Trajectory Following of a Vehicle via Computationally Improved Explicit Nonlinear Model Predictive Controller

Model predictive control is one of the advanced methods to solve trajectory following problems of autonomous ground vehicles due to its predictive capability. It can deal with multiple inputs, outputs, and set-point signals. However, this technique causes a heavy computational burden to the controller due to its multi-objective optimization approach, limiting its real-time applications. In this study, an explicit nonlinear model predictive controller (E-NMPC) is proposed to control front steering angle and rear wheel tractive torques simultaneously to provide trajectory tracking of a vehicle in NATO double lane change (DLC) maneuvers with consideration of reduced computation time. Artificial neural networks enabling offline learning processes are used to generate a computationally-efficient control law without going through extensive optimization methods. Namely, main contribution of this study is establishment of a neural-network based E-NMPC technique for the considered control problem that removes the need of online optimization and replace the classical NMPC that requires online-optimization with high computational load. The simulated results revealed that time elapsed during one execution cycle can be significantly reduced with the proposed method for various cases, compared to the classical online NMPC method.