SuperMimic paper: one differentiable world model replaces cluster-scale RL for humanoid control
alex_verem · x · 2026-11-15
This is the arXiv source (2610.28480) for SuperMimic by UC Berkeley researchers including Pieter Abbeel and Koushil Sreenath. Key points:
- One differentiable world model of shared dynamics replaces model-free RL pipelines (per-motion specialists + lossy distillation, or cluster-scale brute force)
- World model and tracking policy train together; backprop-through-time over short imagined rollouts gives analytic gradients
- 3 hours on a single GPU beats a SOTA baseline trained 7 days on 128 GPUs, reducing GPU-hours by 1000x+ (25/129,514 failures vs 179)
- Zero failures after 9.8 hours on two GPUs; policy transfers to a physical Unitree G1 for diverse acrobatic motions
- Removes compute as the bottleneck, making motion-tracking scaling laws practical to measure
Related event: UC Berkeley's SuperMimic Trains Humanoid Control on a Single GPU in 3 Hours(2 posts)→
More from Embodied
- RaiSim vs Newton benchmark: CPU robot sim beats GPU by up to 95x at practical batch sizes — ChongZzZhang · 2026-11-15
- Researcher teaches a Unitree G1 humanoid to ride a mechanical bull after CoRL — RemiCadene · 2026-11-15
- PrimeSim: world model trained on 30k hours of robot demos runs near real-time on one A100 — jiqizhixin · 2026-11-15
- Viral claim: Claude runs a robot arm on a real factory task with zero training — but the model name doesn't exist — IgorCarron · 2026-11-15
- NEO-VLA ditches visual encoders for robot actions, hitting 90.9% on LIBERO frozen — 机器之心 · 2026-11-15
- Zipline's Keller Cliffton on 3M drone deliveries and what breaks at 10x scale — a16z Podcast · 2026-11-15