Download Datasets#

We provide pre-generated datasets for all four benchmark tasks, hosted on Hugging Face. Each dataset contains mindmap-formatted data ready for training, along with the original HDF5 demonstration files.

Available Datasets#

Cube Stacking: nvidia/PhysicalAI-Robotics-mindmap-Franka-Cube-Stacking
Mug in Drawer: nvidia/PhysicalAI-Robotics-mindmap-Franka-Mug-in-Drawer
Drill in Box: nvidia/PhysicalAI-Robotics-mindmap-GR1-Drill-in-Box
Stick in Bin: nvidia/PhysicalAI-Robotics-mindmap-GR1-Stick-in-Bin

Each dataset includes:

Original HDF5 file from Isaac Lab Mimic for data generation
10 demonstrations in mindmap format, ready for training

Prerequisites#

Before downloading, ensure you have:

Hugging Face account registered
Read-access token ready for authentication

Download Instructions#

Install the Hugging Face Hub CLI:
```
pip install -U "huggingface_hub[cli]"
```
Authenticate with your token:
```
hf auth login
```

Download a dataset for your chosen task:

hf download \
   nvidia/PhysicalAI-Robotics-mindmap-Franka-Cube-Stacking \
   --repo-type dataset \
   --local-dir <LOCAL_DATASET_PATH>

hf download \
   nvidia/PhysicalAI-Robotics-mindmap-Franka-Mug-in-Drawer \
   --repo-type dataset \
   --local-dir <LOCAL_DATASET_PATH>

hf download \
   nvidia/PhysicalAI-Robotics-mindmap-GR1-Drill-in-Box \
   --repo-type dataset \
   --local-dir <LOCAL_DATASET_PATH>

hf download \
   nvidia/PhysicalAI-Robotics-mindmap-GR1-Stick-in-Bin \
   --repo-type dataset \
   --local-dir <LOCAL_DATASET_PATH>

Extract the demonstration files:

python mindmap/scripts/untar_demos.py \
    --demos_dir <LOCAL_DATASET_PATH> \
    --num_processes 10 \
    --remove_tar

Dataset Structure#

After extraction, your downloaded dataset will have the following structure:

📂 <LOCAL_DATASET_PATH>
├── 📂 demo_00000/
├── 📂 demo_00001/
├── 📂 ...
├── 📂 demo_00009/
├── <HDF5_FILE_NAME>.hdf5
└── README.md

Each demo_XXXXX/ folder contains the mindmap-formatted data for one demonstration, while the HDF5 file contains the original Isaac Lab Mimic data.

For detailed information about the dataset structure, see Dataset Structure.

Note

Replace <LOCAL_DATASET_PATH> with your desired local directory path for storing the dataset.