Skip to content

Research · October 2026 edition

Who Is Buying Robot Training Data: Deals, Paid Programs and Dataset Releases

From 1 Jan 2023 to 28 Oct 2026, 88 deals, paid programs and dataset releases for robot, embodied-AI, AV and drone training data are on record, involving 57 buyers.

Read the month's story: Robot Training Jobs in October 2026: The Month in Data

Not final yet. The month has ended, yet stragglers in the data may still land up to 3 Nov 2026; after that the edition is locked permanently.

Covers
3 Oct 2023 to 28 Oct 2026
Dated by when each event occurred
Calculated
1 Nov 2026, 06:05 UTC
Settles when locked
Where it comes from
Company announcements, filings and reporting by major outlets, each linked beside its record with a verbatim quote kept in the dataset.
Method
Revision 1. Terms defined
CC BY 4.0, citing this report

Each figure here comes as CSV and JSON files, emailed to your work address.

Need robot training data collected? We run the project: teleoperation sessions, first-person video, filmed household tasks or labeled robot video, staffed from the people who apply for this work through us, checked against your standard, starting with a paid pilot.

How we run projects

What the data shows

  1. 1

    Internet majors and embodied-AI body makers (unnamed) appears most often as a buyer (11 records: mixed, egocentric video and simulation).

    Figure 2
  2. 2

    14 programs pay members of the public directly to record or produce robot training data, the latest launched 10 Oct 2026 (Mifeng Technology (觅蜂科技, an embodied-AI data services company)).

    Figure 1
  3. 3

    23 of 88 records state an amount; the rest do not disclose what was paid.

    Figure 1
  4. 4

    2026 has the most records so far (62).

    Figure 3

Figure 1

Deals, paid programs and dataset releases

Who has paid for, licensed, commissioned or released robot training data, from whom, and what kind?

