Distributionally Robust Model Predictive Control with Disturbance Uncertainty Adaptiveness for Obstacle Avoidance of Autonomous Driving

Traditionally, vehicle obstacle avoidance relies on nonlinear model predictive control (NMPC), which suffers from two major drawbacks. First, NMPC incurs higher computational costs compared to its linear counterpart. Second, it may fail in complex scenarios due to simplified modeling of time-varying vehicle dynamics and the presence of noisy disturbances. In this paper, we propose a distributionally robust model predictive control (MPC) framework for autonomous vehicle obstacle avoidance under nonlinear dynamics and unknown disturbance distributions. The Koopman operator is employed to transform the nonlinear vehicle dynamics into a linear representation in a high-dimensional latent space. Distributionally robust optimization based on a Wasserstein ambiguity set, constructed from historical data, is introduced to guarantee robustness against distributional shifts. The robust probabilistic obstacle avoidance constraint is further converted into tractable linear constraints. Averaged simulations on a racing circuit demonstrate that the proposed method preserves nominal tracking performance while improving feasibility and tail-error robustness under disturbance shifts, compared with non-robust Koopman MPC.