Humanoid robots are no longer just lab projects or sci‑fi props. XPeng Iron humanoid robot is already turning screws and moving parts on real car assembly lines in China. By pairing this robot with its Kunpeng super electric system and in‑house AI chips, XPeng is trying to build one shared brain and power system for cars, robots, robotaxis, and even flying vehicles.
If this strategy works, it could speed up factory automation, give XPeng an edge in the global EV race, and change how people work with machines on the factory floor.
How XPeng Iron humanoid robot works on the factory floor
XPeng built Iron to roughly match human size and reach so it can use tools, handle parts, and move in spaces designed for people, as shown in its 2024 AI Day reveal of the Iron humanoid robot and Kunpeng system. Unlike classic industrial robot arms fixed to one spot, Iron is made to walk through the plant and share work cells with human staff.
Body design and movement capabilities
The first public version of Iron is about 178 cm tall and weighs around 70 kg, close to an average adult man. It has more than 60 joints and roughly 200 degrees of freedom, meaning its joints can move in many independent ways so it can bend, twist, and reach in a human‑like fashion.
Iron’s legs include hip, knee, and ankle joints that give it a stable bipedal walk. Its torso has a flexible spine that lets it lean and twist to reach into tight spaces. The arms include shoulders, elbows, and wrists with enough travel to lift tools, move boxes, and work at different heights. Iron’s hands are small and simple compared with a human hand, but its dual‑finger grippers can still hold parts, tools, and boxes with a range of grips.
Later “next‑gen IRON” prototypes shown in 2025, including the version presented at XPeng’s 2025 AI Day with more than 60 joints and around 200 degrees of freedom, use an internal skeleton and artificial muscle system that makes movements smoother and more lifelike. For readers, the key point is that each new version is moving closer to human motion while keeping a machine‑like endurance.
AI brain, vision system, and training
Instead of buying off‑the‑shelf chips, XPeng developed the Turing AI chip that powers both its cars and Iron. The chip is built to run large AI models at very high speed, reaching thousands of trillions of operations per second. That level of compute is what lets the robot process camera images, understand instructions, and plan motions almost in real time.
Iron’s “eyes” are a 720‑degree vision system: a set of cameras arranged so the robot can see all around itself. XPeng relies on a pure‑vision approach similar to the one it uses in its driver‑assist and robotaxi systems. Instead of depending on laser scanners or detailed pre‑made maps, its models learn to parse raw camera images into depth, objects, and free space.
On top of that, XPeng is building a Vision–Language–Action (VLA) model that links what Iron sees, what it is told to do, and the movements it chooses. The company says the same type of model helps its robotaxis convert descriptions of the scene into driving actions. Training these models on driving and factory footage lets Iron learn from millions of real‑world examples, not just from simulated data.
Current roles for Iron in XPeng factories
XPeng has already shown Iron doing real assembly work. In videos from its factories and trade shows, Iron picks up parts from bins, holds tools such as power screwdrivers, and installs fasteners on car bodies. It cooperates with mobile robots that bring materials to the line, then hands off finished parts for the next step.
These early tasks are narrow but important. For XPeng, a humanoid that can walk into an existing workstation and handle screw tightening, parts sorting, or material handling could cut the need to redesign lines around fixed robot arms. Over time, the same basic robot body could be re‑trained or re‑tasked as products and workflows change.
XPeng Iron and the global race for humanoid robots
XPeng is not alone. Tesla’s Optimus, Figure’s Figure 01, and several Chinese startups are all racing to show factory jobs done by walking robots. Iron enters this race with a few unique advantages: a direct link to a carmaker’s large factory network, and a shared chip and software stack with XPeng’s vehicles.
How Iron compares with other factory humanoids
On paper, Iron’s basic dimensions are similar to Tesla’s Optimus and other human‑scale robots: human height, tens of kilograms in weight, and dozens of joints. Its compute budget, measured in the low thousands of TOPS, is also in line with other high‑end platforms.
Where XPeng tries to stand out is in its use of the same Turing AI chip and pure‑vision models across cars, robots, and robotaxis. Rather than design different computers for each product, XPeng is betting that one flexible AI stack can be trained once and deployed many times. That could lower costs and shorten development cycles if it works as planned.
