Developing Central Pattern Generator based periodic motions using tactile interaction

Abstract

Controlling robots by Central Pattern Generators (CPGs) is a widespread bio-inspired technique for the realization of robot motions. The numerous parameters of the CPGs, often specialized for a specific task, are usually set by automatic techniques like genetic algorithms or policy gradient. However, using these approaches leaves the users with little control on the resulting motion, which can be modified only by changing the evaluation function. Conversely manually setting each parameter gives the user full control over the motion, but identifying which parameters should be altered to obtain a desired effect is not intuitive and therefore developing motion requires time and effort. We present a system that allows programming the CPG parameters by interacting with the robot and in particular by intuitively touching the robot, giving the user full control on the resulting motion without requiring a direct modification of the parameter values