The invention provides a roundabout multi-vehicle collaborative decision-making method based on safety reinforcement learning. The method comprises the following steps: constructing a multi-agent reinforcement learning model; each intelligent agent obtains own vehicle state information and state information of other surrounding vehicles in a sensing range through the roundabout automatic driving simulation environment; constructing a self-attention network, taking the state space of each agent as input, and outputting the state space after attention weighting; inputting the state space subjected to attention weighting into an action network and a critic network, and outputting execution actions of all agents; performing driving risk assessment on execution actions of all the agents based on the priority list to obtain corrected safety actions, and continuing to optimize the multi-agent reinforcement learning model until the multi-agent reinforcement learning model is completely converged; and applying the completely convergent multi-agent reinforcement learning model to carry out roundabout multi-vehicle collaborative decision-making. According to the method, the overall traffic efficiency and safety of the roundabout are improved.
本发明提供一种基于安全强化学习的环形交叉口多车协同决策方法,包括:搭建构成多智能体强化学习模型;每个智能体通过环形交叉口自动驾驶仿真环境获取自身车辆状态信息以及感知范围内其他周围车辆的状态信息;构建自注意力网络,将每个智能体的状态空间作为输入,输出经过注意力加权后的状态空间;将经过注意力加权后的状态空间输入action网络和critic网络,输出所有智能体的执行动作;基于优先级列表对所有智能体的执行动作进行行车风险评估,得到修正后的安全动作,继续优化多智能体强化学习模型至完全收敛;应用完全收敛的多智能体强化学习模型进行环形交叉口多车协同决策。本发明的方法提高了环形交叉口的整体交通效率和安全性。
Roundabout multi-vehicle collaborative decision-making method based on safety reinforcement learning
基于安全强化学习的环形交叉口多车协同决策方法
2024-10-01
Patent
Electronic Resource
Chinese
IPC: | B60W CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION , Gemeinsame Steuerung oder Regelung von Fahrzeug-Unteraggregaten verschiedenen Typs oder verschiedener Funktion |
European Patent Office | 2024
|European Patent Office | 2024
|