Methodology

How we collect and check the data

Scope

We cover robots that move and act on their own in the physical world, in five categories: humanoids, quadrupeds, industrial & logistics AMRs, autonomous vehicles, autonomous drones. FPV drones, which are flown by a pilot, are out of scope, as are robot vacuums and other small consumer robots.

Every value has a source

Each specification is stored together with the URL it came from and, where known, the date it was stated. Our build process refuses to publish a value without a source, so this rule is enforced by software rather than by good intentions. We prefer, in order: the manufacturer’s own specification sheet, regulatory filings, the manufacturer’s official announcements, and reporting by established trade or news outlets.

One unit per specification

Manufacturers publish in different units. We convert every value to a single canonical unit (kilograms, metres, km/h, hours) so that two robots can be compared directly. Where we converted a value, the note next to it says so.

Blank means unknown

If a manufacturer has not published a value, the field stays empty (“—”). We do not estimate, extrapolate or fill gaps from unofficial sources.

Comparisons

Robots are compared in full only within their category. Across categories we compare only the values every robot has: mass, dimensions (L × W × H), runtime per charge, top speed, payload.

How leaders are picked

The “Leaders by spec” on the homepage are decided by the published number alone: the highest payload, the longest runtime, the most degrees of freedom, the highest top speed and the lowest published list price. There are no scores and no editorial weighting. A robot whose manufacturer has not published a value cannot lead that spec, and each leader shows how many robots publish the figure at all. The payload-to-mass ratio is the only derived value: payload divided by the robot’s own mass, both as published.

How scores are calculated

Two scores sit on top of the published numbers. They are our calculation, not a manufacturer figure, and the weights are a judgement; that is why the full rule is here, every robot page shows its own breakdown, and the homepage lets you set your own weights. Current version: v1. Any change to the rule gets a new version and a changelog entry. Scores currently cover humanoids only.

Performance Score (0–100) is a weighted mean of five specs:

• Payload: 30%
• Runtime: 25%
• DoF: 20%
• Top speed: 15%
• Hand DoF: 10%

1. Each spec is scaled against the best published value in the category: value ÷ best × 100. The robot with the highest payload gets 100 for payload; one with half that payload gets 50.
2. A spec the manufacturer has not published is left out, never estimated, and the weights of the remaining specs are rescaled to add up to 100%. Each robot page shows which specs counted.
3. A robot is scored only if at least 3 of the five specs are published, including payload or runtime. Otherwise the score is shown as “–”.

Value Score is the Performance Score per $10,000 of the manufacturer’s published list price in US dollars, base configuration, before tax and shipping. Robots with no published price or price on request get no Value Score.

Limits you should know.

• Manufacturers do not measure specs the same way. One robot’s payload may be the load on a single arm, another’s a two-handed lift; runtimes are given without test conditions. The notes next to each value say what was published.
• Prototypes and pre-order robots are scored on announced figures that nobody has verified independently. Read the score together with the status column.
• Scores are relative: when a new robot sets a record, other robots’ scaled values for that spec go down.

Data sources

Manufacturer websites and spec sheets, regulatory databases, company press releases and funding announcements, and trade publications. The exact source is linked next to every value.

Corrections

If you find an error or a newer official figure, tell us with a link to the source. Corrections are made in the data file for that robot, so every change and its source are recorded.