Enabling Accurate Trajectory Prediction of Human Driven Vehicles in the Hybrid Driving Scenario

Accurate assessment of the surrounding traffic dynamics is crucial for autonomous driven vehicles (AVs). Specifically, in the hybrid driving scenario, the behaviors of human driven vehicles (HVs) have a significant impact on AVs, due to that HVs usually don't share data with other vehicles. Thereby, it is of high importance for AVs to understand the intentions of surrounding HVs and predict their trajectories. In this paper, we propose a trajectory prediction framework for HVs in the hybrid driving scenario based on the collaboration of multiple AVs. Specifically, we first represent the interactions among different AVs by combining dynamic graphs and dual graphs. Then, an attention network is constructed for feature sharing and integration among AVs, based on which the future trajectory of HVs is predicted accordingly. To validate the performance of the proposed framework, we generate trajectory datasets of the hybrid driving scenario based on the joint simulation of CARLA and SUMO. Experimental results show that our approach outperforms the baselines in terms of prediction accuracy.