RealSource World is a large-scale real-world robotics manipulation dataset collected using RS-02 dual-arm humanoid robot. This dataset contains diverse long-horizon manipulation tasks performed in real-world environments, with detailed annotations for atomic skills and quality assessments.

Key Features

14+ million frames of real-world dual-arm manipulation demonstrations.
11,428+ episodes across 35 distinct manipulation tasks.
71-dimensional proprioceptive state space including joint positions, velocities, accelerations, forces, torques, and end-effector poses.
Multi-camera visual observations (head camera, left hand camera, right hand camera) at 720×1280 resolution, 30 FPS.
Fine-grained annotations with atomic skill segmentation and quality assessments for each episode.
Diverse scenes including kitchen, conference room, convenience store, and household environments.
Dual-arm coordination tasks demonstrating complex bimanual manipulation skills.

News

[2025/12] RealSource World dataset fully uploaded with 35 tasks. Download Link
[2025/11] RealSource World released on Hugging Face. Download Link

Changelog

Version 1.2 (December 2025)

Data Format Upgrade & Fixes
- Upgraded codebase_version to v2.1
- Extended proprioceptive state space from 57 to 71 dimensions
- Added joint acceleration data (LeftJoint_Acc/RightJoint_Acc): 14 dimensions (7 × 2)
- Fixed timing synchronization issues in some data
- Optimized data structure for improved completeness and consistency
- All datasets reprocessed for high-quality output

Version 1.1 (December 2025)

Full Dataset Upload
- Complete release of all dataset files on Hugging Face
- 35 manipulation tasks with 11,428 episodes
- Total size: ~525GB
- Total files: ~104,000+

Version 1.0 (November 2025)

Initial Release
Released RealSource World dataset on Hugging Face
29 manipulation tasks with 11,428 episodes
14+ million frames of real-world dual-arm manipulation demonstrations
57-dimensional proprioceptive state space
Multi-camera visual observations (head, left hand, right hand)
Fine-grained annotations with atomic skill segmentation
Complete camera parameters (intrinsics and extrinsics) for all episodes
Quality assessment for each episode

Quick Start

Download Dataset

To download the complete dataset, you can use the following commands. If you encounter any issues, please refer to the Hugging Face official documentation.

# Ensure git-lfs is installed (https://git-lfs.com)
git lfs install

# When prompted for password, use an access token with read permissions.
# Generate from your settings page: https://huggingface.co/settings/tokens
git clone https://huggingface.co/datasets/RealSourceData/RealSource-World

# If you want to download only file pointers without large files
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/RealSourceData/RealSource-World

If you only want to download a specific task from the RealSource World dataset, for example Arrange_the_cups, follow these steps:

# Ensure Git LFS is installed (https://git-lfs.com)
git lfs install

# Initialize an empty Git repository
git init RealSource-World
cd RealSource-World

# Set remote repository
git remote add origin https://huggingface.co/datasets/RealSourceData/RealSource-World

# Enable sparse checkout
git sparse-checkout init

# Specify folders and files to download
git sparse-checkout set Arrange_the_cups scripts

# Pull data from main branch
git pull origin main

Dataset Structure

Folder Hierarchy

RealSource-World/
├── Arrange_the_cups/
│   ├── data/
│   │   └── chunk-000/
│   │       ├── episode_000000.parquet
│   │       ├── episode_000001.parquet
│   │       └── ...
│   ├── meta/
│   │   ├── info.json          # Dataset metadata and feature definitions
│   │   ├── episodes.jsonl     # Episode-level metadata
│   │   ├── episodes_stats.jsonl  # Episode statistics
│   │   ├── tasks.jsonl        # Task descriptions
│   │   ├── sub_tasks.jsonl    # Fine-grained sub-task annotations
│   │   └── camera.json        # Camera parameters for all episodes
│   └── videos/
│       └── chunk-000/
│           ├── observation.images.head_camera/
│           │   ├── episode_000000.mp4
│           │   └── ...
│           ├── observation.images.left_hand_camera/
│           │   ├── episode_000000.mp4
│           │   └── ...
│           └── observation.images.right_hand_camera/
│               ├── episode_000000.mp4
│               └── ...
├── Arrange_the_items_on_the_conference_table/
│   └── ...
├── Clean_the_convenience_store/
│   └── ...
└── ...

