Multi-Agent System Production Scheduling 2026
Multi-Agent System Production Scheduling Optimization
Autonomous software agents are replacing centralized solvers for NP-hard scheduling problems across job shops, cloud manufacturing, and distributed supply chains. This dataset spans roughly 40 sources from 2005 to 2025 covering MARL, hybrid metaheuristics, and digital twin integration.
From Centralized Solvers to Distributed Agent Architectures
Multi-agent system production scheduling assigns autonomous software agents—representing machines, jobs, workers, and transport vehicles—to negotiate, bid, and cooperatively solve scheduling problems that are computationally NP-hard in centralized formulations. The technology addresses canonical problem types including Job Shop Scheduling, Flexible Job Shop Scheduling, parallel machine scheduling, and distributed multi-factory scheduling under both static and dynamic conditions.
Core mechanisms include Contract Net Protocol and auction-based negotiation among job and machine agents, multi-agent reinforcement learning for dynamic dispatching, genetic algorithms and Tabu Search embedded within agent decision logic, ontology-based knowledge sharing for interoperability across heterogeneous agent populations, and digital twin synchronization coupling agent decisions with real-time shop floor data.
The innovation timeline reveals three phases: early foundations from 2005 to 2014 established agent-based manufacturing management and hybrid GA/Tabu Search MAS scheduling; a mid-stage convergence from 2015 to 2020 integrated MAS with Industry 4.0 infrastructure including ontology-based coordination, cloud manufacturing, and energy-aware scheduling; and recent filings from 2021 to 2025 signal convergence on learned, self-adaptive MAS architectures within cyber-physical production systems.
In this dataset, innovation is broadly distributed across roughly 10 named assignees spanning academic institutions and commercial software vendors. No single commercial entity dominates the patent landscape in this dataset; academic inventors including Zhejiang University, Ben-Gurion University, and Pusan National University account for a substantial share of protected IP alongside commercial players such as SAP SE, Blue Yonder, and AVEVA Software.
Assignee Filing Patterns and Temporal Trends in This Dataset
Among the 14 patent records with jurisdiction and assignee metadata in this dataset, filings are concentrated in the US jurisdiction and distributed across academic and commercial entities. The temporal distribution shows accelerating activity from 2018 onward, with 2025 representing the most recent filing year.
Top Assignees by Patent Filing Count — Retrieved Records (Dataset Snapshot)
In this dataset, Zhejiang University, B.G. Negev Technologies, PEREIRA ANA MARIA DIAS MEDUREIRA, Blue Yonder Group, and Tata Consultancy Services each hold 2 patent records, while AVEVA, SAP SE, Shuoyunke, Pusan National University, and SR University each hold 1.
↗ Click bars to explorePatent Filings by Jurisdiction — Retrieved Records (Dataset Snapshot)
In this dataset, the US jurisdiction accounts for 9 of 14 patent records, followed by India with 3, China with 2, and EP and Germany with 1 each, reflecting strong academic and commercial US filing activity.
↗ Click bars to exploreKey Application Domains for MAS Production Scheduling
Multi-agent scheduling systems are deployed across discrete manufacturing, cloud platforms, distributed supply chains, intelligent logistics, and satellite operations. Each domain exploits different agent coordination mechanisms and presents distinct real-time disruption profiles.
Discrete Job Shop Manufacturing
The dominant application domain in this dataset. MAS-based scheduling handles variable job routings, machine breakdowns, and rush orders in real time. A 2023 MARL study benchmarks agent-based dispatching against FIFO, SPT, and EDD rules across five job-arrival scenarios, targeting minimization of tardiness and flow time. A decentralized CNP-based approach targets production performance maximization in open job-shop configurations.
Dynamic DispatchingCloud Manufacturing Platforms
MAS scheduling is applied to on-demand manufacturing service platforms where tasks arrive dynamically and must be matched to distributed cloud resources. A 2018 study uses an extended Contract Net Protocol for many-to-many negotiation in cloud manufacturing with dynamic task arrivals. Agents representing service providers and manufacturing tasks coordinate without centralized control, enabling scalable resource allocation.
