PhysicalAI Robotics GR00T X-Embodiment Sim
PhysicalAI Robotics GR00T X-Embodiment Sim is a large-scale simulation dataset developed by NVIDIA for post-training the GR00T N1 foundation model. Released in March 2026 under CC BY 4.0, it contains trajectories from multiple robot embodiments performing manipulation and loco-manipulation tasks across Isaac Lab simulation environments. The dataset covers cross-embodiment bimanual manipulation, humanoid tabletop manipulation, robot arm kitchen tasks, and loco-manipulation scenarios. With 422K downloads making it the most downloaded physical AI dataset on Hugging Face, it represents the primary simulation training source for NVIDIA's GR00T humanoid robot foundation model. Data is generated using NVIDIA Isaac Sim and Isaac Lab, providing perfect ground truth state information across diverse robot configurations not available in real-world collection.
| Year | 2026 |
|---|---|
| Embodiments | Franka Panda, Isaac Lab robots, humanoid robots, quadrupeds |
| Modalities | rgb, proprioception, depth |
| Task categories | manipulation, pick-and-place, loco-manipulation, bimanual |
| Data format | parquet, zarr |
| License | CC BY 4.0 |
| Access | open — commercial use permitted |
| Maintainer | NVIDIA |
| Origin country | US |
What is it?
PhysicalAI Robotics GR00T X-Embodiment Sim is NVIDIA's primary simulation dataset for post-training the GR00T N1 humanoid robot foundation model. Released in March 2026 under CC BY 4.0, it contains synthetic trajectories generated in Isaac Lab and Isaac Sim across multiple robot embodiments performing manipulation and loco-manipulation tasks. With 422K downloads it is the most downloaded physical AI dataset on Hugging Face. The dataset covers cross-embodiment bimanual manipulation, humanoid tabletop manipulation, robot arm kitchen tasks, and loco-manipulation scenarios with perfect ground truth state information unavailable in real-world collection.
Who is it for?
Researchers and engineers working with NVIDIA's GR00T foundation model ecosystem — training VLA models, fine-tuning generalised robot policies, and developing sim-to-real transfer pipelines. Particularly valuable for teams using Isaac Lab who want to extend GR00T N1 capabilities to new tasks or embodiments.
Key specifications
- Source: NVIDIA Isaac Sim and Isaac Lab simulation
- Embodiments: Franka Panda, humanoid robots, quadrupeds, cross-embodiment
- Tasks: Bimanual manipulation, tabletop manipulation, kitchen tasks, loco-manipulation
- Format: Parquet, Zarr
- License: CC BY 4.0 — commercial use permitted
- Access: Open — Hugging Face
How it compares
The GR00T X-Embodiment Sim dataset is simulation-only — complementing the real-robot PhysicalAI GR00T Teleop G1 dataset (187 episodes, real Unitree G1 hardware). Together they form the two-track training approach: sim for breadth and scale, real for fidelity and specificity. Open X-Embodiment covers more real-robot embodiments (22) but lacks the simulation depth and GR00T-specific task structure.
Limitations and access notes
Synthetic data only — policies trained exclusively on this dataset require sim-to-real adaptation before real-world deployment. CC BY 4.0 permits commercial use with attribution.
Frequently asked questions
What is the GR00T X-Embodiment Sim dataset?
PhysicalAI Robotics GR00T X-Embodiment Sim is NVIDIA's simulation dataset for post-training the GR00T N1 humanoid robot foundation model. Generated in Isaac Lab and Isaac Sim, it contains cross-embodiment trajectories covering manipulation and loco-manipulation tasks with perfect ground truth state information. It is the most downloaded physical AI dataset on Hugging Face with 422K downloads.
Can GR00T X-Embodiment Sim be used commercially?
Yes. The dataset is licensed under CC BY 4.0, which permits commercial use with attribution to NVIDIA.
How does this differ from GR00T Teleop G1?
GR00T X-Embodiment Sim is synthetic simulation data generated in Isaac Lab across multiple embodiments. GR00T Teleop G1 is real-robot teleoperation data collected on a physical Unitree G1. They are complementary — simulation for scale and diversity, real data for fidelity. NVIDIA uses both in the GR00T training pipeline.
How do I access GR00T X-Embodiment Sim?
The dataset is available on Hugging Face at huggingface.co/datasets/nvidia/PhysicalAI-Robotics-GR00T-X-Embodiment-Sim. No registration required.
What is Isaac Lab?
Isaac Lab is NVIDIA's open-source robot learning framework built on Isaac Sim. It provides environments, reward functions, and training infrastructure for reinforcement learning and imitation learning on GPU-accelerated physics simulation. The GR00T X-Embodiment Sim dataset was generated using Isaac Lab environments.