Cooperative Trajectory Prediction Using IVC and Accuracy-Aware Attention with Inaccurate GPS Data

As autonomous driving vehicles have emerged, Under-Resourced Automated Vehicle (URAV) and Connected Automated Vehicle (CAV) could coexist in similar regions. For safe driving in the coexistence scenario, URAV and CAV should accurately predict future trajectories of surrounding vehicles (i.e., neighbors). To predict the trajectories by URAV, two primitives are required: 1) positioning via low-price devices (e.g., GPS) and 2) Inter-Vehicle Communication (IVC). However, using only these primitives could induce inaccurate trajectory prediction because of inaccurate GPS data and failure in packet transmission via IVC. To address the limitations of URAV, in this paper, we propose a novel scheme for trajectory prediction whereby URAV cooperates with CAV by exploiting the powerful resources of CAV via IVC. For improving the accuracy of cooperative trajectory prediction, our scheme employs three modules: 1) Cooperative Position Share and Compensation (CPSC) to mitigate positioning error of GPS data, 2) Accuracy-aware Social Predictor (ASP) to differentiate the influence of neighbors on prediction based on their position accuracies, and 3) Worst Error Mitigation (WEM) to reduce maximum prediction error. To our knowledge, our scheme is the first to enable accurate trajectory prediction of neighbors with limited resources via cooperation with CAV. Intensive evaluation using QUALNET and Pytorch demonstrates that our scheme significantly outperforms previous approaches in terms of average and maximum prediction errors.