Graph-Based Spatial-Temporal Convolutional Network for Vehicle Trajectory Prediction in Autonomous Driving

Forecasting the trajectories of neighbor vehicles is a crucial step for decision making and motion planning autonomous vehicles. This paper proposes graph-based spatial-temporal convolutional network (GSTCN) to predict future trajectory distributions all using past trajectories. tackles spatial interactions graph (GCN), captures temporal features with neural (CNN). The are encoded decoded by gated recurrent unit (GRU) generate distributions. Besides, we propose weighted adjacency matrix describe intensities mutual influence between vehicles, ablation study demonstrates effectiveness our scheme. Our evaluated on two real-world freeway datasets: I-80 US-101 in Next Generation Simulation (NGSIM). Comparisons three aspects, including prediction errors, model sizes, inference speeds, show that can achieve state-of-the-art performance.