China Can Build Kung Fu Robots — So Why Can’t They Work in a Factory?
China can build humanoid robots that sprint, dance, fight, and perform choreographed routines.
Some can move fast enough to make headlines.
Some look almost ready for science fiction.
And at the 2026 World Humanoid Robot Games in Beijing, one Chinese robot even ran 100 meters in 8.64 seconds — faster than Usain Bolt’s human world record.
That sounds like the future has arrived.
But there’s a problem.
Ask those same robots to do something much less glamorous — like reliably complete an eight-hour factory shift — and the story changes fast.
That gap between what robots can demonstrate and what robots can actually do economically, repeatedly, and safely may be the most important robotics story of 2026.
And China is now spending heavily to close it.
The Demo Problem
Robots look amazing in short videos.
They can:
Kick
Punch
Dance
Run
Jump
Perform synchronized routines
Carry out carefully scripted tasks
Those demonstrations are useful.
They prove balance.
They prove actuator performance.
They prove mechanical control.
They show that humanoid hardware is advancing quickly.
But a factory does not care whether a robot can perform kung fu.
A factory cares whether a robot can:
Repeat the same task thousands of times
Adapt when objects move slightly
Avoid damaging products
Work safely around humans
Recover from mistakes
Operate for long periods
Justify its cost
That is a much harder problem.
Why Factories Are So Unforgiving
A viral demo lasts a few seconds.
A factory shift lasts hours.
That difference changes everything.
Imagine a robot picking up parts from a conveyor belt.
In a demo, the part is exactly where engineers expect it to be.
In a real factory:
The part may be rotated.
The lighting may change.
Another worker may move nearby.
The conveyor may stop.
A box may be slightly damaged.
A sensor may become dirty.
The robot has to understand what changed and recover.
That ability — reliable adaptation — is where humanoid robots still struggle.
Reuters reported that despite rapid hardware progress, Chinese humanoids remain much less useful in real industrial environments than their flashy public demonstrations suggest.
Conventional Robots Are Still Better at Many Factory Jobs
Here’s the uncomfortable truth for humanoid robotics.
Factories already have robots.
And many of them are extremely good.
Industrial robotic arms can:
Weld
Paint
Assemble
Pick and place
Move materials
Perform repetitive precision tasks
They are fast.
They are proven.
They are reliable.
They do not need to look human.
So if a humanoid robot costs more, moves slower, and makes more mistakes, factory managers have a very simple question:
Why should we replace the machine that already works?
That is one reason humanoid robots face a much higher bar than the public often realizes.
So Why Build Humanoids at All?
Because the world was built for humans.
Doors.
Stairs.
Shelves.
Tools.
Vehicles.
Workstations.
Warehouses.
Factories.
Almost everything around us was designed for a body with:
Two arms.
Two legs.
Hands.
Human height.
That creates the long-term argument for humanoids.
Instead of rebuilding every workplace around specialized robots, companies could theoretically build one general-purpose machine that uses the same environment humans already use.
That is enormously attractive.
If it works.
China Is Betting That Scale Will Solve the Problem
China has used this strategy before.
Electric vehicles.
Solar panels.
Batteries.
Industrial manufacturing.
The pattern is familiar:
Invest early.
Build supply chains.
Create many competitors.
Push production volumes up.
Drive costs down.
Accept inefficiency in the beginning.
Then try to dominate the market once the technology matures.
Humanoid robots may be the next industry where China tries the same playbook.
Reuters reported that more than 150 Chinese companies are competing in the humanoid sector, while government spending exceeded $230 million in the first half of 2026 alone.
That is not a small experiment.
It is an industrial strategy.
Unitree Shows Both the Hype and the Reality
Unitree is one of the biggest names in Chinese robotics.
The company became famous for robot dogs and later moved aggressively into humanoids.
Its robots are visually impressive.
They are fast.
They are relatively affordable compared with many Western competitors.
And the company has become a symbol of China's robotics ambitions.
Unitree CEO Wang Xingxing has said robots may eventually reach a “ChatGPT moment,” where software intelligence improves so dramatically that robots become much more useful. He has suggested that robots could one day perform a large share of household tasks, while also acknowledging that the necessary software leap is still years away.
That tension is important.
The hardware looks futuristic.
The intelligence is not there yet.
The Real Bottleneck Is the Brain
For years, robotics was mostly a hardware problem.
Motors.
Actuators.
Batteries.
Sensors.
Mechanical design.
Those things still matter.
But now the hardest problem is increasingly the software.
