# Autonomous or remotely piloted humanoid: IEEE Spectrum’s six clues for decoding a video

> A humanoid video does not always say who is holding the controls. The six clues IEEE Spectrum gives, several concrete cases (Optimus, 1X’s NEO, the Nucleus tape) and the questions to ask before believing a robot works on its own, based on pages read on October 30, 2026.

Published on 2026-10-30 · Updated on 2026-11-08

## A video does not say who is in control

When a humanoid folds laundry or stocks a shelf, the video does not always tell you whether it decides on its own, follows a sequence written in advance or is piloted by someone remotely. Evan Ackerman, IEEE Spectrum’s robotics editor, sees in a January 23, 2024 article a communication problem that the whole robotics community struggles with.

Teleoperation is not a deception in itself. Ackerman points out that humans remain the gold standard for delicate or creative tasks. The problem, in his view, is what the public perceives: a demo that looks autonomous but is actually teleoperated misrepresents the state of the technology and can minimize the real contribution of other researchers. He does not usually think the roboticists posting such videos mean to mislead. Because teleoperation does not scale, it is also used to collect training data: he cites the company Sanctuary AI and Mobile ALOHA from Stanford’s IRIS Lab, which its video description presents as a low-cost, whole-body teleoperation system for data collection.

The observation still holds. Humanoid.guide (article of June 20, 2026, updated September 21) draws on a Let’s Data Science summary of an AFP-JIJI report, published in the Japan Times, on the Robotics Summit in Boston, in late May. The site lists among its examples humanoids mixing cocktails, running marathons or folding laundry, notes that some demonstrations remain teleoperated or limited to chores scripted in advance without naming them, and reports that 1X’s NEO, according to AFP-JIJI, was being steered by a person off to the side. It also quotes Chris Matthieu, of RealSense, who says most of the humanoids you see are teleoperated or follow very specific paths and chores. These remarks reach us third-hand: we have not read the original report.

## Ackerman’s six clues

Ackerman lists six; here are the first three. A visible operator: legs or a gloved hand in the frame, though their absence proves nothing. The information that comes with the video: description, project page, paper, credits. The type of task: unstructured manipulation, fabrics and long multistep tasks are, in his view, the most likely to be teleoperated.

The last three. A result that is too impressive: human-level performance that comes out of nowhere, with no earlier work, is suspect, and nothing tells you whether you are watching the first take or the five hundredth. A pace that is too fast: a complex task takes time to sense and plan, so a long multipart task carried out without a pause raises questions. Movements that are too human: small purposeless motions, such as fingers moving with no relation to the grasp, are usually a sign of a human in the loop. Ackerman admits a surprising autonomous demo is possible, “improbable, perhaps, but not impossible.”

He also asks that the context be in the video itself, not in the description, the subtitles or a link, since videos get shared and remixed. It should say whether it is sped up or slowed down, whether several experiments are edited into one, whether power, compute or localization are external, and how the robot is controlled: human in the loop, human supervised, scripted actions, partial or full autonomy. These recommendations come from an article he wrote for the IEEE Robotics and Automation Society.

## Optimus: the robot’s own word proves nothing

After Tesla’s We, Robot event in October 2024, the question arose whether the [Optimus](/robots/tesla-optimus) robots on show were controlled by humans. Entrepreneur (article by Sherin Shibu, October 14, 2024) quotes an attendee who asked an Optimus whether it was remote-controlled: the robot reportedly answered that it was assisted by a human that day, not yet fully autonomous. The same article reports that Electrek saw a Tesla employee with a signaling device near each robot, and notes that Tesla had not responded to these reports at the time of publication.

The opposite case exists. In a TechCrunch snippet (March 17, 2025), Maxwell Zeff reports that a humanoid from the startup IntBot, at the information booth of Nvidia’s GTC conference, says no human controls it. Zeff notes that teleoperated demos have happened before and, in the post that goes with the video, that it is unclear whether the claim is true; he recalls that Optimus turned out to be remotely controlled. What the robot, or its maker, says is not proof.

## 1X’s NEO: remote piloting is part of the product

1X says so on its order page, read on October 30, 2026. [NEO](/robots/1x-neo) “arrives with basic autonomy for early owners” and will gain capabilities over time. For a complex task it does not know, a 1X expert can remotely supervise its actions at scheduled times (“Scheduled Expert Mode”), and its owner can pilot it from anywhere with the mobile app and a virtual reality device. The page lists $20,000 ($200 refundable deposit) or $499 a month, with deliveries announced only in the United States, starting in 2026. It also lists dual 8.85-megapixel 90 Hz stereo fisheye cameras and four microphones.

BGR (Elias Nash, February 2, 2026) spells out what that means at home: anything beyond the most basic tasks will require a human at the controls remotely. Sessions are scheduled in the app; a 1X employee then operates from the company’s California headquarters and can see through NEO’s eyes and hear through its ears. The article cites an option to opt out of certain collection and sharing of the live video feed, notes that the owner can end a session at any time and that the robot’s ears light up blue when a teleoperator has permission. 1X’s chief executive, Bernt Børnich, wants a fully automated NEO one day, but admits, again according to BGR, that it may only come from gathering large amounts of data with early users’ permission.

