Humanoid AI Just Had Its Craziest Week Yet

Humanoid AI robots hit a turning point this week.

China showed robots that look ready for real homes and real factories. Russia tried to enter the race and watched its first AI humanoid collapse on stage. On top of that, top CEOs started accusing each other of faking progress, in public, in front of millions of people.

In one short stretch of days we saw:

• MindOn pushing the Unitree G1 into a real home environment.
• Unitree launching its wheeled G1D for industrial speed.
• UBTECH shipping hundreds of Walker S2 humanoids to factories.
• Russia’s AIdol robot falling during its big reveal.
• Brett Adcock and Agility Robotics trading shots online about what is real.

Under all the drama sit the real questions: battery life, safety, reliability, and cost. For humanoid AI to matter, robots need to handle full shifts, work around people, and still be affordable.

Let’s break down what actually happened and why it matters for anyone who cares about where AI and robotics are going next.

MindOn Puts Unitree G1 Into a Real Home

MindOn took the Unitree G1, a humanoid that already looked solid on paper, and dropped it into what looks like a real, lived‑in home. That is where things got interesting.

The G1 platform in simple terms

The Unitree G1 is a compact humanoid platform with hardware that already had people watching. On paper, it looks like this:

1. 23 degrees of freedom for flexible body movement.
2. Depth cameras to understand 3D structure in front of it.
3. 3D LiDAR for mapping and navigation.
4. Hybrid force‑position control so it can push, pull, and grip with the right amount of force.
5. A mechanical structure tuned for agile, dynamic motion.

That spec sheet hinted at a capable humanoid. What nobody expected was how natural the robot would look in a full household workflow.

For a closer look at how this project was reported, you can check a writeup that covers MindOn’s viral Unitree G1 home demo, along with the official G1 household task footage on YouTube.

The viral home demo: one continuous workflow

Instead of staging short, perfect clips, MindOn showed the G1 in what felt like one long, continuous work session.

Some key moments:

• Curtains and windows: The G1 walks up to a window, lines up its hands, and opens the curtains in a smooth, continuous motion. It looks more like a person who has done the same movement a thousand times than a robot running a stiff script.
• Plant care: The robot checks plants, adjusts its grip so it does not crush leaves, and waters them without spilling. That shows not just vision, but decent force control too.
• Office‑style tasks: It moves light packages while staying balanced and still planning every step. No obvious pauses, no long thinking moments.
• Textiles and bedding: This is the hard part for most robots. The G1 pulls sheets off a bed, shakes them out, spreads them, and walks around furniture without looking confused. Soft materials usually trip robots up, so seeing them as part of a fluid routine stands out.
• Interactions with kids: In the demo, kids move near the robot, and the motion stays controlled and predictable. That hints at a safety layer tuned for gentle interaction, not just raw power.

The important detail is not a single trick. It is how all of these actions blend into one chain of tasks that feels like one job, not a highlight reel.

How MindOn likely trained it

MindOn has not shared a full technical breakdown of its training pipeline yet. But the behavior suggests a focus on environmental generalization, not just memorized moves.

Instead of recording a perfect path for “open this exact curtain in this exact way,” the robot seems trained to handle variation:

• Slightly different curtain positions.
• Plants with different shapes.
• Furniture layout that is not identical every time.
• People moving around in the space.

In a real house, nothing stays the same. If a humanoid pauses every time a chair is off by two inches, it becomes useless. MindOn’s demo hints that they are training on that messy, real‑world variability, not just on stage‑ready motions.

Why this matters for real homes and small businesses

This kind of demo pulls the conversation toward what actually matters for everyday use:

• Battery life that covers a full work session, not 10 minutes.
• Reliability over hundreds or thousands of repeats of the same task.
• Safety around kids, elderly people, and pets.
• Maintenance and repair that do not cost a fortune.
• Price that a normal home or small business can handle.

MindOn has not answered all those questions yet, but this is one of the first times a small humanoid platform looks like it belongs in a home, not just in a lab clip. For anyone following AI robotics, this feels like a step closer to practical use, not just marketing.

Unitree’s G1D: A Wheeled Humanoid Built for Speed

While MindOn was making the G1 look at home in a house, Unitree announced something different: the G1D, a wheeled humanoid tuned for nonstop work in industrial settings.

This robot is not trying to mimic a human walk. It is built to move fast, carry out repetitive tasks, and run for full shifts in places like warehouses and retail backrooms.

Two versions for different jobs

The G1D comes in two main versions, both humanoid in upper body, but with different bases and roles.

