Methodology
One of the first French-language media outlets run by AI agents.
At a glance
- Where the figures come from. Every value is taken from a source quoted word for word, with its publication date and its retrieval date. With no proof, the cell stays empty: “not disclosed”.
- Four levels of evidence. Stated by the manufacturer, shown by the manufacturer in a demonstration, reported by an independent third party, measured in an independent test. The level is shown next to the value.
- Four stages for the robot. Deployed (available to buy), pilot (limited deliveries), demo (seen working), announced only (no known demonstration or sale).
- Who writes. AI tools sort the sources and write the briefs and articles, under the supervision of the editor, who checks afterwards. A reported error is corrected, and the correction is published with its date in the corrections log. Every text written with AI carries a note next to its date, linking to this page.
- Vuntum tests. No Vuntum test published to date: no value on the site is measured by Vuntum itself.
The principle: one value, one piece of evidence
Every value published on Vuntum is a claim tied to an exact quote from an identified source, to a publication date and to a date of retrieval. A value without a validated claim is never displayed. An empty cell means Not disclosed: it is not a negative answer.
The list of sources and their confidence scores is public, as is the log of corrected errors. For who publishes Vuntum, and with what resources, see the presentation of the editorial team; for what the site does and how to report an error, the who we are page.
What we follow
- Followed first
- Humanoid robots and physical AI (robot models, hands, actuators, sensors, simulation) come before everything else: discovery, writing, checking and publication. Humanoids for the general public come first, then the other humanoids, then legged robots, mobile manipulators and utility robots. A humanoid may have no legs: a torso on wheels with an AI is a humanoid and stays followed first.
- Followed less
- Industrial robots, arms and defence (outside automatic publication) come after the humanoids, with the same checks.
- No longer followed
- Toy robots, vacuum and floor-washing robots that are not humanoids, lawn mowers, pool and window robots and other small household robots are no longer followed (decision of 13/11/2026): no new sheet, no brief, no watch, existing pages set to noindex and out of the sitemap, nothing deleted. Humanoids, torso on wheels included, are never touched by this withdrawal. Defence: no brief or article about a military or defence robot is published automatically (decision of 13/11/2026); the sheets of defence robots stay complete.
Editorial scope updated on 13 November 2026. The same text is given to AI systems by the MCP server (get_editorial_scope tool).
Levels of evidence
| Level | Meaning | Weight |
|---|---|---|
| Manufacturer claim | The manufacturer states it. No independent verification. | 0.45 |
| Manufacturer demonstration | Shown in a demonstration. Conditions not controlled. | 0.5 |
| Independently reported | Reported by an independent third party. | 0.7 |
| Independent test | Measured in a third-party test. | 0.9 |
| Tested by Vuntum | Measured by Vuntum itself. No value at this level yet. | 1 |
| Unverified | No level of evidence established. | 0.1 |
Two statuses override the level of evidence: disputed (two credible sources contradict each other) and stale (the last check is older than the freshness limit of the criterion).
Vuntum autonomy scale
| Level | Definition |
|---|---|
| A0 - Teleoperated | A human commands every movement live. |
| A1 - Assisted | The robot stabilises and secures its movements; the human decides on the action. |
| A2 - Scripted tasks | The robot carries out preprogrammed sequences on its own in a known environment. |
| A3 - Supervised autonomy | The robot chains tasks together on its own; a human watches and takes over when needed. |
| A4 - Task autonomy | The robot completes a whole task without supervision in an open environment. |
| A5 - General-purpose autonomy | The robot performs new tasks without specific instructions. No known product. |
Status: four stages
Every profile carries a stage saying where the robot stands: deployed, pilot, demo or announced only. It is computed from the validated claims and the commercial status, never typed by hand, and a profile with no proof has no stage. What the maker writes about itself never raises the stage beyond announced only.
| Level | Rule |
|---|---|
| Deployed | Available to buy: commercial status “shipping” and a validated availability claim, with its source. |
| Pilot | Limited deliveries: status “limited shipping” and a validated availability claim. |
| Demo | Seen working: at least one validated value rests on a demonstration or a test. |
| Announced only | Announced only: Vuntum has not yet recorded any proof of a demonstration or a sale. |
Estimated European price
This is not a fact, it is a calculation. It is always shown next to the original official price and labelled Vuntum estimate:
official price x dated exchange rate x (1 + customs duty) x (1 + VAT) + flat fees
The exchange rate used is dated and sourced. Customs duty, VAT and actual fees vary by country and sales channel: the estimate does not commit any seller.
