Reinforcement Learning Strategy-Based Adaptive Tracking Control for Underactuated Dual Ship-Mounted Cranes: Theoretical Design and Hardware Experiments

Abstract

As a flexible transportation equipment, the dual ship-mounted crane (DSMC) systems are widely used to transport cargos/goods under complex marine and harbor environments. However, automatic control of such complex systems still faces significant challenges due to their underactuated characteristics, unexpected sea wave disturbances, and uncertain system parameters. Most existing control methods are based on accurate dynamics model or linearized models, which can hardly suppress unknown interferences or may badly decreasing control effects when there exist system uncertainties. To solve the above problems, a reinforcement learning based adaptive tracking control method is proposed in this article, which can obtain a satisfactory control performance without accurate system parameters. Specifically, an actor and a critic neural network are constructed to execute the reinforcement learning (RL) algorithm, for which, the actor-network executes the control input, and the critic network judges the control performance and feedback reinforcement signal to the action network. In addition, a robust integral of the sign of error feedback signal is introduced to improve the robustness of the system. Based on Lyapunov stability theory, it is proved that the tracking error can converge to zero asymptotically under the proposed controller. Finally, hardware experimental results show the effectiveness and robustness of the proposed controller.