Reproduce “REASSEMBLE: A Multimodal Dataset for Contact-rich Robotic Assembly and Disassembly” 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
Daniel Sliwowski, Shail Jadav, Sergej Stanovcic, Jedrzej Orbik, Johannes Heidersberger, Dongheui Lee
The REASSEMBLE dataset provides a comprehensive, multimodal resource for contact-rich, long-horizon robotic assembly and disassembly, utilizing the standardized NIST Assembly Task Board #1. It includes a unique combination of RGB, proprioceptive, force-torque, audio, and event camera data, along with detailed hierarchical annotations for multi-task learning and anomaly detection.
Key insights
- The 'Insert' action is the most challenging, accounting for the highest number of failures due to its multi-step nature and demand for precise alignment and force application.
- Force-torque data reveals distinct patterns corresponding to different action phases (e.g., free-space movement, contact, pushing, twisting), underscoring its value for understanding contact dynamics.
- Randomizing the NIST Assembly Task Board #1 and object poses for each trial successfully created diverse interaction points, which is essential for training generalizable robotic policies.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.