# Viola Zhou and Kinling Lo report in Rest of World on China’s robot-training centers [Canonical reading](https://escapemeridian.com/readings/viola-zhou-and-kinling-lo-report-in-rest-of-world-on-china-s-4ac43576) On January 7, 2026, Viola Zhou and Kinling Lo report in Rest of World that workers in China’s robot-training centers repeatedly act out ordinary movements so humanoid robots can learn from them. Kim, a computer science student working under a pseudonym, calls trainers “cyber-laborers” and says the work is “a bit boring.” The centers, funded by local governments and operated by robotics companies, are meant to address a shortage of movement data that is difficult to obtain from the internet or produce synthetically. The article places this effort within China’s push to develop embodied intelligence and describes how shared training facilities could give smaller startups access to data and equipment they cannot afford themselves. But the scale of the infrastructure does not resolve whether collecting human movements this way is the best route to more capable robots. Researchers are also exploring digital simulation and data from robots already at work. Analyst Marco Wang warns of potential overcapacity, while robotics researcher Ken Goldberg says the approach is slow and its cost effectiveness is unknown. The article ends with Goldberg’s qualified hope: “It’s a noble effort, and there’s a lot of hope right now that this will succeed.” Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source Rest of World · Article Original source date: Jan 7, 2026 (2026-01-07T11:00:00+00:00). Date basis: Date stated on the publisher’s page. Added to Escape Meridian: 2026-11-09T09:33:23.925Z. First observed: 2026-11-09T04:37:40.277Z. Retrieved: 2026-11-09T04:39:34.643Z. - [In Chinese data factories, workers teach humanoid robots boring tasks](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Limits Text as retrieved on November 9, 2026; originally published January 7, 2026. Kim is identified by a pseudonym, and the article says he is not authorized to speak to the media. The article says researchers are debating the best approach; it does not establish whether the centers produce data that improves robot capability or is cost-effective. ## Open question Can repeated human demonstrations produce enough useful movement data, at a cost and pace that make these centers worthwhile? ## Source-stated Viola Zhou and Kinling Lo report that trainers repeatedly perform everyday movements to generate data for humanoid robots. > At the facilities, human trainers mimic movements like folding clothes and wiping the table to teach the robots. - [Supporting source](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Source-stated Kim describes the work as repetitive and somewhat boring, while calling trainers “cyber-laborers.” > “We call ourselves cyber-laborers,” Kim told Rest of World, speaking under a pseudonym because he is not authorized to talk to the media. “It’s a fine gig, though a bit boring.” - [Supporting source](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Source-stated The article says robotics data is more complex than the data used to train large language models, and cannot easily be scraped from the internet or produced synthetically. > But robotics requires far more complex data sets — including visual information, joint motions, rotations — which cannot be easily scraped from the internet or produced synthetically. - [Supporting source](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Source-stated The article reports that researchers are debating whether large-scale human movement data collection is the best way to build fully intelligent robots. > Researchers are still debating if large-scale data collection is the most effective way to build fully intelligent robots. - [Supporting source](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Source-stated The article says shared facilities may help smaller startups, while reporting a concern that the infrastructure could lead to overcapacity. > While these facilities can provide data for smaller startups that cannot afford their own training space, the infrastructure buildup also risks causing overcapacity, Marco Wang, an analyst with Interact Analysis, told Rest of World. - [Supporting source](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Source-stated Goldberg says the cost effectiveness of the data factories is unknown and that producing enough data will take a long time, even with many workers. > The cost effectiveness of China’s data factories is unknown, Ken Goldberg, a robotics researcher at the University of California, Berkeley, told Rest of World. - [Supporting source](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Source-stated Goldberg expresses hope about the effort while emphasizing its slow pace. > “It’s a noble effort, and there’s a lot of hope right now that this will succeed,” Goldberg said. “But it’s slow. Even if you have hundreds of people working, it’s going to take a long time to get enough data.” - [Supporting source](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Source and contributor context Rest of World · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Rest of World Rest of World reports technology and its social effects outside a US-only frame. - [Explore this context](/entities/rest-of-world) - [Reference](https://restofworld.org/) - [Reference](https://restofworld.org/feed/latest/rss.xml) ## Editorial provenance Version 1. Publication ID: 6729419e-9e57-e595-e738-42e5971e17a6. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-09T09:29:00.310Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model. Observation: c4a54721aa300ff3a810d01d744375febe02c82b2480bf3b7c1b5d818ae4e4c0. Source text hash: a8959dd9fe4e346de138c211c3e151b5f28f8b29f02ba1c474917a3fa716cb26. ## Revision and source-check history ### Version 1 — 2026-11-09T09:33:23.925Z On January 7, 2026, Viola Zhou and Kinling Lo report in Rest of World that workers in China’s robot-training centers repeatedly act out ordinary movements so humanoid robots can learn from them. Kim, a computer science student working under a pseudonym, calls trainers “cyber-laborers” and says the work is “a bit boring.” The centers, funded by local governments and operated by robotics companies, are meant to address a shortage of movement data that is difficult to obtain from the internet or produce synthetically. The article places this effort within China’s push to develop embodied intelligence and describes how shared training facilities could give smaller startups access to data and equipment they cannot afford themselves. But the scale of the infrastructure does not resolve whether collecting human movements this way is the best route to more capable robots. Researchers are also exploring digital simulation and data from robots already at work. Analyst Marco Wang warns of potential overcapacity, while robotics