The invention discloses a traffic light-free intersection vehicle scheduling method based on an asynchronous reinforcement learning algorithm. The invention provides a vehicle scheduling method for modeling a cooperative control problem of intelligent network connection vehicles at an intersection into a partially observable Markov decision process. In a driving scene of an intersection without a signal lamp, an RSU (Roadside Unit) and a plurality of intelligent network connection vehicles are arranged. Each intelligent networked vehicle has a local neural network as an acceleration decision model, and shared model parameters stored on the RSU are respectively used for assigning values to the local neural networks of the intelligent networked vehicles. And interactively calculating a parameter updating gradient through the local neural network and the traffic environment, and uploading the parameter updating gradient to the RSU. And the RSU updates the parameters of the shared model by using the parameter updating gradient sent by the intelligent networked vehicle. By applying the method, the intelligent networked vehicle can asynchronously upload and acquire data to the RSU, the limitation that all agents must synchronously upload data in a traditional reinforcement learning method is eliminated, the method is more suitable for a real scene, the training efficiency of a neural network can be improved in a high-density intersection scene, and the training efficiency of the neural network is improved. And the passing efficiency and the safety performance of the intelligent network connection vehicle are further improved.
本发明公开了一种基于异步强化学习算法的无信号灯交叉路口车辆调度方法。将智能网联车辆在路口的协同控制问题建模成部分可观测马尔可夫决策过程,提出了一种的车辆调度方法。在无信号灯交叉路口驾驶场景中,有一个路侧单元(Road Side Unit,RSU)和多辆智能网联车辆。每辆智能网联车辆都有一个本地神经网络作为加速度决策模型,RSU上存储共享模型参数分别用来给智能网联车辆的本地神经网络赋值。通过本地神经网络与交通环境交互计算参数更新梯度,并将参数更新梯度上传给RSU。RSU利用智能网联车辆发来的参数更新梯度更新共享模型参数。运用该方法,智能网联车辆可以异步向RSU上传和获取数据,解除了传统强化学习方法对所有智能体必须同步上传数据的限制,更适用于现实场景,且在高密度路口场景可以提升神经网络的训练效率,进一步提升智能网联车辆的通行效率和安全性能。
Signal-lamp-free intersection vehicle scheduling method based on asynchronous reinforcement learning algorithm
一种基于异步强化学习算法的无信号灯交叉路口车辆调度方法
2023-10-31
Patent
Electronic Resource
Chinese
Multi-Intersection Signal Control Based on Asynchronous Reinforcement Learning
DOAJ | 2025
|Signal lamp-free intersection control method and device based on deep reinforcement learning
European Patent Office | 2023
|European Patent Office | 2023
|Signal-lamp-free intersection vehicle safe passing method
European Patent Office | 2020
|European Patent Office | 2025
|