An online genetic algorithm based model predictive control autopilot design with experimental verification

Wasif Naeem, Robert Sutton, J. Chudley, Fraser Dalgleish, S. Tetlow

International Journal of Control · 2005 · 43 citations · 18 references

Concepts

TL;DR

AUV research is nearing maturity but is constrained by high development costs and a lack of test equipment, limiting most work to simulations. The study aims to demonstrate a low‑cost GA‑based MPC autopilot on a real AUV and to provide a platform for other UK research groups. The authors implemented a GA‑based MPC controller on the AUV and described the vehicle’s hardware and software configuration. The first real‑time deployment of an online GA‑based MPC on an AUV achieved successful LOS tracking with and without an umbilical, demonstrating robust performance under disturbances and modeling uncertainty.

Abstract

Autonomous underwater vehicle (AUV) research and development is on the verge of reaching maturity yet applications are very few. The cost associated with an AUV development particularly of the onboard sensors, power requirements and underwater testing have imposed a significant constraint on its development. Numerous ideas regarding underwater vehicle control have been proposed in the literature however the lack of test equipment confines one to simulations only. The aim of this paper is to introduce fresh results on the application of a genetic algorithm based model predictive controller (GA-MPC) to an actual AUV. These results are significant since to the authors’ knowledge, this is the first known application of an online GA operating in an underwater vehicle in real time. The current hardware and software configuration of the vehicle is also elucidated. The broader aim of the research is to produce a low cost AUV facility to be exploited by other underwater research groups within the UK. The results mainly consists of line of sight (LOS) tracking missions using the GA-MPC autopilot with and without an umbilical. It is demonstrated that the controller performs remarkably well in a real time environment despite the existence of disturbances and ever present modelling uncertainty.

References

18