Distributed And Interactive Prediction Of Future Trajectories Of Surrounding Vehicles For Autonomous Driving

Autonomous vehicles are believed to have the great potential to improve road safety,reduce traffic congestion and relieve drivers from driving burden.Accurate trajectory prediction of surrounding vehicles(SVs)is critical to autonomous driving systems.In mixed traffic flows,SVs with different kinds of behaviors and styles brings complexity to the environment,which requires considering road geometry and interaction among SVs when anticipating their future trajectories.This paper presents a long-term interactive trajectory prediction method using a hierarchical multi-sequence learning network.Firstly,a deep convolutional neural network as the traffic environment understanding module is designed to take into account the road geometry,other road users and other traffic elements.Gridding and numeralizing turn these elements into several matrices where the value matches collision risks.The convolutional-pooling layer extracts sptial features from the traffic elements and the fully connected layer produces vectors representing the environment semantic.Secondly,a two-layer encoder-decoder Structural-LSTM network as the groupinteractive trajectory predicting module is designed to produce future trajectories of SVs by modeling the interactions among multiple SVs.The proposed Structural-LSTM unit assigns multiple LSTMs for each interacting SV,which shares theirs cell states and hidden states with their spatial-neighboring LSTMs by a radial connection,and then recurrently analyze the output state of itself as well as the other LSTMs in a deeper layer,based on whose output to predict trajectories for the target SV.Finally,the proposed method is evaluated on the NGSIM dataset,and its results show that satisfying accurate prediction performance of surrounding vehicle long-term trajectories is accessible,e.g.,longitudinal and lateral RMS error can be reduced to less than 1.93 m and 0.31 m over 5s time horizon,respectively.The discussion also reveals the reasonability of the predicted trajectories in dense traffic flows under highway scenario.