OTRetarget: Joint Robot and Object Motion Retargeting via Optimal Transport
OTRetarget jointly retargets humanoid and object motion from human demos using entropic optimal transport.
OTRetarget jointly transfers human demonstrations to a humanoid and the objects it manipulates, preserving ground and object contacts despite different body proportions. Interactions are represented with signed distances, closest surface points, and relative directions, then mapped across geometries using entropic optimal transport inside constrained inverse kinematics. On OMOMO it reports an 87% robot-object interaction Jaccard score and 8.7 mm depth error, versus 28% and 29.3 mm for OmniRetarget. Retargeted references trained reinforcement-learning whole-body policies that were demonstrated on a physical G1 humanoid for two-handed box pick-and-place.
- Jointly optimizes robot and object poses without rescaling the scene.
- OMOMO result: 87% interaction Jaccard and 8.7 mm depth error.
- OmniRetarget baseline scored 28% Jaccard and 29.3 mm error.
- RL whole-body policies ran two-handed box placement on a physical G1.
- Products
- G1
From the original article
1 min read at huggingface.co
Transferring human motion to humanoid robots requires adapting the demonstrated motion to the robot morphology while preserving interactions with the environment. This is particularly challenging for loco-manipulation tasks, where contacts with the ground and manipulated objects must remain consistent despite differences in body proportions.
By Guillaume Besset, Erwann Carn, Timothée Carecchio · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.