Evolution of Locomotion Controllers for Legged Robots

Abstract

The construction of a locomotion controller for legged robots is a difficult and time consuming process. This task can be automated through the use of evolutionary techniques and an appropriately chosen control system. The presented approach utilizes a genetic algorithm that evolves the parameters for a spline controller. The controller outputs its control signals to a virtual robot in a mechanical simulator, enabling the robot designer to preview the robots locomotion patterns. This approach has shown to produce walking patterns for a number of robots with differing morphology. Extensions to the basic spline controller al-low various forms of sensory information to be encoded enabling the robot to maneuver over irregular terrain. The resulting system has shown to be a simple and efficient method for automated controller design.