Evolving electronic circuits for robotic control

Abstract

This paper details the implementation of a hardware genetic algorithm used to evolve a digital circuit capable of controlling the motion of a robot. The digital circuit was a virtual FPGA comprised of functional elements layered in four columns. The virtual FPGA was configured by a bit stream generated by the hardware genetic algorithm which supports only mutation. The robot was in the form of a simulated balancing beam which has 32 outputs connected to the digital circuit and one input for motor control. It was found that after an average of 40000 generations a digital circuit could be evolved to balance the beam for more than five minutes.