![[2602.23259] Risk-Aware World Model Predictive Control for Generalizable End-to-End Autonomous Driving](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.23259] Risk-Aware World Model Predictive Control for Generalizable End-to-End Autonomous Driving
Summary
This paper presents the Risk-aware World Model Predictive Control (RaWMPC) framework aimed at enhancing generalization in end-to-end autonomous driving systems by enabling reliable decision-making without expert supervision.
Why It Matters
The advancement of autonomous driving relies heavily on the ability to generalize from expert demonstrations. This research addresses critical safety concerns by introducing a method that allows autonomous systems to evaluate risks and make safer decisions in previously unseen scenarios, thus potentially improving the reliability of autonomous vehicles in real-world applications.
Key Takeaways
- RaWMPC enhances decision-making in autonomous driving without expert supervision.
- The framework utilizes a world model to predict outcomes of driving actions.
- It incorporates a risk-aware strategy to expose models to hazardous behaviors.
- RaWMPC outperforms existing methods in both familiar and unfamiliar driving scenarios.
- The approach provides better interpretability of decision-making processes.
Computer Science > Computer Vision and Pattern Recognition arXiv:2602.23259 (cs) [Submitted on 26 Feb 2026] Title:Risk-Aware World Model Predictive Control for Generalizable End-to-End Autonomous Driving Authors:Jiangxin Sun, Feng Xue, Teng Long, Chang Liu, Jian-Fang Hu, Wei-Shi Zheng, Nicu Sebe View a PDF of the paper titled Risk-Aware World Model Predictive Control for Generalizable End-to-End Autonomous Driving, by Jiangxin Sun and 6 other authors View PDF Abstract:With advances in imitation learning (IL) and large-scale driving datasets, end-to-end autonomous driving (E2E-AD) has made great progress recently. Currently, IL-based methods have become a mainstream paradigm: models rely on standard driving behaviors given by experts, and learn to minimize the discrepancy between their actions and expert actions. However, this objective of "only driving like the expert" suffers from limited generalization: when encountering rare or unseen long-tail scenarios outside the distribution of expert demonstrations, models tend to produce unsafe decisions in the absence of prior experience. This raises a fundamental question: Can an E2E-AD system make reliable decisions without any expert action supervision? Motivated by this, we propose a unified framework named Risk-aware World Model Predictive Control (RaWMPC) to address this generalization dilemma through robust control, without reliance on expert demonstrations. Practically, RaWMPC leverages a world model to predict the conseq...
