Four principles
- Like for like. Robots are ranked only within their own category. We never claim a robot vacuum is better or worse than a humanoid.
- Evidence over claims. Figures the maker states count in full, figures reported by media or retailers count half, unconfirmed figures count for nothing, and real-world proof is scored separately from paper specifications.
- Honest about uncertainty. Robots with few sourced figures are pulled towards their category average, and every rank shows an evidence grade.
- Stable and independent. The criteria stay fixed until the next annual review. Ranks are recalculated once a month and change only when evidence changes. Nothing can be bought, and there are no manual overrides.
Score (0–100) = 70% Performance + 30% Real-world proof. Price does not affect the rank: "best overall" and "best value" are different questions.
1. Performance (70%)
For each robot we take the category's key performance figures and compare them with every other robot in the same category, as percentiles (where lower is better, such as noise, lower ranks higher). Each figure is weighted by its source. Evidence shrinkage: performance = (n × robot's average + 4 × category average) ÷ (n + 4), where n is the number of sourced figures, so one impressive number cannot beat five good ones.
- Humanoid robots: Degrees of freedom, Hand DOF, Payload, Runtime, Walking speed
- Robot dogs: Speed, Payload, Runtime, Degrees of freedom
- Cobots: Payload, Reach, Repeatability (lower is better), Tool speed
- Drones: Flight time, Transmission range, Top speed, Wind resistance
- Service robots: Payload, Runtime, Tray load
- Air taxis (eVTOL): Range, Cruise speed, Top speed, Passengers, Payload
2. Real-world proof (30%)
A fixed checklist, the same for every robot, each item backed by a sourced value:
| Evidence | Points |
|---|---|
| Commercial status: shipping 50, pre-order 30, pilot deployment with customers 30 | 50 |
| Deployment or shipment figures published (units deployed) | 20 |
| Safety or regulatory certification published (e.g. ISO/IEC safety standards, aviation type certification progress) | 15 |
| Durability rating published (IP rating) | 10 |
| At least one figure independently verified (test lab, regulator, peer-reviewed paper) | 5 |
Evidence grade
| A | Strong: at least 80% of the category's performance figures are sourced, plus real-world proof beyond being on sale (deployments, certification or independent data) |
| B | Good: at least 60% of the category's performance figures sourced |
| C | Fair: at least 40% sourced |
| D | Thin: fewer than 40% sourced — not ranked; listed on the watchlist until more figures are published |
Who is ranked
Advanced robots only (from October): humanoids, robot dogs, cobots, drones, service and delivery robots, and air taxis. Household robots are compared in RankedRobot's buying guides with the same method. Within these categories, commercial robots only: on sale, on pre-order or deployed with customers in a pilot, not discontinued, with at least 8 sourced values and at least 40% of the category's performance figures sourced (grade C or better). Everything else is on the watchlist until the evidence exists.
Ties and small categories
Robots whose scores are less than 1 point from the first robot in their group share its position, shown as "=2". A category with fewer than 5 ranked robots shows scores but no positions.
Monthly updates
The specifications behind the ranking live in the RankedRobot database, which is updated every day. The Robot 500 is fixed once a month, on the 1st, so ranks don't move from day to day. A robot added mid-month joins at the next update. Every month is kept in the archive, so any past rank can be checked.
What the Robot 500 is not
It is not a hands-on test and not a buying recommendation for your specific needs. It ranks published evidence and says so wherever evidence is thin. It is not a popularity vote; it does not measure AI quality or autonomy, because there is no common published measure yet. See known limitations.
How the criteria can change
- Once a year only. Criteria change only with a new edition, at most once a year (the next review is due in January).
- Announced first. Proposed changes are published at least 30 days before they take effect, with the reason and their effect on the ranking.
- Never to move a particular robot, in response to a maker's request or pressure, or for payment.
Independence
Rankings are never for sale; no sponsorship, advertising or affiliate relationship can change a score. Makers can reply to any rank at [email protected]; every correction is published in the corrections log.
Method v2.0 (26 September 2026): like-for-like category rankings, a real-world proof checklist and evidence grades.