About Physical AI Radar
Evidence-first intelligence for physical AI. We report what a robot can actually do, how independently it has been demonstrated, and how commercially real it is — as three separate answers, never one score.
Why we exist
Humanoid robotics is covered by announcement. A staged demonstration, a letter of intent and a working production line all arrive as the same kind of news story, and by the time they reach a spreadsheet they look identical. We built this to keep them apart.
Every figure on this site is computed by a frozen scoring engine from evidence that was captured durably, classified against a fixed vocabulary, reviewed by two people and promoted through a controlled workflow. If we cannot evidence something, we say so rather than estimating it.
Editorial standards
- No pay-to-score
- No manufacturer can pay to appear, to rank higher, or to change a score. We take no sponsorship, placement fees or commission from any company we cover. If that ever changes, it will be disclosed on this page before it takes effect, not after.
- Independence is measured, not assumed
- Every source is classified for independence. A manufacturer describing its own robot is recorded as first party and earns no independent credit, however detailed it is. A corporate parent reporting on its own subsidiary is partially independent. We publish the numerator and the denominator so you can see how thin the independent evidence is.
- Conflicts and separation
- Research, review and publication are performed by different people, and no one can approve their own work. Any commercial relationship that later arises with a covered company will be disclosed on the affected profiles.
- We publish unflattering results
- Several robots on this site score 0.0 for capability. That is not an editorial judgement about the robot; it means no eligible evidence earned any credit under the frozen methodology. We do not suppress low results, and we do not soften them.
- Corrections are welcome and tracked
- Manufacturers and readers can challenge anything here. See Corrections and manufacturer response.
What our scores are not
Our scores are not a safety rating, not a certification, not a purchase recommendation and not a measure of build quality. Evidence Confidence describes how well evidenced a robot is, not how good it is. "Evidence Ready" means the evidence is ready for publication — not that the robot is ready for your factory.
How to read a profile
Start with the state, not the number. A dimension is either calculated or marked INSUFFICIENT DATA, and those are different answers rather than a high and a low score. Then read the source ledger: it shows every source we used, how it was classified, and which ones the engine actually read. The methodology and evidence standard explain the rules in full.