EvoDrive: Pareto Evolution for Safety-Critical Autonomous Driving via Self-Improving LLM Agents
Researchers have developed EvoDrive, a novel framework that uses LLM agents to generate safety-critical scenarios for autonomous driving systems. This approach aims to improve the validation and enhancement of self-driving technology by maximizing adversariality while maintaining realism. EvoDrive employs an actor-critic architecture grounded in simulators, with a self-evolving world evaluator to optimize simulation budgets and a Pareto archive to preserve diverse trade-offs between attack and realism. AI
IMPACT Enhances autonomous driving safety by generating more realistic and adversarial test scenarios.