Robot training data in United States

35 companies with workers here, 26 datasets collected here and 64 sourced events. Reported robot-data pay runs $20 to $48 an hour, 2.8× to 6.6× the local minimum wage.

Work types

Minimum wage

• $1,257 a month, about $7.25 an hour (ILOSTAT, 2026)

Pay reported

Companies with workers here

Datasets collected here

• BridgeData V2 60,096 trajectories, 24 environments arxiv.org ↗ Major outlet
• RT-1 robot dataset 130k+ episodes, 700+ tasks, 17 months github.io ↗ Major outlet
• Ego4D 3,670 hours, 74 locations meta.com ↗ Secondary
• NVIDIA PhysicalAI GR00T X-Embodiment Sim About 423,000 simulated trajectories huggingface.co ↗ Major outlet
• Physical Intelligence pi0 pretraining data 10k+ hours github.com ↗ Major outlet
• EgoDex 829 hours, 194 tasks arxiv.org ↗ Major outlet
• DoorDash Tasks chore filming bgov.com ↗ Major outlet
• Kiwibot remote piloting from Colombia thehustle.co ↗ Major outlet
• Tesla Optimus camera-rig data collection eweek.com ↗ Secondary
• Scale AI physical AI data factories 1000+ hours of demonstration data per day (company claim) scale.com ↗ Secondary
• Project Go-Big human video figure.ai ↗ Secondary
• PhysicalAI-Robotics-GR00T-X-Embodiment-Sim incl. 240k humanoid tabletop trajectories, 72K kitchen arm trajectories huggingface.co ↗ Major outlet
• Open-H-Embodiment 750 hours; 120,000 trajectories huggingface.co ↗ Major outlet
• GR00T-Dreams synthetic trajectories nvidia.com ↗ Secondary
• ABC (XDOF x Berkeley AI Research) 130,000 trajectories; 300 h sim; 100 h evals techcrunch.com ↗ Major outlet
• Appen egocentric robotics video ~60,000 hours/month capacity abc.net.au ↗ Major outlet
• Tesla Optimus helmet-camera motion data 4+ usable hours per worker per shift aol.com ↗ Secondary
• Apptronik Robot Park data yahoo.com ↗ Secondary
• Humanoid Everyday 10.3k trajectories; 3M+ frames; 260 tasks arxiv.org ↗ Major outlet
• RT-1 robot data arxiv.org ↗ Major outlet
• BEHAVIOR-1K 1,000 activities; 50 scenes; 9,000+ objects arxiv.org ↗ Major outlet
• EgoMimic arxiv.org ↗ Major outlet
• Waymo Open Motion Dataset 100,000+ scenes; 570+ hours; 1750 km arxiv.org ↗ Major outlet
• WOD-E2E 4,021 segments; about 12 hours arxiv.org ↗ Major outlet
• Waymo Open Dataset (Perception) 1150 scenes x 20 s arxiv.org ↗ Major outlet
• ABC (XDOF x BAIR) 130,000 trajectories; 300 h sim; 100 h evals techcrunch.com ↗ Major outlet

