Stochastic MPC for Autonomous Vehicles in Uncertain Situations

Abstract

Abstract In this thesis, an Model Predictive Control (MPC) based trajectory planning algorithm is first introduced for controlling trucks on highways. Given the uncertainties that exist between theoretical models and real vehicles, this study further analyzes these uncertainties and proposes an Stochastic Model Predictive Control (SMPC) based trajectory planning algorithm. The algorithm avoids collisions by tightening constraints and is validated in the CARLA simulation environment. Experimental results show that the SMPC-based trajectory planning algorithm has obvious advantages in terms of safety performance compared with the standard MPC. However, the method also sacrifices certain driving performance and increases computational complexity, which is mainly due to the tightened traffic constraints. This study not only verifies the effectiveness of SMPC in handling uncertainty and enhancing safety but also provides both an experimental and theoretical basis for future work.