Japanese SCHAFT Takes the Gold at DARPA Robot Challenge 51
savuporo writes "The two days of DARPA's humanoid robotics challenge are now over. 16 teams entered in three categories — custom built humanoid, DARPA supplied Atlas platform, and a non-humanoid form — and competed in eight different tasks. The all-Japanese SCHAFT team scored 27 out of 32 maximum points, followed by IHMC Robotics and Tartan Rescue, with 20 and 18 points. The tasks included challenges like driving a vehicle, climbing ladders and walls, using handheld tools to cut through walls, etc. All robots had a mix of autonomy and teleoperated controls to accomplish the tasks. Full details on scores can be found here. The eight teams that scored highest will get continued funding from DARPA to compete in the final challenge in 2014. Two NASA teams also entered, and the JPL-built non-humanoid RoboSimian placed 5th, whereas the JSC built and touted 'Valkyrie' came out of competition with zero points. Team SCHAFT and Boston Dynamics (building the Atlas platform) were recently acquired by Google."
Reader mikejuk says the scores "[make] the performance sound better than it actually was": " Each task could take 30 minutes and most of the robots took their time and moved as slow as ice. It seems that the teams were precomputing every move and taking a lot of time rather than getting on with the task as quickly as possible. As a result there is farther to go in creating useful rescue bots than the scores might suggest."
I made it through 6 hours (Score:4, Interesting)
Let's talk basics, pre-robotics. Video production 101 at the community college. Whoever put the show on could not sync audio and video, and the final 'closing ceremony' was nothing but a 'technical difficulty' screen. Face it, when we can't use technology to broadcast an event, 60 years after TVs invention, we can not do anything that resembles rescue oriented robotics.
The robots themselves, were lacking. Not the hardware platforms, they seemed ok, but the software sucked. Dabbling around robotics myself, I understand why, and acknowledge the teams efforts. But the fact remains; even for a first attempt, I saw nothing 'promising'. We (as a planet) spend too much emphasis on blinking light arduinos, and far too little time encouraging software skills. Again, from personal experience, I can see how the teams had to use compilers with no remote debugging, probably archaic monolithic code ('C' and Assembler), hard to use cross compiling, and basically the kind of stuff you would only force on a development team if you wanted them to fail.
And while I said I thought the hardware looked OK, I will make an observation; 90% of the time the robots stood there doing nothing, that stupid single LIDAR was spinning its ass off. Was that just to keep its grease warm, or was it indeed a huge bottle neck to have only 1 apparently limited LIDAR? To me, the chassis builder (BDI) should have provided at least 3 LIDARS at varying elevations, and the software to analyse point cloud data and provide solid models of the environment should have been on their own independent processors. I get the feeling entirely too much time was required for the developers to do that themselves. Its something everybody needed to do, it should be provided with the hardware.
Then again, I could be completely wrong. I looked at the simulator code back when it was released but have long since forgotten it. Government funded, intended to be open, but I haven't seen where anyone has published anything. Maybe just too early.