Learning-based robust optimal tracking controller design for unmanned underwater vehicles with full-state and input constraints

Abstract

In this article, the optimal tracking control problem for unmanned underwater vehicles (UUVs) with full-state and input constraints under the presence of external disturbances and internal dynamic uncertainties is addressed. To achieve preassigned state constraints on UUVs, the traditional UUVs model is transformed into an unconstrained one by using two different nonlinear mappings (NMs). Then the robust tracking control problem of traditional UUVs model under position/Euler angles and velocity constraints is transformed to an optimal control problem of the transformed system without any constraints. A learning-based optimal control method is designed to solve the optimal control problem of the transformed system by employing the optimized backstepping (OB) paradigm and reinforcement-learning (RL) technique, achieving uniformly ultimately boundedness (UUB) subject to optimal cost. To deal with lumped disturbances for the velocity control loop, a neural-network (NN) identifier is employed and incorporated into actor–critic architecture, attaining robust tracking performance. Due to the adopted nonquadratic cost function with respect to the control input, the optimal control solution is established in the form of a hyperbolic tangent function to handle the input constraints. Compared with traditional PID method and MPC approach, the proposed controller can improve tracking performance of UUV by 32.04% and 79.64%, respectively.