CrowdE: Crowd Simulation on Antelope Island
In modern day games, many agents are navigating independently through detailed environments. By using the massive computational capabilities of the graphical co-processor, exposed by the NVIDIA CUDA programming environment, we aimed to significantly increase the amount of game units. Based on the flow field algorithm found in literature, a novel
hybrid system was developed using a first, coarse A* pathfinding step to
eliminate unnecessary work during the potential field generation. Additionally, an optimized potential solver was implemented using persistent threads and inter-block communication. Results show that our implementation achieves considerable speedups compared to existing path planning
systems. Up to 100,000 agents can be simulated in real-time on a mainstream GPU.
Link to the master's thesis (summary in English, main text in Dutch): http://search.ugent.be/meercat/x/view...
The texture data used in this demo was made available by Utah's State Geographic Information Database (see http://gis.utah.gov/hro2006).
Developed by the Multimedia Lab of Ghent University - IBBT
http://multimedialab.elis.ugent.be/about
http://www.ugent.be/en
http://www.ibbt.be/en