UC Berkeley's SuperMimic Trains Humanoid Control on a Single GPU in 3 Hours

UC Berkeley researchers, including Pieter Abbeel, released SuperMimic (CoRL 2026), which uses a single differentiable world model to train humanoid whole-body control on one GPU in about 3 hours, cutting compute requirements by roughly 1000x.

2026-11-15 ~ 2026-11-15 · 2 related posts

• Berkeley's SuperMimic trains humanoid control on 1 GPU in 3 hours, cutting GPU-hours by 1000x — alex_verem · 2026-11-15
• SuperMimic paper: one differentiable world model replaces cluster-scale RL for humanoid control — alex_verem · 2026-11-15