Vehicle Trajectory Prediction Model Based on Transformer and GCN

The study of vehicle trajectory prediction problem is of great significance to the field of autonomous driving. Most of the existing vehicle traj ectory prediction methods are designed around high-definition maps and rely too much on map information, however, for many situations in reality, map information is not always valid. Therefore, to address the problem of unreliable map information in the vehicle trajectory prediction task, this paper proposes a vehicle traj ectory prediction model based on Transformer and graph convolutional neural network(GCN). First, the processed data are fed into different Transformers to encode the temporal information of vehicles in each scene. Secondly, a graph convolutional neural network combined with an attention mechanism is used to efficiently obtain the interactions between vehicles. Finally, the traj ectory interaction features are decoded using a multimodal decoder and trajectory prediction is accomplished. Experimental results on the Argoverse dataset show that the model outperforms other existing map-free prediction methods in certain performance metrics, and reaches the level of other map-based traj ectory prediction models in certain performance metrics.