Independent intelligence on humanoid capability.
Rankings, comparisons, deployment evidence and market signals. Every score links back to public evidence.
- Published
- 27 July 2026
Start from what you need to know.
Every research surface, grouped by the question it answers.
Judge capability
Verify proof
What one sector is actually buying.
Sector reads sit next to the ranking because a robot is only as useful as the work a specific industry will pay for.
How evidence changes the score.
Take one capability — say, moving totes off a line without a human standing by. This is how far it can be scored as the record improves.
- 01Vendor claimLow ceiling
A press release or spec sheet says the robot can do the task.
Counts, but capped. A claim on its own can never carry a measure to the top of its range.
- 02Demo shownMedium ceiling
Public video or on-stage run showing the task performed.
The capability becomes visible. Cut-aways, speed-ups and off-screen operators keep the ceiling below field evidence.
- 03Named pilotHigh ceiling
An identifiable customer or site running the robot on real work.
The measure can now clear the demo ceiling — someone outside the vendor has put the robot in front of their own operation.
- 04Measured operationFull range
Third-party uptime, throughput or task metrics from that site.
Only here can a measure reach the top of its range, because performance has been observed rather than described.
- Fixed denominator
- Every robot is read against the same 40 measures, so a thin public record shows up as a lower score rather than a smaller test.
- Unknown is not incapable
- A measure with no public evidence scores zero against the denominator. That records our ignorance, not the robot's limits.
- Claims are capped
- A vendor statement can raise a measure off the floor. It can never place one at the top.
- Every score links to a source
- Each measure carries the dated record behind it, so any reader can check the evidence rather than trust the number.