Designing an optimal ship course-keeping system for an unknown object model via adaptive dynamic programming approach

Abstract

The paper discusses the problem of ship autopilot design based on the adaptive (approximate) dynamic programming approach (ADP). The main objective of this paper is to propose the design of a ship's course controller within the framework of computational intelligence techniques, i.e. (in our case) based on model-free optimal regulation. This method is attractive in that it does not require knowledge of the system model to generate optimal process control. The key idea is to iteratively solve the Hamilton-Jacobi-Bellman (HJB) equation based on a simple value function approximator structure. In this way, by introducing such basic elements as evaluation and improvement, we are able to carry out an intelligent optimization process based only on the data received online. Thus, this is a kind of data-driven control methodology. The simulations of the ship course change process confirmed the high efficiency of the proposed controller. An important feature of this approach is also the elimination of the tuning problem.