Events

• 2026-10 Scale AI sells physical AI demonstration data
  Scale runs data factories, home collectors and a robotless egocentric harness (undated page, fetched Oct 2026). scale.com ↗ Secondary
• 2026-09-30 Innodata opens a motion-capture lab for humanoid data
  Listed data vendor moves from LLM data into physical-AI motion capture. therobotreport.com ↗ Secondary
• 2026-09-27 Pakistan's World Context records 8,000 factory workers
  Wearable-glove and head-camera data sold to US and Chinese robotics firms. arabnews.pk ↗ Secondary
• 2026-09-20 Appen collects 60,000 hours of egocentric video a month
  Up to 5,000 contributors worldwide record first-person video for robot training. abc.net.au ↗ Major outlet
• 2026-09-18 OpenAI posts robotics jobs incl. data collection facilities
  27 robotics openings, pay up to $500K base. businessinsider.com ↗ Major outlet
• 2026-09-11 Mecka AI nears $500M valuation
  Sequoia-led round for a firm paying people to record everyday tasks. techcrunch.com ↗ Major outlet
• 2026-09-04 XDOF in talks at $1.2B valuation
  Teleop and egocentric data vendor plans global operator workforce. techcrunch.com ↗ Major outlet
• 2026-09 Figure Helix 2.5: Index pretraining lifts zero-shot home success from 9% to 56%
  Controlled test in 30 unseen homes credits Index human-video pretraining. figure.ai ↗ Secondary
• 2026-08-25 Figure launches Index crowdsourced robot-data app
  Creators in 108 countries upload task video; $15M paid; $1B+ committed to data and compute. figure.ai ↗ Secondary
• 2026-08-20 Micro1 hits $500M gross run rate
  Human-data vendor whose recorders include Nigerians and Argentines. techcrunch.com ↗ Major outlet
• 2026-08-04 Scale AI agrees $12.5M misclassification settlement
  McKinney v. Scale AI settles; fund covers California contributors from Dec 2020 to Feb 2026 (date is the report date). kingsiegel.com ↗ Secondary
• 2026-07-29 The Bot Company hit with class action over unpaid contract-worker wages
  Class status granted June 29; suit names Botco and staffing agencies. yahoo.com ↗ Secondary
• 2026-07-17 microagi raises $55M, Germany's largest seed
  Head-cam task data licensed to frontier AI labs. yahoo.com ↗ Secondary
• 2026-07-17 Agility opens Fremont humanoid training facility
  60,000 sq ft site where Digit learns new skills before field deployment. techcrunch.com ↗ Major outlet
• 2026-06-30 Apptronik opens 90,000 sq ft Robot Park data factory with Google DeepMind
  Apollo 2 fleets generate teleop and autonomous data for Gemini Robotics. thestar.com.my ↗ Secondary
• 2026-06-25 BitRobot, Hugging Face and Unitree release HIW-500 in-the-wild humanoid dataset
  500+ hours of G1 whole-body teleoperation in real Southeast Asian homes. zhidx.com ↗ Secondary
• 2026-06-17 XDOF launches as a robot-data foundry with 20 customers
  Teleoperation and body-sensor data collection for frontier labs it cannot name. techcrunch.com ↗ Major outlet
• 2026-06-17 XDOF exits stealth with $70M for robot data pipelines
  Berkeley GELLO team builds teleop, egocentric and annotation services for frontier labs. techcrunch.com ↗ Major outlet
• 2026-05-26 Human Archive deploys 1,000+ headsets in India
  YC-backed startup raises $8.2M; consent and revenue terms for workers undisclosed. aiweekly.co ↗ Secondary
• 2026-05-01 Meta acquires humanoid startup ARI
  Meta folds a humanoid foundation-model team into Superintelligence Labs. techcrunch.com ↗ Major outlet
• 2026-04-16 Instawork launches Robotics Lab and robot-data certification
  Hourly-work marketplace pivots to selling wearable-captured robot data. businessinsider.com ↗ Major outlet
• 2026-03-27 Physical Intelligence in talks to raise $1B
  Bloomberg-sourced report of a round valuing the robot-brain startup above $11B. techcrunch.com ↗ Major outlet
• 2026-03-19 DoorDash launches Tasks app
