ROS), including many of the available tools that are commonly used in robotics. With the help of different examples, the tutorials offer a great starting point to learn programming robots. You will learn how to create software including simulation, to interface sensors and actuators, and to integrate control algorithms.

The course consists of a guided tutorial and exercises with increasing level of difficulty working with an autonomous robot. We provide recordings of the lectures and give an introduction to the exercises. From the course website, you can download all the material including exercise sheets and templates, and use the provided Virtual Machine (VM) image to start programming right away.

Objectives:

• ROS architecture: Master nodes, topics, messages, services, parameters and actions
• Console commands: Navigating and analyzing the ROS system and the catkin workspace
• Creating ROS packages: Structure, launch-files, and best practices
• ROS C++ client library (roscpp): Creating your own ROS C++ programs
• Simulating with ROS: Gazebo simulator, robot models (URDF) and simulation environments (SDF)
• Working with visualizations (RViz) and user interface tools (rqt)
• Inside ROS: TF transformation system, time, bags

Course 1:

Course 2:

Course 3:

Course 4:

Download the full presentation here.

P. Fankhauser, D. Jud, M. Wermelinger, M. Hutter, “Programming for Robotics – Introduction to ROS”, ETH Zurich, 2017. DOI: 10.13140/RG.2.2.14140.44161

If you liked this tutorial, you may also enjoy these:

• Python programming your NAO robot
• Robotics, maths, python: A fledgling computer scientist’s guide to inverse kinematics
• Sticky business: Five adhesives tested for 3D printing
• Adjust how you learn and quickly pick up robotics programming
• 3-D printing hydraulically-powered robots, no assembly required

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

The feedback system enables human operators to correct the robot’s choice in real-time – Jason Dorfman, MIT CSAIL

For robots to do what we want, they need to understand us. Too often, this means having to meet them halfway: teaching them the intricacies of human language, for example, or giving them explicit commands for very specific tasks. But what if we could develop robots that were a more natural extension of us and that could actually do whatever we are thinking?

A team from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and Boston University is working on this problem, creating a feedback system that lets people correct robot mistakes instantly with nothing more than their brains.

Using data from an electroencephalography (EEG) monitor that records brain activity, the system can detect if a person notices an error as a robot performs an object-sorting task. The team’s novel machine-learning algorithms enable the system to classify brain waves in the space of 10 to 30 milliseconds.

While the system currently handles relatively simple binary-choice activities, the paper’s senior author says that the work suggests that we could one day control robots in much more intuitive ways.

“Imagine being able to instantaneously tell a robot to do a certain action, without needing to type a command, push a button or even say a word,” says CSAIL director Daniela Rus. “A streamlined approach like that would improve our abilities to supervise factory robots, driverless cars and other technologies we haven’t even invented yet.”

In the current study the team used a humanoid robot named “Baxter” from Rethink Robotics, the company led by former CSAIL director and iRobot co-founder Rodney Brooks.

The paper presenting the work was written by BU PhD candidate Andres F. Salazar-Gomez, CSAIL PhD candidate Joseph DelPreto, and CSAIL research scientist Stephanie Gil under the supervision of Rus and BU professor Frank H. Guenther.  The paper was recently accepted to the IEEE International Conference on Robotics and Automation (ICRA) taking place in Singapore this May.

Past work in EEG-controlled robotics has required training humans to “think” in a prescribed way that computers can recognize. For example, an operator might have to look at one of two bright light displays, each of which corresponds to a different task for the robot to execute.

The downside to this method is that the training process and the act of modulating one’s thoughts can be taxing, particularly for people who supervise tasks in navigation or construction that require intense concentration.

Rus’ team wanted to make the experience more natural. To do that, they focused on brain signals called “error-related potentials” (ErrPs), which are generated whenever our brains notice a mistake. As the robot indicates which choice it plans to make, the system uses ErrPs to determine if the human agrees with the decision.

The team uses EEG brain signals to detect if the person notices a mistake – Jason Dorfman, MIT CSAIL

“As you watch the robot, all you have to do is mentally agree or disagree with what it is doing,” says Rus. “You don’t have to train yourself to think in a certain way –  the machine adapts to you, and not the other way around.”

ErrP signals are extremely faint, which means that the system has to be fine-tuned enough to both classify the signal and incorporate it into the feedback loop for the human operator.

In addition to monitoring the initial ErrPs, the team also sought to detect “secondary errors” that occur when the system doesn’t notice the human’s original correction.

“If the robot’s not sure about its decision, it can trigger a human response to get a more accurate answer,” says Gil. “These signals can dramatically improve accuracy, creating a continuous dialogue between human and robot in communicating their choices.”

The feedback system enables human operators to correct the robot’s choice in real-time – Jason Dorfman, MIT CSAIL2

While the system cannot yet recognize secondary errors in real time, Gil expects the model to be able to improve to upwards of 90 percent accuracy once it can.

In addition, since ErrP signals have been shown to be proportional to how egregious the robot’s mistake is, the team believes that future systems could extend to more complex multiple-choice tasks.

Salazar-Gomez notes that the system could even be useful for people who can’t communicate verbally: a task like spelling could be accomplished via a series of several discrete binary choices, which he likens to an advanced form of the blinking that allowed stroke victim Jean-Dominique Bauby to write his memoir “The Diving Bell and the Butterfly.”

“This work brings us closer to developing effective tools for brain-controlled robots and prostheses,” says Wolfram Burgard, a professor of computer science at the University of Freiburg who was not involved in the research. “Given how difficult it can be to translate human language into a meaningful signal for robots, work in this area could have a truly profound impact on the future of human-robot collaboration.”

Read the paper here.

The project was funded in part by Boeing and the National Science Foundation.

If you enjoyed this article from CSAIL, you might also be interested in:

• MIT CSAIL’s 6-foot-tall NASA humanoid robot has landed
• Researchers add a splash of human intuition to planning algorithms
• Wearable AI that can detect the tone of a conversation
• Ingestible robots, glasses-free 3-D, and computers that explain themselves
• Design, simulate and build a custom drone

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

MIT Professor Regina Barzilay has struck up new research collaborations, drawn in MIT students, launched projects with local doctors, and begun empowering cancer treatment with the machine-learning insight that has already transformed many areas of modern life.
Photo: Lillie Paquette/School of Engineering

Computer scientist Regina Barzilay is working with MIT students and medical doctors in an ambitious bid to revolutionize cancer care. She is relying on a tool largely unrecognized in the oncology world but deeply familiar to hers: machine learning.

Barzilay, the Delta Electronics Professor of Electrical Engineering and Computer Science, was diagnosed with breast cancer in 2014. She soon learned that good data about the disease is hard to find. “You are desperate for information — for data,” she says now. “Should I use this drug or that? Is that treatment best? What are the odds of recurrence? Without reliable empirical evidence, your treatment choices become your own best guesses.”

Across different areas of cancer care — be it diagnosis, treatment, or prevention — the data protocol is similar. Doctors start the process by mapping patient information into structured data by hand, and then run basic statistical analyses to identify correlations. The approach is primitive compared with what is possible in computer science today, Barzilay says.

These kinds of delays and lapses (which are not limited to cancer treatment), can really hamper scientific advances, Barzilay says. For example, 1.7 million people are diagnosed with cancer in the U.S. every year, but only about 3 percent enroll in clinical trials, according to the American Society of Clinical Oncology. Current research practice relies exclusively on data drawn from this tiny fraction of patients. “We need treatment insights from the other 97 percent receiving cancer care,” she says.

To be clear: Barzilay isn’t looking to up-end the way current clinical research is conducted. She just believes that doctors and biologists — and patients — could benefit if she and other data scientists lent them a helping hand. Innovation is needed and the tools are there to be used.

Barzilay has struck up new research collaborations, drawn in MIT students, launched projects with doctors at Massachusetts General Hospital, and begun empowering cancer treatment with the machine learning insight that has already transformed so many areas of modern life.

Machine learning, real people

At the MIT Stata Center, Barzilay, a lively presence, interrupts herself mid-sentence, leaps up from her office couch, and runs off to check on her students.

She returns with a laugh. An undergraduate group is assisting Barzilay with a federal grant application, and they’re down to the wire on the submission deadline. The funds, she says, would enable her to pay the students for their time. Like Barzilay, they are doing much of this research for free, because they believe in its power to do good. “In all my years at MIT I have never seen students get so excited about the research and volunteer so much of their time,” Barzilay says.

At the center of Barzilay’s project is machine learning, or algorithms that learn from data and find insights without being explicitly programmed where to look for them. This tool, just like the ones Amazon, Netflix, and other sites use to track and predict your preferences as a consumer, can make short work of gaining insight into massive quantities of data.

Applying it to patient data can offer tremendous assistance to people who, as Barzilay knows well, really need the help. Today, she says, a woman cannot retrieve answers to simple questions such as: What was the disease progression for women in my age range with the same tumor characteristics?

What a machine can see

Working closely with collaborators Taghian Alphonse, chief of breast radiation oncology at Massachusetts General Hospital (MGH); Kevin Hughes, co-director of the Avon Comprehensive Breast Evaluation Center at MGH; and Constance Lehman, the chief of the breast imaging division at MGH, Barzilay intends to bring data science into clinical research nationwide. But first, she’s content with connecting her world with theirs.

Barzilay’s work in natural language processing (NLP) enables machines to search, summarize, and interpret textual documents, such as those about cancer patients in pathology reports. Using NLP tools, she and her students extracted clinical information from 108,000 reports provided by area hospitals. The database they’ve created has an accuracy rate of 98 percent. Next she wants to incorporate treatment outcomes into it.

For another study, Barzilay has developed a database that Hughes and his team can use to monitor the development of atypias, which help identify which patients are at risk of developing cancer later in life.

Machines are good at making predictions — “Why not throw all the information you have about a breast cancer patient into a model?” she says — but Barzilay is wary of having the recommendations arrive as highly complex, computational recommendations without explanation. Jointly with Tommi Jaakkola, a professor of electrical engineering and computer science at MIT, and graduate student Tao Lei, she is also developing interpretable neural models that can justify and explain the machine-based predictive reasoning.

Barzilay is also looking at how new tools can help do preventive work. Mammograms contain lots of information that may be hard for a human eye to decipher. Machines can detect subtle changes and are more capable of detecting low-level patterns. Jointly with Lehman and graduate student Nicolas Locascio, Barzilay is applying deep learning for automating analysis of mammogram data. As the first step, they are aiming to compute density and other scores currently derived by radiologists who manually analyze these images. Their ultimate goal is to identify patients who are likely to develop a tumor before it’s even visible on a mammogram, and also to predict which patients are heading toward recurrence after their initial treatment.

Ultimate success, Barzilay says, will involve drawing on computer science in unexpected ways, and pushing it in a variety of new health-related directions.

Outside her door, several of Barzilay’s students are talking ideas, hunching over laptops, and drinking coffee. An object set against the back wall resembles an odd coatrack. Guided by an idea from Taghian, six undergraduate students, led by graduate student Julian Straub, built a device that uses machine-learning to detect lymphedema, a swelling of the extremities that can be caused by the removal of or damage to lymph nodes as part of cancer treatment. It can be disabling and incurable unless detected early. Because of their high cost, these machines — lymphometers — are rare in the U.S.; very few hospitals have them.

Students have created an affordable version. And they hope to start testing this device at MGH in a couple of months. “These students are doing amazing work,” says Barzilay. “These innovations will make a really big difference. It is an entry point. There is so much to do. We are just getting started.”

If you liked this article, you may also be interested in:

• What is machine learning?
• Brian Cox presents Science Matters: Machine learning and artificial intelligence
• 5 global problems that AI could help us solve
• Open Bionics partners with NHS for feasibility study to develop hands for amputees

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

Tutorial 1: Speech

Helpful hints:

Follow these steps to create a box that we could put our python code into, and have our robot speak: “Hello”. “Hello I am a robot”.

1. Right click the main section box and go to “add new box”
2. Create a name for the box
3. Input main image
4. Select the type of box

Now we’re having a look at the script. Go into the scripts and click ok. And there we have our NAO little box. If we click into it, we have the python code. What we are going to do is change some of the codes to get our robot to speak.

1. Double click the box
2. Select the piece of code you need to alter to get our robot to speak(“def onInput_onStart(self)”)
3. Take out the pass section
4. Type in “ttsproxy = ALProxy(“ALTextToSpeech”)”. (Make sure you get the comment in there).
5. Underneath that, we are going to put “ttsproxy.say” and add the comment “Hello, I am a robot”.

The first line could create an object that gives us access to the robot’s text to speech capabilities. This object is assigned to the variable ttsProxy. We can access the object later through the name ttsProxy. I am going to mount the top line, and the alspeech is there. The second line calls a different function site. This function takes an argument “Hello, I am a robot” and then he uses a string or sequence of characters between two double quoted marks. The robot will speak the string that will pass to the say message aloud.  There are your simple first steps into python and choreograph.

You can now, hopefully, get your NAO robot to talk by using your python programming skills. Now have a play with it’s behavior to see if you can get your NAO to say a longer sentence, or maybe even sing a song. Have a play, and use your imagination!

Tutorial 2: Walking

Helpful hints:

No we’re going get your robot to walk using Python. The NAO can be program to walk to any point using the in-build Python Programming Language instead of using the drag and drop boxes in the choregraphe. Python is more expressive and allows you to do things like computing trigonometric calculations on the robot.

1. Right click the main section box and go to “add new box”
2. Create a name for the box
3. Input main image
4. Select the type of box

First, we want our robot to stand up. Drag a stand up icon into the box. I am actually using the NAO key so the 3D robot in the robot view, and this because we need to put the robot on connecting so we can see what’s going on quite clearly. Oversee when you are using this on the real robot and never forget to trigger on the motors and connect everything up properly.

Open the Code Box

1. Go to “def onInput_onStart(self):”
2. Write “motionProxy = ALProxy(“ALMotion”)”
3. On the next line write “walkTo(0.2,0.0,0.0)”
4. End it with “self.onStopped()”

The first line is a proxy called ALmotion, which what is we created here, which allows us to call the motion functions up. The second line is the walkTo, which moves the robot to a specific distance. Now, look on the decimal places: see that .2 is X and Y and the radius, so he’s going to move .2 forward. Then if I start playing with the other numbers here, he’ll move to the Y radius and then to the other turning point as well. The final bit is the self.on. To stop this, we will call and tell the choregraphe to stop the procedure when it is at the end and just simple loop and loop and loop.

If we are good, we should see this now on the 3D NAO. So if I hit play, we can see it walking along. Well done guys, you can now get your NAO Robot to walk by using
your Python programming skills. Tread carefully!

If you liked this article, you may also enjoy these:

• Robotics, maths, python: A fledgling computer scientist’s guide to inverse kinematics
• Sticky business: Five adhesives tested for 3D printing
• Meet RoboGabby: She’s on a mission to teach more girls about robotics
• Adjust how you learn and quickly pick up robotics programming
• 3-D printing hydraulically-powered robots, no assembly required

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

An Air Force airman at the control module of an MQ-9 Reaper. Credit: John Bainter/USAF

Machine ethics offers an alternative solution for artificial intelligence (AI) safety governance. In order to mitigate risks in human-robot interactions, robots will have to comply with humanity’s ethical and legal norms, once they’ve merged into our daily life with highly autonomous capability. In terms of technical challenges, there are still many open questions in machine ethics. For example, what is deontic logic and how can it be used for improving AI safety? How do we fashion the knowledge representation for ethical robots? What about the adaptiveness of the ethical governance? These are all significant questions for us to investigate.

In this interview, we invite Prof. Ronald C. Arkin to share his insights on robot ethics, with a focus on its technical aspects.

Date: February 1st 2017

Interviewee: Prof. Ronald C. Arkin, Regents’ Professor and Associate Dean in the College of Computing at the Georgia Institute of Technology and Executive Committee Member of IEEE Global Initiative for Ethical Considerations in Artificial Intelligence and Autonomous Systems

Interviewer: Dr. Yueh-Hsuan Weng, Co-founder of ROBOLAW.ASIA and Fellow at CLAST AI & Law Research Committee (Beijing) and Tech and Law Center (Milan).

Ron Arkin

Q1 WENG: Thank you for agreeing to an interview with us. Can you tell us a little about your background?

ARKIN: I have been a roboticist for 35 years, so it’s hard to tell [only] a little, but I will do my best. Currently I am a Regents’ Professor and Associate Dean in the College of Computing at the Georgia Institute of Technology. I got my Ph.D. from the University of Massachusetts in 1987, and have been working on robotics since that time. I actually have degrees in chemistry as well—a bachelor’s and a master’s in organic chemistry. So, I have a rather diverse background and followed a non-traditional path compared to most computer scientists.

Q2 WENG: What is your definition of “Ethical Robots”?

ARKIN: That’s an interesting question! I have been asked that one before. I would say an ethical robot is a robotic system that is able to comply with the ethical and social norms of human beings—in some cases in accordance with international law, in other cases in accordance with national or local law. In addition, also being able to comport with humans in a natural manner, obey, and adhere to the same principles as an ethical person.

Q3 WENG: Could you please tell us why you are particularly interested in “Embedded Ethics”? Why is this concept so important?

ARKIN: That’s why you need to read the preface of my book [Governing Lethal Behaviour in Autonomous Robots]. It’s a rather long story. But let me try to give you a short version. Again, it dates back to a meeting—the first international symposium in Roboethics in 2004—where I was exposed to international conventions, e.g., the Geneva convention. It also led me to understand the Vatican’s position on how they viewed appropriate ethical interactions. Perhaps even more important than that one experience was the fact that we roboticists that have been in the field for some time finally started to have success, and we were seeing our technology move out of the laboratory community and into the real world. As such, it wasn’t hard to see how this could go in a variety of different directions.

Some robotics technology could benefit humanity, some not so much. At that point, one has to ask, “who is responsible for this?” So I assumed that, in part, it was myself. I believe that I have to take the responsibility for the technology that I helped to create. The issue is that we, as roboticists, had actually succeeded in many ways in terms of creating technology that we only envisioned when we first started that is going to have potentially profound impact on our society and the world as a whole. As such, we should be proactive in the ways in which we move forward [in order] to get a better understanding of the consequences of the work, and the social implications of the technology we are designing.

https://www.youtube.com/watch?v=jOJwkvihYtc

Q4 WENG: The IEEE Global Initiative for Ethical Considerations in Artificial Intelligence and Autonomous Systems has published their report “Ethically Aligned Design” in December 2016. As a member of the IEEE Global Initiative, could you please tell us about your role?

ARKIN: I serve a variety of roles. First, I am on the executive committee of that particular initiative, so I work at a high-level with about 30 other people or so. We have committees that large. I also co-chair the committee in Affective Computing. Joanna Bryson is my co-chair on that.

Affective Computing is concerned with the issues surrounding the illusion of emotions we can potentially create in these artifacts, and the ways human beings can bond to them. We are exploring a variety of different things, including robot intimacy, or the notion of applying robotic nudges and moving people in certain directions subtly, and in so doing to change the way people act and behave using these technologies, while trying to understand what may or may not be appropriate guidelines based on ethical considerations.

It’s about 15-20 people in that committee. We are one of the newer committees and we only have a couple of pages in the initial design report. But actually, in the June meeting in Austin, for the next release of the document 2.0, we will have as much as the other committees do. Finally, I am also a member of the humanitarian committee. I think it is called economic and humanitarian. I don’t know the exact title, but I am interested in the humanitarian side, and potential uses of robotics for humanitarian applications.

Q5 WENG: What is “Deontic Logic”? Why does it matter to machine ethics?

ARKIN: It may or may not matter. It’s just one approach to a particular problem. Deontic logic is a logic of prohibitions, obligations and permissions. It is one type of framework that is more transparent in order to get a better understanding of the way in which systems operate in the real world. Then it explains its reasoning to a human being, as compared to say deep learning or other forms of neural networks which may indeed provide good results but provide no explanatory capability to humans after the decision has been made.

The notion of prohibitions basically tells the robot some things it cannot do. The notion of permissions describe the things it is potentially allowed to do. And the notion of obligations says things that the robot must do under specific circumstances. It provides a nice framework of being able to explore morality. Others have already explored this, like Bringsjord at RPI who has applied deontic logic in this framework. We are certainly not the first in that particular space. Much of the work we have done has dealt with first-order predicate logic as a basis. It doesn’t have to be deep because it is so explicit in many cases.

Q6 WENG: Your hybrid embedded ethics framework aims for governing lethal behavior of military robots. Can we apply the same technical framework to service robots? For example, personal care robots might have some non-lethal, but still risky autonomous behaviors that need to be regulated by machine ethics.

