DexCtrl: Towards Sim-to-Real Dexterity with Adaptive Controller Learning

TL;DR

DexCtrl introduces an adaptive learning framework that jointly optimizes actions and controller parameters to improve sim-to-real transfer in dexterous manipulation, reducing manual tuning and enhancing robustness.

Contribution

The paper presents a novel adaptive controller learning method that automatically tunes control parameters during execution, addressing the sim-to-real gap in dexterous manipulation.

Findings

Improved transfer performance across various dexterous tasks.

Enhanced robustness in real-world force interaction scenarios.

Reduced need for manual controller tuning.

Abstract

Dexterous manipulation has seen remarkable progress in recent years, with policies capable of executing many complex and contact-rich tasks in simulation. However, transferring these policies from simulation to real world remains a significant challenge. One important issue is the mismatch in low-level controller dynamics, where identical trajectories can lead to vastly different contact forces and behaviors when control parameters vary. Existing approaches often rely on manual tuning or controller randomization, which can be labor-intensive, task-specific, and introduce significant training difficulty. In this work, we propose a framework that jointly learns actions and controller parameters based on the historical information of both trajectory and controller. This adaptive controller adjustment mechanism allows the policy to automatically tune control parameters during execution, thereby mitigating the sim-to-real gap without extensive manual tuning or excessive randomization. Moreover, by explicitly providing controller parameters as part of the observation, our approach facilitates better reasoning over force interactions and improves robustness in real-world scenarios. Experimental results demonstrate that our method achieves improved transfer performance across a variety of dexterous tasks involving variable force conditions.