ROSCon 2015 recap and videos – Part 5
ROSCon is an annual conference focused on ROS, the Robot Operating System. Every year, hundreds of ROS developers of all skill levels and backgrounds, from industry to academia, come together to teach, learn, and show off their latest projects.
Here is the next set of posts from the OSRF blog, along with videos.
Amit Moran (Intel): Introducing ROS-RealSense: 3D Empowered Robotics Innovation Platform
While Intel is best known for making computer processors, the company is also interested in how people interact with all of the computing devices that have Intel inside. In other words, Intel makes brains, but they need senses to enable those brains to understand the world around them. Intel has developed two very small and very cheap 3D cameras (one long range and one short range) called RealSense, with the initial intent of putting them into devices like laptops and tablets for applications such as facial recognition and gesture tracking.
Robots are also in dire need of capable and affordable 3D sensors for navigation and object recognition, and fortunately, Intel understands this, and they’ve created the RealSense Robotics Innovation Program to help drive innovation using their hardware. Intel itself isn’t a robotics company, but as Amit explains in his ROSCon talk, they want to be a part of the robotics future, which is why they prioritized ROS integration for their RealSense cameras.
A RealSense ROS package has been available since 2015, and Intel has been listening to feedback from roboticists and steadily adding more features. The package provides access to the RealSense camera data (RGB, depth, IR, and point cloud), and will eventually include basic computer vision functions (including plane analysis and blob detection) as well as more advanced functions like skeleton tracking, object recognition, and localization and mapping tools.
Intel RealSense 3D camera developer kits are available now, and you can order one for as little as $99.
Michael Aeberhard (BMW): Automated Driving with ROS at BMW
BMW has been working on automated driving for the last decade, steadily implementing more advanced features ranging from emergency stop assistance and autonomous highway driving to fully automated valet parking and 360° collision avoidance. Several of these projects were presented at the 2015 Consumer Electronics Show, and as it turns out, the cars were running ROS for both environment detection and planning.
BMW, being BMW, has no problem getting new research hardware. Their latest development platform is the 335I G. This model comes with an advanced driver assistance system based around cameras and radar. The car has been outfitted with four low-profile laser scanners and one long-range radar, but otherwise, it’s pretty close (in terms of hardware) to what’s available in production BMWs.
Why did BMW choose to move from their internally developed software architecture to ROS? Michael explains how ROS’ reputation in the robotics research community prompted his team to give it a try, and they were impressed with its open source nature, distributed architecture, existing selection of software packages, as well as its helpful community. “A large user base means stability and reliability,” Michael says, “because somebody else probably already solved the problem you’re having.” Additionally, using ROS rather than a commercial software platform makes it much easier for BMW to cooperate with universities and research institutions.
Michael discusses the ROS software architecture that BMW is using to do its autonomous car development, and shows how the software interprets the sensor data to identify obstacles and lane markings and do localization and trajectory planning to enable full highway autonomy, based on a combination of lane keeping and dynamic cruise control. BMW also created their own suite of RQT and rviz plugins specifically designed for autonomous vehicle development.
After about two years of experience with ROS, BMW likes a lot of things about it, but Michael and his team do have some constructive criticisms: message transport needs more work (although ROS 2 should help with this), managing configurations for different robots is problematic, and it’s difficult to enforce compliance with industry standards like ISO.
Jerry Towler (SwRI): Mapviz – An Extensible 2D Visualization Tool for Automated Vehicles
ROS already comes with a fantastic built-in visualization tool called rviz, so why would you want to use anything else? At Southwest Research Institute, Jerry Towler explains how they’ve created a new visualization tool called Mapviz that’s specifically designed for the kind of large-scale outdoor environments necessary for autonomous vehicle development. Specifically, Mapviz is able to integrate all of the sensor data that you need on top of a variety of two-dimensional maps, such as road maps or satellite imagery.
