Trajectory Prediction Method Involving Strong Map Coupling for Automated Vehicles

Vehicle trajectory prediction is a vital link con-necting the perception module and decision-making module of autonomous vehicles and is closely related to driving safety. Considering that the prediction model under the blessing of high definition (HD) maps has a significant performance improvement, this paper proposes a vehicle trajectory prediction method with strong map coupling based on Generative Adversarial Network (GAN). A graph query mechanism is proposed to realize the fusion of vehicle state features and map topology. In this paradigm, the map's guidance and constraints on the trajectory are described according to the available map nodes at the vehi-cle's current position. To maximize the use of the rich information of the HD map, a strongly map-coupled discriminator model is proposed to judge whether the predicted trajectory follows the guidance and constraints of the lane centerline, thereby improving the usability of the prediction. Experiments on the Argoverse dataset show that our prediction method outperforms state-of-the-art prediction systems, including a GAN, Jean and TNT. The strong coupling of the HD map significantly improves the accuracy of the predicted trajectory.