{ "meta": { "buildDate": "2026-09-13", "lastVerified": "2026-09-13", "license": "CC BY 4.0", "licenseUrl": "https://creativecommons.org/licenses/by/4.0/", "cadence": "weekly, every Monday, and the same day for any announcement by a data regulator", "note": "Every record is a verdict about a document. Verified means fetched at its publisher and quoted. Reported means a reliable secondary source carries it and the primary could not be reached. Announced means a body said it will act and no document exists. Absent means the ledger searched and found nothing, and says where. Open means a question nobody has settled. Every figure carries the source that printed it and the date it was true. No figure is summed across sources, converted between units, or forecast.", "source": "https://www.gage.academy/tools/embodied-ai-data-ledger", "attribution": "GAGE (Global Academy of Generative-AI Education), Embodied AI Data Ledger", "documentation": "https://www.gage.academy/tools/embodied-ai-data-ledger/data", "methodology": "https://www.gage.academy/tools/embodied-ai-data-ledger/method" }, "records": [ { "id": "EAD-2026-0001", "slug": "china-nda-embodied-ai-data-standards-symposium", "title": "China's data regulator says it will advance embodied AI data standards, twelve days after seven companies asked", "kind": "standard", "verdict": "announced", "jurisdiction": "CN", "answer": "On 10 September 2026 the National Data Administration held an embodied AI symposium chaired by its director, Liu Liehong, and said it would advance embodied AI data standards when timely and guide local data authorities. On 29 August seven private companies had asked the same agency for public data infrastructure for embodied AI and common data standards. No standard, draft or timetable has been published.", "key_facts": [ "The 10 September release says the agency will advance embodied AI data standards when timely (适时) and guide local data systems to carry out the work in an orderly way. The word is conditional, not a commitment to a date.", "Speakers named in the release include the Beijing Humanoid Robot Innovation Center, the Shijingshan humanoid data training center, JD Group, and the Institute of Automation of the Chinese Academy of Sciences.", "At the 29 August meeting in Guiyang, seven private firms asked for public data infrastructure for embodied AI, better data standards and token metering rules, and faster building of high quality industry datasets.", "MLex reported the August meeting on 31 August naming three of the firms. The interval between the ask and the regulator's answer is twelve days, not the ten some coverage gives." ], "figures": [ { "label": "Private companies at the 29 August meeting that asked for embodied AI data infrastructure", "value": 7, "unit": "companies", "as_of": "2026-08-29", "source": 1 } ], "for_a_robot_maker": "Nothing binds yet. What has changed is that the body that sets China's data policy has put embodied AI data on its own agenda, at the industry's request, and has said local authorities will be guided. A maker collecting hours in China should expect a standard for that data to come from this agency and from the standards bodies it works with, and should read the two national standard plans already in drafting first.", "sources": [ { "name": "National Data Administration, embodied AI symposium release", "url": "https://www.nda.gov.cn/sjj/jgsz/jld/llh/llhldhd/0913/20260913104730557195636_pc.html", "type": "primary", "date": "2026-09-13" }, { "name": "National Data Administration, digital economy private enterprise symposium release", "url": "https://www.nda.gov.cn/sjj/jgsz/jld/llh/llhldhd/0829/20260829150219449533485_pc.html", "type": "primary", "date": "2026-08-29" }, { "name": "MLex, report on the private enterprise meeting", "url": "https://www.mlex.com/mlex/artificial-intelligence/articles/2519467", "type": "secondary", "date": "2026-08-31" } ], "related_ids": [ "EAD-2026-0002", "EAD-2026-0005", "EAD-2026-0006" ], "tags": [ "National Data Administration", "China", "embodied AI data standards", "Liu Liehong" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0002", "slug": "china-nda-high-quality-dataset-action-plan-2026", "title": "China's June 2026 dataset action plan names real machine interaction data for embodied AI", "kind": "standard", "verdict": "verified", "jurisdiction": "CN", "answer": "The National Data Administration issued its implementation plan for building high quality industry datasets on 3 June 2026. It names embodied AI among five innovation fields and calls for faster building of real machine interaction datasets for physical interaction, environment perception and motion control. It is a policy plan, not a standard, and sets no data volume.", "key_facts": [ "Document number 国数科基〔2026〕25号, issued by the National Data Administration, published on its site on 8 June 2026.", "The plan calls for accelerating real machine interaction datasets in key scenarios covering physical interaction, environment perception and motion control, to empower embodied AI.", "Embodied AI is one of five named innovation fields; the plan does not fix a number of hours, datasets or training grounds." ], "figures": [], "for_a_robot_maker": "This is the policy signal behind the training grounds: the state wants real machine interaction data built at industry scale and has said so in a numbered document. A maker in China can expect provincial money and infrastructure to follow it. A maker outside China should read it as the reason the supply side there is moving while the rule side elsewhere is not.", "sources": [ { "name": "National Data Administration, implementation plan on high quality industry datasets", "url": "https://www.nda.gov.cn/sjj/zwgk/zcfb/0608/20260608172117399715004_pc.html", "type": "primary", "date": "2026-06-08" } ], "related_ids": [ "EAD-2026-0001", "EAD-2026-0003" ], "tags": [ "National Data Administration", "China", "high quality datasets", "policy" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0003", "slug": "china-embodied-ai-training-grounds-count", "title": "China has over 70 embodied AI training grounds in use, per a CAICT report the ledger could only reach through state media", "kind": "supply", "verdict": "reported", "jurisdiction": "CN", "answer": "A research report released on 18 August 2026 by the China Academy of Information and Communications Technology with the Shanghai humanoid robot center counted more than 70 embodied AI training grounds built and in use as at the end of June 2026, with forty odd more under construction or planned, across more than half of provincial regions in three clusters. The ledger reached the report only through CCTV and Xinhua reproductions.", "key_facts": [ "CCTV, reproduced on a government portal: over 70 training grounds built and in use nationwide, and forty odd under construction or planned. The Xinhua summary carried by IT Home gives the second figure as 46.", "The three clusters are the Yangtze River Delta, the Beijing, Tianjin and Hebei region, and the Pearl River Delta.", "The 86 percent figure is the share of grounds that include industrial manufacturing scenes, not the share aimed only at manufacturing.", "The report PDF is not on the academy's own report listings, so every figure here rests on a state broadcaster's reproduction and is labeled that way." ], "figures": [ { "label": "Training grounds built and in use, at least", "value": 70, "unit": "training grounds", "as_of": "2026-06-30", "source": 0 }, { "label": "Under construction or planned, per the Xinhua summary; CCTV said forty odd", "value": 46, "unit": "training grounds", "as_of": "2026-06-30", "source": 1 }, { "label": "Share of grounds that include industrial manufacturing scenes", "value": 86, "unit": "percent", "as_of": "2026-06-30", "source": 1 } ], "for_a_robot_maker": "The supply of recorded hours is being built as public infrastructure in China, mostly for factory work. A maker who needs manipulation data in industrial scenes can buy or partner for it there in a way no other jurisdiction offers. A maker who needs home, hospital or outdoor data will find the same report says most grounds are not built for that.", "sources": [ { "name": "CCTV report reproduced on the Digital China government portal", "url": "https://www.digitalchina.gov.cn/2026/xwzx/spbb/202608/t20260819_5360963.htm", "type": "secondary", "date": "2026-08-19" }, { "name": "Xinhua summary carried by IT Home", "url": "https://www.ithome.com/0/992/357.htm", "type": "secondary", "date": "2026-08-20" }, { "name": "TechNode, report on the training ground count", "url": "https://technode.com/2026/08/21/china-has-more-than-70-operational-embodied-ai-training-grounds-report-says/", "type": "secondary", "date": "2026-08-21" } ], "related_ids": [ "EAD-2026-0004", "EAD-2026-0008", "EAD-2026-0009", "EAD-2026-0010" ], "tags": [ "training grounds", "China", "CAICT", "embodied AI training data" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0004", "slug": "caict-ten-million-hours-estimate", "title": "The hours gap: a CAICT report puts the need on the order of ten million hours and the world's usable supply at under a million", "kind": "supply", "verdict": "reported", "jurisdiction": "CN", "answer": "The same August 2026 training ground report from the China Academy of Information and Communications Technology estimates that an embodied AI foundation model of about 55 billion parameters would need effective real data on the order of ten million hours, while usable high quality real data worldwide is on the order of hundreds of thousands to a million hours. The ledger reached the figures through a Xinhua summary, not the report.", "key_facts": [ "The report's framing, per the Xinhua summary: a 55 billion parameter model needs about 1.1 trillion tokens; if all of it came from real robot data, that is roughly 21,200 robots working a full year, or effective data at the ten million hour scale.", "The same summary quotes the report: currently only hundreds of thousands to a million hours of usable high quality real data exist worldwide.", "Both figures are one institute's estimate under its own assumptions. No standard defines what counts as an hour of usable data, which is an open question on this ledger." ], "figures": [ { "label": "Effective real data the report says one foundation model needs, order of magnitude", "value": 10000000, "unit": "hours", "as_of": "2026-08-18", "source": 0 }, { "label": "Upper bound of usable high quality real data worldwide, per the report", "value": 1000000, "unit": "hours", "as_of": "2026-08-18", "source": 0 } ], "for_a_robot_maker": "If the estimate is even roughly right, the gap between what a model needs and what exists is at least an order of magnitude, and whoever closes it holds the input everyone else must buy. That is the case for training grounds, for company gyms, and for synthetic data, and it is why the question of what counts as an hour matters more than it looks.", "sources": [ { "name": "Xinhua summary of the CAICT training ground report, carried by IT Home", "url": "https://www.ithome.com/0/992/357.htm", "type": "secondary", "date": "2026-08-20" }, { "name": "CCTV report reproduced on the Digital China government portal", "url": "https://www.digitalchina.gov.cn/2026/xwzx/spbb/202608/t20260819_5360963.htm", "type": "secondary", "date": "2026-08-19" } ], "related_ids": [ "EAD-2026-0003", "EAD-2026-0025", "EAD-2026-0023" ], "tags": [ "CAICT", "ten million hours", "embodied AI training data", "data gap" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0005", "slug": "china-gbz-218-1-embodied-ai-data-quality-real-data", "title": "China has published a national guiding document on embodied AI data quality for real data, GB/Z 218.1-2026", "kind": "standard", "verdict": "verified", "jurisdiction": "CN", "answer": "GB/Z 218.1-2026, Artificial intelligence, embodied AI data quality specification, part 1: real data, is listed as published on the national standards platform of the State Administration for Market Regulation. It is a guiding technical document under the national information technology standards committee, registered