Each dataset folder contains:

data/ - Parquet files for observations and actions
meta/ - JSONL files for episodes, tasks, and annotations
videos/ - MP4 videos from multiple camera angles

Tasks

The following 35 tasks are included in this dataset:

Category	Task List
Kitchen Tasks	Arrange the cups, Stack the cups, Clean the kitchen counter, Tidy up the cooking counter, Cook rice using an electric rice cooker, Steam rice in a rice cooker, Steam potatoes, Steam buns, Make toast, Prepare the bread
Organization Tasks	Organize the TV cabinet, Organize the magazines, Organize the pen holder, Organize the repair tools, Organize the toys, Organize the glass tube on the rack, Replace the tissues and arrange them, Tidy up the children's room, Tidy up the conference room table, Place the books, Place the hairdryer, Place the slippers, Take down the book, Hang out the clothes to dry
Convenience Store Tasks	Clean the convenience store, Replenish tea bags
Industrial Tasks	Cable plugging, Move industrial parts to different plastic boxes, Pack the badminton shuttlecock
Daily Life Tasks	Collect the mail, Put the milk in the refrigerator, Refill the laundry detergent, Prepare the birthday cake, Take out the trash

Understanding the Dataset Format

This dataset follows the LeRobot v2.1 format. Each task directory contains:

data/: Parquet files storing time-series data (proprioceptive state, actions, timestamps)
meta/: JSON/JSONL files containing metadata, episode information, and annotations
videos/: MP4 video files from three camera perspectives

Key Metadata Files

meta/info.json: Contains dataset-level metadata, including:
- Total number of episodes, frames, videos
- Feature definitions (shapes and names of actions and observations)
- Video specifications (resolution, codec, frame rate)
Robot type and codebase version
meta/episodes.jsonl: Each line is a JSON object representing one episode, containing:
episode_index: Episode identifier
length: Number of frames in the episode
tasks: List of task descriptions
videos: Video file paths for each camera
meta/sub_tasks.jsonl: Fine-grained annotations for each episode, including:
task_steps: List of atomic skill segments with start/end frames
success_rating: Overall task success rating (1-5 points)
quality_assessments: Detailed quality metrics (pass/fail/valid)
notes: Annotation metadata
meta/camera.json: Camera intrinsics and extrinsics for each episode

Loading and Using the Dataset

This dataset is compatible with the LeRobot library. Here's how to load and use it:

from lerobot.common.datasets.lerobot_dataset import LeRobotDataset

# Load a specific task
dataset_path = "RealSource-World/Arrange_the_cups"
repo_id = "RealSourceData/RealSource-World"

# Initialize the dataset
dataset = LeRobotDataset(dataset_path, repo_id=repo_id)

# Access episode data
episode_0 = dataset[0]  # First frame of the first episode

# Episode metadata
episode_info = dataset.episode_data[0]

# Iterate through episodes
for episode_idx in range(len(dataset.episode_data)):
 episode_length = dataset.episode_data[episode_idx]["length"]
 print(f"Episode {episode_idx} has {episode_length} frames")

# Visualize episode
dataset.show_video(episode_idx=0, video_key="observation.images.head_camera")

Data Format Details

Data Fields

Each parquet file in data/ contains the following columns:

Column	Description
`observation`	Multi-modal observations (images, state)
`action`	Robot control commands
`reward`	Reward signals (optional)
`terminated`	Episode termination flag
`truncated`	Truncation flag
`language_instruction`	Natural language task description

Proprioceptive State (71 Dimensions)

The observation.state field contains comprehensive proprioceptive information:

Index Range	Component	Description
0-15	Joint Positions	7 joints × 2 arms + 2 grippers = 16 DOF
16	Lift Position	Mobile base lift height
17-22	Left Arm Force/Torque	6D force (fx, fy, fz, mx, my, mz)
23-28	Right Arm Force/Torque	6D force (fx, fy, fz, mx, my, mz)
29-35	Left Arm Joint Velocities	7 joints = 7 DOF
36-42	Right Arm Joint Velocities	7 joints = 7 DOF
43-49	Left End-Effector Pose	Position (x, y, z) + Quaternion (qw, qx, qy, qz)
50-56	Right End-Effector Pose	Position (x, y, z) + Quaternion (qw, qx, qy, qz)
57-63	Left Arm Joint Accelerations	7 joints = 7 DOF
64-70	Right Arm Joint Accelerations	7 joints = 7 DOF

State Field Names

[
 "LeftFollowerArm_Joint1.pos", ..., "LeftFollowerArm_Joint7.pos",
 "LeftGripper.pos",
 "RightFollowerArm_Joint1.pos", ..., "RightFollowerArm_Joint7.pos",
 "RightGripper.pos",
 "Lift.position",
 "LeftForce.fx", "LeftForce.fy", "LeftForce.fz",
 "LeftForce.mx", "LeftForce.my", "LeftForce.mz",
 "RightForce.fx", "RightForce.fy", "RightForce.fz",
 "RightForce.mx", "RightForce.my", "RightForce.mz",
 "LeftJoint_Vel1", ..., "LeftJoint_Vel7",
 "RightJoint_Vel1", ..., "RightJoint_Vel7",
 "LeftEnd_X", "LeftEnd_Y", "LeftEnd_Z",
 "LeftEnd_Qw", "LeftEnd_Qx", "LeftEnd_Qy", "LeftEnd_Qz",
 "RightEnd_X", "RightEnd_Y", "RightEnd_Z",
 "RightEnd_Qw", "RightEnd_Qx", "RightEnd_Qy", "RightEnd_Qz",
 "LeftJoint_Acc1", ..., "LeftJoint_Acc7",
 "RightJoint_Acc1", ..., "RightJoint_Acc7"
]

