Social-Learning Coordination of Collaborative Multi-Robot Systems Achieves Resilient Production in a Smart Factory

This paper presents a novel resilient production problem in a multi-robot system (MRS) driven smart factory that suffers diverse degradation of accuracy among heterogeneous robots. The inevitable degradation in production robots’ accuracy undermines desirable productivity and efficiency. Although traditional maintenance and calibration strategies can be employed, they fail to support the continuity, flexibility and agility required in smart factories. Instead of pursuing accurate task execution relative to a consistent global reference, we formulate an innovative MRS coordination strategy, where a collaborative MRS autonomously pursues relative accuracy against accuracy degradation toward resilient operation. This new coordination strategy introduces challenges including adapting to time-dynamic production flows, difficulties in observing production robots’ accuracy, and the absence of a comprehensive architecture of resilient MRS in a smart factory. We propose a computational approach with social learning and AI to overcome these challenges. A cyber-physical MRS model is proposed, in which the physical domain represents the time-dynamic production flows, while the cyber domain forms a partially connected wireless network to automate the collection of data regarding production flows, accuracy degradation, and peer robot measurements. Social learning based on such data is employed to collaboratively estimate accuracy and AI decision-making, thereby enabling adaptive coordinated task execution against accuracy degradation. We further investigated a group decision-based predictive maintenance against point failures caused by accuracy degradation. Computational experiments demonstrate that the proposed approach improves the effective rate and mean-time-to-fail performances against accuracy degradation and point failures in a scalable manner. Note to Practitioners—The practical problem motivating our work revolves around resilient production, which is affected by diverse accuracy degradations among heterogeneous robots. This issue is particularly relevant for implementing smart factories with multi-robot systems (MRS), including production and transportation robots. We propose a cyber-physical multi-robot system (CPMRS) to comprehend the dynamics of MRS-driven smart factories, aiming to facilitate productivity, efficiency, and resilience through MRS coordination, as opposed to accurate production task execution achieved by frequent maintenance and calibration, which sacrifices the flexibility and agility of smart factories. CPMRS allows the autonomous collection and exchange of information about production flows, accuracy degradations, and peer robot measurements with limited wireless communications. CPMRS further enables computational solutions with social learning and reinforcement learning for robots as well as group-decision-based predictive maintenance. It facilitates autonomous coordinated task execution and resilient production. Computational experiments demonstrate a 124 percent improvement in the mean time to failure (MTTF) and a 317.27 percent improvement with predictive maintenance. As in our proposal to general manufacturing, further research may be adapted to specific manufacturing scenarios. This research can benefit broader applications, such as smart logistics, warehouses, and even smart cities, where similar collective actions among multiple robots or agents are required.