Control of an AUV with completely unknown dynamics and multi-asymmetric input constraints via off-policy reinforcement learning

This paper investigates a novel model-free optimal controller for nonlinear autonomous underwater vehicles (AUVs). It is considered that the AUV considered as the case study is subject to multi-asymmetric constrained inputs. To achieve the optimal controller, a performance index function with exponential discounted value term and input hyperbolic function is developed. Since it is assumed that the AUV dynamics are completely unknown, a model-free integral reinforcement learning (RL) strategy is established. The suggested approach uses the sampled data pairs of input and states. To implement the model-free Integral RL optimal controller, a neural network structure is suggested to estimate the performance index function and control policy. Finally, a numerical simulation and comparative results are given to verify the effectiveness of the proposal.