  Couriers paid to film chores for AI and robotics data. techcrunch.com ↗ Major outlet
• 2026-02-06 Open-H-Embodiment released
  Open-H-Embodiment: 750 hours; 120,000 trajectories; 4.5 TB huggingface.co ↗ Major outlet
• 2026-02 Waymo admits Philippines-based remote operators
  US senators objected to overseas remote assistance for robotaxis. businessinsider.com ↗ Major outlet
• 2026-01-14 Skild AI raises $1.4B Series C
  SoftBank-led round values Skild at over $14B; pretraining leans on human video. skild.ai ↗ Secondary
• 2026-01-13 1X releases world model so NEO learns from video
  1X says NEO can now learn from internet-scale video. techcrunch.com ↗ Major outlet
• 2025-11-02 Inside Tesla's Optimus data-collection lab
  Business Insider reports 8-hour shifts in heavy camera backpacks, with injuries and cuts. aol.com ↗ Secondary
• 2025-11-02 Micro1 pays people in Brazil, Argentina, India and US to wear recording glasses
  Smart-glasses capture of daily actions for humanoid training. latimes.com ↗ Major outlet
• 2025-11-02 Business Insider reports on Tesla Optimus data collection operators
  Workers in five-camera helmets repeat motions hundreds of times per 8-hour shift. aol.com ↗ Secondary
• 2025-10-30 WOD-E2E released
  WOD-E2E: 4,021 segments; about 12 hours arxiv.org ↗ Major outlet
• 2025-10-28 1X opens NEO preorders with remote 'Expert' teleoperation
  Owners must accept that a remote 1X operator may see inside their homes to train NEO. engadget.com ↗ Secondary
• 2025-10-09 Humanoid Everyday released
  Humanoid Everyday: 10.3k trajectories; 3M+ frames; 260 tasks arxiv.org ↗ Major outlet
• 2025-09-18 Figure launches Project Go-Big
  Figure says it trained robot navigation only on egocentric human video filmed in Brookfield homes. figure.ai ↗ Secondary
• 2025-09-18 Figure launches Project Go-Big with Brookfield homes
  Figure collects egocentric human video in Brookfield properties; Helix learns navigation from human video only. figure.ai ↗ Secondary
• 2025-09-17 Figure and Brookfield partner on pretraining data
  Brookfield homes, offices and logistics sites become venues for humanoid data capture. figure.ai ↗ Secondary
• 2025-09-16 Figure raises $1B+ Series C at $39B
  Humanoid maker earmarks funds for human video and multimodal data collection. figure.ai ↗ Secondary
• 2025-09-12 xAI lays off 500 data annotators
  About a third of the Grok annotation team cut in a pivot to specialist tutors. techcrunch.com ↗ Major outlet
• 2025-08-26 Tesla moves Optimus data collection to camera rigs
  Business Insider reports Optimus collectors switch from mocap suits and teleoperation to helmet and backpack camera rigs. eweek.com ↗ Secondary
• 2025-08 GlobalLogic lays off 200+ Google AI raters
  Gemini and AI Overviews raters cut in two rounds amid pay and organizing disputes. allwork.space ↗ Secondary
• 2025-07-16 Scale AI cuts 14% of staff and 500 contractors
  Interim CEO says the data-labeling business overextended after the Meta deal. storyboard18.com ↗ Secondary
• 2025-06-12 Meta takes 49% of Scale AI; Google plans exit
  Meta invests $14.3B and hires Wang; Google, Scale's largest customer, moves to cut ties. techcrunch.com ↗ Major outlet
• 2025-06-04 Amazon builds 'humanoid park' test course in San Francisco
  The Information reports Amazon will test humanoid delivery robots on an indoor obstacle course. thestar.com.my ↗ Secondary
• 2025-05-19 Surge AI misclassification suit
  Alleged contractor misclassification and docked pay. aol.com ↗ Secondary
• 2025-05-18 NVIDIA unveils GR00T-Dreams synthetic trajectory blueprint
  Cosmos world models generate robot training trajectories; N1.5 data made in 36 hours. nvidia.com ↗ Major outlet
• 2025-03-18 NVIDIA releases GR00T N1 and open physical AI dataset
