Research On Intelligent Vehicle Path Planning Algorithm For Unsignalized Intersections

Path planning technology is an important component of research in the field of intelligent driving,which has received widespread attention and has become a research hotspot for research institutions and enterprises.Especially at unsignalized intersections,where road topology is complex,the vehicle environment is constantly changing,and there are many potential traffic safety hazards,it significantly restricts the efficient passage of intelligent vehicles.Therefore,this paper focuses on the path planning problem of intelligent vehicles in unsignalized intersection environments.Combining the current situation of vehicle passage at intersections,a dynamic path planning algorithm is proposed to achieve efficient and safe passage of intelligent vehicles.The main contributions of this paper are as follows:(1)This article proposes an Enhanced Dijkstra Path Planning Algorithm to address the problem of traditional path planning algorithms being unable to adapt to the dynamic and complex environment of unsignalized intersections.Firstly,a system model of unsignalized intersections is established based on the scenario.On the basis of intersection grid,the algorithm considers the directional weight,safety weight,and priority weight of the intelligent vehicle in each grid,and formulates the dynamic grid assignment principle.Consequently,the algorithm can flexibly plan the shortest path for the vehicle to pass through in real-time based on its environment.Simulation results show that compared with the Dijkstra algorithm and the Elite Ant Colony algorithm,the Enhanced Dijkstra algorithm improves the performance of intelligent vehicles in terms of travel time and conflict reduction rate at unsignalized intersections.(2)In order to further improve the path planning ability of intelligent vehicles at unsignalized intersections and enable them to predict optimal paths based on environmental status,this paper proposes a path prediction model based on CNN convolutional neural networks.First,the path data planned by the enhanced Dijkstra,ant colony algorithm,and Neural RRT algorithms are encoded into a mixed path dataset.Then,the CNN convolutional neural network is trained to obtain the path prediction model,which can predict the next optimal path based on the traffic environment in which the intelligent vehicle is located.The simulation results show that the CNN path prediction model has stronger adaptability to dynamic and complex environments compared with the enhanced Dijkstra algorithm,and has improved the efficiency of intelligent vehicle traffic and reduced conflict rates.(3)To enable intelligent vehicles to dynamically plan the optimal path,this paper proposes a DQN value network model.Based on the current dynamic environment of vehicles at unsignalized intersections,a reward and penalty function is designed to ensure that the vehicle plans the shortest path while avoiding conflicts.To improve the training speed of the model,this paper combines the DQN model with the CNN path prediction model using the greedy value.An adaptive learning rate is applied to improve the accuracy of the model prediction,and the convergence of the learning rate on the Loss is demonstrated.Based on the Python-SUMO simulation platform,the results verify that the DQN value network model can enable intelligent vehicles to pass through unsignalized intersections efficiently and safely.(4)To examine the path planning ability of the proposed algorithm model for unsignalized intersections,a simulated intersection experiment was designed to compare and analyze the traffic efficiency of the algorithm model at unsignalized intersections.The speed change curve and number of stops of the intelligent car were analyzed through obstacle avoidance experiments.The experimental results showed that the DQN value network model had better traffic efficiency and safety in path planning at unsignalized intersections.