Source registry

Market data

• F1Shanghai Stock Exchange — "Shanghai Stock Exchange to review Unitree Robotics IPO on June 1, with over 4.2 billion yuan fundraising plan". Type: filing. Date: 2026-05-26. Retrieved: 2026-08-27.english.sse.com.cnSupports: Unitree listing cards (review date, ¥4.202B fundraising plan, no fewer than 40.4464M new shares, FY2025 revenue ¥1.699B, net profit ¥590M excluding non-recurring items, 60.13% core gross margin).
• A5Counterpoint Research — "Global Humanoid Robot Shipments Soar Nearly 300% YoY in H1 2026, Driven by Commercial Deployments". Type: analyst. Date: 2026-08-19 (page); byline 2026-08-20. Retrieved: 2026-08-28. Review after: 2027-01-01. Review reason: FY2026 forecast period closes; check for a full-year actual or revised estimate.counterpointresearch.comSupports: humanoid shipment cards (H1 2026 >22,000 units, ~+300% YoY; FY2026 forecast >50,000 units, +210%); who-ships table (AGIBOT first at 9,700 units, then Unitree, Galbot, UBTECH, Leju; top five 86%); application-mix cards (entertainment/performance/data/research >60%, manufacturing 13%, warehousing 5%).
• P10Caixin Global — "Galbot Raises $362 Million in Fresh Funding, Eyes Hong Kong IPO". Type: press. Date: 2026-03-03. Retrieved: 2026-08-27.caixinglobal.comSupports: Galbot funding card (¥2.5B round, ≈$362M, valuation >¥20B, highest-valued unlisted humanoid firm).
• P11South China Morning Post — "Unitree Robotics surges 629% to US$66 billion valuation in Shanghai share debut". Type: press. Date: 2026-08-19. Retrieved: 2026-08-27.scmp.comSupports: Unitree trading cards (IPO price ¥150.80, day-one open ¥1,100 / +629%, day-one close ¥845 / +460%, ≈¥342B closing market capitalisation). Intraday and closing valuations are printed separately, not reconciled.
• A6International Federation of Robotics — "US robot installations rose 11% to 38,000 units in 2025". Type: analyst. Date: 2026-06-18. Retrieved: 2026-08-27.ifr.orgSupports: the preliminary US-only 2025 figure (38,000 units, +11%) quoted in the note refusing 621,000. World Robotics 2026 [A7] restates it as 38,428 (+12%).
• A1Counterpoint Research — "Global Humanoid Robot Installations Reach 16,000 Units in 2025…". Type: analyst. Date: 2026-01-14. Retrieved: 2026-10-09.counterpointresearch.comSupports: humanoids 2025 table, Counterpoint row — “around an additional 16,000 units of humanoid robots were installed worldwide, with China alone accounting for more than 80% of the installations”. An analyst estimate from Counterpoint’s Humanoid Robot Research; the release states no height limit and no survey sample.
• P1The Robot Report — "Galbot brings in $300M…". Type: press. Date: 2025-12-23. Retrieved: 2026-08-25.therobotreport.comSupports: Galbot funding card ($300M round, $800M cumulative, $3B valuation).
• A2IFR — World Robotics 2025 Industrial Robots, Executive Summary. Type: analyst. Date: 2025-09-25. Retrieved: 2026-08-25.ifr.orgSupports: industrial year table, 2024 column — installations 542,076 / China 295,045 / Japan 44,453 / stock 4,663,698 / density 177. The 2025 column is filled from World Robotics 2026 [A7], not from this report.
• A7IFR — World Robotics 2026 Industrial Robots, Executive Summary. Type: analyst. Date: 2026-09-24. Retrieved: 2026-10-09.ifr.orgSupports: industrial year table, 2025 column — installations 603,307 (+11%) / China 354,226 (+20%) / stock 5,079,078 (+9%); US 38,428 (+12%) in the 621,000 note. States no Japan figure in text and no global robot density.
• A8International Federation of Robotics — "Five Million Robots now Operate in Factories Globally". Type: analyst. Date: 2026-09-24. Retrieved: 2026-10-09.ifr.orgSupports: industrial year table, 2025 column — Japan 36,219 (−19%); China at 59% of global deployments.
• A9IFR — World Robotics 2025 Service Robots, Executive Summary. Type: analyst. Date: 2025. Retrieved: 2026-10-09.ifr.orgSupports: earlier-edition line for 2024 (report pp. 11–12) — professional >199,000 units (+9%) / medical close to 16,700 (+91%) / consumer close to 20.1 million (+11%); 944 known service robot producers. The summary does not state the survey sample; that is in [A16].
• A16International Federation of Robotics — "Service Robots See Global Growth Boom". Type: analyst. Date: 2025-10-07. Retrieved: 2026-10-09.ifr.orgSupports: earlier-edition line for 2024 — “All numbers at the World Robotics Service Robots 2025 report are based on a sample of 294 service robot suppliers.” IFR’s press-conference slides for the same edition print 293 of 944; the release’s 294 matches the report’s Sources & Methods.
• A10International Federation of Robotics — "Global Sales of Professional Service Robots Surge 24%". Type: analyst. Date: 2026-09-30. Retrieved: 2026-10-09.ifr.orgSupports: service cards and table for 2025 — professional “almost 250,000” units (+24%) / consumer “almost 34.2 million” (+37%) / medical +19% with no unit total, surgical 8,400 (+57%); transportation and logistics 117,500 units (+21%, 47%); “about 7,000 units of full-size humanoids (above 140 cm)” sold in 2025. States that the figures are from World Robotics 2026 Service Robots.
• A11IFR — World Robotics 2026 Service Robots, Executive Summary. Type: analyst. Date: 2026. Retrieved: 2026-10-09.ifr.orgSupports: humanoids 2025 table (report p. 15) — “Almost 6,900 humanoids were reported sold globally in 2025”, 171 known humanoid manufacturers; professional sales +24% “excluding medical robots” and more than 1,000 known producers (p. 13). Its text states no professional, medical or consumer unit total; its one chart gives the top five application groups in thousands of units.
• A13International Federation of Robotics — Market Presentation, World Robotics Press Conference 2026. Type: analyst. Date: 2026-09-24. Retrieved: 2026-10-09.ifr.orgSupports: service cards and table for 2025 — medical 15,300 units (+19%), professional 250,000, consumer 34.2 m, “All numbers based on a sample of 238 producers of total 1001 manufacturers. Data is not projected to the whole industry.” (slide 28); application-group units and shares (slide 29) and growth rates (slide 30); “Nearly 7,000 humanoid robots” sold, “Full-size humanoids, height above 140cm” (slide 32); 174 humanoid manufacturers identified (slide 33).
• A14IFR — World Robotics 2026 Service Robots, Sources & Methods. Type: analyst. Date: 2026. Retrieved: 2026-10-09.ifr.orgSupports: the sample definition printed with the 2025 service figures (report pp. 30–31) — 238 suppliers, 22 of them with no sales yet, from a statistical population of 1,001; “sample data … not projected to the whole industry … a minimum level of sales”; “The IFR SD strongly discourages the creation of time-series data, compiling data from different issues of this publication.”
• A3IFR — World Robotics 2025 Industrial Robots, Foreword. Type: analyst. Date: 2025-09-25. Retrieved: 2026-10-09.ifr.orgSupports: robot density 2024 as first published (report p. 3) — global 177; Korea 1,220 / Singapore 818 / China 567 / Germany 449 / Japan 446 robots per 10,000 persons employed in manufacturing. IFR’s density release of 2026-04-08 [A12] restates China as 166 and the world as 132; both stay on the page. The World Robotics 2026 executive summary and foreword state no density figure (read 2026-10-09).
• A12International Federation of Robotics — "Robot Density Surges in Europe, Asia, and Americas". Type: analyst. Date: 2026-04-08. Retrieved: 2026-10-09.ifr.orgSupports: robot density 2024 as restated — world 132; Korea 1,220 / Singapore 818 / Germany 449 / Japan 446 / Sweden 377 robots per 10,000 employees in manufacturing; China 166, 22nd worldwide, “based on updated labor market data issued by China's National Bureau of Statistics”. Regional averages (Western Europe 267, North America 204, Asia 131) are not carried. States no 2025 density.
• A4IFR — World Robotics 2024 Service Robots, Executive Summary. Type: analyst. Date: 2024. Retrieved: 2026-10-09.ifr.orgSupports: earlier-edition line for 2023 (report p. 11) — professional “more than 205,000” units (+30%) / medical “more than 6,100” units (+36%); 921 known service robot producers. The summary states neither a consumer unit total nor the survey sample; those are in [A15].
• A15International Federation of Robotics — Presentation, World Robotics 2024 press conference. Type: analyst. Date: 2024-09. Retrieved: 2026-10-09.ifr.orgSupports: earlier-edition line for 2023 (slide 31) — consumer 4.1 million units (+1%); medical 6,200 units; “All numbers based on a sample of 298 companies. Data is not projected to the whole industry.”
• P2Agility Robotics — press releases. Type: press. Date: 2025. Retrieved: 2026-08-25.agilityrobotics.comSupports: Agility ~$641M cumulative funding (incl. $400M Series C 2025).
• P31X Technologies — "1X secures $100M in Series B…". Type: press. Date: 2024. Retrieved: 2026-08-25.1x.techSupports: 1X $125M+ cumulative funding.
• P4LimX Dynamics — news. Type: press. Date: 2026. Retrieved: 2026-08-25.limxdynamics.comSupports: LimX ~$400M cumulative funding.
• P5Apptronik — "Apptronik closes over $935M Series A". Type: press. Date: 2026-02. Retrieved: 2026-08-25.apptronik.comSupports: Apptronik $935M Series A.
• P6Multiples.vc — UBTECH Robotics valuation multiples. Type: aggregator. Date: 2023 IPO. Retrieved: 2026-08-25.docs.multiples.vcSupports: UBTech ~$940M pre-IPO equity (excludes 2023 HK IPO proceeds).
• P7RobotHub — Fourier Intelligence company profile. Type: aggregator. Date: 2025. Retrieved: 2026-08-25.robothub.appSupports: Fourier ~$230M funding.
• P8Shanghai Stock Exchange (English) — Kepler stake-transaction filing. Type: exchange filing (estimate). Date: 2026. Retrieved: 2026-08-25.english.sse.com.cnSupports: Kepler ~$100M industry estimate (undisclosed).
• P9Qiming Venture Partners — reprint of 中国企业家 interview with He Wang. Type: press. Date: 2025-09-11. Retrieved: 2026-08-27.qimingvc.com/cn/node/7399Supports: Galbot cumulative funding >¥2.4B; 100 signed smart pharmacies; Bosch China / 博原资本 MoU and JV 博银合创 (June 2025). Not a specification source. USD $800M from The Robot Report is a different article and is not summed with this RMB figure.

Robots & engineering models

• M1Unitree — unitree_mujoco. Type: official simulation repository. License: BSD-3-Clause. Commit inspected: 4134cb5. Retrieved: 2026-08-26.github.com/unitreerobotics/unitree_mujocoSupports: R1, G1, H1, H1-2, and H2 MJCF / mesh links.
• M2Unitree — unitree_model. Type: official model dataset. License: BSD-3-Clause. Retrieved: 2026-08-26.huggingface.co/datasets/unitreerobotics/unitree_modelSupports: B2, G1, H1, H1-2, and H2 USD model links.
• M3GalaxyGeneralRobotics — galbot_one_golf_description. Type: official robot-description repository. License: Apache-2.0. Retrieved: 2026-08-26.github.com/GalaxyGeneralRobotics/galbot_one_golf_descriptionSupports: Galbot G1 model links, and the browser-assembled G1 preview built from urdf/galbot_one_golf.urdf at commit 7c06d4d.
• M4GalaxyGeneralRobotics — galbot_s1_description. Type: official robot-description repository. License: Apache-2.0. Retrieved: 2026-08-26.github.com/GalaxyGeneralRobotics/galbot_s1_descriptionSupports: Galbot S1 model links, and the browser-assembled S1 preview built from urdf/galbot_s1.urdf at commit cb758e5.
• M5Universal Robots — Universal_Robots_ROS2_Description. Type: official ROS 2 robot-description repository. Retrieved: 2026-09-10.github.com/UniversalRobots/Universal_Robots_ROS2_DescriptionSupports: UR20 remaining source-only because UR20/UR15/UR18/UR8LONG/UR30 meshes use Graphical Documentation terms. UR5e is expanded separately as a labeled recipe product at M33. Other repository content, including UR5e meshes, is BSD-3-Clause.
• M6Boston Dynamics — Spot SDK Mechanical Interfaces. Type: official SDK documentation. License: Boston Dynamics SDK terms. Retrieved: 2026-08-26.github.com/boston-dynamics/spot-sdkSupports: Spot official detailed-CAD availability through the Boston Dynamics Support Center; no model is mirrored here.
• M7Blue Robotics — BlueROV2 documentation, “3D Model”. Type: official product documentation. Retrieved: 2026-08-26.docs.bluerobotics.com/brov2/#3d-modelSupports: BlueROV2 assembled CAD view/download link hosted on GrabCAD; host download terms apply.
• M8Unitree — unitree_ros. Type: official ROS robot-description repository. License: BSD-3-Clause. Commit pinned: 80b9643. Retrieved: 2026-09-10.github.com/unitreerobotics/unitree_rosSupports: browser-assembled previews of G1 (counted 29-DoF whole assembly; 23-DoF and three-finger hands remain variants), H1, H1-2, H2, R1, R1 AIR, B2, B2-W, Go2, Go2-W, Aliengo, Z1, Dex5-1, Dex3-1, and Dex1-1, plus the URDF and mesh links beside them. Unitree G1 is not Galbot G1. Go2-W is the wheeled-foot Go2 description, not B2-W. Aliengo is a distinct research quadruped, not Go2.
• M9AgiBot (AgibotTech) — agibot_x2_urdf. Type: official robot-description repository. License: Mulan PSL v2. Commit pinned: 77f43eb. Retrieved: 2026-08-26.github.com/AgibotTech/agibot_x2_urdfSupports: X2 Ultra model link and browser-assembled preview from X2_URDF-v1.4.0/X2-Ultra.urdf. The repository README maps v1.4.0 to the nameplate reading “X2 Ultra -N”. The quantitative sheet is the AgiBot X2 product table on the humanoids page, not this URDF.
• M10AgiBot (AgibotTech) — A3-A3U-robot-model. Type: official robot-description repository. License: Mulan PSL v2. Commit pinned: 589f508. Retrieved: 2026-08-26.github.com/AgibotTech/A3-A3U-robot-modelSupports: A3 and A3 Ultra model links and browser-assembled previews from the T3.0 variants. No specification sheet is claimed.
• M11Fourier Intelligence (FFTAI) — Wiki-GRx-Models. Type: official robot-description repository. License: GPL-3.0. Commit pinned: 7d96c75. Retrieved: 2026-08-26.github.com/FFTAI/Wiki-GRx-ModelsSupports: GR1T1, GR1T2, and GR2 v3.0.1 model links and live previews. Because the assets are GPL-3.0, no render derived from them is published on this site; those cards load from the vendor repository only. GR-2 quantitative numbers come from the Fourier product page, not this URDF. GR1T1/GR1T2 remain model-only.
• M12ANYbotics — anymal_b_simple_description. Type: official robot-description repository. License: BSD-3-Clause. Commit pinned: 988b5df. Retrieved: 2026-08-26.github.com/ANYbotics/anymal_b_simple_descriptionSupports: ANYmal B model link and browser-assembled preview. ANYbotics publishes this as a simplified simulation description, not as production ANYmal C or D. No specification sheet is claimed.
• M13ROBOTIS — turtlebot3. Type: official robot-description repository. License: Apache-2.0. Retrieved: 2026-08-26.github.com/ROBOTIS-GIT/turtlebot3Supports: TurtleBot3 Burger and Waffle Pi model links, and the browser-assembled previews built from turtlebot3_description/urdf/ at commit fc817ce.
• M14Pollen Robotics — reachy_description. Type: official robot-description repository. License: Apache-2.0. Retrieved: 2026-08-26.github.com/pollen-robotics/reachy_descriptionSupports: Reachy model links, and the browser-assembled preview built from urdf/reachy.URDF at commit d532bb7.
• M15TheRobotStudio — SO-ARM100. Type: official robot-description repository. License: Apache-2.0. Retrieved: 2026-08-26.github.com/TheRobotStudio/SO-ARM100Supports: SO-ARM101 model links, and the browser-assembled preview built from Simulation/ at commit 7629d2a.
• M16EngineAI — engineai_ros2_workspace. Type: official robot-description repository. License: BSD-3-Clause. Retrieved: 2026-08-26.github.com/engineai-robotics/engineai_ros2_workspaceSupports: EngineAI PM01 model links, and the browser-assembled preview built from serial_pm_v2.urdf at commit c41043f.
• M17Rethink Robotics — baxter_common. Type: official robot-description repository. License: BSD-3-Clause. Retrieved: 2026-08-26.github.com/RethinkRobotics/baxter_commonSupports: Baxter model links, and the browser-assembled preview built from baxter_description/urdf/baxter.urdf at commit 6c4b0f3.
• M18ROS-Industrial — kuka_experimental. Type: consortium-maintained description repository, not a vendor publication. License: Apache-2.0. Retrieved: 2026-08-26.github.com/ros-industrial/kuka_experimentalSupports: the finding that plain URDFs for four KUKA arms exist and assemble, but are published by the ROS-Industrial consortium rather than by KUKA.
• M19Blue Robotics — Technical Reference, BlueBoat vehicle row. Type: official product/CAD documentation. Retrieved: 2026-08-28.bluerobotics.com/learn/technical-referenceSupports: BlueBoat official vehicle and component CAD download links. Model publication is not a payload or performance source.
• M20Booster Robotics — booster_assets. Type: official robot-description and motion-data repository. License: BSD-3-Clause. Commit pinned: 38a0ae8. Retrieved: 2026-08-30.github.com/BoosterRobotics/booster_assetsSupports: K1 22-DoF, T1 23-/29-DoF, and T2 31-DoF upstream identities, published URDF / MuJoCo XML formats, model links, licenses, and live browser assemblies. This is model-publication evidence, not a source for hardware performance.
• M21LimX Dynamics — humanoid-description. Type: official full-size humanoid description repository. License: Apache-2.0. Commit pinned: 02adfbd. Retrieved: 2026-08-30.github.com/limxdynamics/humanoid-descriptionSupports: HU_D04 base / gripper / dexterous-hand model ids; URDF, SRDF, MJCF, USD, and mesh paths; live HU_D04 base assembly. The repository does not map HU_D04 to the Oli product name, so the landscape does not infer that identity.
• M22LimX Dynamics — tron2-robot-description. Type: official modular robot-description repository. License: Apache-2.0. Commit pinned: 9939c22. Retrieved: 2026-08-30.github.com/limxdynamics/tron2-robot-descriptionSupports: TRON2A publication record and the explicit separation of dual-arm, wheel-foot, sole-foot, combined humanoid, and centaur configurations. These variants are not collapsed into one payload class.
• M23Leju Robotics — kuavo-ros-opensource. Type: official ROS / simulation repository. Repository-root license: not declared when inspected. Commit pinned: 2be4881. Retrieved: 2026-08-30.github.com/LejuRobotics/kuavo-ros-opensourceSupports: Kuavo xacro / URDF and simulator-resource publication; version- and configuration-generated state. Internal model directories are not treated as proof of a retail SKU mapping.
• M24Hello Robot — stretch4_urdf. Type: official robot-description Python package. License: Clear BSD in LICENSE.md (GitHub SPDX reports NOASSERTION). Commit pinned: cd810ce. Retrieved: 2026-08-30.github.com/hello-robot/stretch4_urdfSupports: Stretch 4 / SE4 composable xacro, batch and end-effector configuration, STL publication, and Python URDF export. No browser card is claimed because a complete robot URDF must be generated.
• M25UC Berkeley Hybrid Robotics — Berkeley-Humanoid-Lite-Assets. Type: official open-research hardware assets. License: CC BY-SA 4.0. Commit pinned: fc90fed. Retrieved: 2026-08-30.github.com/HybridRobotics/Berkeley-Humanoid-Lite-AssetsSupports: URDF, MJCF, USD, STL paths, asset license, and live browser assembly. Hardware costs and performance remain sourced to the research paper/page.
• M26Stanford Robotics — ToddlerBot. Type: official open-research hardware and software repository. License: MIT code; CC BY-NC-SA 4.0 design and STL assets. Commit pinned: e337f3b. Retrieved: 2026-08-30.github.com/hshi74/toddlerbotSupports: exact toddlerbot_2xc URDF / MJCF artifact, asset terms, source links, and live browser assembly. No derived static render is re-hosted.
• M27NASA / OpenHumanoids — val_description. Type: maintainer-identified Valkyrie robot-description package. License: NASA 1.3. Commit pinned: 5257ff7. Retrieved: 2026-08-30.github.com/openhumanoids/val_descriptionSupports: Valkyrie source-only entry; the README says catkin generates the complete URDF from xacro and included meshes. The lone checked-in multisense URDF is not the complete robot.
• M28Hello Robot — stretch_urdf. Type: official Stretch 3 robot-description Python package. License: Clear BSD in LICENSE.md. Commit pinned: c8c6e27. Retrieved: 2026-09-08.github.com/hello-robot/stretch_urdfSupports: Stretch 3 / SE3 checked-in URDF stretch_description_SE3_eoa_wrist_dw3_tool_sg3.urdf and adjacent STL meshes. This is the SE3 standard-gripper (sg3) assembly, not Stretch 4 / SE4 xacro (M24) and not the tablet/nil/pro end-effector variants in the same directory.
• M29Fetch Robotics — fetch_ros / fetch_description. Type: official ROS robot-description package. License: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International, declared in fetch_description/package.xml. Commit pinned: d22b98f on default branch ros1. Retrieved: 2026-09-08.github.com/fetchrobotics/fetch_rosSupports: Fetch wheeled mobile-manipulator URDF fetch_description/robots/fetch.urdf and package meshes. Because the description assets are CC BY-NC-SA 4.0, no derived static render is re-hosted; the card loads live only. This is Fetch, not Freight, and not a Galbot G1 substitute.
• M30ANYbotics — anymal_c_simple_description. Type: official simplified simulation description. License: BSD-3-Clause. Commit pinned: 87b6851. Retrieved: 2026-09-08.github.com/ANYbotics/anymal_c_simple_descriptionSupports: ANYmal C simple-description URDF urdf/anymal.urdf and Collada meshes. ANYbotics publishes this as a simplified simulation description, not as production ANYmal C and not as Generation D. It is not merged with the ANYmal B simple description (M12) or the ANYmal D datasheet.
• M31Kinova — ros_kortex / kortex_description. Type: official ROS robot-description repository. License: BSD-3-Clause style at repository root; package.xml license BSD. Commit pinned: c3280d8 on noetic-devel. Retrieved: 2026-09-10.github.com/Kinovarobotics/ros_kortexSupports: Kinova Gen3 7-DoF vision arm URDF kortex_description/arms/gen3/7dof/urdf/GEN3-7DOF-VISION_ARM_URDF_V12.urdf (9453 bytes, SHA256 262a15a7c86fc484b99087be8d787d00d442241421fcc274c8da26117d03a9bf) and the eight STL paths under arms/gen3/7dof/meshes/. This is the vision configuration without a gripper. NOVISION and BRAKES files are not counted separately. No mesh-specific license file was found beside the root LICENSE.
• M32Franka Robotics — franka_description. Type: official ROS 2 robot-description repository. License: Apache-2.0 (LICENSE and NOTICE). Package 2.9.0. Commit pinned: 7aeeddc. Retrieved: 2026-09-10.github.com/frankarobotics/franka_descriptionSupports: the vendor scripts/create_urdf.py fr3 --robot-ee franka_hand recipe, model-specific Xacro under robots/fr3/, and package://franka_description/ meshes. The browser preview is a labeled recipe product, not an upstream-published standalone URDF. This is FR3, not FR3v2, FP3, FER/Panda, or FR3 Duo. Quantitative FR3 numbers stay on industrial-arms M10, not this description.
• M33Universal Robots — Universal_Robots_ROS2_Description UR5e recipe. Type: official ROS 2 robot-description repository. License: BSD-3-Clause for UR5e meshes and description files. Package ur_description 4.3.1. Commit pinned: 89bbe79. Retrieved: 2026-09-10.github.com/UniversalRobots/Universal_Robots_ROS2_DescriptionSupports: the official launch/view_ur.launch.xml xacro recipe with ur_type:=ur5e, config/ur5e/default_kinematics.yaml as generic factory defaults (not a calibrated robot), and BSD-3-Clause UR5e meshes. The browser preview is a labeled recipe product, not an upstream-published standalone URDF. This is UR5e, not UR5 or UR20.

