Market Maps

Physical AI's Research Labs and National Agencies, Mapped

The physical AI landscape map tracks the companies. This one tracks who trains the people and funds the science behind them. Two groups: the university and corporate labs producing the field's core research, and the government agencies whose grants, competitions, and industrial policy decide what gets built next.
Published2026-07-0810 min read
24
Research labs
22
National agencies
11
Countries represented
2026-07
Data as of

46 organizations, 11 countries — representative, not exhaustive. Country codes show HQ location. Logos are favicons from each organization's official site; where none is available, only the name is shown. As of 2026-07.

01How to read this map: research, not commerce

This map covers the layer beneath the companies: the labs producing the field's core research — foundation-model architectures, control algorithms, new sensing hardware — before a startup or product team turns it into something sold, and the agencies whose grants, competitions, and industrial policy decide which of that research gets funded at all.

Inclusion follows the same bar as the companies map: a verifiable, sourced connection to physical AI or robotics — an active lab, a named funding program, a research institute with a citable track record. The list is representative, not exhaustive, chosen for coverage across North America, Europe, Japan, China, and Korea rather than completeness within any one country.

02Research labs

University labs still produce most of the field's foundational research, but the line between academic and corporate research has blurred: several of the labs below sit inside NVIDIA, Google DeepMind, Meta, and Toyota, publishing in the same venues as their university counterparts while also feeding directly into their parent companies' products.

North America

US university labs (MIT CSAIL, Stanford SAIL, CMU Robotics Institute, Berkeley BAIR) trained much of the field's current generation of researchers, and corporate labs — DeepMind's robotics team, Meta FAIR, OpenAI, NVIDIA Research — now sit alongside them, often publishing openly while also feeding their parent companies' commercial models.

MIT CSAILUSUniversity-affiliated lab

MIT's largest research lab, formed in 2003 by merging its Lab for Computer Science and AI Lab; spans robotics, vision, and manipulation research across dozens of groups. MIT CSAIL

Stanford AI Lab (SAIL)USUniversity-affiliated lab

Founded in 1963 by John McCarthy and relaunched as an independent lab in 2004; its Vision and Learning Lab (SVL) drives much of Stanford's robot-perception and manipulation work. Stanford AI Lab

CMU Robotics InstituteUSUniversity-affiliated lab

Founded in 1979 with a $3M Westinghouse grant as the first academic unit devoted solely to robotics; launched the world's first robotics PhD program in 1988. Carnegie Mellon Today

UC Berkeley BAIRUSUniversity-affiliated lab

Berkeley's umbrella AI lab combining vision, ML, and robotics research; its robotics-focused RAIL group (Sergey Levine) is a leading source of reinforcement-learning-based robot control. BAIR

Meta FAIRUSCorporate research lab

Founded in 2013 as Facebook AI Research; its embodied-AI efforts include an open tactile-sensing collaboration with Wonik Robotics on the Allegro Hand. Meta AI

OpenAIUSCorporate research lab

Founded in December 2015; ran a dedicated robotics team through 2021, whose robotic hand "Dactyl" learned to solve a Rubik's Cube one-handed via simulation-trained reinforcement learning. OpenAI

NVIDIA ResearchUSCorporate research lab

NVIDIA's robotics research group, led by Dieter Fox, built Isaac GR00T N1 — announced in March 2025 as, per NVIDIA, the first open foundation model for generalized humanoid reasoning and skills. NVIDIA Newsroom

Toyota Research Institute (TRI)USCorporate research lab

Announced in November 2015 with a $1B, five-year commitment from Toyota; founding CEO Gill Pratt previously ran DARPA's robotics program office, directly linking TRI to the DARPA Robotics Challenge era. Toyota Research Institute

RAI InstituteUSIndependent research institute

Founded in August 2022 by Boston Dynamics founder Marc Raibert as the Boston Dynamics AI Institute, since renamed Robotics & AI Institute; partnered with Boston Dynamics in 2025 to advance Atlas's reinforcement-learning skills. Boston Dynamics

Allen Institute for AI (Ai2)USIndependent research institute

Nonprofit research institute founded in 2014 by Microsoft co-founder Paul Allen; released the open embodied-AI simulation platform AI2-THOR in 2016 and, more recently, the open robotics foundation model MolmoAct. Ai2

NASA JPLUSFederally funded R&D center

Caltech-managed federal lab founded in 1936; has designed, built, and operated all five successful NASA Mars rovers, from Sojourner (1997) to Perseverance, plus the Ingenuity helicopter. NASA JPL

Europe

Europe's strength sits in legged and precision robotics: ETH Zurich's RSL built the ANYmal quadruped, the Max Planck Institute for Intelligent Systems pushes soft and bio-inspired robotics, and Fraunhofer IPA has iterated on service robots since 1989 — a research base that feeds the region's cobot and inspection-robot industry.

