Novel Methods Inspired by Reinforcement Learning Actor-Critic Mechanism for Eye-in-Hand Calibration in Robotics

Eye-in-hand camera calibration is an essential step in robotic vision systems. Traditional methods perform the calibration via geometry computation methods where the calibration points are acquired by moving the robot arm while capturing images of a static object (often a checkerboard) of known geometry. Our work aims to explore deep reinforcement learning (DRL) techniques for solving common calibration optimization problems. Inspired by the actor-critic mechanism of DRL, we propose a novel approach to perform eye-in-hand camera calibration on the robot arm. Numerical and real-world experiments prove the feasibility of the proposed method as such here we show the potential of the reinforcement learning method to solve the camera calibration problem.