Developing a massive real-time crowd simulation framework on the GPU

Crowd simulations are used to imitate the behaviour of a large group of people. Such simulations are used in industries ranging from video-games to public security. In recent years, research has turned to the parallel nature of GPUs to simulate the behaviour of individuals in a crowd in parallel. This allows for real time visualisation and interaction with massive crowds and/or their environment. There are, however, no tool that could facilitate the integration of GPU-accelerated crowd simulations in an existing program like a game. This paper proposes an OpenCL port of the PedSim library. It is a tool that allows CPU-run crowd simulations to be used in another application. It uses the social forces algorithm which is a crowd simulation method using a set of forces to compute each agents motion. Our implementation is compared in performance and functionality with the original PedSim. Finally, we discuss the functionalities missing from our implementation to provide a more complete crowd simulation