Asymmetric Agent Geometries in Synthetic Crowds




Ferreira, Dominic

Journal Title

Journal ISSN

Volume Title



Crowd simulation plays a crucial role in various domains, including urban planning, evacuation analysis, and in the creation of films and games. The representation of agents used in these simulations greatly influences the accuracy of modelling crowd movements and behaviours. Many existing methods lack diversity and use simplified two-dimensional, typically static, agent representations, which constrain the emer gent phenomena and possible applications. By incorporating asymmetric agents, new behaviours and environments can be explored. This thesis investigates two asym metric agent representations that violate many common assumptions requiring novel solutions, particularly in developing a predictive collision avoidance algorithm that accommodates the rotational uncertainty introduced by asymmetric agents. The first model focuses on dynamic agent representations to enhance fidelity and usability in common crowd simulation applications. A mesh-adaptive deformable rep resentation of agents is proposed, which contrasts existing methods that use static primitive geometries. An efficient method for generating elliptical particles that can deform to any mesh and animation state in real time is presented. A physically-based steering algorithm is developed with predictive collision avoidance which accounts for the novel asymmetric agent geometry. The model exhibits realistic packing behaviour of agents under high-density conditions and unpacking when the flow is unconstrained, which is particularly important in evacuation simulation. This approach is exception ally generalizable and supports diverse heterogeneous crowds. The second model considers three-dimensional agent geometries to facilitate the study of human navigation in microgravity environments. The model addresses the complications of a multi-agent non-symmetric particle-based model of astronauts with biomechanically constrained reachable workspace of limbs for steering and col lision avoidance in three dimensions under the conditions of microgravity. A three dimensional predictive collision avoidance algorithm is proposed, which accommo dates the rotational uncertainty of these agents. A multi-layered real-time simulation model is developed for agents aboard spacecraft in microgravity, including the ability for agents to use handles or surfaces to maneuver and float freely through the space-craft. A path finding algorithm is defined over a graph of usable handles, constructing a representation of navigable space. This approach could be used in the planning and safety analysis of future spacecraft design. Both of these models highlight the potential benefits of using asymmetric agent geometries in synthetic crowds research.



Crowd Simulation