Neural processing and control for artificial compound eyes

This paper gives an overview of an analog implementation for neural processing of compound eye sensors in robotic control. The author has successfully produced several working robotic devices guided by compound eye vision. Vision processing and control feedback are accomplished using both electronic analog neural nets and real-time, hardware-in-the-loop software neurons. To date, the author has integrated compound eye controllers with great success on four robotic systems. These systems exhibit robust motion tracking and detection of light sources and patterns. This paper describes an all-analog, 8-ommatidia pendulum ping-pong player and its neural implementation. The concepts given in this paper are fairly general and can be applied to any compound eye sensing system. The author's main finding is the amazing robotic behavior possible using incredibly simple neural control strategies.<>