This paper aims to show how a particle-based solver, originally developed for SPH and multiphysics simulations, can be applied with little modification to simulate human crowd motion in an urban environment. The aim is to demonstrate that fluid-inspired numerical methods can be adapted to model agent-based systems and that such cross-domain reuse can extend the reach of established solvers.
The simulation combines the Discrete Multiphysics (DMP) approach with agent-based modelling (ABM) and a graph-based representation of space. Agents interact via inter-particle forces and move according to Newtonian dynamics. Pathfinding is handled via a greedy best-first search on the graph. The implementation uses Python and Numba, with performance tested on large agent populations in idealised urban layouts.
The model reproduces realistic collective behaviours including lane formation, congestion and incident hotspots. It can simulate how events such as crime or antisocial behaviour emerge from local interactions. Results show that a solver originally used in fluid–structure interaction can accurately capture social-agent dynamics by adapting only the force interactions.
The model has been tested in idealised urban layouts and has not been calibrated against real pedestrian data. Future work could explore quantitative validation, dynamic graph updates or applications in reactive transport and evacuation modelling.
The method offers a simple way to repurpose existing particle solvers for crowd dynamics or similar agent-based systems, reducing the need to develop new codes from scratch. It also opens up new ways to model infrastructure, navigation and behavioural rules using tools familiar to researchers familiar with particle-based methods.
Crowd simulations can support urban planning, safety assessments and public-space design. The method can be extended to model social behaviours under stress or uncertainty, such as panic, intoxication or crime. It offers a simple way to test policies or layouts before real-world trials, using physical analogies to explore social problems.
This paper provides a worked example of how particle-based numerical methods can be extended to simulate systems outside classical physics. It also introduces a lightweight graph–particle framework that bridges discrete agent rules with mechanics, offering a new way to apply familiar computational tools to social and behavioural systems.