in June 2025 with 31 drafting organizations. It is the earliest national level instrument the ledger found that governs the data itself.", "key_facts": [ "Status on the national platform: published, effective on publication. Plan number 20252047-Z-469, registered 20 June 2025, under TC28, the national information technology standardization committee.", "A GB/Z is a guiding technical document, not a mandatory standard; it recommends, it does not bind.", "Thirty one drafting organizations are listed on the platform entry. The publication date field is not shown on the page." ], "figures": [ { "label": "Drafting organizations listed on the platform entry", "value": 31, "unit": "organizations", "as_of": "2026-09-13", "source": 0 } ], "for_a_robot_maker": "A buyer of Chinese real robot data now has a national reference for what quality means: a document to write into a contract and to test a delivered dataset against. It is guidance, so a supplier can decline it, but a supplier who meets it has something to point at that no other jurisdiction offers yet.", "sources": [ { "name": "National standards platform, State Administration for Market Regulation, entry for GB/Z 218.1-2026", "url": "https://std.samr.gov.cn/gb/search/gbDetailed?id=37FC03D2E1586322E06397BE0A0AA17F", "type": "primary", "date": "2026-09-13" } ], "related_ids": [ "EAD-2026-0006", "EAD-2026-0007", "EAD-2026-0021" ], "tags": [ "GB/Z 218.1-2026", "China", "data quality", "embodied AI data standards" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0006", "slug": "china-humanoid-robot-dataset-national-standard-plans", "title": "Two national standards for humanoid robot datasets are in drafting in China, on an eighteen month cycle from June 2025", "kind": "standard", "verdict": "verified", "jurisdiction": "CN", "answer": "The national standards platform lists two national standard plans under the national robotics standardization committee, TC591, noticed on 12 June 2025 with an eighteen month cycle: Humanoid robot datasets, part 1: general principles, and part 2: upper limb manipulation. Drafters named include Beijing Research Institute of Automation for Machinery Industry, Tsinghua University and the Shanghai humanoid robot center. Neither has been published.", "key_facts": [ "Plan 20253226-T-604, Humanoid robot datasets, part 1: general principles. Drafters listed include the Beijing automation institute, Tsinghua, the Shenzhen AIRS institute, Galbot and the Shanghai AI research institute.", "Plan 20253221-T-604, Humanoid robot datasets, part 2: upper limb manipulation. Lead drafter listed as Humanoid Robots (Shanghai) Co.", "Both are national standard plans in research status, under TC591, not under the humanoid robot and embodied AI committee that MIIT set up in December 2025." ], "figures": [ { "label": "Drafting cycle stated in the plan notice", "value": 18, "unit": "months", "as_of": "2025-06-12", "source": 0 } ], "for_a_robot_maker": "The shape of a Chinese humanoid dataset standard is being decided now, by named institutes and companies, and the general principles part will define the vocabulary the rest inherits. A maker who sells into China or buys data from it should read the drafts when they circulate for comment. The same institute leads the ISO work item on the same subject, so the two are likely to rhyme.", "sources": [ { "name": "National standards platform, plan 20253226-T-604, Humanoid robot datasets part 1", "url": "https://std.samr.gov.cn/gb/search/gbDetailed?id=30D202B3ADBFE9F6E06397BE0A0A4001", "type": "primary", "date": "2026-09-13" }, { "name": "National standards platform, plan 20253221-T-604, Humanoid robot datasets part 2", "url": "https://std.samr.gov.cn/gb/search/gbDetailed?id=33C58193480D82D5E06397BE0A0ADE89", "type": "primary", "date": "2026-09-13" } ], "related_ids": [ "EAD-2026-0005", "EAD-2026-0021", "EAD-2026-0001" ], "tags": [ "humanoid robot datasets", "China", "TC591", "national standard plan" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0007", "slug": "china-ydt-6771-2026-embodied-ai-dataset-quality", "title": "An industry standard on embodied AI dataset quality, YD/T 6771-2026, takes effect on 1 November 2026, per a trade paper the ledger could not trace to the MIIT notice", "kind": "standard", "verdict": "reported", "jurisdiction": "CN", "answer": "A communications industry standard, YD/T 6771-2026, Artificial intelligence, key basic technology, embodied AI dataset quality requirements and evaluation methods, was reported as approved by the Ministry of Industry and Information Technology on 24 July 2026 and effective from 1 November 2026, drafted by CAICT with more than forty organizations across eight quality dimensions. The ministry's own notice was not reached.", "key_facts": [ "Reported by People's Posts and Telecommunications News, reprinted on a Shenzhen government portal: approved 24 July 2026, in force 1 November 2026, eight quality dimensions.", "A YD/T is a voluntary industry standard of the communications sector, distinct from the national GB series.", "The humanoid robot and embodied AI standardization committee under MIIT, set up in December 2025, released a standards system document in February 2026; the state media report of it gives no standard numbers." ], "figures": [ { "label": "Drafting organizations, at least, per the trade paper", "value": 40, "unit": "organizations", "as_of": "2026-08-04", "source": 0 } ], "for_a_robot_maker": "If the report is right, there will be a second Chinese quality reference for embodied AI datasets alongside GB/Z 218.1, this one from the telecoms standards line and with evaluation methods attached. A buyer should ask a Chinese supplier which of the two they test against, and should treat this row as reported until the ministry's approval notice is in hand.", "sources": [ { "name": "People's Posts and Telecommunications News, reprinted on the Shenzhen government portal", "url": "https://www.szzg.gov.cn/2026/xwzx/szkx/202608/t20260804_5354035.htm", "type": "secondary", "date": "2026-08-04" }, { "name": "Xinhua, on the humanoid robot and embodied AI standards system document", "url": "https://www.news.cn/20260228/c27e2dfdb0f4496494c7e4991f2e8c2f/c.html", "type": "secondary", "date": "2026-02-28" } ], "related_ids": [ "EAD-2026-0005", "EAD-2026-0006" ], "tags": [ "YD/T 6771-2026", "China", "MIIT", "dataset quality" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0008", "slug": "beijing-humanoid-robot-innovation-center-training-base", "title": "Beijing's humanoid robot innovation center runs a data and training base with over 120 robots in more than 30 scenes", "kind": "supply", "verdict": "reported", "jurisdiction": "CN", "answer": "The Beijing Humanoid Robot Innovation Center's data and training base in Beijing E-Town was described in March 2026 as close to 5,000 square meters, with more than 30 typical scenes and over 120 mainstream robots, having delivered tens of thousands of hours of high quality data with a stated aim of a million hours. The figures come from a district newspaper reprinted on the district government's portal.", "key_facts": [ "Reported in March 2026: close to 5,000 square meters, more than 30 typical scenes, over 120 mainstream robot units, tens of thousands of hours delivered.", "The million hour figure is the center's stated goal, not a measurement, and the ledger does not print it as one.", "The center is one of the bodies named as speaking at the National Data Administration's 10 September 2026 symposium." ], "figures": [ { "label": "Robot units at the base, at least", "value": 120, "unit": "robots", "as_of": "2026-03-20", "source": 0 }, { "label": "Typical scenes, at least", "value": 30, "unit": "scenes", "as_of": "2026-03-20", "source": 0 } ], "for_a_robot_maker": "One named site, with a robot count and a scene count, that a buyer of humanoid manipulation data can approach. The delivered volume is given only as tens of thousands of hours, so any contract should specify the unit and the quality test, which is what the national guiding document exists for.", "sources": [ { "name": "Yicheng Times, reprinted on the Beijing E-Town government portal", "url": "https://kfqgw.beijing.gov.cn/ywdt/kjcgzhgd/kjqy/202603/t20260320_4562488.html", "type": "secondary", "date": "2026-03-20" } ], "related_ids": [ "EAD-2026-0003", "EAD-2026-0009", "EAD-2026-0001" ], "tags": [ "Beijing", "training grounds", "humanoid robot", "embodied AI training data" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0009", "slug": "shijingshan-humanoid-data-training-center", "title": "Beijing's Shijingshan humanoid data training center reports close to 270 robots producing about 20 million data items a year", "kind": "supply", "verdict": "reported", "jurisdiction": "CN", "answer": "The Shijingshan humanoid robot data training center in Beijing was reported in May 2026 as running close to 270 robot units at full load and producing close to 20 million data items a year, up from a figure of more than 6 million a year reported in September 2025. The unit is data items, not hours, and the figures come from Beijing Daily reprinted on the municipal portal.", "key_facts": [ "May 2026, Beijing Daily on the municipal portal: close to 270 robot units training at full load, annual data output close to 20 million items.", "September 2025, an earlier report on a municipal portal: a site of over ten thousand square meters and annual data output above 6 million items.", "Data items is the unit the source used. It is not convertible to hours, and the ledger does not convert it." ], "figures": [ { "label": "Robot units training at full load, approximately", "value": 270, "unit": "robots", "as_of": "2026-05-12", "source": 0 }, { "label": "Annual data output, approximately", "value": 20000000, "unit": "data items per year", "as_of": "2026-05-12", "source": 0 }, { "label": "Annual data output reported eight months earlier, at least", "value": 6000000, "unit": "data items per year", "as_of": "2025-09-26", "source": 1 } ], "for_a_robot_maker": "Two figures eight months apart from the same site show the supply side scaling by a multiple, not a percentage. A buyer should note the unit: an item is whatever the center counts as one, and a contract that pays per item needs the definition written down.", "sources": [ { "name": "Beijing Daily, reprinted on the Beijing municipal government portal", "url": "https://www.beijing.gov.cn/fuwu/lqfw/gggs/202605/t20260512_4647044.html", "type": "secondary", "date": "2026-05-12" }, { "name": "Beijing municipal science and technology portal", "url": "https://kw.beijing.gov.cn/xwdt/kcyx/xwdtshgg/202509/t20250926_4211351.html", "type": "secondary", "date": "2025-09-26" } ], "related_ids": [ "EAD-2026-0008", "EAD-2026-0003" ], "tags": [ "Shijingshan", "Beijing", "training grounds", "embodied AI training data" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0010", "slug": "shanghai-zhangjiang-humanoid-training-ground", "title": "Shanghai's Zhangjiang humanoid training ground opened its first phase with over 100 heterogeneous robots", "kind": "supply", "verdict": "reported", "jurisdiction": "CN", "answer": "The national and local humanoid robot innovation center's training ground in Shanghai's Zhangjiang was reported in February 2025 as having deployed more than 100 heterogeneous robots in its first phase, with a stated hope that year of accumulating ten million data entries. The figures come from Liberation Daily reprinted on the Shanghai municipal portal, and the ten million is an expectation the ledger does not print as a result.", "key_facts": [ "February 2025, Liberation Daily on the Shanghai portal: first phase deployed over 100 heterogeneous robots.", "The same report voiced a hope of ten million data entries that year; no later source reached by the ledger reports the outcome.", "The center co authored the August 2026 CAICT training ground report that counts the national total." ], "figures": [ { "label": "Heterogeneous robots deployed in the first phase, at least", "value": 100, "unit": "robots", "as_of": "2025-02-11", "source": 0 } ], "for_a_robot_maker": "Heterogeneous is the word that matters: