Ranking methodology

Every model is scored 1 to 5 on five capabilities, using only evidence a company has shown in public: videos, customer pilots, deliveries and independent reports. No evidence means a lower score.

1-5 capability scale

We score only what a company has shown in public: videos, pilots, independent reviews. No evidence means a lower score. A 1 maps to 0 and a 5 maps to 100.

Overall score

Locomotion 15%, manipulation 25%, hands 15%, autonomy 25%, deployment 20%.

Technology and commercial

Technology: locomotion 20%, manipulation 30%, hands 20%, autonomy 30%. Commercial: deployment 70% and sales stage 30%.

Value for money

Technology score divided by the order of magnitude of the price (log10). Only models with a public price.

What each score means

Locomotion: 1 stationary or basic wheeled base · 2 slow walking on flat floors · 3 stairs and uneven ground · 4 running, push recovery, getting up after a fall · 5 acrobatics, parkour.

Manipulation: 1 none or scripted gripper · 2 scripted pick-and-place · 3 two-handed work in a real setting · 4 varied objects and tool use · 5 long, complex tasks (laundry folding, assembly) that generalize.

Hands: 1 gripper · 2 simple hand under 6 DoF · 3 6-11 DoF · 4 12+ DoF with touch sensing · 5 near-human 20+ DoF hand with touch sensing.

Autonomy: 1 teleoperated or scripted · 2 limited autonomy in demos · 3 learned end-to-end / VLA policies · 4 hours of autonomous work at customer sites · 5 generalist, language-commanded, independently verified.

Deployment: 1 prototype · 2 sold to researchers, a few units · 3 paid pilots · 4 100+ robots at customers · 5 1,000+ shipped.

Data confidence

Each model carries a confidence label. High: official specs and independent confirmation. Medium: official or trade-press data, not independently confirmed. Low: aggregators or a single source. Pages with too little verified data are kept out of search results until they are enriched.

Changelog

Methodology maintained by Andrei Sheshenin, Editor.