Attention -Based GRU for Driver Intention Recognition and Vehicle Trajectory Prediction

• Published in: 2020 4th CAA International Conference on Vehicular Control and Intelligence (CVCI) pp. 86 - 91
• Main Authors: Hao, Zixu, Huang, Xing, Wang, Kaige, Cui, Maoyuan, Tian, Yantao
• Format: Conference Proceeding
• Language: English
• Published: IEEE 18.12.2020
• Subjects: attention mechanism; Gated Recurrent Unit; intention recognition; Prediction algorithms; Predictive models; Probability; Safety; Testing; Training; Trajectory; trajectory prediction
• DOI: 10.1109/CVCI51460.2020.9338510

In human-machine cooperative decision making and control of intelligent vehicle, the intelligent system needs to understand driver's intention and desired vehicle trajectory in order to assist driver with safety driving in complex traffic scenes. In this paper, a vehicle trajectory prediction encoder-decoder model based on Gated Recurrent Unit (GRU) with attention mechanism is proposed. The proposed model is comprised of intention recognition module and trajectory prediction module. The intention recognition module was employed for recognizing driver's intention and calculating the probabilities of turning-left, lane-keeping, turning-right. The trajectory prediction module predicts vehicle trajectory using GRU decoder with attention mechanism, which takes vehicle historical position as input and predicts future position. Both intention recognition module and the trajectory prediction module share one encoder to save time. The NGSIM dataset was employed for training and testing. The experimental results indicate, comparing with traditional methods, the proposed horizontal-longitudinal decoupling hierarchical trajectory prediction method based on GRU neural network can predict driver's desired vehicle trajectory in a long prediction horizon and the attention mechanism improve the trajectory prediction accuracy at the same times.