At DARPA’s Virtual Robotics Challenge (VRC), everyone’s a winner.
You might think I’m being folksy here, given how when robots are capable of saving our lives in disaster situations, it’s a pretty good thing for all of humankind.
While that’s almost certainly true, the VRC took it more literally. The original plan was to run 26 university-affiliated robotics teams through three virtual trials to produce six winners. In the end, members of nine different teams were awarded advancements.
What happened? According to a DARPA release, several team mergers and no small amount of “good sportsmanship” were involved.
Over the past two weeks, the 26 teams have been judged by their abilities to program a simulated ATLAS robot to complete several skills that would be mega-useful in a disaster situation, especially since they’re all things that would be hazardous for humans. What’s more, DARPA limited communications and vision between the team and the robot, to better reflect the difficulties first responders might face in a real disaster.
Watch the video to see an overview of the three scenarios in action:
The teams all used the Open Source Robotics Foundation (OSRF)’s Gazebo simulator because, as CEO Brian Gerkey predicts, it should translate easily to programming on an actual robot.
“If you come up with a winning solution for [the simulation], then the software that you’ve written for it should, for the most part, transfer to a physical robot in a physical environment and produce qualitatively the same results,” he said in a VRC media roundtable discussion.
The original plan was to award the six winning teams six ATLAS robots with which to advance to the physical part of the DARPA robotics challenge, this December. But with members of nine teams advancing, it’s a little different.
One of the winning teams, Caltech, Jet Propulsion Laboratory, chose to use its own robot and donate its ATLAS to another team. Two other teams merged, and were awarded a seventh ATLAS robot from Hong Kong University. In total, seven teams (comprised of members of nine original teams) will be competing. Whew!
While seven teams advanced, that doesn’t mean they all did equally well. The top performance came from the Institute for Human and Machine Cognition in Pensacola, Fla., with 52 points. The next team to even come close to that was Worcester Polytechnic Institute, from Worcester, Massachusetts, with 39 points. The lowest amount of points a team earned – that had members still advance – was 23 points.
Those score numbers are based on rankings that show how many tasks the team completed, how fast the team completed the task, and how many bits of information they used to communicate with the robot.
Over the next few weeks, DARPA will be publicizing the teams’ performances on its YouTube channel, and then we’ll really be able to see what the difference between a 52 and a 23 looks like.