Reproduce “Learn the Task, Not the Schedule: Reducing Spurious Cross-Arm Coupling in Bimanual Imitation Learning” 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
Shuyuan Hu, Deng-Ke Wu, Yifan Jin, Qianhang Luo, Mingchen Ju, Kai Ye, Bo Zhao
Constraint-preserving Temporal Recombination (CTR) trains bimanual policies to distinguish incidental timing from task-required coordination. It rearranges when recorded arm-local motion segments occur while preserving annotated constraints, exposing policies to varied admissible schedules. On real-robot Sort Letters, π₀.₅’s worst-state success across six progress states rises from 56% with Mixed to 76% with CTR, not establishing robustness across all tasks or states.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.