the site was built to record many robot bodies doing the same tasks, which is the data a cross embodiment model wants. It is also the site whose operator wrote the national count, so a buyer reading that count should know who counted.", "sources": [ { "name": "Liberation Daily, reprinted on the Shanghai municipal government portal", "url": "https://www.shanghai.gov.cn/nw4411/20250211/5e981bfbee7840c6b3056d9d3a6cceea.html", "type": "secondary", "date": "2025-02-11" }, { "name": "Shanghai municipal government portal, follow up report", "url": "https://www.shanghai.gov.cn/nw4411/20250501/1127ab41648c404c87013e8db06e3a14.html", "type": "secondary", "date": "2025-05-01" } ], "related_ids": [ "EAD-2026-0003", "EAD-2026-0006" ], "tags": [ "Shanghai", "Zhangjiang", "training grounds", "heterogeneous robots" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0011", "slug": "china-cross-border-data-rules-robot-recordings", "title": "Hours recorded in China leave the country under the 2024 cross border provisions and the 2025 network data regulations", "kind": "rule", "verdict": "verified", "jurisdiction": "CN", "answer": "Robot recordings that contain personal information or important data are governed on the way out of China by the Cyberspace Administration's provisions on cross border data flows, in force since 22 March 2024, and by the State Council's network data security regulations, in force since 1 January 2025. A handler need not treat data as important unless told or a catalog says so, and personal information below set head counts moves on a standard contract or without a filing.", "key_facts": [ "Cyberspace Administration provisions, article 2: a handler need not declare data for an outbound security assessment as important data unless notified by a department or region or it is publicly listed as such.", "Article 7: a security assessment is required for important data, or for personal information of one million or more people, or sensitive personal information of ten thousand or more, counted from 1 January of the year.", "Article 8: between one hundred thousand and one million people, or under ten thousand sensitive records, a standard contract or certification suffices. Article 6 lets free trade zones publish negative lists.", "State Council regulations, article 62, define important data by field, group, region, precision and scale; article 29 puts catalogs with regions and sectors; article 37 sends important data abroad only through the state assessment." ], "figures": [ { "label": "Head count of personal information at or above which an outbound security assessment is required", "value": 1000000, "unit": "people", "as_of": "2024-03-22", "source": 0 }, { "label": "Head count of sensitive personal information at or above which an assessment is required", "value": 10000, "unit": "people", "as_of": "2024-03-22", "source": 0 } ], "for_a_robot_maker": "Footage of people in a training ground is personal information under Chinese law, and moving it out counts heads. Below the thresholds the road is a standard contract; above them it is a state assessment. Whether the recordings are also important data is not answered by any catalog the ledger found, which is the open question beside this record.", "sources": [ { "name": "Cyberspace Administration of China, Provisions on Promoting and Regulating Cross Border Data Flows", "url": "https://www.cac.gov.cn/2024-03/22/c_1712776611775634.htm", "type": "primary", "date": "2024-03-22" }, { "name": "State Council, Network Data Security Management Regulations, Order 790", "url": "https://www.gov.cn/zhengce/content/202409/content_6977766.htm", "type": "primary", "date": "2024-09-24" }, { "name": "DigiChina, Stanford, translation of the Personal Information Protection Law", "url": "https://digichina.stanford.edu/work/translation-personal-information-protection-law-of-the-peoples-republic-of-china-effective-nov-1-2021/", "type": "secondary", "date": "2021-11-01" } ], "related_ids": [ "EAD-2026-0012", "EAD-2026-0019", "EAD-2026-0027" ], "tags": [ "China", "cross border data", "PIPL", "important data", "Cyberspace Administration" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0012", "slug": "are-robot-training-recordings-important-data-in-china", "title": "Open question: are a training ground's recordings important data under Chinese law? No catalog names robot data", "kind": "question", "verdict": "open", "jurisdiction": "CN", "answer": "Chinese law sends important data abroad only through a state security assessment, and defines it by field, group, region, precision and scale. No catalog the ledger found names robot training recordings. Beijing's 2025 free trade zone negative list covers nine sectors including artificial intelligence and autonomous driving, with 67 scenarios and 612 data fields, and none is robotics. Whether hours of factory footage are important data is unanswered.", "key_facts": [ "The State Council's 2025 regulations define important data at article 62 by field, group, region, precision and scale, and leave the naming to catalogs under article 29.", "Beijing's 2025 negative list, per the municipal government, spans nine sectors, 67 business scenarios and 612 data fields, and adds autonomous driving; robotics is not a named sector.", "A recording of a factory floor could be argued into important data by region and precision, or out of it by the absence of any catalog entry. Nobody with authority has said which." ], "figures": [ { "label": "Data fields on Beijing's 2025 free trade zone negative list", "value": 612, "unit": "data fields", "as_of": "2025-05-11", "source": 1 }, { "label": "Sectors on that list", "value": 9, "unit": "sectors", "as_of": "2025-05-11", "source": 1 } ], "for_a_robot_maker": "Until a catalog speaks, a maker exporting Chinese training data is choosing between two readings with no authority behind either. The conservative road is to treat precise, large scale recordings of industrial sites as if a catalog might name them, and to keep the outbound file to what a standard contract clearly covers.", "sources": [ { "name": "State Council, Network Data Security Management Regulations, Order 790", "url": "https://www.gov.cn/zhengce/content/202409/content_6977766.htm", "type": "primary", "date": "2024-09-24" }, { "name": "Beijing Municipal Government, on the 2025 free trade zone data negative list", "url": "https://english.beijing.gov.cn/latest/news/202605/t20260525_4664345.html", "type": "primary", "date": "2026-05-25" }, { "name": "Cyberspace Administration of China, Provisions on Promoting and Regulating Cross Border Data Flows", "url": "https://www.cac.gov.cn/2024-03/22/c_1712776611775634.htm", "type": "primary", "date": "2024-03-22" } ], "related_ids": [ "EAD-2026-0011", "EAD-2026-0019" ], "tags": [ "important data", "China", "negative list", "open question" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0013", "slug": "agibot-world-beta-largest-open-real-robot-dataset", "title": "The largest open real robot dataset is Chinese, over a million trajectories, and licensed for non commercial use only", "kind": "supply", "verdict": "verified", "jurisdiction": "CN", "answer": "AgiBot World Beta, released on 1 March 2025 by AgiBot of Shanghai with OpenDriveLab, holds 1,003,672 trajectories from 100 robots with a total duration of 2,976.4 hours, under a Creative Commons Attribution NonCommercial ShareAlike 4.0 license. It is the only real robot release above a million episodes the ledger could anchor at its publisher. A 2026 successor on a newer platform gives no trajectory count.", "key_facts": [ "GitHub release notes: Alpha, 92,214 trajectories, 30 December 2024; Beta, 1,003,672 trajectories, about 43.8 terabytes, 1 March 2025.", "The Hugging Face card: over one million trajectories from 100 robots, total duration 2,976.4 hours. Licence CC BY-NC-SA 4.0 on every version.", "AgiBot World 2026, on the G2 platform, lists a size band of one to ten thousand episodes and 13.6 terabytes; the company describes phase one as hundreds of hours of real world data." ], "figures": [ { "label": "Trajectories in AgiBot World Beta", "value": 1003672, "unit": "trajectories", "as_of": "2025-03-01", "source": 0 }, { "label": "Total duration of AgiBot World Beta", "value": 2976.4, "unit": "hours", "as_of": "2025-03-01", "source": 1 }, { "label": "Robots that recorded it", "value": 100, "unit": "robots", "as_of": "2025-03-01", "source": 1 } ], "for_a_robot_maker": "A million trajectories is three thousand hours, which puts the largest open release at a third of a percent of the CAICT need estimate. The license bars commercial use, so a company cannot train a product on it without a separate agreement. The open supply is smaller and more encumbered than the headline count suggests.", "sources": [ { "name": "OpenDriveLab, AgiBot World repository", "url": "https://github.com/OpenDriveLab/AgiBot-World", "type": "primary", "date": "2025-03-01" }, { "name": "AgiBot World Beta dataset card, Hugging Face", "url": "https://huggingface.co/datasets/agibot-world/AgiBotWorld-Beta", "type": "primary", "date": "2025-03-01" }, { "name": "AgiBot World 2026 dataset card, Hugging Face", "url": "https://huggingface.co/datasets/agibot-world/AgiBotWorld2026", "type": "primary", "date": "2026-09-13" } ], "related_ids": [ "EAD-2026-0023", "EAD-2026-0033", "EAD-2026-0004" ], "tags": [ "AgiBot World", "open dataset", "China", "non commercial license" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0014", "slug": "eu-data-act-connected-products-robots", "title": "The EU Data Act reaches robots as connected products: user access since September 2025, access by design for products placed on the market after 12 September 2026", "kind": "rule", "verdict": "verified", "jurisdiction": "EU", "answer": "Regulation (EU) 2023/2854 has applied since 12 September 2025. It gives the user of a connected product a right to the data it generates and to share it with a third party, bars that third party from using the data to build a competing product, and requires products placed on the market after 12 September 2026 to be designed so the data is accessible. The Commission's own explainer names robots among connected products.", "key_facts": [ "Article 50: it applies from 12 September 2025; the article 3(1) design obligation applies to connected products placed on the market after 12 September 2026.", "Article 2(5): a connected product obtains, generates or collects data on its use or environment and can communicate it; recital 14 names agricultural and industrial machinery. The regulation never says robot; the Commission's fact page does.", "Article 6(2)(e): a third party receiving data may not use it to develop a product that competes with the connected product it came from, nor to derive insights about the data holder's economic situation, assets or production methods.", "Article 4(6) preserves trade secrets and article 4(8) lets a holder refuse where it shows it is highly likely to suffer serious economic damage." ], "figures": [], "for_a_robot_maker": "A robot sold into the EU is a connected product, so its buyer can demand the hours it records and hand them to a rival integrator, who may not train a competing robot on them. That is the one rule anywhere on this ledger that allocates recorded hours between maker, customer and third party. It regulates access to the data. It does not produce any.", "sources": [ { "name": "Regulation (EU) 2023/2854, the Data Act, Official Journal text", "url": "https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:32023R2854", "type": "primary", "date": "2023-12-22" }, { "name": "European Commission, Data Act explained", "url": "https://digital-strategy.ec.europa.eu/en/factpages/data-act-explained", "type": "primary", "date": "2025-12-15" } ], "related_ids": [ "EAD-2026-0026", "EAD-2026-0018", "EAD-2026-0015" ], "tags": [ "EU Data Act", "connected products", "robots", "data access" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0015", "slug": "eu-data-union-strategy-data-labs", "title": "The EU promised data labs for the fourth quarter of 2025 and no page of its own names one as