IALH Research Fellow Brandon Haworth has published a new research article entitled Deformable Elliptical Particles for Predictive Mesh-Adaptive Crowds. Collaborating authors include Dominic Ferreira and Liam Shatzel. This article was published in MIG ’24: Proceedings of the 17th ACM SIGGRAPH Conference on Motion, Interaction, and Games.
Abstract:
Crowd simulation is an essential tool in the modern animation toolkit with a wide variety of applications, ranging from urban planning and evacuation analysis to creating believable crowds in both film and games. Modelling the movements and behaviours exhibited in crowds of characters often relies heavily upon the fundamental representation of agents used in the simulation. We propose a mesh-adaptive deformable representation of agents to increase simulation fidelity, generate novel behaviour, and support diversity. Most existing methods use static primitive geometries to represent agents, which neglects the variety in character meshes and animation states. We present an efficient method for generating elliptical particles, which can deform to any mesh and animation state in real-time. The method is straightforward, robust, and exceptionally generalizable. We develop a novel steering methodology for our agent representation method that solves the subsequent challenges of incorporating dynamic asymmetric particle representations. The physically-based algorithm features predictive collision avoidance, incorporating an activation function that encodes the rotational uncertainty of agents. Our model exhibits realistic packing behaviour of agents under high-density conditions, as well as unpacking when the flow is unconstrained. In addition to the compelling qualitative behaviour generated by our model, we present statistical comparisons with existing methods. Our method intrinsically supports the steering of crowds of diverse mesh morphologies without ad-hoc character-specific rules, while affording artist-defined steering components in the character mesh.
To read the full article, see https://doi.org/10.1145/3677388.3696329