Performance Comparison of Simple Reflex Agents Using Stigmergy with Model-Based Agents in Self-Organizing Transportation

Abstract

Multi-agent systems utilizing simple reflex agents are assumed to have a significant competitive disadvantage when compared to more sophisticated agent-based approaches. However, in terms of resilience and adaptivity, this simple design turns out be an advantage when used together with stigmergy. In this paper we show that simple reflex agents that use stigmergy, are fit and flexible enough to outperform rivaling model-based agents in a disturbed transportation setting that simulates a dynamic, real-world industrial shop floor, and have a performance closer to a centralized, monolithic approach which we compare to as gold standard. This leads to opportunities for simpler, but nevertheless more robust agent design for self-organizing, decentralized multiagent approaches just by sharing knowledge of the world and exploiting their environment.