What exactly are automakers aiming for by building humanoid robots?

[[**Automakers are building humanoid robots with the aim of using the auto industry's capital, real factory tasks, and supply chain capabilities to incubate a new business: first validating in manufacturing and store scenarios, then seeking external customers, though stable returns still await scale-up validation.**

**Key Points**

**1. The auto industry supports robots with capital and scenarios**

Existing businesses bear the early investment, factories provide real tasks, and the supply chain builds improvements into the next generation; Optimus has already begun production in Fremont, the external delivery target remains around the end of 2027, and there are plans to build 10,000 to 20,000 units for training.

**2. Continuously transforming production systems is the underlying driver**

Global auto production in 2025 was about 96.4 million vehicles, and the gains from less scrapping, less inventory, and faster ramp-up are considerable; Ford's assembly line, Unimate entering the factory, and BMW's virtual factory all show that the auto industry has long competed through production improvements.

**3. Humanoid robots must iterate in real tasks**

Fixed robotic arms struggle to cover pick-and-place and handling where positions and models change frequently, while humanoids can use workstations and tools designed for people; Figure 02 ran for more than 1,250 hours at BMW, and a forearm failure drove a wrist-structure change in the third generation.

**4. The partnership route determines development responsibility and returns**

Hyundai controls Boston Dynamics with Mobis supplying actuators, BMW partners with Figure for validation, XPeng develops IRON in-house and builds a dedicated production line, and Mojia leverages Chery's channels; internal trials do not equal sustained payment from external customers. 2026-10-23 21:15 What exactly are automakers trying to achieve by building humanoid robots? Feng Wei Feng Dabai Quick Overview

This article is from the WeChat public account: Feng Wei Feng Dabai, author: Text / Feng Dabai

On October 21, US local time, Tesla announced in its third-quarter earnings report that Optimus has begun production at the Fremont factory. In the subsequent conference call, Musk's external delivery target remained around the end of 2027: the robot's hands, durability, and AI all still need further improvement. To obtain training data, Tesla also plans to build 10,000 to 20,000 robots and have them practice various tasks at the "Optimus Academy."

What automakers want is to build cars more efficiently, and they also want to sell robots to new customers. They may also appear first in stores: providing reception and explanations for cars while observing how humans and robots interact.

From the production line to customers' hands, robots must go through task training, troubleshooting, design changes, and mass manufacturing. Tesla's plan lays out exactly this distance: the production line has already started, and next they still need to build tens of thousands of machines and let them practice continuously.

Why is the auto industry positioned to walk this path? Existing businesses can bear the early investment, factories provide real tasks, and engineering and the supply chain can build improvements into the next generation of products.

The auto industry has always been rewriting production methods

The auto market is large enough that continuous improvements in production models can translate into considerable returns. According to data published by the International Organization of Motor Vehicle Manufacturers (OICA), global auto production in 2025 was about 96.4 million vehicles. It covers passenger cars and commercial vehicles, behind which are enormous parts procurement, factory operations, and after-sales needs. For an industry of this scale, scrapping fewer parts, holding less inventory, and bringing new models into production faster are all worth companies continuing to spend money researching.

Existing businesses can provide funding for new projects, but how long they can invest is still constrained by the performance of the core business. When XPeng released its first-quarter results this year, He Xiaopeng proposed using the profitability and cash flow of the auto business to support physical AI research and development, and then cultivate new revenue sources including robots. Since then, the funding sources have expanded: according to XPeng's disclosure, its robotics business completed its first round of equity financing on August 24, raising more than $900 million. This allows development risks to be shared by more investors. If automakers first provide technology, manufacturing, and application conditions, and then bring in external capital for the new business, they have a chance to reduce the pressure on the core business to bear long-term investment alone. Financing supports continued development; whether stable revenue can be generated is still determined by product sales and use.

This kind of investment has a clear historical thread in the auto industry. In 1913, Ford's moving assembly line had vehicles completed station by station, reorganizing mass production. In 1961, Unimate entered the General Motors factory and took over dangerous, repetitive work such as unloading die-casting machines. By the intelligent manufacturing stage, trial and error were moved forward into the digital environment: BMW used a virtual factory to check collision issues between new models and production lines, shortening what had been nearly four weeks of physical testing to three days of virtual simulation.

