SuperMimic: Scaling Humanoid Whole-Body Control with One Differentiable World Model

SuperMimic: Scaling Humanoid Whole-Body Control with One Differentiable World Model

|Authors: Liao, Qiayuan, Shao, Yiyang, Long, Junfeng, Zakka, Kevin, Horowitz, Roberto, Abbeel, Pieter, Sreenath, Koushil

|Venue: arXiv preprint | Year: 2026

|arXiv: 2610.28480

Abstract

Tracking large human-motion datasets has become the dominant route to general, expressive humanoid whole-body control, but scaling across many motions is expensive. Current pipelines use model-free reinforcement learning, either merging per-motion specialists through lossy distillation, or brute-forcing a single policy with cluster-scale compute. SuperMimic amortizes learning across all motions with one differentiable world model of their shared dynamics, fit from logged transitions of a standard rigid-body simulator. The world model and the tracking policy train together in one loop, and backpropagation-through-time over short imagined rollouts gives the policy analytic gradients. Fast imagined rollouts and analytic gradients make policy learning efficient. In 3 hours on a single GPU, SuperMimic surpasses the return of a state-of-the-art motion tracker baseline trained for 7 days on 128 GPUs. It reduces GPU-hours by more than $1000\times$ and fails on only 25 of 129,514 benchmark motions, versus 179 failures for the baseline. After 9.8 hours on two GPUs, SuperMimic achieves zero failures. From 1 to 264 hours of motion, the world model’s prediction error stays stable, and GPU-hours to a target return grow slowly. Our policy transfers to a physical Unitree G1 across diverse and acrobatic motions. SuperMimic removes compute as the bottleneck to scaling humanoid whole-body control, making it practical to measure motion-tracking scaling laws.

Key Contributions

• Tracking large human-motion datasets has become the dominant route to general, expressive humanoi…
• Current pipelines use model-free reinforcement learning, either merging per-motion specialists th…
• SuperMimic amortizes learning across all motions with one differentiable world model of their sha…
• The world model and the tracking policy train together in one loop, and backpropagation-through-t…
• Fast imagined rollouts and analytic gradients make policy learning efficient

Topics

• reinforcement-learning
• control
• benchmark

Code & Data

No code repository linked in paper metadata.

BibTeX

@article{Qiayuan2026_261028480,
  title     = {SuperMimic: Scaling Humanoid Whole-Body Control with One Differentiable World Model},
  author    = {Liao, Qiayuan and Shao, Yiyang and Long, Junfeng and Zakka, Kevin and Horowitz, Roberto and Abbeel, Pieter and Sreenath, Koushil},
  year      = {2026},
  eprint    = {2610.28480},
  archivePrefix = {arXiv},
  primaryClass  = {cs.RO},
  url       = {https://arxiv.org/abs/2610.28480}
}