Search-to-Crash: Generating safety-critical scenarios from in-depth crash data for testing autonomous vehicles

Autonomous vehicles (AVs) are increasingly recognized as a cornerstone of energy-efficient and sustainable transportation. However, their widespread deployment depends critically on rigorous safety validation under realistic and high-risk conditions. Traditional scenario-based testing frameworks, often driven by naturalistic driving data (NDD), suffer from a lack of critical events and heavy data requirements, limiting both testing diversity and energy relevance. To overcome these challenges, this paper proposes a novel Search-to-Crash framework that generates large-scale, safety-critical scenarios directly from in-depth real-world crash data. Leveraging multi-leader particle swarm optimization (MLPSO), each crash case is evolved into a diverse set of concrete scenarios that preserve real-world characteristics while expanding the risk space. This method significantly reduces data dependency and supports energy-conscious safety evaluation of AVs. Simulation experiments using the Baidu Apollo platform demonstrate the framework’s ability to intensify test risk levels, uncovering more system vulnerabilities than original crash data alone. The proposed approach contributes to the development of robust, energy-aware validation pipelines for next-generation autonomous mobility systems.