Learning robot-environment interaction using echo state networks

Abstract

Learning robot-environment interaction with echo state networks (ESNs) is presented in this paper. ESNs are asked to bootstrap a robot's control policy from human teacher's demonstrations on the robot learner, and to generalize beyond the demonstration dataset. Benefits and problems involved in some navigation tasks are discussed, supported by real-world experiments with a small mobile robot. © 2010 Springer-Verlag.