XPENG has commissioned a dedicated production line for its IRON humanoid robot, and the first completed unit autonomously walked off the line. The company says more than 80% of the line’s core processes are automated, bringing automotive-grade manufacturing methods into humanoid robotics as IRON moves toward mass production.
The first completed IRON walked off the line on its own
XPENG has moved its humanoid robot program from prototyping into line manufacturing.
On September 8, the company said its dedicated humanoid robot production line had officially entered operation and that the first completed IRON unit autonomously walked off the line after production.
That detail is more than a launch-stage flourish. A humanoid robot is a tightly integrated machine: actuators, joints, sensors, controllers, compute, power systems, dexterous hands, software, and calibration all have to arrive at the end of the line as one functioning system.
XPENG is positioning the new facility as the bridge between the engineering work behind IRON and the repeatable manufacturing process needed for larger-scale deployment.
The company says the line was designed for scale from the beginning, with more than 80% of its core processes automated.
XPENG is borrowing the manufacturing discipline of a car company
XPENG did not build the robotics line as an isolated lab.
The company says it brought automotive-grade quality systems from its electric-vehicle manufacturing operations into humanoid robotics. The goal is to apply the same kind of process control, repeatability, inspection, and capacity planning used for complex vehicles to a new product category.
That is a logical fit for a humanoid robot.
IRON has dozens of moving joints, high-performance onboard compute, custom control hardware, and mechanical structures that have to work together with consistent tolerances. Moving from a few engineering units to repeatable production means those elements need manufacturing processes that can be measured and reproduced.
XPENG describes the line as high-precision, flexible, and intelligent, with automation concentrated in the critical processes that determine final product consistency.
For a company already operating large vehicle plants, robotics becomes an extension of an existing manufacturing skill set rather than a completely separate discipline.
IRON was already being prepared for mass production
The production line follows a roadmap XPENG has been signaling for months.
The company has been developing IRON as a general-purpose humanoid platform and previously said the robot was moving toward mass production in 2026.
Its earlier public demonstrations focused heavily on the robot itself: human-like movement, dexterous hands, onboard AI compute, and the ability to operate in environments designed around people.
The new announcement shifts the emphasis from the prototype to the factory.
That is an important change in the story. A robot can look impressive in a controlled demonstration and still be far from a product that can be built repeatedly. A dedicated line introduces another layer of engineering: assembly sequence, calibration, inspection, traceability, throughput, and quality control.
XPENG is now showing that layer alongside the robot.
The hardware is built around 76 degrees of freedom
The current IRON platform uses 76 degrees of freedom across the body, according to XPENG.
Each hand contributes 21 degrees of freedom, giving the robot a large mechanical range for hand and finger movement.
The company also uses a fully enclosed flexible lattice structure intended to combine a human-like form with a protected mechanical design.
Those specifications explain why manufacturing matters so much.
A humanoid with this many articulated elements is not assembled like a simple mobile robot. Every joint, actuator, linkage, controller, and sensor relationship has to end up inside a repeatable mechanical system that can move as intended after final assembly.
The same applies to the hands. High dexterity only becomes useful at scale if the geometry, calibration, control electronics, and software behavior remain consistent from one unit to the next.
The new line is therefore part of the product architecture, not just the place where the product happens to be assembled.
Three Turing AI chips put up to 2,250 TOPS on the robot
IRON’s compute stack is another part of the manufacturing challenge.
XPENG says the robot uses three of its Turing AI chips for up to 2,250 TOPS of effective computing power.
That onboard compute is intended to run XPENG’s Physical AI foundation model directly on the robot. The company says this allows IRON to perform complex tasks autonomously without relying on remote operation, while keeping inference latency low.
It also makes IRON part of the same broader technology strategy XPENG has been developing across vehicles, Robotaxi, and robotics.
The interesting manufacturing consequence is that a humanoid unit is not only a mechanical assembly. It leaves the line as an edge-AI computer with motion hardware attached to it.
Compute modules, controllers, actuators, sensor systems, software images, and final calibration all become part of a single production target.
More than 80% automation is aimed at repeatability
XPENG says automation exceeds 80% across the production line’s core processes.
The company connects that automation directly to quality consistency and future capacity expansion.
For humanoid robots, that is especially relevant because scale depends on repeatability. A production system has to turn complex electromechanical assemblies into units that behave consistently enough for software, control models, and downstream testing to work across the fleet.
Automation can also make the manufacturing data more structured.
Every repeated process creates measurable information about tolerances, calibration, component quality, and final-system behavior. Over time, that can give the engineering team a tighter feedback loop between how IRON is designed and how it is actually built.
XPENG’s advantage here is straightforward: it already knows how to connect software-heavy products to automated manufacturing at vehicle scale.
Stores and campuses are the first commercial environments
XPENG says IRON is scheduled to enter mass production by the end of 2026.
Initial commercial rollouts are planned for XPENG’s own stores and campuses, giving the company controlled environments in which to deploy the robots before broader customer delivery.
The company plans an official market launch and deliveries in China and overseas markets in 2027.
That rollout sequence gives the manufacturing program a clear next step.
The new line establishes repeatable production. Internal stores and campuses create real operating environments. Broader delivery then expands the number of places in which the robot can collect experience and perform useful tasks.
For Physical AI, those stages are closely connected. More deployed robots mean more real-world interaction, and a scalable factory is what makes that deployment loop possible in the first place.
The Upgrade Feeling
The most important part of XPENG’s announcement is not simply that another humanoid robot exists.
IRON now has a production system behind it.
A dedicated line, more than 80% automation in core processes, automotive-grade manufacturing methods, 76 degrees of freedom, dexterous hands, and 2,250 TOPS of onboard compute all point toward the same transition: from building a robot that works to building the same robot repeatedly.
That is where humanoid robotics starts looking less like a sequence of demonstrations and more like an industrial product category.
XPENG’s next milestone is scale. The factory is now part of the robotics story.