Motion tracking, openly evaluated.

MotionChallenge

Public checkpoints, shared evaluation rules and independently verified results. Each model has a reproducible adapter with pinned source, weights and dependencies. LAFAN1 is our first dataset.

Model cards describe the released inference path and any benchmark adaptation. Detailed results include per-motion scores, the exact evaluated code and verification records.

Acknowledgments

Thanks to the controller authors, LAFAN1 and LAFAN1 Retargeting Dataset, MuJoCo and mjlab. The website draws presentation ideas from RoboTwin, RoboCasa, RoboChallenge and MimicLite.