Action Space (17 Dimensions)

The action field contains commands sent to the robot:

Component	Description
0-6	Left arm joint positions (7 DOF)
7	Left gripper position
8-14	Right arm joint positions (7 DOF)
15	Right gripper position
16	Lift platform command

Action Field Names

[
 "LeftLeaderArm_Joint1.pos", ..., "LeftLeaderArm_Joint7.pos",
 "LeftGripper.pos",
 "RightLeaderArm_Joint1.pos", ..., "RightLeaderArm_Joint7.pos",
 "RightGripper.pos",
 "Lift.command"
]

Visual Observations

Each episode contains synchronized video from three camera perspectives:

observation.images.head_camera: Head/overhead perspective
observation.images.left_hand_camera: Camera mounted on left end-effector
observation.images.right_hand_camera: Camera mounted on right end-effector

Video Specifications:

Resolution: 720 × 1280 pixels
Frame Rate: 30 FPS
Codec: H.264
Format: MP4

Camera Parameters

Camera parameters for each episode are stored in the meta/camera.json file, keyed by episode_XXXXXX. Camera parameters include intrinsics (camera matrix and distortion coefficients) and extrinsics (hand-eye calibration parameters).

File Structure

The camera.json file contains camera parameters for all episodes:

{
 "episode_000000": {
 "camera_ids": {
 "head": "245022300889",
 "left_arm": "245022301980",
 "right_arm": "245022300408",
 "foot": ""
 },
 "camera_parameters": {
 "head": {
 "720P": {
 "MTX": [[648.57, 0, 645.54], [0, 647.80, 375.38], [0, 0, 1]],
 "DIST": [-0.0513, 0.0587, -0.0006, 0.00096, -0.0186]
 },
 "480P": { ... }
 },
 "left_arm": { ... },
 "right_arm": { ... }
 },
 "hand_eye": {
 "left_arm_in_eye": {
 "R": [[...], [...], [...]],
 "T": [x, y, z]
 },
 "right_arm_in_eye": { ... },
 "left_arm_to_eye": { ... },
 "right_arm_to_eye": { ... }
 }
 },
 "episode_000001": { ... }
}

Camera Intrinsics

Each camera (head, left_arm, right_arm) contains intrinsics for two resolutions:

MTX: 3×3 camera intrinsic matrix

[fx 0 cx]
[0 fy cy]
[0 0 1]

fx, fy: Focal length (in pixels)
cx, cy: Principal point (optical center) coordinates (in pixels)
DIST: 5-element distortion coefficients (k1, k2, p1, p2, k3)
Used for correcting radial and tangential distortion

Available Resolutions:

720P: 720p video parameters (720 × 1280)
480P: 480p video parameters (480 × 640)

Hand-Eye Calibration (Extrinsics)

The hand_eye section contains transformation relationships between robot end-effectors and cameras:

left_arm_in_eye: Transformation from left wrist camera (left end-effector camera) to left manipulator end center
R: 3×3 rotation matrix
T: 3×1 translation vector [x, y, z] (in meters)
Represents the position and orientation of the left wrist-mounted camera relative to the left manipulator end center
right_arm_in_eye: Transformation from right wrist camera (right end-effector camera) to right manipulator end center
- Represents the position and orientation of the right wrist-mounted camera relative to the right manipulator end center
left_arm_to_eye: Transformation from head camera to left manipulator base coordinate system
R: 3×3 rotation matrix
T: 3×1 translation vector [x, y, z] (in meters)
Represents the position and orientation of the head camera relative to the left manipulator base coordinate system
right_arm_to_eye: Transformation from head camera to right manipulator base coordinate system
- Represents the position and orientation of the head camera relative to the right manipulator base coordinate system

These parameters support the following coordinate transformations:

Conversion between robot end-effector poses and camera image coordinates
Conversion between 3D positions in robot space and image pixel coordinates
Multi-view geometry operations and calibration
Conversion between wrist camera coordinate system and end center
Conversion between head camera coordinate system and manipulator base coordinate system

Camera IDs

Each camera has a unique identifier:

head: Head camera ID
left_arm: Left end-effector camera ID
right_arm: Right end-effector camera ID
foot: Foot camera ID (if available)

Sub-task Annotations

Each episode in meta/sub_tasks.jsonl contains detailed annotations:

