Data driven interaction-aware trajectory prediction for urban driving

As an important tool to promote the development of intelligent transportation systems, autonomous driving can effectively reduce human-induced traffic accidents, relieve traffic congestion, and reduce environmental pressure under certain conditions. It is a key technology that needs to be developed urgently. As the core content of autonomous driving, trajectory prediction needs to perform an accurate risk assessment on the current driving scene of the autonomous vehicle and predict the trajectory of the vehicle based on the risk assessment results. Thereby, it can not only reliably make the operation of the autonomous vehicle safe and efficient but also guarantee reasonable interaction with other traffic vehicles. This requires accurate prediction of the future motion trajectories of other traffic vehicles in the driving scene in each predicting cycle to improve the stability and safety of autonomous driving. In this dissertation, the method of interactive perception is used to predict the future trajectory in the urban driving scene. The RNN-based Encoder-Decoder structure design is adopted, and GRU and FC are used for dynamic encoding from vehicle trajectories to extract dynamic features. In Feature extraction, D-GCN is used to capture interactive information, and then RNN and FC decoders are used for decoding to output the predicted vehicle tracks, which can capture more accurate information than traditional methods. The loss function used in this dissertation is the combination of RMSE and MAE, which not only controls the gradient of the loss to a certain extent but also enhances the robustness of training and learning. Through the comparison with other advanced methods and the fitting trajectory results of the final experiment, it is proved that the method described in this dissertation can effectively improve the rationality, safety, and intelligence of the trajectory prediction of autonomous vehicles. The research results ensure the safe and reasonable interaction between autonomous vehicles and other traffic vehicles and promote the improvement of the intelligence level of autonomous vehicles in complex and dynamic urban scenes. Keywords: intelligent driving, trajectory prediction, RNN, D-GCNMaster of Science (Smart Manufacturing