Reproduce “A Flow Matching Framework for Soft-Robot Inverse Dynamics” 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

Hang Yang, Fangju Yang, Yangming Zhang, Ibrahim Alsarraj, Yuhao Wang, Zhenye Luo, Zixi Chen, Ke Wu

A flow-matching framework is presented for learning soft-robot inverse dynamics, enabling stable open-loop feedforward control. The RF-FWD variant reduced simulation RMSE to 3.5 mm and achieved physical tracking at 1.14 m/s with 20.65 mm RMSE, all with sub-millisecond inference.

Key insights

  • Generative transport models, specifically Rectified Flow, can effectively learn complex, non-unique inverse dynamic mappings for soft robots, outperforming deterministic regression approaches in tracking accuracy.
  • Integrating domain-specific knowledge, either through a quasi-static physical prior or a forward-dynamics consistency loss, significantly enhances the physical plausibility and robustness of learned inverse dynamics.
  • The flow-matching framework enables high-speed open-loop control with sub-millisecond inference latency, making it practical for real-time applications in soft robotics.

Summary by alphaXiv.

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