Hierarchical Spatial-Temporal Attention Network for Lane-Changing Behavior Prediction of Autonomous Vehicles on Multi-Lane Highways

Accurate prediction of driving behavior is crucial for autonomous driving to ensure safe navigation while enhancing overall safety and comfort. This research presents a novel hierarchical spatio-temporal attention-based model for predicting driving behavior. By leveraging a combination of Attention-based Temporal Convolutional Networks (ATCN) and Transformer (TF) architecture, the proposed model significantly improves performance. The effectiveness of the model is evaluated using the publicly available NGSIM dataset. Results indicate that the proposed model outperforms existing state-of-the-art models, achieving 96.28% accuracy for lane keeping (LK), 97.01% for lane change to left (LCL), and 93.15% for lane change to right (LCR), resulting in an overall accuracy of 96.29%. These findings highlight the model's potential to improve autonomous vehicle systems across diverse driving scenarios.