open", "kind": "standard", "verdict": "announced", "jurisdiction": "EU", "answer": "The Commission's Data Union Strategy of 19 November 2025 makes scaling access to data for AI its first priority and defines data labs as service providers linking European data spaces to the AI ecosystem, to be established under the AI Factories through EuroHPC, with the first launched in the fourth quarter of 2025. As of 13 September 2026 the Commission's data union page still reads launching, and no release names an operating lab.", "key_facts": [ "COM(2025) 835, section II: data labs are data service providers that link data spaces with the AI ecosystem, offering data pooling, curation, labelling and pseudonymization; the first were to launch in the fourth quarter of 2025.", "The Commission's data union policy page, last updated 18 May 2026, lists launching the first data labs as a current action.", "EuroHPC opened a call on 28 April 2026 to network AI Factory data labs into a shared framework, with an indicative budget of 25 million euro, without naming a lab as operating.", "Nothing in the strategy funds robot data collection; the labs pool and label data from the common data spaces." ], "figures": [ { "label": "Indicative budget of the EuroHPC call to network AI Factory data labs", "value": 25000000, "unit": "euro", "as_of": "2026-04-28", "source": 2 } ], "for_a_robot_maker": "The European instrument nearest to a training ground is a pooling and labelling service that is, on the Commission's own pages, still launching ten months after its target quarter. A maker in Europe who needs hours cannot get them from a data lab today, and the strategy never promised robot data in the first place.", "sources": [ { "name": "European Commission, Data Union Strategy, COM(2025) 835", "url": "https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:52025DC0835", "type": "primary", "date": "2025-11-19" }, { "name": "European Commission, data union policy page", "url": "https://digital-strategy.ec.europa.eu/en/policies/data-union", "type": "primary", "date": "2026-05-18" }, { "name": "EuroHPC Joint Undertaking, call to strengthen the European AI ecosystem", "url": "https://www.eurohpc-ju.europa.eu/eurohpc-ju-launches-call-proposals-strengthen-european-ai-ecosystem-2026-04-28_en", "type": "primary", "date": "2026-04-28" } ], "related_ids": [ "EAD-2026-0016", "EAD-2026-0014" ], "tags": [ "Data Union Strategy", "data labs", "AI Factories", "EuroHPC" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0016", "slug": "no-eu-program-funds-robot-training-data-collection", "title": "No EU or member state program funds robot training data collection as its object", "kind": "supply", "verdict": "absent", "jurisdiction": "EU", "answer": "The ledger searched the Apply AI Strategy, the Data Union Strategy, the euROBIN network of excellence, the Commission's robotics policy page, the EuroHPC calls and France's 2030 robotics program for a public program whose object is collecting robot training data. None was found. euROBIN shares data and software among labs; Apply AI funds robotics foundation models and adoption pipelines; France funds research that needs quality data.", "search": "Searched on 13 September 2026: COM(2025) 723 Apply AI Strategy; COM(2025) 835 Data Union Strategy; CORDIS entry for euROBIN, grant 101070596; the Commission robotics policy page, last updated 27 March 2026; the EuroHPC call of 28 April 2026; the French economy ministry release of 13 June 2025 on the 30 million euro robotics research program; the Robotics Institute Germany site, which refused the fetch.", "key_facts": [ "euROBIN, coordinated by DLR, runs from July 2022 to December 2026 with an EU contribution of 11,499,999 euro; its stated exchange is software, data and knowledge over the EuroCore repository among its own labs.", "The Apply AI Strategy, 8 October 2025, promises a catalyst for European robotics uptake and sectoral acceleration pipelines, and facilitating data pooling among industrial actors; it funds models and adoption, not recording.", "France's 13 June 2025 release launches a 30 million euro robotics research program and says projects need access to quality data; it does not say who will record it.", "The Commission's robotics page names no robotics strategy with a date, and none appeared in the 2026 work program index the ledger checked." ], "figures": [ { "label": "EU contribution to euROBIN, the closest network", "value": 11499999, "unit": "euro", "as_of": "2026-09-13", "source": 1 }, { "label": "France 2030 robotics research program", "value": 30000000, "unit": "euro", "as_of": "2025-06-13", "source": 2 } ], "for_a_robot_maker": "In Europe the hours come from companies or from nowhere. A maker planning on public data infrastructure like China's will not find it in any program the ledger could read, and the one European operator building gyms at scale is a private firm on this ledger under Germany.", "sources": [ { "name": "European Commission, Apply AI Strategy, COM(2025) 723", "url": "https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX%3A52025DC0723", "type": "primary", "date": "2025-10-08" }, { "name": "CORDIS, euROBIN project entry", "url": "https://cordis.europa.eu/project/id/101070596", "type": "primary", "date": "2026-09-13" }, { "name": "French Ministry of the Economy, robotics research program release", "url": "https://presse.economie.gouv.fr/", "type": "primary", "date": "2025-06-13" }, { "name": "European Commission, robotics policy page", "url": "https://digital-strategy.ec.europa.eu/en/policies/robotics", "type": "primary", "date": "2026-03-27" } ], "related_ids": [ "EAD-2026-0015", "EAD-2026-0020", "EAD-2026-0003" ], "tags": [ "EU", "robot training data", "euROBIN", "Apply AI Strategy", "absent" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0017", "slug": "gdpr-factory-footage-personal-data-biometric", "title": "Under the GDPR, factory footage of a worker is personal data, and biometric data only when processed to identify them", "kind": "rule", "verdict": "verified", "jurisdiction": "EU", "answer": "Video that shows an identifiable person is personal data under article 4(1) of the GDPR. It becomes biometric data under article 4(14), and special category data under article 9, only when it results from specific technical processing to uniquely identify a person; recital 51 says so for photographs and the European Data Protection Board says footage is not in itself biometric. Robot training recordings of workers fall on the first side of that line unless a maker builds identification in.", "key_facts": [ "GDPR article 4(1): personal data is any information relating to an identified or identifiable natural person. Article 4(14): biometric data results from specific technical processing of physical, physiological or behavioural characteristics.", "Recital 51: photographs are biometric only when processed through a specific technical means allowing unique identification or authentication.", "EDPB Guidelines 3/2019 on video devices, version 2, adopted 29 January 2020: footage of an individual cannot in itself be considered biometric data under article 9 if it has not been specifically technically processed.", "The same guidelines set three criteria for biometric data: the nature of the data, a specific technical processing, and the purpose of uniquely identifying a person." ], "figures": [], "for_a_robot_maker": "Recording people in a European workplace to train a robot needs a lawful basis, information to the people recorded, and a retention rule, and the usual basis for workers is not consent. It does not need the article 9 regime unless the pipeline identifies individuals. Blurring faces before training removes the harder question and often the personal data itself.", "sources": [ { "name": "Regulation (EU) 2016/679, the GDPR", "url": "https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32016R0679", "type": "primary", "date": "2016-05-04" }, { "name": "European Data Protection Board, Guidelines 3/2019 on processing of personal data through video devices", "url": "https://www.edpb.europa.eu/sites/default/files/files/file1/edpb_guidelines_201903_video_devices_en_0.pdf", "type": "primary", "date": "2020-01-29" } ], "related_ids": [ "EAD-2026-0027", "EAD-2026-0031", "EAD-2026-0011" ], "tags": [ "GDPR", "biometric data", "video", "workers", "EDPB" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0018", "slug": "eu-ai-act-machinery-regulation-robots-annex-i", "title": "The AI Act reaches robots as machinery through the old Machinery Directive, and learned safety behaviour needs third party assessment from January 2027", "kind": "rule", "verdict": "verified", "jurisdiction": "EU", "answer": "Annex I of the AI Act lists Directive 2006/42/EC on machinery, not the 2023 Machinery Regulation. The Regulation repeals the Directive with effect from 14 January 2027 and reads references to it as references to itself, so robots reach Annex I as machinery from that date. Its Annex I Part A puts machinery and safety components with self evolving behaviour using machine learning under mandatory third party conformity assessment. Article 6(1) of the AI Act applies from 2 August 2027.", "key_facts": [ "AI Act Annex I, section A, item 1: Directive 2006/42/EC on machinery. The string 2023/1230 does not appear in the AI Act.", "Machinery Regulation article 52(2): Directive 2006/42/EC is repealed with effect from 14 January 2027 and references to it are construed as references to the Regulation. Article 54: it applies from 14 January 2027.", "Machinery Regulation Annex I Part A, items 5 and 6: safety components and machinery with fully or partially self evolving behaviour using machine learning approaches ensuring safety functions, third party assessment mandatory.", "AI Act article 6(1): high risk where an AI system is a safety component of, or is, a product under Annex I legislation that needs third party assessment; article 113(c) applies it from 2 August 2027.", "The Machinery Regulation never uses the word robot in its operative text; industrial robots are machinery by the article 3 definition, an assembly with a drive system and at least one moving part." ], "figures": [], "for_a_robot_maker": "What a robot learned from its hours becomes a conformity question in Europe from 2027: a safety function that keeps learning is assessed by a third party, and the AI Act's high risk duties, including its data governance article, attach through the machinery route. The training data is not regulated as data; it is regulated as the behaviour it produces.", "sources": [ { "name": "Regulation (EU) 2024/1689, the AI Act, Official Journal text", "url": "https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:32024R1689", "type": "primary", "date": "2024-07-12" }, { "name": "Regulation (EU) 2023/1230, the Machinery Regulation, Official Journal text", "url": "https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:32023R1230", "type": "primary", "date": "2023-06-29" } ], "related_ids": [ "EAD-2026-0014", "EAD-2026-0019" ], "tags": [ "EU AI Act", "Machinery Regulation", "Annex I", "robots", "conformity assessment" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0019", "slug": "can-chinese-training-ground-hours-train-a-robot-sold-in-the-eu", "title": "Open question: can hours recorded in a Chinese training ground train a robot sold in the European Union?", "kind": "question", "verdict": "open", "jurisdiction": "EU", "answer": "No rule on this ledger answers it. The Data Act governs access to data a connected product generates, not where a model's training data came from. The GDPR reaches the recordings only if they contain personal data of people in the Union. China's outbound rules gate the export by head count and by whether the hours are important data, which no catalog has said. The AI Act's data governance duties arrive with the machinery route in 2027.", "key_facts": [ "The Data Act allocates data a connected product generates among user, holder and third party; it says nothing about the provenance of training data.", "AI Act article 10 sets data governance duties for high risk systems, and for robots those duties attach