Autonomous driving needs to accurately predict the trajectory of surrounding vehicles and pedestrians to reduce the risk of accidents 1. Traditional methods (such as physical models and machine learning) have limitations in complex scenarios and high-dimensional data processing, while existing deep learning models (such as LSTM, GRU) lack the ability to model long-term predictions and real-time interactions. Therefore, more advanced algorithms are needed to improve the accuracy and adaptability of trajectory prediction. To solve the above problems, this paper proposes a gated cyclic unit model (STA-GRU) that combines temporal and spatial characteristics with attention mechanism, aiming to improve the accuracy and robustness of trajectory prediction in complex traffic scenarios. In the simulation test of NGSIM data set, the mean RMSE of 1s is 0.32284. This model is capable of enhancing the accuracy of vehicle trajectory prediction. As a result, it furnishes crucial data support for traffic signal control, traffic flow optimization, and other related aspects. By doing so, it effectively improves traffic efficiency and mitigates congestion-related accidents. Furthermore, the incorporation of the spatio - temporal attention mechanism augments the interpretability of the model, facilitating a more in - depth understanding of its operational principles and prediction outcomes.
Liu et al. (Wed,) studied this question.