Lane-changing trajectory prediction based on multi-task learning

Abstract As a complex driving behavior, lane-changing (LC) behavior has great influence on traffic flow. Improper often leads to accidents. Numerous studies are currently being conducted predict lane change trajectories minimize dangers. However, most of their models focus how optimize input variables without considering the interaction between output variables. This study proposes LC trajectory prediction model based multi-task deep learning framework improve safety. Concretely, in this work, coupling effect lateral and longitudinal movement is considered process. Trajectory changes two directions will be modeled separately, information completed under framework. In addition, fragments clustered by features, type recognition added as an auxiliary task. Finally, process style Long Short-Term Memory (LSTM). The training testing with data collected simulator, that proposed method expresses better performance compared several traditional models. result showed can enhance accuracy Advanced Driving Assistance System (ADAS) reduce accidents caused changes.