The Carmakers Building Humanoids: Why Auto Companies Think They Own the Robot Future

This month, a humanoid robot called Xiaodong is expected to start working in a car factory. Not a robot lab, not a demo booth, a factory. Dongfeng Motor says Xiaodong will handle material sorting and quality inspection on its production lines, with small-batch trial production of the robot itself beginning by the end of the year. Meanwhile XPeng has an operating humanoid production line and a mass-production target of December. BYD just published a design patent for its own humanoid robot on October 9.

Notice something? None of these companies are robotics companies. They are car companies. The humanoid robot race, it turns out, is increasingly being run by automakers.

The Chinese carmaker wave

Dongfeng's announcement, made by chief engineer of intelligent technology Zhang Zhenlin during the company's 10th science and technology innovation week, is the most detailed. Xiaodong will enter factories in October 2026, mainly targeting manufacturing work like sorting and quality inspection. Small-batch trial production follows by year-end. The ambitious part is the timeline after that: Zhang said the company aims to bring the robot's working capabilities on par with those of human workers by the end of 2027.

Dongfeng also has a second machine: Yuanzai, a quadruped robot dog with autonomous navigation, multimodal perception, and intelligent interaction, aimed at campus patrol and smart services. The company says Yuanzai will reach commercial use before the humanoid, starting with customer guidance at dealerships. Both machines debuted at the 2026 World Robot Conference in August.

XPeng is moving faster still. The company put a humanoid robot production line into operation on September 8, 2026, and has said its IRON robot will begin mass production at the end of this year, with launch and delivery in China and overseas in 2027. CEO He Xiaopeng took direct charge of the robotics business earlier this year. And Chery's incubated robotics arm, Aimoga Robotics, has already delivered more than 3,000 robots globally, including more than 2,000 overseas, and is reportedly preparing an initial public offering to fund further expansion. Li Auto, Seres, and Leapmotor have all disclosed embodied intelligence plans this year too.

Then there is BYD. The company had said back in July that it would unveil a humanoid robot in August. That debut never happened. Instead, on October 9, China's National Intellectual Property Administration published a BYD design patent for an "embodied intelligent robot," with patent drawings showing a two-armed, two-legged machine with articulated fingers, intended for service, education, and industrial settings. The patent covers appearance only, so it reveals nothing about capability or timeline. An earlier social media tease, "Xiao Di," was reported at 1.61 meters tall, 58.5 kilograms, with 31 degrees of freedom and a dexterous hand with plus or minus 1 millimeter repeatability. But until BYD actually unveils a product, it remains paperwork.

The humanoid robot race is increasingly being run by automakers. The supply chains, the manufacturing discipline, and the AI stacks built for cars transfer almost directly to embodied robotics.

Why carmakers think they can win

The logic is consistent across the industry, and Dongfeng's Zhang spelled it out: developing embodied intelligence products lets automakers reuse technology across multiple areas, including automotive AI models, speeding up commercialization and mass production. The supply chains built for electric vehicles, batteries, vision sensors, electric motors, compute hardware, and precision assembly, are nearly identical to what a humanoid robot needs. The factory that stamps car doors can stamp robot limbs.

The AI side transfers too. The perception software that lets a car see lane markings and pedestrians is conceptually the same technology that lets a robot navigate a factory floor. Both problems are, at their core, about moving safely through a physical world while understanding what is around you. This is why Tesla's Optimus program, the reference point for the whole sector, is structured as an extension of its self-driving effort rather than a separate robotics company.

There is also a manufacturing-discipline argument. Building a million cars a year forces a company to master things robotics startups are still figuring out: supply chain management at scale, quality control across thousands of suppliers, and cost reduction curves. Agility Robotics built its RoboFab factory for up to 10,000 Digit robots a year, but that is a rounding error next to automotive volumes. XPeng has talked about a 1 million robots per year target by 2030, which only makes sense if you think of a robot as a car with legs.

The gap between the roadmap and the reality

Here is the awkward part. The production timelines are accelerating faster than the technology is. An industry analysis published October 10 identified four persistent roadblocks that no current platform has solved at once: dexterity, battery life, sim-to-real transfer, and unit economics.

The battery ceiling is the most brutal. Most humanoid platforms run two to four hours on a charge. Industrial shifts run eight to twelve. That gap forces constant recharging or battery swaps, which is exactly the kind of downtime that kills factory economics. Dexterity is equally stuck: human hands have 27 degrees of freedom and thousands of tactile sensors, and current robots still fail regularly at grasping fragile items, opening varied packaging, or manipulating tools in unstructured settings. Simulation-trained AI policies still break when they meet real contact dynamics, actuator friction, and sensor noise. And industrial-grade units cost $30,000 to $150,000 each, with five-year operating costs on top, which makes the economic case against human labor hard to close at today's task-completion rates.

This is why the smart deployments are narrow. Boston Dynamics deliberately scopes its Atlas manipulation work to tightly constrained tasks, only asking the hardware to do what it can actually do. The carmakers' October deployments follow the same pattern: sorting and quality inspection are repetitive, controlled-environment tasks, the one category of factory work where today's humanoids can plausibly succeed.

What to watch next

The next twelve months are the sector's honesty test. XPeng's end-of-2026 mass production target and Dongfeng's year-end trial production will collide with the physical realities of batteries, hands, and unit economics. Demonstrations and pilot deployments are not the same as confirmed commercial adoption, and a production line that walks robots off autonomously is still only the beginning.

But dismiss the carmakers at your peril. Every robotics company right now is trying to become a manufacturer. The automakers already are one. If the humanoid robot's biggest problems turn out to be manufacturing problems, the companies with the factories win. If they turn out to be AI and dexterity problems, the timeline slips again, and the incumbents with the most real-world deployment data, like Aimoga's 3,000 delivered units, gain the advantage.

Either way, the venue has changed. The humanoid robot is no longer just a Silicon Valley science project. It is an industrial product, being built in the same countries, by some of the same companies, that built the electric car. The robot race and the car race are the same race now.