ARKIN: I believe so. Indeed the whole point of the architecture is trying to design the systems to make them generalizable and not ad hoc for particular sets of circumstances. We have yet to look at service robotics in a traditional sense, but we have already developed something we refer to as the intervening Ethical Governor, which we are using for health care applications—particularly for assisting the management of two human beings in a particular relationship, like a patient and their care giver.

A robot hands a medication bottle to a person. Photo credit: Keith Bujak. Source: Georgia Tech News Center

We are doing work for the National Science Foundation, joint with Tufts University, looking at how in early stage Parkinson’s there is a condition called the facial masking. The loss of muscle movements in the face of the patient can produce stigmatization, and the patient-care giver relationship deteriorates due to the loss of non-verbal communication. We just had a paper accepted at ICRA describing the architecture that we have and how it has been modified from the military version. We generalized it from the military constraint-based version to a rule-based version.

But also, instead of using the international humanitarian law, we are using occupational therapy guides as a basis for generating the responses when two human beings in the relationship step outside of acceptable bounds. So the acceptable bounds now are not defined by international humanitarian law, but they are defined by occupational therapy—such as when a patient or a care giver start to yell at each other, or one starts to leave the room. The hard part is recognizing when dignity is being abridged, particularly in the patient. So we model the moral emotions of empathy, shame and embarrassment. We tried to ensure that the empathy level in the care giver remains within acceptable bounds and that the patient shame and embarrassment doesn’t rise to unacceptable levels. So the intervening Ethical Governor intervenes when a really inappropriate event occurs. It then provides strong corrective action by the robot to encourage people to get along better. The moral adaptor component which we used in the ethical architecture—initially with the guilt model—has now been extended to other moral emotions like shame, embarrassment, guilt and empathy [in order to] try and provide more subtle non-verbal queues [to help the] robot steer the relationship in the correct way before it get out of hand.

Q7 WENG: Regarding your research on machine ethics, what kind of indicators you use to evaluate the final outputs of your project?

ARKIN: In the first one, we were designing a prototype to prove the concept of the architecture. We never make the claim that this is the best way to do it. It is just one way to do it. Then we demonstrate that particular capability through the use of simulations and other aspects. The system was not intended to be fielded, but hopefully others could take those core ideas and potentially put them to use in the real world if they so choose. For the intervening governor, we’re running human subject tests right now—not with Parkinson’s patients yet, but with the elderly population, 50 or older. We’re trying to get a better understanding of the correct sets of rules, and to correct the interventions in a normative fashion, so we can ultimately move on in the final year of the grant to work with actual Parkinson’s patients. Again, using standard HRI technics, such as recording the kinds of reactions they have and parsing the data from the sessions and using pre- and post- evaluation questionnaires.

Q8 WENG: Let’s talk about your proposed machine ethics framework. You had mentioned that it includes four main parts: (1) Ethical Governor (2) Ethical Behavior Control (3) Ethical Adaptor (4) Responsibility Advisor. First of all, what is the major difference between Ethical Governor and Ethical Behavior Control? Is it the difference between deliberative and reactive control, or something else?

ARKIN: The Ethical Governor is basically a standalone component to the architecture which is a bolt on component at the end of the existing system. In the case of military scenario: asking for permission—should it fire or not. This is an intervening stage which either grants or denies permission through the validation of permission-to-fire variable. So it is straightforward thing to be able to do. The Behavior Ethical Control is a different kind of thing as it works purely with the reactive component. What it tries to do is, for each of the individual behaviors themselves, to respond whether or not to engage a target. This is prior to dealing with the Ethical Governor making the decision; trying to make the decision correctly before it gets to the Ethical Governor, where individual behaviors are given ethical guidelines to try and ensure that the behavior is correct itself. Each of the behaviors is provided with appropriate ethical constraints. So the difference is: one is located within the reactive component in each of the individual behaviors that makes the activity of the architecture, and the Ethical Governor is at the end of the architectural chain.

Q9 WENG: What is the importance of Ethical Adaptor? What kinds of machine learning skill that you chose to improve its adaptiveness?

ARKIN: Machine learning is something we haven’t generally explored in this particular context. Especially the military context, we kind of engineer out the adaptiveness and machine learning for fear that the robot would be granted the authority to start to generalize the targets in the field, which we feel is inappropriate. Learning after action is okay. but that’s a different story. The importance of the Ethical Adaptor, and its cognitive models of guilt that we use, is the fact that even the actions the agent takes may inadvertently end up unethical. This is referred to as a “moral luck”. It could be good luck or bad luck. But if it is bad luck, you want to be able to ensure proactively that the bad action doesn’t occur again. That’s actually the functions of cognitive guilt for people as well to reduce the likelihood of reoccurrence of a behavior that resulted in bad actions. By giving this to a robot we hope that if it makes a mistake it will not repeat it. The probability of that behavior reoccurring would be reduced.

The best example is the case of battle damage assessment for a military scenario. For example, if a robot does its assessment and it looks like a released weapon wouldn’t create too much collateral damage, within acceptable limits, then it will release the weapon. But after it does it, the battle damage action is actually more severe then, it initially appeared to be. Under those circumstances, we want the robot in the field to reduce the probability of releasing the weapon again under similar circumstances. That’s, in essence, the analogue of guilt we use, and eventually, it will get to the point that the robot will not engage targets at all. Again, the software is updated to make sure the use is consistent with Geneva convention.

Q10 WENG: The Laws of War (LOW) and Rules of Engagement (ROE) are written in the form of natural language. How did you transfer these norms into machine readable formats?

ARKIN: It’s important to keep it in mind that we do not try and capture all the Rules of Engagement and all the Laws of War. We always advocate that the system should be used in very narrow circumstances and very narrow situations, such as building clearing operations or counter-sniper operations. It’s impossible to capture all human morality and all the Laws of War at this particular point of time. So, the question is what are the relevant Laws of War and Rules of Engagement for these particular narrow situations that the robot will find itself in.

Let’s take a room clearing operation. We try and do this using these constraint-based approaches. The Laws of War basically say that if you see a civilian, you should not shoot them; if you see a surrendered person, you should not shoot them. For aerial vehicles, if you are outside the kill zone (this is the simulations we did) which is designated by the Rules of Engagement, you cannot fire your weapon—explain simply why you cannot do that. If you release the weapon a toward a legitimate target but it would result in damage of cultural property, like a mosque or church or some similar circumstance, that is illegal as well. You are forbidden from doing those sort of things.

So, the initial work we did was more for aerial circumstances. We also computed the collateral damage using a battle damage assessment strategy to determine whether if it is within acceptable limits for a particular target and to restrain the weapon for proportionality. To be able to do this you have to be able to understand the military necessity of the target, which is assigned by a human being. You have to be able to understand the proportionality of the weapon. In other words: Which weapon is sufficient to take out the target but to do no more damage of the surroundings than is absolutely necessary? You have to ensure that you don’t destroy cultural property as well. Basically, adhere to all necessary principles, such as discrimination which is to be able to tell apart combatants and non-combatants. We also state we should restrict the use of these particular systems for high-intensity combat operations, or inter-state warfare, and not the kind of circumstances you find in modern counter- insurgency operations. They should be used with great care and great caution in very narrow circumstances where the predictability of their performance is far more likely to be good.

Q11 WENG: Why did you choose case-based reasoning (CBR) but not rule-based reasoning (RBR) when you designed Responsibility Advisor?
ARKIN: Historically we used case-based reasoning a lot for a variety of different reasons. Part of the reason is to gain information from previous experts. One of the examples in the past was to work with snipers in the laboratory. We can work with experts. It can be legal experts, it can be medical experts, it can be military experts. Again, the primary reason is to be able to readily use the experience of warfighters to guide us in operations.

Q12 WENG: Finally, your proposal “Governing Lethal Behavior: Embedding Ethics in a Hybrid Deliberative/Reactive Robot Architecture” was published in 2007. There have been great technological improvements in artificial intelligence in the past decade. Are there any recent updates from your robot ethics research, or anything that you want to share with us?

ARKIN: The project ended around 2010, so there is additional information. As I say, I would encourage those who are really interested to take a look at the book entitled “Governing Lethal Behavior in Autonomous Robots” . That represents the final statement and outcome for that particular project and includes the significant technical details describing that work. Recent advances include the application in other domains, such as the caregiver-patient relationship [discussed above] using the same underlying principles, including both the Moral Adaptor and Ethical Governor in a slightly different. If we were successful, it may also be useful for assisting other human-human relationships, such as a parent and a teenager who may be at odds with one another or even in marriage consulting. We hope to use robots for the benefit of mankind, whether it’s to make sure that warfare is conducted within internationally agreed upon laws, or whether it’s to ensure that people behave appropriately toward each other

If you enjoyed this article, you may also enjoy:

• How do we regulate robo-morality?
• One being for two origins: A new perspective on roboethics
• Considering robot care, ethics & future tech at #ERW2016 European Robotics Week
• On the ethics of research in robotics, with Raja Chatila
• A European perspective on robot law: Interview with Mady Delvaux-Stehres
• Open Roboethics initiative delivers statement to United Nations CCW

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

National Grid, the UK grid operator, has several ways of ensuring supply can always meet demand. For shorter gaps in generation, it asks electricity suppliers to run their conventional power stations at below maximum potential output and ramp up as needed.

For longer gaps, it ensures power stations, particularly gas-based ones, are kept on standby. Some stations may only be asked to generate power for between several dozen and a few hundred hours a year. Besides contributing to carbon emissions, operating power plants for such short interventions is expensive.

The question is what to do about this problem. We could build less renewable power and make conventional power “greener” instead by removing the CO₂ and burying it underground. Opinion divides on when these carbon capture technologies can be made commercially viable on a large scale. In the UK, unfortunately two government kickstarter projects have floundered due to concerns about costs and departmental disagreements.

An alternative is to install very big (“grid scale”) batteries capable of storing renewable power to be released when required. This has generated a lot of interest lately. But given the costs of current battery technology, grid-scale storage requires expensive upfront investments.

Solutions on demand

While researchers study these problems, the UK is developing an alternative known as demand-side response. One aspect involves rewarding certain electricity consumers for reducing their usage at short notice. This can range from large industrial customers to smaller consumers using power for heating rooms, cooling, lighting or even refrigeration.

Solar thermal energy system in a modern house.

The other aspect of demand response involves asking customers who own equipment that can store power to help balance surges in demand. For example, the owners of a house equipped with solar panels and corresponding battery storage might reduce repayment costs on the equipment by making the battery units available to the grid. Other equipment in this category includes electric vehicles and hospital/university uninterruptible power supply (UPS) units.

Both types of demand response are happening already. Some industrial power customers and certain other companies such as hotel operators have contracts for reducing power, while National Grid has been attracting much bidder interest for power storage schemes and has some underway in parts of the country. This storage is an alternative to deploying large-scale batteries, and promises to be much more economical if we can make it work on a large enough scale.

Once a pool of customers is set up, some devices might not always be available for storage or reducing demand when needed. This needs to be factored into the calculations both to minimise grid disruption and incentivise customers to participate at these times.The problem is that these schemes get more complicated once the pool of customers gets beyond a certain size. Knowing which customers to sign up and what tariffs to offer requires understanding to what extent devices will be available and at what cost, for example.

There can also be undesired effects, such as large simultaneous rebounds in consumption. For example many refrigerators will draw extra power to get their internal temperature below the required level when a demand response period ends.

Finally there’s a potential major security issue: a central system that collects data about energy usage from many devices may be prone to malicious attacks and information tampering. This could undermine both grid balancing and keeping track of what customers are owed.

How AI can help

Emerging artificial intelligence technologies look like providing answers to these challenges. To select the best participants, for example, grid operators will be able to use sophisticated machine-learning techniques to model the behaviour of individual devices and battery storage units by reviewing data from smart meters and sensors.

Smart Grid concept Industrial and smart grid devices in a connected network. Renewable Energy and Smart Grid Technology Smart city design with future technology for living.

Once signed up for grid storage, it should be possible to estimate the useful lifetime of a battery pack or unit by applying prognostic algorithms to its charging/discharging data. Owners will then receive appropriate compensation, plus the added incentive of knowing how long their battery will last.

When it comes to managing devices in the pool, people used to think we could use individual smart meters or control devices to feed a central server in the cloud. But meters are expensive and the short response times require the cloud server to analyse data in milliseconds, which looks unfeasible once many thousands of units are in a pool.

An alternative is to have metering devices which detect demand levels on the grid themselves and reduce power accordingly. These take pressure off the central server and it only requires metering at site level, rather than for every electrical device. But it still leaves you with a complex control problem in coordinating all these individual decisions. We at Heriot-Watt are working on a solution to this using AI-based algorithms.

Another line of AI research draws on insights from algorithmic game theory to develop reward/penalty mechanisms which ensure enough customers in the pool are willing to participate, and actually respond when necessary. Researchers are also optimistic that blockchain protocols, using the same technology as Bitcoin, could underpin a decentralised ledger system that would get round the security risk of having a single storage point for user data.

Numerous AI research groups, both in the UK and elsewhere, have been addressing these challenges, while a number of start-ups have started developing such systems in practice – relatively simple versions of machine learning are now beginning to be used, for instance. The UK has a good chance to be at the forefront of international efforts to make smarter demand response a reality over the next few years.

This article was originally published on The Conversation.

You might also enjoy the following articles:

• Wearable AI that can detect the tone of a conversation
• Artificial intelligence and ethics: Who does the thinking?
• Soft exosuit economies: Understanding the costs of lightening the load
• MIT Media Lab to participate in $27 million initiative on AI ethics and governance

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

bots_alive Gotcha

Modern character AI could be better.

I’m unaware of any NPCs or electronic toy characters that can sustain an illusion of life over more than an hour. They suffer from predictability, simplicity, and inauthentic verbal and nonverbal behavior.

This post is about a new form of character AI. Much like motion capture for scripted animation, this new technique may revolutionize how interactive characters are created, through observation of authentic human-generated behavior.

I’ll explain via a product on which we’re pioneering this technique.

Bots_alive! A robot that plays and finds its way

We launched bots_alive last week on Kickstarter (campaign here). Here’s a 1-minute teaser video with mostly scripted motion.

The product is a smartphone kit that gives simple but lifelike autonomy to a popular RC toy, the Hexbug® Spider.

That video shows the product as we envision it at shipping time. Here you can see how it acts now, with this new character AI technique driving its core functionality already.

Training data to build a character

The character AI for these robots is created in a new way. Here’s how it works.

A human teleoperates the robot through many different scenarios. The teleoperator acts as an improvising puppeteer; he or she follows heuristics yet also empathizes with the robot and authentically responds to what happens, whether it be with curiosity, fear, jubilation, or another emotion.

For bots_alive, the teleoperator watches a screen display of what the system sees and pushes buttons to send commands of forward, back, left, right, forward-right, forward-left, back-right, or back-left. We also consider not pushing a button to be a no-action command.

From these teleoperation sessions, we gather training data containing (a) teleoperation commands and (b) information about the context in which each command was given. A category of machine learning called supervised learning is applied to create a model of the puppeteer, which effectively answers the question: In context X, what is the probability that the teleoperator would have given command Y?

Creating a behavioral model

All of that happens during development. In the end user’s hands, the robot is autonomously controlled by the model. The model of the puppeteer is the character AI.

Let’s assume that the contextual information in the training data is sufficient to justify why the teleoperator chooses one action over another. Then the data contains the spontaneity, uncertainty, and social authenticity* expressed by the human puppeteer. And the autonomous robot retains these qualities if the machine learning is effective.

Why go to the trouble of using machine learning?

We expect this process to create character AI that feels more organic and alive than traditional character AI.

Traditional methods consist of imagining and writing rules for behavior or a finite state machine. It’s much more of an abstract exercise than immersing someone in a situation as the puppeteer.

If you play video games, imagine how much you sometimes empathize with the character you’re controlling. You may even forget the separation between the two of you, experiencing the world as that character. Consider whether your control in these moments is different, both at a high level and in tiny movements, than what the character would do if you had simply written down a set of rules for it to act by.

In that difference lies the promise of this method for developing character AI.

In psychology research methodology, there’s a broad consensus that if you want to know what someone would do in a situation, you don’t ask them what they would do. You put them in the situation and observe. Traditional character AI development is like asking the human developer or designer. Our method is instead analogous to observing a human immersed in the experience of the character. That’s why we expect it to create more authentic characters.

Our method is an application of what’s called learning from demonstration.** For times the teleoperation needs to include human interaction with the character, we keep the teleoperation secret to keep the human interaction partner from changing his or her behavior from what would be done with an autonomous character. We coined this secret version learning from the wizard (or LfW) when it was a research project at the MIT Media Lab. The name combines learning from demonstration with the Wizard of Oz experimental paradigm.

Training as interactive machine learning

In our development, I act as the teleoperator and machine learning programmer. The training process is not simply a bunch of demonstration. Rather, it’s an iterative process of:

• demonstrate
• apply machine learning on the demonstration data set
• observe behavior from the learned model
• create more demonstrations in those contexts the robot isn’t acting satisfactorily
• apply machine learning, and so on.

A general representation of interactive machine learning. From Amershi et al., 2015. Power to the People: The Role of Humans in Interactive Machine Learning. In AI Magazine.

Throughout these iterations, the teleoperator and algorithm designer also reflect on what contextual information still needs to be encoded to improve learning, determine what context cannot be encoded and therefore should be ignored by the teleoperator, and find fun and more delightful behavior than originally puppeteered.

How can we know that it’s better?

We can’t. Not with 100% confidence. But we have already seen that it can make compelling character AI.

We saw it in our study at MIT. In a relatively large randomized experiment, kids interacted either with a human-teleoperated robot or one with autonomous behavior learned from previous teleoperation. Compared to a no-robot condition, the teleoperated robot and the autonomous robot programmed by machine learning elicited similar behavior from their human interaction partners. Curiously, when children were asked whether the robot was human-controlled or autonomous, approximately half in each condition thought it was human-controlled. If you’re familiar with the Turing Test, the popularly known test of the effectiveness of an artificial intelligence, you might recognize that the MIT study constitutes passing a certain narrow and social Turing Test.

Compelling character AI was achieved in another such Turing test, using a technique similar to learning from demonstration, in Unreal Tournament 2004. BotPrize is a yearly academic competition for the bot that best passes as human. The 2012 winner was a bot by Jacob Schrum, Igor Karpov, and Risto Miikkulainen. It combined algorithmic optimization—which can result in markedly unnatural behavior that still achieves the specified goal— of its combat effectiveness with selective replay of recorded human behavior. Twenty-five human judges played alongside the Unreal-playing bots, repeatedly judging whether the other characters were human or artificial during combat. The bot by Schrum and colleagues was judged human more than 51.9% of the time. That percentage was just below the two human characters who were judged most human, at 53.3% and 52.2%. The algorithm is described here.

Lastly, we see compelling character AI now, in our playtesting of bots_alive.

The limitations of this approach are (1) sensing and encoding contextual information and (2) applying machine learning effectively. Both academia and industry are intensely researching how to improve those two limitations. Major strides have been made in the past 10 years, most notably through deep learning, and so it seems likely that further improvement is yet to come.

Our ability to write down behavioral rules that encode authentic behavior with an illusion of life? If there’s a reason why that will improve meaningfully, I don’t see it.

Part 1: Autonomous Systems and Safety

We all rely on machines. All aspects of modern life, from transport to energy, work to welfare, play to politics depend on a complex infrastructure of physical and virtual systems. How many of us understand how all of this stuff works? Very few I suspect. But it doesn’t matter, does it? We trust the good men and women (the disgracefully maligned experts) who build, manage and maintain the infrastructure of life. If something goes wrong they will know why. And (we hope) make sure it doesn’t happen again.

All well and good you might think. But the infrastructure of life is increasingly autonomous – many decisions are now made not by a human but by the systems themselves. When you search for a restaurant near you the recommendation isn’t made by a human, but by an algorithm. Many financial decisions are not made by people but by algorithms; and I don’t just mean city investments – it’s possible that your loan application will be decided by an AI. Machine legal advice is already available; a trend that is likely to increase. And of course if you take a ride in a driverless car, it is algorithms that decide when the car turns, brakes and so on. I could go on.

Algorithmic trading is now commonplace

These are not trivial decisions. They affect lives. The real world impacts are human and economic, even political (search engine results may well influence how someone votes). In engineering terms these systems are safety critical. Examples of safety critical systems that we all rely on from time to time include aircraft autopilots or train braking systems. But – and this may surprise you – the difficult engineering techniques used to prove the safety of such systems are not applied to search engines, automated trading systems, medical diagnosis AIs, assistive living robots, delivery drones, or (I’ll wager) driverless car autopilots.

