Reproduce “FORGE: Force-Guided Exploration for Robust Contact-Rich Manipulation under Uncertainty” 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
Michael Noseworthy, Bingjie Tang, Bowen Wen, Ankur Handa, Chad Kessens, Nicholas Roy, Dieter Fox, Fabio Ramos and 2 more
The FORGE framework enables robotic policies to perform robust contact-rich assembly tasks under significant pose uncertainty by integrating force-guided exploration and dynamics randomization. Policies trained with FORGE achieve superior sim-to-real transfer, exhibit gentler interactions with parts, reduce task completion times, and adapt to varying controller dynamics while demonstrating successful multi-stage assembly.
Key insights
- Explicitly penalizing excessive force during training, coupled with force observations, yields gentler, safer, and more robust contact-rich robot behaviors.
- Randomizing controller gains and robot dynamics significantly improves sim-to-real transfer, allowing policies to adapt to varying real-world system properties without extensive manual tuning.
- Integrating a success prediction mechanism allows for adaptive, efficient task termination and enables automatic tuning of force parameters for novel assembly challenges.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.