What can engineers learn from insect vision?

For a number of years the author has been running in Canberra a program which combines (a) the performance of the visual system of the bee measured by a variety of visual behaviours, (b) the electrophysiology of identified neurons that are vital components of the visual processing, (c) computations with likely algorithms which are adequate models of the insect visual processing and (d) inspiring hardware which implements these findings and puts them into fast processing of flowfields and pattern. The general idea is that human technology would do very well indeed if it could perform as well as common large inserts that fly around by day, and almost anything new that can be discovered about insect visual processing, either in performance or mechanisms, is likely to be useful. The author shows that insects measure range of nearby objects as they fly. They measure angular velocity and parallax by non-directional colour-blind systems. They have "fast" and "slow" motion detectors for gratings and small objects, and self-motion is important for vision of the 3D world. In pattern vision humans distinguish between sharp ongoing vision and poor visual memory, so bees must use a few simple cues. The inclination of one bar or edge can be discriminated by flying bees but not two or more orthogonal bars at the same place. When a large pattern is composed of bold gratings the bees can discriminate their re-arrangement if they subtend more than about 10/spl times/10 facets on the eye. Pattern discrimination, like colour discrimination, seems to depend upon the overlap of broadly tuned neurons which function in groups covering local or even very large regions of the retina. The central combinations of simple templates can then match in number the possible combinations found in patterns in the visual world. Truly parallel circuitry is needed to implement these principles.<>