Why is this? Well, it’s partly because the field of AI and autonomous systems is moving so fast. But I suspect it has much more to do with an incompatibility between the way we have traditionally designed safety critical systems, and the design of modern AI systems. There is I believe one key problem: learning. There is a very good reason that current safety critical systems (like aircraft autopilots) don’t learn. Current safety assurance approaches assume that the system being certified will never change, but a system that learns does – by definition – change its behaviour, so any certification is rendered invalid after the system has learned.

And as if that were not bad enough, the particular method of learning which has caused such excitement – and rapid progress – in the last few years is based on Artificial Neural Networks (more often these days referred to as Deep Learning). A characteristic of ANNs is that after the ANN has been trained with datasets, any attempt to examine its internal structure in order to understand why and how the ANN makes a particular decision is impossible. The decision making process of an ANN is opaque. Alphago’s moves were beautiful but puzzling. We call this the black box problem.

Safety critical systems such as aircraft autopilots do not learn

Does this mean we cannot assure the safety of learning autonomous/AI systems at all? No it doesn’t. The problem of safety assurance of systems that learn is hard but not intractable, and is the subject of current research*. The black box problem may be intractable for ANNs, but could be avoided by using approaches to AI that do not use ANNs.

But – here’s the rub. This involves slowing down the juggernaut of autonomous systems and AI development. It means taking a much more cautious and incremental approach, and it almost certainly involves regulation (that, for instance, makes it illegal to run a driverless car unless the car’s autopilot has been certified as safe – and that would require standards that don’t yet exist). Yet the commercial and political pressure is to be more permissive, not less; no country wants to be left behind in the race to cash in on these new technologies.

This is why work toward AI/Autonomous Systems standards is so vital, together with the political pressure to ensure our policymakers fully understand the public safety risks of unregulated AI.

In my next blog post I will describe one current standards initiative, towards introducing transparency in AI and Autonomous Systems based on the simple principle that it should always be possible to find out why an AI/AS system made a particular decision.

The next few years of swimming against the tide is going to be hard work. As Luke
Muehlhauser writes in his excellent essay on transparency in safety-critical systems “…there is often a tension between AI capability and AI transparency. Many of AI’s most powerful methods are also among its least transparent”.

*some, but nowhere near enough. See for instance Verifiable Autonomy.

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

The Arm

The design I used is based on industrial pallet-packing robots, and at its core has three degrees of freedom, or ‘axes’ on which it can move. Think left to right, in and out and up and down; it basically means that the arm can move in three different ways. The GIFs below show the different types of movement: the entire arm can swivel on its base, the “main arm” translates the elbow in and out from the centre, and the “actuator” drives the forearm, which in turn translates the end-effector up and down. A 3-DOF arm like this has the handy property of having a unique solution for each possible position of its end-effector, which made the IK task approachable to someone like me, who’s never done this type of thing before.

Swing

Main

Actuator

Like many well-studied problems, IK has generic solutions, but these are numerically-based and computationally complex. I wanted a simple and efficient solution, so I decided to approach the control of my arm analytically, using geometric principals and logic. The result of this work was a graphical Python application, which connected to the arm via serial and allowed the user to move the goal position of the end-effector by clicking and dragging, or by sending commands from another application running on the same computer (this IK application made up a key part of my project, but that’s a topic for another time).

Here’s a quick look at what I ended up with:

Heads up: The code in this post is written in Python with Numpy, and there’s maths. I won’t pretend that any of the ideas presented here are optimal (or good, for that matter), but hopefully you’ll be able to scrounge something useful!

Let’s go already

To start with, let’s consider the “swing” action of the arm, as it rotates on its base. Since the main arm, elbow and forearm are all constrained to a single plane (look at the GIFs above: only the swing rotation affects the left-right position of the wrist), all we really have to do here is line up that plane – a line from our top-down perspective – with the goal point, ignoring the goal’s vertical position. If this sounds easy, you’re right: it’s plain old trigonometry. Unfortunately for me, a quirk of my design left the arm’s operation plane just off-centre of the axis of rotation, which made the solution much less obvious.

Using the naive trigonometric approach here results in pointing off to the left of the actual goal (right if the shoulder sits on the other side), and the gap is big enough that this is a real problem. I took a shot at solving this geometrically, but quickly decided it wasn’t worth my time figuring it out. In the end, I resorted to an extremely simple numerical technique called a binary search. I’ve used this in the past for other tough geometric problems, like finding an intersection between a Bezier curve and a circle.

In a binary search (also known as the “Bisection Method”), you essentially come up with arbitrary upper and lower bounds for your variable – in this case, theta – and then examine the midpoint to find whether it’s larger or smaller than your target. If it’s larger, then the midpoint becomes your new upper bound; if it’s smaller, the midpoint is your new lower bound. Repeat until you’ve bounded your variable with sufficient accuracy. A binary search is conceptually simple and computationally cheap, and converges fairly quickly: you halve your search space each iteration. I found 3-5 iterations got me within 0.3 degrees of the true value, well within the limits of what my servo motors could practically target.

On to the solution! What we’re actually searching for is the point at which the vector representing our arm is lined up with the vector from the arm’s shoulder joint to the goal: where the angle between these two vectors is zero. Normally, when taking the angle between two vectors (call them A and B) you get the absolute angle out, with no directionality. However, with this search method it’s important to have a signed distinction between “too high” and “too low”. To get around this, I wrote a simple function to find a signed angle between two vectors, with negative showing that A points counter-clockwise to B, and positive showing that A points clockwise from B:

This works by taking a perpendicular vector from B (pointing clockwise), and dot-ing it with A to see how much they line up (using the dot product for
vector projection).

If A points clockwise of B then np.dot(A, perp) will be positive, and vice-versa.

Note that the normalize() function simply sets a vector’s length to 1, while keeping it pointed in the same direction. Mathematically, it’s just doing: A=|A|^-1 (np.clip() just keeps the dot product between 1 and -1).

With that, we can now perform our binary search and find a decent approximation to the target angle. We need a reasonable starting point, so let’s just use the angle from the origin to the goal (the naive approach I mentioned above). Numpy provides a very useful function for finding the angle np.arctan2(y, x) which resolves the arc-tangent of in the correct quadrant.

Now, it’s important to check which side the shoulder sits on: this lets us decide whether to use start_theta as an upper or lower bound. This is as simple as testing the signed angle between the shoulder offset vector (from the origin) and a vector pointing straight forwards. At the same time, we can also establish the other bound by taking the absolute value of this angle and adding or subtracting it from start_theta, since this angle to the shoulder is always going to be greater than the error of start_theta.

With our initial bounds established, the actual search loop is pretty straightforward. Note the use of an iteration counter: it’s only there in case something goes wrong (e.g. if we input an unsolvable configuration), since in normal operation the loop exits once the desired accuracy is reached. We iteratively update self.swing, which will hold the new target swing angle at the end of the search.

Note that the self.shoulder(angle) function just generates the top-down (x, z) position of the shoulder joint for a given swing angle. Besides the swing angle, there’s one other important result from this stage of IK: the radial distance, from the shoulder joint to the end-effector. The calculation is simple enough:

goal_radial = np.linalg.norm(self.goaltd – self.shoulder(self.swing))

Round in circles

We know which way to point: now, let’s consider the movement of the arm in its ‘plane of operation’, which covers the goal end-effector position and the line extending vertically (+y) from the shoulder joint. The main arm and forearm both move across this plane. The trick here is that we’ll still be using a 2D coordinate system, basically a local frame of reference, where our y value remains the same but the x value is the radial distance of the point we’re interested in. For the goal, we computed this radial distance above in the top-down IK, so we can go ahead and construct goalpl – the ‘planar’ goal – which lies at: goalpl=(goal_radial/goal_y).

Side-on, the main arm and forearm form an upright triangle between the shoulder joint and goalpl with the elbow joint sitting at the peak. It’s important to note that the arm’s sections are different lengths, with the forearm being a bit longer. Now, in this plane we’re solving given the end effector’s position, and the origin is static. That just leaves the position of the elbow joint to solve for, which we can do geometrically, since it’s at a fixed distance from both the origin and effector. If we draw two circles, one fixed about the origin with radius = main arm length, and one centred on the end-effector with radius = forearm length, then the elbow must be at one of the two possible intersections between these circles:

Let’s look at a simple Python class which deals with this intersection calculation. There’s a few possible cases for circle-circle intersection, most of which invalidate our solver: if the circles do not contact, either by being separated or in the case that one contains another, then there is no valid solution; if the circles are coincident (impossible for this design, since the radii differ) then there are infinite possible solutions; however, when the circle borders intersect, we can find two distinct intersection points.

For any case we’re interested in, it’s the top intersection that matters. This means we can just pick the one with the largest y value, and call it a day:

And – that’s it! We have the swing angle, and a valid position for the elbow joint to match the shoulder and goal end-effector positions. That’s all we need to define a valid pose for the arm. If you need to, you can reconstruct the positions in 3D space:

Wrapping up

If you’ve made it this far, I hope you’ve found this interesting and/or enlightening. Want to see how it all fits together? I’ll be uploading the full IK application to GitHub here, along with all the design files for the robotic arm. If you’re feeling ambitious, you should be able to 3D-print and build your own to play around with

Other notes:

• I didn’t design the arm completely from scratch: check out the original “LiteArm” design on Thingiverse.
• The IK application uses PyGame, a no-frills SDL wrapper for Python, to open and draw to the graphical window. I’ll probably switch to a meatier library for further development, but this did the job of basic “drawn-to-screen” stuff nicely.

Written by Alistair Wick for the University of Bristol Engineering Maths blog. The original post, and future follow-ups, are available on Alistair’s website.

If you liked this article, you may also enjoy these:

• TrotBot tackles rough terrain
• Sticky business: Five adhesives tested for 3D printing
• Meet RoboGabby: She’s on a mission to teach more girls about robotics
• Adjust how you learn and quickly pick up robotics programming
• 3-D printing hydraulically-powered robots, no assembly required

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

This year marks the occasion of ROS turning 9 years old! Over the years, ROS has grown into a strong world-wide community. It’s a community with a large variety of interests: from academic researchers to robotic product developers, as well as the many robot users. Academic use of ROS continues to grow. Citations of the first ROS paper “ROS: An Open-Source Robot Operating System” has grown to 2,871.

To get a better sense of what’s happening in the ROS community, if you have not already done so, I highly recommend reviewing the ROSCon 2016 program. You can also find all the video recordings in this gallery. ROSCon 2016 was another great event bringing ROS community members together to share how they’re using ROS to solve their challenges. As the goal of ROSCon is to share information between the entire community, we record the talks and make them available online. We’ve sold out our venues the last two years and are looking forward to another ROSCon next fall!

Part of understanding our growing community is to try to measure it. For the last 6 years we’ve been generating metrics reports. These reports can give a sense of aggregate concerning what’s happening in the ROS community. Our most recent report is from July 2016. David Lu has put together plots of several of the metrics across the last 6 years which can be quite informative.

This year we wanted to dig a little deeper into the code metrics, so we downloaded the source of all of packages listed in the Indigo Igloo rosdistro and ran some analysis.

• The total line count is over 14 million lines of code
• There have been 2477 authors
• And 181509 commits
• Averaging 73.3 commits per author

You can see the commits as a function of month in this graph.

Our committers are active around the world as evidenced by the commits coming in at all hours of the day.

And the git commits record 24 different time zones (out of 39 possible).

Analyzing the repository for significant lines of code using [SLOCCount](http://www.dwheeler.com/sloccount/) shows:

• 4,077,199 significant lines of code.
• This represents an estimated 1,236 person-years of development.
• For a sense of scale, that is an average of 137 developers contributing full time over the last 9 years!

For those of you curious about the breakdown by language lines of code, it is as follows:

• cpp: 2608592 (63.98%)
• python: 553332 (13.57%)
• ansic: 297629 (7.30%)
• xml: 280615 (6.88%)
• lisp: 149439 (3.67%)
• java: 135343 (3.32%)
• ruby: 26484 (0.65%)
• sh: 21120 (0.52%)

This only represents the packages publicly released into the Indigo rosdistro index.

Note that the tools only worked on Git repos, so code from other source control systems was excluded. There are also a few projects which predate ROS but have ported to use ROS and their history is included.

We’re looking forward to continuing growth through 2017 leading up to the 10-year anniversary of ROS. With the Beta 1 version of ROS 2.0 out, there will be space for new development. We’re looking forward to our next release, Lunar Loggerhead, to coincide with Ubuntu’s next release, Zesty Zapus. With both of these, the ROS community can continue to rely on the many libraries, tools, and capabilities they have come to know and enjoy, as well as begin to experiment with the new features in ROS 2.0

Another exciting project to watch is the upcoming TurtleBot 3! The TurtleBot and TurtleBot 2 have been great platforms for learning and prototyping. However by packing that same capability into a smaller platform with more punch, we look forward to it providing another avenue to grow the ROS community.

We write these anniversary posts to help give you a sense of how ROS has been doing over the past year, but we’d certainly encourage you to find out for yourself. Get involved. Write or edit a wiki page. Answer a question on ROS Answers. Come to ROSCon. And, when you’re ready, think about helping to maintain ROS itself, or even contributing a brand new ROS package.

OSRF is doing great, but the long-term success of ROS depends on every member of the incredibly awesome ROS community. If you’re already an active part of the ROS community, we can’t thank you enough; and if you’re not, think about how you can help ROS grow and thrive for the next nine years, and beyond.

If you enjoyed this article, you may also also enjoy:

• Introducing H-ROS: the Hardware Robot Operating System
• #IROS2016 in pictures
• A practical look at latency in robotics: The importance of metrics and operating systems

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

In this episode, Christina Brester interviews Vladimir Stanovov, PhD student and researcher at the Siberian State Aerospace University (Krasnoyarsk, Russia). Stanovov speaks about a speech-controlled wheelchair, which seeks to provide people that are quadriplegic, that is people with partial or total loss of use of their limbs and torso, with the possibility to control their wheelchairs through voice commands. In this interview Stanovov discusses the basic parts of the speech-controlled wheelchair, the fuzzy controller he created, and the trials they had in the medical center.

Below is a video that explains the speech-controlled wheelchair.

Vladimir Stanovov

Vladimir Stanovov received his Bachelor’s (2012) and Master’s (2014) degree at the Siberian State Aerospace University (SibSAU) in Krasnoyarsk, Russia. Stanovov is currently a PhD student in Computer Science and Engineering at SibSAU, where he performs the research under the supervision of Professor Semenkin. Stanovov’s PhD thesis is about evolutionary algorithms, fuzzy logic and their applications in machine learning.

Links:

• Download mp3 (11.6 MB)
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS
• Vladimir Stanovov on reserchgate.net

Research and development of robotic assistive technologies has gained tremendous momentum in the last decade due to several factors such as the maturity level reached by several technologies, the advances in robotics and AI and the fact that more than 700 million of persons have some kind of disability or handicap. For many people with mobility impairments, essential and simple tasks, such as dressing or feeding, require the assistance of dedicated people. Thus, the use of devices providing independent mobility can have a large impact on their quality of life.

A team of researchers from the University of Cassino, under the direction of Professor Filippo Arrichiello, is working on the development of a control architecture for assistive robotic systems operated via a Brain-Computer Interface (BCI) to allow a user to operate a robot using his thoughts. The proposed software architecture relies on widely used frameworks to operate BCIs and robots (namely, BCI2000 for the operation of the BCI and ROS for the control of the robot) integrating control, perception and communication modules developed for the application at hand.

In particular, using the BCI2000 framework, they developed a Graphical User Interface (GUI) based on P300 paradigm to allow the user to select options via a BCI Epoch+. The user can select objects from the scene and the action to perform by focusing attention on the corresponding flashing icons on a screen. By counting how many times the icon he wants to select flashes, the user generates a P300 potential that is detected via the BCI and translated in a selection operation on the GUI.

Messages related to the object and actions selected via the GUI are automatically sent to a 7DOF lightweight robot manipulator Jaco2 that has to operate in the environment. The robot control architecture relies on a perception module that allows to detection and localization of objects, and the face/mouth of the user, using a RGB-D sensor (Kinect one). The motion of the manipulator is then controlled relying on a closed loop inverse kinematic algorithm that simultaneously manages multiple setbased and equality-based tasks. Preliminary experiments have been performed where the user commands the robot via the BCI in order to move objects on a table or to select a bottle from where to drink.

Team members: prof. Filippo Arrichiello, Paolo Di Lillo (PhD student), Daniele Di Vito (PhD student), prof. Gianluca Antonelli, prof. Stefano Chiaverini.

Click here for more information on the Robotics Research Group of the DIEI.

If you enjoyed this article, you might also be interested in:

• Virtual race: Competing in Brain Computer Interface at Cybathlon
• Brain Computer Interface used to control the movement and actions of an android robot
• Can a brain-computer interface convert your thoughts to text?

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

Photo credit: Robert Shields

Will your next doctor be an app? A cost-cutting NHS wants more patients to act as “self-carers,” with some technologized assistance. A series of flowcharts and phone trees might tell parents whose children have chicken pox how best to care for them—no visits to surgeries required. Or a mole-checking app might tell a worrywart when a given skin discoloration looks harmless, and when to go to a dermatologist, by comparing it to thousands of images in a database.

Cost-cutters in the legal field also promise an algorithmically cheapened future. Tax software simplifies the process of filing by walking the filer through a series of questions. Documents that might have taken human attorneys months to read, can be scanned for keywords in a matter of seconds. Predictive policing promises to deploy force with surgical precision.

All these initiatives have some promise, and may make health care and legal advice more accessible. But they are also prone to errors, biases, and predictable malfunctions. Last year, the US Federal Trade Commission settled lawsuits against firms who claimed their software could aid in the detection of skin cancer, by evaluating photographs of the user’s moles. The FTC argued that there was not sufficient evidence to support such claims. The companies are now prohibited from making any “health or disease claims” about the impact of the apps on the health of users unless they provide “reliable scientific evidence” grounded in clinical tests. If algorithms designed merely to inform patients aren’t ready for prime time, why presume diagnostic robots are imminent?

Legal automation has also faced some serious critiques lately. The University of North Carolina legal scholar Dana Remus has questioned the value and legitimacy of the “predictive coding” now deployed in many discovery proceedings. She and co-author Frank S. Levy (of MIT) raise serious questions about more advanced applications of legal automation as well. The future cannot be completely anticipated in contracts, nor can difficult judgment calls be perfectly encoded into the oft-reductionist formulae of data processing. Errant divorce software may have caused thousands of errors in the UK lately, just as US software systems have disrupted or derailed proper dispositions of benefits applications.

Moreover, several types of opacity impede public understanding of algorithmic ranking and rating processes in even more familiar contexts, like credit scoring or search rankings. Consumers do not understand all the implications of the US credit scoring process, and things are about to get worse as “alternative” or “fringe” data moves into the lending mix for some startups. If the consequences of being late on a bill are not readily apparent to consumers, how can they hope to grasp new scoring systems that draw on their social media postings, location data, and hundreds of other data points? At the level of companies, many firms do not feel that Google, Facebook, and Amazon are playing a fair game in their algorithmic rankings of websites, ads, and products. These concerns, too, are stymied by widespread secrecy of both algorithms and the data fed into them.

In response, legal scholars have focused on remediable legal secrecy (curbing trade secrets and improving monitoring by watchdogs) and complexity (forbidding certain contractual arrangements when they become so complicated that regulators or citizens cannot understand their impact). I have recommended certain forms of transparency for software—for example, permitting experts to inspect code at suspect firms, and communications between managers and technical staff. The recent Volkswagen scandal served as yet another confirmation of the need for regulators to understand code.

But there is a larger lesson in these failures of algorithmic ordering. Rather than trying to replace the professions with robots and software, we should instead ask how professional expertise can better guide the implementation of algorithmic decision-making procedures. Ideally, doctors using software in medical settings should be able to inspect the inputs (data) that go into them, restrict the contexts in which they are used, and demand outputs that avoid disparate impacts. The same goes for attorneys, and other professionals now deploying algorithmic arrangements of information. We will be looking at “The Promise and Limits of Algorithmic Accountability in the Professions” at Yale Law School this Spring, and welcome further interventions to clarify the complementarity between professional and computational expertise.

This post was originally published on the website of Nesta.

In this episode, Audrow Nash interviews several researchers presenting their work at the Robotics Science and Systems (RSS) 2016 conference in Ann Arbor, Michigan.

Audrow speaks with Gangyuan Jing from Cornell University about modular robotics; Rico Jonschkowski from Technical University of Berlin about lessons from last year’s Amazon picking challenge, which his team won; Matthew Gombolay from MIT about software to help nurses coordinate; and Dorsa Sadigh from UC Berkeley about developing human-like behavior for autonomous cars.

Links:

• Download mp3 (9.7 MB)
• Link to RSS 2016 homepage
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS

In this episode, Audrow Nash interviews Hugh Herr, Director of the Biomechatronics Group at MIT. Herr talks about the accident that led to the amputation of both of his legs below the knee and how this shaped his rock climbing and academic career. Herr also discusses orthoses and exoskeletons developed by his research group, as well as the future of bionic technology.

