Comparative study of Bayesian-optimized robust sliding mode control strategies for trajectory tracking of four-wheel-steering autonomous vehicles

Accurate path tracking is a key requirement for autonomous vehicles, especially during fast lane-change maneuvers and under uncertain road conditions. This paper presents a unified robust-control framework for trajectory tracking of a four-wheel-steering autonomous vehicle. Four sliding-mode control strategies are designed and compared: saturation-based sliding mode control (SAT-SMC), super-twisting sliding mode control (ST-SMC), saturation-based nonlinear tracking sliding-mode control (SAT-NTSMC), which uses a nonsingular nonlinear power surface, and a super-twisting-inspired nonsingular nonlinear sliding-mode controller, denoted ST-NTSMC for continuity with the controller family. The free gains of all controllers are tuned by Bayesian optimization using the same normalized multi-objective cost function. The cost combines global lateral-position tracking error, full-horizon maximum lateral deviation, virtual control effort, command-rate smoothness, and command-acceleration smoothness. The controllers are first evaluated in a double lane-change maneuver and are then tested under variations in vehicle mass and tire–road friction. A separate percentage overshoot diagnostic is also reported for the plateau and settling portions of the double lane-change path. The proposed super-twisting-inspired ST-NTSMC is compared with the other SMC variants and qualitatively with a previously reported nonlinear model predictive controller in a single lane-change scenario. The results show that ST-NTSMC gives the best overall compromise among the tested methods. In the double lane-change case, it reduces the lateral root-mean-square error by about 78% relative to the SAT-SMC baseline, while also reducing overshoot and virtual-control effort. The robustness tests further show accurate and smooth tracking under increased mass and reduced friction within the tested operating envelope. The selected nonlinear surface provides global asymptotic stability of the ideal on-manifold tracking-error dynamics. Practical finite-time reaching and ultimate-boundedness results are conditional on a bounded aggregate residual, and no global finite-time certificate is claimed for the complete vehicle model. These findings indicate that Bayesian-optimized, super-twisting-inspired ST-NTSMC is a promising control strategy for robust path tracking in four-wheel-steering autonomous vehicles.