{
 "task": "Separate the two stacked cups in the dish and place them on the two sides of the dish.",
  "language": "en",
 "task_index": 0,
 "episode_index": 0,
 "task_steps": [
 {
 "step_name": "Left arm picks up the stack of cups from the center of the plate",
 "start_frame": 100,
 "end_frame": 180,
 "description": "Left arm picks up the stack of cups from the center of the plate",
 "duration_frames": 80
 },
 ...
 ],
 "success_rating": 5,
 "notes": "annotation_date: 2025/11/13",
 "quality_assessments": {
 "overall_valid": "VALID",
 "movement_fluency": "PASS",
 "grasp_success": "PASS",
 "placement_quality": "PASS",
 ...
 },
 "total_frames": 946
}

Quality Assessment Metrics

overall_valid: Overall episode validity (valid/invalid)
movement_fluency: Robot movement fluency (pass/fail)
grasp_success: Grasp action success (pass/fail)
placement_quality: Object placement quality (pass/fail)
no_drop: No objects dropped during task (pass/fail)
grasp_collisions: No collisions during grasping (pass/fail)
arm_collisions: No arm collisions (pass/fail)
operation_completeness: Task completion status (pass/fail)
And other metrics...

Dataset Statistics

Overall Statistics

Total Tasks: 35
Total Episodes: 11,428
Total Frames: 14,085,107
Total Videos: 34,284 (3 cameras × 11,428 episodes)
Robot Type: RS-02 (dual-arm humanoid robot)
Dataset Format: LeRobot v2.1
Video Resolution: 720 × 1280
Frame Rate: 30 FPS

Task Distribution

The dataset contains diverse manipulation tasks across multiple domains:

Category	Task List
Kitchen Tasks	Arrange the cups, Stack the cups, Clean the kitchen counter, Tidy up the cooking counter, Cook rice using an electric rice cooker, Steam rice in a rice cooker, Steam potatoes, Steam buns, Make toast, Prepare the bread
Organization Tasks	Organize the TV cabinet, Organize the magazines, Organize the pen holder, Organize the repair tools, Organize the toys, Organize the glass tube on the rack, Replace the tissues and arrange them, Tidy up the children's room, Tidy up the conference room table, Place the books, Place the hairdryer, Place the slippers, Take down the book, Hang out the clothes to dry
Convenience Store Tasks	Clean the convenience store, Replenish tea bags
Industrial Tasks	Cable plugging, Move industrial parts to different plastic boxes, Pack the badminton shuttlecock
Daily Life Tasks	Collect the mail, Put the milk in the refrigerator, Refill the laundry detergent, Prepare the birthday cake, Take out the trash

Robot URDF Model

RealSource World dataset is collected using the RS-02 dual-arm humanoid robot. For simulation, visualization, and research purposes, we provide the URDF (Unified Robot Description Format) model of the RS-02 robot.

RS-02 Robot Specifications

Robot Type: Dual-arm humanoid robot
Total Links: 46 links
Total Joints: 45 joints
Manipulator Arms: 2 × 7-DOF manipulator arms (left and right)
End Effectors: Dual grippers, each with 8 DOF
Base: Mobile base with wheels and lift platform
Sensors: Head camera, left hand camera, right hand camera

URDF Package Structure

The RS-02 URDF package contains:

RS-02/
├── urdf/
│   ├── RS-02.urdf          # Main URDF file (59KB)
│   └── RS-02.csv           # Joint configuration data
├── meshes/                 # 3D mesh models (46 STL files)
│   ├── base_link.STL
│   ├── L_Link_1-7.STL      # Left arm links
│   ├── R_Link_1-7.STL      # Right arm links
│   ├── ltool_*.STL         # Left gripper components
│   ├── rtool_*.STL         # Right gripper components
│   ├── head_*.STL          # Head components
│   └── camera_*.STL         # Camera mounts
├── config/
│   └── joint_names_RS-02.yaml  # Joint name configuration
├── launch/
│   ├── display.launch      # RViz visualization
│   └── gazebo.launch       # Gazebo simulation
└── package.xml             # ROS package metadata

Using the URDF Model

ROS/ROS2 Users

The URDF model can be used with ROS tools:

Visualize in RViz:

roslaunch RS-02 display.launch

Simulate in Gazebo:

roslaunch RS-02 gazebo.launch

License and Citation

This dataset is released under the CC BY-NC-SA 4.0 license (Attribution-NonCommercial-ShareAlike).

License

This work is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nc-sa/4.0/ or send a letter to Creative Commons, PO Box 1866, Mountain View, CA 94042, USA.

Citation

If you use this dataset in your research, please cite:

@misc{realsourceworld,
  title={RealSource World: A Large-Scale Real-World Dual-Arm Manipulation Dataset},
  author={RealSource},
  howpublished={\url{https://huggingface.co/datasets/RealSourceData/RealSource-World}},
  year={2025}
}

Downloads last month: 88,499