Learning to Simulate Flocking Behaviour with Graph Neural Networks
Date
2023-03-16
Authors
Bobyn, Steven
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Crowds are often modelled as particle systems governed by "social forces"; however, the forces are unknown and difficult to approximate. Graph neural networks (GNNs) are excellent at learning patterns in unstructured data like crowds; GNNs show promise for learning crowd simulation tasks and have demonstrated impressive results simulating complex physics. Flocks are a simplified version of crowds that also exhibit chaotic behaviour; however, they can be simulated with simple rules. In this work, we aim to apply a GNN-based framework to learn to simulate flocking behaviour.
Description
Keywords
Crowd Simulation, Deep Learning, Graph Networks