Compare

Research labs

The labs inventing physical AI: academic, corporate and public research, with their flagship contributions and key people.

Updated 2026-07-10 · 28 entries · JSON

Research axes matrix in physical AI, by lab familyTable crossing six research axes (robotic foundation models / VLA, world models and sim2real, manipulation and grasping, locomotion, humanoids and whole-body control, data and human-robot interaction) with three lab families from the Physical AI Atlas: corporate, academic and public research. Each cell cites representative labs for that axis within that family.6 research axes crossed with 3 lab families (28 labs)CorporateAcademicPublic researchFoundation models(VLA)NVIDIA GEARGoogle DeepMindMicrosoft ResearchStanford IRIS LabBerkeley RAIL LabCMU Robotics InstituteShanghai AI Lab / OpenDriveLabWorld models,sim2realNVIDIA GEARMeta FAIRToyota TRIMIT Improbable AI LabShanghai AI Lab / OpenDriveLabManipulation,graspingRAI InstituteSony AI ZurichToyota TRICMU Robotics InstituteMIT Improbable AI LabEPFL LASAShanghai AI Lab / OpenDriveLabLocomotion(biped, quadruped)RAI InstituteETH Zurich RSLCMU Robotics InstituteKAIST HuboLabX-Humanoid (Beijing)LAAS-CNRS GepettoHumanoids,whole-body controlNVIDIA GEARIIT Genova (iCub/ergoCub)KIT H2TInria HUCEBOTDFKI RICX-Humanoid (Beijing)Data, teleop.,human-robot interactionMeta FAIRStanford IRIS LabEPFL LASAKIT H2TInria HUCEBOTLISPEN (Arts et Métiers)Source: Physical AI Atlas, data/labs.json (28 lab profiles).

Each row of this table is one research axis, each column one lab family from the Atlas (corporate, academic, public research); the names listed in a cell come directly from the corresponding lab's specialty field. A sparsely filled cell does not mean no work happens there in that family, it simply reflects the scope of the 28 labs covered in this edition of the Atlas.

Click any row to reveal details, sources and verification date.

Name Affiliation Country Type Research areas Flagship contribution

GEAR's demos anchor NVIDIA's Physical AGI narrative at GTC keynotes; the group also produced Eureka, an LLM-driven reward design method, and the open-ended agent Voyager.

Key people : Jim Fan (Director of AI, Distinguished Scientist, co-lead de GEAR), Yuke Zhu (co-lead de GEAR, professeur UT Austin)Open outputs : Poids d'Isaac GR00T-N1-2B (Hugging Face, licence NVIDIA Open Model), datasets NVIDIA Physical AI, code d'Eureka, Voyager, VIMA et MineDojo sur GitHub.

Sources : NVIDIA Research (2026-07)NVIDIA Newsroom (2025-03)Hugging Face (2025-03) Last verified 2026-07-10

Heir to RT-2 and RT-X, the team runs partnerships with Boston Dynamics, Apptronik and Agile Robots, and coordinates a 2026 EMEA robotics accelerator program.

Key people : Carolina Parada (Senior Director, Head of Robotics), Aaron Saunders (ex-CTO Boston Dynamics, recruté en 2025)Open outputs : Pas de poids ouverts pour Gemini Robotics/Robotics-ER (accès waitlist et SDK) ; contribution historique à l'initiative ouverte Open X-Embodiment / RT-X (dataset collaboratif multi-labos, 2023).City : Londres

Sources : Google DeepMind (2026-04)Boston Dynamics (2025-02)CNBC (2026-03-24) Last verified 2026-07-10

Yann LeCun left Meta in November 2025 after founding FAIR in 2013, to launch the startup AMI Labs; FAIR continues under Rob Fergus inside Meta Superintelligence Labs, but public sources do not clearly confirm physical AI as a distinct strategic priority in mid-2026.

