Robot videos are easy to make and hard to read. Every story on RWT Robotics carries a label that tells you how much of a claim we could actually verify, and this page explains what those labels mean, how we test demos, and how we fix mistakes.

Why We Label at All

A clip of a humanoid folding laundry can mean very different things. It might be a shipping product working on its own. It might be one good take out of hundreds. It might be sped up four times. A person off camera might be driving it with a VR headset and motion-capture gloves.

Most coverage doesn’t distinguish between these cases. We do. The label sits at the top of the story so you know the evidence level before you read the first paragraph.

The Four Labels

Label What it means What it does not mean
CONFIRMED The capability or fact is backed by the company’s own official statement plus independent verification, or by multiple established outlets with direct access. That the robot is good, safe, cheap, or commercially successful.
DEMO ONLY We saw the robot do it in a company video or staged event, but there is no independent evidence it works outside that setting. That the demo was faked. Many real capabilities start here.
TELEOPERATED A human was controlling the robot, fully or partly, during the task shown. Either the company disclosed it or credible reporting established it. That the hardware is worthless. Teleoperation is a legitimate tool for data collection and remote work.
RUMOR The claim comes from unofficial sources: leaks, insiders, job listings, supply-chain chatter, or social posts. That it is false. It means unverified, and we grade the source (see below).

CONFIRMED

This is our highest bar. For a capability, we want evidence the robot did the task in conditions the company didn’t fully control: a customer deployment, a hands-on session for reporters, a third-party test, or a peer-reviewed result. For a business fact (a funding round, a factory pilot, a price), we want the company’s own announcement or a filing, ideally corroborated by an established outlet.

A press release alone confirms that a company said something. We report it that way: “the company says” rather than “the robot can.”

DEMO ONLY

Most robot news lives here, and that’s fine. A DEMO ONLY label tells you the footage exists and appears genuine, but we can’t yet say how often the robot succeeds, how much the environment was arranged for it, or whether it generalises to new objects and rooms.

TELEOPERATED

We apply this whenever a human is in the control loop for the task shown. That includes full puppeteering, where an operator’s movements map directly onto the robot, and partial assistance, such as a human handling speech, navigation, or recovery from errors while the robot runs some behaviours on its own.

If teleoperation was disclosed in small print, a caption, or a follow-up post, we still apply the label prominently. Disclosure buried in a reply thread doesn’t change what viewers saw.

RUMOR

Unconfirmed reports get this label every time, including ones we think are probably true. We never present a rumor in the headline as fact, and the story states who the source is and how reliable they have been.

Labels Can Change and Combine

A story can carry more than one label. A robot can be CONFIRMED as a real product while a specific task in its launch video is TELEOPERATED. When new evidence appears, we update the label, note the date of the change, and explain what moved it. A DEMO ONLY claim that later shows up in an independent hands-on review gets upgraded. A clip later revealed to be remotely operated gets downgraded.

How We Check a Demo

We work through the same questions for every significant robot video. We don’t need a “yes” to all of them before publishing, but we tell you which ones we couldn’t answer.

Was it autonomous?

  • Does the company explicitly say “autonomous” or “no teleoperation,” and does that claim cover the whole clip or only part of it?
  • Is there a visible operator, tether, headset, or control rig in any frame, reflection, or behind-the-scenes footage?
  • Do the robot’s movements show the telltale signs of human mirroring: hesitation that matches human reaction time, overly fluid gestures, or head movements that track like a person looking around?
  • Did the company give reporters hands-on access, and did those reporters describe seeing the robot run without an operator?
  • Has the company published technical details (a paper, model description, or engineering post) that make the claimed autonomy plausible?

Was it scripted?

A robot can be fully autonomous and still be running a pre-planned sequence in a room arranged precisely for it. We look at:

  • Whether objects sit in suspiciously convenient positions or the scene looks dressed for the shot.
  • Whether there is any variation, interruption, or recovery from error, or whether it is a single clean run.
  • Whether the company shows failures or success rates. Companies that publish failure footage earn more trust from us, not less.
  • Whether the same environment appears across many of the company’s videos, which suggests a tuned demo space.

Was it sped up?

  • Is there a speed indicator on screen (“1x,” “4x,” “real time”)? If there isn’t one, we treat speed as unknown.
  • Do clocks, people, liquids, cloth, or background motion move naturally?
  • Does the claimed task time match what the clip implies?

