Watch Less, Feel More: Sim-to-Real RL for Generalizable Articulated Object Manipulation via Motion Adaptation and Impedance Control

Tan-Dzung Do, Gireesh Nandiraju, Jilong Wang, He Wang

Keywords: Sim-to-Real, Reinforcement Learning, Impedance Control

Abstract: Fine manipulation tasks like articulated object manipulation pose a unique challenge as the object itself represents a dynamic environment. In this work, we present a novel RL-based pipeline equipped with variable impedance control and motion adaptation for generalizable articulated object manipulation, focusing on smooth and dexterous motion during zero-shot sim-to-real transfer. To mitigate the sim-to-real gap, our pipeline diminishes reliance on vision by extracting useful low-dimensional data via off-the-shelf modules and inferring object motion and intrinsic properties via observation history. Furthermore, we develop a well-designed training setting with great randomization and a specialized reward system that enables multi-staged, end-to-end manipulation without heuristic motion planning. To the best of our knowledge, our policy is the first to report 84% success rate for extensive real-world experiments with various unseen objects. Project website: https://watch-less-feel-more.github.io/

Submission Number: 13