Continuous Self-adaptive Calibration by Reinforcement Learning

Abstract

It is well-known that hand-eye calibration plays an important role in the application of vision-based robot systems. Despite traditional calibration methods achieved huge success, the reduction in calibration accuracy whenever the relative hand-eye position changes reflects the fact that such methods are only suitable for scenarios where the components of the robot system are relatively fixed. To tackle this problem, a continuous self-adaptive calibration approach is proposed by applying the deep reinforcement learning algorithm to the calibration task. The experimental results demonstrate that our method can calibrate accurately in more flexible situations where the relative position of the hand and eye changes frequently.