Respecializing swarms by forgetting reinforced thresholds

Vera A. Kazakova, Annie S. Wu, Gita Reese Sukthankar

Abstract: Response threshold reinforcement is a powerful model for decentralized task allocation and specialization in multiagent swarms. In dynamic environments, initial task assignments and specializations must be updated over time to meet changing system needs. The very nature of threshold reinforcement-based behavior can, however, hinder respecialization, limiting its usability in real-world applications. We propose a decentralized forgetting-based extension to response threshold reinforcement and show that it can improve the efficiency and stability of the resulting task assignments under changing system demands.