NVIDIA Omniverse Libraries

Tools for developers and AI agents to build simulation-ready worlds for physical AI.

Start Building With Omniverse

NVIDIA Omniverse™ libraries, part of NVIDIA Agent Toolkit, include open-source and accelerated tools for agents and software developers to integrate prebuilt physical AI simulation capabilities into their applications and services. Agents can call on these capabilities as tools, while Omniverse skills help orchestrate agentic workflows across simulation pipelines.

Fastest ways to start:

Clone

Clone a GitHub repo for the library you want to evaluate.

Run

Run an agent skill for a defined workflow.

Start

Start from a blueprint for a complete physical AI workflow.

Use

Use docs and samples to connect Omniverse capabilities to your application.

Omniverse AI Agent Skills

Use NVIDIA-verified skills to guide bounded agent workflows for OpenUSD, simulation-ready asset prep, synthetic data, neural reconstruction, viewer creation, and scene optimization, with human review at the decision points.

CAD to SimReady

Convert CAD assets into SimReady USD with geometry, materials, and physics for downstream workflows.

Best for: Turning CAD or source assets into reviewable USD outputs for simulation pipelines

Defect Image Generation

Generate anomaly examples and evaluate models for AOI inspection use cases.

Best for: PCBA, metal surface, or glass inspection data workflows

Neural Reconstruction

Reconstruct real-world driving scenes and generate variations for AV training.

Best for: Camera, lidar, radar, or stereo reconstruction workflows for AV and robotics datasets

Realtime Viewer

Inspect OpenUSD scenes with a library-native viewer built from Omniverse tools.

Best for: Creating reviewable OpenUSD viewers without modifying source USD assets

Video Data Augmentation

Augment and enrich simulation data for training physical AI models.

Best for: Enriching and labeling video datasets for physical AI model development

USD Performance Tuning

Profile and tune large digital twin scenes for runtime simulation performance.

Best for: Diagnosing and improving large USD scenes before runtime or simulation use

Omniverse Tools for Agents

Build physical AI and simulation-ready workflows with libraries and services that agents can call, inspect, and report on while developers stay in control.

ovphysx

Open source, USD-native physics library for scalable robotics and digital twin simulation.

Best for: Loading USD scenes, stepping physics, cloning environments, and exchanging simulation data through tensors

ovrtx

GPU-accelerated rendering and sensor simulation libraries built on NVIDIA RTX™.

Best for: Adding visual and sensor preflight to 3D applications and simulation workflows

ovstage

Shared, GPU-capable scene data layer for high-performance simulation and visualization workflows.

Best for: Loading USD scene data once, sharing runtime state across Omniverse libraries, and exchanging simulation data through zero-copy tensors

ovstorage

Cloud-native APIs to manage, store, and synchronize OpenUSD asset data.

Best for: Giving tools or agents a consistent way to read, write, list, and route asset data across storage backends

ovstream

Streaming library for high-throughput, low-latency GPU data sharing.

Best for: Delivering rendered or interactive output from tools such as ovrtx to local or browser clients

ovui

Python UI library for app interfaces, interaction tests, and CI screenshot checks.

Best for: Creating local tools, inspectors, and review apps around USD workflows

SimReady Foundation

Framework and pipeline for creating and validating simulation-ready assets.

Best for: Understanding asset requirements, profiles, and validation targets for supported simulation workflows

simready-validate

Validate USD assets against SimReady specifications and profiles.

Best for: Checking whether assets meet a selected simulation-readiness profile before handoff

usd-agents

Use agents to assign physics, materials, and textures to USD assets.

Best for: Reviewable material, physics, texture, and validation workflows on USD assets 

usd-search

Enable natural language search and reasoning across OpenUSD content.

Best for: Finding assets, prims, dependencies, or similar content across large USD libraries

usd-convert-asset

Convert common 3D asset formats to the interoperable OpenUSD format.

Best for: Scriptable asset ingestion into OpenUSD-based workflows 

usd-convert-gsplat

Convert Gaussian splat data into OpenUSD for simulation and rendering.

Best for: Bringing .ply or .spz Gaussian splat data into OpenUSD workflows 

mujoco-usd-converter

Convert MuJoCo models into OpenUSD for robotics simulation workflows.

Best for: Moving MuJoCo robot or simulation assets into OpenUSD for visualization, inspection, and targeted simulation workflows

urdf-usd-converter

Convert URDF robot assets into OpenUSD for robotics simulation.

Best for: Preparing URDF robots for OpenUSD review and Newton-oriented simulation workflows

usd-exchange

SDK for building interoperable OpenUSD data exchange workflows.

Best for: Developers mapping native application data into well-structured OpenUSD 

usd-optimize

Optimize OpenUSD content for performance and memory.

Best for: Reducing scene complexity, memory use, or runtime cost in OpenUSD applications and pipelines

usd-validation-nvidia

Validate OpenUSD assets against NVIDIA and core USD checks.

Best for: Adding repeatable USD validation checks to local tools, CI, or agent-reviewed pipelines

OpenUSD

Open source framework and standard for interoperable 3D data, SimReady assets, and agentic workflows.

Best for: Building interoperable 3D data workflows that can support simulation-ready content and agent-reviewed pipelines

Simulation Libraries for Physical AI

These foundational simulation libraries power advanced physics, reconstruction, and AI modeling workflows. Use them directly when you need lower-level control, or through Omniverse tools and skills when building agentic physical AI workflows.

Newton Physics

An open source, extensible physics engine built on NVIDIA Warp and OpenUSD.

NuRec

APIs, libraries, and datasets for 3D Gaussian-based neural simulation from real-world data.

PhysX

Open source multi-physics SDK for scalable robotics simulation and modeling.

PhysicsNeMo

A CUDA-X™ library and open source Python framework for building, training, and fine-tuning AI physics models at scale.

NVIDIA Warp

Open source Python framework and CUDA-X library that delivers GPU acceleration for computational physics, AI​, and optimization workflows.

img-alt-text

Omniverse Labs

Explore experimental samples and reference patterns for building physical AI workflows with Omniverse libraries, OpenUSD, and agent-callable tools.

Explore Omniverse Labs

Get Started With Blueprints and Agent Workflows

Start from reference workflows that show how Omniverse libraries, agent skills, and NVIDIA technologies come together for physical AI development.

Digital twins for AI data centers design and operations

DSX Gigawatt-Scale AI Factories

Optimize design, simulation, and operations of AI factories.

Synthetic manipulation motion generation for robotics

GR00T-Mimic Synthetic Robot Manipulation

Generate exponential synthetic manipulation data for robots.

Digital twin for complex industrial robot fleets

Mega Multi-Robot Industrial Fleets

Simulate complex industrial robot fleets before real-world deployment.

Physical AI applications

Agent Skills for Synthetic Data Generation

Turn coding agents into synthetic data experts for physical AI development.

 Digital twin for aerodynamic simulation and design

Digital Twin for Interactive Fluid Simulation

AI-powered virtual wind tunnel for real-time aerodynamic simulation and design.

Synthetic manipulation motion generation for robotics

Digital Twins for Smart Cities

Build, test, and optimize visual AI agents in SimReady digital twins to monitor city-scale operations.

Omniverse Learning Resources