Beijing’s 'Robot School' trains humanoids
A Robot Data Training Center in Beijing is training humanoids across manufacturing, smart homes, and commercial service scenarios—collecting real-world data rather than relying only on simulation. That on‑floor, data‑heavy approach aims to accelerate embodied AI performance by closing the loop between deployed hardware and model training. (english.news.cn)
Phase III of Beijing’s humanoid Robot Data Training Center opened at Shougang Park in Shijingshan district during the Zhongguancun Forum on March 25, 2026. (global.chinadaily.com.cn)) Operator and industry partner listings show RealMan Robotics as a core technology and deployment partner responsible for daily operations of the center. (roboticsandautomationnews.com)) Press reports give two scale figures for the facility: multiple industry outlets and RealMan describe a 3,000-square-meter site hosting 108 robots, while municipal materials and later coverage describe a larger Phase II/III campus of roughly 10,000+ square meters (about 107,000 sq ft) and “more than 100” robots. (roboticsandautomationnews.com)) Publishers tied to the center report annual data yields ranging from “over one million” to “over six million” high-quality multimodal data points produced for model training, and say the platform has built a standardized, full-process data system. (roboticsbusinessnews.com)) The site deploys more than ten real-world scenarios — examples listed include supermarkets, hotels, homes, eldercare, automotive assembly and smart catering — and a heterogeneous robot mix that includes dual‑arm lifting units, wheeled humanoids, drone‑arm systems and quadruped platforms. (industrial-production-worldwide.com)) Organizers frame the center’s mission as closing the simulation-to-reality gap by generating on‑floor, cross‑scenario datasets and providing a closed-loop pipeline from data collection through validation and redeployment for embodied-AI model iteration. (roboticsbusinessnews.com))