Google DeepMindUKCorporate research lab

London-founded, Alphabet-owned lab whose robotics team built Gemini Robotics, adding physical action as an output modality of its Gemini models for direct robot control. Google DeepMind

ETH Zurich RSLCHUniversity-affiliated lab

ETH Zurich's Robotic Systems Lab, led by Marco Hutter, built the ANYmal quadruped and launched "ANYmal Research," an academia-industry collaboration community, in 2017. ETH Zurich RSL

Founded in 2011 with campuses in Stuttgart and Tübingen; its Physical Intelligence department built a magnetically controlled soft "millirobot," about 4mm long, that can walk, crawl, climb, and swim. Max Planck Society

Fraunhofer IPADEIndependent research institute

Stuttgart-based applied-research institute developing the Care-O-bot service-robot family since 1989; Care-O-bot 4 began working as a shopping assistant at a German electronics store in October 2016. Fraunhofer IPA

InriaFRNational research institute

France's national digital-science institute, founded in 1967; its Flowers team built the open-source humanoid Poppy, whose creators went on to found Pollen Robotics (acquired by Hugging Face in April 2025). Pollen Robotics

Founded in 2014 with a £5M gift from Dyson; led by Andrew Davison, inventor of MonoSLAM (2003), the first real-time drift-free visual SLAM system, foundational to today's robot-vision "Spatial AI." Imperial College London

Bristol Robotics LaboratoryUKUniversity-affiliated lab

Founded in 2005 as a joint venture between UWE Bristol and the University of Bristol; the UK's largest co-located multidisciplinary academic robotics research centre, with over 450 researchers. Bristol Robotics Laboratory

China

Chinese labs are the newest entrants here but move fast: BAAI, backed by the science ministry and Beijing's leading universities and tech firms, released the open embodied-AI model family RoboBrain 2.0 in June 2025, and Tsinghua's AI Industry Research institute runs a dedicated embodied-AI and collaborative-robotics program.

Nonprofit institute founded in November 2018, backed by China's Ministry of Science and Technology, the Beijing municipal government, Tsinghua, Peking University, Baidu, and ByteDance; released the open embodied-AI model family RoboBrain 2.0 in June 2025. South China Morning Post

Founded on December 1, 2020 under founding dean Ya-Qin Zhang; its Tsinghua-Dalle Joint Institute of Embodied AI and Collaborative Robotics develops learning-based robot control policies. Tsinghua AIR

Japan

Japan's labs run some of the field's longest lineages: Waseda's humanoid institute traces to the 1973 WABOT-1, and the University of Tokyo's JSK lab and Osaka University's Ishiguro lab have each shaped humanoid and android research for decades — though, as METI's own data shows, that early lead in adoption volume passed to China by 2013.

University of Tokyo JSK LabJPUniversity-affiliated lab

University of Tokyo's Information Systems Engineering Lab, founded in 1978 by Hirochika Inoue; a long-running humanoid-robotics lab now led by Kei Okada, known for musculoskeletal humanoids and open-source robotics software. University of Tokyo

Traces to Ichiro Kato's 1970 WABOT project and is now led by Atsuo Takanishi; WABOT-1 (1973), still displayed on Waseda's Nishi-Waseda campus, is described by the university as the world's first full-scale bipedal humanoid robot. Waseda University

Osaka University Ishiguro LabJPUniversity-affiliated lab

Led by Hiroshi Ishiguro, researching android robotics; under JST's ERATO Ishiguro project (2014–2019) it developed "ERICA," an autonomous dialogue android, jointly announced with JST, Osaka University, ATR, and Kyoto University in 2015. JST

Korea

KAIST's Humanoid Robot Research Center, founded by HUBO creator Jun-Ho Oh, produced the team that won the 2015 DARPA Robotics Challenge — a research lineage now feeding directly into Korea's government-backed humanoid push.

