Because of communication delays in mixed traffic scenarios, autonomous vehicles may receive outdated information about surrounding vehicles, leading to inaccurate trajectory prediction and compromised safety. To address this issue, this study proposes a spatio-temporal delay-aware long short-term memory (SDA-LSTM) model. Using the Next Generation Simulation dataset, this research analyzes the interactions between spatial distribution and motion parameters in multi-vehicle cooperative driving scenarios. It constructs a spatial relationship delay adaptive module to dynamically weight the contributions of heterogeneous neighboring vehicle data to prediction outcomes. By innovatively integrating a delay-aware temporal encoding mechanism in the model framework, it effectively characterizes the differences in time-delay features of historical data. Eventually, an SDA-LSTM with spatio-temporal delay perception capability is established. Experiment results demonstrate that SDA-LSTM surpasses LSTM in prediction accuracy, with an average root mean square error reduction of approximately 52% across all prediction steps. It also exhibits robust capability to capture vehicle motion intentions across both straight-driving and lane-changing scenarios. The prediction errors remain within acceptable ranges for aggressive, normal, and conservative driving styles, with the smallest prediction error observed in normal driving style vehicles. Furthermore, through delay-aware weight heatmaps analysis, the study verifies its adaptive historical data weighting and irrelevant information pruning capabilities.
Cui et al. (Fri,) studied this question.