Synthetic vision visual perception for computer generated forces using the programmable graphics pipeline

Abstract

In visual simulations, the human must make most of her decisions based on the visual cues rendered to her display. On the other hand, synthetic forces have the luxury of basing their decisions on the data contained in the simulation's model. Line of sight calculations are often examples of the synthetic player's excess of information. Current methodologies for determining a synthetic player's line of sight to a target are generally variations of a ray-casting technique. Hiding from a synthetic player "in plain sight" by using shadow, camouflage, or by simply remaining motionless is not possible. Synthetic vision is an alternative to ray-casting. We perform multiple renders from each synthetic player's point of view and temporarily maintain those images in graphics memory. We then execute vertex and fragment shader programs to make comparisons of the stored images. All the renders and calculations are performed on the Graphics Processing Unit (GPU) and the result is returned to the synthetic player in the form of an annotated list of visible targets. Performing these target visibility calculations on the GPU gives the synthetic player a more robust spectrum of visual inputs with which to make decisions, enabling more realistic behaviors.