基于组合测试的自动驾驶测试场景组合泛化生成

Combination Generalization Generation of Autonomous Driving Test Scenarios Based on Combination Testing

摘要: 基于测试场景对自动驾驶技术的可靠性进行验证是当前自动驾驶安全测试领域的重要手段，研究可靠的场景高效生成方法是目前行业研究的热点。从真实道路交通事故数据出发，分析交叉路口下乘用车直行与二轮车冲突典型危险场景的风险元素，基于风险元素取值的重要度顺序和约束关系，通过组合测试方法实现场景的组合泛化生成，并与随机采样场景在场景危险性上进行对比分析。结果表明：组合泛化场景的场景事故率是随机采样场景的3.4倍，组合泛化场景的主车碰撞速度远大于随机采样场景，组合泛化场景具有更高的场景危险性，可以提高自动驾驶的仿真测试效率。

Abstract: Verifying the reliability of autonomous driving technology based on testing scenarios is an important means in the current field of autonomous driving safety testing, and researching reliable and efficient scene generation methods is currently a hot topic in the industry. Starting from real road traffic accident data, this paper analyzes the risk elements of typical dangerous scenarios where passenger cars collide with two wheeled vehicles at intersections. Based on the importance order and constraint relationship of risk element values, a combination testing method is used to achieve the combination generalization generation of scenarios, and a comparative analysis is conducted with randomly sampled scenarios in terms of scene danger. The results show that the scene accident rate of the combined generalization scenario is 3.4 times that of the randomly sampled scenario, and the main vehicle collision speed of the combined generalization scenario is much higher than that of the randomly sampled scenario. The combined generalization scenario has higher scene danger and can improve the simulation testing efficiency of autonomous driving.