Founded by Jun-Ho Oh, creator of the HUBO humanoid (2004); Team KAIST's DRC-HUBO won the DARPA Robotics Challenge Finals in June 2015, completing all eight tasks among 24 international teams to win the $2M prize. KAIST

03National and government agencies

Government money moves differently from venture capital: it runs through named, multi-year programs — DARPA's robotics challenges, Japan's Moonshot goals, Korea's K-Humanoid Alliance — rather than funding rounds, and it comes with policy strings (safety standards, industrial targets, national strategy documents) that shape the field as much as any single company's product roadmap.

North America

US agencies fund robotics through a mix of defense, science, and standards mandates: DARPA's robotics challenges have repeatedly set the field's benchmarks, NSF and NASA co-fund basic research through the National Robotics Initiative, and NIST maintains the measurement standards that let robot performance claims be compared at all.

DARPAUSDefense R&D agency

US Department of Defense agency, founded 1958; its DARPA Robotics Challenge (2012–2015) for disaster-response humanoids was won at the 2015 finals by South Korea's Team KAIST with the DRC-HUBO robot. DARPA

NISTUSStandards agency

US Department of Commerce agency that develops robotics measurement and performance standards; operates a dedicated Robotics Test Facility in Gaithersburg, Maryland for standardized robot testing. NIST

NSFUSResearch-funding agency

Independent US federal agency funding basic robotics research through the multi-agency National Robotics Initiative, now in its third phase (NRI-3.0), co-sponsored with NASA, USDA, and other agencies. NSF

NASAUSNational research institute

Independent US federal agency whose Software, Robotics, and Simulation Division, established in 1990, oversees intelligent and robotic systems supporting human space exploration. NASA

Europe

The EU funds robotics as part of its wider industrial strategy: Horizon Europe channels hundreds of millions of euros a year into AI and robotics, while national bodies — Germany's research ministry, the UK's UKRI, France's CNRS — run their own parallel programs.

The EU's 2021–2027 research framework programme (€95.5B budget); its 2025–26 call round allocated €307.3M to AI and related technologies, including €221.8M for trustworthy AI, robotics, quantum, and photonics. European Commission

BMFTRDEGovernment ministry

Germany's federal research ministry, renamed from BMBF in May 2025 when its education portfolio was spun off and space policy added; its robotics action plan funds a decentralized "Robotics Institute Germany" network with up to €20M over four years. BMFTR

UKRIUKResearch-funding agency

UK government funding body established in 2018 by merging the seven Research Councils, Innovate UK, and Research England; its Robotics Adoption Hubs call offers up to £38M to help organizations adopt robotics. UKRI

CNRSFRNational research institute

France's largest public research organization, founded in 1939; its LAAS-CNRS institute built HILARE in 1977, regarded as the first French autonomous mobile robot able to navigate an unknown environment. LAAS-CNRS

Fraunhofer-GesellschaftDEIndependent research institute

Founded in Munich in 1949; Europe's largest applied-research organization, operating over 74 institutes across Germany — including IPA — that bridge university research and industry. Fraunhofer-Gesellschaft

China

China's robotics policy runs through MIIT's five-year industrial plans, which set explicit revenue and robot-density targets, backed by CAS-affiliated national institutes like CASIA and competitive-grant funding from NSFC.

China's industrial-policy ministry, established in 2008; jointly issued the 14th Five-Year Plan for the Robotics Industry in December 2021, targeting over 20% annual revenue growth and doubled robot density by 2025. gov.cn

Founded in October 1956 as China's first national automation institute, under the Chinese Academy of Sciences; its "Q family" humanoid prototypes have been publicly demonstrated performing instruction-following manipulation tasks. CGTN

Established February 14, 1986; China's principal agency funding competitive, peer-reviewed basic and frontier research, including grant lines for embodied intelligence. NSFC

Japan

Japan runs some of the world's largest named robotics funding programs: METI sets industrial policy, NEDO and JST administer the money, AIST does applied research in-house, and the Cabinet Office's Moonshot program targets AI-robot co-evolution by 2050 — though METI's own strategy papers are explicit that this funding responds to lost ground, not defends a lead.