Providers index

• A1IFR — World Robotics 2026 Industrial Robots, Sources & Methods. Type: analyst / classification. Date: 2026. Retrieved: 2026-10-09. Review after: 2027-09-01. Review reason: Recheck against World Robotics 2027 sources and methods.ifr.orgSupports: providers index — the industrial-robot class scheme used to group entries. The public document names no producers (read 2026-10-09); provider names come from vendor sites.
• A2IFR — World Robotics 2026 Service Robots, Sources & Methods. Type: analyst / classification. Date: 2026. Retrieved: 2026-10-09. Review after: 2027-09-01. Review reason: Recheck against World Robotics 2027 sources and methods.ifr.orgSupports: providers index — the service, medical, consumer, aerial and marine class scheme used to group entries, and IFR’s count of 1,001 known service robot producers. The per-company list is in the paid report and is not cited; provider names come from vendor sites.

Wheeled mobile manipulators

• M1Galbot — G1 Quick Start Guide. Type: manual. Version: v2.2.4. Retrieved: 2026-08-25.developer.galbot.com/docs/g1/2.2.4Supports: G1 family + field tables (geometry, payload, sensors, energy, environment).
• M2Galbot — S1 快速使用手册. Type: manual. Version: v2.0.0. Retrieved: 2026-08-25.developer.galbot.com/docs/s1/2.0.0Supports: S1 family + field tables.
• M3GalaxyGeneralRobotics — galbot_one_golf_description. Type: official description repository. License: Apache-2.0. Retrieved: 2026-08-25.github.com/GalaxyGeneralRobotics/galbot_one_golf_descriptionSupports: G1 URDF / MJCF / USD model-asset links.
• M4GalaxyGeneralRobotics — galbot_s1_description. Type: official description repository. License: Apache-2.0. Retrieved: 2026-08-25.github.com/GalaxyGeneralRobotics/galbot_s1_descriptionSupports: S1 URDF / MJCF / USD model-asset links.
• P1The Robot Report — Galbot $300M round. Type: press. Date: 2025-12-23. Retrieved: 2026-08-25.therobotreport.comSupports: deploy / company press card.
• M5Hello Robot — Stretch 3 product page. Type: manual / product page. Retrieved: 2026-08-27.hello-stretch3.com/stretch-3-productSupports: Stretch 3 listed price $24,950, mass 24.5 kg, 33×34×141 cm (normalized to mm), 2–5 h runtime, Intel NUC 12, 7 manipulator DoF, ROS 2 / Python.
• M6Hello Robot — Stretch 3 Hardware Guide. Type: vendor hardware manual. Retrieved: 2026-08-27.github.com/hello-robot/stretch_hardware_guidesSupports: arm reach 0.52 m, recommended max payload 2 kg, gripper mass 240 g, D435if / D405 / RP-Lidar A1, charger 12 V / 7 A. Not used as a Galbot wrist-center payload.
• M7AgiBot (智元) — AGIBOT G1 general-purpose robot product page. Type: manual / product page. Retrieved: 2026-08-27.agibot.com/products/G1Supports: AgiBot G1 column in the wheeled dual-arm field table. Not Galbot G1 and not Unitree G1.
• M8Pollen Robotics — Reachy 2 product page. Type: manual / product page. Retrieved: 2026-08-29.pollen-robotics.com/reachy-2Supports: Reachy 2 height 136–166 cm (normalized to 1360–1660 mm), full-robot mass up to 50 kg, ~3 kg payload per arm, ROS 2 Humble / Python SDK, stereo + RGB-D, LiDAR, omnidirectional base. Not Reachy v1.
• M9Pollen Robotics — Reachy 2 general specifications (hardware guide). Type: vendor hardware manual. Retrieved: 2026-08-29.docs.pollen-robotics.comSupports: Reachy 2 7-DoF Orbita arms, 3-DoF Orbita neck, fanless industrial CPU PC, AI off-board note.
• M10PAL Robotics — TIAGo product page specification table. Type: manual / product page. Retrieved: 2026-08-29.pal-robotics.com/robot/tiagoSupports: TIAGo height 110–145 cm (normalized to 1100–1450 mm), footprint ø 54 cm (540 mm), 3 kg arm payload without end-effector, 7-DoF arms, 4–5 h / 8–10 h autonomy, Ubuntu LTS, 0–2 arm configurator. Not a walking humanoid and not mixed into Galbot wrist-center cells.

