Russia’s “Digital Swarm” AI Is Reshaping Industrial Production Management

Researchers at Peter the Great St. Petersburg Polytechnic University, or SPbPU, have developed a set of algorithms built on a multi-agent architecture that significantly improves the efficiency of decentralized production optimization compared with conventional methods.

At the core of the system is a distributed architecture in which multiple independent agents – including people, machines, robots, software modules, logistics hubs and even individual production sites – interact with one another and with their environment to keep the broader system operating efficiently. Optimization is achieved through staged communication among intelligent agents that continuously refine production-management scenarios based on individual operational plans and incoming data.

Partnership Between Science and Industry Lowers the Barrier to AI Adoption

Testing showed that the multi-agent algorithms can find solutions with the same level of optimality 10 times faster than classical methods. With longer simulation times, the system can even outperform the target technological parameters of the production environment. Researchers expect the technology to find applications across multiple industries. In the oil and gas sector, for example, it could enable more precise allocation of resources during drilling and geological operations, potentially improving extraction efficiency. A prototype is already being refined using real industrial datasets.

The project is being developed under the federal Priority-2030 program in partnership with Gazprom Neft, Idealstroy and Gazpromneft NTC. Collaboration with industrial companies is expected to produce a suite of software tools with industry-specific libraries designed for rapid integration into existing decision-support systems and for improving production-management quality.

From Standalone Algorithms to an Industrial Ecosystem

The algorithm suite has been integrated into the POLANIS digital platform for multimodal data analysis, which SPbPU researchers are developing as part of one of the university’s three strategic science and technology initiatives: “Artificial Intelligence for Cross-Industry Applications.”

The platform is designed for data management and computer modeling of industrial processes using neural network technologies. The software can process information in virtually any format, configure calculation parameters and run machine learning algorithms. POLANIS is designed as a universal platform that can be deployed across manufacturing, logistics, energy and other sectors to improve operational efficiency and accelerate digital transformation.

The architectural flexibility and open ecosystem behind POLANIS create a foundation for scaling the multi-agent algorithm suite beyond isolated manufacturing tasks. That approach opens the door to cross-industry deployment and allows enterprises to implement adaptive resource management without fundamentally rebuilding their existing IT infrastructure.

Global Demand for Russian Industrial Algorithms

Russia is actively building a scientific and technological base for shifting from isolated AI applications toward scalable multi-agent platforms for industrial optimization.

One example comes from the Genesis Znaniy Group, a Skolkovo resident company whose solutions, developed in partnership with leading Russian scientists, are already in demand both domestically and internationally. For RSC Energia, the company developed multi-agent systems for managing ISS cargo flows, GOUG crew shifts, emergency-response scenarios and other operational applications. For Russian Railways, it created systems for strategic train-traffic management and real-time dispatch coordination. For the Ramenskoye Instrument Engineering Design Bureau, the company developed adaptive multi-agent flight-mission systems for fighter aircraft. For RCC Progress, it created a multi-agent system for managing constellations of spacecraft, effectively a “satellite swarm.”

Genesis Znaniy Group’s foreign clients include EADS in Germany, for which the company developed a multi-agent system for airliner manufacturing management. For Italy’s Iaccubucci / Modular Galley Systems AG, the company built workshop-management systems for aircraft equipment production. Airbus in France also adopted a multi-agent system for aircraft manufacturing management. Other customers include the British companies Multi-Agent Technology and Barloworld, as well as Lego in the United States and Coca-Cola.

Import Substitution as a Launchpad for Exports

Russian technologies already compete successfully on the global market, offering adaptive solutions for decentralized industrial ecosystems. The SPbPU project extends that trajectory naturally. Systems of this kind can provide a reliable foundation for accelerating import substitution and strengthening the technological resilience of Russian industry, while also laying the groundwork for future expansion of Russian high-value software exports.

Each management agent inside the simulated system has its own level of knowledge. Some agents have a better understanding of the current condition of wells and infrastructure, while regional agents see the broader picture, although in less detail. Based on their own data, each agent proposes operational plans and resource-exchange options. The agents then repeatedly coordinate decisions among themselves, gradually improving the outcome. As a result, both local and overall production volumes in the region increase. That is the defining feature of our system: instead of relying on one centralized decision, we use agents that independently propose scenarios and negotiate with one another. This produces more resilient and realistic plans capable of solving management tasks at every level

Alexey Gintsyak

PhD in Engineering, Head of the Digital Modeling of Industrial Systems Laboratory at SPbPU Advanced Engineering School