Trajectory Prediction for Autonomous Driving of New Energy Vehicles based on Graph Transformer Networks

Abstract

Accurate trajectory prediction is crucial for ensuring the safe and efficient operation of autonomous vehicles, especially in complex and dynamic traffic environments. In this paper, we propose a novel approach for trajectory prediction in the context of new energy vehicles (NEVs) by combining Graph Neural Networks (GNNs) and Transformer models. Our model, Graph Transformer Networks (GTNet), effectively captures the spatial-temporal dependencies and interactions among multiple agents, including other vehicles, pedestrians, and traffic infrastructure, to predict future trajectories. Unlike traditional methods that focus only on local interactions, our approach integrates global context through the Transformer model to better handle long-range dependencies. The GNN component captures the underlying structure of the traffic environment, such as road networks and vehicle relationships, while the Transformer model leverages its attention mechanism to focus on relevant information across time and space. This combination allows for a more accurate prediction of future vehicle trajectories, especially in scenarios where the motion of surrounding agents is highly interdependent. We evaluate our approach on real-world datasets and demonstrate its superior performance compared to baseline methods, achieving improved prediction accuracy for both short- and long-term trajectory forecasts. Our results highlight the effectiveness of integrating Graph Neural Networks with Transformer models for enhancing trajectory prediction in autonomous driving, particularly for new energy vehicles navigating dynamic traffic scenarios.