Why carmakers want humanoid robots in their own plants
For car companies, labor is a major cost and bottleneck, especially for tasks that are repetitive, physically hard, or slightly dangerous but still need human‑like dexterity. Classic industrial robots handle welding and painting well, but many final‑assembly jobs still fall to people.
Humanoid robots promise to take on some of these tasks without tearing up and rebuilding the plant. A robot with human‑like reach, ability to climb stairs, and use hand tools can in theory walk into a workstation designed for a person and do a similar job, with fewer changes to the line. If a carmaker owns the robot platform, it also gains more control over data, safety rules, and long‑term costs.
XPeng’s move fits a wider pattern that metameha has covered in other contexts, such as how an AI system enabled a 20‑minute interactive exchange with a humpback whale. In each case, companies try to turn AI models trained on large streams of sensor data into practical systems that act in the physical world.
Limits and open problems for humanoid robots
Despite the buzz, humanoid robots still face clear limits. Battery energy is constrained, especially when a robot walks and carries loads for long shifts. Even with efficient motors and high‑voltage batteries, it is hard to match human endurance.
Manipulation skills also lag behind humans. Iron’s hands are far less capable than a real hand, so many tasks still need custom grippers or human workers. Safety is another concern. A 70 kg robot moving near people must detect contact, respond quickly, and obey strict rules so it does not injure co‑workers.
Finally, cost remains a barrier. Until robots like Iron can be built, maintained, and updated cheaply enough, they will be limited to pilot projects and high‑value tasks. XPeng’s hope is that sharing chips and software with high‑volume EVs will help bring costs down.
Kunpeng super electric system links XPeng cars, robots, and AI
At the same AI Day where Iron debuted, XPeng also launched the Kunpeng super electric system. This is not a single car, but a high‑voltage platform that aims to give XPeng long‑range EVs with very fast charging and better efficiency.
How Kunpeng’s high‑voltage tech works in XPeng EVs
Kunpeng is built on an 800 V silicon‑carbide platform. In simple terms, this means the main power electronics use silicon carbide instead of standard silicon, which wastes less energy as heat when handling large currents. The system includes a 5C “AI” battery pack that can accept very high charge rates, a hybrid silicon‑carbide coaxial drive unit, and a quiet range‑extender engine in some models.
In new extended‑range cars like the XPeng P7+ EREV, XPeng says Kunpeng can deliver around 430 km of pure‑electric driving and up to 1,400 km of combined range when the range extender is used. Fast chargers tied into this system can reportedly bring the battery from 0 to 80% in roughly 12 minutes, which is close to adding about 1 km of range per second on the gauge.
Shared chips and software between EVs, robots, and robotaxis
Kunpeng shows how XPeng wants its EVs, robots, and robotaxis to share not just power hardware, but also computing and AI. The same Turing AI chips that run autonomous driving features in XPeng cars also power Iron, and will sit at the heart of XPeng’s planned Level 4 robotaxis.
This shared stack matters. Data from cars on the road can help improve the models that control Iron’s walking, grasping, and obstacle avoidance. Lessons from robots working in chaotic factory spaces can feed back into how robotaxis handle crowded depots or loading areas. Over time, this web of shared data can make all three product lines more robust.
Other tech moves fit the same pattern. For example, metameha has covered how virtual avatars in VR can change how the brain perceives the body. XPeng’s vision is similar in spirit: one AI core that can “inhabit” different bodies, from a sedan to a flying car to a humanoid worker.
What Kunpeng and Iron say about the future of “physical AI”
XPeng’s leaders describe this as a shift toward “physical AI,” where intelligent systems are tightly tied to the machines that move people and goods. Instead of treating cars as simple vehicles and robots as separate tools, the company treats all of them as embodiments of one underlying AI system.
If the strategy works, XPeng could lower costs and speed up new products, because improvements in one area, such as perception or motion planning, would spread quickly to others. If it fails, however, XPeng risks tying too many bets to one stack, which could be hard to update or replace if a better standard appears.
What XPeng Iron humanoid robots could do next in daily life
Today, Iron’s work is mainly inside XPeng’s plants and showrooms. But XPeng and several media outlets already talk about future roles for humanoids in public and private spaces. These visions are early and may take years to arrive, yet they signal where companies hope the market will go.
