Event
MimicAll: A Cross-Humanoid Motion Tracking Framework for Sharing
2 reports · 2 sources
Jia Li, Lingyun Xu, Yitang Li, and Hao Zhao release MimicAll, a reinforcement learning framework that lets different humanoid robots share a single whole-body motion tracking policy. It uses structured embodiment descriptors to encode global and joint alignment conditions, trains on 4096 variants covering 16 humanoid robot families, and establishes a benchmark spanning five distribution tiers and 464 embodiments for evaluating motion generalization, within-family embodiment variation, and transfer to unseen families. At the same amount of training transitions, a randomized G1 embodiment achieves a 99.4% 30-second success rate on 100STYLE, versus 42.9% for a single-embodiment specialized policy. On a subset of 8 LAFAN1 motions excluding ground and get-up sequences, when tested across 464 embodiments, embodiment-condition mismatch reduces the 30-second success rate of the self-attention method from 61.7% to 47.0%; the authors accordingly claim that embodiment information yields closed-loop gains.
Report timeline
- Three cross-embodiment papers examined: body, motion, and environment data still need manual translation
The article examines three cross-embodiment research papers published between October 24 and 26, arguing that cross-embodiment is more like translation than a spontaneous model capability. MimicAll uses a structured body-condition table covering 16 series and 4,096 body variants.
人形大讲堂 · 微信公众号 ·
- MimicAll: A Structured Embodiment Conditioning Framework for Cross-Humanoid Whole-Body Motion Tracking
MimicAll is a reinforcement learning framework that encodes global and joint-alignment conditions using structured embodiment descriptors, enabling heterogeneous humanoid robots to share a whole-body motion tracking policy. Training covers 4,096 variants across 16 humanoid robot families, and a benchmark of 464 bodies across five distribution strata was built. With the same amount of training transitions, shared training achieved a 99.4% 30-second success rate on randomized G1 bodies in 100STYLE, versus 42.9% for single-body specialized policies.
arXiv cs.RO ·