Humanoids

• M1Unitree — R1 product page. Type: manual / product page. Retrieved: 2026-08-27. Review after: 2026-11-25. Review reason: Mutable vendor product SPA; re-inspect the live table 90 days after retrieval.unitree.com/R1Supports: R1 AIR / R1 family columns.
• M2Unitree — G1 product page. Type: manual / product page. Retrieved: 2026-08-27. Review after: 2026-11-25. Review reason: Mutable vendor product SPA; re-inspect the live table 90 days after retrieval.unitree.com/g1Supports: G1 / G1 EDU columns; field table G1 column.
• M3Unitree — H1 / H1-2 product page. Type: manual / product page. Retrieved: 2026-08-27. Review after: 2026-11-25. Review reason: Mutable vendor product SPA; re-inspect the live table 90 days after retrieval.unitree.com/h1Supports: H1 / H1-2 columns; H-series deep sheet.
• M4Unitree — H2 product page. Type: manual / product page. Retrieved: 2026-08-27. Review after: 2026-11-25. Review reason: Mutable vendor product SPA; re-inspect the live table 90 days after retrieval.unitree.com/H2Supports: H2 column; field table H2 column.
• M5Generation Robots — AgiBot A2 datasheet. Type: reseller datasheet (not a vendor manual). Retrieved: 2026-08-25.static.generation-robots.comSupports: field table AgiBot A2 column (labeled reseller in every cell).
• M6Unitree — unitree_ros. Type: official description repository. License: BSD-3-Clause. Retrieved: 2026-08-25.github.com/unitreerobotics/unitree_rosSupports: engineering-models panel (URDF).
• M7Unitree — unitree_mujoco. Type: official description repository. License: BSD-3-Clause. Retrieved: 2026-08-25.github.com/unitreerobotics/unitree_mujocoSupports: engineering-models panel (MJCF).
• M8Unitree — unitree_model dataset. Type: official model dataset. License: BSD-3-Clause. Retrieved: 2026-08-25.huggingface.co/datasets/unitreerobotics/unitree_modelSupports: engineering-models panel (URDF / MJCF / USD / STL / DAE).
• M9AgiBot — X2 series product specifications. Type: manual / product table. Retrieved: 2026-08-27.agibot.com/products/X2Supports: X2 / X2 Ultra family table (envelope, DoF split, payload excl. EE, 120 N·m, ≈500 Wh, RK3588 / Orin NX). Distinct from the A2 reseller datasheet.
• M10AgiBot AimDK — X2 robot specifications (Ultra edition). Type: vendor SDK/docs table. Retrieved: 2026-08-27.x2-aimdk.agibot.comSupports: footnote only — Ultra battery ≈421 Wh, which conflicts with the product table’s ≈500 Wh. Not used as the cell source.
• M11Fourier Intelligence — GR-2 product page. Type: manual / product page. Retrieved: 2026-08-27.fftai.com/products-gr2Supports: GR-2 height 175 cm (normalized to 1750 mm), mass 63 kg, 12-DoF hands, peak torque 380 N·m, walking speed 5 km/h. The Fourier 文档中心 GR-2 Introduction URL returned HTTP 404 from this checkout.
• P1Fourier Intelligence — “Fourier Unveils the Next-Generation Humanoid Robot GR-2”. Type: press / launch note. Date: 2024-09-26. Retrieved: 2026-08-27.fftai.com/newsroom-newintech/14Supports: GR-2 total 53 DoF and 3 kg single-arm load (press-graded cells).
• M12UBTECH — Walker S2 product page. Type: manual / product page. Retrieved: 2026-10-09.ubtrobot.com/en/humanoid/products/walker-s2Supports: Walker S2 payload 15 kg within 0–1.8 m workspace, waist ±162°, whole-body pitch 125°, autonomous battery swap ≤3 min, RGB binocular stereo, product-page “mass production and delivery” status. Height 176 cm, weight 70 kg, 52 DoF and walking 2 m/s are printed only in a specification image on the page (owebsite-cdn.ubtrobot.com/resources/image/2025/07/25/701620707123269.png, read 2026-10-09), which also prints hand grasp loads of 7.5 kg and 1 kg. Battery energy and runtime are not published.
• M13Leju Robotics — Kuavo 4.1 product introduction (ROS manual). Type: manual / vendor docs. Retrieved: 2026-08-28.kuavo.lejurobot.comSupports: Kuavo 4.1 column — 1.55 m / 50 kg / 626 mm arm, 28-DoF composition (spec table lists 30), 0.4 m/s walk, 60 V 8 Ah, D435, Mofang i9-13900, NUC12 / optional AGX Orin, knee/hip-roll 40 N·m.
• M14Leju Robotics — Kuavo 4pro MaxB product introduction (ROS manual). Type: manual / vendor docs. Retrieved: 2026-08-28.kuavo.lejurobot.comSupports: Kuavo 4pro MaxB column — 1.66 m / 55 kg / 0.79 m arm, 28 DoF excl. EE, 0.4 m/s walk, 5 km/h run, 60 V 6 Ah, Gemini-335L + D405 ×2 + Mid-360, AGX Orin, knee/hip-roll 40 N·m.
• M15Leju Robotics — Kuavo 5 MaxB product introduction (ROS manual). Type: manual / vendor docs. Retrieved: 2026-08-28.kuavo.lejurobot.comSupports: Kuavo 5 MaxB column — 1.73 m / 63.5 kg / 0.79 m arm, 29 DoF excl. EE (adds waist yaw), 0.4 m/s walk, 5 km/h run, 60 V 6 Ah, Gemini-335L + D405 ×2 + Mid-360, GENE-MTH6 + AGX Orin, knee 90 N·m.
• M16Booster Robotics — T1 product page. Type: manual / product table. Retrieved: 2026-08-29.booster.tech/booster-t1Supports: T1 Basic / Standard / Customized family table — 1180 × 470 × 230 mm, ~30 kg, 23-DoF base listing, 31-DoF with gripper, 41-DoF with dexterous hand, 130 N·m peak joint, 10.5 Ah, 2 h walking / 4 h standing, AGX Orin 200 TOPS. Hand DoF is a separate row. RobotShop / Generation Robots reseller pages are not used as cells.
• M17Booster Robotics — official open-source page and booster_robotics_sdk. Type: official SDK / developer documentation. Retrieved: 2026-08-29.booster.tech/open-sourceSupports: T1 and K1 named on the open-source portal; SDK / ROS 2 / Fast-DDS pointer. Hardware numbers come from the T1 and K1 product pages, not from this portal. GitHub: github.com/BoosterRobotics/booster_robotics_sdk.
• M18LimX Dynamics — Oli (CL-3) English product page. Type: manual / product page. Retrieved: 2026-08-29.limxdynamics.com/en/products/oliSupports: Oli 165 cm (normalized to 1650 mm), 31 DoF split (head 2 / waist 3 / arm 7 / leg 6), quick-swap battery named without V/Ah/Wh, head and chest depth cameras, 6-axis IMU, claimed complete URDF and Isaac Sim / MuJoCo / Gazebo support. Mass, payload, runtime unpublished on this page. This is a biped, not a wheeled dual-arm.
• M19Booster Robotics — K1 product page. Type: manual / product table. Retrieved: 2026-08-29.booster.tech/booster-k1Supports: K1 Geek / Education / Professional family table — ~95 cm (normalized to ~950 mm), ~19.5 kg, 22 DoF (head 2 / arm 4 / leg 6), 60 N·m peak joint, 2 Ah vs 5 Ah, 30 min vs 80 min walking at 0.4 m/s, 48 / 117 / 200 TOPS, from $5,999 starting price. Battery energy unpublished (no voltage). Reseller pages are not used as cells. Not T1.
• M20PAL Robotics — TALOS product page. Type: manual / product page. Retrieved: 2026-08-29.pal-robotics.com/robot/talosSupports: TALOS family table — 175 cm normalized to 1750 mm, 95 kg, 6 kg arm payload with the arm fully extended, 1.5 h walking / 3 h stand-by, EtherCAT 2 kHz up to 5 kHz, torque sensors except head / wrists / grippers, F/T on wrists and ankles via ros_control, ROS / Ubuntu LTS, URDF claimed without a commit pin. Total DoF unpublished. Not TIAGo. The ROS wiki is not used as a spec source.
• M21Galbot — ET1 / 银河星仔 product specification page. Type: manual / product table. Retrieved: 2026-10-09. Review after: 2026-12-02. Review reason: Mutable vendor product SPA; re-inspect the live table.galbot.com/et1Supports: ET1 Standard / Pro / Flagship family table — standing height 1250 mm, mass with battery 32 kg, Standard 26 DoF, Pro/Flagship 26 + 12 with dexterous hands, stereo camera, main processor RK3588 + RDK X5, Pro Orin Nano 40TOPS, Flagship NX 100TOPS. Payload, speed, battery, runtime, IP, and list price unpublished. A Buy Now control is not a price: it links to a JD store listing that shows no price without login (read 2026-10-09; table unchanged). Not Galbot G1 and not Unitree G1. Aggregator 173 cm / 65 kg / 48 DoF figures are not copied.
• M22EngineAI — PM01 Chinese product page, 2026 Education and Navigation editions. Type: manual / product table. Retrieved: 2026-09-04.engineai.com.cn/product-pm01.htmlSupports: current edition envelope 1400 × 540 × 295 mm, about 44.5 kg with battery, 686.5 mm leg, 1440 mm arm span, 24 total DoF, subsystem counts as printed, 164 N·m maximum torque, 145 N·m/kg peak density, >2 m/s hardware support, 10,000 mAh quick-release battery, about 2 h runtime and charge, camera / LiDAR split, 8-core CPU, Orin NX 16 GB, and secondary-development listing. The Education entry literally prints 1 single-leg DoF; the Navigation entry prints 6. No correction is inferred.
• M23EngineAI — PM01 English product page. Type: manual / legacy product table. Retrieved: 2026-09-04.en.engineai.com.cn/product-pm01.htmlSupports: version-conflict note only — English Business / Education listings retain 1400 × 535.55 × 252.66 mm, about 42 / nearly 43 kg, 23 / 24 DoF, 4-core base CPU, Q90H 145 N·m and 130 N·m/kg. These values are not merged into the Chinese 2026 columns.
• M24Fauna Robotics — Sprout Creator Edition product page. Type: manual / product page. Retrieved: 2026-09-04.faunarobotics.com/productSupports: 29 DoF, 107 cm, 22.7 kg, 2-DoF neck, 6-DoF arms, 5-DoF legs, 1-DoF grippers, ZED 2i, four time-of-flight sensors, torso IMU, AGX Orin 64 GB, 1 TB SSD, swappable battery, 3–3.5 h runtime, E-stop, and link to the vendor specification PDF.
• M25Fauna Robotics — Sprout Creator Edition Specifications. Type: vendor specification PDF. Retrieved: 2026-09-04.cdn.prod.website-files.com/Sprout_Specifications.pdfSupports: 107 × 45 × 21 cm envelope, 48 cm arm, 22.7 kg, payload matrix by pose and duration, construction, compute, sensing, 46.8 V and 5000 / 10,000 mAh packs, runtime, connectivity, APIs, simulation compatibility, safety layers, IP20 indoor environment limits, and warranty. Payloads are theoretical per arm and may not be supported by the current teleoperation controller.
• M26NEURA Robotics — 4NE1 product page. Type: manual / product page. Published: 2025-05-11. Modified: 2026-08-12. Retrieved: 2026-09-04.neura-robotics.com/products/4ne1Supports: 4NE1 identity, 180 cm, 80 kg, 5 km/h, vendor payload label 10–100 kg, 360° perception, sensor skin, remote operation, exchangeable forearms, and the warning that wheel and seventh-axis models exist. Generic-page values are corroborated and version-scoped by the Gen 3.5 datasheet.
• M27NEURA Robotics — 4NE1 Gen 3.5 Datasheet, V6 / 02.01.2026. Type: vendor datasheet. Retrieved: 2026-09-04.neurarobotics.px.media/4NE1Gen3.5Datasheet.pdfSupports: Gen 3.5 height 180 cm, mass 80 kg, speed 5 km/h, 12-DoF own hands, 10–100 kg application-dependent payload label, exchangeable forearms, 360° perception, sensor skin, Thor T5000 label, 24/7 operating-time label, autonomous mode, advanced battery-management label, and optional human-detection-sensor caveat. The payload range is not interpreted as arm load.
• M28NEURA Robotics — 4NE1 Gen 3.5 reservation page. Type: vendor commerce / status page. Published: 2026-01-02. Modified: 2026-05-18. Retrieved: 2026-09-04.neura-robotics.com/product/4ne1-reservationSupports: Gen 3.5 biped identity, integrated high-dexterity hands, Wi-Fi 6 / Ethernet / Python / ROS 2 / C++ interfaces, digital twin and teleoperation access, estimated €98,000 unit price for 1–19 and €60,000 for 20+ excluding tax and shipping, refundable €100 reservation fee, and expected end-of-2026 availability. Reservation is not delivery.
• M29EngineAI — T800 English product page. Type: manual / product page. Retrieved: 2026-10-09.en.engineai.com.cn/product-t800Supports: T800 Basic / Open Source / Pro / Max columns — 173 cm, 75–85 kg, total DoF 25 / 25 / 43 / 46 with per-limb split, 450 N·m, ≥2 m/s, 3–5 h, sensing, compute and development support. Body text conflicts (29 DoF, up to 4 h) and the 5 kg hand payload stay in the claim boundary.
• M30EngineAI — T800 Chinese product page. Type: manual / product page. Retrieved: 2026-10-09.engineai.com.cn/product-t800Supports: cross-check of the T800 columns — RK3588S base compute; Open Source 3D LiDAR not listed. It names the second edition 开发版, gives it the same head wide-FOV camera as Basic, and prints no Thor upgrade clause. Its Pro column is misaligned and is not used.
• M31Fourier — About Fourier N1 (documentation centre, old-version host). Type: manual / vendor documentation. Last modified: 2026-06-23. Retrieved: 2026-10-09.support-old.fftai.comSupports: Fourier N1 sheet — introduction “stands 1.3 meters tall and weighs 38 kg”, 23 DoF (5 per arm, 1 waist, 6 per leg), maximum stable running speed 3.5 m/s; core parameter table 1245 × 441 × 202 mm, arm span 1240 mm, ≈39 kg, ≈460 W, walking 2.7 km/h, running 10.8 km/h, peak joint torque 144 N·m, battery 475 Wh, ≤3.2 kg, ≈2 h, charging ≈3.5 h, Intel Core i7-13700H. The same path on support.fftai.com returns HTTP 404 (2026-10-09). Payload is not published.
• M32Fourier — N1 product page. Type: manual / product page. Retrieved: 2026-10-09.fftai.com/ecosystem/n1Supports: Fourier N1 sheet, product-page values — height 130 cm, weight 38 kg, maximum speed 3.5 m/s, battery 2 h, 23 DoF, “over 2 hours of continuous runtime per charge”. A script-rendered page; the strings are in the site’s language bundle. Payload is not published.

Quadrupeds

• M1Unitree — B2 product page. Type: manual / product page. Retrieved: 2026-08-25. Review after: 2026-11-23. Review reason: Mutable vendor product SPA; re-inspect the live table 90 days after retrieval.unitree.com/b2Supports: B2 column (standing/walking load separate rows).
• M2Boston Dynamics — Spot product page. Type: manual / product page. Retrieved: 2026-08-25.bostondynamics.com/products/spotSupports: Spot column.
• M3Unitree — unitree_ros. Type: official robot-description repository. License: BSD-3-Clause. Retrieved: 2026-08-25.github.com/unitreerobotics/unitree_rosSupports: engineering-models panel.
• M4Unitree — unitree_mujoco. Type: official simulation repository. License: BSD-3-Clause. Retrieved: 2026-08-25.github.com/unitreerobotics/unitree_mujocoSupports: engineering-models panel.
• M5Unitree — unitree_model dataset. Type: official model dataset. License: BSD-3-Clause. Retrieved: 2026-08-25.huggingface.co/datasets/unitreerobotics/unitree_modelSupports: engineering-models panel.
• M6Unitree — Go2 product page. Type: manual / product page. Retrieved: 2026-08-27. Review after: 2026-11-25. Review reason: Mutable vendor product SPA; re-inspect the live table 90 days after retrieval.unitree.com/go2Supports: Go2 AIR / PRO / X / EDU family table (payload, speed, battery mAh, listed prices). Not used for B2 standing/walking load.
• M7ANYbotics — ANYmal technical specifications, Generation D. Type: vendor datasheet PDF. Document: V.1 – May 2022, ANY-TECR-000082-00. Retrieved: 2026-08-29.anybotics.com/anymal-technical-specifications.pdfSupports: ANYmal D family table — 930 × 530 × 890 mm walking envelope, 50 kg / 55.7 kg with inspection payload, 5.7 kg payload-unit mass, 10 kg nominal / 15 kg reduced mounted payload, 1.3 m/s max / 0.75 m/s recommended, IP67, 932.4 Wh, 90–120 min. Standing transport load and total DoF are unpublished. Not ANYmal B.
• M8Ghost Robotics — Vision 60 product page Technical Specifications block. Type: manual / product table. Retrieved: 2026-08-29.ghostrobotics.io/vision-60Supports: Vision 60 family table — 950 × 570 × 685 mm standing, 419 mm ride height, 51 kg tare, 10 kg stated payload capacity with user-selectable compensation, 0.9 / 1.2 / 2.4 m/s (working towards 3.0), 3.15 h walking at 0.9 m/s / 21 h standby, IP67, −40 to 55 °C with cold start not below −20 °C, 3 DoF/leg 12-motor, NVIDIA Xavier 32 GB, 5× RGB 1080p, 4× D435, dual-antenna RTK GPS. Marketing “up to 2.5 m/s” is not copied over the spec block. Standing transport load unpublished. Not B2 standing load.

Wearables

• M1German Bionic — “Exia Explained #5 — the world's most powerful exoskeleton”. Type: company-published specification. Retrieved: 2026-08-25.germanbionic.comSupports: Exia assistance figure (up to 38 kg adaptive lift support).
• M2Hypershell — Hypershell X series product page. Type: manual / product page. Retrieved: 2026-08-27.hypershell.techSupports: X Go / X Pro gait-wearable table (motor power, headline mass, battery type, pack count, capacity per pack, assisted range, intelligent modes, IP, anti-cold rating, medical-device status, dated price). Torque, motor count, speed assistance, charge time and fit range are not attributable to a named model on this page.

End-effectors & components

• M1Unitree — Dex5-1 product page. Type: manual / product page. Retrieved: 2026-08-25. Review after: 2026-11-23. Review reason: Mutable vendor product SPA; re-inspect the live table 90 days after retrieval.unitree.com/Dex5-1Supports: Dex5-1 / Dex5-1P dexterous-hand table.
• M2Robotiq — 2F-85 / 2F-140 Instruction Manual, §6 Specifications. Type: manual. Retrieved: 2026-08-25.assets.robotiq.comSupports: 2F-85 / 2F-140 parallel-gripper table.
• M3Unitree — unitree_ros. Type: official robot-description repository. License: BSD-3-Clause. Retrieved: 2026-08-25.github.com/unitreerobotics/unitree_rosSupports: engineering-models panel.
• M4Unitree — unitree_model dataset. Type: official model dataset. License: BSD-3-Clause. Retrieved: 2026-08-25.huggingface.co/datasets/unitreerobotics/unitree_modelSupports: engineering-models panel.
• M5Galbot — G1 Teleoperation Master-Slave Arm TM01 Manual. Type: manual. Version: v1.4.0. Retrieved: 2026-08-27.developer.galbot.com/docs/tm01/1.4.0Supports: TM01 accessory table (DoF, size, mass, voltage, power, sampling, mapping latency, USB, IP20, temperature, humidity, 12-month warranty).
• M6Sharpa — Wave product specification table. Type: vendor product page. Retrieved: 2026-08-27.sharpa.com/pages/waveSupports: Sharpa Wave two-column hardware/tactile/interface table. Columns are unnamed on the source; values are not collapsed.
• M7Robotiq — 2F-85 / 2F-140 Instruction Manual, §3.3 Environmental and Operating Conditions. Type: manual. Retrieved: 2026-08-27.assets.robotiq.comSupports: parallel-gripper environment rows (IP40, −10 to 50 °C operating, −30 to 70 °C storage, 20–80% RH, 2 G / 5 G vibration). §6 does not print these; they are in §3.3 of the same manual.
• R1LDA-1B — “Scaling Latent Dynamics Action Model via Universal Embodied Data Ingestion”. Type: research. ID: arXiv:2602.12215. Retrieved: 2026-08-27.arxiv.org/abs/2602.12215Supports: Galbot G1 real-world figures using a parallel-jaw gripper and a Sharpa 22-DoF hand, plus a BrainCo 10-DoF variant on Unitree G1. Not a product-option proof and not a BrainCo Revo 2 identity.
• R2Shaw, Agarwal, Pathak — LEAP Hand: Low-Cost, Efficient, and Anthropomorphic Hand for Robot Learning. Type: research paper. ID: arXiv:2309.06440. Venue: RSS 2023. Retrieved: 2026-08-29.arxiv.org/abs/2309.06440Supports: LEAP Hand class and five-finger anthropomorphic identity. The abstract restates 4-hour / 2000 USD assembly; those cells cite the project page. Active DoF is not extracted from the abstract and stays empty.
• M8LEAP Hand — project page (v1). Type: official project / hardware page. Retrieved: 2026-08-29.v1.leaphand.comSupports: LEAP Hand assembly time (4 hours) and claimed kit cost (2000 USD). The advertised github.com/leap-hand/LEAP_Hand URL returned HTTP 404 from this checkout, so there is no repo pin or 3D card. A later v2 / tendon-driven family named on the same page is not this column. Not Dex5.