Key people : Rob Fergus (Director of AI Research chez FAIR depuis mai 2025)Open outputs : V-JEPA 2 (poids et code ouverts), PARTNR (benchmark et dataset de 100 000 tâches domestiques), Meta Sparsh (encodeur tactile), Meta Digit 360 et Digit Plexus (capteurs et plateforme matérielle), tous publiés fin 2024 et en 2025.

Sources : Meta AI (2025-06)TechCrunch (2025-05-08)MIT Technology Review (2026-01-22) Last verified 2026-07-10

Tedrake's dual role at TRI and MIT CSAIL anchors TRI within the academic robotics ecosystem.

Key people : Russ Tedrake (SVP Large Behavior Models chez TRI, Toyota Professor au MIT), Ben Burchfiel (Senior Research Manager), Naveen Kuppuswamy (Senior Research Scientist, robotique tactile)Open outputs : Dataset DROID (76 000 trajectoires, ~350h, effort inter-institutionnel avec 13 labos), licence CC-BY 4.0 avec code d'entraînement ; Drake, toolkit de simulation robotique open source.City : Los Altos

Sources : Toyota Research Institute (2026-06)Robotics 24/7 (2025-01)Green Car Congress (2023-09-20) Last verified 2026-07-10

Opened a Zurich office in 2025 and signed a donation agreement with the ETH Foundation USA funding joint robotics research with ETH Zurich.

Key people : Marc Raibert (fondateur de Boston Dynamics, dirigeant du RAI Institute), Marco Hutter (dirige le nouveau bureau de Zurich, professeur à l'ETH Zurich)Open outputs : Aucun accès ouvert significatif identifié (code/modèles non publiés).City : Cambridge (Massachusetts)

Sources : RAI Institute (2026-06)Boston Dynamics (2025-02-05)ETH Zurich (2025-09) Last verified 2026-07-10

A cross-lab initiative spanning Redmond, Cambridge (UK), New York and Montreal rather than a single dedicated lab; Magma unifies digital UI agents and physical robot manipulation.

Key people : Ashley Llorens (CVP et Managing Director, Microsoft Research Accelerator)Open outputs : Magma : code et poids ouverts (GitHub microsoft/Magma, Hugging Face microsoft/Magma-8B) ; Rho-Alpha en accès anticipé uniquement au moment de la vérification.

Sources : Microsoft Research (2026-01)The Robot Report (2026-01)Microsoft Research (2025-02) Last verified 2026-07-10

Sony AI also runs a parallel Gastronomy Flagship Project on dexterous culinary manipulation robots, distinct from the commercial AIBO product line.

Key people : Peter Dürr (Director, Sony AI Zurich, lead du projet Ace)Open outputs : Recherche publiée dans Nature ; aucun poids ou code open source identifié au moment de la vérification.City : Zurich

Sources : Sony AI (2026-04)Nature (2026-04)Robotics and Automation News (2026-05-14) Last verified 2026-07-10

As of mid-2026, Stanford's official faculty profile still lists Chelsea Finn as an active Assistant Professor with no documented leave, alongside her role as Physical Intelligence co-founder.

Key people : Chelsea Finn (Assistant Professor, Stanford, William George and Ida Mary Hoover Faculty Fellow, cofondatrice de Physical Intelligence), Karl Pertsch (postdoc), Moo Jin Kim (doctorant, co-auteur OpenVLA)Open outputs : Co-développement d'OpenVLA (modèle VLA 7B, poids ouverts) et d'Octo (politique généraliste open source, avec Berkeley et CMU) ; code Mobile ALOHA et RoboCrowd publiés en open source.City : Stanford (Californie)

Sources : Stanford Profiles (2026-07)IRIS Lab (2026-07)Stanford Report (2024-04) Last verified 2026-07-10

Chelsea Finn, Dhruv Shah and Kate Rakelly are among the lab's notable alumni; the separate 'Berkeley Humanoid' project belongs to Koushil Sreenath's Hybrid Robotics lab, not RAIL.

