Event
Pieter Abbeel on SuperMimic Training Speedup of Over 1,000x
1 reports · 1 sources
Pieter Abbeel shared SuperMimic and called it a game changer: the method trains a single policy for humanoid whole-body control on 129,514 motions, and using 1 GPU and 3 hours it surpasses the previous best result that used 128 GPUs and 7 days, reducing GPU hours by more than 1,000x. Abbeel said the key lies in two points: using a world model that is faster than conventional simulation, and collecting far more signal than policy gradients through backpropagation through time.
Report timeline
- SuperMimic cuts GPU hours for humanoid full-body control training by over 1,000x
SuperMimic trains a humanoid full-body control policy on 129,514 motions, surpassing the previous best result of 128 GPUs over 7 days with 1 GPU in 3 hours, reducing GPU hours by over 1,000x. Pieter Abbeel shared it and called it a game changer, noting that the key lies in using a world model faster than conventional simulation and in collecting far more signal than policy gradients through backpropagation through time. The project page is https://supermimic.github.io, and the demo is at https://supermimic.github.io/demo/.
Pieter Abbeel ·