FF-GAN: A Fuzzy Fractal-Based Generative Adversarial Network Trajectory Prediction Method for Complex Tasks

View at publisher Summarise with VIRAT
Authors: Tian Yang, Gang Wang, Jian Lai, Yang Wang
Cited by: 0
References: 50

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Pedestrian trajectory prediction is crucial in areas like autonomous driving. However, some prediction methods based on generative adversarial networks (GAN) still have some shortcomings, such as limited adaptability to complex situations and inadequate optimization of the model’s loss. To address these two issues, we propose a trajectory prediction method based on fuzzy fractals. Firstly, to enhance the model’s adaptability to complex situations, we model the conservative soft output states of the GAN discriminator network using triangular fuzzy numbers. Secondly, we convert the output of the discriminator network into belief outputs in evidence theory through normalization processing. This is because evidence theory can distinguish between ‘uncertainty’ and ‘ignorance’, making it widely applicable in dealing with complex uncertain problems. Thirdly, given the effectiveness of fractals in handling information complexity, we embed the information fractal formula into the model’s loss function, introducing a fuzzy fractal loss term to optimize the model’s detailed loss. Experiments on the ETH/UCY dataset reveal that this novel approach significantly enhances trajectory prediction accuracy, particularly for trajectories with heightened uncertainty, such as when pedestrians halt, move within confined spaces, or abruptly alter directions.

Fields of Study & Subjects

Autonomous Vehicle Technology and Safety Evacuation and Crowd Dynamics Adversarial Robustness in Machine Learning

Publication details

Journal / source
Cybernetics & Systems
Volume
57
Issue
7
Pages
1744–1767
Publication date
25 March 2026
Language
EN