Humanoid learning-data collection: the data-factory lane

status: open, opened 2026-10-15, last moved 2026-10-23. Robots humanoids funding

• The thread: the humanoid-robot learning-data supply chain as its own industry — physical data-collection facilities, worker-collected task data (the gig-economy layer the tick flagged).

• 2026-10-16 (~4:45 AM ET, note.com): the lane's first dedicated Japanese facility — J-HRTI Kanto Data Factory reported to begin full-scale operation (presentation dated Aug-26). Watch: the operator behind J-HRTI, collection scale, which humanoids consume the data. https://note.com/tsumura_official/n/n980ef711d37e

• 2026-10-18 (two items, ~4:46 AM ET): (1) Global Times centers a feature on Chinese robotics firms rolling out components and data systems around the Humanoid Robot Data Training Center in Beijing's western Shijingshan District — the lane's first named mainland-China physical facility, with dozens of robots photographed on its floor (https://www.globaltimes.cn/page/202610/1372460.shtml). (2) MSN Money syndication puts robot trainers' top pay at ₹85,000/month (roughly $1,000 USD), an upgraded figure over the Oct 15 record of 4,000–6,000 yuan (top end ~$900 at the prior outlet's conversion) for chore-repetition training work (https://www.msn.com/en-in/money/topstories/humans-are-now-training-robots-by-doing-household-chores-in-china-the-pay-can-go-up-to-85000-a-month/ar-AA2etH3E). Watch: who operates the Shijingshan center, which humanoids consume its data, whether the ₹85,000 figure is a verified tier or headline rounding.

• 2026-10-18 (editorial fold): merged a duplicate thread — the lane thread's scope already claimed the gig-economy layer; the BI feature and the 4,000–6,000-yuan pay-rate first development now live on the survivor. Its dated entries follow, verbatim:

• 2026-10-15 (first record): Business Insider's feature on the race to train robots on human work: thousands of workers recording everyday tasks as robot training data (africa.businessinsider.com, ~8:53 AM ET; main-site version ~8:54 AM ET), including a Sydney startup paying workers to wear AI glasses that collect task data (MSN syndication, ~8:52 AM ET) and a private chef recording a three-course meal — the capture's first private-culinary training scenario (burrata salad, baked salmon, tiramisu). New human-worn gig-collection lane beside the data-layer ticks on file: DexForce's sim training (October 15), Config Intelligence's real-world action hours (October 14), the Philippines' planned physical-AI data factory (October 14). No startup name in capture. Watch: the Sydney startup's identity, pay rates, whose humanoids consume the data, labor-framing backlash. https://africa.businessinsider.com/news/inside-the-race-to-train-robots-on-human-work/g1njf6z

• 2026-10-15 (~12:13 PM ET, RobotsBeat): the tick's first development, hours after it opened — the China leg with the record's first pay-rate data: worker accounts reveal robot trainers repeating chores 200 times a day for reported monthly pay of 4,000–6,000 yuan, a range the outlet's own conversion tops out at about $900 a month. Answers the tick's pay-rate watch item directly and adds the second geography beside the Sydney AI-glasses scene; the Sydney startup's name and rates remain open. Watch: which companies employ the trainers, whose humanoids consume the data, whether Western programs' rates land in or above the yuan band, labor-framing backlash. https://robotsbeat.com/china-robot-trainers-chores-200-times-900-month/

• 2026-10-22 (~4:46 PM ET, Forbes — John Koetsier): the lane's first fresh raise in this window — a robotics startup has raised $30 million to attack the humanoid-robot training-data shortage, with the piece framed around giving robots more training data to make them smart (https://www.forbes.com/sites/johnkoetsier/2026/10/22/theres-not-enough-robot-training-data-this-startup-plans-to-fix-that/). Development of the lane's facility-and-method records (DexForce's sim training, Beijing's Shijingshan data-training center, J-HRTI's Kanto facility, the human task-data workforce): a capital layer aimed at the data shortage itself. The company name and the collection-versus-synthetic split are not recoverable from capture. Watch: the startup's identity, whose humanoids consume its data.

• 2026-10-23 (~5:05 PM ET, RobotsBeat): the identity watch answers — Forbes's October 22 "$30 million to attack the humanoid-robot training-data shortage" startup is Maxinsights, raising a $30 million seed round led by Peak XV, with two named products: MaxLoop for training-data quality control and MaxGlove, a glove capturing touch and force from everyday work (https://robotsbeat.com/maxinsights-30m-seed-robot-training-data-maxloop-maxglove/). Watch: whose humanoids consume Maxinsights' data, the collection-versus-synthetic split, and whether MaxGlove's human-worn capture joins the lane's gig-collection pay-rate records.