BuyerFromKind of dealDate
Genesis AIThe Next WebMaxinsightsContractEgocentric video23 Oct 2026Maxinsights said it supplied all 200,000 hours of first-person data behind GENE-26.5, the robot model from Genesis AI, and is developing the MaxGlove force-sensing glove with Genesis AI.Not disclosed
Dyna RoboticsThe Next WebMaxinsightsContractEgocentric video23 Oct 2026Maxinsights describes itself as the largest data provider behind Dyna-2, the robot foundation model from Dyna Robotics.Not disclosed
EncordWearable Devices Ltd. (GlobeNewswire press release, via Yahoo Finance)Wearable Devices Ltd.PartnershipMixed22 Oct 2026Wearable Devices signed a strategic collaboration agreement with Encord (Cord Technologies, Inc.) under which Encord may purchase Wearable Devices' wrist-worn neural sensing devices and use the data they collect, with device sales and usage-based compensation to Wearable Devices for qualifying data-collection activity.Not disclosed
Internet majors and embodied-AI body makers (unnamed)China Securities Journal (via 10jqka)Noitom Robotics (诺亦腾机器人)ContractMixed19 Oct 2026Noitom Robotics' market and PR director told China Securities Journal that internet majors and embodied-AI body makers are on the list of customers buying embodied-AI data, and that single-customer data orders in the industry have grown from tens of thousands of hours toward a million hours.Not disclosed
Industry customers of the Beijing humanoid data base (unnamed)China Securities Journal (via 10jqka)Beijing Innovation Center of Humanoid RoboticsContractMixed10 Oct 2026The head of the Beijing Innovation Center of Humanoid Robotics' embodied-AI data and training base in Yizhuang said it has delivered nearly 30,000 hours of high-quality data to outside customers, with more than 70 percent of its capacity serving industry customers.nearly 30,000 hours delivered externally
Mifeng Technology (觅蜂科技, an embodied-AI data services company)China Securities Journal (via 10jqka)Device holders collecting data through partner Jiahe Shengxin (嘉合盛鑫)Paid programEgocentric video10 Oct 2026A partner at Jiahe Shengxin, a collaborator of Shanghai Mifeng Embodied Intelligence Technology (觅蜂科技), told China Securities Journal that each device holder who completes data collection tasks earns 10 to 30 yuan per hour of data collected.10 to 30 yuan per hour collected (CNY)
unnamed frontier robotics labs and Big Tech firmsBetaKitMecka AIContractEgocentric video7 Oct 2026Mecka AI, which pays people to record everyday tasks with body sensors and phones, said it converts that first-person data into training data for multiple frontier robotics labs and Big Tech firms, and claimed a revenue run-rate above 100 million dollars in June 2026.surpassed a $100-million USD run-rate in June
micro1 (for an unnamed customer project)AlignList (micro1 job posting)public contributorsPaid programAV driving data4 Oct 2026micro1 posted a paid program for contributors with a 360 degree action camera to record their vehicles navigating real-world roads from a trailing third-person view for a customer project, paying 15 to 20 US dollars per hour per approved video.$15-20 USD/hour per approved video
UK Ministry of Defence (for up to 12 UK companies)GOV.UK (Ministry of Defence)Ukraine Ministry of Defence (Avengers AI Labs)PartnershipDrone data25 Sep 2026The UK became the first international partner to access Ukraine's Avengers AI Labs database of drone imagery and video, and the UK Ministry of Defence opened a competition giving up to 12 UK companies access to train AI models for drone swarms.Not disclosed
Frontier AI labs, Microsoft, Amazon and robotics companies like 1XForbes AustraliaMicro1ContractMixed22 Sep 2026Forbes Australia reports that micro1 counts frontier AI labs, Microsoft, Amazon and robotics companies like 1X as customers, and that micro1 has also entered the robotics training data business (the article does not say which customer buys which kind of data).Not disclosed
Kinetic Blocks marketplace buyers (humanoid model developers)Robotics & Automation Newsdata suppliers listing on Kinetic BlocksMarketplaceMixed1 Sep 2026Oslo startup Kinetic Blocks opened a gated beta of a two-sided marketplace where suppliers list humanoid robot training data with commercial licence terms and approved buyers training humanoid models purchase and download it; sellers set prices and the platform takes a margin.Not disclosed
Figure AIFigure AIIndex Creators (members of the public paid to film tasks at home and at work)Paid programEgocentric video25 Aug 2026Figure AI came out of stealth with Index, an app that pays people to record real household and workplace tasks to train its Helix robot model, and said its Creators have earned 15 million dollars to date and that it is committed to spend over 1 billion dollars on data and compute over the next 12 months.$15M earned by Creators to date; commitment to spend over $1B on data and compute in the next 12 months
LG ElectronicsThe Investor (The Korea Herald)NVIDIA (technology partner: data augmentation and synthetic data tools; LG robots collect the data in-house)PartnershipMixed18 Aug 2026LG Electronics said it expects to collect 100,000 hours of humanoid robot training data with its CLOiD robots at its Yangjae data factory in Seoul by the end of 2026, with the data processed through NVIDIA's data augmentation and synthetic data tools under the companies' expanded physical AI partnership.Not disclosed
Ukrainian defense companiesMinistry of Defence of UkraineUkraine Ministry of Defence (Avengers Labs)LicenceDrone data10 Aug 2026Ukraine's Ministry of Defence opened Avengers Labs, a platform built on 5 million annotated battlefield frames mostly from the DELTA combat system, to Ukrainian defense companies for training their own AI models for unmanned systems; the first companies signed licensing agreements with the ministry. Defence Minister Fedorov had announced in March 2026 that battlefield data would be opened to allies for drone AI training.Not disclosed
RekaRekaPaid collectors recording their own routines through ClaruDataset releaseEgocentric video6 Aug 2026Reka released RekaDaily-10k, more than 10,000 hours of unscripted first-person household video recorded by paid collectors through Claru, Reka's data engine, under an Apache 2.0 licence.Not disclosed
Leading robotics companies (unnamed)Innodata (Accesswire press release, via FinancialContent)InnodataContractEgocentric video6 Aug 2026Innodata said in its second-quarter 2026 results that it ran successful egocentric data-collection pilots with leading robotics companies and is now scoping enterprise-scale multi-modal programs.Not disclosed
