<p>Multi-vehicle trajectory prediction is critical for traffic safety, autonomous driving, and transportation mobility. However, existing methods suffer from inadequate spatial interaction modeling, weak long-term temporal dependency capture, and insufficient explicit driving intent modeling, which collectively hinders prediction accuracy in dynamic traffic. To address these issues, we propose a BiLSTM-based model integrating a joint spatiotemporal attention mechanism and an explicit driving intent module. The encoder employs a BiLSTM with temporal attention to extract key historical features and convolutional neural networks (CNNs) to capture spatial interactions, while a joint attention mechanism fuses these representations with driving intent embedded as a high-dimensional vector augmenting the process. The BiLSTM decoder generates future trajectories using the fused features. Evaluated on the next generation simulation (NGSIM) dataset, our model achieves an overall RMSE reduction of approximately 11.9% compared to the baseline LSTM and 4.8% compared to the convolutional social-LSTM (CS-LSTM) for one-second future trajectory predictions, validating the efficacy of joint spatiotemporal attention and intent fusion in modeling complex traffic interactions.</p>