Ibots Learn Genuine Team Solutions

Abstract

“Ibots” (Integrating roBOTS) is a computer experiment in group learning. It is designed to understand how to use reinforcement learning to program automatically a team of robots with a shared mission. Moreover, we are interested in deriving genuine team solutions . These are policies whose form strongly depends on the number of robots composing the team, on their individual skills and weaknesses, and on any other mission boundary condition which makes it worth to prefer “at a team level” certain solutions to others. The Ibots learn to accomplish the integration mission by means of a reinforcement signal which measures their performance as a team. This form of payoff leads to genuine team solutions. Benefits and drawbacks of using a single team payoff as opposed to individual robot payoffs are discussed.