Multi-Task Collaborative Attention-Based Temporal Convolutional Network for Lane Change Prediction

Lane change prediction is critical for autonomous vehicles in mixed traffic, which can improve traffic efficiency and reduce traffic risk. However, accurate lane change prediction is a great challenge due to complex traffic environments and uncertain human intentions. Therefore, a multi-task collaborative attention-based Temporal Convolutional Network (TCN) is proposed to establish the connection between environments and intentions. First, an attention-based TCN is proposed to extract the underlying features, where the global attention is used to drive TCN to focus on relevant information of the surrounding environment. Secondly, this paper proposes a multi-task Collaborative method, which predicts both lane change intention and Lane Change Time (LCT). The extensive experimental results on highD dataset show that our method has encouraging accuracy. The proposed method improves the accuracy of lane change intention prediction by 7.69% compared to existing methods.