In the field of autonomous driving vehicles, accurately predicting vehicle trajectories is crucial for ensuring safety and enhancing ride comfort. However, predicting the trajectory of a target vehicle is challenging due to complex interaction dynamics with nearby vehicles in different scenarios. To improve prediction accuracy, we introduce a novel spatiotemporal neural network with contrastive learning capabilities. This method delves deep into potential encodings within interaction scenes from both local and global perspectives and incorporates future trajectory features through a Conditional Variational Autoencoder (CVAE). We rigorously evaluate this approach through ablation experiments using the US-101 and I–80 datasets from the National Highway Traffic Safety Administration's Next Generation Simulation (NGSIM) dataset. The research results demonstrate that our method outperforms the current state-of-the-art techniques in terms of performance.
Spatio- Temporal Neural Network with Contrastive Learning for Vehicle Trajectory Prediction
2023-10-20
396930 byte
Conference paper
Electronic Resource
English