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

2026-05-18 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning · Depth: Expert, medium

Summary

Researchers Edoardo Caldarelli, Franco Coltraro, Adrià Colomé, Lorenzo Rosasco, and Carme Torras have developed a novel model predictive controller for dynamic robotic cloth folding. This system addresses the challenge of folding cloth with fast motions by integrating physics-based cloth simulation with efficient, kernel-based Koopman operator regression. The Koopman operator regression, a machine learning technique for nonlinear system identification, generates a linear surrogate model of the cloth dynamics. This model, trained with data from a high-fidelity simulator, replaces costly nonlinear models within the model predictive control algorithm to efficiently generate fast folding trajectories for robotic manipulators. Experiments in both simulated and real-robot environments demonstrate that this Koopman operator-based approach efficiently generates fast folding trajectories to unseen poses without compromising folding accuracy.

Key takeaway

For research scientists developing robotic manipulation systems for deformable objects, this work demonstrates a viable path to achieving both speed and precision. You should consider integrating Koopman operator regression with model predictive control to linearize complex dynamics, enabling faster and more accurate trajectory generation for tasks like cloth folding. This approach can significantly improve simulation-to-reality transfer and overall system performance.

Key insights

Koopman operator regression enables efficient, accurate robotic cloth folding via linearizing complex cloth dynamics.

Principles

• Linear models simplify complex nonlinear dynamics.
• Physics-based simulation aids real-world robot control.

Method

The method uses kernel-based Koopman operator regression to linearize cloth dynamics, then integrates this surrogate model into a model predictive controller for efficient trajectory generation.

In practice

• Apply Koopman operators for nonlinear system identification.
• Use high-fidelity simulators for training data.

Topics

• Robotic Cloth Folding
• Koopman Operator Regression
• Model Predictive Control
• Nonlinear System Identification
• Cloth Dynamics Simulation

Code references

• NEAR-the-future/RC-Koopman

Best for: Research Scientist, AI Scientist, Robotics Engineer

Related on AIssential

• We Hired 10 Humans to Train a Robot to Fold Shirts — Robots can learn complex manipulation tasks like t-shirt folding through teleoperated data collection.
• Learning Control-Affine Reduced-Order Models via Autoencoders — Simultaneous training of autoencoders and state-space models enables robust control-affine reduced-order model identification.
• Making Foresight Actionable: Repurposing Representation Alignment in World Action Models — Aligning visual and action representations in World Action Models improves robot manipulation performance and robustness.
• LadderMan: Learning Humanoid Perceptive Ladder Climbing — LadderMan enables robust humanoid ladder climbing and manipulation through perceptive, sim-to-real learning.
• Japan Science and Technology Agency Develops NVIDIA-Powered Moonshot Robot for Elderly Care — Robots can predict 3D interaction forces from RGB images using simulation-trained AI.
• Learning to Fold: prizewinning solution at LeHome Challenge 2026 (1st place online, 2nd offline) — A bimanual garment folding system leverages a VLA policy with an RL loop, where the policy network also serves as its own value function.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.