Lilly interviews Stephanie Schneider, a PhD candidate at Stanford working on unconventional space robotics. Schneider explains her work on Reachbot, a long-reach crawling and anchoring robot, which repurposes extendable booms for mobile manipulation. They discuss the challenges and exciting elements of robotic prototyping for low-gravity or otherwise unique environments.

Stephanie Schneider

Stephanie Schneider is a Ph.D. candidate in Aeronautics and Astronautics in the Autonomous Systems Laboratory at Stanford, working with Professor Marco Pavone. She received her BS in Mechanical Engineering from Cornell University in 2014. Stephanie’s research interests include real-time spacecraft motion-planning, grasping and manipulation in space, and adaptive control for autonomous robotics.

Links

• Download mp3 (26.6 MB)
• Subscribe to Robohub using iTunes, RSS, or Spotify
• Support us on Patreon

Last night, I attended Stanford’s unveiling of their newest research vehicle for self-driving. An old Delorean has been heavily modified in order to perform drifting experiments – where you let the rear wheels skid freely.

Stanford managed to get Jamie Hyneman of Mythbusters to host the event so there was a good crowd. He asked “Why a Delorean?” What they should have said was:

“The way I see it, if you’re going to build a self-driving drifting car, why not do it with some style?”

But, instead, they got into the technical reasons for choosing a Delorean.

They called the car Marty and it was launched the day before “Back to the Future Day” — Oct 21, 2015, the day in the second movie where Marty travels into the future.

But back to the present. This car, with rear wheel drive and central engine mount, is not a great car to drive. The engineers have removed the engine and replaced it with dual electric motors from Renovo, creating a car able to drive the two rear wheels independently. This means the software is able to spin the wheels at different rates, and do things that no human driver could ever do, including special types of drifting. The car is already able to turn tighter doughnuts (circles) than a human could.

Normally, drifting is a bad idea. It means a loss of control and a loss of power – the connection of the tires and the road is the sole tool you have to drive and control the car. You would only give it up if you absolutely had to. Perhaps the research will show that there are times where you might want to.

Drifting is usually done for show — it will rarely help you in a race — but Stanford’s team wants to discover whether the robot’s ability to do inhuman driving might offer more “outs” in a dangerous situation, like trying to avoid a collision. A car might twist its wheels (perhaps some day all of its wheels) and spin them at different speeds to enable it to take a path which could avoid an accident.

In effect, it’s like making a vehicle that can drive like a Hollywood stunt car. In movies, stunt drivers often make fairly improbable and impossible moves to avoid accidents. A classic Hollywood scene involves a car titling two wheels to get through a tiny gap. The Stanford team did not propose this, and it’s a pretty hard thing to do, but it’s one way to envisage the general idea.

Up to now, research on accident avoidance has been fairly low-key. After all, the main task is to be able to drive safely in the lane you are supposed to be in. But eventually, teams will focus on what to do when things go wrong. For now, though, the priority is to make sure things don’t go wrong. Someday, they may even focus on the infamous trolley problem.

Generally, drift or not, robots should become very good at avoiding accidents. They will have detailed knowledge of the physics of their tires, they will calculate without panic and will be able to drive with full confidence, missing obstacles by very thin margins while staying safe. A human can’t navigate a space only a few inches wider than the car with confidence, but a robot could. A robot will always use the optimal combination of steering and braking, which humans need a lot of training to achieve. Your tires can give you braking force or steering force, but you must reduce one to get more of the other, so often the best strategy is to brake first and then steer, though the human instinct is to do both.

Stanford’s car is not super autonomous. It is meant to do test algorithms in private open spaces. So it won’t be avoiding obstacles or plotting lanes on a highway, it will be testing how a computer can get the most use from the car’s tires.

This article originally appeared on robocars.com.

In a surprise move today, Toyota held a press conference (see video below) announcing a substantial investment in robotics and AI research to develop “advanced driving support” technology, with former Program Manager of DARPA’s DRC Gill Pratt directing the overall project as Executive Technical Advisor. Toyota will allocate USD$50M over the next five years in a partnership with MIT’s CSAIL (headed by Daniela Rus) and Stanford’s SAIL (headed by Fei-Fei Li) to develop research facilities in Stanford and Cambridge.

Two main themes from the press conference underline the tone of this effort:

First,  Toyota doesn’t refer to autonomous cars but to ‘advanced driving support’ . And this makes sense because, unlike Google (with its fully autonomous car), Toyota’s origins are in car manufacturing and not in software development. Car manufacturers have for some time been introducing ever rising levels of autonomy in their cars and Toyota is following that vector. There are legal advantages to having the driving be ultimately responsible for the car’s movements and therefor any possible accidents; when a vehicle doesn’t respond as expected, a costly product liability suit could follow, as Toyota well knows. Thus what appears at one level to be a more conservative approach to autonomy on the road may in essence be a way to bypass potential legal nightmares and bring autonomous driving technology to the market sooner.