Research systems

• R1LATENT — “Learning Athletic Humanoid Tennis Skills…”. Type: research paper. ID: arXiv:2603.12686. Retrieved: 2026-08-25.arxiv.org/abs/2603.12686Supports: research results table (robot body, success rates, sensing); LATENT paper pin.
• R2LATENT — official project page and video gallery. Type: research project page. Retrieved: 2026-08-25.zzk273.github.io/LATENTSupports: project videos; racket-MJCF research asset.
• R3Galbot — LATENT highlight video (official host). Type: official media. Retrieved: 2026-08-25.oss-cn-beijing.galbot.com/online/blog/tennis.mp4Supports: LATENT pin and detail highlight video.
• R4Galbot — official research highlight videos on the company OSS blog. Type: official media. Retrieved: 2026-08-26.oss-cn-beijing.galbot.com/online/blogSupports: WAM-TTT, LDA-1B, GraspVLA, GroceryVLA, NavFoM, and DexNDM pin videos.
• R5WAM-TTT — “Steering World-Action Models by Watching Human Play at Test Time”. Type: research paper. ID: arXiv:2607.06988. Retrieved: 2026-08-26.arxiv.org/abs/2607.06988Supports: WAM-TTT paper pin.
• R6WAM-TTT — official code repository. Type: research code. Retrieved: 2026-08-26.github.com/Cocoyawn/WAM-TTTSupports: WAM-TTT code link.
• R7Humanoid-GPT — “Humanoid Generative Pre-Training for Zero-Shot Motion Tracking”. Type: research paper. ID: arXiv:2606.03985. Retrieved: 2026-08-26.arxiv.org/abs/2606.03985Supports: Humanoid-GPT paper pin; Unitree G1 real-robot boundary.
• R8Humanoid-GPT — official project page and home-scene video. Type: research project page. Retrieved: 2026-08-26.qizekun.github.io/Humanoid-GPTSupports: Humanoid-GPT pin video (videos/home.mp4).
• R9LDA-1B — “Scaling Latent Dynamics Action Model via Universal Embodied Data Ingestion”. Type: research paper. ID: arXiv:2602.12215. Retrieved: 2026-08-26.arxiv.org/abs/2602.12215Supports: LDA-1B paper pin; Galbot G1 and Unitree G1 eval boundary.
• R10LDA-1B — official project page. Type: research project page. Retrieved: 2026-08-26.pku-epic.github.io/LDASupports: LDA-1B project link; EI-30k dataset figure.
• R11StereoVLA — “Enhancing Vision-Language-Action Models with Stereo Vision”. Type: research paper. ID: arXiv:2512.21970. Retrieved: 2026-08-26.arxiv.org/abs/2512.21970Supports: StereoVLA paper pin.
• R12StereoVLA — official project page and demo video. Type: research project page. Retrieved: 2026-08-26.shengliangd.github.io/StereoVLA-WebpageSupports: StereoVLA pin video (videos/1-1.mp4).
• R13DexNDM — “Closing the Reality Gap for Dexterous In-Hand Rotation”. Type: research paper. ID: arXiv:2510.08556. Retrieved: 2026-08-26.arxiv.org/abs/2510.08556Supports: DexNDM paper pin.
• R14DexNDM — official project page. Type: research project page. Retrieved: 2026-08-26.meowuu7.github.io/DexNDMSupports: DexNDM project link.
• R15NavFoM — “Embodied Navigation Foundation Model”. Type: research paper. ID: arXiv:2509.12129. Retrieved: 2026-08-26.arxiv.org/abs/2509.12129Supports: NavFoM paper pin.
• R16NavFoM — official project page. Type: research project page. Retrieved: 2026-08-26.pku-epic.github.io/NavFoM-WebSupports: NavFoM project link.
• R17GraspVLA — “a Grasping Foundation Model Pre-trained on Billion-scale Synthetic Action Data”. Type: research paper. ID: arXiv:2505.03233. Retrieved: 2026-08-26.arxiv.org/abs/2505.03233Supports: GraspVLA paper pin.
• R18GraspVLA — official project page. Type: research project page. Retrieved: 2026-08-26.pku-epic.github.io/GraspVLA-webSupports: GraspVLA project link.
• R19TrackVLA — “Embodied Visual Tracking in the Wild”. Type: research paper. ID: arXiv:2505.23189. Retrieved: 2026-08-26.arxiv.org/abs/2505.23189Supports: TrackVLA paper pin.
• R20TrackVLA — official project page and teaser video. Type: research project page. Retrieved: 2026-08-26.pku-epic.github.io/TrackVLA-webSupports: TrackVLA pin video (static/videos/teaser_new.mp4).
• R21He Wang homepage — GroceryVLA product-model listing (no public arXiv) and UniPart (ECCV 2026, no arXiv on homepage). Type: official homepage. Retrieved: 2026-08-26.hughw19.github.ioSupports: GroceryVLA pin identity as a product model, not a paper; UniPart listing and name-collision note.
• R22DAPL — “Emerging Extrinsic Dexterity in Cluttered Scenes via Dynamics-aware Policy Learning”. Type: research paper. ID: arXiv:2603.09882. Retrieved: 2026-08-26.arxiv.org/abs/2603.09882Supports: DAPL paper pin.
• R23HDFlow — “Hierarchical Diffusion-Flow Planning for Long-horizon Tasks”. Type: research paper. ID: arXiv:2605.04525. Retrieved: 2026-08-26.arxiv.org/abs/2605.04525Supports: HDFlow paper pin.
• R24UrbanVLA — “A Vision-Language-Action Model for Urban Micromobility”. Type: research paper. ID: arXiv:2510.23576. Retrieved: 2026-08-26.arxiv.org/abs/2510.23576Supports: UrbanVLA paper pin and outdoor eval figure.
• R25TrackVLA++ — “Unleashing Reasoning and Memory Capabilities in VLA Models for Embodied Visual Tracking”. Type: research paper. ID: arXiv:2510.07134. Retrieved: 2026-08-26.arxiv.org/abs/2510.07134Supports: TrackVLA++ paper pin.
• R26SoFar — “Language-Grounded Orientation Bridges Spatial Reasoning and Object Manipulation”. Type: research paper. ID: arXiv:2502.13143. Retrieved: 2026-08-26.arxiv.org/abs/2502.13143Supports: SoFar paper pin.
• R27FoldNet — “Learning Generalizable Closed-Loop Policy for Garment Folding”. Type: research paper. ID: arXiv:2505.09109. Retrieved: 2026-08-26.arxiv.org/abs/2505.09109Supports: FoldNet paper pin.
• R28Dexonomy — “Synthesizing All Dexterous Grasp Types in a Grasp Taxonomy”. Type: research paper. ID: arXiv:2504.18829. Retrieved: 2026-08-26.arxiv.org/abs/2504.18829Supports: Dexonomy paper pin.
• R29Uni-NaVid — “A Video-based VLA Model for Unifying Embodied Navigation Tasks”. Type: research paper. ID: arXiv:2412.06224. Retrieved: 2026-08-26.arxiv.org/abs/2412.06224Supports: Uni-NaVid paper pin.
• R30MM-Nav — “Multi-View VLA Model for Robust Visual Navigation via Multi-Expert Learning”. Type: research paper. ID: arXiv:2510.03142. Retrieved: 2026-08-26.arxiv.org/abs/2510.03142Supports: MM-Nav paper pin.
• R31LIMMT — “Less is More for Motion Tracking”. Type: research paper. ID: arXiv:2606.06953. Retrieved: 2026-08-26.arxiv.org/abs/2606.06953Supports: LIMMT paper pin.
• R32CL3R — “3D Reconstruction and Contrastive Learning for Enhanced Robotic Manipulation Representations”. Type: research paper. ID: arXiv:2507.08262. Retrieved: 2026-08-26.arxiv.org/abs/2507.08262Supports: CL3R paper pin.
• R33FetchBot — “Learning Generalizable Object Fetching in Cluttered Scenes via Zero-Shot Sim2Real”. Type: research paper. ID: arXiv:2502.17894. Retrieved: 2026-08-26.arxiv.org/abs/2502.17894Supports: FetchBot paper pin.
• R34DexVLG — “Dexterous Vision-Language-Grasp Model at Scale”. Type: research paper. ID: arXiv:2507.02747. Retrieved: 2026-08-26.arxiv.org/abs/2507.02747Supports: DexVLG paper pin.
• R35GraspADMM — “Improving Dexterous Grasp Synthesis via ADMM Optimization”. Type: research paper. ID: arXiv:2603.13832. Retrieved: 2026-08-26.arxiv.org/abs/2603.13832Supports: GraspADMM paper pin.
• R36HumanTracker — “Towards Comprehensive and Human-Aligned Motion Tracking Benchmark”. Type: research paper. ID: arXiv:2608.13555. Retrieved: 2026-08-26.arxiv.org/abs/2608.13555Supports: HumanTracker benchmark pin.
• R37OmniSpatial — “Towards Comprehensive Spatial Reasoning Benchmark for Vision Language Models”. Type: research paper. ID: arXiv:2506.03135. Retrieved: 2026-08-26.arxiv.org/abs/2506.03135Supports: OmniSpatial benchmark pin.
• R38Open6DOR — official project page for open-instruction 6-DoF object rearrangement. Type: research project page. Retrieved: 2026-08-26.pku-epic.github.io/Open6DORSupports: Open6DOR benchmark pin.
• R39DexGraspNet — “DexGraspNet: A Large-Scale Robotic Dexterous Grasp Dataset…”. Type: research paper. ID: arXiv:2210.02697. Retrieved: 2026-08-26.arxiv.org/abs/2210.02697Supports: DexGraspNet dataset pin.
• R40DexGraspNet 2.0 — “Learning Generative Dexterous Grasping in Large-scale Synthetic Cluttered Scenes”. Type: research paper. ID: arXiv:2410.23004. Retrieved: 2026-08-26.arxiv.org/abs/2410.23004Supports: DexGraspNet 2.0 dataset pin.
• R41GAPartNet — “Cross-Category Domain-Generalizable Object Perception and Manipulation via Generalizable and Actionable Parts”. Type: research paper. ID: arXiv:2211.05272. Retrieved: 2026-08-26.arxiv.org/abs/2211.05272Supports: GAPartNet dataset pin.
• R42DREDS — “Domain Randomization-Enhanced Depth Simulation and Restoration…”. Type: research paper. ID: arXiv:2208.03792. Retrieved: 2026-08-26.arxiv.org/abs/2208.03792Supports: DREDS transparent/specular depth dataset pin.
• R43ASGrasp — “Generalizable Transparent Object Reconstruction and 6-DoF Grasp from RGB-D Active Stereo”. Type: research paper. ID: arXiv:2405.05648. Retrieved: 2026-08-26.arxiv.org/abs/2405.05648Supports: ASGrasp transparent-object pin.
• R44STOPNet — “Multiview-based 6-DoF Suction Detection for Transparent Objects”. Type: research paper. ID: arXiv:2310.05717. Retrieved: 2026-08-26.arxiv.org/abs/2310.05717Supports: STOPNet transparent suction pin.
• R45ScissorBot — “Learning Generalizable Scissor Skill for Paper Cutting”. Type: research paper. ID: arXiv:2409.13966. Retrieved: 2026-08-26.arxiv.org/abs/2409.13966Supports: ScissorBot paper pin.
• R46RoboHanger — “Learning Generalizable Robotic Hanger Insertion for Diverse Garments”. Type: research paper. ID: arXiv:2412.01083. Retrieved: 2026-08-26.arxiv.org/abs/2412.01083Supports: RoboHanger paper pin.
• R47D3RoMa — “Disparity Diffusion-based Depth Sensing for Material-Agnostic Robotic Manipulation”. Type: research paper. ID: arXiv:2409.14365. Retrieved: 2026-08-26.arxiv.org/abs/2409.14365Supports: D3RoMa paper pin.
• R48BODex — “Scalable and Efficient Robotic Dexterous Grasp Synthesis Using Bilevel Optimization”. Type: research paper. ID: arXiv:2412.16490. Retrieved: 2026-08-26.arxiv.org/abs/2412.16490Supports: BODex dataset/method pin.
• R49DyWA — “Dynamics-adaptive World Action Model for Generalizable Non-prehensile Manipulation”. Type: research paper. ID: arXiv:2503.16806. Retrieved: 2026-08-26.arxiv.org/abs/2503.16806Supports: DyWA paper pin.
• R50NaVid-4D — official project page for egocentric RGB-D VLN. Type: research project page. Retrieved: 2026-08-26.lhrrhl0419.github.io/NaVid4DSupports: NaVid-4D special-env pin.
• R51UniDexGrasp++ — “UniDexGrasp++: Improving Dexterous Grasping Policy Learning via Geometry-aware Curriculum and Iterative Generalist-Specialist Learning”. Type: research paper. ID: arXiv:2304.00464. Retrieved: 2026-08-26.arxiv.org/abs/2304.00464Supports: UniDexGrasp++ dataset/method pin.
• P1CCTV News (reprint) — “world's first fully autonomous” claim. Type: press. Date: 2026-03-17. Retrieved: 2026-08-25.digitalchina.gov.cnSupports: attributed “world's first” boundary.
• P2Beijing Daily / Beijing government report. Type: press. Date: 2026-04-09. Retrieved: 2026-08-25.beijing.gov.cnSupports: attributed “world's first” boundary; poster image source.
• P3Qiming Venture Partners — founder profile / 中国企业家 interview. Type: press. Date: 2025-09-11. Retrieved: 2026-08-27.qimingvc.com/cn/node/7399Supports: paper→body map product claims for GraspVLA, GroceryVLA, TrackVLA retail/tracking.
• P4Beijing Science and Technology Commission / Beijing Economy and IT — AstraBrain-WBC 0.5 release report. Type: press / government report. Date: 2026-06-24. Retrieved: 2026-08-27.kw.beijing.gov.cnSupports: Humanoid-GPT row — nearest public paper to the WBC 0.5 announcement; not a checkpoint-identity proof.
• R52Chi et al. — Berkeley Humanoid Lite: An Open-source, Accessible, and Customizable 3D Printed Humanoid Robot. Type: research paper. ID: arXiv:2504.17249. Retrieved: 2026-08-29.arxiv.org/abs/2504.17249Supports: Berkeley Humanoid Lite research hardware table — 800 mm, 16 kg, under $5,000 (U.S.), Intel N95, 6S 4000 mAh, ~30 min, CAN 2.0 at 1 Mbps / 250 Hz. Total DoF is not copied as a single figure. Not the later aggregator ~$15k claim.
• R53Berkeley Humanoid Lite — official project site. Type: research project page. Retrieved: 2026-08-29.lite.berkeley-humanoid.orgSupports: Berkeley Humanoid Lite project link on the research pin and hardware caption.
• R54HybridRobotics — Berkeley-Humanoid-Lite repository. Type: research code / hardware files. License: MIT code; CC BY-SA 4.0 assets. Retrieved: 2026-08-29.github.com/HybridRobotics/Berkeley-Humanoid-LiteSupports: license row and URDF / MJCF / USD presence. No browser assembly is published on this site.
• R55Shi et al. — ToddlerBot: Open-Source Miniature Humanoid for Loco-Manipulation Data Collection. Type: research paper. ID: arXiv:2502.00893. Venue: CoRL 2025. Retrieved: 2026-08-29.arxiv.org/abs/2502.00893Supports: ToddlerBot hardware table — 0.56 m normalized to 560 mm, 3.4 kg, 30 active DoF excluding end-effectors (7/arm, 6/leg, 2 neck, 2 waist), under $6,000, Jetson Orin NX 16GB, experiment lift 1484 g, experiment walking endurance 19 min. Not a Unitree body. Project-page 2.0 extras are not copied over these cells.
• R56ToddlerBot — official project site. Type: research project page. Retrieved: 2026-08-29.toddlerbot.github.ioSupports: ToddlerBot project link on the research pin and hardware caption. Paper cells win over project-page extras.
• R57hshi74/toddlerbot — official code repository. Type: research code / hardware files. Retrieved: 2026-08-29.github.com/hshi74/toddlerbotSupports: ToddlerBot code-repository row. Stanford-TML/toddlerbot returned HTTP 404 from this checkout. No browser 3D assembly is published here.

