AI
OTRetarget: Joint Robot and Object Motion Retargeting via Optimal Transport
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 28/09/2026, 20:00:00 UTC
Briefing created: 03/10/2026, 06:34:31 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 02/10/2026, 13:18:24 UTC.
Briefings
OTRetarget introduces a unified approach using entropic optimal transport to jointly retarget robot and object motion from human demonstrations. The method uses signed distances and closest surface points to represent surface interactions, which are transferred across human, robot, and object geometries. Validation on OMOMO achieved an 87% robot-object interaction Jaccard score and 8.7 mm depth error, significantly outperforming OmniRetarget. The approach was tested on a physical G1 humanoid with whole-body policies trained via reinforcement learning for two-handed box pick-and-place tasks.
Key points
- OTRetarget uses entropic optimal transport to transfer signed distances, closest surface points, and relative directions across human, robot, and object geometries.
- The method optimizes robot and object poses at each frame to balance contact preservation with motion style without rescaling the scene or demonstration.
Context and limitations
Validation was limited to OMOMO and a physical G1 humanoid; no quantitative results are provided for other humanoid platforms or specific locomotion tasks.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 03/10/2026, 06:34:31 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeihtms5rsqwolpeonpvpbauuca2xknhco6qmoc7lin6dp6p37yz3sy