Research On Intelligent Vehicle Path Planning Method Of Roundabout

Path planning technology is the basis of autonomous navigation of intelligent vehicles,which can improve the driving efficiency of intelligent vehicles and has a broad development space and application prospects.Aiming at the problem of unsatisfactory path planning effect caused by characteristics such as large changes in road curvature at roundabouts,this paper was supported by National Key Research and Development Program of China(Grant No.2018YFB1600600)to construct a path planning method for intelligent vehicles in roundabouts scenarios based on global path and local path planning methods,and the main research contents are as follows:(1)An improved fast search random tree algorithm is proposed for the low efficiency of vehicle global planning path search under roundabouts.Firstly,the applicability of the algorithm is improved by amplifying the obstacles;Subsequently,the best sampling point is determined by an improved heuristic function in the sampling process,which makes the sampling point biased to the target point and improves the algorithm planning efficiency;Finally,the paths are further optimized by removing redundant nodes.Through comparison experiments,it is found that the path planned by the improved algorithm is 33.5% lower in the number of nodes,24.6% lower in the running time and 12.1% lower in the path length compared with the traditional algorithm.(2)A trajectory prediction model based on dynamic convolutional social long and shortterm memory networks is designed for vehicle trajectory prediction in roundabout intersection scenarios.Firstly,the long and short-term memory network is employed as the encoder and decoder of vehicle trajectory data,and the social pool concept is introduced to extract the interaction features of surrounding vehicles.The preliminary prediction results of the model are compared with the observed data in the prediction time domain,and the weights of the neural network nodes are adjusted to optimize the prediction results.Finally,simulation experiments are conducted on the proposed model.The results show that the proposed trajectory prediction model outperforms the traditional prediction model,and the root mean square error in each time domain is reduced compared to the traditional long and short-term memory network model,and the root mean square error in predicting the trajectory in the next 5s is reduced by 30.4%compared with the traditional algorithm.(3)An improvement of the artificial potential field method using a depth-deterministic policy-based gradient algorithm for local path planning of vehicles under circular intersections.Firstly,a path planning method combining global path and local path is proposed by referring to the global planning path of the roundabout,predicting the trajectory of the surrounding vehicles according to the obtained trajectory model and performing traffic conflict discrimination;Then a reward function based on the artificial potential field is designed to guide the vehicles to the target point by continuously assigning the reward value.Comparative experimental analysis of the improved algorithm shows that the improved algorithm can successfully plan to local paths,and the improved algorithm reduces the planning time by 8.4%compared with the deep Q-network optimized artificial potential field method in a multi obstacle scenario.