through Annex I machinery from 2027; the article speaks of relevance, representativeness and bias, not of country of origin.", "On the Chinese side, personal information in the footage counts heads under the 2024 provisions, and whether the recordings are important data has no catalog answer.", "A commercial license on the hours is a separate question again: the largest open Chinese dataset is non commercial." ], "figures": [], "for_a_robot_maker": "The honest position today is that the buying side has no rule against it and the selling side has a rule whose scope is undecided. A maker doing it should document the export road taken in China, keep personal data out of the shipped hours, and be ready to show representativeness under article 10 when the machinery route opens.", "sources": [ { "name": "Regulation (EU) 2024/1689, the AI Act, Official Journal text", "url": "https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:32024R1689", "type": "primary", "date": "2024-07-12" }, { "name": "Cyberspace Administration of China, Provisions on Promoting and Regulating Cross Border Data Flows", "url": "https://www.cac.gov.cn/2024-03/22/c_1712776611775634.htm", "type": "primary", "date": "2024-03-22" }, { "name": "Regulation (EU) 2023/2854, the Data Act, Official Journal text", "url": "https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:32023R2854", "type": "primary", "date": "2023-12-22" } ], "related_ids": [ "EAD-2026-0011", "EAD-2026-0012", "EAD-2026-0018", "EAD-2026-0013" ], "tags": [ "open question", "cross border", "EU", "China", "training data provenance" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0020", "slug": "neura-robotics-gyms-physical-ai-training-data", "title": "Europe's nearest equivalent to a training ground is one company: NEURA Robotics has ten gyms under development", "kind": "supply", "verdict": "verified", "jurisdiction": "DE", "answer": "NEURA Robotics said on 22 July 2026 that ten NEURA Gyms were under development, with a stated target of five locations across Europe, the United States and China in operation before the end of the year, and announced a partnership with RWTH Aachen University for a gym there. The release names a TUM RoboGym at Munich Airport as well. It gives no hours, episodes or terabytes, so the ledger prints no data volume.", "key_facts": [ "NEURA, 22 July 2026: ten NEURA Gyms currently under development; five locations across Europe, the United States and China named as the year end target.", "The same release announces NEURA Gym RWTH Aachen with RWTH Aachen University, and names the TUM RoboGym at Munich Airport, powered by NEURA.", "The release states no data volume of any kind. Any hours figure attached to the gyms in coverage is not the company's." ], "figures": [ { "label": "Gyms under development, per the company", "value": 10, "unit": "gyms", "as_of": "2026-07-22", "source": 0 }, { "label": "Locations the company named as its year end target", "value": 5, "unit": "gyms", "as_of": "2026-07-22", "source": 0 } ], "for_a_robot_maker": "A European source of recorded hours exists, and it is a competitor's. A maker who partners for gym data is buying from a company building its own robots, which is the position the Data Act's competing product rule was written for on the customer side and says nothing about here. The count to watch is hours, and the company has not published one.", "sources": [ { "name": "NEURA Robotics, release on NEURA Gym and RWTH Aachen", "url": "https://neura-robotics.com/neura-robotics-rwth-aachen-neura-gym-physical-ai/", "type": "primary", "date": "2026-07-22" } ], "related_ids": [ "EAD-2026-0016", "EAD-2026-0003" ], "tags": [ "NEURA Robotics", "gyms", "Germany", "physical AI training data" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0021", "slug": "iso-26264-1-humanoid-robot-datasets-committee-draft", "title": "ISO is drafting a humanoid robot datasets standard, ISO/CD 26264-1, at committee draft stage", "kind": "standard", "verdict": "verified", "jurisdiction": "GLOBAL", "answer": "ISO/CD 26264-1, Humanoid robot datasets, part 1: general requirements, is under development in ISO technical committee 299 at stage 30.60, close of comment period, as read on the ISO catalog on 13 September 2026. Its abstract covers a dataset life cycle from planning to decommissioning, with quality, security and privacy requirements, for training, validation and test datasets of machine learning models in humanoid robots of any locomotion. A Beijing portal reports the new work item passed with 88 percent.", "key_facts": [ "ISO catalog entry 93011: status under development, stage 30.60 close of comment period, edition 1, technical committee ISO/TC 299.", "Abstract: a humanoid robot dataset life cycle framework from planning to decommissioning, with cross cutting quality, security and privacy requirements, for industrial, service and medical domains and any locomotion.", "A Beijing science and technology portal, reposting a WeChat account, reports the proposal was led by the Beijing Research Institute of Automation for Machinery Industry and passed the new work item vote with 88 percent in April 2026; that report is secondary.", "The same institute leads China's national standard plan for humanoid robot datasets part 1." ], "figures": [ { "label": "New work item approval vote, per the Beijing portal", "value": 88, "unit": "percent", "as_of": "2026-04-28", "source": 1 } ], "for_a_robot_maker": "This is the first international standard the ledger found whose object is the training data of a humanoid robot rather than the robot. A committee draft can change, but the life cycle frame and the quality, security and privacy split are visible now, and the drafting is led from China. A maker who wants the definitions to fit its data should be in TC 299 through a national body before the enquiry stage.", "sources": [ { "name": "ISO catalog, ISO/CD 26264-1 Humanoid robot datasets, part 1: general requirements", "url": "https://www.iso.org/standard/93011.html", "type": "primary", "date": "2026-09-13" }, { "name": "Beijing science and technology portal, on the ISO new work item", "url": "https://www.ncsti.gov.cn/kjdt/yqdy/yqdt/202604/t20260428_245186.html", "type": "secondary", "date": "2026-04-28" } ], "related_ids": [ "EAD-2026-0006", "EAD-2026-0022", "EAD-2026-0025" ], "tags": [ "ISO 26264", "humanoid robot datasets", "ISO/TC 299", "international standard" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0022", "slug": "iso-iec-5259-data-quality-machine-learning-not-robot-specific", "title": "The published international standards on data quality for machine learning, the ISO/IEC 5259 series, are general and name no robot", "kind": "standard", "verdict": "verified", "jurisdiction": "GLOBAL", "answer": "ISO/IEC 5259-1:2024, the first part of the data quality for analytics and machine learning series, was published in July 2024 by ISO/IEC JTC 1/SC 42, with parts 2 to 4 the same year and part 5 in 2025 per the catalog package listing. They frame data quality across the data life cycle for any machine learning use. Europe's AI standards committee, CEN and CENELEC JTC 21, lists datasets among its subjects and names no robot data item.", "key_facts": [ "ISO catalog entry 81088: ISO/IEC 5259-1:2024, status published, publication date July 2024, edition 1, 19 pages, ISO/IEC JTC 1/SC 42. The catalog's package listing bundles parts 1 to 5.", "The series covers overview and terminology, quality measures, quality management requirements, a process framework and a governance framework; none of it is written for robot recordings.", "CEN and CENELEC JTC 21's own page: its work includes standards on datasets, bias, computer vision, cybersecurity, robustness, logging and natural language processing. No robot data work item is named.", "The EU Rolling Plan for ICT Standardisation 2026 entry on robotics and autonomous systems lists no robot dataset work item." ], "figures": [], "for_a_robot_maker": "A generic data quality standard exists and can be written into a data contract today, in any jurisdiction, but it will not tell a supplier what a good hour of manipulation data is. The robot specific answer is being written in ISO/TC 299 and in China's standards bodies, and Europe's AI standards committee is not writing one.", "sources": [ { "name": "ISO catalog, ISO/IEC 5259-1:2024", "url": "https://www.iso.org/standard/81088.html", "type": "primary", "date": "2026-09-13" }, { "name": "CEN and CENELEC, artificial intelligence topic page", "url": "https://www.cencenelec.eu/areas-of-work/cen-cenelec-topics/artificial-intelligence/", "type": "primary", "date": "2026-09-13" } ], "related_ids": [ "EAD-2026-0021", "EAD-2026-0005" ], "tags": [ "ISO/IEC 5259", "data quality", "machine learning", "JTC 21" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0023", "slug": "open-x-embodiment-dataset", "title": "Open X-Embodiment: over a million real robot trajectories from 22 embodiments, pooled from 60 datasets, under CC BY 4.0", "kind": "supply", "verdict": "verified", "jurisdiction": "GLOBAL", "answer": "Open X-Embodiment, released by Google DeepMind with 21 institutions in October 2023, pools 60 existing robot datasets from 34 labs into over a million real robot trajectories across 22 robot embodiments, 527 skills and 160,266 tasks. The code is Apache 2.0 and the other materials Creative Commons Attribution 4.0, so it is usable commercially. The million figure is from the project page; the paper abstract gives the embodiments, skills and tasks.", "key_facts": [ "Project page: over one million real robot trajectories, 22 robot embodiments, 60 existing datasets from 34 robotic research labs.", "Paper abstract, arXiv 2310.08864, first submitted 13 October 2023: 22 different robots, 21 institutions, 527 skills and 160,266 tasks.", "Licence, per the repository: code under Apache License 2.0, all other materials under Creative Commons Attribution 4.0 International." ], "figures": [ { "label": "Real robot trajectories, at least, per the project page", "value": 1000000, "unit": "trajectories", "as_of": "2023-10-13", "source": 0 }, { "label": "Robot embodiments", "value": 22, "unit": "embodiments", "as_of": "2023-10-13", "source": 1 }, { "label": "Skills", "value": 527, "unit": "skills", "as_of": "2023-10-13", "source": 1 } ], "for_a_robot_maker": "This is the open, commercially usable pool, and it is a federation of academic datasets recorded for other purposes, which is why a training ground that records to one specification is worth building. A maker can start from it; nobody on this ledger claims a product was trained on it alone.", "sources": [ { "name": "Open X-Embodiment project page", "url": "https://robotics-transformer-x.github.io/", "type": "primary", "date": "2023-10-13" }, { "name": "Open X-Embodiment paper, arXiv 2310.08864", "url": "https://arxiv.org/abs/2310.08864", "type": "primary", "date": "2023-10-13" }, { "name": "Google DeepMind, open_x_embodiment repository", "url": "https://github.com/google-deepmind/open_x_embodiment", "type": "primary", "date": "2023-10-13" } ], "related_ids": [ "EAD-2026-0013", "EAD-2026-0033", "EAD-2026-0024" ], "tags": [ "Open X-Embodiment", "open dataset", "cross embodiment", "CC BY 4.0" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0024", "slug": "human-egocentric-video-as-robot-training-proxy", "title": "The largest open supply of hours is human, not robot: Ego4D holds over 3,670 hours of first person video", "kind": "supply", "verdict": "verified", "jurisdiction": "GLOBAL", "answer": "The open datasets with the most hours are of people, recorded head mounted: Ego4D, released 17 February 2022, over 3,670 hours from 923 participants in 74 locations; Ego-Exo4D, released 13 December 2023, 1,286.3 hours of skilled activity from 740 camera wearers; EPIC-KITCHENS-100, 100 hours in 45 kitchens under a non commercial license. Ego4D's license is a set of per university agreements, not a Creative Commons license, and some permit commercial development.", "key_facts": [ "Ego4D site: over 3,670 hours of daily life activity video, 923 unique participants, 74 locations in 9 countries, released 17 February 