Price range of upcoming releases
On the list of upcoming releases, the price shown is first the one the manufacturer announced, with its source and date. When no price is announced, the page says “Price not disclosed”.
In that case only, a range may be added, labelled “Vuntum estimate”: it is never a price. It runs from the lowest to the highest observed price of robots already available to buy, first those of the same manufacturer, otherwise those of the same form and use. Every robot counted is listed under the range, with a link to its profile and source.
Conditions: at least three robots compared, each with an observed price in the same currency checked within the last 30 days; the dearest no more than four times the cheapest (beyond that, a range would tell nothing); no range at all when the manufacturer announced a price. Without these conditions, the page shows no estimate.
Rankings
Each ranking rests on a single criterion, a sourced value comparable from one robot to the next: weight, speed, payload, official list price. There is no score and no average, and the page states its criterion, its sources, its date of update and what is left out (a variant’s value, a measurement under particular conditions, a price in another currency).
The bars on each page are proportional to the published value, compared with the largest of the list. They do not measure a robot’s quality. The page’s legend recalls what each level of proof is worth.
Two proof-based rankings: method and state of the data
Two rankings are about what a robot really can do: “The most intelligent robot” (the most gifted) is the one that does the most things well, on its own, shown in real video, teleoperation stated; “The most useful at home” is the one whose household tasks are really demonstrated, with no hidden teleoperation. Neither rests on an announcement alone.
Criteria: the number of tasks completed with no human intervention, filmed in real conditions and dated; declared teleoperation (a teleoperated demonstration does not count as autonomous); the level of proof of each demonstration. Excluded: announcements without video, computer-generated images, edited footage and any demonstration whose autonomy is not established.
State of the data: 4 published robots have a household task stated by a source and 112 have a declared possible teleoperation, but these statements are announcements. No filmed demonstration with established autonomy is recorded yet: every robot is therefore excluded for lack of proof, the threshold of four robots is not reached and these two rankings are not published.
Publication
Vuntum publishes automatically, within safeguards, whatever passes all of its checks. A value is published only after a second, independent reading by another AI model, which checks the value against the quote from the source. A value that only a media outlet reports, without having measured it, is published only if a second independent source gives the same value (the manufacturer’s page or another outlet), and it is refused if that second source does not turn up within 30 days. An article is published only if each of its factual sentences points to a value that is already published and its figures come from those values. A brief is published only if each of its sentences rests on a quote found word for word in the source, its figures appear in those quotes, and a review by another model finds nothing to object to.
Values that are already published are re-read at random every day: a value that is contradicted is withdrawn and the correction is displayed. A circuit breaker suspends all automatic publication in the event of abnormal behaviour (too many refusals, too many publications from a single source). The editor checks after the fact and can reverse anything. Reports from readers are handled first.
Never published: buying recommendations, accusations, unestablished security or privacy information, rumours, and anything a source does not allow us to verify.
Who answers for what is published
Vuntum is published by Antoine, publication director. The identification details the law requires are in the legal notice.
Before publication, a text or a value must pass: the exact quote found word for word in the source, the figures found in that quote, a second reading by another AI model and, for a photo, a visual check with the author and the licence stated. Whatever fails one of these checks is not published.
After publication, values already online are re-read at random every day, an automatic audit cross-checks the published data against the pages they cite every day, and the editor receives each morning a summary of what was put online, refused and reported the day before. The editor can withdraw or correct any content.
Corrections. Every correction is dated in the corrections log, and an important correction is also flagged on the page concerned. To report an error: report form.
The role of AI
AI tools sort the sources, extract the values, translate, write the articles and briefs, and review the work of another model. Their output is checked mechanically: the quote must be an exact substring of the source, numbers must be found in the quote, and an identifier absent from the input data invalidates the output. Every text written with AI carries a note next to its date.
Collection
We respect robots.txt, platform terms and rate limits. We use no workaround for anti-bot protection. When collecting sources for briefs, articles and robot profiles, we neither download nor rehost any video. Our own short videos may reuse brief, credited excerpts of third-party videos. Source snapshots of the pages we consulted are kept internally as evidence and are never republished: only short quotes are displayed, with attribution.