Transcript below.

Among topics discussed is the lower-body exoskeleton below. This exoskeleton is the first lower-body exoskeleton to decreases the user’s energy expenditure when walking (measured by oxygen consumption). It works by assisting the calf muscle and using the body’s joints, rather than heavy mechanical joints. Hugh Herr says, “Subjects noticed that their legs felt heavier and awkward when they took the exoskeleton off.”

Below is a TED talk Hugh Herr gave in 2014. In this talk, Herr describes disability as a failure of technology, he discusses the design of his own prostheses and his lab’s research, and there is a dance performance by Christian Lightner and Adrianne Haslet-Davis, who lost her lower left leg in the Boston Marathon bombings.

Hugh Herr

His research program seeks to advance technologies that promise to accelerate the merging of body and machine, including device architectures that resemble the body’s musculoskeletal design, actuator technologies that behave like muscle, and control methodologies that exploit principles of biological movement. His methods encompass a diverse set of scientific and technological disciplines, from the science of biomechanics and biological movement control to the design of biomedical devices for the treatment of human physical disability.

His research accomplishments in science and technology have already made a significant impact on physically challenged people. The Transfemoral Quasipassive Knee Prosthesis has been commercialized by Össur Inc., and is now benefiting amputees throughhttp://biomech.media.mit.edu/people-hugh-herr/out the world.  In 2006, he founded the company iWalk Inc. to commercialize the Powered Ankle-Foot Prosthesis and other bionic leg devices.  Professor Herr’s work impacts a number of academic communities. He has given numerous invited and plenary lectures at international conferences and colloquia, including the IVth World Congress of Biomechanics, the International Conference on Advanced Prosthetics, the National Assembly of Physical Medicine and Rehabilitation, World Economic Forum, Google Zeitgeist, Digital Life Design, and the TEDMED Conference. He is Associate Editor for the Journal of NeuroEngineering and Rehabilitation, and has served as a reviewer for the Journal of Experimental Biology, the International Journal of Robotics Research, IEEE Transactions on Biomedical Engineering, and the Proceedings of the Royal Society: Biological Sciences. He has been invited to participate in joint funding proposals from other universities and corporations, and has served on research review panels including the National Institute of Health, the National Institute on Disability and Rehabilitation, and the Department of Veterans Affairs. In 2007, He was presented with the 13th Annual Heinz Award for Technology, the Economy and Employment. His work has been featured by various national and international media, including Scientific American Frontiers, Technology Review, National Geographic, the History Channel, and CNN.

Links:

• Download mp3 (18.3 MB)
• Link to the Biomechatronics Group’s webpage
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS

The transcript is edited for clarity.

Hugh Herr:
My name is Hugh Herr and I’m a professor at MIT. I co-direct the Center for Extreme Bionics. I also am the founder of a company called BionX Medical Technologies.

Audrow Nash:
Can you tell me a bit about how you have lost your legs from the knee down?

Hugh Herr:
I was an avid mountain climber. I started mountain climbing when I was seven years old. When I was in my early teenage years I was considered a child prodigy in climbing. When I was seventeen I was out with my climbing partner at the time. Our goal was to climb Mt. Washington in the winter by ascending Huntington’s Ravine, which is about an eight hundred foot, very steep ice wall. We ascended very quickly the eight hundred foot wall, and then decided to continue towards the summit. The weather conditions got worse and worse. Even though we went just five minutes above the head-wall of Huntington Ravine, we were not able to retrace our tracks and we descended the mountain.

By the time we got to tree line it was clear that we had gone off course, but we were stuck at that point. We really didn’t have the rational choice of retracing our tracks and going towards the summit. At that point, wind speeds were so high that one could not even stand. We were forced to go down this ravine system, it’s called the Great Gulf Region. It’s the wilderness side of Mt. Washington. It was there that we spent several days in most extreme bushwhacking that one can imagine. The average depth of snow was to the waist. Sometimes it was to the chest. We were just simply trapped in a white maze.

We dug in when we were not able to move any more out of exhaustion by creating kind of these snow caves. During the day we would walk. We would probably make two miles of progress in a complete kind of marathon effort in the deep snow. We made it within a few miles of the roadway and then couldn’t walk anymore because of severe frostbite. We were later discovered, or found by a person out snowshoeing for the day, and we were plucked from the mountain via helicopter and then treated for severe frostbite and hypothermia.

Audrow Nash:
Then, so from there because of the severe frostbite you had both of your legs amputated from the knee down?

Hugh Herr:
That’s right. After a few months of effort by my medical team to “save” my biological limbs, they were amputated. We were on the mountain late January. My legs were amputated mid-March.

Audrow Nash:
Afterwards, you wanted to return to rock climbing. You were told otherwise, though.

Hugh Herr:
Yeah. I didn’t know what I would be able to do with my new body, and my father said, “If you want to climb, you should climb.” I really had no clear examples of what life would be using prostheses. My limbs are amputated. I went through about a month or two of healing, and then I was fitted with my first pair of artificial limbs. I was shocked at their lack of technological sophistication. They were actually made of plaster of Paris and it was recommended that I not walk without canes or crutches because the limbs might actually break. They were kind of these crazy trainer limbs. The first weekend I went home from the rehabilitation center they didn’t allow me to take my legs because they knew what I was capable of. The next weekend they were stupid enough to allow me to take my legs and I went climbing with my brother, Tony.

Audrow Nash:
How did you begin to adjust your prosthetic limbs for climbing and various purposes?

Hugh Herr:
I quickly abandoned the notion that the limbs needed to look human, and I quickly focused on function. I thought to myself what would be optimal designs for the vertical world of rock and ice climbing. What emerged was a series of different feet for rock and ice surfaces, so feet that could stand on small rock edges the width of a coin, feet that could wedge into rock fissures even where the human foot could not penetrate, feet that could penetrate ice walls. I made my legs exceedingly lightweight to increase my strength to weight ratio. Through design, I was able to etch out a few advantages. There were disadvantages, but I gained enough advantages that it was only twelve months after my limbs were amputated that I was climbing better with artificial limbs than I’d ever achieved with normal biological limbs before the accident.

Audrow Nash:
How did you transition into academia from this, and how did you know that that is the way you should pursue, or continue?

Hugh Herr:
To be honest, the men in my family, my father and brothers and grandfathers, were in the business of construction. Given how uncomfortable artificial limbs were at the time, I couldn’t imagine being on a construction site for the rest of my life. I also couldn’t imagine being on the construction site for the rest of my life. I decided to go to college, which I’d never plan to do. My intent was to be the best climber in the world before the accident. I went to college and I signed up for very basic math courses and computer science courses. I developed an extraordinary passion for the topics and couldn’t stop studying. I just absolutely loved the material. It’s interesting because science and a passion for science and math actually replaced somewhat my passion for climbing mountains.

Audrow Nash:
Then you began to apply what you were learning to your own prostheses with what you were learning? When did you start doing that?

Hugh Herr:
I did have the goal of continuing to design prostheses. The experience of designing new climbing legs and succeeding so wildly beyond everyone’s expectation, including my own, was very inspirational for me. I realized the power of technology to heal, to rehabilitate, even to extend human capability beyond innate capabilities. That was also a clear motivation in my acquisition of knowledge and going back to school. My first patent was obtained, issued very close to the time when I was graduating from undergraduate school in physics. It was on the mechanical interface between the residuum and the artificial limb. The idea was to use fluid bladders in certain configurations. I later proposed to my future MIT advisor that I would build this interface and have actuation pumps and pressure sensing and continually modulate the pressure field around the residuum.

Audrow Nash:
On the spot where the prosthetic limb interfaces with the residual limb is what you’re talking about?

Hugh Herr:
Yes.

Audrow Nash:
So you would use some sort of fluid to make a vacuum so that it would stay very well attached?

Hugh Herr:
Yes, a vacuum offering suspension, but also offering loading support.

Audrow Nash:
Loading support.

Hugh Herr:
A pressure field that would be very comfortable around the end of the limb, enabling a person to walk without discomfort.

Audrow Nash:
Can you tell me a bit about the prostheses that you’re wearing now?

Hugh Herr:
I’m wearing two bionic legs. Both legs have three microprocessors and twelve sensors. The microprocessors are brains, if you will. The devices control a muscle-like actuator that moves and stiffens and powers my synthetic ankles.

Audrow Nash:
What sensors are on board?

Hugh Herr:
The sensors measure position, speed, acceleration, force and temperature, as well as joint position and speed.

Audrow Nash:
What sensing are you doing, contact with the ground?

Hugh Herr:
No. The device has tendon-like series springs, and also a parallel elasticity. The torque sensing is of the torque that the series elastic actuator sees and the parallel spring sees.

Audrow Nash:
I see.

Hugh Herr:
Either one can produce net torque on a joint.

Audrow Nash:
Can you describe how they look a little bit?

Hugh Herr:
They look interesting. There’s two black batteries that sit on top, and there’s black anodized metal shields coming around the ankle. Then inside those shields are all the electronics and then the motor system. Then attached to that is a synthetic foot made of again, black carbon composite.

Audrow Nash:
What kind of grip on the bottom?

Hugh Herr:
I wear the carbon foot, and glued to that carbon foot is a typical rubber that you’d find on the bottom of a shoe.

Audrow Nash:
What kind of battery life?

Hugh Herr:
We get a few thousand fast walking steps. For me and my lifestyle, that’s fine to get through the day. If one wants to walk say, ten thousand steps, one would just take a spare battery or two. The batteries are modular. You just snap them in like a power tool.

Audrow Nash:
I see. How does this compare to other prostheses?

Hugh Herr:
All other foot/ankle prostheses in the world are human powered, meaning the energy of the human attached to them drives the movement. The legs that I’m wearing are bionic. They inject mechanical energy into a person’s stride in a way that’s similar to what the muscles used to do that were lost on amputation.

Audrow Nash:
Can you talk a bit about how they are fit to your residual limbs?

Hugh Herr:
At MIT I have a team of people thinking about fit. What is the nature of comfort? What are the mathematics of comfort? What is the science of comfort? We build mathematical descriptions of the tissues of the ends of the residuum and theories of how the shape of the structure that comes around the residuum and the stiffness of that shape to optimize comfort. Basically, the end of my limb goes into a cup-like device that supports my weight, and so the shape of that cup is critically important to comfort. We mathematically derive that shape, and then we 3D print the structure.

Audrow Nash:
You’re identifying where there are hard and soft spots in the residual limb so that you can match them or match them with their complement, so soft to hard, hard to soft so that it’s more comfortable?

Hugh Herr:
Yeah. Then what’s very complex is the shape, the equilibrium shape of the synthetic skin, if you will. That’s based on tissue, how compliant the tissues are, that shape. Mathematically we understand the compliance of the tissues in all the regions.

Audrow Nash:
What do you mean, compliance of the tissue?

Hugh Herr:
How soft or how stiff they are.

Audrow Nash:
Okay. What about stretch of say, the skin? Does that apply as well?

Hugh Herr:
In a different way. I wear a second skin silicone liner, so the design of that second skin relates to the skin’s strain field, or the amount of stretch in the skin.

Audrow Nash:
I see. Are they attached by this vacuum principle or how would you attach them to the residual limb?

Hugh Herr:
There’s a number of ways to attach. You can have a pin and a lock at the base of the limb. You can have suction. You can have a sleeve that spans from the biological leg across the socket. There’s a number of ways to hold it on.

Audrow Nash:
What have been some challenges in designing this?

Hugh Herr:
The interface. If you ask a thousand persons with limb amputation what’s the number one problem they want solved, probably all thousand would say the mechanical interface, the mechanical attachment. Why? Because it’s uncomfortable. It’s a very, very important problem. It’s a very complex problem because no one understands what comfort is for any device, or there’d be a shoe or a prosthesis. We’re developing that science, and it’s very exciting. We want to be able to produce sockets that are comfortable. We want to produce them fast at low cost.

Audrow Nash:
What are a means of doing this? You 3D print the part. How do you model the residual limb?

Hugh Herr:
We take an MRI image of the limb, and that tells us where the shape of the bones and where the skin is and where the muscles are. We then use robotic palpation tools to compress the tissues and measure displacement speeds and forces, which tell us the fundamental constants of the tissues. We then build a continuum mechanical model. If you have a mathematical description of the biological segment, you can then apply pressure fields. The model will tell you how the tissues deform. Then we define particular tissue deformations that are healthy and comfortable.

Audrow Nash:
Is the goal is to make this cheap and replicable for the general person?

Hugh Herr:
Yeah. My vision of the future is that each human individual will have a digital representation of their body. When they need any type of a bionic device, that digital body will be used to compute an optimal personalized bionic device for the human.

Audrow Nash:
This can refer to clothing as well.

Hugh Herr:
Correct. Clothing, shoes, bras, bike seats, bionic limbs, exoskeletons, neural implants that go inside the body. Everything will be personalized.

Audrow Nash:
Everything is optimized, and this relies heavily on 3D printing or additive manufacturing?

Hugh Herr:
3D printing is a tool. It’s not the dominant science. That is to say, one could build these interfaces with molding processes and not using digital fabrication. Digital fabrication is sometimes nice because it reduces the fabrication frequency, but it’s not necessary. It’s not the secret sauce.

Audrow Nash:
What’s the role of industry and academia in producing these prostheses?

Hugh Herr:
As stated, I’m a professor. The typical model of a professor in the US is they receive some type of grant monies and they conduct research with a team comprising both staff and students. There are inventions. Patents are filed. Patents are owned by the university. The university then licenses the intellectual property to founders of a company. Those founders can be the professors and students that invented, and often that’s the model, the most successful model of translation. In my case that’s exactly what happened. There was patents generated. I’m the inventor. My students are the inventors. A company’s established and then there’s a contractual relationship between the company and the university, licensing the IP within a certain domain of use. That’s what was done with my company, BionX Medical Technologies.

Audrow Nash:
What role does the company play in developing the technology?

Hugh Herr:
At university we do science. We do research that’s publishing. We test hypotheses. We do things that have never been done before, never been tested before. At the company, basically crude prototypes that we build at the university are commercialized. What that means is getting them light enough, most important strong enough, durable enough, manufacturable, at the right price point. Then all the sales and clinical services necessary to distribute that product globally.

Audrow Nash:
What does it look like when someone puts on the BiOM ankle for the first time?

Hugh Herr:
The design in the ankle is biomimetic, inspired by nature, by the human body. The algorithms that are running on these small computers on the ankles control the motor as if the motor were made of muscles and tendon and the whole structure was bone and whatnot. The whole thing moves as if it was made of flesh and bone, even though it’s made of synthetics. Because of that, when it’s fit to a human, the human is used to those dynamics, so there’s little to no training required. Often in minutes the person is saying things like they have their life back, they have their leg back, and they’re either crying out of happiness or laughing out of happiness. It’s fun to go to these fittings, actually.

Audrow Nash:
I’d like to move into talking about exoskeletons. First, would you define what an exoskeleton is and then tell us how it relates to your work with prostheses.

Hugh Herr:
My definition of an exoskeleton is a device that attaches to the body, intimately to the body, that augments physicality. What I mean by augments, it enables a human being to do something that’s beyond natural capability for an innate, healthy body. The word orthoses is a medical term for a robot that attaches to the body that enables a person with a disability or some condition to move more naturally. Exoskeletons are wearable robots that augment.

Audrow Nash:
Can you tell me a bit about metabolic cost as a parameter for designing exoskeletons?

Hugh Herr:
One goal in exoskeletal design is to augment a human by pedal locomotion. An excellent metric of evaluation is the amount of food energy a person requires when using the exoskeleton versus not using the exoskeleton. Why? My view is that if the exoskeleton increases energy levels of the human, the human won’t want to use it. It won’t have sufficient value. It’s a very important metric of evaluation.

Audrow Nash:
Basically they’ll discard it and not use it if the metabolic cost is higher than the task, or higher than it would be to accomplish otherwise?

Hugh Herr:
You know, unless they want to get exercise. If they want to actually be augmented and enhance physicality, no they won’t use it.

Audrow Nash:
How do you measure this?

Hugh Herr:
Typically for aerobic exercise it’s measured by monitoring how much oxygen a person uses when they breathe and how much carbon dioxide is expelled. With those two rates one can compute the calories burned per unit of time.

Audrow Nash:
Would you describe the exoskeleton used to decrease metabolic costs for walking?

Hugh Herr:
The first exoskeleton that was envisioned and published in the world that I’m aware of was in the nineteenth century, over a hundred years ago. It was a Russian inventor called Jagen. His dream was to augment the Russian Army, Russian soldiers. It wasn’t until 2014 that someone succeeded in building an exoskeleton that augments walking and running. My group was the first to accomplish that goal. It’s an exoskeleton that spans the foot and comes up about mid-calf, just below the knee. Fundamentally it’s an artificial calf muscle. The calf muscle is the most important muscle in bi-pedal walking. It supplies about eighty percent of the power to walk. It’s where humans are most inefficient, so the exoskeleton adds an artificial calf muscle and injects energy into the gait like a calf muscle to reduce the metabolic cost of the biological calf.

Audrow Nash:
What does it look like and what kind of actuators are you using?

Hugh Herr:
The actuators are electric based. It looks very interesting. It’s like, this crazy structure, this crazy shoe that goes higher than is normal and has these springy struts that go up on the side of the leg, and then a motor that sits on kind of a shin guard that pulls on that strut and powers movement.

Audrow Nash:
Yes, and so it’s using some sort of tether and you’re pulling that, and that’s how you flex the calf to augment walking?

Hugh Herr:
Flex the ankle, yes.

Audrow Nash:
You’re using the body as its own joint for this. Correct?

Hugh Herr:
It’s very important that exoskeletons are very, very, very light weight. Whenever you add weight to the legs, metabolic cost increases. In an effort to reduce the weight or the mass of the device, we actually use the body’s joints as the bearing instead of putting a synthetic bearing in the device.

Audrow Nash:
What kind of sensing is done with the device?

Hugh Herr:
All the electronics are up high by design so you can walk through a deep, deep puddle of water and not affect the system. We measure positions and speeds and accelerations, essentially. Then from those data we can compute torques, forces, given a model of the structure.

Audrow Nash:
You can infer where the person is in their gait and then determine if you should pull to assist the gap?

Hugh Herr:
Essentially.

Audrow Nash:
Why did you choose to revolve around the ankle?

Hugh Herr:
Again because the calf muscle is the most important muscle in walking.

Audrow Nash:
Are there other advantages to it as well, being that there’s not so much squishy …

Hugh Herr:
Yeah. The shin bone is where the leg is most stiff, most rigid, and it’s a tremendous plate of bone to apply forces on from an exoskeleton.

Audrow Nash:
Can you only use this for walking? Could someone use it for running or other activities?

Hugh Herr:
The device is very versatile. It’s powered, it has smart computation, so yes you can walk, run up and down hills, whatever you want to do. It’s as versatile as the human leg itself.

Audrow Nash:
What kind of efficiency gains are we getting?

Hugh Herr:
There’s a number of peer-reviewed manuscripts. Again, we were the first in 2014 to augment the human in a peer review publication. That first publication, the human subjects were wearing a twenty-three kilogram backpack and we augmented by eight percent. We later conducted a separate study on unloaded humans, just regular walking, and augmented by an average ten percent. The variation is huge, so you have N subjects. Some subjects are as high as almost thirty percent, some are low, and everything in between.

Audrow Nash:
What do you believe creates that variation?

Hugh Herr:
It’s a really interesting question. No one really knows, but everyone in the field has observed this phenomenon, that when you put a single exoskeletal design on N people, you get huge variation. My belief is you take ten people and you could distribute them in terms of athletic capability. Some people have a very smart spinal cord and they can exploit any tool or device very, very quickly. Others are clueless. That’s a possibility where training may be effective to mitigate that variance.

Audrow Nash:
Now, people that use it, what do they say it feels like?

Hugh Herr:
The human body is extraordinary, so a person begins to use it, they very quickly … Initially they feel the energy that it provides, but very quickly the human body gets used to it. Then you take it off and you kind of stumble, and your own biological normal legs feel heavy, awkward and slow. This tells us that wearing bionic structures will just be commonplace in the future, and innate, normal bodies will just be boring.

Audrow Nash:
Now, what kind of upper limit do you think we can get on efficiency with this device?

Hugh Herr:
With the foot/ankle device it’s not known. I would say at least twenty-five percent, but it may be even higher, which is very, very exciting.

Audrow Nash:
Then will you benefit, you think, from augmenting other muscles or other movements?

