Fusing transportation rules and diverse motion behaviors for trajectory prediction in traffic intersections

Area of Science:

• Autonomous Driving
• Robotics
• Computer Vision

Background:

• Trajectory prediction is crucial for autonomous driving safety.
• Traffic intersections present complex challenges due to agent interactions and traffic signals.
• Existing models often fail to capture nuanced driver attention and signal influences.

Purpose of the Study:

• To develop an advanced trajectory prediction model for traffic intersections.
• To address limitations in current models regarding agent interactions and traffic signal impacts.
• To improve the accuracy and robustness of autonomous vehicle navigation.

Main Methods:

• A directed graph was designed to model agent attention and interactions.
• An action prediction method was developed using traffic signal information.
• Multiple t-distributions were employed to forecast diverse agent motion patterns.

Main Results:

• The proposed model demonstrated superior performance compared to twenty-four existing trajectory prediction methods.
• Experiments were conducted on two distinct traffic intersection datasets.
• The algorithm effectively captured agent interactions and traffic signal influences.

Conclusions:

• The novel model significantly enhances trajectory prediction accuracy at intersections.
• It provides a more comprehensive approach by integrating agent attention, traffic signals, and motion diversity.
• This research contributes to safer and more reliable autonomous driving systems.