2022.", "Ego4D's license document is a draft set of per institution agreements; Bristol's permits commercial or non commercial product development.", "Ego-Exo4D: 1,286.3 hours of skilled human activities, 740 camera wearers, 13 cities, released 13 December 2023; the agreement covers research and commercial use.", "EPIC-KITCHENS-100: 100 hours of unscripted egocentric footage from 45 kitchens, 89.9 thousand action segments, Creative Commons Attribution NonCommercial 4.0, with commercial licensing by request." ], "figures": [ { "label": "Hours in Ego4D, at least", "value": 3670, "unit": "hours", "as_of": "2022-02-17", "source": 0 }, { "label": "Hours in Ego-Exo4D", "value": 1286.3, "unit": "hours", "as_of": "2023-12-13", "source": 1 }, { "label": "Hours in EPIC-KITCHENS-100", "value": 100, "unit": "hours", "as_of": "2020-07-01", "source": 2 } ], "for_a_robot_maker": "Human hours are a proxy, and the proxy is bigger than the real thing: the two Ego4D sets alone hold more hours than the largest open robot dataset. Whether they count toward the ten million is the open question on synthetic and proxy hours. Their licenses differ set by set, and one of them is a draft.", "sources": [ { "name": "Ego4D project site", "url": "https://ego4d-data.org/", "type": "primary", "date": "2022-02-17" }, { "name": "Ego-Exo4D project site", "url": "https://ego-exo4d-data.org/", "type": "primary", "date": "2023-12-13" }, { "name": "EPIC-KITCHENS project site", "url": "https://epic-kitchens.github.io/", "type": "primary", "date": "2026-09-13" } ], "related_ids": [ "EAD-2026-0025", "EAD-2026-0023", "EAD-2026-0004" ], "tags": [ "Ego4D", "egocentric video", "human demonstration", "open dataset" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0025", "slug": "do-synthetic-hours-count", "title": "Open question: do synthetic hours count? NVIDIA generated the equivalent of 6,500 human hours in eleven, and no standard says what an hour is", "kind": "question", "verdict": "open", "jurisdiction": "GLOBAL", "answer": "NVIDIA's press release of 18 March 2025 says it generated 780,000 synthetic trajectories in eleven hours, which it calls the equivalent of 6,500 hours of human demonstration data, and that mixing them with real data improved its humanoid model by 40 percent; its technical blog the same day says over 750,000. The CAICT need estimate is in real hours. ISO's draft humanoid dataset standard covers training, validation and test sets without saying how synthetic and real hours compare.", "key_facts": [ "NVIDIA newsroom, 18 March 2025: 780,000 synthetic trajectories, the equivalent of 6,500 hours or nine continuous months of human demonstration data, in eleven hours; a 40 percent improvement over real data alone.", "NVIDIA developer blog, same day: over 750,000 synthetic trajectories in eleven hours. The two company sources disagree by thirty thousand.", "The CAICT estimate of ten million hours is stated as effective real data. Whether a synthetic hour is a fraction of a real one, or none, or more, is not defined anywhere the ledger found.", "ISO/CD 26264-1's abstract speaks of training, validation and test datasets and of quality requirements; the catalog entry does not mention synthetic data." ], "figures": [ { "label": "Synthetic trajectories generated in eleven hours, per the press release", "value": 780000, "unit": "trajectories", "as_of": "2025-03-18", "source": 0 }, { "label": "Human demonstration hours the company says that equals", "value": 6500, "unit": "hours", "as_of": "2025-03-18", "source": 0 } ], "for_a_robot_maker": "If synthetic hours count at par, the ten million hour gap closes in weeks of compute and the training grounds are a detour. If they count for little, the grounds are the moat. Every buyer of hours is betting on the exchange rate, and no standards body has published one. A maker should keep the two kinds of hours separately counted in its own records until someone does.", "sources": [ { "name": "NVIDIA newsroom, Isaac GR00T N1 release", "url": "https://nvidianews.nvidia.com/news/nvidia-isaac-gr00t-n1-open-humanoid-robot-foundation-model-simulation-frameworks", "type": "primary", "date": "2025-03-18" }, { "name": "NVIDIA developer blog, Isaac GR00T N1", "url": "https://developer.nvidia.com/blog/accelerate-generalist-humanoid-robot-development-with-nvidia-isaac-gr00t-n1/", "type": "primary", "date": "2025-03-18" }, { "name": "ISO catalog, ISO/CD 26264-1", "url": "https://www.iso.org/standard/93011.html", "type": "primary", "date": "2026-09-13" } ], "related_ids": [ "EAD-2026-0034", "EAD-2026-0004", "EAD-2026-0021", "EAD-2026-0024" ], "tags": [ "synthetic data", "open question", "NVIDIA", "hours" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0026", "slug": "who-owns-the-hours-a-bought-robot-records", "title": "Open question: who owns the hours a robot records at a customer's site? Only the EU has a rule, and it is about access, not ownership", "kind": "question", "verdict": "open", "jurisdiction": "GLOBAL", "answer": "A deployed robot records its customer's factory every shift. In the European Union the Data Act gives the customer access to that data and lets them pass it to a third party who may not build a competing robot with it, without saying who owns it. The ledger found no equivalent rule in China or the United States. Whether a maker may train its next model on a customer's floor is a contract question everywhere.", "key_facts": [ "Data Act articles 4 and 5: the user may access readily available data and have it shared with a third party; article 6(2)(e) bars the third party from a competing product. The Regulation does not use the word ownership.", "China's cross border provisions and network data regulations govern classification and export; neither allocates a connected product's data between maker and customer.", "No United States federal rule on robot training data exists on this ledger, and the state biometric laws speak to people in the footage, not to the customer's rights in it." ], "figures": [], "for_a_robot_maker": "Whatever the contract says is the rule, in every jurisdiction but one, and in that one the customer can hand the hours to your competitor's integrator. A maker whose business model is learning from deployed fleets should write the training right into the sale, say so in plain words to the customer, and in Europe design the product so the customer's access is real.", "sources": [ { "name": "Regulation (EU) 2023/2854, the Data Act, Official Journal text", "url": "https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:32023R2854", "type": "primary", "date": "2023-12-22" }, { "name": "State Council, Network Data Security Management Regulations, Order 790", "url": "https://www.gov.cn/zhengce/content/202409/content_6977766.htm", "type": "primary", "date": "2024-09-24" } ], "related_ids": [ "EAD-2026-0014", "EAD-2026-0011", "EAD-2026-0028" ], "tags": [ "open question", "data ownership", "deployed fleets", "Data Act" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0027", "slug": "is-a-person-in-robot-training-footage-a-data-subject", "title": "Open question: does a person in a training ground's footage have a say? Three regimes, three different answers, none written for robots", "kind": "question", "verdict": "open", "jurisdiction": "GLOBAL", "answer": "A worker recorded by a robot is a data subject under the GDPR, with a right to information and a lawful basis owed to them. Under China's Personal Information Protection Law, sending their footage abroad needs separate consent and notice of the recipient. In the United States, Washington's biometric law excludes video recordings from the definition altogether, and Illinois' text could not be reached. No rule anywhere was written for a person whose movements train a machine.", "key_facts": [ "GDPR and the EDPB video guidelines: footage of an identifiable person is personal data; it is biometric only when processed to identify. Workers are owed a lawful basis and information, and consent is a weak basis at work.", "PIPL article 39, per the Stanford DigiChina translation: before sending personal information abroad the handler notifies the individual of the recipient's name, contact, purpose, method and categories, and obtains separate consent.", "Washington RCW 19.375.010(1) excludes a physical or digital photograph, video or audio recording or data generated from it from the definition of biometric identifier; 19.375.020 requires notice and consent to enrol one for a commercial purpose.", "Whether a person's gait, grip and habits, learned by a model, remain their personal data after the footage is deleted is not addressed by any of the three." ], "figures": [], "for_a_robot_maker": "The safe course is the same in all three places and stricter than any of them requires: tell the people you record, give them a way to say no, keep faces and names out of the training set, and keep the raw footage only as long as the standard you test against needs. What a model retains of a person is the question the next decade of this ledger will fill in.", "sources": [ { "name": "European Data Protection Board, Guidelines 3/2019 on processing of personal data through video devices", "url": "https://www.edpb.europa.eu/sites/default/files/files/file1/edpb_guidelines_201903_video_devices_en_0.pdf", "type": "primary", "date": "2020-01-29" }, { "name": "DigiChina, Stanford, translation of the Personal Information Protection Law", "url": "https://digichina.stanford.edu/work/translation-personal-information-protection-law-of-the-peoples-republic-of-china-effective-nov-1-2021/", "type": "secondary", "date": "2021-11-01" }, { "name": "Washington State Legislature, RCW 19.375.010", "url": "https://app.leg.wa.gov/rcw/default.aspx?cite=19.375.010", "type": "primary", "date": "2026-09-13" } ], "related_ids": [ "EAD-2026-0017", "EAD-2026-0031", "EAD-2026-0011" ], "tags": [ "open question", "data subjects", "workers", "biometric", "consent" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0028", "slug": "no-us-federal-standard-or-rule-on-robot-training-data", "title": "No United States federal standard or rule governs robot training data; one NIST project names datasets as a deliverable", "kind": "standard", "verdict": "absent", "jurisdiction": "US", "answer": "The ledger searched the White House AI Action Plan of July 2025, NIST's AI Risk Management Framework and robotics pages, the National Science Foundation's robotics programs, DARPA's 2026 physical intelligence request and the Congressional record for a federal instrument on robot training data. None exists. NIST's Physical AI and Data Generation for Robotics project lists datasets among its deliverables and has released one of manufacturing objects; that is the whole federal footprint.", "search": "Searched on 13 September 2026: the AI Action Plan PDF at whitehouse.gov, text extracted; nist.gov, the AI RMF page and the Physical AI and Data Generation for Robotics project page updated 24 April 2026; nsf.gov, the National Robotics Initiative 3.0 sunset letter NSF 22-081 and the Foundational Research in Robotics page; darpa.mil, the 29 April 2026 rethinking robotics request for information; govinfo.gov for bills naming humanoid robots; whitehouse.gov for any 2026 executive action on robotics.", "key_facts": [ "NIST project page, updated 24 April 2026: develop metrics, test methods, standards, software, prototypes and datasets to promote the adoption of AI enhanced robotics; disseminated a dataset of manufacturing objects and assemblies.", "The AI Action Plan names robotics among physical world innovations and asks for a convening on robotics and drone supply chains; its dataset language is about scientific datasets, not robot training data.", "NSF sunset the National Robotics Initiative 3.0 in May 2022; the successor program page carries no dataset language. DARPA's April 2026 request on physical intelligence has none either.", "The NIST AI Risk Management Framework page does not mention robotics, embodied or physical AI." ], "figures": [], "for_a_robot_maker": "In the United States the data side of embodied AI is unregulated at the federal level and unfunded