Hugh Herr:
We don’t know. What’s intriguing is our recent paper on the device. What’s intriguing is when you add exoskeletal torque and power at the ankle, not only do you reduce the power requirements of the biological ankle, but also you reduce the power requirements of the knee and hip. Even though the exoskeleton doesn’t span the knee and hip we nonetheless, reduce muscular effort at those joints tremendously. It begs the question do you really need an exoskeleton that spans the whole leg, or is the foot/ankle sufficient?

Audrow Nash:
Why would you think that it decreases the expenditure at those muscles?

Hugh Herr:
It’s just the dynamics of the system. The body, when you give it power, it uses that power distributively across the entire leg in an optimal way. It’s a new gait. It’s a different gait … than normal.

Audrow Nash:
How does it look compared to normal?

Hugh Herr:
It looks normal to an untrained eye. It doesn’t look like a bizarre walk. It is a distinctly new gait. You give the human energy from a bionic structure and the body quickly figures out how to ultimately use that energy to minimize its own metabolic food energy. That’s what humans are really good at. We’re inherently lazy and we minimize energy really, really well. We’re terrible at detecting high stress levels and probabilities of injury. We injure our knee and we’re surprised. We injure our hip and we’re surprised. An incredible value of exoskeletons is to give the human something that they’re bad at. One class of exoskeletons in the future will be exoskeletons that enable us to do athletic performances without a threat of injury.

Audrow Nash:
How would those work?

Hugh Herr:
I don’t know, but they would have to detect high levels of stress and fatigue in biological structures and then tell the human to stop or to move in a different way. Imagine a world where our top athletes never injure. Imagine the impact to human athletic performance. Performance would just go through the roof, because often what mitigates top performance is injury and recovering from injury and not training harder because you might injure, yada, yada, yada.

Audrow Nash:
What other future applications of exoskeletons do you see?

Hugh Herr:
Another form of augmentation is maintaining one’s inherent physicality independent of age. Imagine a world when you run your best marathon time when you’re sixty, and you’ve been running marathons since the age of fifteen. Why? Because you’re a more skilled athlete. As you age, technology eliminates the impact of age related degeneration by neural implants, maybe by regenerative medicine, maybe by exoskeletons or various forms of augmentation. That’s a very palpable form of augmentation. Most people ethically would be fine with that and most people would be excited about maintaining quality of life as they age.

Audrow Nash:
What are some bottlenecks in exoskeletons? What technological bottlenecks are there?

Hugh Herr:
I’m not sure there are technological bottlenecks. I think that largely the components are sufficient. It’s really a problem with design and architecture. The human machine interaction is something that’s unknown, so putting a bionic structure on a human and adapting its control in real time in a kind of human machine optimization is critically important to the future of this area of design. Imagine putting on a device and you just starting to walk and run and it adjusts its behavior to optimize its performance with you in a collaborative effort.

Audrow Nash:
Is this one of your research interests moving forward?

Hugh Herr:
Absolutely.

Audrow Nash:
What are some of your near-term research goals?

Hugh Herr:
I want to do three things first and foremost. One is I want to advance better muscle-like actuators. Perhaps actuators that are better than biological ones, kind of the engine of BionX. I want to understand electrical interface between the peripheral human nervous system and devices, so how to talk to nerve endings, essentially. I also want to understand how to attach machines to the body mechanically in a comfortable way. Those three innovations will really define bionics in this century. If you solve all three, you’re more or less done. You’ve solved bionics, limb bionics that is.

Audrow Nash:
Thanks for being on Robots Podcast.

Hugh Herr:
Thank you.

In this episode, Audrow Nash interviews Emo Todorov, Director of Movement Control Laboratory at the University of Washington, about a physics-based optimization method for controlling robots. Todorov describes how his physics-based method can be used to solve problems and discusses results in simulation and on hardware.

Emo Todorov

Emo Todorov obtained his PhD in Cognitive Neuroscience from MIT in 1998. Since then he has worked as Postdoctoral Felow in the Gatsby Computational Neuroscience Unit at UCL, Research Scientist in Biomedical Engineering at USC, Assistant Professor in Cognitive Science at UCSD, and is now Associate Professor in Applied Mathematics and Computer Science & Engineering at UW. He is generally interested in intelligent control of complex systems, in both engineering and biology. His current focus is autonomous robot control through high-performance numerical optimization. He is also the founder of Roboti LLC and developer of the MuJoCo physics simulator.

Links:

• Download mp3 (15.8 MB)
• Todorov’s lab website
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS

In this episode, Audrow Nash interviews Karl Iagnemma, a Principal Research Scientist at the Massachusetts Institute of Technology and the CEO of nuTonomy, about autonomous vehicles in urban environments. Iagnemma discusses the market for autonomous cars, why nuTonomy is being developed and, at least initially, deployed in Singapore, and the technology of autonomous cars.

A video of nuTonomy’s autonomous vehicle is below.

Karl Iagnemma

Karl Iagnemma is a principal research scientist in the Mechanical Engineering department at the Massachusetts Institute of Technology, where he directs the Robotic Mobility Group. He holds a B.S. from the University of Michigan, and an M.S. and Ph.D. from MIT, where he was a National Science Foundation Graduate Fellow. He has performed postdoctoral research at MIT, and has been a visiting researcher at the NASA Jet Propulsion Laboratory and the National Technical University of Athens (Greece). Karl is also co- founder and CEO of nuTonomy, a venture-backed startup company focused on developing software for fully autonomous passenger vehicles.

Links:

• Download mp3 (17.6 MB)
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS

Most of D’Andrea’s work shown at this latest demonstration is dedicated to pushing the boundary of what can be achieved with autonomous flight. One key ingredient is localization: To function autonomously, robots need to know where they are in space. Previously his team has relied on an external high-precision motion capture system available in his Flying Machine Arena at ETH Zurich for positioning. In his previous TED talk on robot athletes for example, you can clearly see the reflective markers required by the motion capture system. This meant that most algorithms were difficult to demonstrate outside the lab. It also meant that the system had a single point of failure (SPOF) in the centralized server of the mocap system — a highly problematic point for any safety critical system.

In another world first, Raff and his team are now showing a newly developed, doubly redundant localization technology from Verity Studios, a spin-off from his lab, which gives flying machines, and robots in general, new levels of autonomy. For the live demonstrations, all flying machines use on-board sensors to determine where they are in space and on-board computation to determine what their actions should be. There are no external cameras. The only remote commands the flying robots receive are high level ones, such as “take-off” or “land”.

The nature of this performance was also unprecedented in that it did not use safety nets to separate the audience from the action, and in that it even saw dozens of vehicles fly directly above the audience.

The performance included demonstrations of various safety systems for flying machines, including a large, high-performance quadcopter that is fully redundant and that utilizes a state-of-the-art failsafe algorithm (see previous Robohub article) among its many other onboard and offboard safety features.

Following extensive technical discussions on the various failure modes of the positioning system and of each flying machine as well as on the systems’ numerous redundancies and safety features, the organizers of TED had full faith in the systems’ safety and reliability. So much so that they decided to indemnify the venue against any lawsuits to overcome its ban on drones.

Here is the video:

And here is a more in-depth breakdown of what you are seeing, based on a transcript of the presentation and information from Raff D’Andrea’s team:

Intro video – Institute for Dynamic Systems and Control (IDSC), ETH Zurich

The video shows some of the previous work of D’Andrea’s group on aerial construction: A 6-meter tall tower built out of 1500 foam bricks by four autonomous quadcopters over a three day period in front of a live audience and a rope bridge built by three autonomous quadcopters in the Flying Machine Arena.

Live demo 1: Tail-Sitter – IDSC, ETH Zurich

The Tail-sitter. Photo: Bret Hartman / TED

“This is a so-called tail-sitter. It’s an aircraft that tries to have its cake and eat it: Like other fixed wing aircraft, it is efficient in forward flight, much more so than helicopters and variations thereof. Unlike most other fixed wing aircraft, however, it can hover, which has huge advantages for take-off, landing, and general versatility. There is no free lunch, unfortunately, and one of the challenges with this type of aircraft is its susceptibility to disturbances, such as wind-gusts, while hovering. We are developing control architectures and algorithms that address this limitation. [Pushes the vehicle, then grabs it and throws it] The idea is to allow the aircraft to autonomously recover no matter what situation it finds itself in. [Second throw] And through practice, improve its performance over time. [Third throw, violently spinning the vehicle]”

Demo 2 (video of demo the night before): Monospinner – IDSC, ETH Zurich

The Monospinner. Photo: Marla Aufmuth / TED

“When doing research we often ask ourselves abstract, fundamental questions that try to get at the heart of a matter. One such question: What is the minimum number of moving parts necessary for controlled flight? This line of exploration may have practical ramifications; take helicopters, for example, which are affectionately known as “machines with over 1000 moving parts all conspiring to cause bodily harm”. Turns out that decades ago skilled pilots were flying RC airplanes with only 2 moving parts: a propeller and a tail rudder. We recently discovered that it could be done with only 1.”

“This is the monospinner, the world’s mechanically simplest, controllable flying machine, invented just a few months ago. It has only one moving part, a propeller. There are no flaps or hinges, no control surfaces or valves, no other actuators. Just one propeller. Even though it is mechanically simple, there is a lot going on in its electronic brain to keep it stable and move through space in a controllable fashion. Even so, it does not yet have the recovery algorithms of the tail-sitter, which is why I have to throw it just right.”

Live demo 3: Omnicopter – IDSC, ETH Zurich:

The Omnicopter. Photo: Bret Hartman / TED

“If the monospinner is an exercise in frugality, this machine here, the omnicopter, with its 8 propellers, is an exercise in excess. What can you do with this surplus? The thing to notice is that it is highly symmetric, and as a result it is ambivalent to orientation. This gives it an unprecedented capability: It can move anywhere in space, independently of where it is facing, and even of how it is rotating. It has its own complexities, mainly due to the complex, interacting flows of its 8 propellers. Some of this can be modeled, while the rest can be learned on the fly. [Omnicopter takes off, manoeuvres around the stage, lands]”

Live demo 4: Fully redundant multicopter – Verity Studios

Fully redundant quadcopter. Photo: Ryan Lash / TED

“If flying machines are going to find their way into our daily lives, they will need to become extremely safe. This machine here is a prototype being developed by Verity Studios. It actually consists of two separate, two propeller flying machines, each capable of controlled flight. One wants to rotate clockwise, the other one counter clockwise. When all systems are operational, it behaves like a high-performance, 4 propeller quadrocopter. [Quadcopter takes off and flies an arc across the stage] If anything fails, however – propeller, motor, electronics, even a battery pack – it is still able to fly, albeit in a degraded way [One half of the quad is de-activated, the quadrocopter starts to spin and fall, the failsafe mode kicks in and the quadrocopter recovers, flies to the back of the stage, and performs a controlled landing at the takeoff spot].”

Live demo 5: Synthetic swarm, 33 small flying machines – Verity Studios

Photo: Bret Hartman / TED

“This last demonstration is an exploration of synthetic swarms. The large number of coordinated autonomous entities offers a radically new palette for aesthetic expression. We have taken commercially available micro-quadrocopters, each weighing less than a slice of bread, and equipped them with our localization technology and custom algorithms. Because each unit knows where it is in space and is self-controlled, there really is no limit to their number. [Swarm performance]”

Raffaello D’Andrea wrapped up with: “Hopefully these demonstrations will motivate you to dream up new revolutionary roles for flying machines. For example, the ultra-safe flying machine [points to the fully redundant multicopter] has aspirations to become a flying lampshade on Broadway. The reality is that it is difficult to predict the impact of nascent technology. And for folks like us, the real reward is the journey, and act of creation. It is also a continual reminder of how wonderful and magical the universe we live in is; that it permits clever, creative creatures to sculpt it in such spectacular ways. The fact that this technology has huge commercial potential is just icing on the cake.”

More photos for your viewing pleasure:

Photo: Bret Hartman / Ryan Lash / TED

Photo: Bret Hartman / Ryan Lash / TED

Photo: Bret Hartman / Ryan Lash / TED

Photo: Steven Rosenbaum / Waywire

Myon walking with human support. Source: Neurorobotics Research Lab

The idea of connecting brain-inspired models of computation to robots is probably as old as the discipline of robotics itself; as far back as 1950, neurophysiologist William Grey Walter had already conceived of an autonomous mobile platform that was controlled by brain-inspired analog circuitry[1]. Today, researchers are connecting robotics with neuroscience in order to both build intelligent robots and to better understand the brain. The workshop Advances in Biologically Inspired Brain-Like Cognition and Control for Learning Robots at IROS (Hamburg) brought together experts from diverse fields in brain-based robotics, neurorobotics, artificial neural networks and machine learning to discuss the state of the art.

Learning About Animal Locomotion Control using Robots

Salamander CPG model tested with a amphibious salamander-like robot (Ijspeert et al., 2007).

In contrast to studying animal locomotion though simulation, biorobots allow researchers to observe physical interaction with the real world, while at the same time avoiding the ethical issues of animal experimentation. In the first talk, Auke Ijspeert from EPFL in Switzerland gave an introduction to his research on animal locomotion and biorobotics, and spoke about the advantages of using robots as an experimental tool.

After introducing the four essential ingredients of animal motor control (descending modulation from higher brain areas, reflexes, the musculoskeletal system, and central pattern generators – CPGs) Ijspeert’s talk focused on his group’s efforts to implement walking and swimming in a salamander-like robot using CPGs. CPGs are assumed to be neural circuits that are implemented in the spinal cord of vertebrates, and while there is no clear experimental evidence, researchers in the field suspect that they are responsible for implementing motion primitives that are controlled via modulation signals from the brain. In the salamander robot, this modulation enabled a CPG modeled as a system of coupled oscillators to switch between walking and swimming locomotion.

Further studies on the salamander robot include the incorporation of sensory feedback, which considerably improves performance in the presence of external perturbations. Ijspeert concluded by presenting recent research that showed how adding a CPG to a sensory-driven model of human bipedal locomotion can simplify the control of walking speed.

Employing Neural and Physical Dynamics for Robot Control

The second major theme of the workshop featured talks from three invited speakers. Manfred Hild from the Neurorobotics Research Laboratory at Beuth Hochschule für Technik Berlin, Germany presented on Self-Exploration of Autonomous Robots Using Attractor-Based Behavior Control. Rather than developing explicit models for his Myon [2] and Semni [3] robots, the goal of Hild’s research is to acquire implicit body representations through self-exploration.

The information gathered during this process is stored in sensorimotor manifolds that allow for behavior control. Learning these manifolds through naïve motor babbling is infeasible due to the complexity and size of the resulting data. In Hild’s approach, all joints are controlled independently by relatively simple cognitive sensorimotor loops. The embodied interplay of these individual loops enables attractor-based behavior control through transitions between different attractor states of the system formed by the robot and the environment. By varying the parameters of the individual cognitive sensorimotor loops, it is possible to learn sensorimotor manifolds in a highly efficient manner.

By contrast, Joni Dambre from Reservoir Computing Lab at Ghent University in Belgium leverages dynamical systems for robot control. In her talk Practical approaches to exploiting body dynamics in robot motor control, she provided an overview of her group’s research on applications of the reservoir computing paradigm in robotics. In particular, the focus was on low-cost compliant robots whose dynamics are not analytically tractable and therefore cannot be controlled using standard methods from classical robotics. The idea of reservoir computing was originally conceived independently by Jaeger[4] and Maas[5] in 2001 and 2002, respectively. In the original publications, the reservoir was modeled as a randomly generated recurrent network of artificial neurons or spiking neurons. Rather than training all connections of the network individually, only a small set of linear readout units is adapted with a simple learning algorithm-like linear regression. The internal weights of the complex reservoir network remain constant. Based on this concept, Dambre’s group has developed a method for embedding complex tunable CPGs into neural networks as well as an approach for creating inverse models using reservoir computing. In another study, the reservoir network was replaced by the physical dynamics of a tensegrity robot, which was then able to learn CPGs. Combining reservoir computing with morphological computation therefore provides an easy means of obtaining embodiment. If the body dynamics are not rich enough for the desired control task, it is possible to couple them with a neural reservoir. As an example, Dambre presented new results on learning to walk with the Oncilla robot.

The last presentation on neural dynamics was given by Herbert Jaeger from the MINDS group at Jacobs University in Bremen, Germany. In his talk Generating and Modulating Complex Motion Patterns with Recurrent Neural Networks and Conceptors, Jaeger (one of the inventors of the reservoir computing approach) outlined a novel computational principle called conceptors. Conceptors are an extension to Jaeger’s echo state network approach to reservoir computing and enable a recurrent network of artificial neurons to store and reproduce dynamical patterns. For each pattern stored in the neural network, a corresponding conceptor is generated. By later inserting this conceptor as a neural filter into the state update loop of the reservoir network, the corresponding pattern can be retrieved. Since a single reservoir can store several different motion patterns at the same time, it is also possible to morph patterns by linearly blending between different conceptors. This was demonstrated using the example of recorded human motion patterns.

Training and Understanding Deep Neural Networks

The recent successes of deep learning techniques have brought new momentum to the field of machine learning, and have heralded a renaissance in artificial neural networks. Deep neural networks can achieve superhuman performance in traffic sign recognition[6], learn to play computer games[7], and automatically generate appropriate responses to e-mail messages[8] — rendering them a highly interesting tool for commercial applications ranging from autonomous driving to automated image tagging. It is thus no surprise that companies like Facebook, Google and Microsoft have founded their own deep learning research laboratories.

In his talk on Training and Understanding Deep Neural Networks for Robotics, Design, and Perception, Jason Yosinski from the Creative Machines Lab at Cornell University shed light on some quite different applications and properties of deep neural networks. At the beginning, he demonstrated his work on central pattern producing networks (CPPNs) for generating 3D models. The shape of the generated forms is controlled by an evolutionary algorithm. On endlessforms.com, users can try out this algorithm live by selecting their preferred forms as a starting point and see how they evolve over different generations. Yosinski has also applied CPPNs for evolving gaits on the four-legged robot, Aracna.

The second part of Yosinksi’s talks was dedicated to methods for visualizing and understanding the inner workings of deep neural networks like AlexNet[9]. He demonstrated how CPPNs can be used to generate artificial images that are falsely recognized with high confidence by deep convolutional neural networks. This approach was used in the deep visualization toolbox[10], an open source tool for visualization and analysis of deep neural networks.

The Neurorobotics Platform of the Human Brain Project

In the last talk of the workshop, Stefan Ulbrich from FZI Forschungszentrum Informatik in Karlsruhe, Germany presented The Neurorobotics Platform[11] of the Human Brain Project[12]. This platform is developed within the Neurorobotics Subproject of the Human Brain Project and will provide both roboticists and neuroscientists with tools to connect robot models to large-scale brain simulations running on supercomputers or neuromorphic hardware. All functionality is made accessible via a modern web-based user interface that displays a live visualization of the simulated robot, its environment, and live data such as brain activity or robot state parameters. A separate closed-loop engine connects and synchronizes the robot and brain simulation components. The data exchanged between brain and robot is processed by user-definable transfer functions. Dedicated designers will enable users to define their own robots, environments and experiments. Alternatively, ready-to-run setups will be made available in a catalogue.

After outlining the roadmap for developing new features for the Neurobotics platform, Ulbrich closed his presentation with a video showing the platform in action. Interested users will be able to access and try out the Neurorobotics platform after its public release in April 2016.

https://www.youtube.com/watch?v=b_RjzdlN0y4

Poster Session

The invited talks were complemented by a poster session that presented range of topics, including a review of the role of time in neurobiological learning algorithms, an approach to cognitive walking, and a neural architecture for autonomous task recognition, learning and execution. Topics with a special focus on neurorobotics included the cerebellar control of a compliant robot arm and select experiments from the Neurorobotics Platform.

Final thoughts

The variety of different topics in the workshop and diverse backgrounds of the invited speakers demonstrates that there is a lot of research being done in machine learning and biologically inspired cognition, and much of it goes beyond standard deep convolutional neural networks. Instead of asking ‘which of the proposed methods will outperform all of its competitors,’ the question should rather be ‘how can these methods and their individual strengths be combined to build better robots?’ This workshop was a first important step on this way.

To foster future collaboration, a call for contributions to a special issue of the journal Frontiers in Neurorobotics on the topic of the workshop will soon be announced. Further information about the workshop and abstracts of all talks and posters are available on the www.neurorobotics.net.

[1] http://www.nature.com/scientificamerican/journal/v182/n5/pdf/scientificamerican0550-42.pdf

[2] http://www.neurorobotik.de/robots/myon_de.php

[3] http://www.neurorobotik.de/robots_de.php

[4] H. Jaeger (2001): The “echo state” approach to analysing and training recurrent neural networks. GMD Repo

[5] Maass, W., Natschläger, T., & Markram, H. (2002). Real-time computing without stable states: A new framework for neural computation based on perturbations. Neural computation, 14(11), 2531-2560.