Independence
No paid placement in comparisons. Any affiliate link is flagged where it appears. Conflicts of interest are declared. An error that is reported is corrected and recorded in the corrections log.
Technical detail
For developers and readers who want to recompute: the score formula, freshness limits by criterion with each criterion’s name in the API, and the badge rules.
Confidence score
The score is deterministic and explainable. It does not depend on any impression formed by a model:
confidence = 0.40 x source confidence + 0.35 x level of evidence + 0.15 x independent corroborations + 0.10 x freshness
Freshness limits by criterion
| Criterion | Unit | Freshness (days) | Risk class |
|---|---|---|---|
Manufacturer suggested priceKey in the API: price.msrp | - | 30 | Medium |
Manufacturer estimated priceKey in the API: price.estimated | - | 30 | Medium |
Street priceKey in the API: price.street | - | 30 | High |
Pre-order depositKey in the API: price.preorder_deposit | - | 30 | High |
Monthly subscriptionKey in the API: price.subscription_monthly | - | 30 | Medium |
AvailabilityKey in the API: availability.state | - | 30 | Medium |
Announced first shipping dateKey in the API: availability.first_shipping_date | - | 90 | Medium |
Product stageKey in the API: lifecycle.stage | - | 180 | Medium |
Announcement dateKey in the API: lifecycle.announced_date | - | 3650 | Low |
HeightKey in the API: dimensions.height | m | 180 | Low |
WidthKey in the API: dimensions.width | m | 180 | Low |
DepthKey in the API: dimensions.depth | m | 180 | Low |
DimensionsKey in the API: dimensions.envelope | - | 180 | Low |
WeightKey in the API: mass.weight | kg | 180 | Low |
Battery runtimeKey in the API: battery.runtime | s | 180 | Low |
Battery capacityKey in the API: battery.capacity | Wh | 180 | Low |
Charge timeKey in the API: battery.charge_time | s | 180 | Low |
Hot-swappable batteryKey in the API: battery.hot_swap | - | 180 | Low |
Maximum speedKey in the API: mobility.max_speed | m/s | 180 | Low |
Locomotion typeKey in the API: mobility.type | - | 365 | Low |
Climbs stairsKey in the API: mobility.stairs | - | 365 | Medium |
Degrees of freedomKey in the API: mechanics.dof_total | count | 365 | Low |
Degrees of freedom per armKey in the API: mechanics.dof_per_arm | count | 365 | Low |
PayloadKey in the API: payload.carry | kg | 180 | Low |
Industrial robot typeKey in the API: arm.type | - | 365 | Low |
ReachKey in the API: arm.reach | m | 365 | Low |
RepeatabilityKey in the API: arm.repeatability | m | 365 | Low |
Maximum tool speedKey in the API: arm.max_tcp_speed | m/s | 365 | Low |
Number of axesKey in the API: arm.axes | count | 365 | Low |
MountingKey in the API: arm.mounting | - | 365 | Low |
ControllerKey in the API: arm.controller | - | 365 | Low |
Hand typeKey in the API: hands.type | - | 365 | Low |
Functional handsKey in the API: hands.functional | - | 365 | Medium |
Autonomous navigationKey in the API: navigation.autonomous | - | 365 | Medium |
MappingKey in the API: navigation.mapping | - | 365 | Low |
Teleoperation supportedKey in the API: teleoperation.supported | - | 365 | Medium |
Vuntum autonomy levelKey in the API: autonomy.level | - | 365 | High |
Household tasksKey in the API: tasks.household | - | 365 | Medium |
CameraKey in the API: sensors.camera | - | 365 | Low |
MicrophoneKey in the API: sensors.microphone | - | 365 | Low |
LiDARKey in the API: sensors.lidar | - | 365 | Low |
Cloud processingKey in the API: privacy.cloud_processing | - | 180 | High |
On-device processingKey in the API: privacy.local_processing | - | 180 | High |
A human teleoperator can view the streamKey in the API: privacy.human_teleoperator_can_view | - | 180 | High |
Server locationKey in the API: privacy.server_location | - | 180 | High |
Data retentionKey in the API: privacy.data_retention | - | 180 | High |
Public SDKKey in the API: software.sdk_public | - | 180 | Medium |
Cloud APIKey in the API: software.api_cloud | - | 180 | Medium |
Local controlKey in the API: software.local_control | - | 180 | Medium |
ROS compatibilityKey in the API: software.ros | - | 180 | Low |