A Wall Street Journal test, as quoted by John Gruber on Daring Fireball on October 31, 2025, shows how far there is to go: the NEO tried by journalist Joanna Stern was piloted by an operator in a virtual reality headset, nearly fell over while closing a dishwasher and took two minutes to fold a shirt. The Daring Fireball note describes NEO, as it stands, as entirely remote-controlled by 1X employees, which contrasts with the “basic autonomy” the company announces. We have read no independent measurement of how much work NEO does on its own.

## Nucleus: a two-hour tape that gives its share of teleoperation

According to AI Weekly (October 2, 2026), which draws on Interesting Engineering, Nucleus posted on X on October 1 nearly two hours of uncut factory footage: a humanoid picks parts, loads shelves, handles products and moves a cart. Its chief executive, Melvin Schwarz, puts it at roughly 60% autonomous and 40% teleoperated, and the hesitations and human interventions were left in. The company says it regularly runs shifts of four to six hours and treats teleoperation corrections as training signals.

That is the level of transparency one can ask for, and it has limits. AI Weekly notes that the 60% could be measured by time, by tasks completed or by number of interventions, and that Nucleus has not said which. The tape was produced by the company, on a shift it chose, with no independent verification. According to Interesting Engineering, the figure, which the company did not explain, describes this clip rather than Nucleus’s capabilities in general.

## An AI in control, but how?

On October 22, 2026, at the World Physical AI Conference in Hangzhou, Unitree founder Wang Xingxing showed a [G1](/robots/unitree-g1) copying a painting. According to his presentation, the robot was controlled by GPT-6, OpenAI’s model, with no pretraining or pre-planned algorithm. The Chinese site MyDrivers (October 23, 2026) relays these remarks without saying it checked them. It does not say whether the demo was live or recorded, how the model was connected to the robot, or how much human intervention there was. A claim of this kind fits no category until the method is published.

## Questions to ask before believing a robot works alone

Who is in control, and for what share? Nucleus gives a figure, 1X describes a mode, and Ackerman asks that every video do so. Is the footage continuous, at real speed, or edited? Does the task hold up when objects, layout or conditions change? Humanoid.guide sees that as the weak point: system integration under variation. A robot that folds laundry in one setup does not prove it can work unsupervised when objects, layouts and failures change. The site considers more convincing repeated manipulation under novel perturbations, peer-reviewed evaluations of vision-language-action models and real deployments without persistent teleoperation.

For a humanoid meant for the home, one more question applies: who can see inside my house, and when? At 1X the answer runs through scheduled sessions and an option to opt out of some data sharing, which BGR describes and which are worth reading closely before ordering. For other makers’ demos, the six clues remain.

## Sources

- [IEEE Spectrum, article by Evan Ackerman of January 23, 2024, read on October 30, 2026](https://spectrum.ieee.org/robot-teleoperation-autonomy)
- [Humanoid.guide, article of June 20, 2026 (updated September 21), read on October 30, 2026](https://humanoid.guide/robotics-summit-report-flags-humanoid-teleoperation-limits/)
- [Entrepreneur, article by Sherin Shibu of October 14, 2024, read on October 30, 2026](https://entrepreneur.com/business-news/elon-musk-reveals-tesla-optimus-robots-remotely-controlled/481252)
- [TechCrunch, snippet by Maxwell Zeff of March 17, 2025, read on October 30, 2026](https://techcrunch.com/snippet/2982030/this-well-dressed-robot-tells-me-it-is-autonomous)
- [1x.tech, order page read on October 30, 2026](https://www.1x.tech/order)
- [BGR, article by Elias Nash of February 2, 2026, read on October 30, 2026](https://bgr.com/2086758/future-robot-companion-human-stranger-control)
- [Daring Fireball, note by John Gruber of October 31, 2025 on Joanna Stern’s test (Wall Street Journal), read on October 30, 2026](https://daringfireball.net/linked/2025/10/31/stern-1x-neo)
- [AI Weekly, alert of October 2, 2026 on the Nucleus tape (citing Interesting Engineering), read on October 30, 2026](https://aiweekly.co/alerts/nucleus-posts-2-hour-humanoid-factory-tape-60-autonomous)
- [MyDrivers, article of October 23, 2026 on Unitree’s Hangzhou demo, read on October 30, 2026](https://news.mydrivers.com/1/1159/1159979.htm)

---

HTML version: https://vuntum.com/en/news/autonomous-or-teleoperated-how-to-read-a-humanoid-robot-demo

Factual data: CC BY-NC 4.0. Attribution: “Source: Vuntum (https://vuntum.com)” with a link to the page quoted. Editorial text: all rights reserved, short quotation with attribution.
