Jamie Snape, Jur van den Berg, Stephen J. Guy, and Dinesh Manocha
Department of Computer Science, University of North Carolina at Chapel Hill
Simulations of two hundred crowd agents with randomly chosen destination locations at the four corners of the level.
The RVO2 Library integration with Unreal Development Kit (UDK) provides a custom subclass of the UnrealScript class GameCrowdAgent that interfaces with RVO2 Library using the DLLBind functionality provided in recent releases of UDK beta for the Microsoft Windows platform. Level creators may spawn crowds using the RVO2 Library integration in a near-identical workflow to the built-in UDK crowd system. All relevant parameters of the RVO2 Library integration may be configured within UDK Editor.
RVO2 Library (above left): QuickTime Movie (18.6 MB), Vimeo Video, YouTube Video
UDK crowd system (above right): QuickTime Movie (18.3 MB), Vimeo Video, YouTube Video
Average collision count (per agent per frame, lower is better)
RVO2 Library: 0.0026
UDK crowd system: 0.0989
Average simulation rate (frames per second, higher is better)
RVO2 Library: 30.43
UDK crowd system: 26.80
Empty scene: 40.85
(UDK March 2011 beta 2, 2.53 GHz Intel Core 2 Duo processor, 4 GB memory, NVIDIA GeForce 9400M graphics, DirectX 9.0c, 200 crowd agents, 4000 frames)
This work was supported by ARO under Contract W911NF-10-1-0506, by NSF under Award 0917040, Award 0904990, and Award 1000579, and by DARPA and RDECOM under Contract WR91CRB-08-C-0137.
Jur van den Berg, Stephen J. Guy, Jamie Snape, Ming C. Lin, and Dinesh Manocha, "RVO2 Library: Reciprocal Collision Avoidance for Real-Time Multi-Agent Simulation."