Manufacturing
How manufacturers apply AI and NVIDIA technologies: digital twins for plant and line planning, robot cells rehearsed in simulation, camera-based inspection and safety monitoring, and optimized internal transport, with notes on when the tools a plant already licenses are enough.
The problem
Every hour a production line stands still costs output, so launches, changeovers and new equipment are planned months ahead and still bring surprises at ramp-up. Engineering, automation and quality teams each keep their own picture of the plant, and the knowledge of how a line really behaves often sits with a few experienced people.
Quality is a second pressure point. Many defects are small, rare or look different under changing light and material, so rule-based machine vision needs frequent retuning and manual checks do not keep pace with output. Equipment failures between fixed maintenance intervals add downtime that the schedule did not foresee.
Staff shortages and growing product variety push plants toward more robots and flexible cells, yet each new robot must be programmed, risk-assessed and wired into existing controllers and execution systems, which slows adoption on brownfield sites.
The approach
NVIDIA's manufacturing page groups its offer into product development, engineering simulation and AI-enabled factories, where an industrial digital twin combines physics, IoT data and design and maintenance records. A plant twin is typically assembled from CAD and layout data in OpenUSD with Omniverse libraries. Foxconn, for example, runs a factory twin platform on Omniverse and uses Isaac Sim to refine robot tasks such as screw tightening and cable insertion before they reach the line. PepsiCo works with Siemens Digital Twin Composer, which is built on Omniverse libraries, in selected US plants, a reminder that many manufacturers will meet this technology through their automation vendor rather than build it themselves.
On the shop floor, Metropolis provides customized vision models for automated visual inspection and video analytics agents that track activity across several cameras; Foxconn uses Metropolis-based agents to flag unauthorized entry into restricted areas. For internal transport, Foxconn links cuOpt to its material control systems to optimize automated guided vehicle paths. Cosmos 3 reasoning models can add alerts and dense captions to quality inspection video.
Simpler routes often do the job. A single line change can usually be studied in the process simulation module of the CAD or automation suite a plant already pays for, a smart camera with built-in tools covers many fixed, well-lit checks, and vibration or temperature sensors with standard analytics handle much predictive maintenance. GPU-based twins and custom vision models earn their cost when layouts change often, a proven line is copied to other sites, or defects vary too much for fixed rules.12345
Conceptual architecture
Diagram as a list
Applications & solutions
- CAD, PLM, MES and controller dataSupplies geometry, process steps, machine states and quality recordsConnects to Factory twin built with Omniverse libraries on OpenUSD
- Factory twin built with Omniverse libraries on OpenUSDShared model of lines, cells and material flow for planning changesConnects to Robot cell simulation (Isaac Sim), Internal transport optimizer (cuOpt)
- Robot cell simulation (Isaac Sim)Rehearses robot tasks and produces synthetic images before commissioningConnects to Inspection and safety models (Metropolis, Cosmos)
- Quality, maintenance and dispatch applicationsActs on results and feeds actual performance back to the twinConnects to Factory twin built with Omniverse libraries on OpenUSD
Models & frameworks
- Inspection and safety models (Metropolis, Cosmos)Detect defects and restricted-area events in line camera videoConnects to Line-side inference on edge GPUs
Inference & runtime software
- Line-side inference on edge GPUsRuns vision models next to the cameras with low latencyConnects to Quality, maintenance and dispatch applications
Accelerated computing
- Internal transport optimizer (cuOpt)Plans guided vehicle paths that the twin can testConnects to Quality, maintenance and dispatch applications
- RTX servers or workstationsRender and simulate the twin and robot cellsConnects to Factory twin built with Omniverse libraries on OpenUSD, Robot cell simulation (Isaac Sim)
Technologies and their roles
NVIDIA Omniverse6
Factory twin libraries
Rendering, physics and OpenUSD scene handling behind plant twins, used directly by Foxconn and inside Siemens Digital Twin Composer; some libraries are labeled pre-release.