黎曼动力 (maker of the Riemann-1.0 embodied world-action model)China News Service (中新网)Lightwheel (光轮智能) and Noitom Robotics (诺亦腾机器人)PartnershipMixed6 Aug 2026黎曼动力 announced strategic partnerships with Lightwheel and Noitom Robotics for high-quality embodied data collection, model training and evaluation, and plans to complete the collection and training of 1 million hours of embodied data by the end of 2026.plan: 1 million hours of embodied data collected and trained on by the end of 2026
Physen AIThe Star (South China Morning Post report)A sanitation company (street cleaners)PartnershipMixed1 Aug 2026Shenzhen-based Physen AI partnered with a sanitation company to fit street cleaners with data-collection devices, capturing their movements as they swept roads to help train robots for real-world environments.Not disclosed
EncordTechCrunchZander LabsPartnershipMixed26 Jul 2026Encord is running a trial with German neuroscience startup Zander Labs, recording brain activity with EEG headsets while its in-house 'pilots' perform manual tasks, to build a brain wave tagged robotics dataset for customers.Not disclosed
More than 20 robotics and foundation model companies (unnamed)Digital News AsiaRopediaContractMixed24 Jul 2026Ropedia says it already serves more than 20 robotics and foundation model companies across North America, China and Singapore (a company claim).Not disclosed
Allen Institute for AI (Ai2)SiliconANGLERopediaContractMixed23 Jul 2026Ropedia's CEO said the company supplied more than half of the data sources used to train Ai2's MolmoMotion model, a 3D motion forecasting model.Not disclosed
Google DeepMindApptronikApptronikPartnershipTeleoperation30 Jun 2026Apptronik opened its expanded Robot Park data collection facility in Austin, where Apollo 2 humanoids generate training data through teleoperation and autonomous runs; under its research partnership the data feeds Google DeepMind's Gemini Robotics models.Not disclosed
BitRobot NetworkHumanoids Dailyhuman teleoperators in 12 homes in Southeast AsiaDataset releaseTeleoperation25 Jun 2026BitRobot Network, with Hugging Face and Unitree Robotics, released HIW-500, a humanoid teleoperation dataset of more than 500 hours and 23,000 episodes recorded on Unitree G1 robots in 12 real homes in Southeast Asia.Not disclosed
unnamed frontier AI labs and robotics companies (about 20 customers)TechCrunchXDOFContractTeleoperation17 Jun 2026Robot data startup XDOF launched with 70 million dollars in funding and said it already sells teleoperation and egocentric training data to about 20 customers, including several frontier AI labs it cannot name.Not disclosed
Enabled IntelligenceDefenseScoopUndisclosed sources of Ukraine war drone footageMarketplaceDrone data16 Jun 2026US data-labeling firm Enabled Intelligence added more than half a million hours of labeled Ukraine conflict drone footage to its EView training-data library, available to approved users in the US, Ukraine and NATO countries; it did not disclose where the footage came from or who its customers are.Not disclosed
InstaworkNewslaundry (The News Minute)Market vendors in BengaluruPaid programEgocentric video16 Jun 2026Instawork paid vendors in Bengaluru's New Thippasandra market to wear head-mounted phone cameras while they worked, generating egocentric footage for robot training; the hourly rate fell from Rs 300 to about Rs 200 within days.Rs 300 per hour at first, later about Rs 200 per hour
星动纪元, 它石智航, 灵心巧手, 零次方, ByteDance, 无界动力 and 章鱼动力 (named as Wuwen Zhixing customers)36Kr (Lieyun, 猎云网)Wuwen Zhixing (无问智科)ContractSimulation16 Jun 202636Kr's Lieyun reports that Wuwen Zhixing, a physical-AI data company, counts 星动纪元, 它石智航, 灵心巧手 and 零次方 as long-term customers, signed orders with ByteDance, 无界动力 and 章鱼动力 in the first quarter of 2026, and has orders on hand worth hundreds of millions of yuan.orders on hand worth hundreds of millions of yuan (CNY)
Leading embodied-AI customers (unnamed)36Kr (Lieyun, 猎云网)Yiren Technology (弈人科技)ContractMixed16 Jun 202636Kr's Lieyun reports that Yiren Technology, which turned its autonomous-driving perception network to embodied-AI data collection, won new embodied-AI data orders worth more than 100 million yuan in the first quarter of 2026 alone from leading customers.new embodied-AI data orders above 100 million yuan (CNY) in Q1 2026
ObjectwayseNCA (AFP)Household and factory workers in IndiaPaid programEgocentric video11 Jun 2026Data company Objectways pays Indian workers to film first-person video of housework and factory tasks through its app and supplied head cameras; a Chennai recorder said she earns 250 rupees an hour.250 rupees an hour
More than 100 Ukrainian defence companiesMinistry of Defence of UkraineUkraine Ministry of Defence (Brave1 Dataroom)LicenceDrone data11 Jun 2026Ukraine's Ministry of Defence said more than 100 Ukrainian companies have gained access to Brave1 Dataroom, a secure environment of structured combat datasets, and are training AI models on it, and that Ukrainian and international companies can train on millions of annotated frames from real combat sorties.Not disclosed
20 core customers of the Wuxi consortium (unnamed)The Paper (澎湃新闻), account of Wuxi Observation (无锡观察)Wuxi city-level embodied-AI dataset consortium (led by Huazhi (Wuxi) AI, a Wuxi Data Group subsidiary)ContractMixed11 Jun 2026At the Wuxi embodied-intelligence supply chain conference, Wuxi announced that its city-level 1,000-company, million-hour dataset consortium has signed 30,000 hours of data orders and 20 core customers, against about 100,000 hours of market data demand it has matched.30,000 hours of data orders signed
A leading domestic embodied-AI maker in China (unnamed)Eastmoney (东方财富网)Haitian Ruisheng (海天瑞声)ContractMixed8 Jun 2026Haitian Ruisheng, a Chinese AI training data company, told institutional investors it has signed a real-robot data collection contract with a leading domestic embodied-AI maker and has started order talks with several top embodied-AI makers, large tech firms and local governments.Not disclosed
Toyota Research Institute and WeaveThe Spokesman-Review (Tribune News Service)EncordContractMixed3 Jun 2026Encord, which records and sells movement data to companies racing to bring humanoids to homes, offices and factories, named Toyota Research Institute and Weave (laundry-folding robots) among its clients and charges up to 1,000 dollars per hour for training data.Encord charges clients up to $1,000 per hour for training data
Niantic SpatialThe AI InsiderSpexi Geospatial (network of drone pilots)PartnershipDrone data28 May 2026Niantic Spatial and Spexi partnered so that Spexi, which runs a network of more than 10,000 drone pilots, became a preferred drone imagery provider for training Niantic Spatial's real-world foundation models for physical AI.Not disclosed
Human ArchiveTechCrunchHome services, hotel and restaurant workers in IndiaPaid programEgocentric video26 May 2026Y Combinator backed Human Archive partners with Indian home services, hotel and restaurant companies to have workers wear camera caps that capture egocentric video for robot training, paying workers a base hourly rate.$1 per hour base rate
