At the DARPA Robotics Challenge Finals this summer, we heard from NASA that they were preparing to send two Valkyrie robots to U.S. universities “for active research of high-level humanoid behaviors” through a competitive selection process. Yesterday, NASA announced the winners: Russ Tedrake’s lab at MIT, and a group led by Taskin Padir, who just moved to Northeastern University from WPI.
Here’s where NASA is coming from on this; note that they refer to Valkyrie as “R5” even though “Valkyrie” is a way cooler name:
NASA is interested in humanoid robots because they can help or even take the place of astronauts working in extreme space environments. Robots, like NASA’s R5, could be used in future NASA missions either as precursor robots performing mission tasks before humans arrive or as human-assistive robots actively collaborating with the human crew. R5 initially was designed to complete disaster-relief maneuvers, however, its main goal is to prove itself worthy of even trickier terrain -- deep space exploration.
The robotics labs at MIT and Northeastern will both get their very own superhero robot on loan, plus tech support from NASA and $250,000 a year for two years for robot support and pizza parties. Here are the two proposals that were selected, representing the entirety of the information that NASA was willing to share with us about what the R5 robots are going to be, you know, doing:
- Robust Autonomy for Extreme Space Environments: Hosting R5 at Massachusetts Institute of Technology in Cambridge, Massachusetts, led by principal investigator Russ Tedrake
- Accessible Testing on Humanoid-Robot-R5 and Evaluation of NASA Administered (ATHENA) Space Robotics Challenge—Northeastern University in Boston, Massachusetts, led by principal investigator Taskin Padir
For a little more detail, we spoke briefly with Russ Tedrake at MIT about what his plans are for Valkyrie. Along with two other MIT CSAIL robotics professors (Leslie Kaelbling and Tomas Lozano-Perez) and (we assume) a big pile of students and postdocs, Tedrake will be developing the kind of high-level autonomy algorithms that were only partly necessary for the DARPA Robotics Challenge. The DRC was very highly structured, and the robots were allowed to focus more on not falling over than on having to make any kind of decision based on their ability to reason. For long term useful autonomy, however, robots will be far more valuable if they can determine on their own whether they should turn a valve, or where to cut into a wall, or which switch to toggle. The low-level motion primitives that were developed for the ATLAS robots at the DRC were a very important building block toward this goal, but Tedrake is now looking forward to “our chance to finish what we’ve started.”
In addition to improving Valkyrie’s high-level reasoning, there’s also a lot of potential for more robust (and even useful) interaction with the environment. Unintentional interaction with the environment generally meant a fall for robots during the DRC, but Tedrake wants his robots to become much more capable:
On both [Valkyrie and ATLAS], my personal goal is to enable them to have a much much richer interaction with the environment. Instead of touching the world with some small carefully enumerate set of potential contact points, I want them to make contact all over their body when appropriate, e.g. to recover from a push by putting an elbow out to catch the wall, or to carry around big heavy boxes by making contact all along the arms and upper legs. That’s a computational challenge for that runs through the entire perception, planning, and control stack. And I think the failure to reason about unexpected contact was the biggest weakness that was exposed in terms of robustness at the DRC. I got a taste, for the first time, at the DRC of watching my humanoid operate off the belay (no safety harness). I want to develop the controls that would make it so we never put these robots on a belay ever again.
We definitely agree that a major weakness of the DRC robots is that (with a few exceptions) they weren’t really designed to work off belay, and most teams did not have any confidence that their robot would be able to recover from a fall. Obviously, this is not realistic for long-term autonomy, or even short term autonomy, for that matter: useful robots must be independent, and they must be able to recover from falls if they happen.
The idea of robots leveraging environmental contact instead of avoiding it is also an interesting one. For example, we’ve seen proposals for robots that can use contact with the environment to help them navigate rugged terrain:
And also experiments showing that robots can use contact with objects to enable them to move things that would otherwise be too heavy:
This may not be exactly what Valkyrie ends up doing at MIT, but it should give you some idea of why it’s a useful problem to take look at.
Meanwhile, over at Northeastern, Taskin Padir has posted an article about what his team will be focusing on. Here’s an excerpt:
Our overarching goal in this basic and applied research and technology development effort is to advance humanoid robot autonomy for the success of future space missions. We will achieve this goal by (1) establishing a tight collaborative environment among our institutions (Northeastern University (NEU) and the University of Massachusetts Lowell (UML)) and NASA’s Johnson Space Center, (2) leveraging our collective DARPA Robotics Challenge (DRC) experience in humanoid robot control, mobility, manipulation, perception, and operator interfaces, (3) developing a systematic model-based task validation methodology for the Space Robotics Challenge (SRC) tasks, (4) implementing novel perception based grasping and human-robot interaction techniques, (5) providing access to collaborative testing facilities for the SRC competition teams, and (6) making the developed software available to the humanoid robotics community. Successful completion of this project will not only progress the technological readiness of humanoid robots for practical applications but also nurture a community of competitors and collaborators to enhance the outcomes of the SRC to be administered by NASA in 2016.
We’ll have more details for you about NASA’s loaner Valkyries in the near future, and we’re looking forward to lots of fun new videos from MIT and Northeastern as they start playing with their new toys. By which I mean, as they start conducting very important and serious research with their government supplied hardware.
[ NASA ]
Evan Ackerman is a senior editor at IEEE Spectrum. Since 2007, he has written over 6,000 articles on robotics and technology. He has a degree in Martian geology and is excellent at playing bagpipes.