Risk-Aware World Model Predictive Control for Generalizable End-to-End Autonomous Driving

Researchers developed Risk-aware World Model Predictive Control (RaWMPC), a new framework for autonomous driving that makes safe decisions without relying on expert demonstrations. The system uses a world model to predict consequences of multiple actions and selects low-risk options through explicit risk evaluation, showing superior performance in both normal and rare driving scenarios.

• →RaWMPC addresses the generalization problem in end-to-end autonomous driving by eliminating reliance on expert demonstrations.
• →The framework uses a world model to predict outcomes of risky driving behaviors and systematically avoids catastrophic scenarios.
• →A risk-aware interaction strategy exposes the world model to hazardous behaviors during training to improve safety predictions.
• →Self-evaluation distillation transfers risk-avoidance capabilities from the world model to an action proposal network.
• →Experimental results show RaWMPC outperforms existing methods in both standard and out-of-distribution driving scenarios.