arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2610.28480cs.RO
收藏

SuperMimic:用一个可微分世界模型扩展人形机器人全身控制

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

UC Berkeley(加州大学伯克利分校)

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

AI总结:

SuperMimic用可微分世界模型分摊人形运动学习,效率远超基线,可迁移到Unitree G1,消除了计算瓶颈。

摘要:

跟踪大型人类运动数据集是实现通用、高表现力人形机器人全身控制的主流途径,但跨多种运动扩展的成本很高。当前的流程采用无模型强化学习,要么通过有损蒸馏合并各运动的专用策略,要么通过集群规模计算蛮力训练单个策略。SuperMimic利用一个可微分世界模型分摊所有运动的学习成本,该模型基于标准刚体模拟器记录的转换数据拟合,捕捉运动的共享动力学。世界模型与跟踪策略在同一循环中联合训练,通过短时间想象展开的随时间反向传播为策略提供解析梯度。快速想象展开和解析梯度使策略学习效率极高。在单个GPU上训练3小时后,SuperMimic的回报超过了在128个GPU上训练7天的最先进运动跟踪器基线;它将GPU小时数减少了1000倍以上,在129514个基准运动中仅25个失败,而基线有179个失败。在两个GPU上训练9.8小时后,SuperMimic实现零失败。从1小时到264小时的运动数据来看,世界模型的预测误差保持稳定,达到目标回报所需的GPU小时数增长缓慢。该策略可迁移到物理Unitree G1机器人,适用于各种复杂运动。SuperMimic消除了扩展人形机器人全身控制的计算瓶颈,使测量运动跟踪扩展定律成为可能。

英文摘要:

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.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