Kiyotaka Ise, Toyota Motor Corporation Senior Managing Officer, repeatedly stated hybrid cars as an example of a specific technology that Toyota pioneered (along with Honda) and brought to market even while it was borderline unprofitable due to its large cost at the time. This offers a possible hint of what may follow.

The second interesting point made clear during Toyota’s Q&A is that this project is only part of the auto manufacturer’s investment in the field. $50M is a lot of money, but for a profitable car maker the size of Toyota, it isn’t much. It looks like Toyota wants to establish a strong presence in the US by collaborating with two of the best research centers in the world (the budget will be equally split between the two universities) … but being a truly global company, it is developing other projects elsewhere as well. It’s worth mentioning that almost all large car manufacturers have similar research projects, and they usually collaborate with research institutes, universities or with each other.

In this respect, Toyota’s decision to bring on Pratt as the project lead appears astute. Pratt’s high visibility role as Program Manager of DARPA’s DRC has left him with significant experience managing and promoting large, big-budget projects, and has provided him with deep connections into US research labs.

You can read Toyota’s press release here for all the details. Automaton has also an excellent coverage of the story (see Toyota Announces Major Push Into AI and Robotics, Wants Cars That Never Crash) and says they will be releasing more in-depth stories about the deal early next week. You can watch the full press conference below:

Source: Stanford University  

In episode twelve we talk with Andrew Ng, Chief Scientist at Baidu, about how speech recognition is going to explode the way we use mobile devices and his approach to working on the problem.

We also discuss why we need to prepare for the economic impacts of machine learning. We’re introduced to Random Features for Large-Scale Kernel Machines, and talk about how using this twist on the Kernel trick can help you dig into big data. Plus, we take a listener question about the size of computing power in machine learning.

In particular, Anderson promoted one idea that deserves the attention of the drone industry: that by operating in a legal “grey area” (read: deliberately violating regulations that are perceived as unreasonable and unenforceable) drone enthusiasts will force regulatory movement, and this is a good thing. Anderson gave two examples of circumstances that challenge current regulations: Lego Mindstorms being classified by regulators as a cruise missile guidance system (and hence violating export controls), and flying drones in contravention to the FAA regulations governing the use of unmanned aircraft.

Technologists in general (and Silicon Valley in particular) take a dim view of regulations promulgated under the old order (e.g. Lyft, Uber, AirBnB), and it’s true that some aspects of the FAA regulations are out-of-date and no longer reasonable. But while I’m not completely outraged when technologists upend the status quo to make something new (and improved), this attitude works better in industries that don’t involve fast-flying heavy objects.

Let’s remember that the FAA’s primary goal is to keep people from getting killed by aircraft. When it comes to aircraft regulation – manned or unmanned – people on the ground are the main beneficiaries. We are not talking about regulating a relatively low-risk activity (like renting your room out with AirBnB); we are talking about regulating heavy objects hurtling over the heads of people who haven’t consented to be part of the experiment.

Our society rightly asks the government to ensure that activities that impose risk on others — especially those who did not consent — be minimized. We need to update our regulations, but aircraft operators need to respect the letter and spirit of the law as it stands.  What standards do we have if not the law? If we follow the path of breaking all the rules, someone is going to accidentally kill a sympathetic bystander. Beyond the personal tragedy this will create, such an accident will set back our industry and the benefits we can provide to society by a decade.

In Afghanistan, one of my planes almost smacked into a helicopter, but it was the helicopter – not the drone –  that had come, without clearance or radio calls, into an active artillery firing ROZ (restricted operating zone: this is an airspace control measure that makes sure that aircraft don’t run into artillery fire).  Similarly, the first full-sized drone and manned aircraft collision involved a C-130 violating airspace control measures around an airfield.

Pilots are not infallible and often break the rules. The best drone operators have a different safety culture: one where there is proper approval for everything, because every move will be recorded and second guessed. Military drone safety culture is a great example of this approach. I hope this culture will permeate the civilian unmanned aircraft community as well.

Although breaking the rules might move us toward our unmanned enabled future a little bit faster, this is an incredibly dangerous path for our industry and our bystanders.  The closing thought of VLAB Drones was that unmanned aircraft will eventually make the airspace safer for all users.  The panel wondered if this was hyperbole, but it is inevitable when drones have a strong safety culture. Conversely, as long as we are the irresponsible jerks of the air, safety conscious regulators like the FAA will be unsupportive of us flying.  When we, as a civilian unmanned aircraft industry, can be counted on as strong safety partner, and when people in general and commercial aviation are learning safety lessons from us, there will be no more foot dragging. We will get our airspace and the drone revolution will finally be here.

In the meantime: seriously, don’t do anything that could kill anyone, please. Especially if it is illegal. You will ruin it for the rest of us.