Reproduce “One-Shot Learning from Demonstration of Contact-Rich Robotic Manipulation by Identifying Physical Interactions” 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

A. H. G. Overbeek, H. van der Kooij, M. Vlutters

A Learning from Demonstration (LfD) method explicitly models physical interactions—identifying what, where, and when they occur—from a single demonstration to enable robust and adaptive contact-rich robotic manipulation. This approach enhances task reproduction by adapting to unknown geometric variations and generalizing with known prior information.

Key insights

  • Explicitly modeling the 'what, where, and when' of physical interactions provides a human-interpretable framework for one-shot LfD, facilitating debugging and safety guarantees for complex tasks.
  • Combining compliant hybrid position/force control in interaction frames with contact-aware monitoring and judiciously relaxed tolerances offers intrinsic robustness to minor unknown geometric variations by allowing for earlier or later contact events.
  • Online trajectory adaptation based on actual poses achieved at segment ends is crucial for managing larger, previously unknown geometric variations, preventing mechanical interference and improving task completion.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.