Benchmarks

• R1RLBench — “RLBench: The Robot Learning Benchmark & Learning Environment”. Type: research paper. ID: arXiv:1909.12271. Retrieved: 2026-09-03.arxiv.org/abs/1909.12271Supports: RLBench identity, 100 unique tasks, Franka Panda, V-REP / CoppeliaSim, first public date 2019-09-26.
• R2RLBench — official project site. Type: research project page. Retrieved: 2026-09-03.sites.google.com/view/rlbenchSupports: RLBench public-code row.
• R3CALVIN — “CALVIN: A Benchmark for Language-Conditioned Policy Learning for Long-Horizon Robot Manipulation Tasks”. Type: research paper. ID: arXiv:2112.03227. Retrieved: 2026-09-03.arxiv.org/abs/2112.03227Supports: CALVIN identity, 4 environments, 34 subtasks, ~24 h play, 20K language directives, Franka Panda 7-DOF, PyBullet, first public date 2021-12-06.
• R4CALVIN — official GitHub repository linked from the project page. Type: research code. Retrieved: 2026-09-03.github.com/mees/calvinSupports: CALVIN public-code row. The project site itself is HTTP-only and is not used as the locator.
• R5LIBERO — “LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning”. Type: research paper. ID: arXiv:2306.03310. Retrieved: 2026-09-03.arxiv.org/abs/2306.03310Supports: LIBERO identity, 130 language-conditioned tasks in 4 suites, 50-demonstration training protocol (not a second task count), first public date 2023-06-05. Not a 65k-demo figure from third-party pages.
• R6LIBERO — official project site. Type: research project page. Retrieved: 2026-09-03.libero-project.github.ioSupports: LIBERO public-code row.
• R7ManiSkill2 — “ManiSkill2: A Unified Benchmark for Generalizable Manipulation Skills”. Type: research paper. ID: arXiv:2302.04659. Retrieved: 2026-09-03.arxiv.org/abs/2302.04659Supports: ManiSkill2 identity, 20 task families, 2000+ objects, 4 million+ demonstration frames, stationary or mobile-base / single-arm or dual-arm / rigid or soft-body, SAPIEN, first public date 2023-02-09.
• R8ManiSkill2 — official project site. Type: research project page. Retrieved: 2026-09-03.maniskill2.github.ioSupports: ManiSkill2 public-code row.
• R9Meta-World — “Meta-World: A Benchmark and Evaluation for Multi-Task and Meta Reinforcement Learning”. Type: research paper. ID: arXiv:1910.10897. Retrieved: 2026-09-03.arxiv.org/abs/1910.10897Supports: Meta-World identity, 50 distinct manipulation tasks, Sawyer, MuJoCo, first public date 2019-10-24.
• R10Meta-World — official project site. Type: research project page. Retrieved: 2026-09-03.meta-world.github.ioSupports: Meta-World public-code row.
• R11RoboCasa — “RoboCasa: Large-Scale Simulation of Everyday Tasks for Generalist Robots”. Type: research paper. ID: arXiv:2406.02523. Retrieved: 2026-09-03.arxiv.org/abs/2406.02523Supports: RoboCasa identity, 100 tasks (25 atomic + 75 composite), 120 kitchen scenes, 2,509 objects, 153 categories, 100K trajectories, language goals / LLM composite tasks, first public date 2024-06-04. Abstract 2,500+ / 150+ stay attributed to the abstract; body figures win the scale row.
• R12RoboCasa — official project site. Type: research project page. Retrieved: 2026-09-03.robocasa.aiSupports: RoboCasa public-code and engine/renderer rows.
• R13SIMPLER — “Evaluating Real-World Robot Manipulation Policies in Simulation”. Type: research paper. ID: arXiv:2405.05941. Retrieved: 2026-09-03.arxiv.org/abs/2405.05941Supports: SIMPLER identity for Google Robot and WidowX setups, ~1,500 paired eval episodes across both embodiments, SAPIEN default, first public date 2024-05-09. No policy success rates are copied.
• R14SIMPLER — official project site. Type: research project page. Retrieved: 2026-09-03.simpler-env.github.ioSupports: SIMPLER public-code row.
• R15FurnitureBench — “FurnitureBench: Reproducible Real-World Benchmark for Long-Horizon Complex Manipulation”. Type: research paper. ID: arXiv:2305.12821. Retrieved: 2026-09-03.arxiv.org/abs/2305.12821Supports: FurnitureBench identity, 8 models, 5000+ demonstrations, 200+ hours, first public date 2023-05-22.
• R16BEHAVIOR-1K — “BEHAVIOR-1K: A Benchmark for Embodied AI with 1,000 Everyday Activities and Realistic Simulation”. Type: research paper. ID: arXiv:2403.09227. Retrieved: 2026-09-03.arxiv.org/abs/2403.09227Supports: BEHAVIOR-1K identity, 1000 activities, 50 scenes, 9000+ objects, OmniGibson, first public date 2024-03-14.
• R17BEHAVIOR — official project site. Type: research project page. Retrieved: 2026-09-03.behavior.stanford.eduSupports: BEHAVIOR-1K public-code row.
• R18ALFRED — “ALFRED: A Benchmark for Interpreting Grounded Instructions for Everyday Tasks”. Type: research paper. ID: arXiv:1912.01734. Retrieved: 2026-09-03.arxiv.org/abs/1912.01734Supports: ALFRED identity, 25,743 English directives, 8,055 expert demonstrations, 120 indoor scenes, AI2-THOR simulation, first public date 2019-12-03. Abstract 25k stays attributed to the abstract.
• R19Room-to-Room — “Vision-and-Language Navigation: Interpreting visually-grounded navigation instructions in real environments”. Type: research paper. ID: arXiv:1711.07280. Retrieved: 2026-09-03.arxiv.org/abs/1711.07280Supports: R2R identity, 21,567 instructions, average 29 words, Matterport3D real imagery, first public date 2017-11-20. Not Habitat-Sim.
• R20Room-to-Room — official project site. Type: research project page. Retrieved: 2026-09-03.bringmeaspoon.orgSupports: R2R public-code row.
• R21OpenEQA — official project site for embodied question answering. Type: research project page. Venue: CVPR 2024. Retrieved: 2026-09-03.open-eqa.github.ioSupports: OpenEQA identity, 1600+ questions, 180+ real environments. No arXiv id is invented; arXiv:2404.06600 is a different paper.
• R22HumanoidBench — “HumanoidBench: Simulated Humanoid Benchmark for Whole-Body Locomotion and Manipulation”. Type: research paper. ID: arXiv:2403.10506. Retrieved: 2026-09-03.arxiv.org/abs/2403.10506Supports: HumanoidBench identity, 15 manipulation + 12 locomotion tasks, Unitree H1 with two Shadow Hands as the primary agent, MuJoCo, first public date 2024-03-15. Optional other models are not this column and not Galbot G1. Not HumanTracker.
• R23HumanoidBench — official project site. Type: research project page. Retrieved: 2026-09-03.humanoid-bench.github.ioSupports: HumanoidBench public-code row.
• R24Open X-Embodiment — “Open X-Embodiment: Robotic Learning Datasets and RT-X Models”. Type: research paper. ID: arXiv:2310.08864. Retrieved: 2026-09-03.arxiv.org/abs/2310.08864Supports: OXE identity, 1M+ trajectories, 22 embodiments, 60 datasets, 34 labs, 527 skills, 160266 tasks (body). Abstract uses 22 robots, 21 institutions, 527 skills, 160266 tasks. Both stay attributed. First public date 2023-10-13. Not an evaluation suite.
• R25Open X-Embodiment — official project site. Type: research project page. Retrieved: 2026-09-03.robotics-transformer-x.github.ioSupports: OXE public dataset / distribution row.
• R26DROID — “DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset”. Type: research paper. ID: arXiv:2403.12945. Retrieved: 2026-09-03.arxiv.org/abs/2403.12945Supports: DROID identity, 76k trajectories, 350 hours, 564 scenes, 86 tasks, Franka Panda stack, first public date 2024-03-19. Trajectories and hours are the same corpus in two units. Abstract collector counts are not reconciled here.
• R27DROID — official project site. Type: research project page. Retrieved: 2026-09-03.droid-dataset.github.ioSupports: DROID public dataset row.
• R28BridgeData V2 — “BridgeData V2: A Dataset for Robot Learning at Scale”. Type: research paper. ID: arXiv:2308.12952. Retrieved: 2026-09-03.arxiv.org/abs/2308.12952Supports: BridgeData V2 identity, 60,096 trajectories, 24 environments, first public date 2023-08-24. Not SIMPLER’s WidowX eval column.
• R29BridgeData V2 — official project site. Type: research project page. Retrieved: 2026-09-03.rail-berkeley.github.io/bridgedataSupports: BridgeData V2 public dataset row.
• R30Habitat — “Habitat: A Platform for Embodied AI Research”. Type: research paper. ID: arXiv:1904.01201. Retrieved: 2026-09-03.arxiv.org/abs/1904.01201Supports: Habitat-Sim platform card — several thousand fps single-threaded and over 10,000 fps multi-process on one GPU on a Matterport3D scene. Not the Room-to-Room instruction count. First public date 2019-04-02.
• R31Habitat — official project site. Type: research project page. Retrieved: 2026-09-03.aihabitat.orgSupports: Habitat-Sim platform identity and project link.
• R32VLABench — “VLABench: A Large-Scale Benchmark for Language-Conditioned Robotics Manipulation with Long-Horizon Reasoning Tasks”. Type: research paper. ID: arXiv:2412.18194; ICCV 2025. Retrieved: 2026-09-04.arxiv.org/abs/2412.18194Supports: 100 task categories (60 primitive + 40 composite), 2,164 objects / 163 categories, six capability dimensions, standard 7-DoF Franka Emika Panda with parallel gripper, MuJoCo + dm_control, task-success and Progress Score metrics, first public date 2024-12-24.
• R33VLABench — official project site and linked MIT-licensed repository. Type: research project page. Retrieved: 2026-09-04.vlabench.github.ioSupports: public code, evaluation episodes, primitive-task fine-tuning datasets, ICCV 2025 acceptance, and incomplete standard VLA / VLM leaderboard status.
• R34RoboTwin 2.0 — “RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulation”. Type: research paper. ID: arXiv:2506.18088. Retrieved: 2026-09-04.arxiv.org/abs/2506.18088Supports: 50 dual-arm tasks, 100,000+ trajectories, five embodiments, RoboTwin-OD’s 731 objects / 147 categories, five randomization axes, clean / hard success-rate protocol, and Aloha-AgileX / ARX-X5 / Piper / Franka / UR5 embodiment identities. Method scores are not copied.
• R35RoboTwin 2.0 — official project site with linked code, data, documentation, and leaderboard. Type: research project page. Retrieved: 2026-09-04.robotwin-platform.github.ioSupports: over-50-task project wording, public MIT-licensed code, object / trajectory downloads, and public leaderboard access.
• R36HumanoidBench — official GitHub repository at commit cb1189039151c8aadaaa987b442da54383c87fab (2025-09-18, “Update LICENSE”). Type: research code. Licence: MIT (UC Berkeley RLL). Retrieved: 2026-09-07.github.com/carlosferrazza/humanoid-benchSupports: env-explorer official MJCF fragment identity for the twenty published courses — stair, hurdle, maze, pole, slide, crawl, highbar_simple, cabinet, door, window, room, truck, basketball, spoon, insert_normal, push, package, cube, powerlift, and sit_hard — plus the hinge and slide ranges those fragments declare. Not scores and not a robot specification.
• R37LIBERO — official GitHub repository at commit 8f1084e3132a39270c3a13ebe37270a43ece2a01. Type: research code. Licence: MIT (root LICENSE). Retrieved: 2026-09-08.github.com/Lifelong-Robot-Learning/LIBEROSupports: libero_task_map 130 entries (Spatial 10, Object 10, Goal 10, LIBERO-10 10, LIBERO-90 90), wooden_cabinet.xml / short_cabinet.xml group-1 visual .msh plus group-0 collision boxes, ControlEnv defaults Panda / OSC_POSE / 20 Hz / horizon 1000, and set_init_state calling set_state, sim.forward, check_success, _post_process, and _update_observables. LIBERO-100 is the 90+10 union class, not another mapped hundred. Code MIT does not grant third-party mesh or demonstration rights.
• R38robosuite — official GitHub tag 1.4.0 at commit fbee5844ff5632f5b5698e204ec5357ca50be0df. Type: research code. Licence: MIT (ARISE). Retrieved: 2026-09-08.github.com/ARISE-Initiative/robosuiteSupports: pinned LIBERO requirement robosuite==1.4.0 and the official Panda / PandaGripper / RethinkMount assets in the native ControlEnv composed Spatial workcell. Not a current-robosuite API substitute. Python-side arena/robot/object merge is not a standalone URDF.
• R39LIBERO — root MIT LICENSE at the same pin as R37. Type: licence notice. Retrieved: 2026-09-08.github.com/Lifelong-Robot-Learning/LIBERO LICENSESupports: the only licence-named file in the pinned tree. It covers the benchmark code and authored XML, not automatically every third-party object mesh or external demonstration archive.
• R40LIBERO-datasets — Hugging Face dataset yifengzhu-hf/LIBERO-datasets at revision f13aa24a3da8c43c7225569f28c562979fa0e35a. Type: demonstration archive. Licence: Apache-2.0 (dataset card). Retrieved: 2026-09-08.huggingface.co/datasets/yifengzhu-hf/LIBERO-datasetsSupports: official Spatial wooden-cabinet demonstration file identity — 748,271,833 bytes, LFS sha256 7ff1ee60447875c152d8c7819f499d23dbc0eb4d4bf045225d5d87a2cdad99bd, git oid ce6d89257f7ea3a0c4df8a2b1f262f03ae1f62cd. Filename matches the mapped Spatial task. This site imports official demo_0 by restoring the embedded model_file (99,547 bytes, sha256 e9fbc436272154f12aef7fd1d6aa9dc0ffb7d326fe2c621007d777f691aaaba6) through ControlEnv.reset_from_xml_string, then set_init_state, as state replay of one human demonstration; that is not a suite score and not ControlEnv.step physics evaluation. Source seed is unpublished. Code MIT is not this grant.
• R41Meta-World — official Farama-Foundation/Metaworld tree at commit 6e01ad7e2ffb2302e4dca04f796fcd8837df8540. Type: research code. Licence: MIT (root LICENSE). Retrieved: 2026-09-08.github.com/Farama-Foundation/MetaworldSupports: ENV_CLS_MAP 50 unique v3 task ids, declared package 3.1.1, MuJoCo==3.3.0, Gymnasium>=1.1, SawyerXYZEnv mocap XYZ action Box[-1,1]^4 scaled 0.01, base step terminated=False with TimeLimit 500, and the official sawyer_reach_v3.xml include graph. Git tag v3.1.1 is 7ea2b501c4a698c8533cdc55a396fe2734e2649d, not this commit; this pin is not treated as a released build. Root MIT covers authored code/XML, not automatically mesh or texture attribution.
• R42Meta-World — Farama documentation, basic usage. Type: official documentation. Retrieved: 2026-09-08.metaworld.farama.org/introduction/basic_usageSupports: gym.make Meta-World/MT1 construction, current v3 task names, and the Gymnasium five-tuple step contract. A docs example that swaps terminated/truncated names is not the pinned SawyerXYZEnv.step return order.
• R43Meta-World — Farama documentation, expert / scripted trajectories. Type: official documentation. Retrieved: 2026-09-08.metaworld.farama.org/benchmark/expert_trajectoriesSupports: upstream SawyerReachV3Policy as a scripted-policy demonstration route, not a learned SOTA checkpoint and not a published MT/ML score.
• R44DROID — official gresearch robotics/droid_100/1.0.0 TFRecord shard 00000. Type: dataset object. Licence: CC-BY-4.0 (dataset). Retrieved: 2026-09-09.storage.googleapis.com/gresearch/robotics/droid_100/1.0.0Supports: object generation 1710528095074666, 27,488,816 bytes, sha256 37d04d3586fbde386d14122320d8f7d866b37007ff7a8d96625e72a3994dda70. Two framed records at offset 0 / 14,343,905 payload bytes and offset 14,343,921 / 13,144,879 payload bytes. Dataset CC-BY-4.0 does not automatically license unrelated assets.
• R45DROID — official gresearch robotics/droid_raw/1.0.1 RAIL and RPL metadata, HDF5, MP4, and sidecar timestamps for the two selected episode identities. Type: raw recording objects. Licence: CC-BY-4.0 (dataset). Retrieved: 2026-09-09.storage.googleapis.com/gresearch/robotics/droid_raw/1.0.1Supports: exact episode identities RAIL+80edfcb1+2023-04-17-14h-48m-05s and RPL+cb4f6842+2023-06-05-20h-15m-38s; raw metadata success=true; selected 166/238 RLDS samples versus raw 167/239; selected elapsed 11.742 s / 17.05 s versus raw 11.812 s / 17.122 s; three non-stereo MP4 streams each; RPL sidecars only. Generation and sha256 pins stay on the committed episode manifests.
• R46DROID — recorder source pin droid-dataset/droid at commit 33ae6a67274f36d2e29525b86f23a56616ef43a7. Type: research code. Retrieved: 2026-09-09.github.com/droid-dataset/droidSupports: public recorder/code identity for the DROID stack. This pin is not an exact historical build of the 2023 collection software. Dataset metadata reports r2d2_version 1.1 separately from this commit.
• R47DROID — KarlP/droid improved cam2base_extrinsics.json at commit bcb840c3b496533e0adf548a54b51f2f00057837. Type: third-party calibration table. Retrieved: 2026-09-09.huggingface.co/KarlP/droidSupports: neither selected episode UUID is present in that table. Registered 3D overlay and known camera extrinsics stay unpublished for these two recordings. Raw metadata extrinsics are unvalidated and are not applied.
• R48RoboCasa — official robocasa/robocasa tree at commit 4f8a2980def75a55dff96b990745b83540425f09. Type: research code. Licence: MIT (root LICENSE / README code grant). Retrieved: 2026-09-10.github.com/robocasa/robocasaSupports: declared package 1.0.1, git describe v1.0-29-g4f8a298 with no release tag on this SHA, setup.py pins MuJoCo==3.3.1 / NumPy==2.2.5 / Numba==0.61.2 / SciPy==1.15.3 / LeRobot==0.3.3 plus gymnasium, ATOMIC_TASK_DATASETS 65 keyword keys and COMPOSITE_TASK_DATASETS 252 (317 dataset-task entries), KitchenEnvMeta 374 registered classes after excluding MG_Robocasa_Env / PickPlace / ManipulateDoor / OpenDoor / CloseDoor, 60 positive layout ids and 60 positive style ids with negatives as groups (test −1 → 1–10, train −2 → 11–60), download --type groups tex / tex_generative / fixtures_lw / objs_objaverse / objs_aigen / objs_lw, and official Box URLs in box_links_assets.json. README separates MIT code from CC BY 4.0 assets/datasets and advertises a ~10GB full asset set. Official Box fixture/texture/object URLs returned HTTP 404 from this checkout on 2026-09-10. 317 dataset keys are not 365 verified previews; 60×60 is not a verified scene product. Paper 100/120 identity stays on R11/R12.
• R49RoboCasa — official installation documentation. Type: official documentation. Retrieved: 2026-09-10.robocasa.ai/docs/build/html/introduction/installation.htmlSupports: install from the robocasa repository and the instruction to use the robosuite master branch rather than a compatibility-pinned 1.4.0 wheel. Isolated first-kitchen runtime uses robosuite master 5ce6643f3092639d08f7b0f90ed1c6a84f50552c (declared package 1.5.2; git tag v1.5.2 is 824ac14c…, not this commit) with MuJoCo 3.3.1. That is not LIBERO’s robosuite 1.4.0 and not Meta-World MuJoCo 3.3.0.
• R50RoboCasa — official overview documentation. Type: official documentation. Retrieved: 2026-09-10.robocasa.ai/docs/build/html/introduction/overview.htmlSupports: current official docs advertise RoboCasa365 / 365 tasks. That branding is not the paper’s 100 tasks / 120 scenes (R11) and is not a verified working-preview or explorer-card count. This slice catalogues 317 dataset-task entries and assembles two layout-1 / style-1 task entries on one kitchen.
• M1RoboCup Federation — official organisation site. Type: official organisation page. Retrieved: 2026-09-03.robocup.orgSupports: RoboCup competition identity and ongoing-status card. Not scores or payloads.
• M2DARPA — Robotics Challenge programme page. Type: official programme page. Retrieved: 2026-09-03.darpa.mil/research/programs/darpa-robotics-challengeSupports: DARPA Robotics Challenge historical identity; the programme page states it is complete. Not scores or payloads.
• M3XPRIZE — ANA Avatar XPRIZE page. Type: official prize page. Retrieved: 2026-09-03.xprize.org/prizes/avatarSupports: ANA Avatar XPRIZE identity card. Not scores or payloads.
• M4CYBATHLON — ETH Zurich official championship site. Type: official organisation page. Retrieved: 2026-09-03.cybathlon.ethz.chSupports: CYBATHLON identity card. Not scores or payloads.
• M5EUROBENCH — official project site. Type: official organisation page. Retrieved: 2026-09-03.eurobench2020.euSupports: EUROBENCH identity as an EU benchmarking framework for biped systems. Organisation page, not a spec table.

RoboTwin native recording

• R1RoboTwin — official code at commit 96c1feab536306b50c26af200044fcdf126e8904. Type: research code. Licence: MIT. Retrieved: 2026-09-11.github.com/RoboTwin-Platform/RoboTwinSupports: handover_block, official expert planning and stored-path replay, native check_success, ALOHA AgileX configuration, and the separate take_action evaluation limit. The locally recorded native audit and execution protocol are site evidence, not an upstream leaderboard or learned-policy result.
• R2RoboTwin Team / Tianxing Chen — official asset distribution at revision 785feb15aa4a4f532395ad2b1d2be5f28cb561ad. Type: source asset distribution. Licence: separately declared MIT. Retrieved: 2026-09-11.huggingface.co/datasets/TianxingChen/RoboTwin2.0Supports: selected ALOHA AgileX embodiment assets. Archive/member hashes and asset grant · attribution and permission notice. Published GLBs preserve native visual buffers through reversible partitioning; they are not vendor engineering models. Per-component rights chains were not independently verified.

