Imagine a factory where every machine, robot, and quality sensor has its own little brain — but none of them talk to each other. That's the problem most manufacturers face when they try to switch from mass production to custom orders. MAS4AI built a system of AI agents that act like a team of coordinators, each one managing a different part of production and sharing information so the whole line adapts on the fly. They tested it in real factories making cars, bicycles, bearings, and wood products.
Most factories today run automation systems that operate in silos — each machine or robot has its own controller with no way to coordinate with the others. When a customer wants a custom product or the production plan changes, someone has to manually reconfigure each system. This makes mass customization expensive, slow, and error-prone, putting European manufacturers at a disadvantage against competitors who can adapt faster.
The team built a multi-agent AI system where specialized software agents handle planning, knowledge sharing, and machine learning across the production floor. Concrete deliverables include a hierarchical planning agent, an agent knowledge access system with semantic logic, and integrated demonstrators tested in automotive, contract manufacturing, bicycle, bearings, and wood processing industries with documented final results.
If you are an automotive manufacturer dealing with growing demand for vehicle customization while your automation systems can't talk to each other — this project developed AI planning agents that coordinate across production modules so your line reconfigures itself for each order. Tested in an actual automotive demonstrator with 18 consortium partners.
If you are a contract manufacturer juggling different product specs from different clients on the same equipment — this project built a multi-agent system where AI agents handle scheduling, quality, and machine coordination across 4 hierarchy layers. The system was demonstrated across 5 industrial sectors including contract manufacturing.
If you are a wood processing company struggling to automate custom cuts and finishes because your machines each run their own proprietary software — this project created an agent knowledge access system using semantic web technologies that lets different machines share production knowledge. Demonstrated with final results in the wood processing industry.
The project does not publish licensing costs or implementation pricing. Since the coordinator is DFKI (a German AI research center) and 6 SME partners were involved in development, commercial terms would need to be negotiated directly with the consortium. Implementation cost will depend on your existing automation infrastructure.
The project ran demonstrators across 5 real industrial sectors — automotive, contract manufacturing, bicycle industry, bearings production, and wood processing — and published final demonstrator results. With 11 industry partners in the consortium, the system was designed for production environments, not just labs.
This was a publicly funded RIA project with 18 partners across 7 countries. IP is typically shared among consortium members according to their grant agreement. Contact DFKI (the coordinator) or specific industrial partners to discuss licensing options for individual components.
A core goal of MAS4AI was solving the interoperability problem — the fact that different automation systems can't communicate with each other. The agent knowledge access system uses semantic web technologies specifically to bridge different proprietary systems. Integration effort will depend on your current setup.
The project ran for 3 years (2020-2023) to develop and test the full system. Based on available project data, the system includes planning agents, knowledge representation, and machine learning components that were integrated in phases. A focused deployment of specific modules would likely take less time than the full research cycle.
The project produced 6 demonstrator deliverables with both initial and final results across real industrial settings. Prototypes exist for the planning agent, knowledge access system, and hierarchical planning integration. This is past the research stage but would likely need engineering work to adapt to your specific production setup.
This is a large, industry-heavy consortium with 18 partners from 7 countries, and a 61% industry ratio — meaning the majority of partners are companies, not universities. Of those, 6 are SMEs, suggesting both large manufacturers and smaller tech providers were at the table. The coordinator is DFKI, Germany's leading AI research center, which brings credibility and deep technical capability. With 11 industry partners and demonstrators running in automotive, contract manufacturing, bicycle, bearings, and wood processing, this project was clearly built to solve real factory problems rather than produce academic papers. For a business looking to adopt this technology, the breadth of industrial partners means there are likely multiple entry points for collaboration or licensing.
DFKI (German Research Center for Artificial Intelligence) in Germany — a well-known AI institute with a technology transfer office
Want an introduction to the MAS4AI team? SciTransfer can connect you with the right consortium partner for your production setup.
