Warehouse digital twin
A simulation of a warehouse's layout, robots, people and flows used to test slotting, fleet sizes and safety rules, and to train and check the AI that runs on site, before anything changes in the real building.
The problem
Warehouses change constantly: new product lines, seasonal peaks, added autonomous mobile robots and new pick stations. Testing these changes on a live floor disrupts operations, and robots, forklifts and people sharing aisles raise safety questions that are hard to study during production.
Planners need a way to try layouts, fleet sizes and routing rules virtually, and AI teams need realistic, site-specific data to train and test perception and robot behavior.
The approach
A warehouse twin combines a 3D model of the building and racking, simulated robots and sensors, and the operational logic that drives them. With NVIDIA's stack, the scene is built in OpenUSD with Omniverse libraries, Isaac Sim simulates mobile robots and their sensors and produces synthetic training data, and cuOpt can plan routes and assignments; NVIDIA lists Omniverse digital twins for fleet simulation among cuOpt's integrations. On the live floor, the Warehouse Operations example in the VSS Blueprint tracks people and forklifts from cameras and verifies near-miss events.
Many warehouse questions, such as slotting or labor planning, can be answered with discrete-event simulation or warehouse management analytics without a 3D twin. A physically based twin earns its cost when robots, sensors or camera AI must be trained and tested against the specific site.1234
Conceptual architecture
Diagram as a list
Applications & solutions
- Warehouse management and order dataSupplies orders, inventory and movement history to drive scenariosConnects to Routing and assignment optimizer (cuOpt), Robot and sensor simulation (Isaac Sim)
- Robot and sensor simulation (Isaac Sim)Runs fleets, sensors and people in the virtual siteConnects to Synthetic data for perception models
- Operations and safety dashboardShows events and compares plans with what happened
Models & frameworks
- OpenUSD warehouse sceneBuilding, racking, conveyors and robots as one sceneConnects to Robot and sensor simulation (Isaac Sim)
- Synthetic data for perception modelsLabeled images and video of site-specific situationsConnects to Camera analytics on site (VSS warehouse example)
Inference & runtime software
- Camera analytics on site (VSS warehouse example)Tracks people and forklifts and verifies near missesConnects to Operations and safety dashboard
Accelerated computing
- Routing and assignment optimizer (cuOpt)Computes routes and task assignments that the simulation then testsConnects to Robot and sensor simulation (Isaac Sim)
- RTX GPUs for simulationRender and simulate the twinConnects to Robot and sensor simulation (Isaac Sim)
Technologies and their roles
omniverse15
Scene and simulation libraries
OpenUSD scene handling, RTX sensor rendering and physics for the virtual warehouse; NVIDIA labels the ovrtx and ovphysx libraries pre-release and not enterprise-supported.
isaac6
Robot fleet simulation
Isaac Sim simulates mobile robots and sensors and generates synthetic data; Isaac ROS runs on the real robots.
cuopt3
Routing and assignment
Open source GPU optimization engine for vehicle routing and planning problems, with Omniverse integration for fleet simulation.
metropolis4
Camera-based safety analytics
The VSS Warehouse Operations example tracks people and forklifts and verifies near-miss events.
cosmos7
Synthetic data variety
World foundation models can generate additional synthetic video for training perception models.
What you need first8
- Building drawings, rack layouts and robot specifications
- Order and movement history to drive realistic scenarios
- Simulation and robotics skills, or an integration partner
- RTX-class GPUs for simulation; Isaac Sim does not support GPUs without RT Cores, such as A100 or H100
- A camera placement plan and privacy assessment if video analytics is included
Risks and how to reduce them
- Simulated behavior does not match the real fleet
- Calibrate robot speeds, sensor noise and congestion against measurements from the floor.
- Worker privacy with camera tracking
- Track anonymously where possible, set retention limits, inform staff and follow local labor and data protection rules.
- Building more twin than the decision needs
- Model only what the first decision requires and extend later.
- Licensing of blueprint components9
- VSS deployment files are Apache 2.0 while its microservices use an evaluation license; confirm production terms with NVIDIA.
Related
Sources
- NVIDIA Omniverse for developers (opens in a new tab)
- NVIDIA Isaac Sim (developer page and FAQ) (opens in a new tab)
- NVIDIA cuOpt (product page) (opens in a new tab)
- NVIDIA VSS Blueprint documentation: Introduction (opens in a new tab)
- NVIDIA Omniverse documentation (opens in a new tab)
- NVIDIA Isaac (developer page) (opens in a new tab)
- NVIDIA Cosmos repository README (opens in a new tab)
- Isaac Sim documentation: system requirements (opens in a new tab)
- NVIDIA VSS Blueprint documentation: License Information (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.