Announced / no-sheet

• P1Galbot — ET1 / 银河星仔 launch post. Type: official announcement. Date: 2026-08-19. Retrieved: 2026-09-03.weibo.comSupports: ET1 identity and demonstrated functions. Hardware numbers come from the 2026-09-03 product page, not from this launch post.
• P2China Daily — ET1 launch and AstraBrain-Agent overview. Type: press. Date: 2026-08-21. Retrieved: 2026-08-25.chinadaily.com.cnSupports: ET1 identity and AstraBrain-Agent description.
• P3每日经济新闻 (NBD) — ET1 release report. Type: press. Date: 2026-08-19/20. Retrieved: 2026-08-25.nbd.com.cnSupports: ET1 demonstrated functions.
• P4Figure — “Ramping Figure 03 Production”. Type: official announcement. Date: 2026-04-29. Retrieved: 2026-08-25.figure.aiSupports: Figure 03 status card (>350 units, one/hour).
• P5Tesla — Q2 2026 filing (via ir.tesla.com); TechCrunch 2026-07-22. Type: filing / press. Date: 2026-07-22. Retrieved: 2026-08-25. Review after: 2027-01-01. Review reason: The stated later-2026 production window closes; verify production status from a filing or official update.ir.tesla.comSupports: Tesla Optimus status card (Fremont construction; production later 2026).
• P6Boston Dynamics — “Unveils New Atlas…”. Type: official announcement. Date: 2026-01-05. Retrieved: 2026-08-25.bostondynamics.comSupports: electric Atlas status card.
• P7Apptronik — “Welcome to Robot Park…” (and $935M Series A, 2026-02). Type: official announcement. Date: 2026-06-30. Retrieved: 2026-08-25.apptronik.comSupports: Apollo 2 status card.
• P8Genesis AI — “Introducing Eno…” PR Newswire. Type: official announcement / press. Date: 2026-06-16. Retrieved: 2026-08-29. Review after: 2027-01-01. Review reason: The targeted end-2026 industrial-deployment window closes; check for an official specification or deployment report.prnewswire.comSupports: Eno identity — first general-purpose robot, wheeled lifting-tower architecture, waitlist, targeted industrial deployments toward end-2026. Official PR does not publish whole-body DoF, height, mass, or 3–5 kg/arm.
• P9Business Insider — Genesis AI unveils a robot with no head or legs. Type: press. Date: 2026-06. Retrieved: 2026-08-29.businessinsider.comSupports: Eno name = “one” backward; non-biped form note. Not a specification source.
• P10Genesis AI — GENE-26.5 press release. Type: official announcement. Date: 2026-05-06. Retrieved: 2026-08-29.genesis.ai/press/press-release-gene-265Supports: Eno architecture card and GENE-26.5 demo list (meal prep, pipetting, wire harnessing, Rubik’s cube, piano). No parameter count, API, or benchmark table.
• P11TechCrunch — Genesis AI launches with $105M seed from Eclipse and Khosla. Type: press. Date: 2025-07-01. Retrieved: 2026-08-29.techcrunch.comSupports: Eno architecture card — lab identity and seed context. Same article is foundation-models-p1 for the policy page.
• P12Forbes — John Koetsier private-preview write-up of Eno. Type: press. Date: 2026-06-16. Retrieved: 2026-08-29. Access: bot-blocked (HTTP 403 on 2026-08-29 and again on 2026-10-09, from two hosts); the direct article URL remains the citation. Review after: 2027-01-31. Review reason: Re-probe the direct article after the observed HTTP 403 while keeping access state separate from review state.forbes.comSupports: press-only Eno numbers that stay out of tables — ~20 actuated hand DoF, unequal finger lengths, ~3–5 kg per arm, ~4–6 h runtime. Hand DoF is not robot DoF. Not used as a sheet.
• P131X Technologies — NEO home robot official launch / Early Access page. Type: official announcement / marketing. Date: 2025-10-28. Retrieved: 2026-08-29. Review after: 2027-01-01. Review reason: The announced US 2026 delivery window closes; verify delivery status and reconcile the conflicting hand figures.1x.tech/discover/neo-home-robotSupports: NEO status card — $20,000 or $499/month Early Access; US 2026 / other markets from 2027. Marketing also prints mass/lift/hand-DoF figures that conflict with a later 1X post; those stay unpublished as a sheet.
• P14Agility Robotics — official press-release hub. Type: press / company newsroom. Retrieved: 2026-08-29.agilityrobotics.com/press-releasesSupports: Digit status card as a commercial humanoid with press and RaaS language. Investor-deck pricing is not copied as a list price. Same hub is market-p2 for funding.
• P15XPeng — Robotics funding / IRON mass-production announcement. Type: official pressroom. Retrieved: 2026-08-29. Review after: 2027-01-01. Review reason: The end-2026 mass-production target closes; verify production and the stated 2027 delivery sequence.xpeng.com/pressroomSupports: IRON status card — mass production targeted end-2026, store/campus then 2027 deliveries. Aggregator height / mass / TOPS figures are not copied.
• P16Kepler Robotics — Forerunner K2 GITEX debut, PR Newswire. Type: official announcement / press. Date: 2024-10-21. Retrieved: 2026-08-29.prnewswire.comSupports: Kepler K2 status card — industrial humanoid entering customer testing. Conflicting height/mass/DoF/price figures across marketing pages are not copied into a column.
• P17Kepler Robotics — official site (Forerunner series). Type: official marketing. Retrieved: 2026-08-29.gotokepler.comSupports: Kepler Forerunner series identity on the K2 card. Not used as a quantitative sheet.
• P18Sanctuary AI — “Releases New Generation of AI Robots for High Quality Data Capture”. Type: official announcement. Date: December 2024. Retrieved: 2026-08-29.sanctuary.aiSupports: Phoenix Generation 8 status card — wheeled base as a stated design decision; data-capture and manufacturing-cost claims. No height, mass, payload, or hand-DoF figures are copied from aggregators.
• P19Astribot — company homepage (星尘智能 S1). Type: official marketing. Retrieved: 2026-08-29.astribot.comSupports: Astribot S1 status card — company identity and Design-for-AI (DFAI) claim. The English /s1 product URL returned HTTP 404 from this checkout. No height, mass, DoF, payload, or price is copied. Not Galbot S1.
• P20Robotera — STAR1 English product URL. Type: official marketing / SPA. Retrieved: 2026-08-29.robotera.com/en/products/star1Supports: Robotera STAR1 status card as identity only. Live page is a Chinese SPA with no extractable specification table. Aggregator numbers are not copied. XBot-L is a separate no-sheet name.
• P21MagicLab — company homepage. Type: official marketing / SPA. Retrieved: 2026-08-29.magiclab.topSupports: MagicLab status card as identity only. Live page is a Chinese SPA with no extractable product table.
• P22Noetix Robotics — company homepage. Type: official marketing / SPA. Retrieved: 2026-08-29.noetixrobotics.comSupports: Noetix status card as identity only. Live page is a Chinese SPA with no extractable product table.
• M1Galbot — official product site. Type: official product catalogue / SPA. Retrieved: 2026-09-03. Review after: 2026-12-02. Review reason: Mutable vendor product SPA; re-inspect the live table 90 days after retrieval.galbot.comSupports: ET1 now appears in live product navigation beside G1 and S1. Hardware numbers come from the dedicated /et1 specification page (humanoids-m21), not from this homepage.
• M2Galbot — developer-site sitemap. Type: official developer documentation index. Retrieved: 2026-10-09.developer.galbot.com/sitemap.xmlSupports: ET1 public-developer boundary after the product table appeared. The 573-URL sitemap, last-modified 2026-09-24, covers G1, S1, TM01, and an SDK that now adds a G3 target beside G1 and S1; it exposes no ET1, 银河星仔 / Xingzai, or AstraBrain manual, SDK target, or description entry.
• M3GalaxyGeneralRobotics — official public GitHub organization. Type: official code / robot-description index. Retrieved: 2026-10-09.github.com/GalaxyGeneralRoboticsSupports: ET1 public-repository boundary after the product table appeared. The 17 public repositories still expose no ET1 or Xingzai-named SDK or robot description. GalbotSDK names G1/S1/G3; Humanoid-GPT and LATENT target Unitree G1 and do not establish ET1 support.

Industrial arms

• A1IFR — World Robotics 2026 Industrial Robots, Sources & Methods. Type: analyst / classification. Date: 2026. Retrieved: 2026-10-09. Review after: 2027-09-01. Review reason: Recheck against World Robotics 2027 sources and methods.ifr.orgSupports: industrial-robot class definition and kinematic types (§1.7.3, report p. 34; robot types §1.9, p. 36).
• M4FANUC America — SCARA robots (SR-series). Type: manual / product page. Retrieved: 2026-08-25.fanucamerica.comSupports: SCARA family table (SR-3iA / SR-6iA / SR-12iA).
• M2ABB — GoFa collaborative robots page. Type: manual / product page. Retrieved: 2026-08-25.abb.comSupports: GoFa 12 collaborative column.
• M3FANUC — Collaborative Robot CRX series page and catalog. Type: manual / product page. Retrieved: 2026-08-25.fanuc.co.jpSupports: CRX-10iA collaborative column.
• M1Universal Robots — UR20 Technical Specification (tech sheet). Type: manual. Retrieved: 2026-08-25.universal-robots.comSupports: UR20 collaborative column.
• M5FANUC America — M-2iA / M-3iA delta data sheets. Type: manual / data sheet. Retrieved: 2026-08-25.fanucamerica.comSupports: delta family table (M-2iA / M-3iA).
• M6ABB — Product specification CRB 15000 (GoFa), 3HAC077390-001 Rev. X. Type: manual / product specification PDF. Retrieved: 2026-08-27.search.abb.comSupports: GoFa 12 (CRB 15000-12/1.27) mass, mounting, protection, environment, power, torque, accuracy.
• M7FANUC — Robot CRX series brochure, RCRALL(E)-08. Type: manual / catalogue PDF. Retrieved: 2026-08-27.fanuc.co.jpSupports: CRX-10iA mass, protection, environment, wrist moment and inertia, controller, rated voltage.
• M8FANUC — Robot SR series brochure, RSCARA(E)-10. Type: manual / catalogue PDF. Retrieved: 2026-08-27.fanuc.co.jpSupports: SCARA per-axis motion range and speed, J4 load inertia, push-down force, user cables and air, environment, controller.
• M9FANUC — Delta robot brochure (M-2iA / M-3iA / DR-3iB), DELTA(E)-01. Type: manual / catalogue PDF. Retrieved: 2026-08-27.fanuc.co.jpSupports: delta work envelope, J4 speed, mounting, mass, drive, environment. M-2iA/6H is not in this brochure.
• M10Franka Robotics — Franka Research 3 product page. Type: manual / product page. Retrieved: 2026-08-29.franka.de/franka-research-3Supports: FR3 research-arm table — 7 axes, 3 kg payload, 855 mm reach, <±0.1 mm position repeatability, 7 integrated torque sensors, 1 kHz motion control, 94.5% workspace efficiency. A datasheet download control exists; no public PDF URL was extracted. Not mixed into the 6-axis collaborative table. Not the older Panda name.

Mobile robots

• A1IFR — World Robotics 2026 Service Robots (AP5 logistics). Type: analyst / classification. Date: 2026. Retrieved: 2026-10-09. Review after: 2027-09-01. Review reason: Recheck against World Robotics 2027 sources and methods.ifr.orgSupports: AP5 logistics class definition (Table 1.3, report p. 28).
• M1MiR — MiR250 specifications. Type: manual / product page. Retrieved: 2026-08-25.mobile-industrial-robots.comSupports: MiR250 column.
• M2MiR — MiR600 specifications. Type: manual / product page. Retrieved: 2026-08-25.mobile-industrial-robots.comSupports: MiR600 column.
• M3OTTO by Rockwell Automation — OTTO 600 spec sheet (OTTO-DS001B-EN_EU-MAR2025), linked from the data-sheets page. Type: manual / data sheet. Retrieved: 2026-08-25.ottomotors.comSupports: OTTO 600 column.
• M4Locus Robotics — Locus Vector datasheet. Type: manual / data sheet. Date: 2026. Retrieved: 2026-08-25.locusrobotics.comSupports: Locus Vector column.
• M5Locus Robotics — Locus Origin datasheet. Type: manual / data sheet. Date: 2026. Retrieved: 2026-08-25.locusrobotics.comSupports: Locus Origin in-aisle sheet.

Professional service

• A1IFR — World Robotics 2026 Service Robots (AP classes). Type: analyst / classification. Date: 2026. Retrieved: 2026-10-09. Review after: 2027-09-01. Review reason: Recheck against World Robotics 2027 sources and methods.ifr.orgSupports: AP1–AP8 class cards (Table 1.3, report p. 28); food delivery grouped in AP5, not AP8 hospitality (p. 29).
• M1Lely — Lely Vector feeding product page. Type: manual / product page. Retrieved: 2026-08-25.lely.comSupports: feeding sheet (accuracy, slope, power).
• M2Kärcher — KIRA B 50 scrubber dryer page. Type: manual / product page. Retrieved: 2026-08-25.kaercher.comSupports: cleaning sheet.
• M3Pudu Robotics — BellaBot product introduction (Academy). Type: vendor manual / academy spec table. Last modified: 2026-05-06. Retrieved: 2026-08-27.academy.pudutech.comSupports: BellaBot AP8 sheet (565 × 537 × 1290 mm, 55 kg, 4 trays, 10 kg/layer, 0.5–1.2 m/s, 12–24 h, IP20). Tray kg is not an arm payload. The marketing product URL returned no parseable numeric table from this checkout.

Medical robots

• A1IFR — World Robotics 2026 Service Robots (AP6 medical). Type: analyst / classification. Date: 2026. Retrieved: 2026-10-09. Review after: 2027-09-01. Review reason: Recheck against World Robotics 2027 sources and methods.ifr.orgSupports: AP61–AP64, AP69 class cards (Table 1.4, report p. 29).
• M1Intuitive — da Vinci X/Xi Instrument & Accessory Catalogue. Type: manual / catalog. Date: 2024. Retrieved: 2026-08-25.intuitive.comSupports: da Vinci X instrument diameter, EndoWrist DoF, articulation.
• M2Intuitive — official system manuals (systems_i_a). Type: manual. Retrieved: 2026-08-25.manuals.intuitive.comSupports: system configuration, environment, regulatory/status statements.

Consumer robots

• A1IFR — World Robotics 2026 Service Robots (AC classes). Type: analyst / classification. Date: 2026. Retrieved: 2026-10-09. Review after: 2027-09-01. Review reason: Recheck against World Robotics 2027 sources and methods.ifr.orgSupports: AC1–AC3 class cards (Table 1.2, report p. 26).
• M1Roborock — Qrevo Edge product page (US store). Type: manual / product page. Retrieved: 2026-08-25.us.roborock.comSupports: floor-cleaning sheet.
• M2Pollen Robotics — Reachy Mini product page. Type: manual / product page. Retrieved: 2026-08-29.pollen-robotics.com/reachy-miniSupports: Reachy Mini identity, Lite $399 / Wireless $499, Lite host-over-USB compute, Wireless Raspberry Pi Compute Module 4, Python SDK named without a version pin. Apache 2.0 was not extracted from this page. Not Reachy 2.
• P1Hugging Face + Pollen Robotics — “Reachy Mini – The Open-Source Robot for Today's and Tomorrow's AI Builders”. Type: official announcement / blog. Date: 2025-07-09. Retrieved: 2026-08-29.huggingface.co/blog/reachy-miniSupports: Reachy Mini dimensional table — 11 in / 28 cm awake height stored as 280 mm, 6.3 in / 16 cm width as 160 mm, ~9 in / 23 cm sleep as 230 mm, 3.3 lb / 1.5 kg, 6 head DoF, 4 mics, 5 W speaker, wide-angle camera. The later product page wins for wireless compute (CM4 vs this post’s Raspberry Pi 4).

Aerial robots

• A1IFR — World Robotics 2026 Service Robots (type C aerial). Type: analyst / classification. Date: 2026. Retrieved: 2026-10-09. Review after: 2027-09-01. Review reason: Recheck against World Robotics 2027 sources and methods.ifr.orgSupports: aerial class definition (Table 1.1, report p. 25); drones generally left out of the statistics (p. 23).
• M1DJI — Matrice 400 support / technical data. Type: manual / product page. Retrieved: 2026-08-25.dji.comSupports: Matrice 400 column.
• M2DJI — Matrice 350 RTK specifications. Type: manual / product page. Retrieved: 2026-08-25.enterprise.dji.comSupports: Matrice 350 RTK column.
• R1learnsyslab / gym-pybullet-drones contributors — gym-pybullet-drones 2.2.0 PID example with the CF2X model, at 7ebad1ec. Type: research code and native simulation output. License: MIT code; PyBullet zlib; CF2 mesh credits Lakshman Kumar. Retrieved: 2026-10-09.Pinned PID exampleSupports: simulated flight recording. Capture, independent replay and media evidence · Gate, diagnosis, tuning record and attribution. Qualification failed under an authored tracking-accuracy gate: the first trial measured RMSE 0.0894 m against 0.05 m. The 577-sample recording is a second trial, declared after that failure with the reference velocity supplied to the unchanged controller; it met the same gate (0.0110 m) and is not counted as a pass. Zero qualified tasks, additional browser assemblies or reproduced suite scores.

