Networked model predictive control for vehicle collision avoidance

This thesis investigates and develops Networked Model Predictive Control (Net-MPC) for large-scale Networked Control Systems (NCS). The developed Net-MPC strategies are then applied to collision avoidance of networked vehicles.The basic objective of this thesis, pertaining to the methodology of Net-MPC, especially targets the study of the distribution of the control problem of a homogeneous NCS, where the agents are dynamically decoupled and the coupling is achieved in the objective function or in the constraints using Distributed MPC (DMPC). The Net-MPC strategies Centralized MPC (CMPC), Cooperative DMPC (Coop. DMPC), and Non-Cooperative DMPC (Non-Coop. DMPC) are also studied in detail. A novel Net-MPC strategy called Priority-Based Non-Cooperative DMPC (PB-Non-Coop. DMPC) is introduced in this thesis and compared with the other strategies. This strategy reduces the computation time in comparison with the existing strategies of Non-Coop. DMPC.Collision avoidance of networked vehicles can be seen as an attempt to solve a large-scale control problem by means of computing the steering angles that drive the vehicles along their reference trajectories, while simultaneously avoiding collisions. Due to the collision avoidance constraints, the resulting optimization problem is non-convex. This thesis therefore presents different methods to solve this non-convex optimization problem. As the number of vehicles increases, the solution of such large-scale, centralized, non-convex optimization problems becomes computationally more prohibitive. Therefore, the optimization methods are implemented in the frameworks of CMPC, Coop. DMPC, and PB-Non-Coop. DMPC. The effectiveness of the developed Net-MPC strategies and optimization methods is validated in simulations and in experiments using model-vehicles