Intelligent bounded robust adaptive neural network controller design for fully actuated autonomous underwater vehicles with guaranteed performance using a novel reinforcement learning method

In this paper, a novel amplitude-limited reinforcement learning controller is proposed for fully actuated autonomous underwater vehicles (AUVs) in the presence of the saturating actuators in six degrees-of-freedom with a guaranteed performance. At first, a second-order error dynamic model is expanded through a nonlinear transformation in terms of constrained posture errors to unconstrained ones. The generalized saturation functions are employed to diminish the actuator saturation risk by bounding the transformed errors. An effective combination of a reinforcement learning strategy, a critic function, actor-critic neural networks and a robust adaptive controller is integrated to secure the robustness of the controller versus nonlinear-in-parameter uncertain terms, the effects of the dynamics that cannot be modeled and exogenous disturbances. A Lyapunov-based stability approach is employed to prove that all time-variant variables in the control system will stay semi-globally uniformly ultimately bounded. The tracking errors will also show a funnel convergence behavior by the designed controller. Eventually, the results of simulation along with a comparative study demonstrate the importance and control objectives of the proposed algorithm and confirm the theoretical contributions.