Driving Risk Analysis And Motion Planning Of The Left-Turn Vehicle At Unsignalized Intersection

Due to the lack of traffic control and the difference of human drivers’ styles,the vehicle interaction at unsignalized intersections is complicated and the randomness of the movement trajectory is strong.With the development of wireless communication technology,connected automated vehicles can adjust their own motion state by collecting the motion state information of nearby human-driving vehicles in real time,thus reducing the potential collision risk.At present,the prior knowledge of the driving risk of left-turning vehicles at unsignalized intersections is insufficient,which makes the driving risk factors not fully considered in the left-turning motion planning and control algorithm of intelligent networked vehicles.Therefore,this paper aims at the quantitative assessment of left turn risk of human-driving vehicles and the left turn motion planning of connected automated vehicles based on this,and carries out the following research:(1)The driving risk of left-turn vehicles is analyzed.A visualization method is used to manually extracted the trajectory dataset to left-turn conflicts.Considering the driving characteristics of the different vehicles in the inter-lane,the sequential clustering method Kshape is used to cluster the driving styles of vehicles.According to the clustering results,the left-turn motion features in the conflict zone is described.Aim at angle crash,this paper used the distribution of relative acceleration to represent the collision probability,and the deceleration rate to avoid the crash to represent the collision severity.A comprehensive risk index has been established to calculate the left-turn risk variation.The results show that the interaction stability is negatively correlated with the time-varying coefficient of the longitudinal/ lateral acceleration rate.Aggressive drivers have the most unstable turning risk and the highest fluctuation peak.Normal drivers have moderate turning risk and the suitable adjustment time.Conservative drivers have the lowest fluctuation and the stable driving state.(2)A hierarchical framework of intelligent connected vehicle decision making and motion planning for left-turn maneuvers is proposed.The decision-making layer includes driving risk awareness and speed planning module.The former uses Bayes theorem to estimated and the probability of the different risk levels and calculate the coefficient.According to the risk coefficient,the reward strategy of RL agent is adjusted to fit the actual feedback of the maneuvers bring to environment,which combined with Soft Actor-Critic called after RA-SAC algorithm.The latter described the speed decision as a multi-objective planning problem.Gap acceptance and particle swarm algorithm are used to solve the solution and as the subgoal of the controller.The motion planning layer solve the parameter of longitudinal cruise and path tracking based on RL agent.(3)The vehicle motion planning controller simultaneously running a longitudinal speed adjuster and a steering tracker.The steering tracker is composed of pure pursuit algorithm and PID controller.The RL agent is used to make decisions on throttle/brake pressure and the basic look-forward distance of the pure pursuit algorithm.A joint Matlab/Simulink and Prescan simulation platform was established for the RL training and model evaluation.The results show that the average reward of RA-SAC algorithm improved by 10.39%,and the average training step decrease by 31.06% compared to the SAC.Moreover,RA-SAC showed the best performance in driving safety.In terms of travel efficiency and comfort,RA-SAC improved by2.73%,26.88% and 34.59% compared with SAC,TD3 and DDPG,respectively.The proposed model can ensure driving safety under the environment vehicles with different driving styles,which has good adaptability and robustness.