A genetic algorithm-based model predictive control autopilot design and its implementation in an autonomous underwater vehicle

Abstract

The control of any underwater vehicle has always been a challenging task and is certainly an important and necessary feature of an autonomous underwater vehicle (AUV). This paper describes the implementation of a genetic algorithm (GA)-based model predictive controller for an AUV named Hammerhead which is being developed jointly by the Universities of Plymouth and Cranfield. To the present authors' knowledge, this is the first successful application of a GA in realtime optimization for controller tuning in the marine sector and thus the paper makes an extremely novel and useful contribution to control system design. The advantages of using model predictive control (MPC) include its constraint handling and disturbance rejection properties commonly present in an underwater environment. The use of GAs generalizes MPC to employ linear as well as nonlinear process models. Furthermore, it supports the inclusion of various types of objective function without having to change the controller structure. The model required for MPC is extracted using system identification techniques on actual AUV data obtained from full-scale in-water tests. A description of Hammerhead AUV is outlined and simulation and experiment data are shown which were obtained by optimizing the cost function online using the GA. Results demonstrate good tracking behaviour despite the presence of disturbances and ever-present modelling uncertainty.