Evolution of gaits of a legged robot

Abstract

Gaits of a legged robot were generated from scratch using evolutionary robotic techniques. A physical robot was shared by 50 individuals (software invoked control processes that run the entire robot) in each generation. Each program competed for a better score for its gait and ability to lift the body. A genotype that describes leg motions of each robot was defined and used in behavior generation and reproduction of offspring. An evaluation function that favors early generation of collaborative movements of legs for efficient forward motion of the robot was defined and successfully tested in the experiment.