An optimised nonlinear model predictive control based autopilot for an uninhabited surface vehicle

Abstract In this paper a modified nonlinear model predictive control (NMPC) algorithm based on a genetic algorithm (GA) is proposed for the design of an autopilot for the Springer uninhabited surface vehicle (USV). In place of seeking the exact global solution for NMPC at every sampling time, suboptimal control sequences satisfying the constraints are implemented. The GA decreases the cost function within the sampling interval and the best chromosome represents the optimal control sequence at that time and so on. This requires less computational demands without deteriorating much to the control performance. The NMPC autopilot is benchmarked against a PID autopilot scheme to guide the USV in the presence of disturbances through different line-of-sight waypoints stored in a mission planner. The USV was presented as a recurrent neural-network model and the simulation results show that NMPC autopilot scheme outperforms its PID alternative.