Study On Parking Path Planning And Tracking Control Of Intelligent Vehicle

As an important part of intelligent transportation system,intelligent driving vehicles can alleviate the problems of traditional vehicles in environment,safety,traffic and so on.As an application scenario of L4 technology in the development of intelligent driving,parking scene has great potential in improving the user’s "last mile" driving experience.Intelligent vehicle path planning is the necessary guarantee for intelligent driving vehicle to complete automatic driving behavior,and another important link in the development of intelligent vehicle is lateral control or automatic steering control.In this paper,the path planning algorithm and lateral tracking control of intelligent vehicle in parking scene are studied.This paper proposes a parking path planning algorithm which can be applied to the real parking lot and meet the needs of vehicle driving,and makes the vehicle track the planned path stably.The main contents of this paper are as follows(1)Intelligent vehicle system structure,environment model and vehicle model building.The intelligent vehicle system structure is established to meet the demand of vehicle parking,which is divided into hardware perception system and software decision system.The real car park map is transformed into probability value grid map by hardware system structure.In order to provide accurate map for intelligent vehicle software system,the probability information of grid map is further analyzed,and transformed into binary grid map.Finally,in order to analyze the dynamics and kinematics of the intelligent vehicle in the parking environment,the vehicle coordinate system,vehicle dynamics model,vehicle steering control model and preview tracking model are established,which lays the foundation for the research of parking path planning and tracking control in this paper.(2)Parking path planning based on optimal rapid exploring random tree(RRT *).Firstly,the driving path is divided in the binary grid map and the obstacles are inflated.Three circle approximation model is used as collision detection model to ensure driving safety.RRT *algorithm with target bias is used to search the complete path in the configuration space.Dubins curve and Reeds-Shepp curve are used to connect the search tree nodes in RRT *algorithm to form a vehicle drivable path.Finally,aiming at the problem of discontinuity of the second-order derivative in the planning path after the curve connection,the cubic spline curve is used to further smooth the planned path to form a planning path that meets the riding comfort of vehicles.(3)Trajectory Tracking Control of Intelligent Vehicle Based on DDPG(Deep Deterministic Policy Gradient)Method of Reinforcement Learning.Aiming at the problem of lateral control of intelligent vehicle in the process of trajectory tracking,a trajectory tracking control method of intelligent vehicle based on DDPG method of reinforcement learning is proposed.Firstly,the tracking control of intelligent vehicle was described as a reinforcement learning process based on Markov Decision Process(MDP).The main body of reinforcement learning was the Actor-Critic framework composed of actor neural network and critic neural network.The reinforcement learning environment included vehicle model,tracking model,road model and reward function.Then,the learning agent of the proposed method was updated by DDPG,in which the replay buffer was used to solve the problem of sample correlation,and the neural network with the same structure was copied to solve the problem of update divergence.(4)Simulation verification experiment of intelligent vehicle parking path planning and tracking control.In the simulation experiment verification,the intelligent vehicle parking scene and tracking control scene are built respectively.In the path planning simulation experiment,Dubins curve is used to connect the nodes searched by RRT * algorithm when the vehicle parking motion does not need reversing behavior;Reeds-Shepp curve is used to connect the nodes searched by RRT * algorithm when the vehicle parking motion needs reversing behavior.In vehicle tracking control,the DDPG based tracking controller is compared with Deep Q-learning(DQN)and Model Predictive Control(MPC)methods in different scenarios,and the speed adaptability and curvature adaptability of the proposed tracking control method are verified.