Fuzzy reinforcement learning for an evolving virtual servant robot

Abstract

This work presents our research in the application of reinforcement learning algorithms for the generation of autonomous intelligent virtual robots, that can learn and enhance their task performance in assisting humans in housekeeping. For the control system architecture of the virtual agents, two algorithms, based on Watkins' Q(λ) learning and the zeroth-level classifier system (ZLCS), are incorporated with fuzzy inference systems(FlS). Performance of these algorithms is evaluated and compared. A 3D application of a virtual robot whose task is to interact with virtual humans and offer optimal services on everyday in-house needs is designed and implemented. The learning systems are incorporated in the decision-making process of the virtual robot servant to allow itself to understand and evaluate the fuzzy value requirements and enhance its performance.