Key people : Sergey Levine (Professor, directeur du lab), Andrew Wagenmaker (postdoc)Open outputs : BridgeData V2 (dataset et code, GitHub rail-berkeley/bridge_data_v2), co-développement de Q-Transformer avec Google DeepMind, co-leadership d'Octo, SERL (Sample-Efficient Robotic RL) en open source.City : Berkeley (Californie)

Sources : RAIL Lab (2026-07)UC Berkeley (2026-07)arXiv (2023-08) Last verified 2026-07-10

Skild AI, a 2023 CMU spin-off co-founded by Pathak and Gupta, raised a $1.4B Series C led by SoftBank in January 2026 at a valuation above $14B to build a general-purpose robot foundation model.

Key people : Deepak Pathak (Assistant Professor, Robotics Institute ; CEO de Skild AI), Abhinav Gupta (Professor, Robotics Institute ; co-fondateur et président de Skild AI), David Held (Associate Professor), Katerina Fragkiadaki (Associate Professor)Open outputs : Code de RMA et travaux dérivés publiés en open source (GitHub) ; publications RSS/CoRL en accès arXiv.City : Pittsburgh

Sources : CMU Robotics Institute (2026-07)arXiv (RSS 2021) (2021-07)The Robot Report (2026-01) Last verified 2026-07-10

Rapid Locomotion via Reinforcement Learning reached record speeds of 3.9 m/s on the MIT Mini Cheetah and transferred to the Unitree Go1 across natural terrain.

Key people : Pulkit Agrawal (Steven and Renee Finn Chair Assistant Professor, MIT EECS, directeur du lab), Gabe Margolis (doctorant, auteur principal de Rapid Locomotion)Open outputs : Code de Rapid Locomotion via Reinforcement Learning (GitHub Improbable-AI/rapid-locomotion-rl) ; publications arXiv en accès libre pour Visual Dexterity.City : Cambridge (Massachusetts)

Sources : Improbable AI Lab (2026-07)arXiv (Science Robotics 2023) (2022-11)arXiv (RSS 2022) (2022-05) Last verified 2026-07-10

Hutter co-founded ANYbotics, the spin-off commercializing ANYmal; RSL trains its policies via massively parallel simulation running thousands of simulated ANYmal robots on a single workstation.

Key people : Marco Hutter (Head of Lab, Director ETH Centre for Robotics, cofondateur d'ANYbotics), Cesar Cadena (Senior Scientist)Open outputs : Organisation GitHub leggedrobotics (300+ dépôts), notamment rsl_rl (framework RL), elevation_mapping_cupy (cartographie GPU) et open3d_slam (SLAM point-cloud).City : Zurich

Sources : ETH Zurich RSL (2026-07)Science Robotics (2024-03)ETH Zurich News (2024-03-13) Last verified 2026-07-10

The complementary BioRob Lab at EPFL (Auke Ijspeert) works in 2026 on bio-inspired locomotion controllers trained by reinforcement learning on humanoid robots.

Key people : Aude Billard (fondatrice et directrice du lab depuis 2006, directrice du Swiss Innovation Booster on Robotics)Open outputs : DS Motion Generator, iCub Whole Body Task Controller, MLDemos (toolkit ML), datasets Crowdbot et LASA Handwriting Dataset, tous publiés sur la page dédiée Source Code and Datasets.City : Lausanne

Sources : EPFL (2026-07)EPFL LASA (2026-07)EPFL (2026-07) Last verified 2026-07-10

Generative Bionics, a spin-off co-founded by Pucci, raised 70 million euros for 'Made in Italy' humanoid robots; the ergoCub-INAIL project was recognized in New Delhi for ethical and sustainable AI use.

