Fuzzy Reward-Based Cooperative Reinforcement Learning for Bio-Insect and Artificial Robot Interaction

In this paper, we address our on-going research that is for interaction between artificial robots and a bio-insect. The research motivation and research goal were introduced in [1]. In order to report a progress of this project, this paper contains advanced framework and simulation results. When we did experiments using real bio-insects, their movement showed a little randomness. For this reason, fuzzy logic is employed to drive the model-free bio-insect towards a desired point. The framework formulated in this paper is based on fuzzy reward system and fuzzy expertise measurement system. Fuzzy reward system uses three inputs and an output resulting in numerical value within −1 to 1. Fuzzy expertise measurement system is inspired by area of expertise. In area of expertise method, it uses expertise measurement equation for finding expert agent. Based on area of expertise method, our method uses three expertise measurements to calculate score of individual agent. Based on this score, agents can share their intelligences with weighted scores. Simulation results demonstrate the validity of the framework established in this research.Copyright © 2009 by ASME