Real-Time Interactive Reinforcement Learning for Robots

Abstract

It is our goal to understand the role real-time human in-teraction can play in machine learning algorithms for robots. In this paper we present Interactive Reinforce-ment Learning (IRL) as a plausible approach for train-ing human-centric assistive robots by natural interac-tion. We describe an experimental platform to study IRL, pose questions arising from IRL, and discuss ini-tial observations obtained during the development of our system.