AUTHOR=Hong Tinghe , Li Weibing , Huang Kai 

TITLE=A reinforcement learning enhanced pseudo-inverse approach to self-collision avoidance of redundant robots

JOURNAL=Frontiers in Neurorobotics

VOLUME=Volume 18 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2024.1375309

DOI=10.3389/fnbot.2024.1375309

ISSN=1662-5218

ABSTRACT=Redundant robots are more flexible than non-redundant ones, at the cost of increased collision risks when the end-effector moves close to the robot's own links. Redundant degrees of freedom (DoFs) provide the possibility of collision avoidance. However, it is still challenging to determine an appropriate inverse kinematics (IK) solution among the infinite solutions. In this work, a reinforcement learning (RL) enhanced pseudo-inverse approach is proposed for self-collision avoidance of redundant robots. The RL agent takes effect in the redundancy resolution process of a pseudo-inverse method to attain an appropriate IK solution to avoid self-collisions during task execution. Moreover, an improved replay buffer is employed to enhance the performance of the RL algorithm. Simulations and experiments demonstrate that the proposed method effectively reduces the risk of self-collision of redundant robots.