Dexterous in-hand manipulation of slender cylindrical objects through deep reinforcement learning with tactile sensing

Abstract

Continuous in-hand manipulation is an important physical interaction skill, where tactile sensing provides indispensable contact information to enable dexterous manipulation of objects. In this work, we propose a learning-based framework for dexterous in-hand manipulation that controls the pose of a thin cylindrical object, such as a long stick, to track various continuous trajectories, through multiple contacts of three fingertips of a dexterous robot hand with tactile sensor arrays. We extract the contact information between the stick and each fingertip from the high-dimensional tactile information and show that the robot can effectively learn a policy to achieve the task. The policies are trained with deep reinforcement learning in simulation and successfully transferred to real-world experiments, using coordinated model calibration and domain randomization. We compare the effectiveness of different types of tactile information and find out that the policies trained with contact center positions achieve best tracking results. The sim-to-real performances are validated through real-world experiments.