Data engineering for Physical AI, with sources attached.
OpenBot is the data engineering layer for Physical AI: a source-backed catalog of robot datasets and models, openbot-data for local validation, and an API platform with explicit product-state labels.
What is public now
Labels describe current availability—not benchmark validation or paid maturity.
- Catalog · Live
- Source-backed dataset and model records.
- openbot-data · Open source
- Local robot-video and LeRobot dataset inspection, readiness, and repair planning. Local only.
How to read OpenBot
Use the catalog to narrow choices, then verify the upstream source.
Follow the source
Every record links to its dataset, model, paper, or repository page.
Read confidence with readiness
Readiness is a navigation aid. Confidence and missing evidence stay visible. How scores work
Keep full datasets local
openbot-data runs locally without an account or dataset upload.
How scores work
Every record has one headline score: Selection readiness. The other numbers you may see are its breakdown or ordering signals. None of them is a benchmark result, a task-performance claim, or a legal conclusion.
Selection readiness · the headline score
A 0–100 score from the published Catalog evaluation. It asks whether the source evidence is strong enough to inspect, compare, or adopt a record, and averages the dimensions that carry evidence.
- Dataset dimensions
- Access and governance · Schema and signal coverage · Policy training · World-model · Failure and recovery · Download and processing
- Model dimensions
- Access and governance · Artifact availability · Training and loading reproducibility · Training data requirements · Evaluation evidence · Deployment
- A dimension without evidence stays Not scored; it is never counted as zero.
- Verdicts: Strong at 80 and above, Useful from 60, Limited below 60.
- Confidence averages every dimension, so unresolved ones pull it down. The score is labelled Provisional while confidence is below 50% or fewer than half the dimensions carry evidence.
- A record with no scored dimension shows Not scored instead of a number.
Engineering checks (OBRS) · a breakdown
The OpenBot Readiness Standard runs pass/fail checks on record metadata and reported test evidence across standardization and loaders, physical data quality, semantic richness, real-world validation, and license compliance. Detail pages list the checks passed and the gaps found. It is not a separate score and not an independent certification.
Recommended order · not a score
The default Catalog order and homepage picks weigh source evidence, usability, adoption such as downloads and stars, freshness, and distinct value. That ranking decides what you see first; it is not shown as a quality score.