Vehicle Trajectory Prediction Using Hierarchical Graph Neural Network for Considering Interaction among Multimodal...

Area of Science:

• Autonomous Systems
• Artificial Intelligence
• Robotics

Background:

• Predicting surrounding vehicle trajectories is crucial for autonomous vehicle safety and functionality.
• Existing methods struggle to account for unobservable maneuvers and numerous maneuver combinations, limiting interaction modeling.

Purpose of the Study:

• To propose a novel hierarchical graph neural network (HGNN) for predicting vehicle trajectories.
• To effectively model and consider interactions among multiple, potentially unobservable, vehicle maneuvers.

Main Methods:

• Developed a hierarchical graph neural network architecture.
• Represented relationships among vehicle maneuvers in a graph structure to capture interactions.
• Evaluated the model using both public and real-world driving datasets.

Main Results:

• The proposed HGNN demonstrated superior prediction performance compared to previous methods.
• The model showed significant improvements, particularly in highly interactive driving situations.
• Accurate prediction of multiple maneuvers and their interactions was achieved.

Conclusions:

• The hierarchical graph neural network effectively addresses the limitations of prior trajectory prediction methods.
• The approach enhances the ability of autonomous vehicles to navigate complex, interactive environments.
• This work contributes to safer and more reliable autonomous driving systems.