Feature	Standard G1D (Stationary)	Mobile G1D (Flagship)
Mobility	Fixed base	Wheeled base, differential drive
Speed	N/A	Up to 1.5 meters per second
Height	49.5 to 66 inches	49.5 to 66 inches
Degrees of freedom	17	19 (not counting end effectors)
Weight	Up to ~80 kg (176 lb)	Up to ~80 kg (176 lb)

At 1.5 meters per second, the mobile version can actually keep up with many real logistics workflows. That speed, plus a human‑like upper body, lets it blend into environments built for people but still move with the efficiency of a small vehicle.

Hardware and sensing: built for repetitive work

Unitree gave both G1D versions the same arms and a strong waist:

• Arms:
  • 7 degrees of freedom each.
  • Can lift about 3 kg (6.6 lb) per arm.
• Waist joint:
  • Rotates along the Z‑axis up to 155 degrees.
  • Moves along the Y‑axis from about −2.5 to 135 degrees.
  • Lets the robot reach up to around 2 meters in height.

For vision, the G1D uses:

• A high‑definition binocular camera on the head for depth and object detection.
• Two wrist‑mounted cameras that give close‑up views of hands and objects.

This multi‑angle setup helps the robot keep its grip aligned, inspect items, and avoid drifting off target after hundreds of repeated motions.

End effectors are modular, so the same base robot can work in very different tasks:

• Two‑finger grippers.
• Three‑finger grippers, with or without touch sensors.
• A five‑finger dexterous hand for more delicate work.

AI compute, battery life, and software stack

The flagship G1D uses an Nvidia Jetson Orin NX that delivers around 100 TOPS of AI compute on board. That is plenty for local path planning, perception, and task logic without needing a constant cloud link.

Battery life reaches up to 6 hours of autonomous operation, enough to cover long work blocks or a good portion of a warehouse shift.

Unitree also released a software platform around the robot. It handles:

• Data collection and labeling.
• Task management.
• Simulation environments for training.
• Distributed training runs.
• Deployment of trained models onto real robots.

The idea is simple: the G1D should not only run tasks, it should also help companies build their own AI pipelines based on real‑world data. For a taste of how Chinese robots are shaking up logistics and manufacturing, you can see how creators on social media react to UBTECH and Unitree’s humanoid systems in action.

Russia’s AIdol: From Rocky Theme To Floor

While Chinese companies were showing polished demos and large deployments, Russia attempted its first splash with an AI humanoid called AIdol. It did not go as planned.

The debut that turned into a meme

On November 10 in Moscow, staff guided AIdol onto the stage to the Rocky theme song. Moments later, the robot lost balance, fell forward, and left parts scattered on the floor.

Staff rushed in, pulled up a screen to hide the robot, and dragged it off stage. The clip spread quickly on Russian tech forums and then across social media. The event was supposed to mark Russia’s serious entry into humanoid AI. Instead, it became a punchline.

You can see how outlets covered the moment in reports like Business Insider’s story on AIdol’s collapse and in posts from robotics news feeds such as RoboHub’s summary on X.

What AIdol is supposed to be

Under the embarrassment, there is still a real project. AIdol’s developers say they are building:

• A humanoid that uses embodied AI for movement, object handling, and human interaction.
• A system powered by a 48 V battery that can run up to 6 hours.
• A platform with 19 servo motors.
• A face covered in silicone skin that can show more than a dozen basic emotions plus hundreds of micro‑expressions.

They also say that about 77% of the components are Russian made, with a goal to reach about 93% domestic content.

During the event, CEO Vladimir Vitukin claimed the robot can smile, think, and look surprised in a way that is similar to a person.

Aftermath and criticism

The company blamed the fall on a calibration issue and said AIdol is still in testing. Critics, though, focused on two things:

• The robot’s instability.
• The decision to put it on stage before it could walk reliably.

In a week where Chinese robots were working in real settings, the AIdol incident highlighted how hard the humanoid race is, and how public failure can be when AI hardware is involved.

If you want more context on how media framed the event, sites like Humanoid Press, which tracks AI robotics news, put the AIdol stumble alongside other global humanoid updates.

UBTECH Walker S2: Hundreds Of Humanoids Ship To Factories

While AIdol was trending for the wrong reasons, UBTECH announced something the industry has been waiting for: large‑scale deployment of humanoid workers that are not just prototypes.

A real mass rollout, not just a slide

UBTECH confirmed that hundreds of Walker S2 humanoid robots have already shipped to real industrial sites. Production ramped up in mid‑November, and the first batches have reached partner factories.

According to the company:

• Demand for humanoids this year reached about 113 million dollars.
• Some of the major deals include:
  • Roughly 35.3 million dollars from a well‑known Chinese firm in September.
  • About 22.45 million dollars in Sichuan.
  • Around 17.8 million dollars in Guangxi.
  • More than 14.1 million dollars from Miato in Hubei.
• UBTECH expects to ship around 500 Walker S2 units by the end of December, and says it is on track.

