1 Large-Scale Agent-based Pedestrian and Crowd Simulation in Real-Time

Abstract—This paper deals with the implementation details and results of simulating a city populated by a large number of pedestrians. The goal of the simulation was to, as realistically as possible, simulate large numbers of people going about their daily lives, interacting with each other and the city environment, in real-time. We also simulated dense crowds and realistic collision avoidance techniques, and tried to replicate some observations of previous studies. Simulators based on the mechanics of human interaction can easily become inconveniently complex and/or resource expensive. As this has been the main risk during the project we’ve been careful in the implementation to keep coupling as low as possible and to construct interfaces that allow for scaling and adding of new behaviour in “isolation”, without having to modify prior code. The concern for performance was just as real – the simulation, after all, was to support thousands of interacting pedestrians walking about in real-time. In the end, the resulting simulation turned out to be a good and efficient representation of inner-city pedestrians, and was mostly fine in handling the issues of denser crowds. This may potentially be extended for use in city and public transport planning, producing large amounts of data for data mining or as a basis for further development into city life dynamics and the artificial intelligence of individuals in a populated environment