Key people : Giorgio Metta (Directeur scientifique de l'IIT), Daniele Pucci (coordinateur de l'AMI Lab et du projet ergoCub, cofondateur de Generative Bionics), Lorenzo Natale (responsable du Humanoid Sensing and Perception Lab), Marco Maggiali (Facility Coordinator, iCub Tech)Open outputs : YARP (Yet Another Robot Platform, middleware open source v3.0) et logiciels iCub/ergoCub publiés sur GitHub (organisation robotology).City : Genova

Sources : IIT (2026-07)IIT Open Talk (2026-05)EurekAlert! (2026-05) Last verified 2026-07-10

Founding director Sami Haddadin left MIRMI in January 2025 and now serves as VP Research at MBZUAI in Abu Dhabi.

Key people : Lorenzo Masia (Executive Director depuis octobre 2024, robotique portable et assistive), Achim Lilienthal (Deputy Director, perception multimodale), Angela Schoellig (Industry & International)Open outputs : Aucun dépôt ou dataset open source spécifique confirmé au moment de la vérification.City : Munich

Sources : MIRMI-TUM (2025-04-10)Wikipedia (2026-07)MBZUAI (2026-07) Last verified 2026-07-10

Asfour leads the Robotics Institute Germany (RIG), a national initiative federating German robotics research, with dedicated Physical AI events in 2025-2026.

Key people : Tamim Asfour (Full Professor, directeur du H2T, porte-parole du Robotics Institute Germany, IEEE Fellow depuis janvier 2026)Open outputs : ArmarX (architecture logicielle cognitive open source), Master Motor Map (format de capture de mouvement), Simox, IVT, RobotEditor, et datasets KIT Human Motion Database, KIT Motion-Language Dataset et Bimanual Actions Dataset.City : Karlsruhe

Sources : KIT H2T (2026-07)KIT H2T (2026-07)Robotics Institute Germany (2026-01) Last verified 2026-07-10

Oxa raised 47 million dollars in a 2021 Series B and 140 million dollars in a 2023 Series C to deploy its GPS-independent autonomous driving software stack.

Key people : Nick Hawes (Director de l'ORI depuis 2022), Ingmar Posner (Founding Director, aujourd'hui Deputy Director, cofondateur d'Oxa), Paul Newman (Professor, head du Mobile Robotics Group, cofondateur d'Oxa)Open outputs : Oxford RobotCar Dataset et Oxford Radar RobotCar Dataset, en accès ouvert sur robotcar-dataset.robots.ox.ac.uk.City : Oxford

Sources : Oxford Robotics Institute (2026-07)Oxford Robotics Institute (2026-07)International Journal of Robotics Research (2017-01) Last verified 2026-07-10

The team won the IEEE T-ASE Best Paper Award in 2024 for distributed auction algorithms in multi-robot task assignment, and an ICRA Best Paper Award on multi-robot systems in 2019.

Key people : Javier Alonso-Mora (Professor of Mobile and Collaborative Robotics, directeur du TU Delft Robotics Institute, ERC Starting Grant 2021), Laura Ferranti (fondatrice du Reliable Robot Control Lab), Holger Caesar (Assistant Professor, perception de véhicules autonomes)Open outputs : Code de planification multi-robots publié sur github.com/tud-amr.City : Delft

Sources : TU Delft Robotics Institute (2026-07)TU Delft AMR Lab (2026-07)AMS Institute (2026-07) Last verified 2026-07-10

Founded in 2018 with over 100 members and around 40 PhD students, LISPEN co-leads the 'Prototyping/Design' axis of France's national TIRREX robotics infrastructure (EquipEx+ 2021-2026) alongside CEA List, and Béarée directs the French national robotics research group GDR Robotique.

