Chen, yinsong (Beijing Institute of Technology), Wang, Kaifeng (Beijing Institute of Technology), Meng, Xiaoqiang (Beijing Institute of Technology), Li, Xueyuan (Beijing Institute of Technology), Li, Zirui (Beijing Institute of Technology), Gao, Xin (Beijing Institute of Technology)

Red-Team Multi-Agent Reinforcement Learning for Emergency Braking Scenario

Scheduled for presentation during the Regular Session "S27b-Safety and Risk Assessment for Autonomous Driving Systems" (TH-EA-T27), Thursday, November 20, 2025, 13:30−13:50, Broadbeach 3

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

当前关于安全关键型决策的研究 场景通常依赖于低效的数据驱动场景 生成或特定的建模方法，这些方法无法 在真实环境中捕获极端情况。要解决 本期，我们提出红队多智能体加固 学习框架，其中具有 干扰功能被视为红队代理。通过积极干扰和探索，红队 车辆可以在数据之外发现极端情况 分配。框架使用 Constraint Graph 表示马尔可夫决策过程，确保 红队车辆遵守安全规则，同时 不断颠覆自动驾驶汽车 （AV）。一个 构建策略威胁区模型，量化 红队车辆对 AV 构成威胁，诱发更多 提高 场景。实验结果表明，所提 框架对自动驾