Advanced trajectory prediction framework integrating diverse driving styles for autonomous vehicles

With the prevalence of autonomous vehicles, accurately predicting the motions of surrounding vehicles has become one of the key safety concerns for autonomous driving in mixed traffic. However, human drivers’ personalized driving habits and necessary vehicle interactions complicate the motion prediction process. To address these challenges, this paper proposes an innovative personalized trajectory prediction framework that comprehensively considers individual driving styles and their interactions with surrounding vehicles. This framework comprises two main components: driving style classification and vehicle trajectory prediction. Initially, principal component analysis (PCA) and fuzzy c-means (FCM) algorithms are combined to effectively capture and categorize multiple driving styles. Based on the style information, a global-local enhanced attention module with residual bidirectional long short-term memory network (GLEAResBiLSTM) is constructed to integrate local and global information to obtain the predicted trajectory. Lastly, the next-generation simulation (NGSIM) dataset is used to validate the high performance of the proposed prediction framework. By comparing with the conventional methods, the proposed method reduces the root mean square error (RMSE) by 34.12% in 5 s predictions, which intensely shows the enhanced accuracy and innovative practicality of the proposed prediction framework for autonomous driving.