Steering-lag-compensated LPV model predictive control for autonomous vehicle trajectory tracking: Design and real-vehicle validation on a steer-by-wire platform

Omitting steering actuator dynamics can degrade trajectory tracking under high-speed, low-friction, and rapidly varying-curvature conditions. This paper reports the integration and real-vehicle implementation of steering-lag-compensated linear parameter-varying model predictive control (LC LPV-MPC) on the tested steer-by-wire vehicle. A first-order-plus-delay-time (FOPDT) model is identified from on-vehicle data. The identified delay time is not explicitly modeled; instead, the first-order steering lag and measured actual front-wheel steering angle are incorporated into a five-dimensional model. Dynamic curvature feedforward and soft tracking-error constraints are included to reduce curvature-induced offset and the risk of quadratic programming (QP) infeasibility. Hessian normalization and active-set warm starting are used. Simulations and closed-course real-vehicle experiments cover single lane change (SLC), double lane change (DLC), and right-angle turn (RAT) maneuvers. LC LPV-MPC maintained bounded trajectory tracking under the tested real-vehicle conditions and achieved lower lateral tracking errors than a feedforward linear quadratic regulator (LQR+FF) at comparable SLC and DLC speeds. Hardware-in-the-loop testing on MABX III yielded a maximum complete controller-task turnaround of 0.361 ms, below 0.73% of the 50-ms control period. The constant-cornering-stiffness tire model is only a local control-oriented approximation, with limitations evident in the RAT maneuver. Local asymptotic stability is established only for the analyzed frozen, unconstrained nominal models. The analysis does not establish uniform LPV stability or constrained closed-loop stability.