When a video is sped up and the company says so, we report the real-time figure. When it doesn’t say so and the motion looks accelerated, we state that the speed is unverified.

Was it teleoperated?

This overlaps with the autonomy check but deserves its own pass, because teleoperation is often disclosed after the fact. We check the company’s posts and replies in the days after a launch, statements to reporters, and event coverage. A well-known case is Tesla’s October 2024 robotaxi event, where Optimus robots serving drinks and talking to guests were widely reported by established outlets to have had human assistance. That’s the kind of situation where we would label the event footage TELEOPERATED even if the product itself is real.

How We Grade Rumor and Leak Sources

Every RUMOR story names the type of source and phrases the claim to match. We use four grades:

  1. Official. The company itself, a regulator, a court filing, or another primary document. These usually move a story out of RUMOR entirely.
  2. Proven reporter. A journalist or outlet with a public track record of accurate reporting on the company or sector. We write “Outlet X reports.”
  3. Community or datamine. Findings from job listings, patent filings, trademark records, supplier documents, firmware, or app code. We describe exactly what was found and avoid stretching it into conclusions the evidence doesn’t support.
  4. Anonymous or mixed record. Unnamed insiders, social accounts, and leakers whose past claims have been hit and miss. We write “an anonymous account with a mixed track record claims” and keep the claim out of the headline.

We report about leaks. We don’t host, embed, or point to leaked files, footage, or code, and we don’t explain how to obtain them. We never speculate about the real identity of an anonymous source.

How the Humanoid Robot Tracker Is Built

The tracker is a structured table of humanoid robots: who makes them, what they are claimed to do, and where they stand. It follows the same rules as our stories, with a few additions.

Where the data comes from

Each value comes from either an official source (the manufacturer’s website, spec sheet, press release, investor materials, or regulatory filing) or an established outlet with a record of accurate robotics or technology reporting. We don’t fill fields from social media posts, fan wikis, or other aggregators, because errors spread quickly between trackers that copy each other.

Unknown values stay empty

If a company hasn’t published a figure and no reliable outlet has reported one, the cell stays blank. We don’t estimate, extrapolate, or carry over a number from a different model. An empty cell is information: it tells you the company hasn’t disclosed it.

Price is the most common example. A “target price” mentioned in an interview is recorded as a stated target, with the context, rather than as a retail price.

Last-checked dates

Every row carries a last-checked date showing when we most recently verified that entry against its sources. Specs and status change often in this sector, and a figure that was accurate eight months ago might not be now. If a row looks stale, treat it with more caution, and feel free to tell us.

Confidence notes

Some fields need more than a value. Confidence notes explain things like:

  • A spec is the company’s claim and hasn’t been independently tested.
  • Figures differ between the company’s site and its launch presentation.
  • A “deployment” refers to a pilot with a single customer, not commercial sales.
  • A capability is shown only in DEMO ONLY or TELEOPERATED footage.

Where a status field applies (prototype, pilot, limited availability, on sale), we use the most conservative description the evidence supports.

What We Don’t Do

  • We don’t accept payment for labels, tracker placement, or coverage.
  • We don’t upgrade a label because a company asks us to. We upgrade it when the evidence changes.
  • We don’t treat investor decks, funding totals, or valuations as proof that a robot works.
  • We don’t invent specifics. If we couldn’t verify it, we say so or leave it out.

Corrections

We will get things wrong sometimes. When we do, we fix it openly.

  • Factual errors are corrected in the story or tracker row, with a dated note at the bottom describing what changed. We don’t silently edit.
  • Label changes are noted in the same way, with a short explanation of the new evidence.
  • Significant errors, the kind that changed the meaning of a story, get a correction note at the top of the piece as well as the bottom.
  • Tracker updates refresh the last-checked date, and any corrected value is noted in that row’s confidence notes.

If you spot an error, an outdated figure, or footage you believe was mislabeled, contact us through the channels listed on the site. Tell us what is wrong and, where possible, point to the source that shows it. Companies are welcome to send corrections too, and we apply the same evidence standard to them as to anyone else.

Why This Matters to Us

RWT Robotics is a new, independent site. We don’t have a long history to point to, so the method on this page is how we earn trust. Robotics is making real progress, and that progress is easier to see when the staged clips, sped-up takes, and puppeteered demos are clearly marked for what they are.