Xpeng IRON Targets Stores in 2026 After Lifelike Humanoid Debut

A Guangzhou stage walk puts IRON in the spotlight

Xpeng plans to put its next generation IRON humanoid robot in its own retail locations in 2026 and targets mass production by the end of that year. At the company's AI Day in Guangzhou, a strikingly human walk brought attention to the machine that could eventually greet customers, guide visitors and answer questions in public spaces.

IRON crossed the brightly lit stage with a stride resembling a model's walk, shifting its weight from hip to hip while controlling its feet and toes to stay balanced. The demonstration quickly drew attention on social media, where some viewers wondered whether a person was wearing a robotic shell.

An engineer responded by carefully cutting open part of the robot's outer covering on stage. The opening exposed the metal structure and actuators inside a leg. IRON then walked across the floor again, showing the machinery behind its lifelike movement. Actuators are the components that turn control signals into physical motion.

As of November 8, 2025, Xpeng's plan is to begin with service roles in settings such as car showrooms, shopping malls and office lobbies, followed by broader commercial deployment. The robot combines custom computing chips, a vision centered AI system and mechanics designed to imitate the human body.

Xpeng calls this wider strategy physical AI: artificial intelligence that controls machines moving through the real world. IRON connects that ambition to the company's work on cars and robotaxis, sharing software foundations with vehicles while adapting them to a machine that walks, reaches and interacts with people.

How the body produces a natural stride

IRON follows the human body's structure from its internal frame to its outer surface. A flexible spine and a network of bionic muscles allow the torso to bend, twist and recover its balance. Those movements help the robot shift its posture as it walks or reaches.

Xpeng says IRON has 82 degrees of freedom across its body, including 22 in each hand. A degree of freedom describes an independent movement that a mechanism can make, such as bending a joint or rotating it. More available movements give engineers more ways to shape a gesture, grip an object or adjust the robot's position.

The feet include passive toe movement to smooth the roll from heel to toe during a step. Passive movement allows part of the foot to yield mechanically as weight moves over it, helping the stride look more natural. Small harmonic joints also help the hands match human hand size, making them better suited to tools and everyday objects designed for people.

Walking on two feet requires continual adjustments at the ankles, knees, hips and spine. People make these corrections without consciously planning each one. A robot must coordinate thousands of timed commands to achieve a similar result. Xpeng combines motion learned from data with mechanical choices such as yielding toes and a layout that gives the hands and shoulders human proportions.

The flexible spine, hands and feet also work together when the robot leans or reaches. Holding a small box, extending an arm or stepping around a display stand changes how weight is distributed. IRON's design aims to keep the whole body stable through those changes.

Skin, expressions and a customizable appearance

IRON's outer layer is synthetic skin that covers the body and feels warm to the touch. Xpeng designed it to make close interactions feel less cold than contact with exposed metal or hard plastic. In a service setting, that could matter during a handshake or a handoff.

A curved 3D display wraps part of the head. It acts as a changing face, showing simple emotions, animations or practical cues such as service status. Those visual signals can help a visitor recognize whether the robot is listening, processing a request or finishing a task.

Xpeng also supports customization of body shapes and sexes, hairstyles, colors and clothing. A robot working in a showroom could therefore have an appearance suited to the brand and surroundings, while another in an office lobby could be dressed differently.

The hands are central to this social role as well as to handling objects. With 22 degrees of freedom on each side, they are designed for readable conversational gestures and delicate grips. Possible service tasks include holding a phone to scan a code, picking up a brochure without crumpling it or passing an item to a visitor.

A visual route from seeing to moving

IRON's control system is built around a vision centered model that connects what the robot sees with what it should do next. Xpeng describes the route as vision to implicit token to action. In simple terms, visual information becomes an internal representation that the system can use to choose a movement.

The approach avoids first turning every image into words before deciding how to act. Language remains useful for conversation and planning, but that extra conversion can introduce delay when the immediate job is to step aside or pick up a dropped card.

A robot balancing on two feet must repeatedly update its foot placement and posture. Learning direct connections between visual input and motor output can shorten the interval between noticing a change and responding to it. IRON's system is intended to recognize a scene, predict intent and move with minimal detours.

Xpeng combines several models to handle different parts of that work. VLT, short for Vision Language Task, is the core engine for behavior directed toward a goal. VLA, or Vision Language Action, comes from the automotive team and provides continuous control of the robot's 82 degrees of freedom. VLM, or Vision Language Model, supports conversation and following instructions.

Working together, these systems are intended to let IRON speak with a guest, understand the surrounding store and then walk, point or hand over a brochure as part of one interaction. The conversation and movement functions have different jobs, but they must coordinate for the service to feel continuous.

Three Turing chips supply the computing power

IRON carries three Turing AI chips developed by Xpeng, with combined computing performance of about 2,250 TOPS. TOPS means trillion operations per second, a measure of how many mathematical operations processors can perform.

The computing capacity is intended to support several neural networks at the same time, including systems for perception, dialogue and motion. Neural networks are AI models that learn patterns from data. Running them on the robot allows it to process routine tasks without sending that work to a distant server.

Some presentations at the event referred to a higher effective computing figure for particular modes. The stated figure of about 2,250 TOPS describes the combined performance of the three chips, while those other references concerned how performance was counted in specific operating modes.

Xpeng's aim is a compact computing platform that can keep pace with movement and interaction in real time. A robot speaking to someone while watching its surroundings and maintaining balance needs several processes to run together, rather than taking turns one task at a time.

A solid state battery, braking and privacy rules

Xpeng says IRON is the first humanoid robot to use an all solid state battery. A conventional lithium ion cell uses a liquid electrolyte to carry charge. A solid state design replaces that liquid with a solid electrolyte.

