NVIDIA Shows Developers Using AI Agents to Build Omniverse Simulations
NVIDIA detailed a series of developer projects that combine frontier AI agents with Omniverse libraries to build and test physics-based simulations.
Quick answer
What did NVIDIA announce about developers using AI agents with Omniverse simulation tools?
NVIDIA published a blog post describing developers across its teams who used frontier AI agents, including Astra and Claude Fable 5, alongside NVIDIA Omniverse libraries to build simulation applications covering warehouse automation, gas-turbine testing, robot training, autonomous-driving workflows and a browser-based space station model.
Key takeaways
- NVIDIA described a set of internal and ecosystem developer projects that pair frontier AI agents with NVIDIA Omniverse libraries to build simulation applications.
- Projects include reconstructing a warehouse building from site scans, a what-if gas-turbine analysis tool, a SimReady kitchen scene with more than 2,000 available library assets, and a warehouse simulation that NVIDIA says sustained 30 frames per second with 80 conveyors, 40 arms and 800 boxes.
- NVIDIA says developers also used AI agents to test flexible cable routing for electronics assembly, convert a robot CAD model into a SimReady asset, and build a humanoid warehouse simulator with first- and third-person views.
- An autonomous-driving project called Zero to Alpamayo built a reusable San Francisco street simulation, and a separate sensor-validation project used Astra and Claude Fable 5 agents to compare simulated and recorded camera and lidar data to improve digital twins.
- In a project called Robo Olympics, NVIDIA says a simulated Unitree G1 humanoid cleared a single hurdle in 64 of 100 simulation trials while developers used the results to refine the robot's timing and control.
NVIDIA details developer projects pairing AI agents with Omniverse
NVIDIA published a blog post outlining a series of projects in which developers across its teams used frontier AI agents, including Astra and Claude Fable 5, together with NVIDIA Omniverse libraries to build and test physics-based simulations. According to NVIDIA, developers direct the agents through natural-language instructions, review the results and guide further changes, while Omniverse libraries supply GPU-accelerated physics, rendering and sensor simulation.
Warehouse and industrial examples
NVIDIA says David Wehr, from its simulation technology team, used Astra to reconstruct a building from panoramic photos and point-cloud measurements, producing an editable, simulation-ready model for evaluating warehouse layouts before physical changes are made. In a separate project, Pradeep Kaushik of the Omniverse developer relations team directed Astra to build a what-if analysis application around a simulated gas turbine and generator, using tools including Cantera and SciPy, with six test procedures covering scenarios such as load changes, a stuck fuel valve and an exhaust-sensor error.
NVIDIA also describes a warehouse simulation built by principal engineer Bhavik Sanganee to test performance at scale. The company says the simulation sustained 30 frames per second at full HD resolution during a two-minute test with 80 conveyors, 40 arms and 800 boxes, completing 640 successful box placements, and that doubling the scene again reduced performance, which testing traced to the physics and robot-control pipeline.
Robotics and asset-creation projects
According to NVIDIA, Paul Kind of its Omniverse documentation team used an AI agent to assemble a kitchen scene from the SimReady Central Library, which the company says offers more than 2,000 simulation-ready USD assets, combining 180 placements from 63 unique library assets with 18 custom geometry pieces.
Other projects described by NVIDIA include a robot simulation for flexible cable routing in electronics assembly built by robotics technical marketing manager Oyindamola Omotuyi using Isaac Lab and Newton physics; a conversion of an ABB Robotics YuMi CAD model into a SimReady asset, which NVIDIA says completed four cube pick-and-place cycles in simulated testing; and a humanoid warehouse simulator with first- and third-person views built by Omniverse product manager Frank DeLise.
Autonomous driving and sensor validation
NVIDIA says Doyub Kim built Zero to Alpamayo, a reusable simulation environment based on San Francisco's Market Street, to test how scene, sensor and driving-model changes affect autonomous-vehicle simulation outcomes, and ran a separate experiment varying weather and lighting conditions. In another project, Ashley Reid directed Astra and Claude Fable 5 agents over roughly three days to compare simulated camera and lidar outputs against recorded data, creating two digital twins from scratch and improving two existing ones based on measured discrepancies.
NVIDIA also describes a project called Robo Olympics, led by Tae Kim, in which a simulated Unitree G1 humanoid cleared a single hurdle in 64 of 100 trials, along with a robotic disassembly test on a car suspension and an International Space Station model built for browser viewing using NASA assets and Omniverse rendering tools.
Source: NVIDIA Blog, "Into the Omniverse: How Developers Turn Ideas Into Simulations With Frontier AI Agents," published October 29, 2026.
Frequently asked questions
- What tools did developers use in these projects?
- NVIDIA says developers used frontier AI agents such as Astra and Claude Fable 5 together with NVIDIA Omniverse libraries, including ovrtx for rendering, ovstage for scene updates, ovphysx for physics and ovui for interfaces, along with OpenUSD for storing models.
- What is SimReady Central Library?
- NVIDIA says SimReady Central Library, announced at GTC Berlin, gives developers access to more than 2,000 simulation-ready USD assets from equipment manufacturers and asset providers for robotics simulation and AI factory digital twin workflows.
- What results did the warehouse stress test report?
- NVIDIA says a warehouse simulation with 80 conveyors, 40 arms and 800 boxes sustained 30 frames per second at full HD resolution during a two-minute test, with the arms completing 640 successful box placements, before performance dropped again when the scene was doubled further.
- What was tested in the autonomous-driving project?
- NVIDIA says a manager on its simulation technology team directed Astra to build Zero to Alpamayo, a reusable simulation environment based on San Francisco's Market Street, to test how changes to scenes, sensors or driving models affect autonomous-vehicle simulation results.