Tesla, Boston Dynamics and Figure AI (named by Egolab.AI as clients)Scroll.inEgolab.AIContractEgocentric video20 May 2026Indian startup Egolab.AI says in its company documents that it packages first-person footage from factory workers into datasets for global AI companies, naming Tesla, Boston Dynamics and Figure AI among them (the claim is Egolab's own and the report does not confirm it with the named companies).Not disclosed
JD.comDeep News (via Tiger Brokers)Suqian residents and other recruited data collectors (paid per qualified data)Paid programEgocentric video20 May 2026JD.com opened an embodied-intelligence data collection community in Suqian, Jiangsu, where trained residents wear its JoyEgoCam wearable while doing household chores, and plans to mobilize over 100,000 employees and 500,000 external people to collect data, targeting 5 million hours of real-world human video within a year.Not disclosed
Longyu Big DataYicai GlobalHome-based data collectors in ChinaPaid programDemonstrations15 May 2026Chinese data firm Longyu Big Data hires home-based collectors, many of them stay-at-home mothers, to record task demonstrations with gripper devices for embodied AI clients, at a stated monthly salary.CNY3,000 to CNY4,000 a month (home-based collectors)
Frontier physical AI teams and industrial companies (unnamed customers)PR Newswire (Lightwheel press release)LightwheelContractSimulation6 May 2026Lightwheel said it closed approximately 100 million dollars in orders in the first quarter of 2026 across simulation, data generation, evaluation and deployment systems for frontier physical AI teams and industrial companies.approximately $100 million in orders closed in Q1 2026
Ai2 (Allen Institute for AI)Ai2Ai2 in-house, curated with support from Cortex AIDataset releaseDemonstrations5 May 2026Ai2 released MolmoAct 2 with the MolmoAct 2-Bimanual YAM dataset, which it describes as the largest open-source bimanual tabletop manipulation dataset, with over 720 hours of two-arm robot demonstrations such as towel folding and grocery scanning; Ai2 says it curated the dataset with support from data company Cortex AI.Not disclosed
25 AV company partners (including Wayve)TechCrunchUber (AV Labs)PartnershipAV driving data1 May 2026Uber's CTO said the company has partnerships with 25 AV companies, including Wayve, and is building an AV cloud of labeled sensor data that partners can query to train their models, adding that Uber's goal is not to make money from this data.not charged for now (Uber's CTO: the goal is not to make money out of the data)
Autoware FoundationAutomotive WorldNATIX NetworkPartnershipAV driving data13 Apr 2026NATIX Network joined the Autoware Foundation as a Premium Member to supply multi-camera driving data for the foundation's first open-source end-to-end autonomous driving model. NATIX says its dataset spans more than 250 million kilometres from over 270,000 drivers.Not disclosed
Embodied-AI developers buying on the Baidu data supermarket36Kr (Lieyun, 猎云网)Baidu AI Cloud with partner embodied-AI companies (零次方, 灵生, Fourier, 纬钛科技, 拓元智慧, 枢途科技, 松应科技)MarketplaceMixed10 Apr 2026Baidu AI Cloud and several embodied-AI companies launched a beta embodied-intelligence data supermarket, a marketplace with a tiered, extensible data labelling system for robot training data.Not disclosed
Micro1MIT Technology ReviewContract data recorders in more than 50 countriesPaid programEgocentric video1 Apr 2026Palo Alto data company Micro1 pays thousands of contract workers in India, Nigeria, Argentina and other countries to film household chores, and sells the footage to robotics companies; a recorder in Nigeria told MIT Technology Review he is paid $15 an hour.$15 an hour (one recorder in Nigeria)
Robotics companies (market-wide spend estimated by micro1's CEO)MIT Technology ReviewMicro1 and similar real-world data vendorsContractEgocentric video1 Apr 2026Micro1's CEO Ali Ansari told MIT Technology Review he estimates robotics companies are now spending more than 100 million dollars each year to buy real-world data from Micro1 and similar companies (an executive's estimate, not an audited figure).more than $100 million each year (estimate by Micro1's CEO)
Google DeepMindPR Newswire (Agile Robots press release)Agile RobotsPartnershipMixed24 Mar 2026Agile Robots and Google DeepMind formed a strategic research partnership to run Gemini Robotics models on Agile Robots' industrial platform and to improve performance through robot deployment, data collection, model training and iteration.Not disclosed
DoorDashTechCrunchDoorDash couriers (Dashers)Paid programMixed19 Mar 2026DoorDash launched a stand-alone Tasks app that pays couriers in select US markets to film everyday tasks and record speech for AI and robotics systems, with pay shown upfront per task.Not disclosed
Universal RobotsUniversal RobotsScale AIPartnershipTeleoperation16 Mar 2026Universal Robots unveiled the UR AI Trainer, developed with Scale AI, which records motion, force and visual data from human-guided robot demonstrations, and said UR and Scale AI will release a large-scale industrial dataset collected on UR robots later in 2026.Not disclosed
InstaworkLos Angeles TimesLos Angeles gig workersPaid programEgocentric video12 Mar 2026Staffing marketplace Instawork handed out headband phone mounts to Los Angeles residents and paid them to film themselves doing chores as robot training data; one worker was paid $80 for two hours of footage.$80 for two hours of footage
Google DeepMind, Figure, AgiBot, ByteDance, Geely and BYD (named Lightwheel users)EqualOceanLightwheelPartnershipSimulation11 Mar 2026Simulation data company Lightwheel supplies simulation assets and synthetic datasets for robot training; an NVIDIA case study names Google DeepMind, Figure, AgiBot, ByteDance, Geely and BYD as partners that use them, and EqualOcean reported in March 2026 that Lightwheel had completed RMB 1 billion in combined financing.Not disclosed
Build AIScroll.inEgolab.AIData acquisitionEgocentric video1 Mar 2026Build AI, a US egocentric data company, acquired Egolab.AI, an Indian startup that collects first-person footage from factory workers with head-mounted cameras for robotics companies.Not disclosed
DaBoss.AIPR Newswire (Robo.ai)Robo.ai (subsidiary, Embodied AI data joint venture in Dubai)ContractMixed12 Feb 2026Nasdaq-listed Robo.ai said its subsidiary won its first commercial embodied AI data order: 30,000 hours of robot training data (RGB-D vision, 6-DoF motion trajectories, force and tactile feedback) for Silicon Valley data provider DaBoss.AI over 12 months. DaBoss is also Robo.ai's joint venture partner; no price was disclosed.Not disclosed