Cloud SchedulingDistributed Multi-Factory Supply Chains
MAS enables collaborative scheduling across geographically distributed factories and supply chain partners. A 2020 study introduces a multi-agent mechanism for adapting production plans to supply chain disruptions. Blue Yonder Group holds two US patents (filed 2015 and 2018) on solving supply chain campaign planning problems involving major and minor setups, representing an industrially deployed commercial MAS scheduling position.
Supply Chain MASIntelligent Logistics and AGV Integration
Several works target integration of material handling—autonomous guided vehicles and unmanned aerial vehicles—with production scheduling within a unified MAS framework. A 2021 study applies multi-agent deep reinforcement learning to the joint Flexible Job Shop Scheduling and AGV coordination problem, treating vehicle assignment and job scheduling as cooperative sub-tasks. A separate 2021 study addresses scheduling UAV and AGV operations in indoor manufacturing environments.
Logistics MASKey Patent Assignees in MAS Production Scheduling (Retrieved Records)
In this dataset, 10 named assignees hold patent records spanning academic institutions and commercial software vendors. No single entity accounts for a majority of filings in retrieved records; academic institutions including Zhejiang University and B.G. Negev Technologies at Ben-Gurion University each hold 2 US-active patents representing distinct RL-based scheduling architectures.
Top Assignees by Filing Count in Retrieved Records (Dataset Snapshot)
↗ Click bars to exploreZhejiang University
Zhejiang University holds 2 US-active patents in this dataset, filed in 2022 and 2025, both titled “Adaptive-learning intelligent scheduling unified computing frame and system for industrial personalized customized production.” The 2025 filing extends the architecture to automatic algorithm selection based on classified dynamic events, distinguishing global static planning from event-triggered rescheduling without manual intervention. Both patents are active in the US jurisdiction and cover deep reinforcement learning and neural network-based scheduling for mass customization scenarios.
United StatesB.G. Negev Technologies — Ben-Gurion University
B.G. Negev Technologies & Applications Ltd. at Ben-Gurion University holds 2 US patents in this dataset, filed in 2021 and 2023, both covering a “Multi-objective scheduling system and method” based on a master-slave Deep Reinforcement Learning architecture with hierarchical agents. The 2021 filing is listed as US active, covering multi-objective optimization through cooperative hierarchical agent structures. This represents the current IP frontier for DRL-based multi-objective scheduling within the retrieved records.
United StatesFrontier Technologies in MAS Production Scheduling (2022–2025)
The most recent filings and publications in this dataset point to four converging frontiers: human-machine game-theoretic scheduling with digital twins, adaptive deep RL with automatic algorithm selection, Asset Administration Shell standardization for agent interfaces, and ensemble metaheuristic schedulers.
Human-Machine Game-Theoretic Scheduling with Digital Twins
Two CN patents filed by Shuoyunke (Tianjin) Technology Co., Ltd. in October and November 2025 introduce a hierarchical distributed multi-agent platform integrating six specialized agents: a Disturbance Monitoring Agent, Digital Twin Agent, Human Expert Agent, Machine Optimization Agent, Game Coordination Agent, and Execution Agent. This human-in-the-loop, game-equilibrium approach balances production stability and time efficiency, representing a frontier not yet well documented in the broader academic literature. Both patents are active in the CN jurisdiction.
Adaptive Deep RL with Automatic Algorithm Selection
Zhejiang University’s 2025 US active patent extends earlier work to an adaptive-learning unified computing frame that automatically selects optimization algorithms based on classified dynamic events, without manual intervention. The architecture distinguishes between global static planning and event-triggered rescheduling, with neural network-based algorithm evaluation selecting the best solver for each event class. This represents the current IP frontier for self-adaptive MARL scheduling in the retrieved records.
