Robustness Evaluation of Linear and Predictive Control Strategies in High-Speed Autonomous Racing

This research focuses on the design and robustness analysis of several control strategies for the dynamics of autonomous racing vehicles based on a nonlinear bicycle model. Tire dynamics are modeled using the well-known Pacejka "magic formula" to accurately describe lateral tire forces as a function of slip angle. The nonlinear model is linearized around a specific operating point and used for controller design and comparative evaluation. Four control methods are investigated: Linear Quadratic Regulator (LQR), Generalized Predictive Control (GPC), Model Predictive Control (MPC) with finite horizon, Dynamic Matrix Control (DMC) and Incremental Model Predictive Control (IMPC). The controllers are evaluated under various operational scenarios, including nominal conditions, measurement noise, model uncertainties, actuator errors, and external disturbances. System performance is assessed using common error metrics such as Integral Absolute Error (IAE), Integral Square Error (ISE), and total control effort. Simulation results show that all controllers maintain system stability in most scenarios, but exhibit significant differences in robustness and control effort. The LQR, despite its simplicity and low computational complexity, is sensitive to disturbances and actuator errors. MPC provides smooth control with low effort but suffers from bias in the presence of uncertainties. DMC faces high control effort and limited robustness. Because of high order of states GPC can't control the system. Overall, IMPC achieves the best balance between path-tracking accuracy, robustness, and stability—particularly under noise and uncertainty—albeit at the cost of increased control effort.

کلیدواژه ها:

predictive control ، autonomous racing car ، linear control ، path tracking