As a successful control method in both engineering and academia, model predictive control (MPC), especially nonlinear one (NMPC) has been extensively researched notonly for improving its performance but also for extending its application fields. Originally proposed for complex interacting industrial process, MPC is well suited to deal with nonlinearities and constraints in a much more straightforward way than other methods. Autonomous vehicle related research, with the nonholonomic constraint and mechanical saturations, becomes such a new and promising field for exploring MPC.
This dissertation concentrates on the design of a model predictive control architecture based on a discrete time nonlinear car model to solve regulation ("parking") problem. Discrete time MPC is proposed here not only to overcome the difficulties encountered by smooth feedback stabilization for nonholonomic systems but also to integrate the input and state constraints into the controller design process. An important consideration for finite horizon NMPC, stability is achieved by considering terminal state constraints combined with a terminal state penalty in the cost function, as well as the terminal controller design. The generated trajectory satisfies minimum curvature requirements and obstacle avoidance is also realized by considering distance constraints in the open-loop optimization process.
It is well known that a primary concern for NMPC strategies is the evaluation of their control performance, especially robustness. Many researchers show the existence of robustness as a byproduct of stability which is achieved by monotonicity of the cost function. However the design of a control architecture within the MPC frame and the analysis of its robustness to additive uncertainties are far from well solved together as a complete topic. A robustness analysis is provided for the designed MPC control architecture so that the bound for additive uncertainties could be found under which the closed-loop system is input-to-state stable. The results are fit for general cases where more than one control values solved from the optimal problem are applied to real systems.
As a further example of the possibility of applying MPC in vehicle industry, energy efficient cruise control is proposed to realize optimal energy management for vehicles.
This concept is realized based on MPC strategy with an adaptive prediction horizon.