Reproduce “Learning Forward & Reverse Skills from a Single Unfinished Demonstration for Constrained Manipulation Tasks” 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
Yexin Hu, Haoyi Zheng, Johannes Heidersberger, Dongheui Lee
A framework from TU Wien and DLR enables robots to learn both forward and reverse skills for contact-rich manipulation from a single, potentially unfinished demonstration, utilizing screw primitive modeling and admittance control. This approach demonstrated 100% success in forward and reverse tasks across various constrained scenarios using complete demonstrations.
Key insights
- Geometric screw representations provide a robust and generalizable way to model constrained motions in contact-rich tasks, outperforming simple trajectory-based methods.
- A twist-direction-based trajectory segmentation algorithm accurately identifies distinct contact phases, proving more reliable for short, constrained motions than existing segmentation methods.
- Integrating explicit screw primitive modeling with interaction-aware admittance control enables robust skill execution and extrapolation beyond the demonstrated trajectory, even from unfinished data.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.