ClaruClaru (job posting)Remote wearable camera contributors (US)Paid programEgocentric video6 Feb 2026Robotics data company Claru posted a remote US role paying contributors per accepted recording hour to capture egocentric video for computer vision and robotics research.$35 to $40 per hour (per accepted recording hour)
OpenAIBusiness Insider Japan (translation of a Business Insider report)about 100 contract data collectors hired through a contracting agencyContractTeleoperation28 Jan 2026Business Insider reported, citing insiders, that OpenAI runs a robotics lab in San Francisco where about 100 contract workers on round-the-clock shifts teleoperate Franka robot arms with 3D-printed GELLO controllers to collect household-task training data, and that its contracting agency posted a robotics operator role for a second lab in Richmond, California.Not disclosed
Uber AV partners (Waymo, Waabi, Lucid Motors and others)TechCrunchUber (AV Labs)PartnershipAV driving data27 Jan 2026Uber launched an AV Labs division that sends its own sensor-equipped cars into cities to collect driving data for its autonomous vehicle partners, such as Waymo, Waabi and Lucid Motors. At launch TechCrunch reported that no partner contracts had been signed yet; by June 2026 Uber said it would put 500 data-collection Hyundai Ioniq 5s on the road that year.Not disclosed
ValeoAutomotive WorldNATIX NetworkPartnershipAV driving data23 Jan 2026NATIX Network and Valeo partnered to build an open-source multi-camera world foundation model for autonomous driving, with the first version trained on 5,000 hours of multi-camera driving data collected through NATIX's VX360 network in the US, Europe and Asia.Not disclosed
Ukrainian defense developersMinistry of Defence of UkraineUkraine Ministry of Defence (with Palantir)Dataset releaseDrone data21 Jan 2026Ukraine's Ministry of Defence, with the Ministry of Digital Transformation, the Armed Forces, a military intelligence research institute and Palantir, launched Brave1 Dataroom, a secure environment where vetted Ukrainian defense developers train AI on structured visual and thermal datasets of aerial targets, starting with drone interception. By June 2026 the ministry said more than 100 companies had access.Not disclosed
AgiBot (智元创新(上海)科技股份有限公司)STAR Market Daily (科创板日报, Cailian Press)Hubei Humanoid Robot Innovation Center (湖北人形机器人创新中心)ContractDemonstrations15 Jan 2026The Hubei Humanoid Robot Innovation Center signed a data service agreement with AgiBot and delivered thousands of hours of humanoid robot training data, collected to AgiBot's specifications; STAR Market Daily called it the first customized inter-company humanoid data transaction in China.thousands of hours of humanoid robot training data delivered
NEURA RoboticsNEURA RoboticsBosch (Robert Bosch Robotics GmbH)PartnershipDemonstrations14 Jan 2026NEURA Robotics and Bosch's robotics subsidiary agreed to jointly collect real-world work, movement and environmental data in Bosch facilities using sensor suits, as training data for humanoid robots.Not disclosed
HumanoidHumanoidSchaefflerPartnershipTeleoperation13 Jan 2026UK robotics company Humanoid and Schaeffler signed a strategic technology partnership that, beyond robot deployments and actuator supply, covers joint collection of robot training data through teleoperation and synthetic data generation.Not disclosed
NVIDIANVIDIA NewsroomNVIDIA (planned data-collection drives)Dataset releaseAV driving data5 Jan 2026At CES 2026 NVIDIA presented its Physical AI Open Datasets for autonomous vehicles, more than 1,700 hours of driving data on Hugging Face, alongside the Alpamayo reasoning model family. The dataset card says the 1,700 hours were recorded on planned data-collection drives in 25 countries and 2,500+ cities, licensed for AV development.Not disclosed
undisclosed buyer on the Jiangsu Data ExchangeIT之家 (IT Home)Jiangsu Zhujing Intelligent Technology (江苏箸境智能科技有限公司)MarketplaceMixed3 Jan 2026The Jiangsu Data Exchange said it completed what it called China's first on-exchange sale of an embodied AI dataset: about 25,000 structured clips of robot video with joint angle, current and torque data across office, supermarket, catering and housekeeping scenes, supplied by Jiangsu Zhujing Intelligent Technology. The buyer and price were not disclosed.Not disclosed
Beijing Innovation Center of Humanoid RoboticsGlobal TimesBeijing Innovation Center of Humanoid Robotics (in-house teleoperation)Dataset releaseTeleoperation1 Dec 2025The Beijing Innovation Center of Humanoid Robotics open-sourced RoboMIND 2.0 in December 2025, a real-world dataset of more than 310,000 dual-arm manipulation trajectories across six robot embodiments and 739 tasks; in September 2026 the center said RoboMIND downloads had passed 20 million. The first RoboMIND release (December 2024) was collected through human teleoperation.Not disclosed
Build AIScrollfactory workers filmed with Build AI head-mounted camerasDataset releaseEgocentric video1 Dec 2025US startup Build AI, which collects head-mounted camera footage of factory workers to train general purpose robots, made 100,000 hours of egocentric factory video public on Hugging Face in December 2025 (Egocentric-100K), after its 10,000-hour Egocentric-10K release.Not disclosed
Sunday RoboticsSunday Robotics (job posting)Memory Developers (remote and in-house contractors)Paid programDemonstrations19 Nov 2025Home robot maker Sunday recruits 'Memory Developers' who use company hardware to record demonstrations of household tasks, paid per hour of approved data for remote contributors and hourly for in-house contractors.$30 per hour of approved data during screening, up to $60 per hour after (remote); $35 per hour (in-house)
Generalist AIGeneralist AImultiple data foundry partners (unnamed)PartnershipDemonstrations4 Nov 2025Generalist AI said its GEN-0 robot foundation model is pretrained on more than 270,000 hours of real-world manipulation data collected in thousands of homes, warehouses and workplaces, growing by 10,000 hours a week, and that its pretraining datasets are collected together with multiple data foundry partners whose operations it compares and gives feedback on.Not disclosed
EncordLos Angeles TimesObjectwaysContractDemonstrations2 Nov 2025San Francisco data platform Encord contracts Indian data labeling firm Objectways to collect human demonstration data, such as staff in Karur filming themselves folding towels, for robotics customers.Not disclosed
Physical Intelligence and Dyna RoboticsLos Angeles TimesEncordContractDemonstrations2 Nov 2025The Los Angeles Times reports that Encord, the data platform that contracts Objectways to collect human demonstration data, works with robotics companies such as Physical Intelligence and Dyna Robotics.Not disclosed