researcher Ken Goldberg says the approach is slow and its cost effectiveness is unknown. The article ends with Goldberg’s qualified hope: “It’s a noble effort, and there’s a lot of hope right now that this will succeed.” ## Original source Rest of World · Article Original source date: Jan 7, 2026 (2026-01-07T11:00:00+00:00). Date basis: Date stated on the publisher’s page. Added to Escape Meridian: 2026-11-09T09:33:23.925Z. First observed: 2026-11-09T04:37:40.277Z. Retrieved: 2026-11-09T04:39:34.643Z. - [In Chinese data factories, workers teach humanoid robots boring tasks](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Limits Text as retrieved on November 9, 2026; originally published January 7, 2026. Kim is identified by a pseudonym, and the article says he is not authorized to speak to the media. The article says researchers are debating the best approach; it does not establish whether the centers produce data that improves robot capability or is cost-effective. ## Open question Can repeated human demonstrations produce enough useful movement data, at a cost and pace that make these centers worthwhile? ## Source-stated Viola Zhou and Kinling Lo report that trainers repeatedly perform everyday movements to generate data for humanoid robots. > At the facilities, human trainers mimic movements like folding clothes and wiping the table to teach the robots. - [Supporting source](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Source-stated Kim describes the work as repetitive and somewhat boring, while calling trainers “cyber-laborers.” > “We call ourselves cyber-laborers,” Kim told Rest of World, speaking under a pseudonym because he is not authorized to talk to the media. “It’s a fine gig, though a bit boring.” - [Supporting source](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Source-stated The article says robotics data is more complex than the data used to train large language models, and cannot easily be scraped from the internet or produced synthetically. > But robotics requires far more complex data sets — including visual information, joint motions, rotations — which cannot be easily scraped from the internet or produced synthetically. - [Supporting source](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Source-stated The article reports that researchers are debating whether large-scale human movement data collection is the best way to build fully intelligent robots. > Researchers are still debating if large-scale data collection is the most effective way to build fully intelligent robots. - [Supporting source](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Source-stated The article says shared facilities may help smaller startups, while reporting a concern that the infrastructure could lead to overcapacity. > While these facilities can provide data for smaller startups that cannot afford their own training space, the infrastructure buildup also risks causing overcapacity, Marco Wang, an analyst with Interact Analysis, told Rest of World. - [Supporting source](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Source-stated Goldberg says the cost effectiveness of the data factories is unknown and that producing enough data will take a long time, even with many workers. > The cost effectiveness of China’s data factories is unknown, Ken Goldberg, a robotics researcher at the University of California, Berkeley, told Rest of World. - [Supporting source](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Source-stated Goldberg expresses hope about the effort while emphasizing its slow pace. > “It’s a noble effort, and there’s a lot of hope right now that this will succeed,” Goldberg said. “But it’s slow. Even if you have hundreds of people working, it’s going to take a long time to get enough data.” - [Supporting source](https://restofworld.org/2026/china-robots-training-centers-workers/) ## Source and contributor context Rest of World · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Rest of World Rest of World reports technology and its social effects outside a US-only frame. - [Explore this context](/entities/rest-of-world) - [Reference](https://restofworld.org/) - [Reference](https://restofworld.org/feed/latest/rss.xml) ## Editorial provenance Version 1. Publication ID: 6729419e-9e57-e595-e738-42e5971e17a6. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-09T09:29:00.310Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model. Observation: c4a54721aa300ff3a810d01d744375febe02c82b2480bf3b7c1b5d818ae4e4c0. Source text hash: a8959dd9fe4e346de138c211c3e151b5f28f8b29f02ba1c474917a3fa716cb26. ### uncertain — 2026-11-09T09:29:00.310Z — anthropic/claude-opus-5.5 Publication: 6729419e-9e57-e595-e738-42e5971e17a6 attribution — pass: The reading correctly attributes the article to Viola Zhou and Kinling Lo at Rest of World. It attributes quotations to Kim (pseudonym), Marco Wang of Interact Analysis and Ken Goldberg of UC Berkeley, as the source does. This is one reported article, not a listing of other pieces. > By Viola Zhou and Kinling Lo summary — uncertain: The reading names embodied intelligence as the policy frame but never explains it, and the definitions list is empty. The source glosses it as AI in physical form; a short parenthetical would fix this. Separately, the closing line says the article ends with Goldberg's hope quote. The article actually ends on his caveat, but it's slow and will take a long time. Calling the hope qualified partly covers this, but ending on the caveat or quoting both sentences would be more faithful. > embodied intelligence — or artificial intelligence in the physical form — as a priority for the country date — pass: The reading's January 7, 2026 matches originalPublished (2026-01-07T11:00Z) and the article's dateline. The evidence is a publisher page, so no archive disclosure is needed. The modifiedAt timestamp precedes publication and does not indicate a later edit. The reading uses no hindsight framing, and it does not present the Goldman Sachs forecast or the Unitree listing plan as outcomes. > 7 January 2026 support — pass: The claims are source-stated and match their excerpts. The summary's mention of equipment slightly extends the source's phrase about training space, but it does not materially change the meaning. Goldberg's cost-effectiveness point and Wang's overcapacity concern are attributed correctly. The question follows from the source's stated debate and from Goldberg's remarks on pace and cost. > While these facilities can provide data for smaller startups that cannot afford their own training space media — pass: No media was selected and no verified candidates were supplied, so there is no media identity or relevance issue to check. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/4ac435765e7e39384b71a6e7c3e504eef64cd1e27929dc2f5e9daf3ebdd9b7d6/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/viola-zhou-and-kinling-lo-report-in-rest-of-world-on-china-s-4ac43576.json)