A Step Towards Incremental On-Board Evolutionary Robotics

Abstract

We apply evolutionary algorithm (EA) to the design of controller for adaptive robots. EAs can be successful for more complicated tasks, where traditional engineering methods struggle to provide a good solution. However a simple evolutionary process is computationally expensive and works only for tasks of limited complexity. Incremental evolution can provide a solution. Complex tasks are divided into easier steps. This article gives an overview of incremental evolution and explains our incremental evolution experiments. We use EA as an on-board learning method to solve the novel task encountered by the robot in its environment. A robot equipped with a simulator of itself evolves a routine to overcome an obstacle in the execution of its task. We demonstrate embedded incremental evolution of a robot program for a task with computationally limited LEGO 1 robots.