What happened: China’s largest humanoid-robot data-training centre has opened in Beijing’s Shijingshan district, with nearly 200 human trainers teaching around 100 humanoid robots everyday tasks in staged “real-world” scenarios. Yes, it’s basically robot school, except the students can bench-press you.

Why it matters: The centre is explicitly about building robot “brains” with high-quality multimodal data, including failure cases. Leju Robotics says it can generate more than 8 million records a year and even supply data to AI model companies, because nothing says “progress” like industrial-scale homework.

Wider context: The piece underlines a recurring reality in humanoids: the body is getting good enough to ship, but robust behaviour in messy environments is still the bottleneck. The industry is drifting from flashy gala backflips toward pipelines, supervision, and repeatable training, aka the boring stuff that actually works.

Background: The facility was co-built with the local government and uses 16 scenario types spanning manufacturing, smart home, eldercare, and more. Researchers cited in the article stress the gap between rehearsed demos and real factories or homes, and note safety concerns when robots weigh over 220 pounds.

Do androids dream of domestic feats? — The Independent

Droid Brief Take: If you’re wondering where “autonomous humanoids” actually come from, it’s not magic, it’s a small army of humans patiently teaching robots not to drop cups. The future arrives via supervision, data, and a lot of quiet embarrassment.

Key Takeaways:

• Data Factory: Leju’s training centre reportedly produces over 8 million multimodal records annually, treating failure data as a first-class ingredient so robots can learn recovery, not just perfect rehearsals.
• Teleop Scaffolding: Trainers wear headsets and demonstrate motions while robots mirror them, highlighting how much “embodied intelligence” currently rides on human-guided behaviour collection before autonomy can pretend it arrived alone.
• Deployment Reality Check: The article says robots from the centre have started factory work at around 70% of a skilled worker’s efficiency on standardised tasks, while home use remains riskier given robot mass and unpredictable environments.