Optimal Control Models for the Autonomous Underwater Vehicle LoLo

Examiner: Jonas Mårtensson

Abstract: The main goal with this thesis was to develop an optimal control model for the autonomous underwater vehicle (AUV) LoLo and evaluate whether reference tracking using Model Predictive Control (MPC) based on a linear dynamics model describing all six degrees of freedom (DOF) is a suitable method for waypoint navigation. MPC is an advanced receding horizon optimal control method capable of including constraints in the optimization. The drawback with this method is that it is computationally heavy and since the dynamic behavior of LoLo is both complex and non-linear there are two open research questions: Does MPC improve reference tracking and is it computationally feasible? Optimal control requires accurate dynamics models of the targeted system to function and ensure robustness. At the start of this thesis LoLo was an unmodelled vehicle and therefore modelling, model analysis and linearization techniques constitutes a big part in this project and creating a non-linear dynamics model was a big part of this project. To ensure that reference tracking using optimal control strategies was feasible, state-feedback solutions together with standard techniques for stability analysis, controllability and observability were investigated. A linear quadratic regulator (LQR) was designed using a linear time-invariant (LTI) dynamics model augmented with integral action and error dynamics to create a performance reference for the MPC implementation. The final step during this thesis was to implement MPC reference tracking using integral action on both states and inputs and the results from this controller was then compared to the results from the LQR controller in terms of performance. A model generator capable of creating 6 DOF non-linear dynamics models with varying complexity has been designed. This generator was used to derive two different linear models describing the dynamic behavior of LoLo. These linear models were analyzed and the necessary model parameters were estimated using simulation and then compared to physical data and live testing. Results from the LQR controller using an augmented error dynamics model are promising in terms of reference tracking. Due to a time varying reference and an under-actuated vehicle the resulting performance of the MPC controller does not match the results from the LQR controller. Overall, model-based optimal control show potential as a method for dynamic control and waypoint navigation for AUV applications, but further work with parameter estimation, model validation and control objective definition is required to ensure feasibility in a physical implementation.