一句话：提出 Flexcrash 平台，通过生成与仿真混合交通场景支持自动驾驶车辆测试。
自动驾驶车辆利用人工智能来减少事故并提升燃油效率，但需与人类驾驶者共享道路；现有原型尚未达到这些目标，因此亟需更好的开发与测试方法。当前 AV 测试实践大量依赖仿真，而现有 AV 工具多聚焦单个 AV 实例或忽略人类驾驶者，可能生成大量无关的混合交通测试场景。Flexcrash 平台针对这些问题，支持混合交通场景的生成与仿真，使测试人员能够识别现实的临界场景，交通专家能够创建新数据集，监管机构能够扩展消费者测试基准。论文的核心贡献是提供一个面向混合交通的仿真测试平台，以弥补 AV 测试中对人类驾驶行为与多车交互建模的不足。
Autonomous vehicles (AV) leverage Artificial Intelligence to reduce accidents and improve fuel efficiency while sharing the roads with human drivers. Current AV prototypes have not yet reached these goals, highlighting the need for better development and testing methodologies. AV testing practices extensively rely on simulations, but existing AV tools focus on testing single AV instances or do not consider human drivers. Thus, they might generate many irrelevant mixed-traffic test scenarios. The Flexcrash platform addresses these issues by allowing the generation and simulation of mixed-traffic scenarios, thus enabling testers to identify realistic critical scenarios, traffic experts to create new datasets, and regulators to extend consumer testing benchmarks.
@inproceedings{paperbot8529,
title = {The Flexcrash Platform for Testing Autonomous Vehicles in Mixed-Traffic Scenarios},
author = {Alessio Gambi and Shreya Mathews and Benedikt Steininger and Mykhailo Poienko and David Bobek},
booktitle = {ISSTA},
year = {2024},
doi = {10.1145/3650212.3685299}
}