PhysicalAI-Robotics-GR00T-X-Embodiment-Sim
VerifiedCollection of 9k bimanual robot trajectories for GR00T N1 post-training.
What is PhysicalAI-Robotics-GR00T-X-Embodiment-Sim?
PhysicalAI-Robotics-GR00T-X-Embodiment-Sim contains simulation trajectories from multiple bimanual robot embodiments performing manipulation tasks. These datasets support post-training of the GR00T N1 model with examples including threading, tray lifting, and assembly operations.
It is useful for researchers and developers working on robotics foundation models and embodied AI, particularly those fine-tuning models on diverse robot hardware configurations.
What you can build with PhysicalAI-Robotics-GR00T-X-Embodiment-Sim
Fine-tune GR00T N1 policies
Load specific subsets like bimanual_panda_gripper.Threading to continue training the GR00T N1 model on additional simulation trajectories for improved bimanual manipulation.
Benchmark multi-embodiment control
Compare policy performance across the 9k trajectories from different robot hands and grippers on tasks such as LiftTray to measure generalization.
Develop imitation learning baselines
Use the trajectory data to train and evaluate behavior cloning or diffusion policies for bimanual robot tasks in simulation before real-world deployment.
Load PhysicalAI-Robotics-GR00T-X-Embodiment-Sim
from datasets import load_dataset
ds = load_dataset("nvidia/PhysicalAI-Robotics-GR00T-X-Embodiment-Sim")- 1pip install datasets
- 2from datasets import load_dataset
- 3dataset = load_dataset('nvidia/PhysicalAI-Robotics-GR00T-X-Embodiment-Sim')
- 4Select a subset such as 'bimanual_panda_gripper.Threading'
- 5Iterate over trajectories for training loops
PhysicalAI-Robotics-GR00T-X-Embodiment-Sim: pros & cons
Pros
- +9k simulation trajectories across multiple bimanual embodiments
- +Ready-made named task subsets with 1000 examples each
- +Designed specifically for GR00T N1 post-training
- +Accessible directly via Hugging Face datasets library
Cons
- –Simulation data only, no real-robot recordings
- –Limited to listed bimanual panda tasks
- –Requires NVIDIA GR00T ecosystem for full intended use
Frequently asked questions
A collection of approximately 9k simulation trajectories from bimanual robot embodiments for post-training GR00T N1.
User reviews
Verified reviews from the community shape this listing's rating.
Loading reviews…
Similar datasets
Other ai & machine learning options worth comparing.