混合交通环境下自动驾驶车在无信控交叉口运行特征仿真分析

Simulation Analysis of Operational Characteristics of Autonomous Vehicles at Unsignalized Intersections in Mixed Traffic Environments

摘要:

针对混合交通环境下自动驾驶车在无信控交叉口的通行效率与安全问题，通过深度强化学习训练了两种车辆行为策略，在仿真环境中搭建了4类典型场景并进行测试。结果表明：两种策略下自动驾驶车通过无信控交叉口的平均成功率均达到99.5%以上，交叉口车辆的平均刹车时间均维持在较低水平；与基于碰撞时间的传统策略相比，平均通行时间分别缩短了26.8%和18.1%，同时，平均碰撞比例维持在较低水平（0.18%、0.02%）。提出的基于深度强化学习的自动驾驶车行为策略能够高效、安全地协调无信控交叉口的通行，可为类似研究提供参考。

Abstract:

Aiming at the traffic efficiency and safety issues of autonomous vehicles at the unsignalized intersections in mixed traffic environments, two vehicle behavior strategies are trained through deep reinforcement learning, and four typical scenarios are built and tested in the simulation environment. The results demonstrate that under both strategies, the average success rate of autonomous vehicles passing through unsignalized intersections reached over 99.5%, and the average braking time of vehicles at intersections remain at a relatively low level. Compared with the traditional strategy based on collision time, the average passage time has been shortened by 26.8% and 18.1% respectively. Meanwhile, the average collision ratio remains at a relatively low level (0.18%, 0.02%). The proposed behavior strategy for autonomous vehicles based on deep reinforcement learning can efficiently and safely coordinate the traffic at unsignalized intersections, which can provide a reference for similar research.