Distributed Constraint Optimization for Task Allocation in Self-Adaptive Manufacturing Systems

Adaptive manufacturing systems consist of many autonomous agents working together in an ever-changing environment. Therefore, collectively deciding which agent performs what task is a key issue and widely studied. However, many approaches towards this issue assume (partially) centralized control, require implementing proprietary algorithms, or cannot provide any guarantees regarding their runtime or communication overhead. To address these problems, we investigate the use of distributed constraint optimization (DCOP) in this context: We present a DCOP model built on freely available algorithms to distribute the problem among the agents that cooperate to solve it. Furthermore, we compare this decentralized approach to a centralized one by measuring the runtime in a set of system configurations with an increasing number of agents. While the DCOP approach works well in small system configurations, our results indicate poor scalability compared to the central approach when increasing the number of agents. We conclude that, although the DCOP approach has desirable properties, it is unsuitable for larger practical applications with dozens or hundreds of agents.