The cooperative reinforcement learning in a multi-agent design system

Abstract

This paper presents a multi-agent cooperative reinforcement learning approach in cooperative design system. For effectively speed up the learning process, this approach adopts dynamic niche technology grouping design agents, and selects the optimal design agent in every groups. The selected agents make reinforcement learning via interaction with designers and carry on cooperative learning each other, and then spread the learned knowledge in respective groups. The radius of the niches and selected design agents are dynamically adjusted during cooperative reinforcement learning process.