roboterradar/humanoid-robot-radarscore

Roboterradar Humanoid & Quadruped Robot Dataset

Curated editorial assessments of 21 commercially relevant humanoid robots (16) and quadruped robots (5), with a frozen scoring methodology, evidence grades and a complete source register.

This Hugging Face repository is a versioned distribution mirror. The canonical, citeable publication is the Zenodo release:

• Version 1.0.0 DOI: https://doi.org/10.5281/zenodo.21797689
• Concept DOI for all versions: https://doi.org/10.5281/zenodo.21797688
• Living dataset page: https://roboterradar.de/datensatz/
• Normative RadarScore 2.0 methodology: https://roboterradar.de/methodik/radarscore-2.0.md

Data as of 2026-08-05 · License: CC BY 4.0, subject to the quotation carve-out below.

Dataset contents

Each model is rated on six axes using RadarScore 2.0. Five axes are evidence-coded: mobility, fine motor skills, autonomy, availability and maturity. Every one of the 105 evidence-coded assessments includes a level, an evidence grade and a review date:

• A: independently verified
• B: credibly documented
• C: manufacturer claim only; scoring is capped
• D: disputed or unsupported; scoring follows the documented D-grade rules

The sixth axis, price-performance, is a separately computed metric. The release contains 313 documented evidence entries and a retained editorial rating/review log with 359 entries. The included offline reproduction gate recomputes 147 published values.

Viewer configurations

The Dataset Viewer exposes the three CSV tables as separate configurations because their schemas differ:

Configuration	Rows	Purpose
models	21	Model master data, prices, availability and computed scores
assessments	105	One row per model × evidence-coded axis
sources	313	Full evidence register with URLs, access dates and citation fields

The complete nested dataset, methodology, codebook, audit records, checksums and reproduction script are preserved byte-for-byte from Zenodo under v1.0.0/.

Load with Python

from datasets import load_dataset

models = load_dataset(
    "roboterradar/humanoid-robot-radarscore",
    "models",
    split="train",
)

assessments = load_dataset(
    "roboterradar/humanoid-robot-radarscore",
    "assessments",
    split="train",
)

sources = load_dataset(
    "roboterradar/humanoid-robot-radarscore",
    "sources",
    split="train",
)

For direct pandas imports, the CSV files are RFC 4180-compliant UTF-8 with BOM:

import pandas as pd

models = pd.read_csv(
    "v1.0.0/roboterradar-models-v1.0.0.csv",
    encoding="utf-8-sig",
)

Integrity and reproduction

The files under v1.0.0/ are the complete 17-file Zenodo release. From that directory:

sha256sum --check SHA256SUMS
node reproduce.mjs

reproduce.mjs runs with Node.js 20+ and no package dependencies. Without importing project code, it reads the package files, recomputes every published score covered by the gate and cross-checks the JSON against all three CSV tables. Its rule tables are checked against the production implementation in the source repository; this demonstrates agreement, not an independently derived methodology. Its optional live-source diagnostic is documented in the package README and does not replace the frozen release audit.

Scope and limitations

• Not laboratory measurements. Roboterradar did not physically test the devices for this dataset; ratings are derived from documented public evidence.
• Not peer-reviewed research. This is transparent editorial work by an independent German robotics publication. The package openly documents the AI-assisted counter-review and QA.
• Not a complete census of all robots. Inclusion follows the criteria recorded in the frozen methodology.
• Manufacturer-only statements are graded C and capped. Missing, disputed or unsupported evidence is represented explicitly rather than silently inferred.
• preis_leistung: null means that no qualifying published EUR price for the rated variant was documented; it is not a poor price-performance score.

Roboterradar's independence and affiliate disclosure are documented at https://roboterradar.de/rechtliches/transparenz/.

License and third-party quotations

The project's own contribution—ratings, structure, documentation and evidence descriptions—is licensed under CC BY 4.0. Third-party wording in citation fields is excluded from that grant; rights remain with the respective rightsholders. The quotations are included as the verification mechanism under the German quotation exception. See THIRD_PARTY_NOTICES.md for the precise boundary.

If you reuse quotation text itself, assess the necessary rights independently. For most analyses, use the structured ratings and source URLs rather than republishing the quotations.

Citation

Katawazi, André (2026): Roboterradar Humanoid & Quadruped Robot Dataset — curated editorial ratings with evidence grades. Version 1.0.0, data as of 2026-08-05. Roboterradar. https://doi.org/10.5281/zenodo.21797689 (CC BY 4.0).

@dataset{katawazi_roboterradar_2026,
  author    = {Katawazi, André},
  title     = {Roboterradar Humanoid \& Quadruped Robot Dataset --- curated editorial ratings with evidence grades},
  year      = {2026},
  month     = {8},
  publisher = {Roboterradar},
  version   = {1.0.0},
  doi       = {10.5281/zenodo.21797689},
  url       = {https://doi.org/10.5281/zenodo.21797689}
}

Corrections are published as new, versioned releases with a public changelog. Reports are welcome at redaktion@roboterradar.de.

Kurzfassung auf Deutsch

Der Datensatz bewertet 16 humanoide Roboter und fünf Laufroboter anhand der eingefrorenen RadarScore-2.0-Methodik. Fünf Achsen tragen Evidenzgrade A–D und konkrete Quellen; die sechste Achse, Preis-Leistung, wird separat berechnet. Es handelt sich um nachvollziehbare redaktionelle Bewertungen, nicht um Labormessungen oder wissenschaftliches Peer Review. Zenodo bleibt die verbindliche Publikation, Hugging Face dient als öffentlicher Vollspiegel und interaktiver Viewer.