Reproduce “POIL: Point-based One-Shot Imitation Learning with Stable Dynamical Systems” 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
Sang Min Kim, Jinwoo Seo, Hyeongjun Heo, Junho Lee, Yonghyeon Lee, Young Min Kim
POIL transfers demonstrations by matching task-relevant object-part points across scenes and tracking them during execution. This shared representation corrects object motion through a rigid-body twist instead of replaying gripper actions. In controlled Mug Insertion trials with ground-truth point trajectories, one demonstration achieved 100% success from Far poses, versus 84% for Point Policy with 50 demonstrations. The guarantee covers terminal control under rigid-object assumptions, not full-pipeline performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.