"To construct what it's calling an "omni-bodied robot brain," Skild trained an AI to "control not just one robot, but a whole multiverse of robots with different bodies," according to a recent blog post. "It cannot memorize the solution for one body, it must find a strategy that works across all of them." "We created a universe with 100,000 different robots and trained our AI to control them all,""
"The video is as disconcerting as it is impressive, demonstrating the effectiveness of an AI that can seemingly be dropped into pretty much any robot body - even a severely mutilated one - and still adapt and move. Even with all four of its limbs lopped off, the robodog starts to hobble around almost immediately - albeit in a far less dignified way."
Skild AI developed an "omni-bodied robot brain" by training a controller across a simulated multiverse of robot morphologies so it must learn general strategies rather than memorizing single-body solutions. The training environment included 100,000 different robots to encourage robustness and transfer. Physical demonstrations show the controller enabling movement after severe damage such as lopped-off limbs, jammed motors, broken legs, stuck wheels, and walking on stilts. The controller often adapts to scenarios unlike those seen in training, producing early signs of robustness and the ability to handle unforeseen real-world failures.
Read at Futurism
Unable to calculate read time
Collection
[
|
...
]