Evaluation of LSTM GAN for Trajectory Prediction in Connected and Autonomous Vehicles

This paper introduces a novel LSTM GAN model for trajectory prediction in Connected and Autonomous Vehicles (CAVs), combining LSTM’s time-series data handling with GAN’s data generation capabilities. Designed to accurately adapt to various driving scenarios, the model’s efficacy is evaluated in both short and long-trajectory prediction contexts. Initial results show that incorporating dynamic features like velocity and acceleration significantly enhances short-term trajectory predictions. However, long-term predictions pose challenges, particularly in accurately determining the final position. Addressing these issues, the paper suggests model refinements and training adjustments, alongside alternative evaluation metrics. This approach aims to improve the model’s stability and accuracy, especially in complex scenarios. The LSTM GAN model emerges as a promising tool for advancing trajectory prediction in autonomous vehicles, offering valuable insights for integrating LSTM and GAN in this field.