Reproduce “Dynamic robotic cloth folding with efficient Koopman operator-based model predictive control” 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

Edoardo Caldarelli, Franco Coltraro, Adrià Colomé, Lorenzo Rosasco, Carme Torras

Researchers developed a Model Predictive Control framework combining a high-fidelity physics-based simulator with an efficient Koopman operator-based data-driven model to enable a robot to perform fast, accurate cloth folding with a single corner grasp and continuous motion. The system achieved successful zero-shot sim-to-real transfer, demonstrating average relative folding errors of 5.18% for wool and 2.93% for denim on a UR5 robot.

Key insights

  • The dimensionality of the Koopman surrogate state and the MPC prediction horizon are crucial parameters that directly impact the accuracy and effectiveness of the dynamic cloth folding trajectories.
  • An initial 'motion towards the center' of the cloth is a pivotal strategy for successful dynamic folding, which naturally emerges when the MPC prediction horizon is sufficiently long.
  • Leveraging Koopman operator regression allows for the linearization of complex, nonlinear cloth dynamics, enabling efficient, derivative-free trajectory computation suitable for real-time optimal control.

Summary by alphaXiv.

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