NVIDIAUber (press release via Nasdaq)UberPartnershipAV driving data28 Oct 2025Uber and NVIDIA announced a joint robotaxi data factory on the NVIDIA Cosmos platform: Uber will collect more than 3 million hours of robotaxi-specific driving data for Level 4 model training and validation, and NVIDIA supplies GPUs and data curation, search and simulation tools. It builds on a January 2025 CES deal pairing Uber driving datasets with Cosmos and DGX Cloud.Not disclosed
CobotScale AIScale AIContractMixed24 Sep 2025Scale AI named Cobot, maker of the Proxie robot, as a customer of its physical AI data engine, and Cobot's CEO Brad Porter said Scale had already done work building a data set for it.Not disclosed
Generalist AIScale AIScale AIContractMixed24 Sep 2025Scale AI named Generalist AI as a customer of its robot data collection and annotation service, and Generalist's CEO Pete Florence called Scale a trusted partner for custom data.Not disclosed
Physical IntelligenceScale AIScale AIContractMixed24 Sep 2025Scale AI said it is providing and enriching robot training data for Physical Intelligence through its Data Engine for Physical AI, which collects data from dedicated robots and human demonstrations at its San Francisco lab and through contributors worldwide.Not disclosed
Figure AIFigure AIBrookfieldPartnershipEgocentric video17 Sep 2025Figure partnered with Brookfield to build a humanoid pretraining dataset by capturing human video in Brookfield's residential, office and logistics properties to train its Helix model. Brookfield also invested in Figure's Series C as part of the deal.Not disclosed
Galaxea AIarXivGalaxea AI (in-house teleoperators)Dataset releaseTeleoperation30 Aug 2025Galaxea released the Galaxea Open-World Dataset with its G0 model: 500 hours and 100,000 demonstration trajectories recorded by teleoperation on one robot platform across 50 real homes and workplaces and more than 150 tasks.Not disclosed
Volkswagen ADMTCoinDeskBee Maps (Hivemapper network contributors)PartnershipAV driving data8 Jul 2025Volkswagen's autonomous-vehicle testing subsidiary ADMT tapped Hivemapper's Bee Maps for real-time street-level imagery, captured by contributors with Hivemapper dashcams who earn crypto rewards, to help validate its self-driving technology for robotaxi pick-ups and drop-offs.Not disclosed
ApplearXivApple (in-house, Apple Vision Pro capture)Dataset releaseEgocentric video16 May 2025Apple researchers released EgoDex, 829 hours of egocentric video with paired 3D hand and finger tracking recorded on Apple Vision Pro, as a dataset for learning dexterous robot manipulation from human video.Not disclosed
Grab (GrabMaps)CointelegraphNATIX Network (Tesla drivers using VX360)PartnershipAV driving data6 May 2025Grab's mapping unit GrabMaps partnered with NATIX Network, which launched VX360, a device built on Grab's hardware platform that lets Tesla drivers collect and share 360 degree vehicle imagery in return for rewards. GrabMaps said the imagery provides training and validation datasets for autonomous driving and physical AI.Not disclosed
NVIDIAPR Newswire (Nexar)NexarPartnershipAV driving data18 Mar 2025Dashcam network Nexar announced a collaboration with NVIDIA to enhance NVIDIA Cosmos world foundation models with Nexar's crowdsourced real-world driving video, which Nexar says covers more than 100 million miles a month; Nexar in turn uses Cosmos and DGX Cloud for its own model training.Not disclosed
NVIDIANVIDIANVIDIA (in-house)Dataset releaseMixed18 Mar 2025At GTC 2025 NVIDIA released an open Physical AI dataset on Hugging Face: 15 terabytes including more than 320,000 robotics training trajectories and up to 1,000 OpenUSD assets, with autonomous vehicle data announced as coming later.Not disclosed
FourierYicai GlobalFourier (in-house)Dataset releaseDemonstrations17 Mar 2025Shanghai humanoid maker Fourier open-sourced Fourier ActionNet, a training dataset for its GRx humanoid robots covering tasks such as pick-and-place of tools, household items and food, together with a data acquisition, training and deployment toolchain.Not disclosed
AgiBotThe Robot ReportAgiBot (in-house teleoperators at its data collection facility)Dataset releaseTeleoperation30 Dec 2024Shanghai humanoid maker AgiBot announced AgiBot World, an open robot learning dataset of more than 1 million trajectories from 100 robots gathered by skilled teleoperators; the complete Beta release of 1,003,672 trajectories followed on 1 March 2025.Not disclosed
Kodiak AIKodiak AIKognicContractAnnotation9 Dec 2024Autonomous truck developer Kodiak AI chose Kognic's sensor-fusion data labeling platform to scale its annotation pipelines for lidar, radar and camera data, integrating its own pre-labeling with Kognic's platform.Not disclosed
Aurora InnovationOutsource AcceleratorUber Scaled Solutions (independent contractors)ContractAnnotation26 Nov 2024Uber's AI training and data labeling unit Scaled Solutions, which pays independent contractors per task, counted self-driving truck developer Aurora Innovation as a client; its contractors help Aurora's software classify objects on the road.Not disclosed
1X TechnologiesMaginative1X Technologies (in-house EVE robot data)Dataset releaseTeleoperation17 Sep 20241X Technologies released over 100 hours of vector-quantized video and raw actions from operating its EVE humanoids at 1X offices, with baseline models and the 1X World Model Challenge, whose first stage pays a 10,000 dollar prize.Not disclosed
TeslaFortuneData collection operators (hourly Tesla hires in Palo Alto)Paid programDemonstrations19 Aug 2024Tesla advertised hourly 'data collection operator' roles in which workers wear motion-capture suits and VR headsets to record movements used to train its Optimus humanoid robot.$25.25 to $48 per hour
Toyota Research InstitutearXivDROID collaboration (Stanford, UC Berkeley and other institutions)Dataset releaseTeleoperation19 Mar 2024DROID, a robot manipulation dataset of 76,000 demonstration trajectories (350 hours) collected by 50 data collectors in North America, Asia and Europe over 12 months, was open sourced; the authors thank Toyota Research Institute for support from data collection to compute.Not disclosed
Meta (FAIR and Project Aria)Carnegie Mellon University School of Computer ScienceEgo4D university consortiumDataset releaseEgocentric video30 Nov 2023Meta's FAIR team and Project Aria, with a consortium of university partners it brought together, announced Ego-Exo4D: simultaneous first-person and external video of skilled human activity recorded with more than 800 participants, with more than 1,400 hours of video to be open sourced in December 2023.Not disclosed
Google DeepMindGoogle DeepMind33 academic robotics labs (21 institutions)Dataset releaseDemonstrations3 Oct 2023Google DeepMind and partners from 33 academic labs pooled robot demonstration data from 22 robot types into the open Open X-Embodiment dataset, with more than 1 million episodes, released alongside the RT-1-X model.Not disclosed

