Trajectory tracking algorithm for autonomous vehicles using adaptive reinforcement learning

The off-shore industry requires periodic monitoring of underwater infrastructure for preventive maintenance and security reasons. The tasks in hostile environments can be achieved automatically through autonomous robots like UUV, AUV and ASV. When the robot controllers are based on prespecified conditions they could not function properly in these hostile changing environments. It is beneficial to count with adaptive control strategies that are capable to readapt the control policies when deviations, from the desired behavior, are detected. In this paper, we present an online selective reinforcement learning approach for learning reference tracking control policies given no prior knowledge of the dynamical system. The proposed approach enables real-time adaptation of the control policies, executed by the adaptive controller, based on ongoing interactions with the non-stationary environments. Applications on surface vehicles under non-stationary and perturbed environments are simulated. The presented simulation results demonstrate the performance of this novel algorithm to find optimal control policies that solve the trajectory tracking control task in unmanned vehicles.