Model predictive control of an autonomous underwater vehicle in an in situ estimated water current profile

Autonomous underwater vehicle control actuation is attained through the use of various methods, including propellors, jets and control surfaces. In order for a vehicle to achieve a desired trajectory and fulfil the mission goals successfully, the input of commands to the control subsystem is needed. Model predictive control (MPC) [2] relies on having a function which determines the future vehicle poses to a horizon given the present vehicle pose, and control actions during this time, and then minimising a cost function, such as the squared distance from the predicted to desired vehicle path. The advantage of MPC over other control methods like PID, Linear Quadratic Regulator (LQR) and its derivatives, is that very little hand tuning is required [10]. The method outlined in [9] allows simultaenous estimates of the vehicle pose and the water current profile in the direction of the Acoustic Doppler Current Profiler (ADCP) beams, including small scale gradients in situ. The position, velocity, attitude and water current estimates from this localisation filter could be used to arrive at control commands in real-time to achieve the desired vehicle trajectory given the predicted water current acting on the vehicle and the vehicle pose for future states. Results in this paper show that even with large delays due to the MPC optimisation stage to arrive at control actions, the controller can accurately track the desired trajectory in the mean estimates from the localisation. The trajectory following accuracy is shown to be limited by the localisation error.