A Framework for Learning Declarative Structure

This paper provides a framework with which a humanoid robot can efficiently learn complex behavior. In this framework, a robot is rewarded by learning how to generate novel sensorimotor feedback—a form of native motivation. This intrinsic drive biases the robot to learn increasingly complex knowledge about itself and its effect on the environment. The framework includes a mechanism for uncovering hidden state in a well-structured state and action space. We present an example wherein the robot, Dexter, learns a sequence of manual skills: (1) searching for and grasping an object, (2) the length of its arms, and (3) how to portray its intentions to human teachers in order to induce them to help.