EgoVerse: Li Fei-Fei's Student Launches International Data Standard for Embodied AI; Only Chinese Company Is Guanglun Intelligence
Meta
Scale AI
Danfei Xu, a former PhD student of Li Fei-Fei, founded the EgoVerse initiative — an ecosystem and data standard for first-person human activity data to train robots. Partners include leading universities and companies such as Meta, Scale AI, and the sole Chinese company Guanglun Intelligence. EgoVerse aims to create a unified "language system" for embodied AI data, akin to ImageNet in computer vision.
Danfei Xu, a professor at the Georgia Institute of Technology and former PhD student of Fei-Fei Li, has initiated the EgoVerse project—an ever-expanding ecosystem of first-person human action data designed for training robots. The current version includes 1,362 hours of demonstrations, approximately 80,000 episodes, 1,965 tasks, 240 scenarios, and 2,087 participants. In addition to academic institutions (Georgia Tech, Stanford, ETH Zurich, UC San Diego), the alliance includes Meta, Scale AI, Guanglun Intelligence, Mecka AI, MicroAGI, and Trace Labs. The only Chinese company is Guanglun Intelligence, which is responsible for data quality control at all stages: from agent-managed on-device collection to unified cloud processing and data value verification through simulation and testing. EgoVerse aims to become the standard for embodied AI data, solving the incompatibility problem of datasets from different teams. Concurrently, Guanglun Intelligence also participates in another infrastructure project—Newton (jointly with NVIDIA, Google DeepMind, Disney Research, and TRI), which sets standards for physical simulation for robots. Thus, the company is the only Chinese participant in both global embodied AI infrastructures.
Source: QbitAI 量子位 —
original