DAS-Sample-Data

<h1 align="center"><strong>DAS: Data Acquisition System</strong></h1> <p align="center">Enable embodied intelligence data acquisition to be as simple and natural as shooting a video.</p> --- ## 📋 Contents - [📦 How to Use the Dataset](#📦-how-to-use-the-dataset) - [📚 Dataset Structure](#📚-dataset-structure) - [License](#license) - [Contact](#contact) ## 📦 How to Use the Dataset Due to Hugging Face's file size limitation of 50GB per file, the dataset has been split into smaller parts. ## 📚 Dataset Structure Purpose: Each HDF5 file corresponds to a single episode and encapsulates both observational data and actions. Below is the hierarchical structure of the HDF5 file: ``` xxx.h5 ├── observations/ │ ├── cameras/ │ │ └── <camera_name_x> (Dataset) │ ├── tactile/ │ │ └── <left or right> (Dataset) │ ├── eef_pos (Dataset) │ └── imu (Dataset) ├── label/ │ └── depthmap (Dataset) └── action (Dataset) (actions mirror the eef_pos data) ``` 💡 For Demo data, it has a frame rate of 10Hz (downsample from 30Hz to 10Hz). The length and width of the image data are downsampled to half the original size. Groups and Datasets: observations/ - cameras/ - Description: Image data from camera. - Datasets: - Type: Dataset - Shape: (num_frames, height=xxx, width=xxx, channels=3) for mid fisheye camera, (num_frames, height=xxx, width=xxx, channels=3) for side wide camera - Data Type: uint8 - Compression: gzip with compression level 4. - tactile/ - Description: Pressure data from tactile sensor. - Datasets: - Type: Dataset. - Shape: (num_frames, 96), row=12, col=8 - Data Type: float32 - Compression: gzip with compression level 4. - eef_pos/ - Type: Dataset. - Shape: (num_frames, 8) - Data Type: float32 - Description: Position and orientation data for each timestep. We obtain high-precision positioning information based on SLAM technology. - Columns: [Pos_X, Pos_Y, Pos_Z, Q_X, Q_Y, Q_Z, Q_W, Gripper_width] - Compression: gzip with compression level 4. - imu/ - Type: Dataset. - Shape: (num_frames, 6) - Data Type: float32 - Description: Angular Velocity and Linear Acceleration data from IMU sensor for each timestep. We align IMU and image data based on timestamp. - Columns: [AngularVel_X, AngularVel_Y, AngularVel_Z, LinearAcc_X, LinearAcc_Y, LinearAcc_Z] - Compression: gzip with compression level 4. label/ - depthmap/ - Type: Dataset - Shape: (num_frames, height=xxx, width=xxx) for mid fisheye camera, (num_frames, height=xxx, width=xxx) for side wide camera - Data Type: float32 - Description: Depth map estimated from RGB image. At present, we have only released the depth map of one camera view. - Compression: gzip with compression level 4. action/ - Type: Dataset - Shape: (num_frames, 8) - Data Type: float32 - Description: Stores action data corresponding to each timestep. Same to eef_pos. - Columns: [Pos_X, Pos_Y, Pos_Z, Q_X, Q_Y, Q_Z, Q_W, Gripper_width] - Compression: gzip with compression level 4. ## License This project is licensed under the [MIT License](https://opensource.org/licenses/MIT). ## Contact For questions or feedback, please reach out to the [e-mail] or visit our [website](https://genrobot.ai/).