Possible roles for Iron beyond the factory
In the near term, XPeng suggests that Iron or its successors could stand at reception desks, guide visitors through showrooms, or act as brand ambassadors at auto shows. In logistics settings, humanoids might handle tasks that require climbing steps, opening doors, or working in tight mixed‑use spaces where installing fixed conveyors and robots would be too costly.
Further into the future, XPeng hints that home helper roles might be possible: simple household chores, basic elder support, or night‑time security patrols. Those ideas overlap with a broader trend of companies pitching humanoids as general‑purpose helpers rather than single‑task machines.
Jobs, safety, and regulation for humanoid robots
If humanoid robots spread beyond factories, the main questions will not be technical but social. People worry, with reason, about robots taking jobs in warehouses, retail, or cleaning. The shift may not remove work altogether, but it could change which skills are in demand.
Safety is another core issue. Laws and standards will need to address how heavy mobile robots behave around children, older adults, or crowds. Rules might cover speed limits in shared spaces, required safety sensors, and how companies must report incidents.
Governments are only starting to think about detailed rules for humanoid robots. For now, companies like XPeng rely on internal safety policies and broader machine‑safety standards. As deployments move from controlled factory floors into malls, offices, or streets, public regulators will likely push for stronger oversight.
How to make humanoid robots useful and trusted
For robots like Iron to be accepted, they must be both useful and predictable. That means clear, simple roles; easy‑to‑understand behavior; and transparent limits. Design choices that make robots look less threatening, such as smooth motion, clear status lights, and friendly but not uncanny faces, can also help.
Cost will matter too. Only when the total cost of ownership undercuts a mix of human labor and simpler machines will customers adopt humanoids at scale. XPeng hopes that building Iron on top of its EV platforms and AI chips will lower costs enough to cross that threshold.
Sources & related information
XPENG – XPENG unveils Kunpeng super electric system and AI‑defined mobility innovations at XPENG AI Day – 2024
XPeng’s official press release on Kunpeng and Iron describes how the company introduced the Kunpeng super electric system, the Turing AI intelligent driving system, and its humanoid AI robot Iron at XPeng AI Day 2024, framing them as parts of a unified AI‑defined mobility strategy.
CarNewsChina – Xpeng showcases humanoid robot Iron used for making its vehicles – 2025
A report from CarNewsChina on Iron in XPeng’s factory covers XPeng’s Iron robot at the 2025 Shanghai Auto Show, noting its 60 joints, 200 degrees of freedom, Turing AI chip with 3,000 TOPS of compute, and demonstrations of the robot performing assembly tasks in XPeng’s factories.
EVMoD – Xpeng showcases humanoid robot Iron used in EV production – 2025
Thai‑language coverage of XPeng’s AI Day reveal summarizes XPeng’s AI Day 2024 technology reveals and explains how Iron is designed for use on XPeng’s EV production lines, including specs such as its joint count, degrees of freedom, and use of XPeng’s own AI hardware.
TechNode – Xpeng targets 2026 with three Robotaxi models and mass‑produced humanoid robot IRON – 2025
TechNode’s feature on robotaxis and IRON outlines XPeng’s plan to launch three self‑developed Level 4 robotaxi models and mass‑produced IRON robots in 2026, all built around the company’s Turing AI chips and Vision–Language–Action model in a broader “physical AI” strategy.
LiveScience – Watch: Chinese company’s new humanoid robot moves so smoothly, they had to cut it open to prove a person wasn’t hiding inside – 2025
LiveScience’s story on the new IRON prototype reports on a newer XPeng IRON prototype presented at XPeng AI Day 2025, describing its lifelike movements, internal skeleton, artificial muscles, and over 80 degrees of freedom, as well as XPeng’s vision for IRON as a future “life companion or colleague.”
Gulf News – Meet ‘Iron’: Made‑in‑China humanoid robot with 200 degrees of freedom – 2024
The Gulf News article introducing Iron that prompted this article introduces Iron’s key specs, explains its role in XPeng’s factories, and briefly describes the Kunpeng super electric system and XPeng’s broader plans for robotaxis and smart mobility
0 Comments