The Earliest Bettor on First-Person View Data in China, Peking University's Lu Zongqing: Latent Space is the Path for Embodied AI

Lu Zongqing, Associate Professor at the School of Computer Science, Peking University, clearly stated at the Intelligence Source Conference that human first-person view video is the key path for achieving scaled data growth in embodied AI. This judgment differs from the current industry's general focus on robot hardware or scene setup, bringing the core competitive point back to the data source and model architecture level.

Regarding the technical route for embodied AI, Lu Zongqing proposed that latent space is the correct direction, rather than simply relying on world models. As a systematic promoter of the latent space route in China, he has long served as a Senior Area Chair for top conferences such as ICML and NeurIPS. His academic background provides theoretical support for this technical judgment, emphasizing breaking through technical certainty bottlenecks through data accumulation.

On the implementation level, the team has accumulated over 500,000 hours of first-person human video data and founded Bein company in 2025 to advance commercialization. In the context where the embodied AI industry has not yet seen certain technology, this early data betting strategy may influence future training paradigms for robot perception and decision models, providing the industry with references for new data solutions.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.