  Open humanoid foundation model plus training data published on Hugging Face. nvidia.com ↗ Major outlet
• 2025-03-18 GR00T X-Embodiment Sim released
  GR00T X-Embodiment Sim: 240k humanoid tabletop + 72K kitchen arm + 9k bimanual + 102 G1 trajectories huggingface.co ↗ Major outlet
• 2025-02-14 Meta sets up humanoid robotics group in Reality Labs
  Internal memo names ex-Cruise CEO Marc Whitten to lead consumer humanoid R&D. thestar.com.my ↗ Secondary
• 2024-12-10 McKinney class action against Scale AI
  Suit in San Francisco alleges contractor misclassification and unpaid work. sfexaminer.com ↗ Major outlet
• 2024-10-31 Physical Intelligence publishes pi0
  Vision-language-action model trained on cross-robot dexterous data (laundry, table cleaning, boxes). arxiv.org ↗ Major outlet
• 2024-10-31 EgoMimic released
  EgoMimic: egocentric human video + 3D hand tracking, co-trained with robot data arxiv.org ↗ Major outlet
• 2024-08-30 Amazon hires Covariant founders, licenses its robot foundation models
  Amazon took Covariant's three founders and about a quarter of staff plus a non-exclusive model license. siliconangle.com ↗ Secondary
• 2024-07-09 Skild AI raises $300M Series A
  Robot foundation-model startup raises a large Series A. skild.ai ↗ Secondary
• 2024-05-22 97 Kenyan AI workers write to President Biden
  Labelers and moderators publish an open letter on pay and trauma ahead of Ruto's state visit. foxglove.org.uk ↗ Major outlet
• 2024-03-19 DROID released
  DROID: 76k trajectories; 350 hours; 564 scenes; 84 tasks arxiv.org ↗ Major outlet
• 2024-03-14 BEHAVIOR-1K released
  BEHAVIOR-1K: 1,000 activities; 50 scenes; 9,000+ objects arxiv.org ↗ Major outlet
• 2024-03 DROID robot manipulation dataset released
  76k teleoperated trajectories gathered by 50 collectors across 13 institutions over 12 months. github.io ↗ Major outlet
• 2024-01-20 Google terminates Appen contract
  Google cancels Appen's rater contract (26% of Appen revenue), services ending 19 March 2024. searchenginejournal.com ↗ Secondary
• 2023-08-24 BridgeData V2 released
  BridgeData V2: 60,096 trajectories; 24 environments arxiv.org ↗ Major outlet
• 2022-12-13 Google RT-1 trained on 130k robot episodes
  Robotics Transformer trained on a 17-month, 13-robot fleet data collection covering 700+ tasks. research.google ↗ Major outlet
• 2022-12-13 RT-1 robot data released
  RT-1 robot data: real-robot demonstrations arxiv.org ↗ Major outlet
• 2021-10-14 Meta and university partners release Ego4D
  Over 2,200 hours of first-person video from over 700 participants in nine countries, released for embodied perception research. meta.com ↗ Secondary
• 2021-04-20 Waymo Open Motion Dataset released
  Waymo Open Motion Dataset: 100,000+ scenes; 570+ hours; 1750 km arxiv.org ↗ Major outlet
• 2019-12-10 Waymo Open Dataset (Perception) released
  Waymo Open Dataset (Perception): 1150 scenes x 20 s arxiv.org ↗ Major outlet

On-the-ground evidence

• Tesla (Optimus data collection operators)wear a motion capture suit and virtual reality headset while performing designated movementsdecrypt.co ↗ Major outlet
• Figure AI (data creators / robot operators)Figure's current postings advertise $30 to $32 an hour for data-collection and robot-operation roles, with its Helix Data Creator night shift listed at $33 an hour.christianandtimbers.com ↗ Secondary
• DoorDash Taskshands washing at least five dishes while wearing a body cameratechcrunch.com ↗ Major outlet
• Scale AI, Encord (data recorders)Data companies like Scale AI and Encord are recruiting their own armies of data recorderstechnologyreview.com ↗ Major outlet
• xAI (AI tutors)about one-third of xAI's 1,500-person data-annotation teamtechcrunch.com ↗ Major outlet

Every row links to its source, and hovering a link shows the quote it rests on. Research compiled 2026-10-05. Method