A robot working in the real world needs to understand:
What am I looking at?
What changed?
What should I do next?
Is this safe?
Did I make a mistake?
How do I recover?
That is embodied intelligence.
And it requires much more than a language model.
A chatbot can give a wrong answer and apologize.
A robot can drop a heavy object.
That difference matters.
Robots Need a Different Kind of Data
Large language models learned from enormous amounts of text.
Robots need something much harder.
They need physical-world data.
Examples of:
Grasping objects
Walking on different surfaces
Opening doors
Using tools
Recovering from slips
Moving around humans
Handling unpredictable environments
And that data is much more expensive to collect.
A trillion words can be scraped from the internet.
A trillion examples of robots manipulating physical objects cannot.
That is why many robotics companies now talk about world models, simulation, teleoperation, and data collection from human workers.
The real race may not be who builds the coolest robot body.
It may be who collects the best physical-world training data.
This Is Why Robot Sports Actually Matter
Robot races and competitions can look like publicity stunts.
But they are also testing grounds.
At the 2026 World Humanoid Robot Games, more than 2,000 robots from 666 teams competed across dozens of events.
That creates opportunities to test:
Balance.
Coordination.
Recovery.
Locomotion.
Control.
Hardware durability.
Performance under pressure.
So even if a 100-meter robot sprint does not directly help a warehouse tomorrow, the technologies developed for those competitions may eventually improve real-world systems.
The problem is assuming that athletic performance automatically means industrial usefulness.
It does not.
The Factory Test Is Much Harder Than the Olympics
A robot sprinting 100 meters has one objective:
Run forward.
A factory robot may have to:
Recognize an object.
Pick it up.
Rotate it.
Inspect it.
Place it precisely.
Avoid another worker.
Notice a defect.
React to a new instruction.
Do all of that repeatedly.
Then do it again tomorrow.
That is a completely different level of intelligence.
It is less exciting to watch.
But economically, it matters much more.
Why Companies Still Want Humanoids
Despite all these problems, the business case is not crazy.
China is aging.
Labor shortages are becoming more important.
Manufacturers want more automation.
Warehouses want flexible labor.
Retail and service industries face staffing challenges.
A robot that could do multiple jobs could be extremely valuable.
The key word is:
Could.
Today, many humanoids are still better at proving what may eventually be possible than delivering an immediate return on investment.
The Danger: Too Much Money, Too Early
There is another risk.
When governments, investors, and companies all rush into the same sector, the industry can become overheated.
Too many companies.
Too many factories.
Too much capacity.
Too little real demand.
Reuters reported concerns that China's humanoid boom may be creating overcapacity before the technology has proven a strong commercial market.
That sounds familiar.
China has experienced brutal price competition in EVs and solar.
Humanoid robotics could follow the same path.
A lot of companies may not survive.
That Might Actually Help the Industry
Industry shakeouts are painful.
But they can also accelerate progress.
Weak companies disappear.
Prices fall.
Suppliers improve.
Standards emerge.
The strongest companies gain scale.
That happened in:
Smartphones.
EVs.
Solar.
Drones.
It could happen again in robotics.
So a future collapse in the number of humanoid startups would not necessarily mean the technology failed.
It may mean the industry is maturing.
The Better Question Isn't “Will Robots Take Our Jobs?”
That question is too simple.
The more useful question is:
Which jobs can robots do reliably enough to justify their cost?
That changes the discussion.
A humanoid probably does not need to replace an entire person.
It could take over specific tasks.
For example:
Moving boxes
Restocking shelves
Night inspections
Handling dangerous materials
Repetitive warehouse work
Basic logistics
Simple machine tending
If a robot becomes good at one narrow task, it can create value long before it becomes a general-purpose worker.
Where Humanoids May Work First
The first successful deployments probably will not look like science fiction.
They may look boring.
And boring is good.
The strongest early use cases are likely to be environments that are:
Structured.
Repetitive.
Predictable.
Controlled.
That could include:
Warehouses
Moving items between fixed locations.
Manufacturing
Loading and unloading machines.
Logistics centers
Carrying containers or sorting materials.
Retail
Restocking or simple back-room tasks.
Inspection
Checking equipment in predictable environments.
Hazardous work
Tasks where sending a human is risky.
The more structured the environment, the easier the robotics problem becomes.
Where Humanoids Will Struggle Longer
Unstructured environments are much harder.
Homes.
Restaurants.
Hospitals.
Construction sites.
Crowded public spaces.
Why?
Because humans are unpredictable.
Objects move.
Rooms change.
