Reproduce “Bridging the Sim-to-Real Gap with Dynamic Compliance Tuning for Industrial Insertion” 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
Xiang Zhang, Masayoshi Tomizuka, Hui Li
Researchers from UC Berkeley and Autodesk Research developed a framework to bridge the sim-to-real gap for contact-rich industrial insertion tasks, achieving robust zero-shot transfer by dynamically tuning robot compliance based on desired contact forces and learned motion plans, outperforming traditional baselines across various complex assembly scenarios.
Key insights
- Decoupling the planning of robot motion and desired contact force from the dynamic adjustment of compliance gains enables more robust sim-to-real transfer.
- Dynamically tuning admittance control gains allows the robot to adapt its interaction force in the real environment to match a desired force profile, mitigating simulation inaccuracies.
- Training both modules with supervised learning on simulation data, augmented with domain randomization and noise, facilitates zero-shot transfer without real-world fine-tuning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.