Key people : Richard Béarée (Professeur des universités en automatique et robotique, directeur du LISPEN depuis janvier 2021), Olivier Gibaru (Professeur, coordinateur du projet H2020 ColRobot, génération de trajectoire), Adel Olabi (Maître de conférences en robotique, calibration et précision de robots), Olivier Thomas (Professeur, dynamique vibratoire et structures intelligentes)Open outputs : Publications en accès ouvert via HAL (profil cv.hal.science/richard-bearee) ; aucun dépôt de code ou dataset open source majeur identifié pour les travaux robotiques du LISPEN.City : Lille

Sources : LISPEN, Arts et Metiers (2026-07)Arts et Metiers (2021-01-04)HAL (2026-07)Arts et Metiers (2018-06)TIRREX (2026-01) Last verified 2026-07-10

Ivaldi and Mouret co-founded the startup Bleu Robotics, building on around 20 years of humanoid robotics research to deliver AI-driven humanoid robots for industry.

Key people : Serena Ivaldi (Directrice de recherche, cheffe d'équipe HUCEBOT depuis octobre 2023), Jean-Baptiste Mouret (Directeur de recherche, algorithmes MAP-Elites, ERC Starting Grant 2014), Pauline Maurice (Chargée de recherche CNRS)Open outputs : Code open source sur GitHub (github.com/hucebot, dont seiko_controller_code) ; participation aux projets européens euROBIN et ENACT (2024-2029).City : Nancy

Sources : Inria (2026-07)LORIA (2026-07)GitHub / IEEE RA-L 2024 (2024-06) Last verified 2026-07-10

The team develops in-house robotic platforms, the Pyrene humanoid and the Solo and Sassa quadrupeds; France's national PEPR Robotique research program launched at LAAS-CNRS in February 2026.

Key people : Nicolas Mansard (Directeur de recherche CNRS, chef d'équipe Gepetto, prix Étoile de l'Europe 2022), Olivier Stasse (chercheur, robotique humanoïde), Florent Lamiraux (chercheur, planification de mouvement), Ludovic Righetti (chercheur, contrôle de robots quadrupèdes)Open outputs : Pinocchio (dynamique et cinématique), Crocoddyl (contrôle prédictif), HPP - Humanoid Path Planner (planification de mouvement sans collision), tous open source sur GitHub (stack-of-tasks).City : Toulouse

Sources : LAAS-CNRS (2026-07)Gepetto Team (2026-07)GitHub (2026-07)CNRS INS2I (2022-12-13) Last verified 2026-07-10

The center is organized into four fundamental research groups (Mechanics and Control, Interaction and Learning, Quantum Computing, Human-Centred Interaction) and four applied teams (Maritime, Space, Terrestrial, Intelligent Healthcare Systems).

Key people : Frank Kirchner (Directeur du Robotics Innovation Center), Rebecca Adam (Responsable de l'équipe Interaction and Learning), Rolf Drechsler (Directeur du département Cyber-Physical Systems, DFKI Brême)Open outputs : Logiciels open source publiés sur github.com/dfki-ric : Rock (Robot Construction Kit, middleware robotique), HyRoDyn (dynamique hybride), Bagel (langage graphique de dataflow), ARC-OPT.City : Brême

Sources : DFKI Robotics Innovation Center (2026-07)TechXplore (2024-06-25)DFKI (2024-06) Last verified 2026-07-10

OpenDriveLab is not an internal Shanghai AI Lab team but a research initiative mainly hosted at the University of Hong Kong; AgiBot (Zhiyuan Robotics) is a separate private startup, not a subsidiary of Shanghai AI Lab, the two organizations run a dataset and research partnership.

Key people : Hongyang Li (cofondateur et lead scientist d'OpenDriveLab, Assistant Professor à HKU, chercheur affilié à Shanghai AI Lab), Zhou Bowen (Directeur et Chief Scientist de Shanghai AI Laboratory depuis juillet 2024)Open outputs : Dataset AgiBot World (GitHub OpenDriveLab/AgiBot-World, Hugging Face agibot-world/AgiBotWorld-Alpha et Beta), poids du modèle GO-1 / GO-1 Air, code de UniAD (Best Paper CVPR 2023) en open source.City : Shanghai

Sources : OpenDriveLab (2026-01)GitHub (2026-01)The Robot Report (2026-06)South China Morning Post (2024-07) Last verified 2026-07-10

Designated in October 2024 as a 'National-Local Jointly Built Embodied Intelligence Robotics Innovation Center', X-Humanoid is legally distinct from the private startup AgiBot / Zhiyuan Robotics, though both are based in China's humanoid robotics ecosystem.

