Animat Vision Active Vision in Arti cial Animals

Animat Vision Active Vision in Arti cial Animals Tamer F. Rabie Doctor of Philosophy Graduate Department of Electrical and Computer Engineering University of Toronto 1999 We propose and demonstrate a new paradigm for active vision research which draws upon recent advances in the elds of arti cial life and computer graphics. A software alternative to the prevailing hardware vision mindset, animat vision prescribes arti cial animals, or animats, situated in physics-based virtual worlds as autonomous virtual robots with active perception systems. To be operative in its world, an animat must autonomously control its eyes and actuated body. Computer vision algorithms continuously analyze the retinal image streams acquired by the animat's eyes, enabling it to locomote purposefully through its world. We describe an initial animat vision implementation within lifelike arti cial shes inhabiting a physics-based, virtual marine world. Emulating the appearance, motion, and behavior of real shes in their natural habitats, these animats are capable of spatially nonuniform retinal imaging, foveation, retinal image stabilization, color object recognition, color stereo obstacle avoidance, and perceptuallyguided navigation. These capabilities allow them to foveate and pursue moving targets of interest, such as other arti cial shes, while exercising the sensorimotor control necessary to avoid collisions and predators. We demonstrate that the animat vision paradigm extends to virtual environments inhabited by virtual humans. Animat vision o ers a fertile approach to the development, implementation, and evaluation of computational theories that profess sensorimotor competence for animal or robotic situated agents.