Vehicle Trajectory Prediction on Interaction Driving Scenarios

Predicting the trajectories of vehicles around an autonomous vehicle can enhance its ability to assess the risk of the driving environment, allowing it to clarify the drivable area and make safe and reliable driving path planning. Current vehicle trajectory prediction studies usually only consider the historical trajectory data of the target vehicle, with less consideration of the effects of vehicle interactions, resulting in inferior prediction accuracy. We propose a model based on Long Short-Term Memory (LSTM) and Transformer to improve the accuracy of trajectory prediction by effectively extracting the interaction effects of the neighboring vehicles on the target vehicle through the multi-head attention mechanism of the Transformer encoder. We experimentally evaluate the model on the public trajectory dataset NSGIM by comparing the differences between predicted and ground truth traj ectories. The results show that our model predicts more accurately compared to existing methods.