SEEM: A Sequence Entropy Energy-Based Model for Pedestrian Trajectory All-Then-One Prediction

Area of Science:

• Computer Science
• Robotics
• Artificial Intelligence

Background:

• Pedestrian trajectory prediction is crucial for autonomous systems.
• Existing models lack diversity, accuracy, and stability.

Purpose of the Study:

• To propose a novel Sequence Entropy Energy-based Model (SEEM) for improved pedestrian trajectory prediction.
• To address limitations of current models in diversity, accuracy, and stability.

Main Methods:

• SEEM utilizes a generator and an energy network.
• Optimizes sequence entropy via local variational inference of f-divergence to maximize mutual information.
• Employs probability distribution clipping and an energy network for trajectory selection (all-then-one strategy).
• Incorporates zero-centered potential energy regularization for training stability.

Main Results:

• SEEM demonstrates superior performance over state-of-the-art methods.
• Achieves significant improvements in diversity, accuracy, and stability.
• Validated on synthetic and public benchmark datasets.

Conclusions:

• SEEM offers a robust solution for diverse, accurate, and stable pedestrian trajectory prediction.
• The proposed methods effectively address key challenges in the field.
• SEEM advances the capabilities of autonomous driving and social robot applications.