As an autonomous vehicle visualization tool, Mapviz works just like you’d expect that it would, which Jerry demonstrated with several demos at ROSCon. Mapviz shows you a top-down view of where your vehicle is, and tracks it across a basemap that seamlessly pulls image tiles at multiple resolutions from a wide variety of local or networked map servers, including Open MapQuest and Bing Maps. Mapviz is, of course, very plugin-friendly. You can add things like stereo disparity feeds, GPS fixes, odometry, grids, pathing data, image overlays, projected laser scans, markers (including textured markers) from most sensor types, and more. It can’t really handle three dimensional data (although it’ll do two-and-a-half dimensions via color gradients), but for interactive tracking of your vehicle’s navigation and path planning behavior, Mapviz should offer most of what you need.
For a variety of non-technical reasons, SwRI hasn’t been able to release all of its tools and plugins as open source quite yet, but they’re working on getting approval as fast as they can. They’re also in the process of developing even more enhancements for Mapviz, and you can keep up to date with the latest version of the software on GitHub.
Matt Vollrath and Wojciech Ziniew (End Point): ROS-Driven User Applications in Idempotent Environments
Matt Vollrath and Wojciech Ziniew work at an ecommerce consultancy called End Point, where they provide support for Liquid Galaxy; a product that’s almost as cool as it sounds. Originally an open source project begun by Google engineers on their twenty percent time, Liquid Galaxy is a data visualization system consisting of a collection of large vertical displays that wrap around you horizontally. The displays show an immersive (up to 270°) image that’s ideal for data presentations, virtual tours, Google Earth, or anywhere you want a visually engaging environment. Think events, trade shows, offices, museums, galleries, and the like.
Last year, End Point decided to take all of the ad hoc services and protocols that they’d been using to support Liquid Galaxy and move everything over to ROS. The primary reason to do this was ROS support for input devices: you can use just about anything to control a Liquid Galaxy display system, from basic touchscreens to Space Navigator 3D mice to Leap Motions to depth cameras. The modularity of ROS is inherently friendly to all kinds of different hardware.
Check out this week’s ROSCon15 video as Matt and Wojciech take a deep dive into their efforts in bringing ROS to bear for these unique environments.
Tom Moore: Working with the Robot Localization Package
Clearpath Robotics is best known for building yellow and black robots that are the research platforms you’d build for yourself; that is, if it wasn’t much easier to just get them from Clearpath Robotics. All of their robots run ROS, and Clearpath has been heavily involved in the ROS community for years. Now with Locus Robotics, Tom Moore spent seven months as an autonomy developer at Clearpath. He is the author and maintainer of the robot_localization ROS package, and gave a presentation about it at ROSCon 2015.
robot_localization is a general purpose state estimation package that’s used to give you (and your robot) an accurate sense of where it is and what it’s doing, based on input from as many sensors as you want. The more sensors that you’re able to use for a state estimate, the better that estimate is going to be, especially if you’re dealing with real-worldish things like unreliable GPS or hardware that flakes out on you from time to time. robot_localization has been specifically designed to be able to handle cases like these, in an easy to use and highly customizable way. It has state estimation in 3D space, gives you per-sensor message control, allows for an unlimited number of sensors (just in case you have 42 IMUs and nothing better to do), and more.
Tom’s ROSCon talk takes us through some typical use cases for robot_localization, describes where the package fits in with the ROS navigation stack, explains how to prepare your sensor data, and how to configure estimation nodes for localization. The talk ends with a live(ish) demo, followed by a quick tutorial on how to convert data from your GPS into your robot’s world frame.
The robot_localization package is up to date and very well documented, and you can learn more about it on the ROS Wiki.
Moritz Tenorth (Magazino): Maru and Toru — Item-Specific Logistics Solutions Based on ROS
It’s not sexy, but the next big thing for robots is starting to look like warehouse logistics. The potential market is huge, and a number of startups are developing mobile platforms to automate dull and tedious order fulfillment tasks. Transporting products is just one problem worth solving: picking those products off of shelves is another. Magazino is a German startup that’s developing a robot called Toru that can grasp individual objects off of warehouse shelves, a particularly tricky task that Magazino is tackling with ROS.
