MuscleMimic: open-source benchmark controls all 354 human muscles, zero-shot on chained movements
TinfoilTricorn · x · 2026-10-04
EPFL's amathis lab released MuscleMimic, a JAX-based motion imitation benchmark and generalist for biomechanically accurate, muscle-actuated full-body models, controlling 354 muscles with zero-shot generalization on chained movements.
- GPU-parallel training via MuJoCo/MJWarp for locomotion and manipulation
- Ships retargeted motion datasets plus C3D tooling for marker fitting and AMASS/SMPL-H export
- Code on GitHub (259 stars), with an arXiv preprint and blog post
More from Embodied
- Meta AI Chief Alexandr Wang Praises Rokid Glasses Running Local LLM Anime Companion — alexandr_wang · 2026-10-04
- Roboticists Should Benchmark Success Rates, Not Moral Thresholds, Argues Physical AI Researcher — micoolcho · 2026-10-04
- Open-source Hermes Gadget SDK turns an ESP32 into a voice terminal for your Hermes Agent — Teknium · 2026-10-04
- Tesla FSD hands-on: 7 cameras react 5x faster than human drivers — tristanbob · 2026-10-04
- Crab Rave: open-sourced jumping crab robot that people loved — Anen-o-me · 2026-10-04
- NVIDIA GEAR's SONIC whole-body control ported to humanoid Asimov, running at 2.6ms per tick on ARM — freelerobot · 2026-10-04