The change aims to improve stability when the battery gets hot and reduce fire risk, an important consideration for a machine operating beside people indoors. Solid electrolytes can also increase energy density, meaning more energy can be stored for a given battery size or weight. That could help a mobile robot operate longer without carrying a heavier pack.

IRON also supports indoor automatic emergency braking, applying a concept familiar from advanced driver assistance systems in cars. If sensors detect a person or a glass wall that the robot did not anticipate, the controller can slow or stop the machine to avoid contact.

Xpeng has added a privacy principle to its discussion of robot behavior. Beyond the three laws associated with science fiction writer Isaac Asimov, the company proposes a fourth duty: protecting personal data. Its stated approach is that sensitive user information should stay on the device during routine operation.

The three onboard chips support that approach by allowing more processing to take place inside the robot. Keeping routine work local reduces reliance on cloud processing and connects the privacy promise to the way the computing system is built.

Why stores come before large factory deployments

Xpeng's first deployment priority is service in public spaces, rather than heavy industrial work. Humanoid robots still face questions about durability, cost and maintenance in demanding factory environments. Starting in the company's own retail locations gives IRON a more structured setting with staff nearby.

The proposed duties include greeting visitors, guiding them through exhibits, answering product questions and directing people during busy launches. Car showrooms, malls and office lobbies all require a robot to navigate around people while communicating in a useful, understandable way.

Xpeng plans to open a software development kit so outside teams can create applications. A software development kit gives developers the tools needed to build functions for a platform. Possible uses include a branded store concierge, an office assistant checking badges or a museum guide whose speech is coordinated with a video wall.

The shared AI foundations with Xpeng's vehicles also give the company a common starting point for perception and planning. Tools used to recognize a stroller at a crosswalk could help a service robot notice a child who has moved away from a parent. The surroundings and required actions differ, but both tasks involve recognizing people and objects before choosing a response.

The deployment schedule separates initial use from manufacturing at scale:

• At AI Day in Guangzhou, Xpeng introduced the next generation IRON and demonstrated its walk, then exposed the machinery inside a leg.
• In 2026, the company plans to begin placing IRON in its own retail locations for service work.
• By the end of 2026, Xpeng targets mass production, with broader commercial deployment to follow the initial placements.

Xpeng has not given a precise date for the first store deployment or a detailed timetable for the broader rollout. The 2026 plans are targets, with the first operating locations intended to provide experience before expansion.

Baosteel trials will test a different environment

Xpeng named Baoshan Iron and Steel, also known as Baosteel, as an ecosystem partner. IRON units are planned for trials in steel facilities, including inspection work and routine rounds.

Steel plants present different conditions from a showroom. Heat, dust and vibration will give engineers practical information about what the robot can withstand and what it would need to work through long shifts. The trials are intended to help define those requirements.

Lessons from the industrial pilots are likely to feed into later versions of IRON. This puts factory testing alongside the initial service plan, while keeping large scale industrial deployment outside the first priority.

Where IRON fits among other humanoids

Tesla continues to show progress with Optimus, Boston Dynamics is moving Atlas toward industrial work after becoming known for dynamic jumping and running, and Chinese companies such as Unitree are pursuing rapid updates. Humanoid robots across the industry are moving from laboratory demonstrations toward staged trials.

Xpeng's immediate focus is lifelike social presence in public spaces. IRON combines a soft outer skin, human proportions, expressive display, privacy rules centered on processing inside the device and a solid state battery intended to suit indoor operation. Its starting role is service rather than heavy lifting.

The Guangzhou demonstration showed the result of Xpeng's work on gait and mechanical coordination. Practical usefulness will depend on the hours IRON spends doing tasks on a showroom floor or during industrial trials. Those settings will bring repeated interactions, reaching movements and navigation demands beyond a short stage walk.

The planned deployments connect the visual appeal of a humanoid to specific work. Greeting a visitor is a different challenge from inspecting a steel facility, but both require reliable movement, perception and task completion. Xpeng is preparing to gather experience in both environments.

IRON joins a wider physical AI push

Xpeng's robot is part of a broader effort that includes cars, robotaxis and low altitude aviation products. The same vision centered model used for IRON also supports driver assistance in the company's vehicles and a robotaxi program expected to expand with new cars in 2026.

The shared software is designed to run on Xpeng's own chips. In vehicles, the computing arrangement includes substantial redundancy, providing backup capacity for systems that must continue operating reliably.

Xpeng has also introduced products under the ARIDGE brand, including a modular ground vehicle that carries a small electric vertical takeoff aircraft. Such an aircraft can rise vertically rather than needing a conventional runway for takeoff.

Together, these projects show how Xpeng is extending its work beyond automobile manufacturing. Its stated ambition covers AI controlled machines moving through traffic, walking across showrooms and eventually flying. For IRON, the next practical step is narrower and more immediate: entering stores in 2026 and turning coordinated movement and conversation into everyday service.

At a Glance

• Xpeng unveiled the next generation IRON at AI Day in Guangzhou with a lifelike walking demonstration.
• An engineer opened part of a leg covering to reveal the metal structure and actuators before IRON walked again.
• The robot has 82 degrees of freedom across its body, including 22 in each hand, plus a flexible spine and passive toe movement.
• Warm synthetic skin, a curved 3D head display and customizable appearance support its planned service role.
• Three Turing AI chips provide about 2,250 TOPS for perception, dialogue and movement.
• Xpeng says IRON uses an all solid state battery and supports indoor automatic emergency braking.
• The company's privacy principle calls for sensitive personal information to remain on the device during routine operation.
• Initial placements in Xpeng stores are planned for 2026, with mass production targeted by the end of that year.
• A planned software development kit will let outside teams build applications.
• Trials with Baoshan Iron and Steel will test inspection and routine rounds in steel facilities.