as public infrastructure. A maker there answers to state biometric law for the people in the footage, to export controls that do not name the data, and to contract for everything else. The nearest thing to a national data effort is a manufacturing institute's data call, on this ledger as a separate record.", "sources": [ { "name": "NIST, Physical AI and Data Generation for Robotics", "url": "https://www.nist.gov/programs-projects/physical-ai-and-data-generation-robotics", "type": "primary", "date": "2026-04-24" }, { "name": "The White House, America's AI Action Plan", "url": "https://www.whitehouse.gov/wp-content/uploads/2025/07/Americas-AI-Action-Plan.pdf", "type": "primary", "date": "2025-07-23" }, { "name": "National Science Foundation, sunset of the National Robotics Initiative 3.0, NSF 22-081", "url": "https://www.nsf.gov/funding/information/dcl-nri-sunset-announcement/nsf22-081", "type": "primary", "date": "2022-05-03" }, { "name": "DARPA, rethinking robotics with physical intelligence", "url": "https://www.darpa.mil/news/2026/rethinking-robotics", "type": "primary", "date": "2026-04-29" } ], "related_ids": [ "EAD-2026-0029", "EAD-2026-0030", "EAD-2026-0032" ], "tags": [ "United States", "absent", "NIST", "AI Action Plan", "robot training data" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0029", "slug": "arm-institute-ai-data-foundry-manufacturing-robot-data", "title": "The US answer to a training ground is a manufacturing institute's data call: the ARM Institute's AI Data Foundry", "kind": "supply", "verdict": "verified", "jurisdiction": "US", "answer": "The Advanced Robotics for Manufacturing Institute's 2025 AI project call, published 5 March 2025, says a key barrier to AI in manufacturing robotics is a lack of real, quality data to train models, and frames an AI Data Foundry and a national repository for AI based manufacturing resources. It is a project call, not a site with robots in it, and no volume of data is stated. NIST's physical AI project is the other federal deliverable naming datasets.", "key_facts": [ "ARM Institute, 5 March 2025: a key barrier to the adoption of AI within robotics for manufacturing is a lack of real, quality data to train AI models.", "The call names an AI Data Foundry and a National Repository for AI Based Manufacturing Resources as the program's frame.", "No robot count, hour count or site is given; the instrument funds projects that would produce data rather than recording it." ], "figures": [], "for_a_robot_maker": "A US maker in manufacturing can apply for money to make data; it cannot walk into a public training ground. The difference between a call and a site is the difference between the two sides of this ledger, and it is the whole of the gap the founding article named.", "sources": [ { "name": "ARM Institute, 2025 AI project call", "url": "https://arminstitute.org/news/ai-project-call-2025/", "type": "primary", "date": "2025-03-05" }, { "name": "NIST, Physical AI and Data Generation for Robotics", "url": "https://www.nist.gov/programs-projects/physical-ai-and-data-generation-robotics", "type": "primary", "date": "2026-04-24" } ], "related_ids": [ "EAD-2026-0028", "EAD-2026-0003" ], "tags": [ "ARM Institute", "United States", "manufacturing", "AI Data Foundry" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0030", "slug": "humanoid-robot-act-senate-bill-data-of-us-persons", "title": "A Senate bill, not a law: the Humanoid ROBOT Act of 2025 would order a report on data of US persons obtained through humanoid robots", "kind": "rule", "verdict": "verified", "jurisdiction": "US", "answer": "S.3275, the Humanoid ROBOT Act of 2025, introduced on 20 November 2025 by Senators Cassidy and Coons, would bar federal procurement of covered humanoid robots, route certain transactions through CFIUS review, and require a Defense Department report including privacy and data security threats relating to the use of data of United States persons obtained through humanoid robots. A House bill of April 2026 on security robotics names humanoids and carries no training data provision. Neither has passed.", "key_facts": [ "S.3275, introduced 20 November 2025: procurement restrictions, CFIUS review, and a report that must cover privacy and data security threats relating to the use of data of United States persons obtained through humanoid robots.", "H.R.8189, the American Security Robotics Act of 2026, introduced 2 April 2026, covers unmanned ground vehicles including a humanoid robot; no training data provision.", "Both are introduced bills as of 13 September 2026. Neither is law, and the ledger prints them as what they are." ], "figures": [], "for_a_robot_maker": "The first federal text to name the data a humanoid collects treats it as a security question about foreign robots, not a governance question about training. If it passes, the report it orders would be the first federal document on the subject. A maker selling humanoids in the United States should watch the procurement and CFIUS provisions more than the data one.", "sources": [ { "name": "Congress, S.3275 Humanoid ROBOT Act of 2025, as introduced", "url": "https://www.govinfo.gov/content/pkg/BILLS-119s3275is/html/BILLS-119s3275is.htm", "type": "primary", "date": "2025-11-20" }, { "name": "Congress, H.R.8189 American Security Robotics Act of 2026, as introduced", "url": "https://www.govinfo.gov/content/pkg/BILLS-119hr8189ih/html/BILLS-119hr8189ih.htm", "type": "primary", "date": "2026-04-02" } ], "related_ids": [ "EAD-2026-0028", "EAD-2026-0032" ], "tags": [ "Humanoid ROBOT Act", "Congress", "United States", "CFIUS" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0031", "slug": "us-state-biometric-law-video-recordings", "title": "US state biometric laws: Washington excludes video recordings from the definition, Texas needs consent for face geometry, and the Illinois text could not be reached", "kind": "rule", "verdict": "verified", "jurisdiction": "US", "answer": "Washington's biometric statute, RCW 19.375, bars enrolling a biometric identifier in a database for a commercial purpose without notice and consent, and excludes a photograph, video or audio recording or data generated from it from the definition. Texas' statute covers a record of hand or face geometry, requires consent and sets a civil penalty of up to 25,000 dollars per violation, per an unofficial mirror. The Illinois statute could not be fetched from the legislature.", "key_facts": [ "RCW 19.375.020(1): a person may not enrol a biometric identifier in a database for a commercial purpose without first providing notice, obtaining consent, and providing a mechanism to prevent later use.", "RCW 19.375.010(1): biometric identifier does not include a physical or digital photograph, video or audio recording or data generated therefrom.", "Texas Business and Commerce Code 503.001, per the only mirror that served the text: capture of a biometric identifier for a commercial purpose needs informing and consent, and destruction within a year of the purpose expiring.", "The same Texas section sets a civil penalty of not more than 25,000 dollars per violation, enforced by the attorney general.", "Illinois, 740 ILCS 14, and its 2024 amendment: ilga.gov refused every connection and every mirror returned an error, so no Illinois figure is printed." ], "figures": [ { "label": "Texas civil penalty ceiling per violation, per the mirror", "value": 25000, "unit": "dollars", "as_of": "2026-09-13", "source": 2 } ], "for_a_robot_maker": "In the state with the clearest text, a robot's video of a worker is not a biometric identifier until something is derived from it to identify them, which is the same line the GDPR draws. Texas reaches face geometry. Illinois, the state with private lawsuits, is the one whose text this ledger could not read, and a maker recording people there should read it before this ledger can.", "sources": [ { "name": "Washington State Legislature, RCW 19.375.020", "url": "https://app.leg.wa.gov/rcw/default.aspx?cite=19.375.020", "type": "primary", "date": "2026-09-13" }, { "name": "Washington State Legislature, RCW 19.375.010", "url": "https://app.leg.wa.gov/rcw/default.aspx?cite=19.375.010", "type": "primary", "date": "2026-09-13" }, { "name": "Texas Business and Commerce Code 503.001, mirror at texas.public.law", "url": "https://texas.public.law/statutes/tex._bus._and_com._code_section_503.001", "type": "secondary", "date": "2026-09-13" } ], "related_ids": [ "EAD-2026-0027", "EAD-2026-0017" ], "tags": [ "biometric law", "Washington", "Texas", "Illinois BIPA", "United States" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0032", "slug": "us-export-controls-robot-training-data", "title": "No US export rule names robot training data; the AI diffusion rule that reached model weights is under rescission", "kind": "rule", "verdict": "verified", "jurisdiction": "US", "answer": "The Bureau of Industry and Security announced on 13 May 2025 that it would rescind the January 2025 framework for artificial intelligence diffusion and issued guidance on chips; the rescission sits in the regulatory agenda as a final rule stage item, and the Government Accountability Office found in May 2026 that the press release itself was a rule. No Export Administration Regulations entry names robot training data or datasets, and the ledger found none.", "key_facts": [ "BIS, 13 May 2025: rescission of the Biden era artificial intelligence diffusion rule announced, with three guidance actions on advanced computing chips; no mention of datasets or robotics.", "Regulatory agenda entry RIN 0694-AJ90, rescinding the framework for artificial intelligence diffusion, final rule stage, against the interim final rule at 90 FR 4544 of 15 January 2025.", "GAO decision B-337935, 12 May 2026: the press release is a rule for purposes of the Congressional Review Act.", "The January 2025 rule created an export classification for certain closed model weights; nothing in it or after it classifies robot training data." ], "figures": [], "for_a_robot_maker": "A US maker can ship recorded hours abroad without an export license, and a foreign maker can buy them, as far as the EAR is concerned today. The controlled item is compute and, on paper, certain model weights. Whether a trained robot policy is a controlled weight is a question the rescission left open rather than answered.", "sources": [ { "name": "Bureau of Industry and Security, rescission announcement", "url": "https://www.bis.gov/press-release/department-commerce-announces-rescission-biden-era-artificial-intelligence-diffusion-rule-strengthens", "type": "primary", "date": "2025-05-13" }, { "name": "Office of Information and Regulatory Affairs, regulatory agenda entry RIN 0694-AJ90", "url": "https://www.reginfo.gov/public/do/eAgendaViewRule?pubId=202504&RIN=0694-AJ90", "type": "primary", "date": "2026-09-13" }, { "name": "Government Accountability Office, decision B-337935", "url": "https://www.gao.gov/products/b-337935", "type": "primary", "date": "2026-05-12" } ], "related_ids": [ "EAD-2026-0028", "EAD-2026-0019" ], "tags": [ "export controls", "BIS", "EAR", "United States", "model weights" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0033", "slug": "droid-dataset-stanford", "title": "DROID: 76,000 demonstration trajectories, 350 hours, one robot arm, under CC BY 4.0", "kind": "supply", "verdict": "verified", "jurisdiction": "US", "answer": "DROID, released by Stanford with twelve other institutions in March 2024, holds 76,000 demonstration trajectories, or 350 hours of interaction data, recorded on a Franka Panda arm across 564 scenes and 86 tasks, under a Creative Commons Attribution 4.0 license. It is the cleanest open figure the ledger has for how many hours a large academic collection effort yields: three hundred and fifty.", "key_facts": [ "Project page: 76,000 demonstration trajectories or 350 hours of interaction data, 564 scenes, 86 tasks; the paper abstract gives 84 tasks.", "Paper, arXiv 2403.12945, first submitted 19 March 2024: the full dataset is open sourced under CC BY 4.0.", "Thirteen institutions recorded it on one robot type, which is the opposite design to Open