Manufacturing

BMW Group Develops Custom Application on NVIDIA Omniverse for Factory Planners

Objective

BMW Group is redefining automotive factory planning and optimization in production with digital twins built with NVIDIA Omniverse™ and OpenUSD. Years before production begins, factory planners work in virtual factories to optimize layouts, robotics, and logistic systems for manufacturing. BMW’s virtual factories span over 1 million square meters—roughly 140 football fields—and support planners making design and process changes every day.

Customer

BMW Group

Use Case

Simulation/Modeling/Design

Products

NVIDIA Ominiverse Enterprise

Developers at BMW Group leveraged NVIDIA Omniverse to build digital twins that are transforming factory planning worldwide.
The BMW team used OpenUSD and the Omniverse Kit SDK to develop FactoryExplorer, a digital twin platform that enables factory planners to collaborate in real time, helping users optimize layout and design of complex manufacturing systems, and develop autonomous robot and vision AI applications prior to real world deployment.

Projected 30% savings from optimized factory planning and highly efficient processes
Reduction in change orders and capital investments
Real-time collaboration across various teams and specialties

Increased stability in product launches

Building a New Suite of Tools for Software-Defined Factories

At BMW’s virtual Debrecen plant, the power and agility of planning industrial manufacturing plants can be seen firsthand. Using FactoryExplorer, a platform built with Omniverse technologies, logistics and production planners plan factory layouts, logistics, and processes in a digital twin of the factory. This allows the team to perform tasks such as visualizing and identifying optimal placement for robots in constrained spaces.

A team of factory planners play critical roles in the planning and operations of the company's factories worldwide. Their jobs are highly complex, with each planner making an average of three changes per week across over thirty factories. Even the slightest miscalculations or mistakes can result in significant real-world costs.

Using virtual environments, they can plan and optimize production processes before committing to real-world construction, layout changes, and capital expenditures. This approach significantly reduces costs and production downtime caused by change orders and process flow re-optimizations in existing facilities. The cost of shutting down a factory, or even parts of assembly lines, can be enormous.

Factory planners have diverse specializations and levels of expertise with complex software. Delivering a highly intuitive platform that could be used without extensive instructions was paramount. Leveraging Omniverse technologies, BMW created a customizable user experience for planning teams that allows them to focus on their relevant functions and most crucial tasks inside the virtual factories.

“If we shut down for two weeks, that’s two weeks where we’re not earning any money whatsoever. And it’s two weeks where we cannot get that time back. “The plant produces new vehicles every minute, so every minute not building a car is the price of a car lost.”

Ross Patrick Krambergar
BMW Group, Virtual Production Expert

Collaborating in Real-Time on Virtual Factories

Connecting multiple teams working on different areas and elements of the same factory demanded a centralized platform. To provide their team with a holistic planning solution, BMW developed FactoryExplorer, an internal metaverse application.

Countless aspects of planning go into building factories and their digital twins, including building structure, vehicles, equipment and kinematics, product and process coupling, logistics, and human simulation. Each of these relies on unique data from different source systems, such as Autodesk Revit, Bentley Microstation, ipolog, and ema.

Building their digital twin platform with Universal Scene Description (OpenUSD) lets BMW bring data from these software tools together to unlock new possibilities for planners. OpenUSD is an open and extensible ecosystem for describing, composing, simulating, and collaborating within 3D worlds.

In the digital twin, planners can select and aggregate source files like CAD data containing buildings or layouts into structured OpenUSD projects that then get published to the digital twin with one click. From within the application, planners can adjust the validity of files, update reference files, and ensure projects are up to date. The result is a well-structured, standardized database that factory planners of all technical levels can tap into.

“This is transformative — we can design, build and test completely in a virtual world.”

Milan Nedeljković
Member of the Board of Management, BMW Group

“This is transformative — we can design, build and test completely in a virtual world.”

Milan Nedeljković
Member of the Board of Management, BMW Group

Collaborating in Real-Time on Virtual Factories

As data gets automatically converted into USD, BMW users can then publish it to its digital twin platform and use it alongside data from other sources

In addition to enabling real-time collaboration, OpenUSD enables non-destructive assembly of data from numerous sources as individual layers. Different users can modify the factory on different layers at the same time, and their edits will not harm the work of colleagues. The stronger layer will win out in composition, but the data from the weaker layer remains accessible.

