Reproduce “Safe and Steerable Geometric Motion Policies for Robotic Dexterous Manipulation” 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

Albert Wu, Riccardo Bonalli, Thomas Lew, C. Karen Liu

SafePBDS introduces a framework for robotic dexterous manipulation that provides hard safety guarantees and external steerability by extending Pullback Bundle Dynamical Systems with Pullback Control Barrier Functions and a Task Manifold Action Interface. This approach successfully enabled 92.5% grasp success on diverse objects and robust palm-down in-hand reorientation on a 23-DOF robotic platform.

Key insights

  • Geometric consistency from Pullback Bundle Dynamical Systems can be combined with hard safety guarantees via Pullback Control Barrier Functions, ensuring robust constraint adherence across different charts and metrics.
  • Low-dimensional external control inputs can effectively steer complex, high-dimensional robotic manipulation behaviors while rigorously preserving safety guarantees, bridging autonomy and higher-level guidance.
  • Principled, model-based control frameworks are capable of achieving advanced dexterous manipulation capabilities, such as robust palm-down in-hand reorientation, without relying on extensive data-driven methods or specialized hardware.

Summary by alphaXiv.

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