Exploring the possibilities of Multi-Agent Reinforcement Learning to solve coordinated cooperative tasks in Flexible manufacturing systems

Abstract

Advances in artificial intelligence and Multi-Agent Systems enable coordinated agents to achieve multiple, often conflicting, objectives—making them ideal for ”flexible factories.” These factories, driven by technologies merging physical, digital, and biological domains, are evolving into ”smart factories.” Modeling production processes as multi-agent systems allows simultaneous optimization of efficiency, waste reduction, sustainability (economic, social, and environmental), cost savings, and downtime reduction. However, the flexibility needed in reconfigurable environments increases the complexity of decentralized control. Small and medium-sized enterprises (SMEs) are a key example, as they often produce small batches or customized goods, requiring constant adaptation. Multi-agent reinforcement learning provides a viable solution, avoiding impractical centralized control in dynamic settings. This work explores multi-agent reinforcement learning for collaborative manufacturing tasks, such as material handling (a nonvalue-adding operation where efficiency is critical). A preliminary case study is presented, using virtual environments to train multiple agents in coordinated material manipulation across varying complexity scenarios.