Gaussian-Process-Based Adaptive Trajectory Tracking Control for Autonomous Ground Vehicles

This paper proposes an adaptive trajectory tracking control algorithm for autonomous ground vehicles. The nonlinear vehicle dynamics are decoupled into two subsystems corresponding to the longitudinal and lateral motions. Each subsystem is augmented with a Gaussian Process to compensate for modeling errors and external disturbances. Based on the augmented subsystems, adaptive control algorithms are synthesized. To give a mathematically correct performance measure, the induced
$\mathcal{L}_2$
-gain of the nonlinear closed-loop system is computed. The efficiency of the learning-based control method is demonstrated on a high-fidelity physical simulator using a digital twin model of the 1/10 scale F1TENTH vehicle platform.