[6] http://people.idsia.ch/~juergen/superhumanpatternrecognition.html

[7] http://arxiv.org/abs/1312.5602

[8] http://googleresearch.blogspot.ch/2015/11/computer-respond-to-this-email.html

[9] Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems (pp. 1097-1105).

[10] http://yosinski.com/deepvis

[11] http://www.neurorobotics.net

[12] http://www.humanbrainproject.eu

In a forest simulation, algorithms segment space into “obstacle-free regions” and then link them together to find a single collision-free route.
Photo courtesy of CSAIL researchers.

Adam Conner-Simons | CSAIL

Getting drones to fly around without hitting things is no small task. Obstacle-detection and motion-planning are two of computer science’s trickiest challenges, because of the complexity involved in creating real-time flight plans that avoid obstacles and handle surprises like wind and weather. In a pair of projects announced this week, researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) demonstrated software that allow drones to stop on a dime to make hairpin movements over, under, and around some 26 distinct obstacles in a simulated “forest.”

One team’s video shows a small quadrotor doing donuts and figure-eights through an obstacle course of strings and PVC pipes. Weighing just over an ounce and clocking in at 3 and a half inches from rotor to rotor, the drone can fly through the 10-square-foot space at speeds upwards of 1 meter per second.

The team’s algorithms — which are available online and were previously used to plan footsteps for CSAIL’s Atlas robot at last year’s DARPA Robotics Challenge — segment space into “obstacle-free regions” and then link them together to find a single collision-free route.

“Rather than plan paths based on the number of obstacles in the environment, it’s much more manageable to look at the inverse: the segments of space that are ‘free’ for the drone to travel through,” says recent graduate Benoit Landry ’14 MNG ’15, who was first author on a related paper just accepted to the IEEE International Conference on Robotics and Automation (ICRA). “Using free-space segments is a more ‘glass-half-full’ approach that works far better for drones in small, cluttered spaces.”

In a second CSAIL project, PhD student Anirudha Majumdar showed off a fixed-wing plane that is guaranteed to avoid obstacles without any advanced knowledge of the space, and even in the face of wind gusts and other dynamics. His approach was to pre-program a library of dozens of distinct “funnels” that represent the worst-case behavior of the system, calculated via a rigorous verification algorithm.

“As the drone flies, it continuously searches through the library to stitch together a series of paths that are computationally guaranteed to avoid obstacles,” says Majumdar, who was lead author on a related technical report. “Many of the individual funnels will not be collision-free, but with a large-enough library you can be certain that your route will be clear.”

Both papers were co-authored by MIT professor Russ Tedrake; the ICRA paper, which will be presented in May in Sweden, was also co-written by PhD students Robin Deits and Peter R. Florence.

Drones in high-density

A bird might make it seem simple, but flight is a highly complicated endeavor. A flying object can change position in six distinct directions — forward/backward (“surge”), up/down (“heave”), left/right (“sway”), and by rotating front-to-back (“pitch”), side-to-side (“roll”), and horizontally (“yaw”).

“At every moment in time there are 12 distinct numbers needed to describe where the system it is and how quickly it is moving, on top of simultaneously tracking other objects in the space that could get in your way,” says Majumdar. “Most techniques typically can’t handle this sort of complexity in real-time.”

One common motion-planning approach is to sample the whole space through algorithms like the “rapidly-exploring random tree.” Although often effective, sampling-based approaches are generally less efficient and have trouble navigating small gaps between obstacles.

Landry’s team opted to use Deits’ new free-space-based technique, which he calls the “Iterative Regional Inflation by semidefinite programming” algorithm (IRIS). They then coupled IRIS with a “mixed-integer semidefinite program” (MISDP) that assigns specific flight movements to each “space-free region” and then executes the full plan.

To sense its surroundings, the drone used motion-capture optical sensors and an on-board inertial measurement unit (IMU) that help estimate the precise positioning of obstacles.

“I’m most impressed by the team’s ingenious technique of combining on- and off-board sensors to determine the drone’s location,” says Jingjin Yu, an assistant professor of computer science at Rutgers University. “This is key to the system’s ability to create unique routes for each set of obstacles.”

In its current form, MISDP has been optimized such that it can’t do real-time planning; it takes an average of 10 minutes to create a route for the obstacle course. But Landry says that making certain sacrifices would let them generate plans much more quickly.

“For example, you could define ‘free-space regions’ more broadly as links between areas where two or more free-space regions overlap,” says Landry. “That would let you solve for a general motion-plan through those links, and then fill in the details with specific paths inside of the chosen regions. Currently we solve both problems at the same time to lower energy consumption, but if we wanted to run plans faster that would be a good option.”

Majumdar’s software, meanwhile, generates more conservative plans, but can do so in real-time. He first developed a library of 40 to 50 trajectories that are each given an outer bound that the drone is guaranteed to remain within. These bounds can be visualized as ”funnels” that the planning algorithm chooses between to stitch together a sequence of steps that allow the drone to plan its flying on the fly.

A flexible approach like this comes with a high level of guarantees that the software will work, even in the face of uncertainties with both the surroundings and the hardware itself. The algorithm can easily be extended to drones of different sizes and payloads, as well as ground vehicles and walking robots.

As for the environment, imagine the drone choosing between making a forceful roll maneuver that will avoid a tree by a large margin, versus flying straight and avoiding a tree by a small amount.

“A traditional approach might prefer the first since avoiding obstacles by a significant amount seems ‘safer,’” Majumdar says. “But a move like that actually may be riskier because it’s more susceptible to wind gusts. Our method makes these decisions in real-time, which is critical if we want drones to move out of the labs and operate in real-world scenarios.”

A clear path to avoiding obstacles

In a related project, a planner searches a library of pre-computed funnels and selects one that doesn’t intersect with the obstacles.
Photo courtesy of CSAIL researchers.

CSAIL researchers have been working on this problem for many years. Professor Nick Roy has been honing algorithms for drones to develop maps and avoid objects in real-time; in November a team led by PhD student Andrew Barry published a video demonstrating algorithms that allow a drone to dart between trees at speeds of 30 miles per hour.

While these two drones cannot travel quite as fast as Barry’s, their maneuvers are generally more complex, meaning that they can navigate in smaller, denser environments.

“Enabling dynamic flight of small, off-the-shelf quadcopters is a marvelous achievement, and one that has many potential applications,” Yu says. “With additional development, I can imagine these machines being used as probes in hard-to-reach places, from exploring caves to doing search-and-rescue in collapsed buildings.”

Landry, who now works for 3D Robotics in California, is hopeful that other academics will build on and refine the researchers’ work, which is all open-source and available on github.

“A big challenge for industry is determining which technologies are actually mature enough to use in real products,” Landry says. “The best way to do that is to conduct experiments that focus on all of the corner cases and can demonstrate that algorithms like these will actually work 99.999 percent of the time.”

Landry’s work was partially supported by the Siebel Scholars Foundation, while Majumdar’s work was supported by a grant from the Office of Naval Research.

Papers

• Aggressive quadrotor flight through cluttered environments using mixed integer programming”
• Funnel Libraries for Real-Time Robust Feedback Motion Planning

If you liked this post, you may also be interested in:

• “Shrinking bull’s-eye” algorithm speeds up complex modeling from days to hours
• The Master Algorithm, with UW’s Pedro Domingos
•  Quadrocopter failsafe algorithm: Recovery after propeller loss
• ‘Tailsitter’ flying robot hovers and recovers easily thanks to new algorithm
• Robots Podcast #194: Embodied quadrotors, with Davide Scaramuzza

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

Design process

As a starting point we derived a physical model of the flying machine, which helped us to understand the behavior of the system. For instance, as the electric ducted fans are placed farther away from the center of the vehicle, the lever arm of the thrust vectoring system increases – but so does the inertia of the system. These are conflicting requirements for high maneuverability, where one seeks to maximize the lever arm, but still keep the inertia small. The physical model helped us to find a suitable tradeoff.

In parallel, the electric ducted fan and the thrust vectoring mechanism were characterized quantitatively by doing measurements on a separate test bench. In this way, the properties of the actuation – e.g. the total thrust, the time constants of the fan, and the thrust vectoring mechanism – could all be determined experimentally.

Figure 1: The test bench for characterizing the electric ducted fan and the thrust vectoring system.

The measurements on the test bench were used to update the physical model of the flying machine and to assess its flight performance in simulations.

Figure 2: Electric ducted fan with exit nozzle and control flaps.

Key Facts

Flying Platform Website

Researchers
Michael Muehlebach
Raffaello D’Andrea

Mechanical and electrical design
Michael Egli
Marc-Andrè Corzillius

Status
Ongoing research project

Hardware design and specifications

The analysis based on the physical model was used to determine the specifications of the hardware design. The three electric ducted fans (Schübeler DS-51-DIA-HST) were aligned on an equilateral triangle of 30cm side length. A carbon fiber honeycomb structure was used to connect the fans, and a 3D printed exit nozzle was mounted on each fan (see Fig. 2) in order to increase the airflow velocity and augment the thrust vectoring capabilities. The exit nozzle also integrated the two servos that actuate the control flaps. The flaps were optimized for a high stall angle using the XFoil software package (see https://en.wikipedia.org/wiki/XFOIL). Three aluminum feet support the structure when the Flying Platform is on the ground. In total it weighs 7.6kg.

Power is delivered by six 2-cell (or alternatively three 4-cell) battery packs that have a capacity of approximately 6.6Ah. The power consumption at hover is roughly 5.3kW at 120A. A PX4 flight computer is used to run the control algorithms. Each fan delivers a maximum thrust of 4.2kg, which results in a payload of approximately 3kg.

Research on control algorithms

The Flying Platform is used for benchmarking novel control strategies that account for actuation limits. In particular, the flaps are limited to a range of approximately 18°. Control algorithms explicitly accounting for this limitation can provide larger stability margins and other performance enhancements.

We are currently investigating model predictive control algorithms. The idea is to explicitly exploit the physical model of the system – including input and state constraints – for making model-based predictions. Since these methods are computationally expensive, we seek to trade off the prediction accuracy with computational cost, and quantify the consequences of such approximations.

If you liked this article, you may also be interested in:

• Watch flying machines weave a rope bridge you can walk on
• Quadrocopter failsafe algorithm: Recovery after propeller loss
• Quadrocopters learn from prior experience to improve slalom flying

   

• Multiple tethered quadrotors performing high-g, high-speed formation maneuvers 

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

In this new lecture series, controls expert Brian Douglas walks you through key concepts in control system theory. Focused on making control theory accessible and intuitive, this series is for anyone who wants to relate control concepts to robotic applications in the real world. 

Presumably you are a human, an as such you have become really good at manipulating your environment by using your hands. You have these incredible actuators that allow you gently and nimbly grab an object, hold it tight, and then move it to a new location. So why wouldn’t you use your hands to grasp everything, and why wouldn’t you build a humanoid robot with hands so it could be as capable as you?

Well, this type of manipulation is called prehensile manipulation — or gasping manipulation — and it’s reasonable to expect a robot to be able to interact with the environment in this manner. However, you may not realize that as a human you are also performing a huge amount of non-prehensile manipulation everyday, both with and without your hands, such as pushing a swinging door open at the store, typing on your keyboard, or walking on two legs.  

A robot that is expected to do well in our human built environment would need to be good at non-prehensile manipulation as well as the prehensile sort, and unfortunately it doesn’t always have a straightforward solution. The difficulty of getting a robot to move like a human was demonstrated at the DARPA Robotics Challenge (DRC), where the robot contestants were expected to perform activities like driving a car, opening a door, and closing a rotating valve. The robots successfully completed a number of very impressive feats, but it was also obvious from the number of robots falling down that we have a long way to go in our efforts to build a human-like robot; not the least of which is building agile bi-pedal robots.

I was recently at the International Conference on Robotics and Automation (ICRA 2015) and I came across a very unassuming robot performing continuous non-prehensile manipulation of a ball around the edge of a butterfly-shaped link. Using a camera to sense the position of the ball, and a motor for actuation, the robot was able to continuously spin the butterfly shape while keeping the ball perched on top. Maksim Surov and Leonid Paramonov developed the algorithms that made this possible, and not only was their display mesmerizing, but their formalized approach to solving these types of problems could go to advancing our ability to generate continuous stability for an agile walking bipedal robot.

Transcript included.

In this episode, Audrow Nash speaks with Sergei Lupashin about Perspective Robotics’ tethered flying camera, the Fotokite Phi, which is currently crowdfunding on Indiegogo. The Phi is a portable light weight GoPro-carrying quadrocopter that flies without a vision system or GPS, and instead uses the tension on the retractable tether to determine where it is in space. According to Lupashin, the tethered leash provides a natural, easy-to-learn and accountable interface for controlling the Phi.

https://www.youtube.com/watch?v=aglTWdI-4Bo

Sergei Lupashin

Sergei Lupashin is the founder of Perspective Robotics, the Swiss company behind the Fotokite. Previous to that he worked in aerospace and participated in robotics competitions such as the DARPA Grand+Urban Challenges (autonomous cars). He was chief architect for the ETH Flying Machine Arena, where he enjoyed torturing little quadrocopters until they learned to do flips. He holds a BS in Elec & Comp Engineering from Cornell, an MSc + PhD in Mechanical Engineering from ETH Zurich (topic: aerial robotics) and is a TED Fellow. Sergei is passionate about building devices that are usable and useful in the real world.

Links:

• Download mp3 (14.3 MB)
• Subscribe to Robots using iTunes
• Subscribe to Robots using RSS
• Link to Fotokite Phi Indiegogo campaign

Transcript

Audrow Nash: Hi welcome to the Robots Podcast. Can you introduce yourself to our listeners?

Sergei Lupashin: I’m the founder of Perspective Robotics, a company based in Zurich, Switzerland that is behind the Fotokite, the Fotokite Pro and the Fotokite Phi.

Audrow Nash: Fotokite Phi is currently crowdfunding on Indiegogo. Can you tell me a bit about the campaign and its goal?

Sergei Lupashin: This is our first attempt at building a product for consumers. We started off by building the Fotokite Pro, which is a specialized, more expensive product designed to be used for broadcast. With the Fotokite Phi we have tried to make something much more accessible both in terms of price and usability. It’s extremely portable and it can be used in lots of different situations.

We are trying to raise a minimum of US $300,000 … It’s a hardware product, so you have to build critical mass to make it accessible for a wider audience. It’s a very exciting step for us.

Audrow Nash: Can you describe the Phi?

Sergei Lupashin: I’m holding one in my hands right now … It’s a small quadrocopter that holds a GoPro, and it weighs only 12 ounces – about the weight of a small can of soda.

What’s unique about it is that you control it via tether. You hold onto what is essentially a glorified dog leash with some electronics, and you launch the Phi just by turning it on, giving the quadrocopter a twist, and letting it fly.

Audrow Nash: This doesn’t have the same sensors as most drones. Of course it has an IMU, but you’ve removed GPS, correct?

It’s a different experience from traditional drones because the interface is so physical: you feel pretty much the smallest motions of the Fotokite in the air through the string.

Sergei Lupashin: Exactly. One of the things we use the tether for is to figure out where the Phi is relative to the user. A normal drone does that by having a GPS receiver or having some sort of machine vision to figure out where it is. In our case, we use a tether that is kept under tension at all times. By calculating where that tension force points, the Fotokite actually knows where it is relative to you.

This changes the dynamics of the system because it’s completely self-contained and the whole experience exists around the user. If you are walking down the street, hiking in the mountains or walking down a trail with this thing attached to you, it will naturally follow with you without any special algorithms.

Audrow Nash: Can you tell me a bit about the leash, how it works, how it keeps constant tension?

Sergei Lupashin: When it’s hovering it doesn’t fly straight up; it flies at a small angle and it stays not above you but to the side. The leash is spring-loaded, so the Fotokite produces a bit of extra force to keep the tether under tension.

You can just point the Fotokite in the direction you want to launch it, but once it’s flying, how do you tell it to turn? How do you tell it to move around? We wanted to get away from the concept of having to use a joystick or your smart phone to do that. Inside the leash are electronics that connect to the Fotokite while it’s flying; to control it you hold down the button on the side of the leash and you rotate your wrist. That motion of your wrist – that motion of the leash – is translated into a command for the Fotokite.

Audrow Nash: What kind of commands can you give? Can you have it rotate to look around?

Sergei Lupashin: Yes. Right now there are just two modes of interface. One mode is in-phase rotation. For example, if I want to take a panorama, I would launch the Fotokite straight up and then I would hold down the first button and rotate my wrist. The Fotokite would then rotate in place and capture the full 360 panorama. The other mode lets you control the position of the Fotokite relative to you.

Audrow Nash: The angle?

Sergei Lupashin: Exactly. We usually speak in terms of ‘elevation’ and ‘orbit’. For example, if you look at the stars in the sky, a star has a certain elevation above the horizon. To control the elevation in our case, holding down a button while you move your wrist up and down will move the Fotokite either from a horizontal position to a vertical one (relative to you), or vice versa. It basically controls the vertical angle where it flies.

Moving your wrist in the horizontal plane will cause the Fotokite to orbit, meaning that it will slowly reorient itself around you. If you keep doing this motion, the Fotokite will complete a circle around you at a given elevation.

You can get really neat matrix-style shots of yourself. Or you can use it to do sophisticated shots, using the Fotokite like a flying steady-cam.

Audrow Nash: How long does it take to train someone to use the Fotokite?

The beauty of the system is that there is nothing active in this tether, so you can attach any string to the Fotokite and it will still work just fine.

Sergei Lupashin: It’s a different experience from traditional drones because the interface is so physical: you feel pretty much the smallest motions of the Fotokite in the air through the string.

We did a number of demos for both expert drone users and people who haven’t touched one before. Getting it to fly is surprisingly straightforward: you show someone how to do it, they’ll repeat the motions, and it will work immediately.

To get really unique shots and smooth video requires a bit of skill. The nice thing is, with the tether, you can use your own physical movements to control it and to keep it from crashing into things. It’s a lower stress experience than what you see with traditional drones.

Audrow Nash: Speaking of crashing into things, what happens if it bumps into an object? Or if someone runs into the tether, the tether becomes taught in a different direction, and that tension is released suddenly?

Sergei Lupashin: In terms of bumping into things, we have protective coverings that extend past the motors, so most of the time an object would bump into the protective coverings first. That helps to prevent crashes, and it also helps to give the user confidence in being around the Phi.

In terms of the tether getting caught in things, the beauty of the system is that there is nothing active in this tether, so you can attach any string to the Fotokite and it will still work just fine.

For example, if you’re filming the rooftop of your house and you’re trying to get over to the other side of it, you can actually fly the Fotokite over the rooftop and then get it such that the tether touches the edge so you can sling-shot it over. It will simply use that new fixed point as the new reference point, and this happens naturally.

Explained in words it sounds weird, but just imagine flying a kite where the lift is generated by the kite itself: if you grab the kite string, and someone pulls it away, it will still fly at the same angle – but relative to that new point.

Audrow Nash: Does this mean that you don’t really need the smart leash? Is it just for controlling the Fotokite, so that it rotates and you can change the vertical angle?

Sergei Lupashin: The smart leash is there to make the system complete. Most folks just want to use the system to capture a picture or video. We didn’t want to make them have to go find a string.

The leash is a convenience: it’s nicely built, it has a spring, and it has electronics that allow control.

The Phi’s arms fold up with the propellers on the inside so that you can put it into its tube-shaped carrying case. It’s about the size of a thermos, and you fit it into your backpack.

I should also mention that the Fotokite Phi folds, because in real life, transportation is a major issue. The Phi is pretty unique in that it has a lock mechanism that makes a satisfying click when it locks. And when you unlock it, the arms fold up with the propellers on the inside so that you can put it into its tube-shaped carrying case. It’s about the size of a thermos, and you fit it into your backpack.

Audrow Nash: And everything fits in there?

Sergei Lupashin: Yes, the leash and everything.

Audrow Nash: The tether is very nice from the perspective of controlling the drone. What are some other advantages of using tethers?

Sergei Lupashin: The tether wasn’t actually the original idea.

I was finishing my PhD and looking at different schemes for aerial systems. This was in Raff D’Andrea’s lab – he has been on your podcast as well. We were exploring some different approaches using gestures – with Kinnect for example. But all of these systems were based on machine vision, which is incredibly sensitive to the conditions you are operating in. Sunrise, lens flare, darkness, water – all of these things confuse the system.

The tether completely changed the dynamics of the conversation between the operator, the vehicle and the public. It makes this device much more acceptable in public.

So the question was: how do we get localization and interaction in the most robust way possible? That’s where the idea of the tether came from: it was originally a technical solution to let a flying vehicle figure out precisely where it is relative to you, without relying on sensitive systems like GPS or machine vision etc.