NVIDIA Isaac2
Robot cell rehearsal
Isaac Sim lets engineers refine assembly tasks such as screw tightening and cable insertion before robots are installed.
NVIDIA Metropolis4
Inspection and floor monitoring
Vision AI platform listing automated visual inspection and multi-camera video analytics agents for factories.
NVIDIA cuOpt7
Internal logistics optimization
Open source GPU solver that Foxconn connects to material control systems to plan guided vehicle paths.
NVIDIA Cosmos5
Video reasoning for quality
Cosmos 3 reasoning models can generate alerts and dense captions for quality inspection footage.
What you need first
- Current CAD, layout and equipment data, with a named owner who keeps the plant model in step with engineering changes
- Read access to controller, MES and quality system data through documented interfaces
- Labeled images of real defects, including rare ones, or a plan to create synthetic examples
- Automation engineers who can check simulated robot programs against the real controllers
- OpenUSD, Python and simulation skills in house or through a systems integrator
- GPU capacity for simulation, checked against each tool's published system requirements
- An agreement with employee representatives wherever line cameras may capture staff
Risks and how to reduce them
- Machine and robot safety on the shop floor
- Robot programs refined in simulation still go through a cell risk assessment, and guarding, light curtains and safety controllers stay in place.
- Monitoring of workers through line and area cameras
- Aim cameras at products and zones rather than people, mask faces where events allow, and settle purpose and retention with works councils before go-live.
- Supplier and customer design data inside the twin
- Equipment CAD from suppliers and customer product geometry often sit under confidentiality terms; restrict access per scene layer and confirm contract rights before sharing the model.
- Pre-release software and differing licenses6
- Some Omniverse libraries are pre-release and not enterprise-supported; review the license and support level of each library before production use.
- Planning on vendor-reported gains
- Most published factory twin results are projections or single-site figures; measure your own baseline and run a pilot before committing capital.
Documented examples
PepsiCo, with Siemens · Food and beverage manufacturing and logistics
PepsiCo and Siemens: plant and warehouse twins with Digital Twin Composer on Omniverse
PepsiCo is converting selected US plants and warehouses into 3D digital twins with Siemens Digital Twin Composer, which Siemens builds on NVIDIA Omniverse libraries. PepsiCo and Siemens report a 20 percent throughput gain on the first deployment; the program is in early pilots.
Pilot
BMW Group · Automotive manufacturing
BMW Group: planning car plants in an Omniverse-based Virtual Factory
BMW Group plans and checks its car plants in a Virtual Factory built on NVIDIA Omniverse and OpenUSD. BMW says digital twins now cover more than 30 production sites, cut collision checks for new models from almost four weeks to about three days, and are projected to lower planning costs by up to 30 percent.
Scaling
Foxconn (Hon Hai Technology Group) · Electronics manufacturing
Foxconn: digital twins for new server plants with Omniverse, Isaac and Metropolis
Foxconn uses NVIDIA Omniverse digital twins to plan production lines, Isaac to simulate robots and Metropolis for camera-based monitoring, from Hsinchu to new server plants in Mexico and the US. Most published results are expectations, such as a forecast energy cut of over 30 percent in Mexico.
In production
Related
Sources
- NVIDIA manufacturing industry page (opens in a new tab)
- Foxconn develops physical AI-enabled smart factories with digital twins (case study) (opens in a new tab)
- PepsiCo announces industry-first AI and digital twin collaboration with Siemens and NVIDIA (opens in a new tab)
- NVIDIA Metropolis (product page) (opens in a new tab)
- NVIDIA Cosmos (product page and FAQ) (opens in a new tab)
- NVIDIA Omniverse (product page) (opens in a new tab)
- NVIDIA cuOpt (product page) (opens in a new tab)
Thank you. Your correction was sent.
The editors check it against the sources. If you left an email address, they may reply about it.