Software, SDK, and simulation

• M1Galbot — Developer Platform (applicable devices and SDK listing). Type: official developer documentation. Retrieved: 2026-08-27.developer.galbot.com/platformSupports: G1 docs v2.2.4, S1 docs v2.0.0, TM01 docs v1.4.0, SDK v1.9.1, and G1/S1 description downloads (URDF / MJCF / USD). Platform Chinese copy currently labels G1 “通用人形机器人”; the landscape does not take that as a form-class change.
• M2Galbot — Galbot SDK documentation. Type: official SDK documentation. Version: 1.9.1. Retrieved: 2026-08-27.developer.galbot.com/docs/SDK/1.9.1Supports: galbot_sdk name, Python and C++ API references, GitHub download pointer.
• M3GalaxyGeneralRobotics — GalbotSDK README. Type: official SDK repository. Retrieved: 2026-08-27.github.com/GalaxyGeneralRobotics/GalbotSDKSupports: SDK V1.9.1, release date 2026-07-03, robot environment V1.17, Python 3.8–3.14, Ubuntu 20–24, C++/Python, G1/S1 badge.
• M4Galbot — developer product handbook (二次开发产品手册). Type: official commercial / developer page. Retrieved: 2026-08-27.developer.galbot.comSupports: three-layer product map; SDK capability areas; FREE / PRO (¥50,000 per robot per year) / ENTERPRISE (¥200,000 per robot per year) subscription copy and SLA hours.
• M5GalaxyGeneralRobotics — galbot_one_golf_description README. Type: official robot-description repository. License: Apache-2.0. Retrieved: 2026-08-27.github.com/GalaxyGeneralRobotics/galbot_one_golf_descriptionSupports: G1 URDF / MJCF / USD formats, preset URDF names, MJCF wheeled/fixed/planar variants, USD entry path.
• M6GalaxyGeneralRobotics — galbot_s1_description README. Type: official robot-description repository. License: Apache-2.0. Retrieved: 2026-08-27.github.com/GalaxyGeneralRobotics/galbot_s1_descriptionSupports: S1 URDF / MJCF / USD formats and Apache-2.0 license.
• M7NVIDIA Isaac Lab — Available Environments. Type: official simulator documentation. Retrieved: 2026-08-27.isaac-sim.github.io/IsaacLabSupports: Galbot cube-stack task IDs, Manager Based workflow, and NVIDIA’s “Galbot humanoid robot” wording for the left-arm gripper task.
• M8Google DeepMind — mujoco_menagerie README. Type: official model collection. Retrieved: 2026-08-27.github.com/google-deepmind/mujoco_menagerieSupports: Unitree G1 listed at 29 DoF (BSD-3-Clause), Sharpa Wave listed at 22 DoF (Apache-2.0), Galbot absent from the published list.
• M9GalaxyGeneralRobotics — Humanoid-GPT README. Type: official research-code repository. License: Apache-2.0. Retrieved: 2026-08-27.github.com/GalaxyGeneralRobotics/Humanoid-GPTSupports: Unitree G1 29-DoF target, MuJoCo-MJX, Apache-2.0, real-robot deploy path. GitHub repository description currently says “Official implementation of AstraBrain-WBC 0.5”; treated as a repo blurb, not a checkpoint-identity proof.
• M10GalaxyGeneralRobotics — LATENT README (default branch: master). Type: official research-code repository. Retrieved: 2026-08-27.github.com/GalaxyGeneralRobotics/LATENTSupports: MuJoCo pipeline, G1TrackingGeneral task name, UnitreeG1 motion path, default 50 Hz preprocess frequency.
• R1Qi et al. — Humanoid-GPT: Humanoid Generative Pre-Training for Zero-Shot Motion Tracking. Type: research. Venue: CVPR 2026. Retrieved: 2026-08-27.arxiv.org/abs/2606.03985Supports: paper link on the eval-body code table; real-robot body remains Unitree G1 per the research page.
• R2Zhang et al. — LATENT: Learning Athletic Humanoid Tennis Skills from Imperfect Human Motion Data. Type: research. Venue: IROS 2026. Retrieved: 2026-08-27.arxiv.org/abs/2603.12686Supports: paper link on the eval-body code table; real-robot body remains Unitree G1 per the research page.
• M11Genesis-Embodied-AI — Genesis README (current title: Genesis World). Type: official open-source repository. License: Apache-2.0. Retrieved: 2026-08-29.github.com/Genesis-Embodied-AI/GenesisSupports: Genesis / genesis-world engine table — Apache-2.0, Python >=3.10,<3.14, PyPI genesis-world, Quadrants backends (CUDA / ROCm / Metal / Vulkan / x86 / ARM64), academic since Dec 2024, “now officially supported by Genesis AI.” No MuJoCo / Isaac Gym speedup factor is copied from this README.
• M12Genesis-Embodied-AI — genesis-world documentation and sibling repository. Type: official docs / repository. Retrieved: 2026-08-29.genesis-world.readthedocs.ioSupports: docs / sibling-repo row. Companion GitHub: github.com/Genesis-Embodied-AI/genesis-world. Not Genesis AI’s commercial product blog.
• P1Genesis AI — “The Role of Simulation in Scalable Robotics, Genesis World 1.0, and the Path Forward.” Type: company blog. Date: 2026. Retrieved: 2026-08-29.genesis.ai/blogSupports: Genesis World 1.0 company-product card (Nyx, Quadrants, physics platform). Not a license, parameter count, or third-party benchmark.
• M13Booster Robotics — official open-source page and booster_robotics_sdk. Type: official SDK / developer documentation. Retrieved: 2026-08-29.booster.tech/open-sourceSupports: software third-party SDK table — T1 and K1 named; SDK / ROS 2 surface. Hardware numbers stay on the humanoids T1 table. GitHub: github.com/BoosterRobotics/booster_robotics_sdk.
• M14Fourier Intelligence (FFTAI) — fourier-grx-N1 SDK docs. Type: official SDK documentation. Retrieved: 2026-08-29.fftai.github.io/fourier-grx-N1Supports: Fourier N1 software listing only. Companion repository: github.com/FFTAI/fourier-grx-N1. No N1 hardware column on this landscape. N1 is not GR-2.
• M15Pollen Robotics — Reachy 2 hardware-guide general specifications. Type: vendor hardware / SDK documentation. Retrieved: 2026-08-29.docs.pollen-robotics.comSupports: Reachy 2 column on the software third-party SDK table. Hardware numbers stay on the mobile-manipulators Reachy 2 table.
• M16Google DeepMind — MuJoCo repository. Type: official open-source repository. License: Apache-2.0 (GitHub API 2026-08-29). Retrieved: 2026-08-29.github.com/google-deepmind/mujocoSupports: peer-engine MuJoCo column — C/C++ physics engine with Python bindings, Apache-2.0. No throughput versus genesis-world is copied.
• M17haosulab — SAPIEN repository. Type: official open-source repository. License: GitHub API SPDX empty (NOASSERTION) on 2026-08-29. Retrieved: 2026-08-29.github.com/haosulab/SAPIENSupports: peer-engine SAPIEN column. License cell says see repository rather than inventing Apache-2.0.
• M18mani-skill — ManiSkill repository. Type: official open-source repository. License: Apache-2.0 (GitHub API 2026-08-29). Retrieved: 2026-08-29.github.com/mani-skill/ManiSkillSupports: peer-engine ManiSkill column — Python GPU-parallel simulator and benchmark. Live org is mani-skill, not haosulab/ManiSkill.
• M19RobotLocomotion — Drake repository. Type: official open-source repository. License: GitHub API SPDX empty (NOASSERTION) on 2026-08-29. Retrieved: 2026-08-29.github.com/RobotLocomotion/drakeSupports: peer-engine Drake column — C++ / Python model-based design toolbox. License cell says see repository rather than inventing Apache-2.0.
• M20Booster Robotics — official GitHub repositories. Type: vendor SDK, ROS 2 interface, RL training/deployment, and description repositories. Retrieved: 2026-08-30.github.com/BoosterRoboticsSupports: developer ecosystem map — booster_robotics_sdk, booster_robotics_sdk_ros2, booster_gym / booster_train / booster_deploy, and booster_assets. Licenses are repository-specific; the assets repository is BSD-3-Clause.
• M21LimX Dynamics — official GitHub repositories. Type: vendor low-level SDK, ROS / ROS 2 deploy, MuJoCo simulation, Isaac Lab training, and description repositories. Retrieved: 2026-08-30.github.com/limxdynamicsSupports: developer ecosystem map — limxsdk-lowlevel, humanoid-rl-deploy-ros2, humanoid-mujoco-sim, humanoid-rl-isaaclab, humanoid-description, and tron2-robot-description. Named repositories inspected declare Apache-2.0.
• M22Leju Robotics — kuavo-ros-opensource. Type: vendor ROS and simulation monorepository. Repository-root license: not declared when inspected. Commit inspected: 2be4881. Retrieved: 2026-08-30.github.com/LejuRobotics/kuavo-ros-opensourceSupports: developer ecosystem map — ROS packages, configuration/version selection, MuJoCo, Gazebo, Isaac Sim, and Kuavo description resources in one repository.
• M23Hello Robot — Stretch 4 official repositories. Type: vendor body API, ROS 2, MuJoCo, examples, and description packages. Retrieved: 2026-08-30.github.com/hello-robotSupports: developer ecosystem map — stretch4_body, stretch4_ros2, stretch4_mujoco, stretch4_examples, and stretch4_urdf. Licenses are repository-specific; stretch4_mujoco reports BSD-3-Clause-Clear and examples MIT.
• M24UC Berkeley Hybrid Robotics — Berkeley-Humanoid-Lite and assets. Type: open research hardware, control, and simulation repositories. Code license: MIT; asset license: CC BY-SA 4.0. Retrieved: 2026-08-30.github.com/HybridRobotics/Berkeley-Humanoid-LiteSupports: developer ecosystem map — build documentation, locomotion/control code, and separately licensed URDF / MJCF / USD assets.
• M25Stanford Robotics — ToddlerBot. Type: open research hardware, runtime, policy, and description repository. Code license: MIT; design/assets: CC BY-NC-SA 4.0. Commit inspected: e337f3b. Retrieved: 2026-08-30.github.com/hshi74/toddlerbotSupports: developer ecosystem map — real-robot and simulation code, multiple actuator/end-effector assemblies, URDF / MJCF, documentation, and Onshape / printable-design link.
• M26NVIDIA — Isaac GR00T developer platform. Type: official developer documentation. Version: GR00T 1.7 Early Access. Retrieved: 2026-09-04.developer.nvidia.com/isaac/gr00tSupports: open reference-platform components; multimodal input and action-chunk output; real, synthetic, and internet-video data mix; post-training for specific embodiments; weights/reference-code access; Early Access excludes production deployment with commercial support.
• M27Google DeepMind — Gemini Robotics On-Device 2 model card. Type: official model card. Publication date: 2026-07-30. Retrieved: 2026-09-04.deepmind.google/models/model-cards/gemini-robotics-on-device-2Supports: on-device VLA topology; text, image, and proprioception inputs; numerical robot-action output; standing bi-arm manipulation scope; Trusted Tester distribution; high-DoF, mobile-platform, and whole-body limitations.
• R3Cadene et al. — “LeRobot: An Open-Source Library for End-to-End Robot Learning”. Type: research paper. ID: arXiv:2602.22818. Publication date: 2026-02-26. Retrieved: 2026-09-04.arxiv.org/abs/2602.22818Supports: end-to-end open library; unified robot middleware and LeRobotDataset schema; asynchronous inference that separates action planning from control execution; remote dataset streaming; 16K+ datasets and 2.2K+ contributors as of September 2025.
• M28VDA and VDMA — VDA 5050 version 3.0 release. Type: official standard release. Version: 3.0. Publication date: 2026-04-20. Retrieved: 2026-09-04.vda.de — VDA 5050 version 3.0Supports: cross-manufacturer mobile-robot data exchange; mixed-fleet master-control scope; free-navigation zones, path sharing, local route planning, standardized power-saving actions, and retained predefined-trajectory concepts.
• M29Open-RMF — official project site and governance statement. Type: official project documentation. Retrieved: 2026-09-04.open-rmf.orgSupports: open modular interoperability across multiple robot fleets and physical infrastructure including doors, elevators, and building-management systems; OSRA governance since 2024.