Moritz Tenorth is Head of Software Development at Magazino. In his ROSCon talk, Moritz describes Magazino’s Toru as “a mobile pick and place robot that works together with humans in a shared environment,” which is exactly what you’d want in an e-commerce warehouse. The reason that picking is a hard problem, as Moritz explains, is perception coupled with dynamic environments and high uncertainty: if you want a robot that can pick a wide range of objects, it needs to be able to flexibly understand and react to its environment; something that robots are notoriously bad at. ROS is particularly well suited to this, since it’s easy to intelligently integrate as much sensing as you need into your platform.
Magazino’s experience building and deploying their robots has given them a unique perspective on warehouse commercialization with ROS. For example, databases and persistent storage are crucial (as opposed to a focus on runtime), and real-time control turns out to be less important than being able to quickly and easily develop planning algorithms and reducing system complexity. Software components in the ROS ecosystem can vary wildly in quality and upkeep, although ROS-Industrial is working hard to develop code quality metrics. Magazino is also working on remote support and analysis tools, and trying to determine how much communication is required in a multi-robot system, which native ROS isn’t very good at.
Even with those (few) constructive criticisms in mind, Magazino says that ROS is a fantastic way to quickly iterate on both software and hardware in parallel, especially when combined with 3D printed prototypes for testing. Most importantly, Magazino feels comfortable with ROS: it has a familiar workflow, versatile build system, flexible development architecture, robust community that makes hiring a cinch, and it’s still (somehow) easy to use.
Michael Ferguson (Fetch Robotics): Accelerating Your Robotics Startup with ROS
Michael Ferguson spent a year as a software engineer at Willow Garage, helping rewrite the ROS calibration system, among other projects. In 2013, he co-founded Unbounded Robotics, and is currently the CTO of Fetch Robotics. At Fetch, Michael is one of the primary people responsible for making sure that Fetch’s robots reliably fetch things. Mike’s ROSCon talk is about how to effectively use ROS as an integral part of your robotics business, including best practices, potential issues to avoid, and how you should handle open source and intellectual property.
Because of how ROS works, much of your software development (commercial or otherwise) is dependent on many external packages. These packages are constantly being changed for the better — and sometimes for the worse — at unpredictable intervals that are completely out of your control. Using continuous integration, consisting of systems that can handle automated builds, testing, and deployment, can help you catch new problems as early as possible. Michael also shares that a useful way to avoid new problems is to not immediately switch over to new software as soon as they are available: instead, stick with long-term support releases, such as Ubuntu 14.04 and ROS Indigo.
While the foundation of ROS is built on open source, using ROS doesn’t mean that all of the software magic that you create for your robotics company has to be given away for free. ROS supports many different kinds of licenses, some of which your lawyers will be more happy with than others, but there are enough options with enough flexibility that it doesn’t have to be an issue. Using Fetch Robotics as an example, Mike discusses what components of ROS his company uses in their commercial products, including ROS Navigation and MoveIt. With these established packages as a base, Fetch was able to quickly put together operational demos, and then iterate on an operating platform by developing custom plugins optimized for their specific use cases.
When considering how to use ROS as part of your company, it’s important to look closely at the packages you decide to incorporate, to make sure that they have a friendly license, good documentation, recent updates, built-in tests, and a standardized interface. Keeping track of all of this will make your startup life easier in the long run. As long as you’re careful, relying on ROS can make your company more agile, more productive, and ready to make a whole bunch of money off of the future of robotics.
If you enjoyed this article, you may also want to read:
- ROSCon 2015 recap and videos – Part 4
- ROSCon 2015 recap and videos – Part 3
- ROSCon 2015 recap and videos – Part 2
- ROSCon 2015 recap and videos – Part 1