Using AI and simulation to bridge the gap between planning and field reality, SNCF Réseau and Sopra Steria are developing an industrial metaverse that enhances productivity, safety and strategic planning across more than 27,000 kilometres of track.
Across France’s vast rail infrastructure, time is a constant adversary. Trains run day and night, placing intense pressure on maintenance and regeneration schedules. For François Pontvianne, Signalisation Tools Manager at SNCF Réseau, improving productivity and reducing project costs meant adopting a more industrial and digitally integrated approach.
“At SNCF Réseau, our role is to maintain and regenerate the national railway infrastructure, which is constantly used day and night,” Pontvianne explains. “To meet rising demands, we must improve productivity and reduce project costs, which means shifting to a more industrial and digitalised approach. We launched an experiment using the metaverse to support our regeneration work, especially in signalisation, where we often face discrepancies between schematics and field conditions. These inconsistencies impact the efficiency of projects, so we needed a way to use trusted data and make planning more responsive and cost-effective.”
What began as an internal awareness campaign quickly evolved. A demonstrator validated early interest and technical feasibility but also revealed challenges. The reliance on powerful desktop computers made scalability impractical. A web-based version followed, offering seamless integration with SNCF’s information systems and remote accessibility for real projects. This pilot phase examined whether the platform could genuinely accelerate workflows and reduce time-to-insight.
From schematics to simulation
The pilot’s outcomes were significant. Engineers could compare AutoCAD schematics with 3D field data, identifying signal misalignments and planning corrections early. Thanks to precise spatial measurement tools, calculating braking distances became simpler. Simulating the driver’s view of shelters and signal masts enabled early detection of clearance violations. A train envelope model helped determine whether new structures interfered with safety profiles.
This approach does more than visualise potential risks; it creates a shared environment where stakeholders can interrogate and resolve issues collaboratively. For instance, maintenance teams can test the feasibility of future installations in parallel with design reviews, ensuring smoother transitions from concept to construction. “We are satisfied with the progress and want to expand the platform to other departments,” Pontvianne adds.
“We are taking a step-by-step approach, but the potential is clear. This is a powerful tool for day-to-day job efficiency and can deliver value across the organisation.
Industrialising a digital platform
Translating these use cases into scalable software required more than digital enthusiasm. It demanded robust engineering. Yoann Yvon, Head of Digital Interaction at Sopra Steria, faced the technical reality of working with high-resolution point clouds, often exceeding 10GB per kilometre of track. Integrating these with asset libraries, CAD files, and business metadata called for a unified, scalable foundation.
“We did not want to build a one-off proof of concept,” Yvon explains. “Our vision was to create a scalable, industrial-grade platform that could become part of SNCF Réseau’s core information system. That meant designing for long-term software engineering practices, with multiple development squads working across locations in France and Spain. We needed a development model that supported DevSecOps principles and could scale.”
The team turned to NVIDIA Omniverse. Its Universal Scene Description (USD) format merged diverse 3D sources into a consistent environment. Version control and authentication ensured users accessed a single, verified model of reality. Omniverse Kit, built on Python, allowed 3D and AI developers to collaborate within the same ecosystem, enabling faster, more integrated iterations.
Yvon describes a modular architecture: a data ingestion pipeline converts LIDAR and scans data into USD, adding geolocation, model alignment, and asset normalisation. A shared repository supports both development and runtime environments. GPU workloads are handled elastically, using dynamic allocation across on-premises servers or Azure cloud instances. This maintains performance while controlling infrastructure costs.
Embedding realism and training intelligence
Planning requires more than data fidelity. It demands realistic simulation. With Omniverse Physics and Isaac Sim, teams simulate daylight, fog, snow, and physical interactions to validate visual and functional integrity. Under different conditions, visibility tests at level crossings or urban junctions become more accurate. These simulations now form part of the safety planning process.
The metaverse also serves as a synthetic data engine. Instead of relying on expensive, manually annotated real-world data, teams generate thousands of virtual LIDAR scans with ground-truth labels. These datasets accelerate the training of AI models for object detection and classification, with consistency and control that field data often lacks.
This represents a fundamental shift in how infrastructure operators can approach AI. Rather than being constrained by the availability and accuracy of legacy data, organisations can now create tailor-made environments to test, train and deploy AI applications at scale. That capability becomes especially powerful when extended beyond signalisation to use cases such as autonomous inspection, defect detection, and asset lifecycle management.
“This project is not just about building a beautiful 3D world,” Yvon continues. “It is about delivering operational efficiency, reducing time to market, and integrating new use cases over time. By relying on open standards like USD, we are building long-term assets that can be extended and repurposed. That is key to making the industrial metaverse a strategic part of an organisation’s infrastructure.”
The architecture of collaboration
David Maurange, Consulting Director at Sopra Steria Next, sees the value of the metaverse not just in simulation or visualisation but in its ability to bring teams together around shared data. “The industrial metaverse is a powerful concept because it allows diverse teams to work together in a unified environment, using the same data,” he says. “That is essential for collaboration across departments and disciplines. Rather than developing bespoke tools for each use case, we can build one environment that adapts to many needs, whether it is signalisation, maintenance, safety, or long-term planning.”
SNCF Réseau now navigates its assets via a Google Earth-style interface, merging spatial and operational context. Users can drag and place virtual components, immediately receiving feedback based on business rules. Asset placement becomes a dialogue between engineering intent and system intelligence.
Training datasets, once a bottleneck, are now abundant. Thanks to synthetic scans of real-world infrastructure, AI models can be developed rapidly. Signal masts, shelters, and safety beacons are dropped into simulated landscapes, captured from every angle, and fed into neural networks.
A replicable roadmap for the industry
The project remains grounded in pragmatism. Use cases are validated iteratively, and departments are onboarded incrementally. The emphasis is on open standards, reuse, and integration with existing tools. What distinguishes this work is not the novelty of digital twins but the maturity of its execution.
“It is important to approach projects like this incrementally,” Maurange concludes. “Start with a strategic vision, then validate use cases step by step. Involve users early, adapt the platform to their feedback, and build on open standards so that you can capitalise on your investment in the long term. That is what we have done with SNCF Réseau, and we believe it offers a roadmap for any organisation looking to develop an industrial metaverse.”
The lessons are clear for organisations managing complex physical infrastructure. Trusted data, unified platforms, and simulation-driven AI are not futuristic concepts. They are active enablers of efficiency and risk reduction. The French railway network may be one of the oldest in the world, but in the hands of a metaverse, it is finding new relevance in the digital age.