Sets Japan's industrial robotics policy; its own 2015 Robot Strategy acknowledged China overtook Japan as the world's largest annual industrial-robot adopter in 2013, and it has since named "physical AI" a priority field. METI

NEDOJPResearch-funding agency

Japan's largest public R&D funding agency; co-organized the World Robot Summit with METI (WRS2018 held October 2018 at Tokyo Big Sight), a flagship program to accelerate robot social implementation. METI

Japan's largest public applied-research institute; its Intelligent Systems Research Institute released a free dataset in September 2025 to support development of dual-arm robot AI models. AIST

Japan's central science and technology funding agency; manages Moonshot R&D Goal 3, targeting AI-robot co-evolution robots that learn, act, and coexist with humans by 2050. JST

Cabinet Office — SIP / Moonshot R&D ProgramsJPCross-ministerial R&D program

The Cabinet Office's cross-ministerial R&D schemes; Moonshot Goal 3 targets AI-robot co-evolution by 2050, alongside a companion SIP program on the rules and infrastructure for human-collaborative robotics. Cabinet Office

RIKEN's national AI research hub, established in April 2016; reorganized its research structure to include a dedicated "Physical Intelligence" group pursuing embodied AI that interacts with the real world. RIKEN AIP

Korea

Korea's KIST runs robotics research as a national institute in its own right, while MOTIE's April 2025 "K-Humanoid Alliance" — 40-plus partners aiming for a domestic humanoid foundation model by 2028 — shows how quickly a government program can follow a lab's research lead.

Founded February 10, 1966 as Korea's first government-funded multidisciplinary research institute; its robotics center reported teaching a robot to catch a falling ball using human forearm EMG signals in 2019. phys.org

Korea's industrial and energy policy ministry; launched the "K-Humanoid Alliance" of 40+ partners in April 2025, targeting a domestic humanoid foundation model by 2028 and about ₩1 trillion in investment through 2030. Korea.net

Other

Smaller, well-funded national programs round out the picture: Singapore's A*STAR runs a National Robotics Programme with fresh government backing, and Australia's CSIRO fields a robotics team, Data61, that has placed in the top ranks of international field-robotics competitions.

A*STARSGNational research institute

Singapore's lead public-sector R&D agency; its National Robotics Programme, launched in 2016, received a further S$60M funding commitment in April 2024 for its next phase. A*STAR

CSIROAUNational research institute

Australia's national science agency; its Data61 unit, with Emesent and Georgia Tech, placed second at the DARPA Subterranean Challenge Final in September 2021 — the first Australian team to place in the top two of a DARPA robotics challenge. CSIRO

04The regional picture

No region monopolizes either side of this map. North America and Europe together account for most of the listed labs, reflecting decades of university and corporate research investment; Japan, China, and Korea are more heavily represented on the agency side, where large, named national programs (Moonshot, the Five-Year Plan for Robotics, the K-Humanoid Alliance) play the role that corporate R&D budgets play in North America.

As with the companies map, HQ location understates how international this research actually is: KAIST's DARPA Robotics Challenge win, Meta's collaboration with Korea's Wonik on tactile hands, and RIKEN's new Physical Intelligence group all involve researchers and institutions crossing the borders this chart draws.

Labs and agencies on this map by region
Research labsNational agencies

North America

11
4

Europe

7
5

China

2
3

Japan

3
6

Korea

1
2

Other

0
2

05How this list is maintained

Like the companies map, this one is maintained rather than published once: organizations are removed if they wind down or are absorbed, and entries are re-checked against primary sources when a name, program, or ministry structure changes (Germany's research ministry, for instance, was renamed from BMBF to BMFTR in May 2025 mid-restructuring). Every specific number or named program in the tables above links to the source it came from.

06FAQ

How is this different from the Physical AI Landscape?

The Physical AI Landscape covers companies — who ships, raises funding, and competes commercially. This map covers the layer underneath: the labs producing the field's core research, and the agencies funding, regulating, and setting targets for it.

Why do some labs sit inside companies (NVIDIA, DeepMind, Meta, Toyota)?

Because academic and corporate research has genuinely blurred in this field — see the research labs section above. Excluding labs like NVIDIA Research or Google DeepMind just because they sit inside a company would misrepresent where the field's research actually happens.

How are organizations selected for this map?

Each entry needs a verifiable, sourced connection to physical AI or robotics — an active lab, a named funding program, or a research institute with a citable track record — confirmed against primary or authoritative secondary sources. Selection aims for global representativeness across North America, Europe, Japan, China, and Korea rather than exhaustive coverage of any single country.

Which country leads physical AI research?

No single country leads across both labs and agencies. North America and Europe host most of the labs on this map, reflecting long-running university and corporate research investment; Japan, China, and Korea run some of the largest named government programs. Even where one country held an early lead — Japan's early dominance in industrial-robot adoption, for instance — its own agencies' data shows that lead had passed to China by 2013.

Related articles

02

Company & Market Data

Share this articlePost on X
02

Research Labs & National Institutions

  • ›Physical AI's Research Labs and National Agencies, Mapped(this article)