NVIDIA Omniverse Replicator

NVIDIA Omniverse's synthetic-data framework that randomizes simulated scenes and automatically outputs labeled training data.

Replicator is the framework inside NVIDIA's Omniverse platform for building synthetic data generation (SDG) pipelines, integrated into Isaac Sim as the omni.replicator extension. Collecting and labeling real images is expensive, whereas in simulation the position, category, and depth of every object are already known exactly, so labels come for free and are perfectly accurate. It has three main pieces: randomizers, which sample assets, materials, lighting, and camera poses following the domain-randomization approach; annotators, which output ground truth such as 2D/3D bounding boxes, semantic and instance segmentation, and depth; and writers, which save the results in whatever format a given model needs. In robotics it is commonly used to train perception models such as object detectors and pose estimators, and the Isaac Sim documentation also gives examples for navigation and manipulation scenes.

ExampleNVIDIA-AI-IOT's open-source pallet-detection model (sdg_pallet_model) was trained entirely on synthetic data generated with Replicator, and can be deployed to Jetson hardware with TensorRT.

Also called

Replicator, omni.replicator

Related

Synthetic Data · Domain Randomization · NVIDIA Isaac Sim · NVIDIA Omniverse · Visual Randomization · Photorealistic Rendering

Sources

Omniverse Extensions Docs: Replicator
Isaac Sim Docs: Synthetic Data Generation (Replicator tutorials)
GitHub: NVIDIA-AI-IOT/sdg_pallet_model

As of

2026-09