Lifelong evolution for adaptive robots

Abstract

Researchers in mobile robotics have for many years been investigating the use of evolutionary techniques for the development of control methods. Most evolutionary robotics work has concentrated on the use of evolution exclusively during the training phase of the robot controller. Few investigators continue evolution throughout the lifetime of a robot. This paper investigates a system that uses a genetic algorithm to train a robot for a generalised test environment, then an evolutionary strategy to investigate the effect of continuing the evolution as the robot progresses with its task. It is concluded that lifelong evolution has an important role to play in producing truly adaptive robots.