X-Embodiment's pooling of many bodies." ], "figures": [ { "label": "Demonstration trajectories", "value": 76000, "unit": "trajectories", "as_of": "2024-03-19", "source": 0 }, { "label": "Hours of interaction data", "value": 350, "unit": "hours", "as_of": "2024-03-19", "source": 0 }, { "label": "Scenes", "value": 564, "unit": "scenes", "as_of": "2024-03-19", "source": 0 } ], "for_a_robot_maker": "Three hundred and fifty hours from thirteen universities is the scale of the open academic supply per project. Against a need estimate in the millions of hours, it says why the money is going into sites that record around the clock, and it is usable commercially, which the largest Chinese set is not.", "sources": [ { "name": "DROID project page", "url": "https://droid-dataset.github.io/", "type": "primary", "date": "2024-03-19" }, { "name": "DROID paper, arXiv 2403.12945", "url": "https://arxiv.org/abs/2403.12945", "type": "primary", "date": "2024-03-19" } ], "related_ids": [ "EAD-2026-0023", "EAD-2026-0013" ], "tags": [ "DROID", "open dataset", "Stanford", "CC BY 4.0" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0034", "slug": "nvidia-synthetic-trajectories-gr00t", "title": "NVIDIA's synthetic supply: 780,000 trajectories in eleven hours, and an open simulated set of about 273,000", "kind": "supply", "verdict": "verified", "jurisdiction": "US", "answer": "NVIDIA said on 18 March 2025 that it generated 780,000 synthetic trajectories in eleven hours for its humanoid foundation model, and publishes a simulated cross embodiment dataset of about 273,000 trajectories under Creative Commons Attribution 4.0. A company post on Hugging Face describes a separate physical AI dataset of more than 320,000 trajectories and 15 terabytes. Its June 2026 Cosmos release names five open datasets and gives no size for them.", "key_facts": [ "NVIDIA newsroom, 18 March 2025: 780,000 synthetic trajectories generated in eleven hours; the developer blog of the same day says over 750,000.", "Hugging Face dataset card, PhysicalAI Robotics GR00T X Embodiment Sim: about 273,000 simulated trajectories, license CC BY 4.0.", "Hugging Face blog post authored by NVIDIA: 15 terabytes representing more than 320,000 trajectories for robotics training, plus up to 1,000 OpenUSD assets.", "The Cosmos 3 developer blog of June 2026 names five open synthetic datasets and states no hours or trajectory counts." ], "figures": [ { "label": "Synthetic trajectories generated in eleven hours, per the press release", "value": 780000, "unit": "trajectories", "as_of": "2025-03-18", "source": 0 }, { "label": "Simulated trajectories in the open GR00T cross embodiment set, approximately", "value": 273000, "unit": "trajectories", "as_of": "2026-09-13", "source": 1 }, { "label": "Trajectories in the physical AI dataset described on Hugging Face, at least", "value": 320000, "unit": "trajectories", "as_of": "2026-09-13", "source": 2 } ], "for_a_robot_maker": "Synthetic supply is the one kind that scales with compute rather than with floor space and robots, and it is licensed for commercial use. Whether it substitutes for recorded hours is the open question on this ledger, and the company's own claim is a 40 percent improvement when mixed with real data, not a replacement of it.", "sources": [ { "name": "NVIDIA newsroom, Isaac GR00T N1 release", "url": "https://nvidianews.nvidia.com/news/nvidia-isaac-gr00t-n1-open-humanoid-robot-foundation-model-simulation-frameworks", "type": "primary", "date": "2025-03-18" }, { "name": "NVIDIA, PhysicalAI Robotics GR00T X Embodiment Sim dataset card, Hugging Face", "url": "https://huggingface.co/datasets/nvidia/PhysicalAI-Robotics-GR00T-X-Embodiment-Sim", "type": "primary", "date": "2026-09-13" }, { "name": "NVIDIA, physical AI dataset post on the Hugging Face blog", "url": "https://github.com/huggingface/blog/blob/main/nvidia-physical-ai.md", "type": "primary", "date": "2026-09-13" }, { "name": "NVIDIA developer blog, Cosmos 3", "url": "https://developer.nvidia.com/blog/develop-physical-ai-reasoning-world-and-action-models-with-nvidia-cosmos-3/", "type": "primary", "date": "2026-06-30" } ], "related_ids": [ "EAD-2026-0025", "EAD-2026-0023" ], "tags": [ "NVIDIA", "synthetic data", "GR00T", "Cosmos", "open dataset" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0035", "slug": "japan-nedo-robotics-data-platform-airoa", "title": "Japan funds a robotics data platform and a physical AI foundation model program through NEDO, with no published data volume", "kind": "supply", "verdict": "verified", "jurisdiction": "JP", "answer": "Japan's New Energy and Industrial Technology Development Organization selected AIRoA in September 2025 to develop data platforms for generative AI foundation models in robotics, running from 1 October 2025 to 31 August 2029 and covering data collection, model development and validation. On 30 June 2026 NEDO named Noetra with AIST for a multimodal foundation model program for AI robots and physical AI, fiscal 2026 to 2030. Neither notice states a volume of data.", "key_facts": [ "AIRoA, 26 September 2025: selected by NEDO for development of data platforms for generative AI foundation models in the robotics field, 1 October 2025 to 31 August 2029, covering data collection, model development and validation.", "NEDO, 30 June 2026: the multimodal foundation model program for AI robots and physical AI, fiscal 2026 to 2030, drew 15 applications; Noetra and AIST were selected. The awardee's release speaks of protecting site data.", "Budget figures for the program appear only in trade press and on ministry pages that refused the fetch, so the ledger prints none." ], "figures": [ { "label": "Applications to the NEDO foundation model program", "value": 15, "unit": "applications", "as_of": "2026-06-30", "source": 1 } ], "for_a_robot_maker": "Japan is the one jurisdiction outside China on this ledger with a public program whose scope includes collecting robot data, run through an agency rather than a strategy document. Its awardee is a consortium, its volume is unpublished, and a maker in Japan should read the AIRoA platform as the door to it.", "sources": [ { "name": "AIRoA, on its selection by NEDO", "url": "https://www.airoa.org/updates/20250926", "type": "primary", "date": "2025-09-26" }, { "name": "NEDO, call page for the AI robot and physical AI multimodal foundation model program", "url": "https://www.nedo.go.jp/koubo/CD3_100431.html", "type": "primary", "date": "2026-06-30" }, { "name": "Noetra, release on its selection", "url": "https://www.noetra.co.jp/pressrelease20260630-2", "type": "primary", "date": "2026-06-30" } ], "related_ids": [ "EAD-2026-0036", "EAD-2026-0016" ], "tags": [ "Japan", "NEDO", "AIRoA", "robotics data platform" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0036", "slug": "korea-ai-hub-robot-grasp-dataset-domestic-only", "title": "South Korea's public robot dataset: 209,403 grasp videos that only Korean nationals may request, for non commercial use", "kind": "supply", "verdict": "verified", "jurisdiction": "KR", "answer": "Korea's AI Hub, run by the National Information Society Agency, publishes a robot behaviour dataset for small object grasping of about 209,403 videos, version 1.2 released 4 December 2024. The dataset page says only Korean nationals may apply, and the hub's usage policy limits use to non commercial research and development and requires a separate agreement with the agency to take data abroad. In March 2026 the trade ministry's alliance discussed securing training data for humanoids.", "key_facts": [ "AI Hub dataset 71713, robot behaviour data for small object grasping: about 209,403 videos, version 1.2 released 4 December 2024; the page states that only domestic nationals may request the data.", "AI Hub usage policy: the data may be used only for non commercial research or development, and export abroad needs a separate agreement with the agency.", "Ministry of Trade, Industry and Resources, 5 March 2026: the AI Robot alliance meeting discussed securing and leveraging training data for humanoid robots; no program detail given." ], "figures": [ { "label": "Grasp videos in the AI Hub dataset, approximately", "value": 209403, "unit": "videos", "as_of": "2024-12-04", "source": 0 } ], "for_a_robot_maker": "A public dataset that a foreign maker cannot download and a domestic maker cannot sell a product on is a research asset, not supply. It is the clearest example on the ledger of a jurisdiction that built the data and fenced it, and the ministry's March 2026 discussion suggests the fence is under review.", "sources": [ { "name": "AI Hub, robot behaviour data for small object grasping", "url": "https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&dataSetSn=71713", "type": "primary", "date": "2024-12-04" }, { "name": "AI Hub, usage policy", "url": "https://www.aihub.or.kr/intrcn/guid/usagepolicy.do?currMenu=151&topMenu=105", "type": "primary", "date": "2026-09-13" }, { "name": "Ministry of Trade, Industry and Resources, AI Robot alliance meeting", "url": "https://english.motir.go.kr/eng/article/EATCLdfa319ada/2521/view?pageIndex=1&bbsCdN=2", "type": "primary", "date": "2026-03-05" } ], "related_ids": [ "EAD-2026-0035", "EAD-2026-0013" ], "tags": [ "South Korea", "AI Hub", "public dataset", "domestic only" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" }, { "id": "EAD-2026-0037", "slug": "uk-no-program-or-standard-on-robot-training-data", "title": "The United Kingdom funds robot adoption, 38 million pounds of it, and nothing on robot training data", "kind": "standard", "verdict": "absent", "jurisdiction": "UK", "answer": "The ledger searched the Innovate UK Robotics Adoption Hubs competition of February 2026, worth up to 38 million pounds, its 2 million pound convening body, the government's AI Opportunities Action Plan one year on, and the British Standards Institution for any instrument on robot training data or embodied AI datasets. None exists. The adoption competitions never mention training data, datasets or embodied AI.", "search": "Searched on 13 September 2026: apply-for-innovation-funding.service.gov.uk competitions 2408 and 2409, open 24 February to 15 April 2026; GOV.UK, AI Opportunities Action Plan one year on, 29 January 2026; bsigroup.com and aistandardshub.org for robot data or embodied AI dataset standards; the UK AI Hardware Plan of 8 June 2026 as indexed, which mentions sensing for embodied AI and no data program.", "key_facts": [ "Innovate UK, Robotics Adoption Hubs, open 24 February 2026, closed 15 April 2026, up to 38 million pounds; the central convening body competition, up to 2 million pounds. Neither page mentions training data, datasets or embodied AI.", "AI Opportunities Action Plan, one year on, 29 January 2026: the only robotics line is 40 million pounds for a network of robotics adoption hubs and an expansion of Made Smarter.", "No BSI standard or work item on robot training data was found." ], "figures": [ { "label": "Robotics Adoption Hubs competition, up to", "value": 38000000, "unit": "pounds", "as_of": "2026-02-24", "source": 0 }, { "label": "Robotics adoption funding named in the action plan update", "value": 40000000, "unit": "pounds", "as_of": "2026-01-29", "source": 1 } ], "for_a_robot_maker": "British public money buys robots for factories; it does not record what they see. A maker in the United Kingdom collecting hours answers to the UK GDPR for the people in them and to nothing written for the data itself, and will find no public site or dataset to draw on.", "sources": [ { "name": "Innovate UK, Robotics Adoption Hubs competition", "url": "https://apply-for-innovation-funding.service.gov.uk/competition/2408/overview", "type": "primary", "date": "2026-02-24" }, { "name": "GOV.UK, AI Opportunities Action Plan: one year on", "url": "https://www.gov.uk/government/publications/ai-opportunities-action-plan-one-year-on", "type": "primary", "date": "2026-01-29" } ], "related_ids": [ "EAD-2026-0016", "EAD-2026-0028" ], "tags": [ "United Kingdom", "absent", "Robotics Adoption Hubs", "Innovate UK" ], "date_added": "2026-09-13", "last_verified": "2026-09-13", "last_modified": "2026-09-13" } ] }