Generating Arachnid Robot Gaits with Cyclic Genetic Algorithms

Learning gaits for walking robots is difficult because the elements of control require repetition of a sequence of integrated steps. Controllers for the six-legged robots, such as the Stiquito, and the eight-legged spider robots, in development, exemplify the difficulties of evolving gaits. The controllers for these robots store the actuator commands in a field-programmable gate array as a list of integers. This list must be ordered in the proper sequence of activations that, when continually repeated, produces a gait. This paper discusses the Cyclic Genetic Algorithm (CGA) used to evolve gait actuation lists, and discusses the extension of the CGA from use on the six-legged ant robot to the eight-legged spider robot. Optimal gaits for the sixlegged were evolved in previous work, expanding the operators used in the CGA produced optimal gaits for the eight-legged robot.