Browse

Unitree family mass is a row id, not a page. The G1 cell resolves to humanoids-m2.

R1 / G1 joints split DoF, named-joint torque, pack, encoder, and runtime. Totals are not invented from mixed subsystems.

Galbot payload is wrist-center arm load. It is not a biped arm-load row and not Stretch’s recommended holding load.

biped-subsystem-dof joins only head / per-arm / per-leg rows that passed the three-metric audit. Combined cells stay out.

Biped arm load + IP lists only subjects with both sourced cells. ET1 arm load stays —, not zero. Galbot G1 cannot enter.

A “task” in one evaluation table is not the same unit as a “task” in another. The sim scale row is six paper-specific definitions.

LATENT tennis is a 29-DoF Unitree G1 experiment. It is not Galbot G1 and not Galbot ET1 hardware.

The live card fetches a commit-pinned Galbot description. A parse proves publication, not payload.

Unitree’s exchange-published listing outranks press for ownership and shipment disclosures. Filings are never hardware sources.

The newest data release opens on value changes. Unchanged claims are omitted. Evidence-id and source URL or retrieval are separate filters.

?q=wrist-center+payload#search keeps class and evidence on every hit. Two characters load the index; the URL is the query.

Claim ids are page::table::row::sorted-subjects. This URL opens the Unitree G1 mass cell, its source, and the correction path.

Robot inspector loads the commit-pinned Galbot G1 description. Joint angles are URDF values, not datasheet payload or DoF.

Explore worlds fetches the official HumanoidBench stair fragment. Overlay boxes are XML geoms, not a score and not Unitree H1.

H1 / H1-2 list price is authored as missing. A dash is not zero, not “contact sales”, and not a number borrowed from a reseller.

Galbot G1 / S1, Reachy 2, PAL TIAGo, and Hello Robot Stretch 3. Eno is announced-only.

Galbot ET1, Unitree, AgiBot, Fourier, UBTECH, EngineAI, Fauna, NEURA, Leju, Booster, LimX, and PAL.

Unitree B2, Boston Dynamics Spot, Unitree Go2, and ANYmal D.

UR20, ABB GoFa, FANUC CRX, SCARA, delta.

MiR, OTTO, and Locus warehouse AMRs.

IFR AP classes; Lely Vector, Kärcher KIRA, and Pudu BellaBot.

IFR AP6 classes; isolated da Vinci X sheet.

IFR AC classes; Roborock Qrevo Edge sheet.

DJI Matrice 400 vs 350 RTK multirotors.

BlueBoat surface USV and BlueROV2 tethered diving ROV in separate classes.

Exoskeletons and gait wearables.

Wheeled dual-arm. Wrist-center payload lives on the mobile-manipulator family sheet.

Biped. Mass, arm load, and 29-DoF eval-body rows live on the humanoid sheets.

Third vendor, same letters. Same-class wheeled peer on the field table; not Galbot and not Unitree.

Vendor sheet: 1250 mm standing height, 32 kg with battery, edition-split DoF. Not Galbot G1, not Unitree G1. No SDK or URDF here.

The Apache genesis-world engine is not Genesis AI the company and not Eno.

Desktop companion. Not Reachy 2, not a mobile manipulator, not a vacuum.

Separate SKU tables. K1 is not T1, and optional hands stay out of base-body DoF profiles.

Generation D datasheet. Not ANYmal B, and mounted payload is not B2 standing transport.

Homepage identity only. The English /s1 URL 404’d here. Not Galbot S1.

Walking research biped. Not TIAGo, and not mixed into the wheeled dual-arm field table.

Officially a wheeled data-capture platform. Aggregator biped figures stay out; a future sheet would sit with mobile manipulators.

SDK line only. The vendor hardware-intro URL 404’d here, so N1 has no humanoid column and is not GR-2.

Every covered robot, component, or wearable with its verified engineering-model state and an explicit gap where no public model is available.

Select products from one source table, share the view, and export claim rows with evidence ids and direct source URLs.

List subjects that have explicit sourced cells in one class. Missing cells stay missing. This is not a ranking.

Reproducible previews assembled from commit-pinned vendor and research-project URDFs, plus a publication-state index for xacro, MJCF, USD, and CAD sources. Publication evidence, never a specification source.

IFR installations and density, service-robot units, humanoid installations, and funding.

Major players per class, with covered systems linked to their source-scoped page.

Dex5 hand, Sharpa Wave, Galbot TM01 teleop accessory, and Robotiq 2F grippers. Hand load is never G1 wrist payload.

Genesis Eno, 1X NEO, Digit, IRON, Kepler K2, Sanctuary Phoenix, Tesla Optimus, Figure, Atlas, and Apptronik. Status cards only. Galbot ET1 now has a vendor sheet.

Paper pins (video, official figure, or text), special-env benchmarks, and a single catalog of homepage papers. LATENT tennis stays research on Unitree G1.

Named evaluation-suite and corpus identity. Sixteen tables, including a focused VLABench / RoboTwin 2.0 sheet. Not a method leaderboard or robot ranking.

IP codes, named machine-safety claims, and hot-swap or swap-time continuity. Not a payload sheet and not a ranking.

GalbotSDK, Genesis World / genesis-world, Isaac Lab tasks, Menagerie presence, and Unitree G1 eval-body code.

GENE-26.5, GR00T N1, OpenVLA, π0.5, Skild, Helix, SmolVLA, SpatialVLA, Gemini Robotics, Octo, RT-2, RDT-1B, FAST. Papers and licenses where they exist; declared gaps everywhere else.

Evidence-backed identities and relationships across robots, models, data, runtimes, organizations, deployments, and standards.

One generated page per reviewed identity. Galbot G1, Unitree G1, and Galbot ET1 stay distinct; shared letters are not a join.

Comparability rules, evidence grades, conflict handling, correction workflow, release identity, and licensing.

Exact claim and source diffs for the last recorded data release. Added is not improved; removed is not discontinued.

One workspace for every live URDF. A parsed tree is publication evidence, not a specification source.

Official HumanoidBench courses: locomotion lanes, articulated cabinet and door joints, and object scenes. Preview, Full scene, and a recorded cabinet replay where published. Animation is not a score.

One dated record per marker. Share ?fresh=due for the review queue. Retrieval date is not an evidence score.

Archival export of pages, tables, typed claims, entities, and sources. Contract: landscape.schema.json.

Temporal identity-and-relation graph plus one static HTML and JSON dossier per entity. Contract: ecosystem.schema.json.

Compact comparison projection plus reviewed profiles. Used by compare, scenarios, and the claim inspector.

Filtered compare-index slices. Same claim cells as the entity dossier; not a second dataset.

Maintenance state only: retrieval date, access, and review event. Not an evidence score.

Retained retrieval metadata, digest queues, and human retain/revise/retire decisions bound to the import-time canonical snapshot. A digest change is a review trigger, not a claim edit. Claims no longer in the current export are marked historical. Machine retrieval is not verification.