Then, when we made the first demo, it turned out to be a really elegant way to interact with the system. The tether completely changed the dynamics of the conversation between the operator, the vehicle and the public. I think this was actually one of the most exciting moments of this process, because as a technologist you don’t foresee it; you see the tether as a technical solution and nothing else. But then it turns out to also have all of these facets that are not technical, but rather about the experience, and how people see the product.

In the end, the tether changes the Fotokite from being another drone into something that is more of a companion flying device. It’s a very personal device … it’s more like a kite, and people treat it that way.

One parallel is to imagine being at the park and having your dog on a leash versus not on a leash: for some people it’s fine, but other people just get freaked out by dogs that are not leashed. I think we see some of that with drones from a privacy perspective. What we found is that having a tether changes the conversation and makes this device much more acceptable in public.

Audrow Nash: There’s accountability because you can see where the operator is?

Anyone – no technical background required – intuitively understands the chain of responsibility with this device.

Sergei Lupashin: Exactly. Anyone – no technical background required – intuitively understands the chain of responsibility with this device.

Audrow Nash: Can you tell me a bit about how this and similar technology has been used in journalism, for example the Fotokite Pro?

Sergei Lupashin: The original inspiration for the Fotokite came from aerial footage of the Bolotnaya Square demonstrations in Russia in 2011. It was after the federal elections, and it was a big demonstration right in the center of Moscow. The footage was captured pretty much by pure chance from a drone – quite a big drone with an SLR camera. There was this one image, a panorama really … it’s amazing how well it transmitted the scale and atmosphere of the event.

The people who took that footage were very lucky that they happened to have the right equipment; they were professionals so they could do it safely, and they happened to have the right permissions. But we’ve seen many other examples where people haven’t been able to get these perspectives and cover events like this as well as they could have been covered. For example, in Ferguson, Missouri where there was a no fly zone was imposed, which many believe was done to keep the media out.

We also found out that having a physical connection usually changes the regulatory aspect. For example, in Switzerland and France, after some conversation with the regulators we were able to get special rules for tethered devices that are much more permissive than traditional drones. From a regulatory perspective, the usual thought process is focused on safety, whereas privacy is more universal and not usually what they are basing rules on. From a safety perspective, in Zurich for example, it was prohibited for a while to fly drones commercially. The Fotokite was allowed based on the tether and specifically based on the idea that the tether restricts the height of the device.

Many regulators think of how much energy would be dissipated if the device falls in an uncontrollable fall, and there is a magic number of 66 joules … if you fall and it dissipates 66 joules then you are magically unsafe.

Audrow Nash: How much is 66 joules?

Sergei Lupashin: This is potential energy, so it’s Mass X Gravity X Height. Imagine a 6 kg bowling ball falling from about a meter …

It’s a safety margin. With the Fotokite’s tether you can restrict the length of the tether and thus the height, so you can guarantee that under no conditions are you going to exceed that safety limit.

Audrow Nash: So you can only extend to the length of the tether unless you can’t get high enough to possibly generate 66 joules on complete impact?

Sergei Lupashin: Yes, that’s right.

There are other aspects they look at, too, like system robustness. For example, how many subsystems or how many components do you have in your critical loop? How many subsystems are there such that if one of them fails you have an uncontrolled system or a catastrophic failure? In a traditional drone, GPS is usually in that critical loop, as is your radio system, and if you lose your GPS or radio you will crash into something. With the Fotokite, there is no GPS and if the radio cuts out, it will just fly in the same position.

Audrow Nash: How does it look for other countries with these exemptions? For example, do you believe the US will permit tethered drones to fly? Do you think they will give exemptions similar to Switzerland and France?

Sergei Lupashin: It’s very hard to predict which way the US regulations are going to go right now. If you look at the history of it, just over the past few months regulations were proposed that were extremely lenient and permissive. Most recently there has been a massive push over certain events … you’ve probably heard of the person shooting drones down over their property. So it’s really hard to predict.

It’s a challenge since all of this technology is new and people don’t quite know how to differentiate it. For example, what is a tethered drone? If you attach a fishing line to a DJI, is it now tethered? I kid you not that there are commercial products out there for sale that offer exactly this, and they promise to get you around regulations.

We are trying to be proactive and come up with commonsense and observable definitions of what it means to be tethered.

We are trying to be proactive and come up with commonsense and observable definitions of what it means to be tethered. For example, I really believe that the tether has to be loadbearing, meaning that under typical flight it should actually be under tension, and it should be rated for some sort of tension that responds to the vehicle.

Audrow Nash: For a hobbyist or someone that just wants to use the Fotokite Phi to take pictures of themselves, do these regulations mean anything outside of countries that have exemptions for tethered drones? Or is it only for commercial use and journalism that the regulations apply?

Sergei Lupashin: Regulations generally apply to commercial use, but there are exceptions specifically generally cities, populated centers, and in the US national parks. If you go for a hike to the Yosemite or Yellowstone National Park in the US, for example, you can’t use a drone right now. These are the kind of the frontlines where we’re going to see how consumer drones get regulated in the future.

Audrow Nash: Going back to the Phi, how does the tether affect battery life?

Sergei Lupashin: In the Pro we use the tether to supply power, so we can fly forever. In the Phi we couldn’t do that because of cost and complexity, so we use a standard one-port battery. The trick there is that, since you’re producing extra tension to keep the tether taut, you are wasting a tiny bit of battery life, but on the other hand, if you imagine how a kite actually uses the wind, you could be using a less than normal power to stay afloat.

The current prototypes fly for around 7-10 minutes. For the production units, we are aiming for 15 minutes of typical flight time. The battery is swappable in the field, so you always have the option of replacing it.

Audrow Nash: How did you decide to move from a journalism drone – the Fotokite Pro – to a consumer drone with Fotokite Phi?

Sergei Lupashin: With the Fotokite Pro we realized after talking to existing drone users that the tether has a unique advantage especially in the niche of live reporting. In journalism you typically want to operate next to people and next to something that’s happening, and you don’t want to add tension to that situation. It’s a very unique niche where the tether makes a huge difference. The powered tether is important because journalists can keep the device afloat and not worry about battery life.

But we always wanted to build something a wider reach, and a few months ago we finally decided it’s now or never, because everything is developing very quickly. The Phi is is a big experiment for us to see how well we can use the tether to build a new user experience.

A lot of people see the Phi for the first time and they think it’s just a quadrocopter on a string, but the deeper story is to look at the engineering behind it, using the tether as a different interface approach.

A lot of people see the Phi for the first time and they think it’s just a quadrocopter on a string, but the deeper story is to look at the engineering behind it, using the tether as a different interface approach.

Audrow Nash: Is this a way of addressing the interaction bottleneck with robotics technology?

Sergei Lupashin: That’s right. Have you tried reprogramming your alarm clock lately? [laughs]

In robotics labs we tend to come up with ridiculously complex systems. I’m sure they can do a lot in principle, but they are very difficult to interact with. I think this somehow echoes with many robotics products that have already come out. An example of that is in Japan when the Fukushima incident happened. Arguably Japan has some of the most advanced robotics, especially for search and rescue, anywhere in the world. But they really struggled to make a difference in that situation. And I think a lot of problems are because of interaction, because of training, and because of the complexity of the systems.

The inspiration for the Phi is really to try to cut across that and come up with a smooth and low-friction user experience, where even someone who has never touched robotics could be shown how to use this device and they could naturally take it, use it themselves, and feel confident and empowered by it.

Audrow Nash: Thank you.

Sergei Lupashin: Thank you.

All audio interviews are transcribed and edited for clarity with great care, however, we cannot assume responsibility for their accuracy.

Tailsitter in the Flying Machine Arena. Source: Robin Ritz, Institute for Dynamic Systems and Control, ETH Zurich.

In August 2014 Google [x] announced that they had been secretly developing a drone delivery program to rival Amazon’s Prime Air. Called Project Wing, the system was based on a mechanically simple “tailsitter” design – a hybrid vehicle that takes off vertically like a multirotor but is also able to fly horizontally like a fixed-wing airplane. Following field tests, however, Google scrapped the design, saying that it was difficult to control and didn’t perform well in high winds. Since then, researchers at ETH Zurich have developed a new algorithm for robustly controlling a tailsitter flying machine in hover position. Using the algorithm, the tailsitter is able to recover from any orientation, including upside down. ETH has not yet demonstrated its algorithm in real-world conditions (e.g., high winds, GPS localization, different payload distributions). However, given the robustness of the solution shown in the video, perhaps Google should not dismiss the tailsitter design too quickly for its next design iteration.  

– Robohub Editors

A tailsitter provides hover capability by pointing the nose and thrust direction upwards. For fast forward flight, the vehicle tilts to a near-horizontal orientation, resulting in efficient lift production due to the conventional wing design. Compared to other fixed-wing, hover-capable aircraft such as tiltrotors and tiltwings, the major advantage of a tailsitter is its mechanical simplicity: no mechanism for changing the direction of the propulsion system has to be added, saving weight and increasing reliability.

The algorithm demonstrated in the video is based on Optimal Control. Optimal trajectories for a set of initial orientations are precomputed and the corresponding initial desired body rates are stored in a small lookup table (500 entries). An outer control loop running at 50 Hz reads the desired body rates for the current orientation from the lookup table, and feeds them to an inner control loop. The inner control loop runs at 1000 Hz and tracks the desired body rates with the two propellers and the two flaps using feedback from a rate gyroscope. The resulting real-time computational complexity is extremely low.

Tailsitter’s propellers and motors. Source: Robin Ritz, Institute for Dynamic Systems and Control, ETH Zurich.

The propellers are driven by brushless Hacker A05-13S motors that are attached to the wings by 3D-printed mountings. While the vehicle is hovering, it is crucial that a large part of the flaps are within the propeller-induced airstream, in order to keep the flaps effective and the system highly maneuverable. The motors are controlled by ZTW Spider 12A ESCs, which run custom SimonK firmware that enables a fast response to thrust set point changes. Since the custom SimonK firmware is tuned for the fastest possible response, the motor controllers get quite hot and are therefore placed in the propeller airstream for cooling.

Tailsitter’s underbelly. Source: Robin Ritz, Institute for Dynamic Systems and Control, ETH Zurich.

Two radios are attached on the bottom side of the vehicle; a WiFly module for high-bandwidth telemetry, and a low-latency, frequency-hopping 2.4 GHz radio link that receives the control commands from the ground station. The total system latency including the motion capture system and the radio link is around 35 milliseconds.

Tailsitter’s flaps. Source: Robin Ritz, Institute for Dynamic Systems and Control, ETH Zurich.

The flaps are actuated by MKS DS65K servos providing a fast response on flap angle changes. The flaps of the original frame are designed for fixed-wing forward flight and do not provide enough control authority for the hover flight regime. Therefore their area is enlarged by carbon fibre sheets. This increases the effectiveness of the flaps by a factor of four. The maximum flap deviation is 30 degrees.

Tailsitter’s flight management unit (FMU). Source: Robin Ritz, Institute for Dynamic Systems and Control, ETH Zurich.

A Pixhawk PX4 Flight Management Unit (FMU) runs the onboard control loop that tracks the desired body rates using a rate gyroscope which is also provided by the FMU. Safety routines that switch off the actuators if the vehicle does not receive control commands anymore or if the battery voltage is too low are also implemented on the FMU.

The algorithm is presented in research paper “A Global Strategy for Tailsitter Hover Control”, submitted to International Symposium on Robotics Research (ISRR), 2015.

Researchers
Robin Ritz and Raffaello D’Andrea
Institute for Dynamic Systems and Control, ETH Zurich
Flying Machine Arena

Technical details
Frame: Robbe Mini Wing RC styrofoam airframe
Electronics: Pixhawk PX4 Flight Management Unit
Battery: Thunder Power RC G6 Pro Lite 25C LiPo 350mAh 2S
Motors: Hacker A05-13S
Motor controllers: ZTW Spider 12A ESCs with SimonK firmware
Propellers: 5×3 GWS EP-5030
Flap servos: MKS DS65K
Infrastructure: Flying Machine Arena

• Multiple tethered quadrotors whip around a pole together at high speeds 
• Quadrocopter failsafe algorithm: Recovery after propeller loss
• Three new quadrotor videos demonstrate agile control and the power of machine learning
• Video: Quadrocopter learns from its mistakes, perfects slalom air racing

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

The testbed allows the quadrocopter’s high speed flight behavior to be characterized in order to determine drag characteristics, propeller efficiency, and the physical limits of the machine. It is also being used to safely develop high-speed maneuvers such as emergency braking.

Note that it is possible to remove the central pole by balancing the forces acting on the strings; this could be then used in performance settings, possibly enhanced by light and sound effects.

This research was conducted at the Flying Machine Arena at ETH Zurich.

Reference:

Maximilian Schulz, Federico Augugliaro, Robin Ritz, Raffaello D’Andrea , “High-speed, Steady Flight with a Quadrocopter in a Confined Environment Using a Tether “, IROS 2015, submitted.

If you liked this article, you may also be interested in:

• Three new quadrotor videos demonstrate agile control and the power of machine learning
• Quadrocopters learn from prior experience to improve slalom flying
• Surrounded by his quadrocopter drones on stage, Raffaello D’Andrea explains feedback control and talks about the coming Machine Revolution
• Video: Watch flying robots cooperate, throw and catch balls
• Quadrocopter failsafe algorithm: Recovery after propeller loss

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

Komekurayama Solar Power Plant owned and operated by TEPCO in Kofu, Yamanashi Prefecture. Source: Sakaori via Wikimedia Commons.

Japan’s power industry is currently centralized, but it aims to deregulate by around 2020. Coupled with this major structural market change, the expansion of thermal, nuclear and renewable power generation will place additional demands on the management of the country’s energy market. Researchers from the Namerikawa lab at Keio University are working with control engineers, power engineers and economists to designing mechanical and control algorithms that can manage this large-scale problem.

The Namerikawa Laboratory of the Department of System Design Engineering in the Keio University Faculty of Science and Technology is conducting theoretical research of distributed control and cooperative control, as well as research concerning social applications in network systems thereof. As applied research, the Namerikawa Lab is conducting world-leading research in the areas of distributed and cooperative control of large-scale, complex systems connected to a network, formation control of unmanned aerial vehicles, and energy management of power networks.

“Since the earthquake and tsunami disaster of March 11, 2011, energy has become an extremely important issue that now involves the problems of nuclear power generation and the use of renewable energy. With all these areas, we no longer have centralized management by power companies as we did in the past, and instead, various companies can now supply energy. In addition, we also now have the option of selecting the company that is best suited for our needs from amongst various energy companies. Hence distributed management is becoming a key issue. “

Japan is aiming to deregulate the power industry by sometime around the year 2020, and with this deregulation, the power network is expected to become extremely complex. In the past, information was consolidated and managed based on centralized management, but after deregulation, it will be necessary to analyze power systems on the demand side and the supply side to ensure stabilized frequency tuning, and an appropriate power market must be created.

Frequency controllers are used to control power generators, and by running the generator itself at a constant rotational speed, we should be able to obtain power with a stable 50 Hz, 100 V supply voltage. However, when we add renewable energy and other power sources to the power network, they act in a form that is like noise, so the first thing that we need to do is implement feedback control with respect to those additional power sources so that a 100 V supply voltage can be continuously maintained at 50 Hz.

In addition to thermal power generation and nuclear power generation, an expansion of renewable energy is also expected to occur in the future. With that expansion, the power supply side will need to implement frequency control in order to maintain a stable supply of power, while the demand side will need to be managed based on a new control theory.

“The second area we need to look at is controlling demand using what is called demand response. Basically, what we want to do is achieve a balance between supply and demand. We want to change the behavior of consumers by changing the price. By doing this, we can shift demand peaks and level out power usage, and as a result, we are basically able to control consumers. This isn’t merely a mechanical and electrical control problem. It also pertains to economic activity, so it is an economic problem and an electric power problem. Trying to figure out how to design the power market in order to deal with these issues has become an extremely large-scale problem, namely a market design problem. When a full deregulation of power transactions has been achieved, there will be a demand to design a power market that is fully capable of supporting this type of deregulation. Therefore, right now we’re working with control engineers, power engineers, researchers, and economists to research new mechanisms for power transactions.”

With consideration of the standpoints of both the power demand side and supply side, the Namerikawa Lab is leading the world in research to develop algorithms to find the most appropriate compromise between the two sides and to quantify satisfaction levels. The lab is also focusing its efforts on research to build sensor agent and multi-agent systems to determine in real time the status of spatially and temporally distributed energy and environments.

“We’re conducting research with a major goal of fusing together this type of energy technology and UAV control technology to achieve optimum management and optimum control of social systems and social infrastructures in the future.”

If you liked this article, you may also be interested in:

• Why is Japan the first to get Dyson’s new 360 Eye?
• Japan’s new robotics push: Funding and deregulation
• Fukushima photo essay: A drone’s eye view
• ShanghAI Lectures: Weidong Chen “Modeling and Control of Mobile Robot Networks”
• The Distributed Flight Array: Modular robots that self-assemble, coordinate and take flight

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

In our laboratory we develop various types of high-speed vision hardware and algorithms that can implement high-speed image processing with a sampling time from 10ms up to 1ms. High-speed vision can provide control data at the same sampling rate as that of the servo controller used for the robot actuators. This means that vision can control actuators just like other sensors e.g. an encoder. Although at present the camera is located off board the robot, it will be attached to the body in future iterations.

In addition, we developed a light-weight, high-power actuator for high-speed motion. Its torque per weight ratio is 3.5 times higher than that of previous products of same actuators.

Those technologies are used in various demonstrations of our robots such as:

• Janken (Rock Paper Scissors)
• batting
• raw egg catching
• regrasping
• dribbling
• picking up with tweezers
• flexible material handling
• book flipping scanning
• 1ms pan/tilt
• micro visual feedback

The running algorithm used in the ACHIRES robot is different from those typically used in other running robots. While most running robots use a method based on ZMP-criteria for maintaining stable and balanced posture, we introduced a very simple algorithm using high-speed performance of a sensory-motor system without ZMP criteria. The aerial posture is recovered to compensate for the deviation from the stable trajectory using high-speed visual feedback.

It took four years to develop ACHIRES, in part because analyzing robot dynamics that are faster than video capture rates requires high speed video analysis. You can see how the abilities of the robot have evolved since the project was first started in 2009:

Although ACHIRES is a research platform with no direct application at the present moment, the combination of high-speed vision and actuation could be applied to various types of high-speed intelligent systems, including high-speed robots, manufacturing systems, aircraft, microscope image control for bio/medical applications, and human-machine interfaces. We believe it will open new era of visual feed back systems.

More info:
Project Website
YouTube channel

Reference: T. Tamada, W. Ikarashi, D. Yoneyama, K. Tanaka, Y. Yamakawa, T. Senoo, M. Ishikawa: High Speed Bipedal Robot Running Using High Speed Visual Feedback, The Robotics Society of Japan The 32nd Annual Conference (RSJ2014) (Fukuoka, 2014)/1B2-03.

Some of AIR’s feature highlights are:

• Robot application development can be done in almost any programming language (C/C++, Java, Python, Ruby, Pearl… even Bash).
• The robot has a distributed fault-tolerant hardware architecture with a state-of-the-art control and positioning system.
• It runs a real-time Linux OS.
• It is equipped with range and proximity sensors, a image processing module and an audio recognition system.
• It weights approximately 0.7 Kg with a circumference of 10 cm and, like an ant, can push, pull or carry up to 9 Kg (more than 10 times its actual weight).

More info:

AI-MAS Group Website
Prof. Adina Magda Florea’s Homepage
Youtube Channel

Last December, an article named “Playing Atari with Deep Reinforcement Learning” was uploaded to arXiv by employees of a small AI company called DeepMind. Two months later Google bought DeepMind for 500 million euros, and this article is almost the only thing we know about the company . Currently our team is trying to replicate their artificial mind, and in this post we describe its inner workings.

The paper describes a single system that learns to play several old video games by selecting actions based on only the screen images and the reward signal (whether the score in the game is increased or not).

At first (before training), the program does not know anything about the game: it doesn’t know the objective of the game (if it should just stay alive, kill something or solve a puzzle), it has no understanding about its influence on the game (the consequences of its actions) and one could even say that it cannot see the objects in the game. By trial and error the system gradually learns how it should behave to receive a reward. Also note that the system uses the same architecture for all different games without any game-specific hints from the programmers other than how many possible actions there are in the game, e.g. left, right, up, down, space (fire) etc.

The reported results show that the system was able to master several games and play some of them better than a human player. This result can be seen as a step towards artificial general intelligence (AGI), which is really fascinating. Let us see in more detail how this was achieved.

