UCL Discovery

Abstract

Computer Graphics have become important for many applicationsand the quality of the produced images have greatly improved. Oneof the interesting remaining problems is the representation of densedynamic environments such as populated cities. Although recentlywe saw some successfulwork on the rendering such environments,the real?time simulation of virtual cities populated by thousands ofintelligent animated agents is still very challenging.In this paperwe describe a platformthat aims to accelerate the developmentof agent behaviours. The platform makes it easy to enterlocal rules and callbacks which govern the individual behaviours.It automatically performs the routine tasks such as collision detectionallowing the user to concentrate on defining the more involvedtasks. The platform is based on a 2D-grid with a four-layered structure.The two first layers are used to compute the collision detectionagainst the environment and other agents and the last two are usedfor more complex behaviours.A set of visualisation tools is incorporated that allows the testingof the real?time simulation. The choices made for the visualisationallow the user to better understand the way agents move inside theworld and how they take decisions, so that the user can evaluate ifit simulates the expected behaviour.Experimentation with the system has shown that behaviours inenvironments with thousands of agents can be developed and visualisedin effortlessly.

Downloads