Comparison of Centralized and Multi-Layer Architectures for Nonlinear Model Predictive Torque-Vectoring and Traction Control

A significant body of literature discusses direct yaw moment controllers for vehicle stability control and torque-vectoring (TV), based on model predictive control. However, the available references lack an analysis of the effect of including or excluding the wheel dynamics in the prediction model in combined longitudinal and lateral acceleration conditions, which is related to the control system architecture. In fact, in the first case, the controller can also fulfill the wheel slip control function, according to a centralized architecture, while in the second case, the tire slip limitation has to be implemented externally, in a multi-layer approach. This study addresses the identified gap by proposing and comparing–through simulations with a high-fidelity vehicle model–centralized and multilayer real-time implementable architectures using nonlinear model predictive control (NMPC) for the TV and traction control (TC) of an electric vehicle with front in-wheel motors. An optimization routine calibrates the main controller parameters, to ensure fairness in the comparison during extreme accelerating-while-cornering maneuvers with transient steering inputs. The results show that the real-time implementable multi-layer architecture with wheel dynamics in the NMPC prediction model, and considering the externally generated TC torque reduction in the TV layer, provides equivalent performance to a centralized set-up.