Socially Guided Intrinsic Motivation for Robot Learning of Motor Skills

  • 2018-04-19 16:47:04
  • Sao Mai Nguyen, Pierre-Yves Oudeyer
  • 1

Abstract

This paper presents a technical approach to robot learning of motor skillswhich combines active intrinsically motivated learning with imitation learning.Our architecture, called SGIM-D, allows efficient learning of high-dimensionalcontinuous sensorimotor inverse models in robots, and in particular learnsdistributions of parameterised motor policies that solve a correspondingdistribution of parameterised goals/tasks. This is made possible by thetechnical integration of imitation learning techniques within an algorithm forlearning inverse models that relies on active goal babbling. After reviewingsocial learning and intrinsic motivation approaches to action learning, wedescribe the general framework of our algorithm, before detailing itsarchitecture. In an experiment where a robot arm has to learn to use a flexiblefishing line , we illustrate that SGIM-D efficiently combines the advantages ofsocial learning and intrinsic motivation and benefits from human demonstrationproperties to learn how to produce varied outcomes in the environment, whiledeveloping more precise control policies in large spaces.

 

Quick Read (beta)

loading the full paper ...