10Kh RealOmni-Open - Gen Robot.AI's open-source embodied intelligence dataset

What is 10Kh RealOmni-Open?

10Kh RealOmni-Open is an open-source embodied intelligence dataset from the Gen Robot.AI team, and is one of the largest open-source datasets in the industry. It contains over 10,000 hours of data, more than 1 million task clips, and a total storage capacity of 95TB. The dataset focuses on 10 common household tasks, with over 10,000 clips for each skill, ensuring skill depth. The data was collected from 3,000 real-world households, with high-resolution footage (1600×1296 pixels, 30fps), sub-centimeter trajectory accuracy, and includes information on the opening and closing angles and displacement of the grippers.

10Kh RealOmni-Open's main functions

• Large-scale data coverageIt contains over 10,000 hours of data and 1 million+ task clips, with a total storage of 95TB, making it a leading open-source embodied intelligence dataset in the industry.
• Focus on core skillsThe dataset focuses on 10 common household tasks, with over 10,000 clips for each skill, ensuring in-depth coverage of the skills.
• High-quality data collectionHigh image clarity (1600×1296 pixels, 30fps), trajectory accuracy reaches sub-centimeter level, and includes the opening and closing angle and displacement information of the gripper.
• Real-world generalizationThe data was collected from 3,000 real families, 99.2% of which were long-duration tasks involving both hands. The average edit length was 1 minute and 37 seconds, recording the complete action process, making it suitable for applications in various real-world scenarios.
• High-efficiency data storageThe data is stored in MCAP format. The first phase of the data has a total duration of 950 hours and contains 39,761 task clips, which can be easily accessed by developers as needed.
• Ease of use supportIt provides web-based tools and open-source toolkits, supports both Chinese and English, follows the CC-BY-SA-4.0 license, allows commercial use, and provides detailed usage instructions.
• Continuous updates and feedbackThe dataset is still being updated to achieve "digitalization of all human skills," and user feedback and suggestions are welcome.

10Kh RealOmni-Open Data Characteristics

• Large scaleIt contains over 10,000 hours of data, 1 million+ task clips, and a total storage of 95TB, making it one of the largest open-source embodied intelligence datasets in the industry.
• Skill DepthFocusing on 10 common household tasks, each skill has over 10,000 clips, ensuring depth and richness in the skills.
• High-quality imagesHigh image clarity, with a resolution of 1600×1296 and a frame rate of 30fps, capable of recording environmental and operational details from all angles.
• High-precision trajectoryThe trajectory accuracy reaches the sub-centimeter level, achieved through high-precision IMU hardware and cloud reconstruction technology, making it suitable for robot learning.
• Multimodal dataIt includes multiple modal data such as visual and auditory data, and records the opening and closing angles and displacement information of the gripper.
• Real-world scene captureThe data comes from 3,000 real families, has strong generalization ability for scenarios and targets, and operates naturally, avoiding the problems of single scenarios and repetitive actions.
• Long-term task log99.2% of the tasks were long-duration two-handed tasks, with an average edit length of 1 minute and 37 seconds, recording the complete action process, which is suitable for learning action logic and cause and effect.
• High-efficiency storage formatThe data is stored in MCAP format. The first phase of data has a total duration of 950 hours and contains 39,761 task clips, which are convenient for developers to call up as needed.
• Usability supportIt provides web-based tools and open-source toolkits, supports both Chinese and English, follows the CC-BY-SA-4.0 license, allows commercial use, and provides detailed usage instructions.

10Kh RealOmni-Open project address

• Hugging Face Model Libraryhttps://huggingface.co/datasets/genrobot2025/10Kh-RealOmin-OpenData

Application scenarios of 10Kh RealOmni-Open

• Home service robot trainingIt provides rich task data for home service robots, helping them learn and optimize daily household skills, such as tidying up items and cleaning tabletops.
• Reinforcement learning researchIt provides large-scale real-world environment data for reinforcement learning algorithms, supports the training and optimization of algorithms in complex tasks, and improves the decision-making ability of robots.
• Research on Robot Vision and Motion CoordinationMultimodal data supports research on robot vision and motion coordination, helping robots better understand and perform tasks.
• Skills transfer and generalization researchBased on diverse home scenarios and tasks, it supports research on the transfer and generalization of robot skills and can adapt to different environmental and task requirements.
• Human-computer interaction optimizationBy recording real human-computer interaction data, we can help optimize the way robots and humans collaborate, and improve the naturalness and efficiency of the interaction.
• Embodied intelligent model developmentIt provides high-quality data support for the development of embodied intelligence models, helping the models to be applied and implemented in real-world scenarios.