robomimic-mh-square-image-dense

This dataset was created using LeRobot.

Dataset Description

A robotics demonstration dataset for training imitation learning and reinforcement learning policies.

Dataset Statistics

• Total Episodes: 300
• Total Frames: 80731
• Total Tasks: 1
• FPS: 20
• Robot Type: robomimic
• Codebase Version: v2.1

Dataset Structure

The dataset directory contains:

• meta/: Metadata files

  • info.json: Dataset configuration
  • episodes.jsonl: Episode metadata
  • episodes_stats.jsonl: Episode statistics
  • tasks.jsonl: Task information

• data/: Parquet data files

  • train-00000-of-00001.parquet: Consolidated dataset

• videos/: Video files (if available)

  • episode_*.mp4: Episode videos

Dataset Metadata

info.json

Contains comprehensive dataset metadata and configuration:

{
  "codebase_version": "v2.1",
  "robot_type": "robomimic",
  "total_episodes": 300,
  "total_frames": 80731,
  "total_tasks": 1,
  "total_videos": 22,
  "total_chunks": 1,
  "chunks_size": 1000,
  "fps": 20,
  "splits": {
    "train": "0:11"
  },
  "data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
  "video_path": "videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4",
  "features": {
    "action": {
      "dtype": "float32",
      "shape": [
        7
      ],
      "names": [
        "eef_delta_pos_x",
        "eef_delta_pos_y",
        "eef_delta_pos_z",
        "eef_delta_rot_rx",
        "eef_delta_rot_ry",
        "eef_delta_rot_rz",
        "gripper_action"
      ]
    },
    "next.done": {
      "dtype": "bool",
      "shape": [
        1
      ],
      "names": [
        "done"
      ]
    },
    "next.reward": {
      "dtype": "float32",
      "shape": [
        1
      ],
      "names": [
        "reward"
      ]
    },
    "observation.state": {
      "dtype": "float32",
      "shape": [
        9
      ],
      "names": [
        "robot0_eef_pos_0",
        "robot0_eef_pos_1",
        "robot0_eef_pos_2",
        "robot0_eef_quat_0",
        "robot0_eef_quat_1",
        "robot0_eef_quat_2",
        "robot0_eef_quat_3",
        "robot0_gripper_qpos_0",
        "robot0_gripper_qpos_1"
      ]
    },
    "observation.images.agentview": {
      "dtype": "video",
      "shape": [
        256,
        256,
        3
      ],
      "names": [
        "height",
        "width",
        "channel"
      ],
      "info": {
        "video.height": 256,
        "video.width": 256,
        "video.codec": "av1",
        "video.pix_fmt": "yuv420p",
        "video.is_depth_map": false,
        "video.fps": 20,
        "video.channels": 3,
        "has_audio": false
      }
    },
    "observation.images.robot0_eye_in_hand": {
      "dtype": "video",
      "shape": [
        256,
        256,
        3
      ],
      "names": [
        "height",
        "width",
        "channel"
      ],
      "info": {
        "video.height": 256,
        "video.width": 256,
        "video.codec": "av1",
        "video.pix_fmt": "yuv420p",
        "video.is_depth_map": false,
        "video.fps": 20,
        "video.channels": 3,
        "has_audio": false
      }
    },
    "timestamp": {
      "dtype": "float32",
      "shape": [
        1
      ],
      "names": null
    },
    "frame_index": {
      "dtype": "int64",
      "shape": [
        1
      ],
      "names": null
    },
    "episode_index": {
      "dtype": "int64",
      "shape": [
        1
      ],
      "names": null
    },
    "index": {
      "dtype": "int64",
      "shape": [
        1
      ],
      "names": null
    },
    "task_index": {
      "dtype": "int64",
      "shape": [
        1
      ],
      "names": null
    }
  }
}

Features

• action (float32) - Shape: [7] - Components: eef_delta_pos_x, eef_delta_pos_y, eef_delta_pos_z ... (+4 more)
• next.done (bool) - Shape: [1] - Components: done
• next.reward (float32) - Shape: [1] - Components: reward
• observation.state (float32) - Shape: [9] - Components: robot0_eef_pos_0, robot0_eef_pos_1, robot0_eef_pos_2 ... (+6 more)
• observation.images.agentview (video) - Shape: [256, 256, 3] - Components: height, width, channel
• observation.images.robot0_eye_in_hand (video) - Shape: [256, 256, 3] - Components: height, width, channel
• timestamp (float32) - Shape: [1]
• frame_index (int64) - Shape: [1]
• episode_index (int64) - Shape: [1]
• index (int64) - Shape: [1]
• task_index (int64) - Shape: [1]

Usage

To load this dataset with LeRobot:

from lerobot.common.datasets import LeRobotDataset
dataset = LeRobotDataset(repo_id="poolvarine/robomimic-mh-square-image-dense")

Or with the Hugging Face Datasets library:

from datasets import load_dataset
dataset = load_dataset("poolvarine/robomimic-mh-square-image-dense")

Citation

If you use this dataset, please cite LeRobot:

@software{ lerobot2024,
  author = { The HuggingFace Team },
  title = { LeRobot: A Python Library for Robot Learning },
  url = { https://github.com/huggingface/lerobot },
  year = { 2024 }
}

License

[More Information Needed]

Dataset Card Contact

[More Information Needed]