People interrupt.
Surfaces vary.
Instructions are vague.
A household robot may need to understand:
“Can you clean this up?”
That sounds easy to a human.
For a robot, it can involve dozens of decisions.
What is “this”?
What counts as clean?
Where do the objects belong?
Which items are fragile?
What can be thrown away?
That is why household robots may take longer than viral demos make it seem.
What I Would Look for Before Believing the Next Robot Demo
This is the part most people should pay attention to.
The next time you see an incredible humanoid robot video, ask five questions.
1. Is the task scripted?
If engineers carefully programmed every movement, that tells you less about intelligence.
2. How many attempts did it take?
A perfect 20-second clip may have required 50 failed attempts.
3. Can it handle changes?
Move the object.
Change the lighting.
Interrupt the task.
Does the robot still work?
4. Can it do the task for hours?
Reliability matters more than one perfect demonstration.
5. Is it economically better than existing automation?
This may be the most important question.
A robot can be technologically impressive and commercially useless at the same time.
The Metric That Matters: Useful Work Per Dollar
Robot companies love showing:
Speed.
Strength.
Agility.
Dexterity.
Those are important.
But businesses care about something else.
How much useful work does the robot produce for the money?
Imagine two machines.
Robot A looks human, walks around, and costs $100,000.
Robot B is an ugly industrial arm that never moves from one spot and costs half as much.
If Robot B performs the required task faster and more reliably, the factory will buy Robot B.
That is why humanoids have to prove more than technical capability.
They have to prove economic value.
China Still Has a Major Advantage
Even if the robots are not ready, China has one huge advantage:
Manufacturing scale.
China already has:
Battery supply chains.
Motor manufacturers.
Sensors.
Electronics.
Precision machining.
Factories.
Robotics suppliers.
EV manufacturing infrastructure.
That ecosystem can make iteration much faster and cheaper.
A robotics startup in another country may have to import dozens of components.
A Chinese company may find many of those suppliers nearby.
That matters.
But America Still Has Strengths Too
The U.S. remains strong in:
AI models.
Software.
Advanced chips.
Research.
Capital.
Frontier machine learning.
Companies like Tesla, Figure, Agility Robotics, and others are also pushing humanoid development.
So the global race may end up looking something like this:
China: manufacturing scale and cost
United States: AI software and frontier models
The winner may be whoever combines both.
The Robot Race Is Really an AI Race
The most interesting thing about humanoid robots is that robotics and AI are becoming the same industry.
A robot without good intelligence is basically a complicated machine.
An AI model without a body can only affect the physical world indirectly.
Put them together and you get:
Embodied AI.
That is why companies like Nvidia, Tesla, Google, OpenAI-linked startups, Chinese robotics firms, and industrial giants are all paying attention.
The prize is not just a robot.
It is a machine that can understand instructions and act in the physical world.
That could be one of the biggest technology markets ever created.
So When Will Humanoid Robots Actually Be Useful?
Probably gradually.
Not suddenly.
There may not be one dramatic moment where everyone wakes up and robots have taken over.
Instead, adoption may look like this:
First:
One robot handles one factory task.
Then:
A fleet handles several tasks.
Then:
Software improves.
Costs fall.
Reliability rises.
More industries adopt them.
Eventually, robots become normal.
That is how many technologies spread.
Slowly.
Then suddenly.
My Take: Ignore the Kung Fu
The kung fu is fun.
The races are impressive.
The dancing is great marketing.
But none of those things tell us whether humanoid robots are ready to transform the economy.
The real signs to watch are much less exciting:
Factory deployment numbers.
Hours worked without intervention.
Failure rates.
Cost per task.
Real customer renewals.
Repeat orders.
Safety incidents.
Those metrics will tell us when humanoid robots have moved from entertainment to infrastructure.
The Bottom Line
China's humanoid robot industry is moving incredibly fast.
The hardware is improving.
Government support is enormous.
Manufacturing scale is real.
Companies like Unitree and UBTech are pushing the field forward.
But the robots themselves are still not as intelligent or reliable as the most impressive demonstrations make them appear.
That does not mean the humanoid boom is fake.
It means the industry is early.
Very early.
And the next phase will be much harder than teaching robots to dance.
The real test is whether they can show up Monday morning...
work all day...
make fewer mistakes than a human...
and still make economic sense.
When that happens, the robot revolution will no longer need viral videos to prove it's real.
Related Reading
The Person Looking at You Might Be Recording You — AI Glasses Are Becoming a Privacy Nightmare
Comments
Post a Comment