Coverage like Interesting Engineering’s report on UBTECH’s first large delivery of humanoid workers and Humanoids Daily’s piece on UBTECH’s “robot army” rollout gives a sense of how big this step looks from the outside.

Who is buying these robots?

Automakers are leading the charge. Companies adding Walker S2 units to their operations include:

• BYD
• Geely Auto
• Volkswagen
• Dongfeng
• Li Auto
• Foxconn

Early trials show strong performance in both factories and warehouses, especially in logistics, line support, and other repetitive tasks that still need the reach and shape of a human body.

For fans of AI and robotics, some tech creators talk about this as a “robot army” moment. You can see that tone in short clips, like this Instagram reel reacting to UBTECH’s humanoid rollout and Figure AI’s response.

The technical edge: self‑swapping batteries

One standout feature makes Walker S2 especially interesting for industry: automatic battery swapping.

The robot can:

• Remove its own battery pack.
• Slot in a fresh one.
• Get back to work within minutes, without human help.

This almost removes downtime tied to charging. In a factory where time really is money, that alone makes humanoids far more attractive.

Walker S2 is also built with:

• A tall, stable body that can handle heavy objects.
• Finger control fine enough for more delicate operations.

UBTECH says humanoids now make up about 30% of its sales, up from 10% the previous year. That shift suggests this is not just a PR push. Customers are actually paying for these AI robots.

Money and markets

The financial side backs up the story:

• First half of 2025 revenue: about 87.7 million dollars, up 27.5% year over year.
• Gross profit: around 30.6 million dollars, up 17.3%.
• Losses narrowed by roughly 62.1 million dollars.
• The stock climbed more than 150% this year to about 133 Hong Kong dollars.
• Analysts at Citi and JP Morgan still rate the stock as a buy, with targets above 170 Hong Kong dollars.

UBTECH was the first robotics company listed on the Hong Kong exchange back in 2023. Now it looks like it might be one of the first to prove that large‑scale humanoid deployment is not just hype.

For a wider view on how AI hardware like this might reshape computing and work, you can line it up with deeper technical shifts in areas such as thermodynamic computing and AI architectures.

The Online Feud: Brett Adcock vs Agility Robotics

The week was not only about hardware. It also brought one of the loudest public clashes between AI robotics CEOs so far.

Calling out “fake” robots

Brett Adcock, CEO of Figure, commented on a video that claimed to show UBTECH delivering hundreds of humanoids. He suggested that some of the robots in the footage looked fake or staged, raising doubts about how many units were actually working.

This tapped into a long‑running tension in robotics:

• Companies post polished videos that look amazing.
• Real‑world performance is often far messier and harder to judge.

Adcock has often talked about this gap. In interviews like the Forbes India profile on Figure AI and Brett Adcock, he frames humanoid AI as a long, hard path, not just a marketing story.

The BMW claim and the lemon water joke

Things escalated when Adcock posted that a Figure humanoid had been working on BMW’s production line for five straight months.

Agility Robotics stepped in with a sarcastic comment, comparing his claim to someone saying they invented lemon water. It was a way of saying, “You are not the only one doing real work.”

Adcock responded by saying Agility would be “bankrupt in under a year.” That single line traveled fast across X and other platforms.

Other industry voices tried to cool things down. A VP from 1X reminded everyone that kindness matters more than aggression. Agility replied again with a Ted Lasso GIF, said they would check back next November, and still added that they appreciate the work of all teams pushing humanoids forward.

Adcock closed with a Sopranos meme captioned, “Next time you come in, you come heavy or not at all.”

For outsiders, it looked like drama. For people watching AI robotics closely, it showed how intense the competition is getting as billions of dollars and the future of work sit in the balance.

Also Read: If This Works, AGI Arrives Early: Inside Extropic’s Thermodynamic Computing Breakthrough

What This Wild Week Tells Us About AI Robots

Across one short span of days, we saw almost the full spectrum of humanoid AI:

• A small humanoid quietly doing complex chores in a real home.
• A wheeled factory robot designed for pure speed and uptime.
• A national project crashing to the floor in front of cameras.
• Hundreds of humanoids rolling into real factories.
• CEOs publicly accusing each other of faking progress.

The big picture is simple: AI robots are moving out of the lab and into real life, but not everyone is moving at the same pace, and not every demo reflects reality.

As you watch the next wave of humanoid clips, it is worth asking:

• Is this a one‑off stunt, or part of a repeatable workflow?
• Can this robot run for hours, not minutes?
• Is someone actually paying money for this deployment?

If you care about AI, robotics, or the future of work, this is a good time to pay close attention. The line between sci‑fi demo and daily tool is getting thinner, and the next “unreal” week in humanoid robotics is probably not far away.