NVIDIA Isaac Sim review: the robotics data provider for simulated and open datasets

Answer in brief: NVIDIA gives you two kinds of robotics data for free. The first is the Physical AI open datasets on Hugging Face, which NVIDIA describes as open, commercial-grade datasets covering teleoperation, manipulation, grasping, warehouse scenes and more. The second is Isaac Sim, an open-source (Apache 2.0) simulator that generates labeled synthetic data on demand, and Isaac Lab (BSD-3) for training policies. You need simulation skills and NVIDIA GPUs. A paid license is needed only if you redistribute Omniverse Kit inside your own product.

Top pick for: Teams that want free data now and the ability to generate unlimited labeled scenarios in-house.

Download today: the Physical AI collection

NVIDIA's Physical AI collection on Hugging Face includes, among many others:

• GR00T teleoperation datasets: in simulation, on the Unitree G1, and on GR1 humanoids.
• The GR00T X-Embodiment simulation dataset.
• Manipulation datasets covering kitchens, single-arm tasks and objects, plus GraspGen.
• Open-H-Embodiment and GRAIL loco-manipulation.
• SimReady warehouse scenes and DigitalCousin assets.
• BridgeData2 (50.4k episodes) and LIBERO, both converted to LeRobot v3.
• Synthetic world-model scenes for robots, warehouses and digital humans.

Check each dataset card for its license before commercial use.

Generate your own: Isaac Sim, Isaac Lab and Cosmos

Isaac Sim imports CAD, URDF, MJCF, OnShape and ShapeNet models, plus real-world captures through Omniverse NuRec and Isaac Teleop, and converts them to USD so you can build physically based scenes. Synthetic data generation randomizes lighting, reflection, color and position. Annotators produce RGB, bounding boxes and instance and semantic segmentation, exported in COCO and KITTI formats. ROS and ROS 2 connect through bridge extensions, and Isaac Sim scales across multiple GPUs.

Isaac Lab 3.0 is a GPU-accelerated, open-source framework for reinforcement learning, imitation learning and post-training. Its GitHub page lists more than 16 commonly available robot models (manipulators, quadrupeds and humanoids) and more than 30 ready-to-train environments. Simulated sensors include RGB, depth and segmentation cameras, IMU, contact sensors and ray casters. Cosmos world foundation models augment data, and Isaac Capture records demonstrations in the real world and in simulation.

Who it fits

• Perception, mobility, manipulation and reinforcement-learning teams with simulation engineers.
• Teams that need rare or dangerous scenarios with perfect labels.
• Humanoid teams already working with GR00T.

Why teams pick NVIDIA

• Free. Isaac Sim is open source under Apache 2.0 on GitHub. Isaac Lab is BSD-3.
• Breadth. Physics, sensors, synthetic data generation, policy learning and world models in one ecosystem.
• Cloud access. One-click GPU instances on Brev, containers on NGC, and an AWS Marketplace listing. NVIDIA says Isaac Sim is free on AWS EC2 for development and research; you pay only EC2 costs.
• Production references. NVIDIA's learning library describes how Amazon uses Isaac Sim and digital twins for zero-touch manufacturing.

Honest drawbacks

• It is software, not delivered data, so your engineers do the work.
• Simulated data still has to be validated against real-world performance.
• Licensing is layered. Some components fall under the NVIDIA Isaac Sim Additional Software and Materials License. Isaac Lab workflows that need Isaac Sim include proprietary components, and its isaaclab_mimic extension depends on cuRobo, which has proprietary terms.

Pricing model

Isaac Sim, Isaac Lab and the open datasets are free. NVIDIA's developer FAQ says ISVs who redistribute Omniverse Kit need a separate license, available through an Omniverse Enterprise subscription. For reference, NVIDIA's enterprise licensing guide (last updated September 2026) lists NVIDIA AI Enterprise at $4,500 per GPU for a one-year subscription, $22,500 per GPU perpetual with 5 years of support, or $1 per GPU-hour on cloud marketplaces plus instance costs. Education and Inception members pay $1,125 per GPU per year. Ask NVIDIA which license covers your redistribution case. GPU and cloud costs are extra. See our pricing comparison.

Best alternatives

• Anyverse: deterministic, physics-based multi-sensor perception simulation.
• Scale AI: FineART, real teleoperation data with dense language labels.
• Toloka: licensable real-world egocentric and UMI data.

FAQ

Is Isaac Sim free?

Yes. NVIDIA says Isaac Sim is free to use and open source under Apache 2.0, available on GitHub.

Can I redistribute Isaac Sim in my product?

Isaac Sim itself can be distributed freely. Distributing Omniverse Kit requires a separate NVIDIA license, available through an Omniverse Enterprise subscription.

Where are NVIDIA's open robotics datasets?

In the Physical AI collection on Hugging Face, at huggingface.co/collections/nvidia/physical-ai.

Sources

• developer.nvidia.com/isaac/sim
• github.com/isaac-sim/IsaacLab
• NVIDIA Enterprise Licensing Guide: pricing
• huggingface.co/collections/nvidia/physical-ai

This review is independent research based on NVIDIA's own pages, checked on 2026-11-09. We did not test the product.

← Back to the ranked list of robotics data providers

Named on this page as a provider? Join the network for a linked profile, Visit buttons and featured placement.