Vehicle Trajectory Prediction with Attention Based on Convolutional Social Pooling

In order to improve the prediction accuracy of autonomous vehicles in complex traffic scenarios under multi-vehicle interactions, this paper proposes an attention mechanism considering the historical trajectories of vehicles to be predicted and the attention mechanism between interacting vehicles based on convolutional social pooling. In the trajectory prediction algorithm, based on the LSTM (Long Short Term Memory) encoder-decoder network, the attention of the interacting vehicles is built in the middle of the encoder and decoder, and is used as the input of the decoder together with the attention of the historical trajectory of the vehicles to be predicted, which solves the problem that the attention mechanism of the historical trajectory of the target vehicles is not fully considered in the interaction. The advantage of the method in this paper is that the influence magnitude between the interacting vehicles of the scene and the influence magnitude of the historical trajectory of the target vehicle are considered in the middle of the network. The validity of the model is verified in the NGSIM (Next Generation Simulation) dataset, and the prediction accuracy is significantly improved when compared with various advanced algorithms.