Datasetpublic

24jihoward/task_data_atm_224_square_center_crop

Real-World Franka Panda Task Data (224px Square Center Crop) Tasks Task Description Train Demos Val Demos task_1 put the pink noodle in the bowl 90 10 task_2 put the bread in the bowl 90 10 task_3 pour the pasta into the pan 89 10 task_4 close the right cabinet door 90 10 Data Format Each demo is an HDF5 file with: root/ actions: (T, 10) float64 — [pos(3), rot6d(6), gripper(1)] agentview/video: (1, T, 3… See the full description on the dataset page: https://huggingface.co/datasets/24jihoward/task_data_atm_224_square_center_crop.

sourceHugging Facemitupdated 6mo agoView on Hugging Face

0likes105downloads

Dataset Card

Real-World Franka Panda Task Data (224px Square Center Crop)

Tasks

Task	Description	Train Demos	Val Demos
task_1	put the pink noodle in the bowl	90	10
task_2	put the bread in the bowl	90	10
task_3	pour the pasta into the pan	89	10
task_4	close the right cabinet door	90	10

Data Format

Each demo is an HDF5 file with:

root/
  actions:               (T, 10) float64  — [pos(3), rot6d(6), gripper(1)]
  agentview/video:       (1, T, 3, 224, 224) uint8
  eye_in_hand/video:     (1, T, 3, 224, 224) uint8
  extra_states/ee_states:(T, 10) float64  — current EE state, same format
  task_emb_bert:         (768,) float32

Image Preprocessing

Images are square center cropped from ZED VGA (672×376):

1.Crop 148px from each side → 376×376
2.Resize → 224×224

Action Format

10D absolute end-effector pose:

—dims 0-2: position (x, y, z) in meters
—dims 3-8: 6D rotation (first two rows of rotation matrix)
—dim 9: gripper width in meters (0=closed, 0.08=open)

Robot Setup

—Robot: Franka Panda
—Cameras: 2x ZED Mini (agentview + eye-in-hand)
—Control: 10Hz
—Data collection: teleoperation via GELLO