Vehicle Trajectory Prediction in Highway Weaving Areas Considering Interaction-Transfer Characteristics

In order to enhance safety in dynamic traffic, autonomous vehicles must predict surrounding vehicles’ trajectories to reduce accident risk. Addressing the issue of insufficient consideration of vehicle interaction transfer characteristics in existing studies, this paper proposes a trajectory prediction framework for highway weaving areas that explicitly accounts for vehicle interaction transfer effects. This method uses a hierarchical dynamic graph to comprehensively represent interaction transfer characteristics and achieves cross-layer coupling at both the feature level and the decision level, which helps to address the complexity of vehicle interactions in mixed traffic environments. The proposed graph attention network for trajectory prediction, which models the transfer characteristics of vehicle interactions, fully exploits temporal continuity and spatial coupling to jointly extract spatiotemporal features. This paper infers latent interaction information transmitted by indirectly adjacent vehicles based on vehicle historical data. The inferred information is integrated with vehicle state features and direct interaction information. A hierarchical dynamic graph is then constructed to account for interaction transfer among vehicles. To overcome insufficient modeling of coupled spatiotemporal dependencies, we design a spatiotemporal dynamic graph attention network model for trajectory prediction. Experiments were conducted on publicly available datasets. The results indicate that the model’s predictions surpass those of the currently advanced baseline models, demonstrating higher prediction accuracy.