AgiBot World, together with Shanghai Artificial Intelligence Laboratory and other institutions, has launched the world's first million-level robot data set based on real scenarios, AgiBot World. Its scale and quality are significantly ahead of existing data sets. This data set covers five core scenarios of home, catering, industry, supermarket and office, and includes more than 80 daily life skills. It provides unprecedented valuable resources for embodied intelligence research and is known as the "ImageNet moment" in the field of embodied intelligence. ". The release of AgiBot World marks China's major breakthrough in the field of embodied intelligence and also contributes to global AI research.
AgiBot Robotics, together with Shanghai Artificial Intelligence Laboratory and other institutions, has released the AgiBot World data set, which is the world's first million-level robot data set based on real scenarios. Compared with Google's Open X-Embodiment, its long-range data scale is increased by 10 times, scene coverage is expanded by 100 times, and data quality reaches industrial-grade standards.
Known as the "ImageNet moment" in the field of embodied intelligence, this open source project demonstrates amazing robotic capabilities. From flower arrangement in the living room to cooking in the kitchen, from supermarket checkout to factory sorting, AgiBot World covers five core scenarios of home, catering, industry, supermarket and office, and includes more than 80 daily life skills.
In terms of hardware configuration, the robot is equipped with a 360-degree surround camera, six-degree-of-freedom dexterous hands and high-precision tactile sensors, with up to 32 degrees of freedom in the whole body. These advanced hardware facilities enable robots to complete diverse tasks ranging from simple grabbing and placing to complex dual-arm collaborative operations.
In terms of data quality control, AgiBot World has established a strict standard system:
Task design invites multiple experts to check
The entire collection process is supervised by a professional team
Data audit adopts dual screening of device and cloud
All actions are manually reviewed frame by frame
Ensure data availability through algorithm validation
Zhiyuan also revealed its planning blueprint for 2025, including open source of full data, release of simulation data sets, launch of large models of embodied bases, release of complete tool chains, and a series of challenges. This series of initiatives is expected to accelerate the development of embodied intelligence and promote the widespread application of robotics technology in real life.
The release of this open source project not only marks an important breakthrough in the field of embodied intelligence in China, but also provides valuable data resources for global robotics research and is expected to become an important milestone in promoting the development of embodied intelligence.
Project page: https://github.com/OpenDriveLab/agibot-world
The open source of the AgiBot World data set will greatly promote research and development in the field of embodied intelligence around the world. Its high-quality data and comprehensive scene coverage are expected to spawn more innovative applications, accelerate the progress of robotics technology, and ultimately benefit human society. . Looking forward to more breakthrough results based on this data set in the future.