Key people : Xiong Youjun (cofondateur et General Manager de X-Humanoid, également cofondateur et CTO d'UBTech Robotics), Tang Jian (CTO de X-Humanoid)Open outputs : Design matériel du robot Tiangong (open source depuis novembre 2024), modèle vision-langage Pelican-VL, dataset RoboMIND.City : Pékin (zone de Yizhuang)

Sources : Global Times (2024-11)Xinhua (2025-03-13)Humanoids Daily (2026-02) Last verified 2026-07-10

One of Japan's oldest robotics labs, founded around 1978 by Hirochika Inoue, with a lineage of emblematic humanoids including Kotaro, Kenshiro, Kengoro and Musashi.

Key people : Masayuki Inaba (Professeur, directeur du JSK Lab), Kei Okada (Professeur, Département Mechano-Informatics), Kento Kawaharazuka (Lecturer, JSK Lab et UTokyo AI Center)Open outputs : Publications en accès libre sur arXiv (nombreux articles 2024-2026 sur Musashi et le contrôle appris) ; pas de release logicielle majeure confirmée au-delà de ces publications.City : Tokyo

Sources : New Atlas (2024-06-13)arXiv (2024-06)Kento Kawaharazuka (page personnelle) (2026-07) Last verified 2026-07-10

Rainbow Robotics, a 2011 KAIST spin-off commercializing HUBO-derived technology, has drawn significant investment from Samsung, illustrating a strong capital link between this academic lab and industry.

Key people : Hae-Won Park (Directeur du Humanoid Robot Research Center depuis février 2020, dirige aussi le DRCD Lab), Jemin Hwangbo (Professeur KAIST, collabore sur le haut du corps du nouvel humanoïde), Jun-Ho Oh (Professeur émérite, créateur du HUBO original en 2004, aujourd'hui CTO de Rainbow Robotics)Open outputs : Aucune publication ouverte de dataset ou de code comparable identifiée ; publications académiques via IROS/ICRA uniquement.City : Daejeon

Sources : KAIST (CV Hae-Won Park) (2026-07)TechXplore (2025-09)Interesting Engineering (2025-09) Last verified 2026-07-10

Fraunhofer's applied funding model (industry contracts plus public base funding) makes IPA a reference partner for German Mittelstand factories piloting robotics and AI, from vision-guided picking to mobile manipulation.

Key people : Werner Kraus (responsable robotique et systèmes d'assistance), Marco Huber (IA industrielle et vision)Open outputs : Publications appliquées, démonstrateurs industriels, participation aux projets européens Horizon (robotique de production).City : Stuttgart

Sources : Fraunhofer IPA (2026-07-10)Fraunhofer IPA (Care-O-bot) (2026-07-10) Last verified 2026-07-10

IML sits at the junction the Atlas tracks between AMR fleets and data: its swarm work prefigures the fleet-learning loops that make warehouse robotics a data business.

Key people : Michael ten Hompel (directeur, chaire de logistique TU Dortmund)Open outputs : Démonstrateurs evoBOT et LoadRunner, publications, projets avec DHL et l'écosystème logistique allemand.City : Dortmund

Sources : Fraunhofer IML (2026-07-10) Last verified 2026-07-10

28 / 28 entries

D·Fairy

Bringing a physical AI solution into European industry?

D·Fairy supports robotics and embodied AI companies on their EMEA market entry: OEM and Tier 1 access, co-funded pilots, AI Act readiness.

Book an intro call