Safe and Efficient Vehicle Motion Control of Articulated Heavy Vehicles - A motion control method using Model Predictive Control and Machine Learning

Abstract

Abstract When braking an articulated heavy vehicle with at least one electric unit, it is best to prioritize electric braking to enhance power regeneration for the battery electric vehicle. However, this approach combined with treacherous road conditions such as low-friction high curvature roads, may lead to unsafe yaw instability modes such as jack-knifing, trailer swing, and combinational spin-out. These unsafe modes can lead to potentially fatal accidents. Therefore, achieving a safe control allocation during manoeuvring is crucial to ensure stability while maximizing the regenerated energy. This thesis describes a method for control allocation of longitudinal forces and steering wheel input to a specified vehicle combination for vehicle motion control. The system was implemented using a non-linear Model Predictive Control (MPC) scheme, which allows the use of constraints to maintain the vehicle within safe operating conditions. To ensure a persistently feasible non-linear model predictive controller that improves vehicle safety, an approximation of the control invariant set was computed. The control invariant set is approximated using an Active Learning algorithm that utilizes support vector machines to approximate the set effectively. Results indicated the ability of the control scheme to manage the vehicle motion under various environmental conditions in near real-time. Energy regeneration could be improved using the MPC while keeping the vehicle combination in a safe mode. The derived method enabled flexible usage of the MPC depending on the vehicle parameters and objectives. Additional extensive analysis is necessary to more accurately approximate the largest possible control invariant set through the application of machine learning.