Flexible job shop scheduling optimization is an important research topic in digital manufacturing science, especially the random occurrence of abnormal disturbances such as machine failures and order changes, which disrupt the original production plan, causing problems such as unreasonable resource allocation, delayed order delivery, and increased production costs. In recent years, distributed multi-agent scheduling methods have been considered to be one of the most effective ways to improve the response speed of manufacturing system disturbances and reduce the negative impact of uncertain disturbances in the production process. In the context of job shop scheduling, designing an embodied scheduling agent that integrates the dynamic behavior of physical entities enables real-time environmental perception and autonomous decision-making based on behavioral feedback during disturbances. This ensures the efficient operation of the production system.

To develop a multi-agent scheduling method for flexible job shops based on embodied agents and to enhance the stability and responsiveness of the system during disruptive events, a metamodel-driven approach for constructing embodied agents in flexible job shops was proposed. By instantiating this model, a scheduling system with a unified structure of embodied agents was achieved. First, to enable agents to perform autonomous decision-making and real-time dynamic adjustments, the concept of embodied intelligence was applied. Based on the existing interaction layer, decision layer, and adaptation layer of job shop scheduling agents, their instruction sets, behavior spaces, and sensory signal sets were associated and encapsulated to form embodied agents with physical bodies and behavioral spaces. Based on the resource composition characteristics of flexible job shops, the elements, relationships, and attributes of embodied agents were analyzed and abstracted. A metamodel for embodied agents in flexible job shops was proposed, enabling the unified modeling of embodied scheduling agents and providing a foundational model for their collaborative scheduling. Second, through instantiation operations such as inheritance, composition, aggregation, dependency, and association applied to the metamodel, a distributed multi-agent scheduling system with a unified structure and self-organizing collaborative operation capability was developed. Finally, a set of distributed multi-agent scheduling strategies was designed based on the different functions of different agents and the different information they can obtain. By integrating these strategies with a Q-game negotiation mechanism, collaborative scheduling among multiple agents was realized, thereby improving the stability of the scheduling method and enhancing its responsiveness to disruptions. This scheduling system, based on embodied agents, enabled the adjustment of scheduling strategies at the individual level when disruptive events occur. This approach effectively reduced the number of information exchanges during the scheduling process, improving the stability of the multi-agent system and enhancing its scheduling optimization capabilities in the face of disruptions.

To validate the advantages of the proposed embodied scheduling agent modeling method and the multi-agent scheduling system, two small-scale manufacturing workshops producing structural components were used as case studies. The proposed method was compared with existing approaches in three aspects: embodied agent modeling, collaborative operation, and scheduling optimization. Experimental results demonstrate that the embodied agent modeling method proposed in this paper ensures model structure consistency, guaranteeing that the model adheres to predefined specifications and rules, thereby providing a unified modeling foundation for the collaborative scheduling of multi-agents. In the embodied multi-agent scheduling system, each agent generates a complete set of feasible individual strategies after evaluating all possible actions. Negotiation and interaction among agents are conducted based on these strategy sets. The number of interactions remains independent of the number of actions selected by the agents, resulting in an average reduction of 60.4% in communication volume and a 32.78% average decrease in computational response time. In terms of scheduling optimization performance, agents enhance the diversity of scheduling strategies during the negotiation process by adjusting their individual scheduling strategies, thereby improving the system’s global optimization capability. Compared to existing methods, the proposed approach achieves an improvement of more than 22.6%.

In summary, the proposed multi-agent scheduling method based on embodied agents not only improves system responsiveness and stability in handling disruptive events but also significantly enhances global optimization performance. This provides an efficient and robust solution for scheduling optimization in flexible job shops and offers new research directions and technical support for intelligent scheduling and collaboration in future manufacturing systems. In the future, further research could focus on applying the proposed method to larger and more complex manufacturing scenarios, integrating real-time sensing technologies and advanced machine learning algorithms to further enhance the adaptability and intelligence of the scheduling system.