Robot foundation models

• P1TechCrunch — Genesis AI launches with $105M seed from Eclipse and Khosla. Type: press. Date: 2025-07-01. Retrieved: 2026-08-29.techcrunch.comSupports: Genesis AI stealth card — $105M seed co-led by Eclipse and Khosla; Bpifrance, HSG, Eric Schmidt, Xavier Niel participating; CEO Zhou Xian; co-founder Théophile Gervet; universal robotics foundation model + horizontal platform.
• P2Genesis AI — GENE-26.5 press release. Type: official announcement. Date: 2026-05-06. Retrieved: 2026-08-29.genesis.ai/press/press-release-gene-265Supports: GENE-26.5 identity, glove/hand claims, demo list, and the company-VLA column. No parameter count, API, weights, or benchmark table.
• P3Genesis AI — GENE-26.5 blog. Type: company blog. Date: 2026-05. Retrieved: 2026-08-29.genesis.ai/blog/gene-26-5-advancing-robotic-manipulation-to-human-levelSupports: Genesis Hand 1.0 “20 active, back-drivable degrees of freedom” as a hand claim, not a robot-body DoF. 1:1:1 glove mapping and efficiency claims stay press-graded.
• R1NVIDIA — GR00T N1: An Open Foundation Model for Generalist Humanoid Robots. Type: research paper. ID: arXiv:2503.14734. Retrieved: 2026-08-29.arxiv.org/abs/2503.14734Supports: GR00T N1-2B column — 2.2B total / 1.34B VLM, 16-action chunk 63.9 ms on L40 bf16, Fourier GR-1 collect, 1X thanked. This column is the N1-2B paper checkpoint, not later N1.6 / N1.7 labels.
• M1NVIDIA — Isaac-GR00T repository. Type: official code / weights. License: Apache-2.0 code; NVIDIA Open Model License named for N1-era weights. Retrieved: 2026-08-29.github.com/NVIDIA/Isaac-GR00TSupports: GR00T code-license rows (Apache-2.0). N1-2B paper-checkpoint column. Later N1.6 / N1.7 version facts use M10–M13. Rechecked 2026-09-08: current default branch is N1.7.
• R2Kim et al. — OpenVLA: An Open-Source Vision-Language-Action Model. Type: research paper. ID: arXiv:2406.09246. Venue: CoRL 2025. Retrieved: 2026-08-29.arxiv.org/abs/2406.09246Supports: OpenVLA 7B column — 7B parameters, 970k Open X-Embodiment episodes, Llama 2 + SigLIP/DINOv2. A Hugging Face “MIT” card does not override the Llama 2 Community License for inherited weights.
• M2OpenVLA — project page and Hugging Face openvla/openvla-7b. Type: official project / weights. Retrieved: 2026-08-29.openvla.github.ioSupports: OpenVLA weights and code-publication rows. Hugging Face: huggingface.co/openvla/openvla-7b.
• R3Physical Intelligence — π0.5: a Vision-Language-Action Model with Open-World Generalization. Type: research paper. ID: arXiv:2504.16054. Venue: CoRL 2025 Oral. Retrieved: 2026-08-29.arxiv.org/abs/2504.16054Supports: π0.5 column — 2B VLM (PaliGemma) + 300M action expert. Experiment results only; not a product API.
• M3Physical Intelligence — openpi repository. Type: official code / checkpoints. License: Apache-2.0. Retrieved: 2026-08-29.github.com/Physical-Intelligence/openpiSupports: π0.5 code-license and checkpoint rows.
• P4Skild AI — company site and S1 blog. Type: company / press. Date: S1 post August 2026. Retrieved: 2026-08-29.skild.aiSupports: Skild S1 column as an in-context “omni-bodied” company model with no public parameter count, weights, or eval table. S1 post: skild.ai/blogs/s1.
• P5Figure — Helix announcement. Type: company post. Date: 2025-02-20. Retrieved: 2026-08-29.figure.ai/news/helixSupports: Helix column — S2 7B VLM, S1 80M, ~500 h teleop, S2 7–9 Hz / S1 200 Hz. Company post, not a paper. Helix 02 is a later post and is not collapsed into this column.
• R4Octo Model Team — Octo: An Open-Source Generalist Robot Policy. Type: research paper. ID: arXiv:2405.12213. Retrieved: 2026-08-29.arxiv.org/abs/2405.12213Supports: prior-policy Octo column — 27M and 93M checkpoints, 800k Open X-Embodiment trajectories.
• M4Octo — official project page. Type: official project / weights. Retrieved: 2026-08-29.octo-models.github.ioSupports: Octo public-checkpoint row.
• R5Brohan et al. — RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control. Type: research paper. ID: arXiv:2307.15818. Retrieved: 2026-08-29.arxiv.org/abs/2307.15818Supports: prior-policy RT-2 column. Weights were not released with the paper.
• P6Physical Intelligence — π0 blog. Type: company blog. Date: 2024. Retrieved: 2026-08-29.pi.website/blog/pi0Supports: prior-policy π0 column as a 2024 blog, not a paper. Parameter split is cited from the π0.5 paper, not invented from this post.
• R6Liu et al. — RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation. Type: research paper. ID: arXiv:2410.07864. Retrieved: 2026-08-29.arxiv.org/abs/2410.07864Supports: prior-policy RDT-1B column — 1B-parameter diffusion Transformer, 1M+ multi-robot episodes, 170M ablation checkpoint named in the repository. Tsinghua TSAIL; not a Galbot or Genesis product model.
• M5RDT — thu-ml/RoboticsDiffusionTransformer and Hugging Face robotics-diffusion-transformer/rdt-1b. Type: official code / weights. License: MIT. Retrieved: 2026-08-29.github.com/thu-ml/RoboticsDiffusionTransformerSupports: RDT-1B weights and MIT license row. Hugging Face: huggingface.co/robotics-diffusion-transformer/rdt-1b. Project: rdt-robotics.github.io/rdt-robotics.
• R7Cadene et al. — SmolVLA: Efficient Vision-Language-Action Models. Type: research paper. ID: arXiv:2506.01844. Retrieved: 2026-08-29.arxiv.org/abs/2506.01844Supports: SmolVLA open-peer column — less than 0.5 billion parameters, fewer than 30k community episodes, SmolVLM-2 plus flow-matching action expert, single-GPU train / consumer-GPU or CPU deploy.
• M6LeRobot — Hugging Face lerobot/smolvla_base. Type: official weights card. Retrieved: 2026-08-29.huggingface.co/lerobot/smolvla_baseSupports: SmolVLA 0.5B card figure and public-weight row.
• R8Qu et al. — SpatialVLA: Exploring Spatial Representations for Visual-Language-Action Models. Type: research paper. ID: arXiv:2501.15830. Retrieved: 2026-08-29.arxiv.org/abs/2501.15830Supports: SpatialVLA open-peer column — 3.5B, Paligemma2, Ego3D + Adaptive Action Grids, 1.1M real-robot episodes (OXE subset + RH20T).
• M7SpatialVLA — official project page. Type: official project. Retrieved: 2026-08-29.spatialvla.github.ioSupports: SpatialVLA identity on the open-peer caption. No public weight dump is copied here.
• R9Kim et al. — Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success. Type: research paper. ID: arXiv:2502.19645. Retrieved: 2026-08-29.arxiv.org/abs/2502.19645Supports: OpenVLA-OFT open-peer column — fine-tuning recipe on OpenVLA 7B, not a new pretrain; LIBERO 76.5% → 97.1%; 26× action-generation throughput vs the OpenVLA baseline in the paper.
• M8OpenVLA-OFT — official project page. Type: official project. Retrieved: 2026-08-29.openvla-oft.github.ioSupports: OpenVLA-OFT identity on the open-peer caption. Still a fine-tune of OpenVLA 7B, not a new pretrain. Published suite checkpoints are M15. Rechecked 2026-09-08.
• R10Gemini Robotics Team — Gemini Robotics: Bringing AI into the Physical World. Type: research paper. ID: arXiv:2503.20020. Retrieved: 2026-08-29.arxiv.org/abs/2503.20020Supports: Gemini Robotics open-peer column — VLA plus Robotics-ER on Gemini 2.0; ERQA 400 questions. No parameter count, weights, or public API.
• P7Google DeepMind — “Gemini Robotics brings AI into the physical world”. Type: company blog. Retrieved: 2026-08-29.deepmind.googleSupports: Gemini Robotics blog identity on the open-peer caption. Not a model card.
• R11Physical Intelligence — FAST: Efficient Action Tokenization for Vision-Language-Action Models. Type: research paper. ID: arXiv:2501.09747. Retrieved: 2026-08-29.arxiv.org/abs/2501.09747Supports: FAST prior-policy column — DCT + BPE action tokenizer, not a standalone VLA; FAST+ on 1M trajectories; π0-FAST on 10k hours; paper claims up to 5× faster training than diffusion VLAs. No policy parameter count is invented. The /blog/fast URL returned HTTP 404 from this checkout.
• M9Physical Intelligence — FAST research page. Type: official research page. Retrieved: 2026-08-29.pi.website/research/fastSupports: FAST research-page / weights row. Not a policy checkpoint.
• M10NVIDIA — Isaac-GR00T current README (N1.7 GA). Type: official code documentation. Retrieved: 2026-09-08.github.com/NVIDIA/Isaac-GR00TSupports: N1.7 as current default-branch GA; Cosmos-Reason2-2B / Qwen3-VL backbone; relative EEF action space; 20K hours EgoScale human video; gated VLM access; Apache-2.0 code; N1.6 kept on the n1d6 tree; RoboCasa / SimplerEnv / real G1 evals named without copying a numeric table. The README’s “commercially licensable under Apache 2.0” sentence is not treated as the weight licence.
• M11NVIDIA — Hugging Face nvidia/GR00T-N1.7-3B. Type: official model card / LICENSE. Retrieved: 2026-09-08.huggingface.co/nvidia/GR00T-N1.7-3BSupports: N1.7-3B model id, NVIDIA Open Model License Agreement named on the card, and the public weight row. Card also says the model is ready for commercial/non-commercial use; that sentence is not copied as a certification.
• M12NVIDIA — Hugging Face nvidia/GR00T-N1.6-3B. Type: official model card. Retrieved: 2026-09-08.huggingface.co/nvidia/GR00T-N1.6-3BSupports: N1.6-3B model id, Eagle-era N1.6 description, NVIDIA License / non-commercial PDF named on the card, and training-mixture wording. Not the N1-2B paper checkpoint.
• M13NVIDIA — Isaac-GR00T GitHub releases. Type: official release tags. Retrieved: 2026-09-08.github.com/NVIDIA/Isaac-GR00T/releasesSupports: n1-release 2025-06-11; n1.5-release 2025-12-15; n1.6-release 2026-04-15; n1.7-release 2026-04-18; n1.6.1-release 2026-04-23. Research-blog URLs named on those tags are not copied as hardware specs.
• M14Physical Intelligence — openpi LIBERO example README. Type: official eval recipe. Retrieved: 2026-09-08.github.com/Physical-Intelligence/openpi — examples/libero/README.mdSupports: π0.5 @ 30k (finetuned) LIBERO table — Spatial 98.8, Object 98.2, Goal 98.0, LIBERO-10 92.4, average 96.85 — for gs://openpi-assets/checkpoints/pi05_libero/ trained with the pi05_libero config. Upstream-reproduced, not site-reproduced. Episodes, seeds, and hardware unpublished on that page.
• M15OpenVLA-OFT — moojink/openvla-oft LIBERO.md and Hugging Face suite checkpoints. Type: official eval recipe / weights. License: MIT code and Spatial card; inherited OpenVLA / Llama 2 terms still apply to the backbone. Retrieved: 2026-09-08.github.com/moojink/openvla-oft — LIBERO.mdSupports: independently trained suite checkpoints including moojink/openvla-7b-oft-finetuned-libero-spatial; combined four-suite checkpoint; eval command, 10×50 default episodes, three paper seeds, A100 / Python 3.10.14 / PyTorch 2.2.0 / custom transformers v4.40.1; 97.1% vs 96.8% four-suite averages. Spatial-only success is unpublished. Fine-tune of openvla-7b, not a new pretrain.
• M16Hugging Face — LeRobot README policy inventory. Type: official library documentation. License: Apache-2.0. Retrieved: 2026-09-08.github.com/huggingface/lerobotSupports: listed VLA / world-model / reward-model names; lerobot-eval LIBERO example command. A README listing is not a pretrained-weight proof and not a SOTA ranking. Hub checkpoints named on the inventory table were confirmed via the Hugging Face models API on the retrieval date; SARM / TOPReward / Robometer and EVO1 had no matching public model id in that check.
• M17LeRobot — Hugging Face world-model cards lerobot/VLA-JEPA-LIBERO and lerobot/VLA-JEPA-Pretrain. Type: official weights cards. License: Apache-2.0 on the inspected VLA-JEPA-LIBERO card. Retrieved: 2026-09-08.huggingface.co/lerobot/VLA-JEPA-LIBEROSupports: public VLA-JEPA checkpoints exist. A world-model card is not a physical-execution success rate. LingBot-VA and FastWAM Hub ids named on the inventory row were seen in the same API check.
• M18OpenVLA-OFT — moojink/openvla-oft commit e4287e94541f459edc4feabc4e181f537cd569a8 (main HEAD). Type: official eval code pin. Retrieved: 2026-09-08.github.com/moojink/openvla-oft — e4287e9Supports: pinned runnable recipe — run_libero_eval.py defaults (--seed 7, 50 trials/task, center_crop true, 8-action chunk, gripper invert, max 220 Spatial steps). A current HEAD is not proof it reproduces the paper’s four-suite averages.
• M19LIBERO — Lifelong-Robot-Learning/LIBERO commit 8f1084e3132a39270c3a13ebe37270a43ece2a01 (master HEAD). Type: official suite code pin. Retrieved: 2026-09-08.github.com/Lifelong-Robot-Learning/LIBERO — 8f1084eSupports: suite code revision used by the pinned OFT Spatial recipe. Not a tagged LIBERO release and not a train/eval split name.
• M20OpenVLA-OFT — Hugging Face API for moojink/openvla-7b-oft-finetuned-libero-spatial revision 6d0231af0e48c5985f1ff86908f4674b84bc049b. Type: official model tree / LFS pointers. Retrieved: 2026-09-08.huggingface.co/api/models/…/tree/6d0231afSupports: checkpoint git revision and LFS SHA-256 oids for the Spatial weight shards, action head, and proprio projector. Inspected via the Hub API without downloading the weight blobs. Card lastModified 2025-06-17.

Safety, ingress, and energy service

• R1IEC — IEC 60529: Degrees of protection provided by enclosures (IP Code). Type: standards-body catalogue. Retrieved: 2026-09-06.webstore.iec.ch/publication/2452Supports: definition of IP codes on the safety page. Does not certify any robot.
• R2ISO — ISO 10218-1 Robotics — Safety requirements for robot systems in an industrial environment — Part 1: Robots. Type: standards-body catalogue. Retrieved: 2026-09-06.iso.org/standard/73933.htmlSupports: ISO 10218-1 scope card. Does not certify any robot.
• R3ISO — ISO 13482 Robots and robotic devices — Safety requirements for personal care robots. Type: standards-body catalogue. Retrieved: 2026-09-06.iso.org/standard/53820.htmlSupports: ISO 13482 scope card. Does not certify any robot.
• R4UL Standards & Engagement — UL 3300 Service, Communication, Information, Education and Entertainment Robots — SCIEE Robots. Type: standards-body shop listing. Retrieved: 2026-09-06.shopulstandards.com/ProductDetail.aspx?productId=UL3300Supports: UL 3300 scope card. Does not certify any robot.

Marine robots

• A1IFR — World Robotics 2026 Service Robots (type B water-based). Type: analyst / classification. Date: 2026. Retrieved: 2026-10-09. Review after: 2027-09-01. Review reason: Recheck against World Robotics 2027 sources and methods.ifr.orgSupports: surface vs diving class definition (Table 1.1, report p. 25).
• M1Blue Robotics — BlueROV2 datasheet. Type: manual / data sheet. Date: 2025-04. Retrieved: 2026-08-25.bluerobotics.comSupports: BlueROV2 sheet.
• M2Blue Robotics — BlueROV2 technical documentation. Type: manual. Retrieved: 2026-08-25.docs.bluerobotics.comSupports: operating-limits reference.
• M3Blue Robotics — BlueBoat datasheet v1.1. Type: manual / data sheet. Date: 2025-01. Retrieved: 2026-08-28.bluerobotics.comSupports: BlueBoat identity, autonomy, geometry, mass, payload definition, motion, conditional endurance, power, interfaces, compute, sensors, software, and communications.
• M4Blue Robotics — BlueBoat current product specifications. Type: manual / product page. Retrieved: 2026-08-28.bluerobotics.com/store/boat/blueboatSupports: current BlueBoat 2/4/6/8-battery endurance conditions, freeboard, draft, static thrust, temperature, and penetrator counts.

ProcTHOR / MolmoSpaces

• R2Ai2 MolmoSpaces and local native execution — NavToObjTask, RB-Y1 source A* navigation recording. Type: research code and recorded simulation evidence. Retrieved: 2026-09-13.Pinned native taskSupports: source task predicate and protocol. Local execution receipts and pose checks support the selected recording, not a suite score or physical hardware result. Attribution and limits.

Forest3D woodland

• R1unitsSpaceLab, input asset authors and local native execution — generated terrain and vegetation with original Clearpath Husky and an authored navigation protocol. Type: research code and recorded simulation evidence. Licence: AGPL-3.0 root; upstream MIT metadata conflict; separate CC0, public-domain and BSD-3-Clause inputs. Retrieved: 2026-09-14.Pinned Forest3D generatorSupports: generated world construction. The input lock, native and geometry evidence, desktop camera proof, and attribution and limits support one authored route. No upstream benchmark score or IFIT wheel-soil result.

Habitat / ReplicaCAD

• R1Meta Habitat contributors and local native execution — Habitat-Lab 0.3.3 source task and oracle. Type: research code and recorded simulation evidence. Licence: MIT code; data assets separately licensed. Retrieved: 2026-09-13.Pinned Habitat-Lab sourceSupports: native NavToObjTask-v0, original horizon and success protocol. Local execution receipts bind the failed navigation recording and independent repeat. Source identity, desktop checks and replay limits.
• R1Open Navigation contributors and local native execution — original warehouse, source Nav2 mission dispatcher and Gazebo sensor observations. Type: research code and recorded simulation evidence. Licence: Apache-2.0 code; separate Creative Commons model terms. Retrieved: 2026-09-13.Pinned original workload benchmarkSupports: one complete warehouse and original navigation mission. Native source and geometry evidence, camera/body desktop proof, and scope, attribution and replay limits. No workload-throughput or learned-policy score.
• R2Meta / Habitat ReplicaCAD contributors — ReplicaCAD v1.6 asset distribution. Type: research dataset. Licence: CC BY 4.0. Retrieved: 2026-09-13.Pinned ReplicaCAD distributionSupports: artist-created apartment variations, object/articulation definitions and navigation meshes. One layout is bound here. YCB and Fetch attribution and separate licences.
• R3Matterport / Habitat — Habitat-Matterport 3D Research Dataset. Type: research dataset and access terms. Retrieved: 2026-09-13.Official HM3D datasetSupports: a separate real-space scan collection for academic, non-commercial research. No HM3D world or episode is published here; ReplicaCAD does not establish HM3D access or coverage.

Genesis native Panda recording

1. R1Genesis contributors — Genesis World 1.4.1 Panda IK, path-planning and grasp tutorial. Type: research code and native simulation output. Retrieved: 2026-09-14.Pinned Genesis tutorialSupports: native CPU simulation and mixed position/force control. Capture, independent replay and media evidence · Inputs, attribution and replay limits. One 701-sample native recording with an authored hold predicate; zero additional browser assemblies or reproduced suite scores.

Newton native manipulation recording

• R1Newton contributors — 1.6.0 native hydro manipulation example, selected pen scene with Franka Research 3. Type: research code and native simulation output. Retrieved: 2026-09-14.Pinned Newton exampleSupports: scripted IK control and upstream final geometric checks. Native capture, replay and media evidence · Inputs, attribution and replay limits. One 721-sample native recording; zero browser world assemblies or reproduced suite scores.

CARLA native driving recording

• R1CARLA contributors — 0.9.16 source, BasicAgent and Town01 native recording. Type: research code and authored simulation output. Retrieved: 2026-09-14.Pinned CARLA sourceSupports: native controller/API semantics and simulation inputs. The route, arrival predicate and 40-second horizon are site-authored. Native capture, replay and media evidence · Protocol and output attribution. One simulation recording; zero browser city assemblies or reproduced suite scores.

Isaac Lab native drawer recording

• R1Isaac Lab / NVIDIA contributors — official Isaac-Open-Drawer-Franka registration, Newton/MJWarp configuration and RSL-RL checkpoint trial. Type: research code and native simulation output. Retrieved: 2026-09-13.Pinned Isaac Lab sourceSupports: registered task, source success above 0.30 m, native 8-second horizon and the exact task/backend checkpoint. Source, runtime, asset, checkpoint and output hashes · Native trial and replay limits. One successful native task; zero browser USD world assemblies; no reproduced suite score.
• M1NVIDIA — Isaac Sim licence FAQ and additional materials terms. Type: vendor licensing guidance. Retrieved: 2026-09-13.Simulation output guidanceCode, native assets and outputs have separate terms. Additional materials licence. This release publishes camera/state outputs, without redistributing original USD, runtime or checkpoint files.

Native simulation frontier

Feasibility matrix and implementation sequence. Source review only: zero new worlds, recordings or reproduced evaluation suites.

• R1Newton contributors — v1.6.0 physics engine and native examples. Type: official research code. Licence: Apache-2.0 code; selected asset terms separate. Retrieved: 2026-09-14.Pinned newton sourceSupports: Python/Warp/MJWarp requirements, bundled cartpole dynamics checks, Panda hydro lift/placement checks and externally pinned asset downloads. Selected source identities; one native FR3 recording is qualified separately.
• R2Genesis-Embodied-AI contributors — Genesis World v1.4.1 engine and Panda grasp tutorial. Type: official research code. Licence: Apache-2.0 code; selected Panda model README separately states Apache-2.0. Retrieved: 2026-09-14.Pinned genesis sourceSupports: CPU-initialized IK/grasp tutorial, exact Panda model path/provenance and declared Python/Quadrants dependencies. Distinct from commercial Genesis AI products. Selected source identities; one native Panda grasp recording is qualified separately.
• R3learnsyslab / gym-pybullet-drones contributors — current source at 7ebad1ec, package version 2.2.0. Type: official research code. Licence: MIT code; original model and pybullet_data notices need separate inventory. Retrieved: 2026-09-14.Pinned drones sourceSupports: CF2X PID trajectory defaults, CtrlAviary reward and horizon semantics, current dependency declarations. Latest GitHub release v1.0.0 from 2021 is a different source generation. Selected source identities; one native flight is recorded, and its qualification failed.
• R4Patryk Cieślak / Stonefish contributors — v1.5 marine simulator and underwater example. Type: official research code. Licence: GPL-3.0-or-later code; third-party and selected asset terms separate. Retrieved: 2026-09-14.Pinned stonefish sourceSupports: C++20/OpenGL build requirements, parsed/programmatic scenario branches, original Girona references and missing mission outcome protocol. Selected source identities; one authored station hold is recorded separately.

BEHAVIOR demonstration replay

• R1StanfordVL / BEHAVIOR contributors — 2026 challenge demonstrations, LeRobot v3.0. Type: official research dataset. Licence: MIT in this dataset repository. Retrieved: 2026-09-13.Pinned demonstration releaseSupports: R1Pro, task turning_on_radio, selected episode 0 / raw id 10 / task instance 1, 1,956 rows and three RGB cameras. Dataset MIT notice · Selected source and frame comparison evidence. This licence does not establish rights for the separate 3D Data Bundle or raw HDF5 distribution.
• R2StanfordVL / BEHAVIOR contributors — OmniGibson v3.9.2 reference reader and installation requirements. Type: research code. Retrieved: 2026-09-13.Pinned BEHAVIOR reference sourceSupports: source channel layout, previous-observation/current-action/next-transition alignment and separate Data Bundle terms. This August 24 release postdates the August 5 dataset revision; it is not proof of the historical producer runtime. Replay scope and source interpretation.