Catalog methodology
Sources before scores.
OpenBot uses source-backed records to help narrow a robot-data decision. Catalog inclusion is not an endorsement, a benchmark result, or permission to use the underlying asset.
Source hierarchy
The strongest available source governs each published field.
- Official dataset or model source
- Publisher documentation, project pages, repositories, model cards, and dataset cards establish identity, access, and declared scope.
- Primary research
- Papers and official technical reports support corpus, benchmark, modality, and training-use relationships.
- Attributed disclosures
- Maintainer correspondence or clearly attributable announcements may guide review, but unpublished information is not presented as public fact.
- Secondary discovery
- Indexes and reporting can surface candidates. They do not override a conflicting primary source.
Record relationships
Deduplication preserves lineage instead of flattening every name into one card.
- Canonical
- One record represents one named source release or corpus. Alternate names point to that record instead of creating another result.
- Derived
- A materially transformed dataset can remain separate when its upstream source and transformation are explicit.
- Composition
- An aggregate remains linked to its upstream records and does not replace them in a source-corpus index.
- Unknown
- Missing evidence stays unknown. OpenBot does not convert an approximate year, marketing claim, or keyword match into a verified fact.
Ego topic review
Current review · 2026-08-31
The source index includes named physical-world, human-worn first-person corpora and synchronized ego–exo corpora. A record must have an attributable source and enough capture evidence to confirm that scope.
Keyword matches, vendor samples, generic onboard-camera data, repackaged aggregates, evaluation-only derivatives, and projects without a named source corpus are excluded from the source index.
Access describes how data is obtained: open, gated, preview, closed, or unknown. It is not a license. License and usage terms remain separate and must be confirmed at the official source.
Paper, training-effect, and dataset links are published only where a reviewed primary paper states the relationship. They do not imply that a dataset alone produces an executable robot policy.
Browse the reviewed Ego indexCorrections and additions
Send the canonical source, the disputed field, and evidence for the proposed change. Submissions enter review; they are not published automatically.