Gaussian-Process-based Adaptive Tracking Control with Dynamic Active Learning for Autonomous Ground Vehicles

TL;DR

The paper addresses robust trajectory tracking for autonomous ground vehicles under modeling uncertainties by coupling online sparse Gaussian Processes with a computationally efficient LPV-LQR baseline, augmented by a dynamic active-learning loop and a formal L2-gain robustness analysis. The method jointly updates GP hyperparameters and inducing points online, plans informative experiments via Bayesian optimization to refine the GP models, and proves performance bounds through a dissipativity-based L2-gain framework. Key contributions include a recursive sparse-GP update scheme, a dynamic experiment-design strategy, and a counterexample-based L2-gain verification, all demonstrated on high-fidelity simulations and real 1/10-scale F1TENTH experiments. The approach enables data-efficient learning-based control with formal guarantees, achieving accurate tracking despite up to ~30% model mismatch while maintaining real-time feasibility.

Abstract

This article proposes an active-learning-based adaptive trajectory tracking control method for autonomous ground vehicles to compensate for modeling errors and unmodeled dynamics. The nominal vehicle model is decoupled into lateral and longitudinal subsystems, which are augmented with online Gaussian Processes (GPs), using measurement data. The estimated mean functions of the GPs are used to construct a feedback compensator, which, together with an LPV state feedback controller designed for the nominal system, gives the adaptive control structure. To assist exploration of the dynamics, the paper proposes a new, dynamic active learning method to collect the most informative samples to accelerate the training process. To analyze the performance of the overall learning tool-chain provided controller, a novel iterative, counterexample-based algorithm is proposed for calculating the induced L2 gain between the reference trajectory and the tracking error. The analysis can be executed for a set of possible realizations of the to-be-controlled system, giving robust performance certificate of the learning method under variation of the vehicle dynamics. The efficiency of the proposed control approach is shown on a high-fidelity physics simulator and in real experiments using a 1/10 scale F1TENTH electric car.

Figures (9)

• Figure 1: Single-track vehicle model and reference trajectory.
• Figure 2: Proposed control architecture for trajectory tracking.
• Figure 3: Waypoint-based parameterization of the reference trajectories: the 2D path (left) and the speed profile (right).
• Figure 4: Initial training trajectory (dashed) and the trajectories obtained by the dynamic active learning (solid)
• Figure 5: Cumulative variance of the GPs after each iteration of the dynamic active learning algorithm.