RoboCOIN - A real robot dataset for dual-armed robots open-sourced by Wisdom Source in collaboration with several universities

What is RoboCOIN?

RoboCOIN is the world's first large-scale dual-arm robot dataset open-sourced by Beijing Zhiyuan Artificial Intelligence Research Institute in cooperation with several enterprises and universities, containing 15 robot platforms, 180,000 real operation trajectories and 421 task scenarios. The most important feature is the use of hierarchical annotation system, the task is disassembled into trajectories, segments and frames at three levels, supporting full-stack learning from macro-planning to micro-control. The supporting open-source CoRobot framework provides standardized data processing tools, which can quickly invoke data with one line of command. The dataset has been applied to improve the performance of visual-verbal action models, with the success rate of complex tasks increasing from 20% to 70%, and the number of downloads exceeding 100,000 in the first week of launch.

Features of RoboCOIN

• Scale and diversity: Contains more than 180,000 demo data covering 421 tasks and 16 different scenarios, such as home, business, factory, and more.
• Multi-figure platform: Data were collected from 15 different robotic platforms covering bi-armed, semi-humanoid and humanoid robots.
• Rich sensor data: Contains multi-view RGB and depth images, as well as detailed kinematic states.
• Pyramid of Hierarchical Competencies: Provides multi-resolution annotation from trajectory-level concepts, segment-level subtasks, to frame-level kinematics, enabling models to learn structurally from global planning to precise control.
• CoRobot Framework: Includes the RTML quality assessment language, an automated annotation toolchain, and a unified multi-object management platform that creates an infrastructure for scalable robotics learning.

RoboCOIN's core strengths

• Large-scale, high-quality data: Providing more than 180,000 demo data covering a wide range of tasks and scenarios, the data is large in volume and high in quality, providing a solid foundation for embodied intelligence research.
• Multi-platform versatility: The data are collected from 15 different robot platforms, covering a wide range of robots of various forms and functions, with strong generalization and adaptability.
• Multimodal data fusion: Integration of multiple sensor data such as RGB images, depth images, and kinematic states provides rich information inputs to the model and enhances the understanding of the environment and task.
• hierarchical labeling system: Using hierarchical annotation from trajectory level to frame level, it supports structured learning from macro-task planning to micro-motion control, and improves the generalization ability of the model.
• Automated annotation tools: Equipped with an automated annotation tool chain to improve the efficiency and quality of data annotation, reduce labor costs, and accelerate the construction and updating of datasets.
• open source sharing: The dataset, toolchain and technical report are fully open-sourced to support free use by individual developers, research institutions and enterprises, and to promote collaborative innovation in the industry.
• Unified Management PlatformThe CoRobot framework realizes the unified management and scheduling of the multifigure platform, supports large-scale data collection and model training, and improves the R&D efficiency.
• Quality assessment system: Introducing the RTML quality assessment language for standardized assessment of data annotation and model performance to ensure data quality and model reliability.

What is RoboCOIN's official website?

• Project website:: https://flagopen.github.io/RoboCOIN/
• Github repository:: https://github.com/FlagOpen/RoboCOIN
• arXiv Technical Paper:: https://arxiv.org/pdf/2511.17441

Who RoboCOIN is for

• Artificial intelligence researchers: Researchers working in the fields of embodied intelligence, robot learning, and reinforcement learning can use RoboCOIN's datasets and tools for model training and algorithm optimization.
• Robotics Engineer: Engineers focused on robotics development and applications can leverage this dataset to improve the ability of robots to operate in complex environments and perform tasks efficiently.
• teachers and students of higher education: Teachers and students of computer science, artificial intelligence, robotics engineering and other related majors can use it for teaching practice and research projects to promote academic research and talent training.
• R&D Team of Technology EnterprisesRoboCOIN can be used to accelerate product development and enhance product competitiveness for R&D personnel of companies committed to intelligent robot product development.
• algorithm developer: Developers interested in machine learning and deep learning algorithms can experiment and innovate with algorithms using RoboCOIN's dataset.
• Industry Application Developers: Developers who are developing robotics applications in the domestic, commercial, industrial, and medical fields can enhance the usefulness and adaptability of their applications based on RoboCOIN.