Base: 88 deals and releases dated 1 Jan 2023 to 28 Oct 2026, each linked to its announcement or reporting.

Most deals do not disclose an amount; a stated amount is shown as its source gives it.

Credit: compiled by RobotTraining.jobs Research, Who Is Buying Robot Training Data: Deals, Paid Programs and Dataset Releases, October 2026 edition.

Figure 2

Who is buying

Which companies appear most often as buyers of robot training data?

BuyerDealsKinds of data
Internet majors and embodied-AI body makers (unnamed)
11
mixed, egocentric video and simulation19 Oct 2026
Frontier AI labs, Microsoft, Amazon and robotics companies like 1X
6
mixed, simulation, egocentric video and demonstrations22 Sep 2026
NVIDIA
4
av driving data and mixed5 Jan 2026
Encord
3
mixed and demonstrations22 Oct 2026
Google DeepMind
3
teleoperation, mixed and demonstrations30 Jun 2026
AgiBot (智元创新(上海)科技股份有限公司)
2
demonstrations and teleoperation15 Jan 2026
Allen Institute for AI (Ai2)
2
mixed and demonstrations23 Jul 2026
Build AI
2
egocentric video1 Mar 2026
Figure AI
2
egocentric video25 Aug 2026
Generalist AI
2
demonstrations and mixed4 Nov 2025
Instawork
2
egocentric video16 Jun 2026
micro1 (for an unnamed customer project)
2
av driving data and egocentric video4 Oct 2026
25 AV company partners (including Wayve)
2
av driving data1 May 2026
More than 100 Ukrainian defence companies
2
drone data11 Jun 2026
1X Technologies
1
teleoperation17 Sep 2024
Apple
1
egocentric video16 May 2025
Aurora Innovation
1
annotation26 Nov 2024
Autoware Foundation
1
av driving data13 Apr 2026
Beijing Innovation Center of Humanoid Robotics
1
teleoperation1 Dec 2025
BitRobot Network
1
teleoperation25 Jun 2026
Claru
1
egocentric video6 Feb 2026
Cobot
1
mixed24 Sep 2025
DaBoss.AI
1
mixed12 Feb 2026
DoorDash
1
mixed19 Mar 2026
Dyna Robotics
1
egocentric video23 Oct 2026
Enabled Intelligence
1
drone data16 Jun 2026
Fourier
1
demonstrations17 Mar 2025
Galaxea AI
1
teleoperation30 Aug 2025
Genesis AI
1
egocentric video23 Oct 2026
Grab (GrabMaps)
1
av driving data6 May 2025
Human Archive
1
egocentric video26 May 2026
Humanoid
1
teleoperation13 Jan 2026
JD.com
1
egocentric video20 May 2026
undisclosed buyer on the Jiangsu Data Exchange
1
mixed3 Jan 2026
Kinetic Blocks marketplace buyers (humanoid model developers)
1
mixed1 Sep 2026
Kodiak AI
1
annotation9 Dec 2024
LG Electronics
1
mixed18 Aug 2026
黎曼动力 (maker of the Riemann-1.0 embodied world-action model)
1
mixed6 Aug 2026
Longyu Big Data
1
demonstrations15 May 2026
Meta (FAIR and Project Aria)
1
egocentric video30 Nov 2023
Mifeng Technology (觅蜂科技, an embodied-AI data services company)
1
egocentric video10 Oct 2026
NEURA Robotics
1
demonstrations14 Jan 2026
Niantic Spatial
1
drone data28 May 2026
Objectways
1
egocentric video11 Jun 2026
OpenAI
1
teleoperation28 Jan 2026
Physen AI
1
mixed1 Aug 2026
Physical Intelligence
1
mixed24 Sep 2025
Reka
1
egocentric video6 Aug 2026
Sunday Robotics
1
demonstrations19 Nov 2025
Tesla
1
demonstrations19 Aug 2024
Toyota Research Institute
1
teleoperation19 Mar 2024
UK Ministry of Defence (for up to 12 UK companies)
1
drone data25 Sep 2026
Ukrainian defense companies
1
drone data10 Aug 2026
unnamed frontier AI labs and robotics companies (about 20 customers)
1
teleoperation17 Jun 2026
Universal Robots
1
teleoperation16 Mar 2026
Valeo
1
av driving data23 Jan 2026
Volkswagen ADMT
1
av driving data8 Jul 2025

