GP-enhanced Autonomous Drifting Framework using ADMM-based iLQR

TL;DR

The paper tackles autonomous vehicle drifting under model mismatch and real-time constraints by proposing a hierarchical framework that augments a nominal drift model with Gaussian Process residual learning and solves the constrained optimal control problem via ADMM-based iLQR. GP residual learning compensates dynamic residuals, while ADMM decomposes the optimization into tractable subproblems, enabling real-time performance. Key contributions include a GP-enhanced residual model for drift dynamics, an ADMM-enabled iLQR solver, and demonstration of significant improvements in lateral tracking (≈38% RMSE reduction) and computation time (≈75% faster than IPOPT) with robustness to friction variations. The approach supports drifting along general paths with improved equilibrium tracking and shows potential for real-world validation on high-fidelity platforms.

Abstract

Autonomous drifting is a complex challenge due to the highly nonlinear dynamics and the need for precise real-time control, especially in uncertain environments. To address these limitations, this paper presents a hierarchical control framework for autonomous vehicles drifting along general paths, primarily focusing on addressing model inaccuracies and mitigating computational challenges in real-time control. The framework integrates Gaussian Process (GP) regression with an Alternating Direction Method of Multipliers (ADMM)-based iterative Linear Quadratic Regulator (iLQR). GP regression effectively compensates for model residuals, improving accuracy in dynamic conditions. ADMM-based iLQR not only combines the rapid trajectory optimization of iLQR but also utilizes ADMM's strength in decomposing the problem into simpler sub-problems. Simulation results demonstrate the effectiveness of the proposed framework, with significant improvements in both drift trajectory tracking and computational efficiency. Our approach resulted in a 38$\%$ reduction in RMSE lateral error and achieved an average computation time that is 75$\%$ lower than that of the Interior Point OPTimizer (IPOPT).

Figures (9)

• Figure 1: The framework integrates three key components: Path Tracking, Optimal Control, and Model Learning. The Path Tracking layer generates the desired states based on predicted errors. The Optimal Control layer leverages the iterative Linear Quadratic Regulator (iLQR) to efficiently solve the nonlinear constrained control problem in real-time, with the Alternating Direction Method of Multipliers (ADMM) aiding in problem decomposing. To account for model inaccuracies and environmental uncertainties, the Model Learning layer employs Gaussian Process (GP) regression to compensate for dynamic residuals.
• Figure 2: Single-track bicycle model for drifting goh2020toward.
• Figure 3: Equilibrium points for $\beta$ and $r$ with fixed $\delta^{eq}=-20$ deg and $R^{eq}$ varying from 20 m to 45 m.
• Figure 4: Error prediction with look-ahead distance.
• Figure 5: The vehicle trajectories for the nominal controller (in blue) and the GP-based controller in the 6th lap (in red) are shown, with yellow points indicating the reference path.