We propose novel gaze control algorithms for active perception in mobile autonomous agents with directable, foveated vision sensors. Our agents are realistic artificial animals, or animats, situated in physics-based virtual worlds. Their active perception systems continuously analyze photorealistic retinal image streams to glean information useful for controlling the animat eyes and body. The vision system computes optical flow and segments moving targets in the low-resolution visual periphery. It then matches segmented targets against mental models of colored objects of interest. The eyes saccade to increase acuity by foveating objects. The resulting sensorimotor control loop supports complex behaviors, such as predation.
Registration: ISBN 978-0-262-51091-2
Copyright: August 4-8, 1996, Portland, Oregon. Published by The AAAI Press, Menlo Park, California.