The Factory Builder platform has transformed the way BMW's teams collaborate on projects, resulting in improved productivity, streamlined communication, and increased agility in responding to changing market demands. Instead of meeting in a physical location, teams can now collaborate inside their factory digital twins.

Customizing BMW’s Industrial Metaverse Platform With Unique Extensions

Using the Omniverse Kit SDK, developers at BMW were able to build a digital twin platform and tools for their unique planning needs. To tailor the platform to the specific needs of planning teams, BMW’s developers built and integrated custom extensions from Omniverse libraries. These extensions deliver unique user experiences and streamline workflows across BMW’s factory planning teams.

A Collision Detection extension helps planners identify potential collisions between equipment, building components, and materials.
A Factory Filter extension enables users to select which payloads or factory areas to load.
A Factory Launcher extension allows planners to launch buildings or factories from different set points in time.
The Navigation and Waypoint extension makes it easier for planners to maneuver and interact with the virtual factory environment.
A Measuring extension helps planners assess whether robots and equipment can fit in designated spaces.

BMW plans to continuously develop new extensions and functionality to meet the needs of its planning teams as their virtual and real-world factories become even more advanced.

New Possibilities With Generative and Agentic AI

BMW is committed to expanding its digital twin platform with advanced generative and agentic AI capabilities. By introducing intelligent AI assistants, BMW will make planning tools easier to use and more responsive to team needs. These new features will help automate tasks, suggest solutions, and enable planners to identify new opportunities and resolve issues more quickly. As the platform grows, BMW’s teams will have greater flexibility and efficiency in factory planning, supporting a wider range of use cases across the organization.

Reimagining Quality Control With Synthetic Data

Beyond the transformations they are driving in factory planning, BMW is redefining quality control by putting AI at the heart of its production lines. Instead of relying solely on human inspectors, prone to fatigue and inconsistency, BMW developed an AI-powered visual inspection system called Presence Detection, which automatically evaluates images of critical components in real time and flags any issues for workers to address.

Training these visual inspection models requires large volumes of defect data, which is difficult to collect, especially during early production runs, where flaws are rare and rework is costly. With NVIDIA Omniverse Replicator and Isaac Sim, the BMW team built a synthetic data pipeline that generates thousands of photorealistic training images with the click of a button. BMW can now train robust AI models before production begins by simulating defects and real-world scenarios virtually.

To further scale this effort, BMW taps into SORDI.ai (Synthetic Object Recognition Dataset for Industries)—a central platform that manages image data across use cases, helping data scientists securely access and reuse datasets company-wide. This groundbreaking initiative aims to accelerate AI training in production by offering the largest, most realistic open-source dataset for the industrial environment, comprising over 800,000 photorealistic images spanning 80 categories. BMW leveraged NVIDIA DGX™ systems with NVIDIA Hopper™ architecture to train deep-learning-based synthetic data-generation models used to develop the SORDI dataset. The deep learning models perform tasks including object detection, image segmentation, image classification, and 6D pose estimation. Lastly, the DGX systems are used to evaluate and test the trained models. This fusion of simulation, AI, and NVIDIA Accelerated Computing is transforming inspection from a manual checkpoint into a smart, scalable process that evolves with every production run.

Together, Replicator and SORDI.ai helped the team manage quality control weeks ahead of schedule, all while cutting the costs and complexity of traditional data collection.

Customizing BMW’s Industrial Metaverse Platform With Unique Extensions

A Collision Detection extension helps planners identify potential collisions between equipment, building components, and materials.
A Factory Filter extension enables users to select which payloads or factory areas to load.
A Factory Launcher extension allows planners to launch buildings or factories from different set points in time.
The Navigation and Waypoint extension makes it easier for planners to maneuver and interact with the virtual factory environment.
A Measuring extension helps planners assess whether robots and equipment can fit in designated spaces.

BMW plans to continuously develop new extensions and functionality to meet the needs of its planning teams as their virtual and real-world factories become even more advanced.

New Possibilities With Generative and Agentic AI

Discover how NVIDIA Omniverse and NVIDIA AI platform technologies can help you build your own industrial digital twins and transform your operations.

Learn More