Base: Every buyer named across 88 records.

Credit: compiled by RobotTraining.jobs Research, Who Is Buying Robot Training Data: Deals, Paid Programs and Dataset Releases, October 2026 edition.

Figure 3

Deals by year and kind of data

Is spending on robot training data growing, and on which kinds of data?

YearEgocentric videoMixedAV driving dataDrone dataTeleoperationSimulationDemonstrationsAnnotationAll
2023100000102
2024000030126
20253440205018
2026Small sample172066634062

Base: All 88 records by the year each was announced.

This year is counted only up to the review date.

Credit: compiled by RobotTraining.jobs Research, Who Is Buying Robot Training Data: Deals, Paid Programs and Dataset Releases, October 2026 edition.

Robotics labs · Fleet operators · Data vendors · Research teams

Buy the data behind this report

This report sums up records we gather ourselves, day after day:

  • 1,432 robot training roles followed, history included

    Every role on 100 employer job boards, read each day since 12 Sep 2026: its pay, its place, and the first and last day it was up.

  • 536 apply clicks over 28 days

    Which robot training roles people really apply to, by kind of work, pay and place. The clicks happen here, so no one else can count them.

  • 25 sourced industry datasets

    Funding, revenue, deals, fleets, deployments and prices across the robot training industry, each record linked to the publication it comes from.

On offer

  • The raw records

    Every robot training role we track: title, employer, place, kind of work, pay as stated, and the days it was first and last open.

  • A regular feed

    New and closed roles every week or month, as file drops or through the data endpoints.

  • A custom cut

    Any figure here by country, city, kind of work, vehicle, employer or date, or your own hiring benchmarked against the market.

Need the people behind this data for your own robots? (you contract them yourself) or (we deliver the robot data).

Investing in robotics, AV or robot-data companies? The Robot hiring radar is these records role by role, every day, as files.

Robot hiring brief: free, once a month

The robot-training job market in figures, on the first weekday of each month: open roles, what opened and closed, who added and who cut, and posted pay by kind of work. To your work email.

How it was measured

Company announcements, filings and reporting by major outlets, each linked beside its record with a verbatim quote kept in the dataset. Each row carries a link to its publication and gives the number exactly as published, with no estimates or conversions. A note under every figure says what it covers.

Buyer
The company that pays for, licenses or receives the data, as the source names it.
Paid program
A program that pays members of the public directly to record or produce robot training data (per minute, per task or per hour).
Dataset release
A large training dataset published or sponsored by a named company, kept because it shows who is building robot data at scale.

Limits of this data

  • Built only from public sources. Anything never announced or reported is absent, and private-company figures are whatever the company or the press said, unaudited.
  • Most deals do not disclose an amount, so the dataset counts deals, not dollars.
  • Where the parties' roles are shared (a joint data effort), the buyer is the party that receives the data for training, as the source describes it.

Collection and classification of listings are explained on the methodology page. Freshness, completeness and consistency are tested daily in the data quality report.

Monthly editions

The live page shows the dataset as of its latest review, and new events join it once they are made public. Every finished month also gets its own edition, locked two days after month end, so a number quoted from it stays fixed.

  • Live pageLatest review
  • October 2026 edition (you are here)Will lock when the next daily refresh runs

Quoting this report

You may quote, chart or republish these figures under CC BY 4.0, provided you link back to this page. This edition is locked: its numbers will not change.

Citation

RobotTraining.jobs (2026). Who Is Buying Robot Training Data: Deals, Paid Programs and Dataset Releases, October 2026 edition. https://robottraining.jobs/research/robot-training-data-deals/2026-10

BibTeX

@misc{robottrainingjobs_robot_training_data_deals_2026_10,
  author = {{RobotTraining.jobs}},
  title = {Who Is Buying Robot Training Data: Deals, Paid Programs and Dataset Releases, October 2026 edition},
  year = {2026},
  howpublished = {\url{https://robottraining.jobs/research/robot-training-data-deals/2026-10}},
}
Files: every figure in one CSV (a row per value) or as JSON, emailed to your work address.Other open data: datasets and API

Keep reading

Browse all 91 reports