Auction-Based MAS vs. MARL-Based MAS: Key Dimensions
Click any row to explore further.
| Dimension | Auction / Negotiation-Based MAS | Multi-Agent Reinforcement Learning (MARL) |
|---|---|---|
| Maturity | Most established paradigm in this dataset; active since at least 2005 | Fastest-growing cluster; most active 2021–2025 in this dataset |
| Core Mechanism | Agents submit bids or engage in multi-round negotiations to allocate operations using economic models | Agents learn dispatching policies from simulation-based experience; adapt to real-time events |
| Disruption Response | Robust to shop floor disruptions without global re-optimization via local bidding | Adapts to machine failures, urgent orders, and variable job arrivals without re-optimization from scratch |
| Representative Work (Dataset) | Iterative combinatorial auction with flexible bidding strategies demonstrating low price of anarchy vs. centralized benchmarks (2021) | MARL targeting tardiness and flow time minimization, benchmarked against FIFO, SPT, and EDD rules across five scenarios (2023) |
| Key Patent Holder (Dataset) | PEREIRA, ANA MARIA DIAS MEDUREIRA — hybrid GA/Tabu Search MAS, US 2011 and 2013 | Zhejiang University — adaptive deep RL scheduling, US 2022 and 2025 (active) |
| Application Fit | Matrix production systems, mass customization, cloud manufacturing many-to-many negotiation | Dynamic job shop, AGV+FJSSP joint scheduling, multi-objective hierarchical scheduling |
| Industry 4.0 Integration | Extended CNP used for cloud manufacturing (2018); ontology-based agent coordination (2019) | Digital twin synchronization via EPHM (2020); AAS-based scheduling agent deployment (2023) |
Frequently Asked Questions: MAS Production Scheduling
Based on the retrieved dataset, MAS technology addresses Job Shop Scheduling, Flexible Job Shop Scheduling, parallel machine scheduling, flow shop scheduling, and distributed multi-factory scheduling under both static and dynamic conditions including machine breakdowns, urgent order insertions, and supply chain disruptions.
In this dataset, Zhejiang University, B.G. Negev Technologies & Applications Ltd. at Ben-Gurion University, PEREIRA ANA MARIA DIAS MEDUREIRA, Blue Yonder Group Inc., and Tata Consultancy Services Limited each hold 2 patent records. AVEVA Software LLC, SAP SE, Shuoyunke (Tianjin) Technology Co. Ltd., Pusan National University, and SR University each hold 1 record in this dataset.
According to the retrieved dataset, Contract Net Protocol (CNP) is an auction-based negotiation mechanism where agents representing machines or jobs submit bids or engage in multi-round negotiations to allocate operations, using economic models to drive local decisions toward globally coherent schedules. It was applied to cloud manufacturing using many-to-many negotiation in a 2018 study.
The retrieved dataset describes digital twin synchronization as a mechanism to couple agent decisions with real-time shop floor data. A 2020 simheuristics framework combines GA optimization with discrete-event simulation synchronized via an Equipment Prognostics and Health Management digital twin. The 2025 Shuoyunke patents incorporate a dedicated Digital Twin Agent within their hierarchical multi-agent platform.
Based on a 2023 paper in this dataset, the Asset Administration Shell (AAS) is an Industry 4.0 standard information model that can be used to model and deploy production scheduling agents, enabling interoperability across heterogeneous factory assets. The paper proposes AAS-based information models establishing a standardization path for plug-and-play scheduling agents.
The most recent signals in this dataset (2022–2025) point to: human-machine game-theoretic scheduling with digital twins (Shuoyunke, CN 2025); adaptive deep RL with automatic algorithm selection (Zhejiang University, US 2025 active); AAS as a universal agent interface for Industry 4.0 interoperability (2023 literature); modular resilient production MAS (2022 literature); and hybrid GA-PSO-GWO ensemble metaheuristic schedulers (SR University, IN 2025 pending).
Data and insights on this page are based on a limited patent and literature dataset and are for reference only. Figures may not represent the complete technology landscape.