The evolution of neural network controllers for the targeted swing of cockroach-like robot

Abstract

In this work, agents are evolved to control the locomotion of legged robots. The work has been done on a 5-DOF simulated cockroach robot leg powered by artificial muscles known as braided pneumatic actuators. The problem was addressed in making a targeted swing movement from a number of possible initial leg configurations. The agents are evolved to successfully compensate for these differences and navigate obstacles in a robust swing.