RoCA: Robust Cross-Domain End-to-End Autonomous Driving

End-to-end (E2E) autonomous driving has recently emerged as a new paradigm, offering significant potential. However, few studies have looked into the practical challenge of deployment across domains (e.g., cities). Although several works have incorporated Large Language Models (LLMs) to leverage their open-world knowledge, LLMs do not guarantee cross-domain driving performance and may incur prohibitive retraining costs during domain adaptation. In this paper, we propose RoCA, a novel framework for robust cross-domain E2E autonomous driving. RoCA formulates the joint probabilistic distribution over the tokens that encode ego and surrounding vehicle information in the E2E pipeline. Instantiating with a Gaussian process (GP), RoCA learns a set of basis tokens with corresponding trajectories, which span diverse driving scenarios. Then, given any driving scene, it is able to probabilistically infer the future trajectory. By using RoCA together with a base E2E model in source-domain training, we improve the generalizability of the base model, without requiring extra inference computation. In addition, RoCA enables robust adaptation on new target domains, significantly outperforming direct finetuning. We extensively evaluate RoCA on various cross-domain scenarios and show that it achieves strong domain generalization and adaptation performance.

End-to-end (E2E) autonomous driving uses a single system that directly maps camera images to driving actions. While recent approaches show great promise, a major challenge remains: models trained in one environment (such as a specific city or weather condition) often struggle when deployed in new, unseen settings. This lack of robustness can lead to unsafe behavior, especially in rare or complex driving situations. In this work, we introduce a new method called RoCA that improves the reliability of autonomous driving systems across different environments. RoCA builds a structured set of representative driving behaviors and trajectories. When the vehicle encounters a new or challenging scenario, it compares the situation against this learned set using a statistical model called Gaussian Process. This allows the system not only to predict safe actions but also to estimate how uncertain it is about its decisions. During training, this uncertainty helps the model focus more on difficult or ambiguous cases, improving its ability to handle rare and safety-critical situations. An important advantage of RoCA is that it improves robustness without increasing computational cost during deployment. It can also adapt efficiently to new environments using limited data and identify which new data would be most useful for learning. We evaluate RoCA across a range of realistic driving scenarios, including simulations and cross-city settings, and show that it consistently outperforms existing methods. Overall, this work moves toward safer and more adaptable autonomous driving systems that can operate reliably in diverse real-world conditions.