Python APIKey in the API: software.python | - | 180 | Low |
C++ APIKey in the API: software.cpp | - | 180 | Low |
MCP compatibleKey in the API: software.mcp | - | 180 | Low |
Home Assistant compatibleKey in the API: software.home_assistant | - | 180 | Low |
Operating systemKey in the API: software.os | - | 365 | Low |
Spoken languagesKey in the API: software.spoken_languages | - | 365 | Low |
Official SDKKey in the API: software.sdk_name | - | 365 | Low |
Open-source SDKKey in the API: software.open_source | - | 365 | Low |
SDK licenceKey in the API: software.sdk_license | - | 365 | Low |
Supported ROS 2 distributionsKey in the API: software.ros2_distributions | - | 180 | Low |
Cloud account requiredKey in the API: software.cloud_required | - | 180 | Medium |
Alternative firmwareKey in the API: software.alt_firmware | - | 180 | Medium |
Natural-language control by an AIKey in the API: software.llm_control | - | 180 | Medium |
Block-based programmingKey in the API: software.visual_programming | - | 365 | Low |
Official simulation modelKey in the API: simulation.model_available | - | 365 | Low |
Isaac Sim simulationKey in the API: simulation.isaac_sim | - | 365 | Low |
Gazebo simulationKey in the API: simulation.gazebo | - | 365 | Low |
MuJoCo simulationKey in the API: simulation.mujoco | - | 365 | Low |
Webots simulationKey in the API: simulation.webots | - | 365 | Low |
Official mobile appKey in the API: control.mobile_app | - | 365 | Low |
Compatible voice assistantsKey in the API: control.voice_assistant | - | 365 | Low |
Onboard computeKey in the API: compute.onboard | - | 365 | Low |
Power consumptionKey in the API: power.consumption | W | 365 | Low |
WarrantyKey in the API: warranty.duration | s | 365 | Medium |
Safety certificationKey in the API: safety.certification | - | 365 | High |
Customer pilotKey in the API: deployment.pilot | - | 3650 | High |
On-site trial or demonstrationKey in the API: demonstration.on_site | - | 3650 | High |
Filmed demonstrationKey in the API: demonstration.filmed | - | 3650 | High |
Official recallKey in the API: safety.recall | - | 3650 | High |
Mission familyKey in the API: mission.family | - | 365 | High |
Ingress protection (IP)Key in the API: compliance.ip_rating | - | 365 | Medium |
CE markingKey in the API: compliance.ce | - | 365 | Medium |
RoHSKey in the API: compliance.rohs | - | 365 | Medium |
FCCKey in the API: compliance.fcc | - | 365 | Medium |
UL certificationKey in the API: compliance.ul | - | 365 | Medium |
ISO 10218 (industrial robots)Key in the API: compliance.iso_10218 | - | 365 | Medium |
ISO/TS 15066 (cobots)Key in the API: compliance.iso_ts_15066 | - | 365 | Medium |
ISO 13482 (personal care robots)Key in the API: compliance.iso_13482 | - | 365 | Medium |
UN 38.3 (battery transport)Key in the API: compliance.un38_3 | - | 365 | Medium |
LicenseKey in the API: repository.license | - | 365 | Low |
GitHub starsKey in the API: repository.stars | count | 30 | Low |
Badges
| Badge | Rule |
|---|---|
| Manufacturer claim | The highest level of evidence among the validated claims is “manufacturer claim”. |
| Manufacturer demo | The highest level of evidence among the validated claims is “manufacturer demonstration”. |
| Independently tested | At least one validated claim comes from an independent test. |
| Tested by Vuntum | At least one validated claim comes from a measurement made by Vuntum. |
| Unverified | No validated claim above the “unverified” level. |
| Teleoperation possible | Validated claim `teleoperation.supported` = yes. |
| Public SDK | Validated claim `software.sdk_public` = yes. |
| Cloud API | Validated claim `software.api_cloud` = yes. |
| Local control | Validated claim `software.local_control` = yes. |
| Pre-order | Commercial status `preorder`, or validated claim `availability.state` = preorder. |
| Available in Europe | Validated claim `availability.state` = available, from an official source, for an EU/EEA market. |
| Sample data | The entity or the value is marked `is_sample`. |
| Stale data | At least one displayed claim is past the freshness limit of its predicate. |
| Disputed | At least one displayed claim is in the `disputed` state. |