Vessel-to-Vessel Motion Compensation with Reinforcement Learning

Actuation delay poses a challenge for robotic arms and cranes. This is especially the case in dynamic environments where robot arm or objects it trying to manipulate are moved by exogenous forces. In this paper, we consider task of using compensate relative motion between two vessels at sea. We construct hybrid controller that combines an Inverse Kinematic (IK) solver with Reinforcement Learning (RL) agent issues small corrections IK input. The solution empirically evaluated simulated environment under several sea states actuation delays. observe more intense waves larger delays have adverse effect on controller's ability vessel motion. RL shown be effective mitigating large parts these errors, both average worst case. Its modest requirement sensory information, combined inherent safety only making adjustments, also makes promising approach real-world deployment.