The Task
The system receives a picture of the game screen (this example image is from the simplest game, Breakout) and after analyzing it, it has to choose an action to take. The action is executed and the system is told whether the score increased, decreased or did not change. Based on this information and playing a large number of games, the system needs to learn to improve its performance in the game.

Take a pause and think about it: try to outline in your head how you would approach this task if someone had asked you to program it. Keep in mind that you do not know in advance which game your system will need to play.

Machine Learning and Artificial Neural Networks
Before we dive into the implementation DeepMind came up with, let’s introduce the concepts of machine learning and artificial neural networks.

A very common task in machine learning is: given the information about an object decide which category it belongs to. In the case of DeepMind we want to decide which action to take given the current state of the game. Machine learning algorithms learn from examples: given many example objects and their categories, the algorithm figures out which features of the object are useful for identifying the category. The algorithm produces a model that minimizes misclassification rate in the training set. This model can later be used to predict the category of an unknown object.

Artificial Neural Networks (ANN) is one of the machine learning algorithms. It is inspired by the architecture of our brain. The network consists of nodes (as our brain consists of neurons) that are connected to each other (as neurons in our brain are connected using axons, dendrites and synapses). For each neuron we must specify the condition under which the neuron should transmit the signal forward (emit a spike in real neurons). By changing the strengths of the connections one can make the network perform computation. Modern architectures of neural networks usually consist of:

• Input layer of neurons (they receive the description of an object)
• Hidden layers (the main part, where the learning happens)
• Output layer (it has one neuron for each category, the category with highest activated node is the predicted class)

After the learning is done we can feed new objects to the network and see scores for each category in the output layer.

DeepMind

In this system an ANN is used to assign an expected reward to each of the possible actions given the current game state. The network can answer a question, e.g. “How good would it be to take this action now?”. The input to the network at any time point consists of the last four game screens, so that system is given not only the last position, but also a bit of information about how the situation in the game has evolved recently. This input is then passed through three successive hidden layers of neurons and finally to the output layer of the ANN.

The output layer has one node for each possible action and the activation of those nodes indicates the expected reward from each of them – here, the action with the highest expected reward (the “best” action) is selected for execution. But how does the network know how to calculate the expected scores in a good and correct way?

The Learning Process

After you have defined the structure of your network there is only one thing you can still change: the strength of the connections. The learning process is, therefore, just changing the weights in a way that for all of the training examples the network is capable to propose actions that lead to positive rewards.

It is an optimisation problem where the goal is to maximize the reward over time. The problem is solved by combining gradient descent with a simple idea related to reinforcement learning: the network attempts to maximize not only the immediate reward, but also the cumulative reward of future actions. This can be achieved by estimating the expected value of the next screen image using its current neural network. To put it another way, traditionally you would use (received reward – estimated reward) as the error signal for gradient descent, but in this case we also take in consideration the reward estimate from the next image.

Summary

To summarize all the pieces we have seen, let us have one final look:

1. Build a network and initialize connections to random values

2. Feed lots and lots of game situations into the network

3. Network proposes the action to take in each of them

4. If the action was good, we reward the system, if it was bad we penalize it

5. System updates its knowledge by updating the weights

6. After several hundreds of games the system outperforms a human player (in the case of Breakout)

This result can be seen as a small step forward from the traditional machine learning in the direction of artificial general intelligence. It might be a very small step (one might say that the system does not “know” or “understand” what it is doing), but the learning capabilities of DeepMind’s system are far larger than its predecessors’  – and the range of problems is can solve without specific hints from the programmers is much wider.

Replicate and Make It Open Source

At the moment we are trying to replicate this successful project and make it available to the whole scientific community for further research. The project is open source and you can jump on board to help on Github!

References

The original paper:
Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Alex Graves, Ioannis Antonoglou, Daan Wierstra, and Martin Riedmiller. Playing atari with deep reinforcement learning.

Our implementation at GitHub:
Reproducing the results of “Playing Atari with Deep Reinforcement Learning” by DeepMind.

The authors of this article are from the Computational Neuroscience Group at University of Tartu’s Institute of Computer Science.

If you liked this article, you may also be interested in:

• Google’s robot and artificial intelligence acquisitions are anything but scary
• Google’s DeepMind acquisition in reinforcement learning
• On internal models, consequence engines and Popperian creatures
• Why robots will not be smarter than humans by 2029

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.

We’ve seen robots on stage before: Humanoids have been featured in theater productions in Switzerland, Austria, and Japan, and industrial robots have played parts in dance performances and other staged events. Flying robots have also played a theatrical role, for example in Shakespeare’s A Midsummer Night’s Dream at Texas A&M, a dance set to a Schubert piano trio at MIT, as well as in advertising. Some robots have even gone pro: Robothespian was designed for acting and is rented out for performances.

But the dance between human and machine has never looked quite like this .

The film’s main plot is straight-forward: Lamps in a repair shop magically come alive, rising from their stands to sneak up on an unsuspecting repairman. As the magic unfolds, and the repairman discovers that the everyday objects of his trade have become animated and full of character, viewers become witness to the evolving dance between man and technology.

Everyone on the team, including myself, was impressed by how well the film captures the stunning, at times hypnotic visuals of the flying lampshades. Everything you see is real — no CGI, wires, or speed-ups/slow-downs were used in the film. 

So how did the lampshades come alive?

This short film, titled “Sparked”, was produced in collaboration between researchers at ETH Zurich’s Flying Machine Arena, their spin-off company Verity Studios, and entertainment company Cirque du Soleil; in other words, it’s quadrocopter magic that makes the lampshades fly.

The film marks a step change from previous work by ETH Zurich’s Flying Machine Arena team, catalyzed by their new spin-off company Verity Studios, which also gets credit for designing the quadrocopters’ choreography. Here, in spite of their computer-controlled precision flights, the flying machines become robot actors that are individuals with their own personalities: Pomponette, with her tassels bustling in the propellers’ airflow; the Medusa with its bizarre tentacles waving in the wind; the Runt, sleek and fast but still glowing and twitching from the electric shock. And Lady Purple, absent from most of the film because of her intractable aerodynamics.

Flying machines up, close, and personal

One of the most striking aspects of this film shoot was watching our actor, Nicolas Leresche, interact with the flying machines. Much of this interaction was captured on camera — and you can see the results in the film, in particular the moment of ‘first contact’ — but, for me, it was what happened off-camera that was most revealing.

Naively unjaded by their joys and dangers, this was Nicolas’ first experience with quadrocopters. As he began interacting with them, tentatively at first, and then — in a matter of seconds — more confidently, even playfully, it became crystal clear that I was not merely observing a human using a robotic tool.

As the quadrocopter’s onboard controller pushed back against Nicolas’ hand, immediately reacting to match his actions, it became apparent that the two interacted in a way we don’t usually get to interact with technology. The interaction here was physical and dynamic, and clearly intuitive for Nicolas. There was no noticeable learning curve, no need for manuals or explanations.

But what was most striking about this interaction was the result. Within seconds, Nicolas had established a deep, almost animate connection with the flying robot actors. As their dance developed, driven by the autonomous machines’ choreography and Nicolas’ script, and became ever more fluid and natural, it became clear that what we were witnessing was a relationship in its making; one of mutual respect and, at the same time, confidence and trust. And once there was trust, magic happened.

In his own words:

Actors think that they are the ones who make objects move. I think that, on the contrary, it’s the objects that make us move. In the case of drones, even more so! They are companions (in an etymological sense), confrères, brothers.

Within hours the entire film crew had become comfortable with the flying machines hovering nearby or brushing past them as they swooshed through the air. On one occasion, a mis-timed dolly motion led to a flying machine hovering just centimeters above the camera man, who continued with his work unfazed. On another occasion, when Nicolas missed his cue, he bumped into one of the flying machines. But they just jostled past each other and continued on with their choreography as if nothing exceptional had happened. On yet another occasion, a drone collided with the moving camera without disturbance.

The only damage during the entire two-day shoot was a single broken lamp, knocked over by one of the quadrocopters’ airflow.

Safety

I have no doubt that this level of confidence and comfort with the technology can in no small part be attributed to the system’s reliability. In spite of the packed shooting schedule (a total of 40 shots over two days) and the high complexity of the project (for example, designing lighting conditions using more than 60 light sources including some moving through space), production went smoothly, and importantly for the quadrocopter team, the flying machines did not add any significant delay. Battery swaps and on-the-spot changes to the choreography were efficient and quick, never exceeding a few minutes. Kudos to the excellent film team from who’s mcqueen for brilliantly doing their part to keep the production on schedule.

Collaboration

One aspect that made this project particularly enjoyable was the collaboration with creatives at Cirque du Soleil. From the start there was a lot of excitement about the potential of this technology for entertainment not just at ETH and Verity Studios here in Zurich but also at Cirque in Montreal. Everyone shared the view that the flying machines could be used to amaze audiences and to tell wonderful stories. What ensued was a collaborative, iterative creation process that culminated in a script and a set design that were imaginative to Cirque du Soleil standards, while exhibiting the abilities of the quadrocopters in their full splendor.

As it turned out, figuring out what is easy and what is difficult to achieve with autonomous quadrocopters can be quite counter-intuitive. Having researchers and creatives on both sides collaborate on the key creative elements of the story from the early phases of the project therefore proved invaluable to its success.

Next steps

Why go to all this trouble? We’ve all seen the amazing things CGI can do on film. I love films and movies, how they can condense experience and add life to our most impossible dreams. But nothing compares to the real experience of getting up, close, and personal with robot actors yourself.

What’s next? Our audiences over the past years have had more ideas than we can count. At this point, only one thing’s for sure: They all want to see more flying robot actors.

Some more snapshots from the film shoot below for your viewing pleasure:

All images credited to the “Sparked” team. Markus Waibel is a former member of ETH Zurich, a co-founder of Verity Studios, and one of the contributors to “Sparked”.

For a couple of years I’ve been thinking about robots with internal models. Not internal models in the classical control-theory sense, but simulation-based models; robots with a simulation of themselves and their environment inside themselves, where that environment could contain other robots or, more generally, dynamic actors. The robot would have, inside itself, a simulation of itself and the other things, including robots, in its environment. It takes a bit of getting your head round. But I’m convinced that this kind of internal model opens up all kinds of possibilities. Robots that can be safe, for instance, in unknown or unpredictable environments. Robots that can be ethical. Robot that are self-aware. And robots with artificial theory of mind.

I’d written and talked about these ideas but, until now, not had a chance to test them with real robots. But, between January and June the swarm robotics group was joined by Christian Blum, a PhD student from the cognitive robotics research group of the Humboldt University of Berlin. I suggested Christian work on an implementation on our e-puck robots and happily he was up for the challenge. And he succeeded. Christian, supported by my post-doc Research Fellow Wenguo, implemented what we call a Consequence Engine, running in real-time, on the e-puck robot.

Here is a block diagram. The idea is that for each possible next action of the robot, it simulates what would happen if the robot were to execute that action for real. This is the loop shown on the left. Then, the consequences of each of those next possible actions are evaluated. Those actions that have ‘bad’ consequences, for either the robot or other actors in its environment, are then inhibited.

This short summary hides a lot of detail. But let me elaborate on two aspects. First, what do I mean by ‘bad’? Well it depends on what capability we are trying to give the robot. If we’re making a safer robot, ‘bad’ means ‘unsafe’; if we’re trying to build an ethical robot, ‘bad’ would mean something different – think of Asimov’s laws of robotics. Or bad might simply mean ‘not allowed’ if we’re building a robot whose behaviours are constrained by standards, like ISO 13482:2014.

Second, notice that the consequence engine is not controlling the robot. Instead it runs in parallel. Acting as a ‘governor’, it links with the robot controller’s action selection mechanism, inhibiting those actions evaluated as somehow bad. Importantly the consequence engine doesn’t tell the robot what to do, it tells it what not to do .

Running the open source 2D robot simulator Stage as its internal simulator our consequence engine runs at 2Hz, so every half a second it is able to simulate about 30 next possible actions and their consequences. The simulation budget allows us to simulate ahead around 70cm of motion for each of those next possible actions. In fact Stage is actually running on a laptop, linked to the robot over the fast WiFi LAN. But logically it is inside the robot. What’s important here is the proof of principle.

Dan Dennett, in his remarkable book Darwin’s Dangerous Idea, describes the Tower of Generate-and-Test; a conceptual model for the evolution of intelligence that has become known as Dennett’s Tower.

In a nutshell Dennett’s tower is set of conceptual creatures each one of which is successively more capable of reacting to (and hence surviving in) the world through having more sophisticated strategies for ‘generating and testing’ hypotheses about how to behave. Read chapter 13 of Darwin’s Dangerous Idea for the full account, but there are some good précis to be found on the web; here’s one. The first three storeys of Dennett’s tower, starting on the ground floor, have:

• Darwinian creatures have only natural selection as the generate and test mechanism, so mutation and selection is the only way that Darwinian creatures can adapt – individuals cannot.
• Skinnerian creatures can learn but only by literally generating and testing all different possible actions then reinforcing the successful behaviour (which is ok providing you don’t get eaten while testing a bad course of action).
• Popperian creatures have the additional ability to internalise the possible actions so that some (the bad ones) are discarded before they are tried out for real.

Like the Tower of Hanoi each successive storey is smaller (a sub-set) of the storey below, thus all Skinnerian creatures are Darwinian, but only a sub-set of Darwinian creatures are Skinnerian and so on.

Our e-puck robot, with its consequence engine capable of generating and testing next possible actions, is an artificial Popperian Creature: a working model for studying this important kind of intelligence.

In my next blog post, I’ll outline some of our experimental results.

Acknowledgements:
I am hugely grateful to Christian Blum who brilliantly implemented the architecture outlined here, and conducted experimental work. Christian was supported by Dr Wenguo Liu, with his deep knowledge of the e-puck, and our experimental infrastructure.

Related blog posts:

A few weeks ago, one of our readers, Javier Lopez, submitted a number of questions to the ORi team. He brought up many interesting points, and we won’t attempt to cover them all in one poll. But one of them had to do with the fact that some people actually enjoy the driving experience, and may not be willing to trade it for convenience. Perhaps this might be just another issue of preference, similar to how some people still prefer to drive a stick shift (standard/manual transmission) over automated transmission.

So we’d like to know what you think. Will you miss driving your non-autonomous car? Why or why not?

Given a choice between crashing into a motorcyclist wearing a helmet vs. a motorcyclist who isn’t wearing one, which one should an autonomous car be programmed to crash into? What about the choice between crashing into an SUV vs. a compact car?

These are some of the dilemma situations Professor Patrick Lin brought forth in his WIRED article, The Robot Car of Tomorrow May Just be Programmed to Hit You.

Lin says that programming a car to collide with any particular kind of object over another seems like a targeting algorithm. Accidents involving human drivers typically involve people’s split-second, instinctive responses to the situation. Your first autonomous car of the near future, however, will probably come equipped with a crash-optimization algorithm – a deliberately designed algorithm that will determine the outcome of all potential crashes. As Lin points out in his article, the pre-meditated nature of such algorithm creates a load of interesting issues that calls for a broader discussion on the topic.

So two weeks ago we, the members of the Open Roboethics initiative and the Robohub family, decided to help continue the discussion by conducting our reader poll on this very topic.

Scenario 1: Motorcyclists with / without a helmet

Given the choice between the two motorcyclists (one wearing a helmet, and one not), we asked our readers who an autonomous car should crash into. Choosing to hit the helmet wearing motorcyclist would perhaps minimize overall harm done. But considering the perspective of the motorcyclist who took that extra few minutes to put on a helmet for safety reasons, such programmed decision by an autonomous car doesn’t seem fair. Indeed, only 2% of our participants said that the car should crash into the biker wearing a helmet because s/he has better chances of survival. Lin points out that choosing to hit the biker wearing a helmet might discourage bikers from wearing helmets.

Crashing into the biker not wearing a helmet wasn’t a popular choice either (10%). In fact, our most popular response (45%) was that the car should hit the brakes and do nothing else, leaving it up to chance to determine which biker gets hit. And 3% of participants provided us with a similar response, where they said the car should run a random number generator to make a random decision. This means that almost half (48%) of our participants advocating for random chance to decide the outcome of the crash.

Relatedly, 30% of our participants said that the car shouldn’t have the ability to detect whether a biker is wearing a helmet or not. Given that a decision made by random chance does not make intelligent use of sensed data, effectively 78% of our participants are voting for autonomous cars to not use certain sensed data in making crash decisions.

But not using available information in crash decisions doesn’t seem like a straight forward solution to crash-optimization. Lin says,

“Not using that information in crash-optimization calculations may not be enough. To be in the ethical clear, autonomous cars may need to not collect that information at all. Should they be in possession of the information, and using it could have minimized harm or saved a life, there could be legal liability in failing to use that information.”

Scenario 2: Hit the SUV or a compact car?

An odd sense of unfairness also exists if we consider another crash scenario. Let’s say the car has a choice between an SUV and a compact car to unavoidably crash into. The car could be programmed to hit the SUV over a compact car, since bigger cars with perhaps better safety ratings could better absorb impact from the collision and minimize overall harm from the crash.

The responses we got are similar to the first scenario: 37% opted for the car to simply engage the brakes and do nothing else regardless of which car gets hit, and 25% said that the cars shouldn’t have the ability to detect the make/model of the cars around it.

A slight difference from the first scenario is that 20% said that the car should crash into the SUV (a minimum overall harm option), and only 3% said the car should crash into the compact car. In the first scenario, the option that yields minimum overall harm (hit the motorcyclist with a helmet) only received 2% support as compared to 20%.

This might have something to do with the fact that people’s decision to drive a certain make or model of a car is not illegal, nor an activity that typically increases risks of driving. So no one is really doing anything wrong by driving a compact car or an SUV, whereas the motorcyclist wearing a helmet has a moral/legal high ground over the other motorcyclist.

Regardless, if we were to have crash-optimization algorithms that biasedly crash into SUVs over smaller cars, it will surely have an impact on the consumer market. The rates of insurance for these supposedly more safe cars may go up, because buying a safer car may also mean a higher probability of being crashed into by an autonomous car – yikes!

Is it OK for a car to always choose to crash into non-law-abiding citizens?

So how does moral or legal high ground of individuals on the road impact people’s decisions about who should be hit? We asked our readers whether it’s OK for a car to always choose to crash into those not following traffic laws over those who are, assuming that the car can detect it. Considering the first scenario, it is illegal to ride a motorcycle without a helmet in many countries/states. According to our results, law-abiding status of people on the road doesn’t seem to be a popular variable to optimize for crashes. The majority of respondents said it is not OK (70%), whereas only 20% said it is OK for cars to biasedly crash into non-law-abiding citizens.

It is true that by answering ‘yes’ to this question, you’d have to be OK with the idea that the car might always choose the one less likely to survive from a crash. Although we have traffic laws in place to regulate the rules of the road and maintain social order, using it to make life/death decisions in this manner may not be something people are comfortable with.

How should an autonomous car respond to unavoidable crashes?

So how should an autonomous car respond to unavoidable crashes? Is there a general rule that people are more in support of? According to our results, majority of our participants (52%) are in favour of minimizing overall harm to both pedestrians and passengers by spreading out harms. But it seems that the rest of the participants are quite divided: 20% of people support to minimize harm to pedestrians at the expense of passengers, while 13% supports the opposite.

This reminds us of a recent reader poll discussion we had on a different crash dilemma situation, in which a large number of participants (64%) said an autonomous car should save the life of its passenger over a child on the road (36%). One of the main reasons for choosing to save the passenger’s life was the notion that a car should always have its passenger’s safety as its priority over that of others. Given that the previous poll result provided such a strong support for prioritizing the passenger’s safety, it is surprising to see a contrasting result on this poll.

To find out what makes people prioritize passenger safety over pedestrians or vice versa, we’ll have to do some more detailed investigations. What is also interesting is that, although people preferred autonomous cars to take random or uninformed decisions in the two specific unavoidable crash scenarios above, the majority of people opt for minimizing overall harm to both pedestrians and passengers – an option that is likely to be the least random and in need of using the most amount of information about the passenger and pedestrians.

For now, it seems that there is much more discussion to be had before autonomous cars start making crash decisions people will be happy with. Check out our latest reader poll: When fully autonomous cars are the norm, will you miss driving?.

The results of the poll presented in this post have been analyzed and written by AJung Moon, Camilla Bassani, and Shalaleh Rismani at the Open Roboethics initiative.

If you liked this article, you may also be interested in:

• Human error as a cause of vehicle crashes
• If death by autonomous car is unavoidable, who should die? Reader poll results
• Should autonomous cars be allowed to speed? Results from our reader poll
• Kids with wheels: Should the unlicensed be allowed to ‘drive’ autonomous cars?
• What should a robot do? Designing robots that know right from wrong
• Morgan Stanley reports on the economic benefits of driverless cars

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.