Spatial-Temporal Attentive LSTM for Vehicle-Trajectory Prediction

Vehicle-trajectory prediction is essential for intelligent traffic systems (ITS), as it can help autonomous vehicles to plan a safe and efficient path. However, still challenging task because existing studies have mainly focused on the spatial interactions of adjacent regardless temporal dependencies. In this paper, we propose spatial-temporal attentive LSTM encoder–decoder model (STAM-LSTM) predict vehicle trajectories. Specifically, attention mechanism used capture relationships among neighboring then obtain global feature. Meanwhile, designed distinguish effects different historical time steps future trajectory prediction. addition, motion feature extracted reveal influence dynamic information vehicle-trajectory prediction, combined with local features represent integrated target at each moment. The experiments were conducted public highway datasets—US-101 I-80 in NGSIM—and results demonstrate that our achieves state-of-the-art performance.