5 ms·

They went around the actual hand-like manipulators to lift boxes from the top & bottom edges like human hands would but with vacuum suckers. Almost reminds me

They went around the actual hand-like manipulators to lift boxes from the top & bottom edges like human hands would but with vacuum suckers. Almost reminds me of the last DARPA challenge for humanoid robots where KAIST cheated with a robot which simply rolled from one place to another & did super slow stairs one leg at a time, which was exactly what the competition was supposed to overcome.

wodenokoto 6y ago
This? https://www.youtube.com/watch?v=v6-heLIg85o https://www.youtube.com/watch?v=v6-heLIg85o Can you explain more about how they "cheated"? it looks like they were awarded 1st prize.
- mitchdoogle 6y ago
  Maybe OP has an ideal these guys didn't live up to. They designed the robot according to the course - there's no sense in which that is cheating.
  - bmcahren 6y ago
    I think he's saying the spirit of the competition was failed because it took over 1 minute per step which if this was a consumer or commercial product...would have been a failure. 4 steps look like it took 5+ minutes. Hard to tell because they sped up the footage 20X to make it look reasonable.
  - villgax 6y ago
    The purpose was to have resilient robots for disaster operations, and to do away with slower robotic platforms.
- dylan604 6y ago
  Did that robot just jerk the plug from the wall by pulling the cord rather than the plug itself? That's such a no-no. Even though there's no actual plug, but what else was that supposed to simulate?
- 6gvONxR4sf7o 6y ago
  Presumably the test designers designed it in a way hoping that if something can pass test X, then it must also be able to do Y, making it a good test of Y abilities. Then someone comes along and figures out how to pass the test without being able to do Y. You see it all the time in ML stuff. "Only a true AI could X" (with an implicit "and that would imply Y") and then people figure out how to build a system that does X but doesn't imply Y. Turns out it's amazingly hard to build good standardize tests that don't fall into "When a measure becomes a target, it ceases to be a good measure."
SwiftyBug 6y ago
Did any of the competitors do better?
- villgax 6y ago
  Boston Dynamic's Atlas would have killed it at its current development stage.

wodenokoto 6y ago

This? https://www.youtube.com/watch?v=v6-heLIg85o https://www.youtube.com/watch?v=v6-heLIg85o Can you explain more about how they "cheated"? it looks like they were awarded 1st prize.

mitchdoogle 6y ago
Maybe OP has an ideal these guys didn't live up to. They designed the robot according to the course - there's no sense in which that is cheating.
- bmcahren 6y ago
  I think he's saying the spirit of the competition was failed because it took over 1 minute per step which if this was a consumer or commercial product...would have been a failure. 4 steps look like it took 5+ minutes. Hard to tell because they sped up the footage 20X to make it look reasonable.
- villgax 6y ago
  The purpose was to have resilient robots for disaster operations, and to do away with slower robotic platforms.
dylan604 6y ago
Did that robot just jerk the plug from the wall by pulling the cord rather than the plug itself? That's such a no-no. Even though there's no actual plug, but what else was that supposed to simulate?
6gvONxR4sf7o 6y ago
Presumably the test designers designed it in a way hoping that if something can pass test X, then it must also be able to do Y, making it a good test of Y abilities. Then someone comes along and figures out how to pass the test without being able to do Y. You see it all the time in ML stuff. "Only a true AI could X" (with an implicit "and that would imply Y") and then people figure out how to build a system that does X but doesn't imply Y. Turns out it's amazingly hard to build good standardize tests that don't fall into "When a measure becomes a target, it ceases to be a good measure."

SwiftyBug 6y ago

Did any of the competitors do better?

villgax 6y ago
Boston Dynamic's Atlas would have killed it at its current development stage.