Reproduce “OTRetarget: Joint Robot and Object Motion Retargeting via Optimal Transport” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Guillaume Besset, Erwann Carn, Timothée Carecchio, Valentin Tordjman-Levavasseur, Fabian Schramm, Yann de Mont-Marin, Justin Carpentier, Ajay Suresha Sathya
OTRetarget jointly retargets robot-object motion by transferring surface interaction targets with entropic optimal transport, then solving both poses under constraints. On OMOMO, it achieves median 87% robot–object Jaccard agreement and 8.7 mm depth error, versus 28% and 29.3 mm for OmniRetarget. A large-box case favors OmniRetarget on depth error; evaluation is limited to one robot and flat ground.
Key insights
- Surface-level interaction targets capture contact geometry more directly than skeletal pose targets, including contact location and direction.
- Allowing object poses to adapt can preserve task interactions when the original trajectory is unreachable, without uniformly scaling the scene.
- Optimal-transport correspondences cover extremal surface samples more evenly than nearest-neighbour matching: 14% were unmatched versus 42% for nearest neighbours in the reported comparison.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.