Learning in a Free Robot

Abstract

A "free" robot is one which "pursues happiness" or reward rather than obeying commands. A program for controlling such a robot has been written, and is in the process of being debugged and tested. It is designed to learn to maximize reward while interacting with an environment. The basic mechanisms are extremely general with respect to the nature of the environment, motor abilities, and senses, while fully utilizing known sensory characteristics. Behavior can be directed toward deferred as well as immediate reward and expected reward duration is taken into account. Compounding is used to develop learned hierarchical sensory and motor structures.