The technologies changed at each stage were different, but the requirements companies faced were connected: the next batch of cars had to be built better, new models had to be delivered on time, and production costs still had to be competitive. The Toyota Production System emphasizes handling quality abnormalities and producing on demand, likewise making process improvements subordinate to overall delivery. To sell the next generation of products well, automakers must continuously transform production systems, and suppliers must also adjust parts, processes, and delivery arrangements accordingly. This long-term pressure is the driving force behind the auto industry's pursuit of productivity changes.

After capital investment, automakers still have to organize scattered technology and manufacturing resources into products. A selling price and delivery target need to be broken down into parts performance, manufacturing costs, quality requirements, and procurement plans so that suppliers can decide what to do and how much to invest. In the robotics business, the automaker's role should also be judged along this relationship: what tasks it proposes, how much cost it allows, and how it plans to procure are the basis for suppliers to develop components for it. The vast industrial chain is mobilized through clear requirements and order expectations, and only then does the industry's scale truly translate into engineering strength.

Cars are also a highly mature and complex consumer product. Factories care about cycle time, yield, and downtime; dealerships are the first to hear consumers' questions about price, design, and features; and the after-sales network constantly encounters failures in real-world use. This feedback influences how the next model is defined, which parts need to be changed, and which features are worth continuing to invest in. After entering the robotics business, the automotive network can provide early opportunities for contact, but robots still need to collect their own usage feedback. Whether handling in the factory is stable, whether interaction in the store is effective, and whether home users are willing to use it every day each correspond to different product requirements.

Of course, electronics manufacturing also has a huge market and precision manufacturing capabilities. IFR's data on industrial robot installations in 2024 shows that the electronics industry is slightly ahead of the automotive industry. Cars deserve to be discussed separately because large-scale production demand, the development of complex electromechanical products, consumer brands, and sales and service networks are concentrated in the same industry, which also continuously undergoes product updates. It is positioned to simultaneously be a user of robots, a product operator, and an early market organizer. Young automakers can also participate in this system; the key lies in how these conditions are used, not in how many years the company has existed.

Machines begin to learn new work

Car factories are not short of robots. Fixed robotic arms are already adept at tasks such as welding and painting, for which mature solutions have long existed. The reason for also researching humanoid robots today is that companies want machines to handle more variable tasks: the positions of objects change, the models being produced change, and the picking, placing, handling, and manipulation originally done by people are also hard to cover entirely with a single set of fixed motions.

Vision, language, and action models are attempting to connect environments, instructions, and actions. The on-device robot model announced by Google DeepMind in 2025 has demonstrated research progress in adapting to new tasks through a small number of demonstrations. For automakers, the direction worth betting on is whether, after a vehicle model change or a part changes position, a robot can continue working through training, reducing the burden of developing a dedicated solution anew each time.

The appeal of the humanoid structure is also related to this. A large number of workstations, tools, and passages were originally designed for human use. Being close to the human operating range gives robots a chance to reuse part of the existing environment. Companies can then try to bring the machine to the task rather than rebuilding a workplace around the equipment each time. Bipedalism is not the best choice for all tasks; wheeled equipment on stable ground can also complete many jobs. What automakers ultimately buy is still the solution that suits the task. The spatial adaptability brought by a humanoid form must prove its value together with action capabilities.

BMW's cooperation with Figure has already shown how real tasks change robots. Figure 02 handles sheet metal part loading at the Spartanburg plant, taking parts from racks and placing them into welding fixtures. According to Figure's project review, over more than 1,250 cumulative hours of operation, the forearm was the main hardware failure point. The third-generation product therefore redesigned the wrist electronic structure, removing the communication distribution board and dynamic cables there. Problems that emerged on the shop floor thus made their way into the next generation's blueprints.

This cooperation gives BMW an opportunity to validate new equipment on existing production lines, while Figure gains failure and maintenance feedback that laboratories find difficult to provide continuously. The two sides seek different returns: BMW needs better production tools, while Figure wants to incorporate improvements into its own products and serve more customers.

Automakers' own intelligentization investments in recent years have also given them opportunities to participate further in development. XPeng uses its self-developed Turing chip in IRON and has also built a dedicated production line for the robot. Chips, computing facilities, and R&D tools can serve different products, but robots still need to build their own motion control and task capabilities. For companies, the significance of such investment is not just building one fewer system: technical resources originally used for cars begin to have opportunities to validate other product directions.

Cars can recognize roads, while robots also need to learn to grasp, contact, and manipulate objects; the constraints faced by the products are not the same. But if a company believes that intelligent technology can continuously expand the physical tasks that can be completed, it has reason to maintain an R&D direction beyond cars. Waiting until the robot market is fully mature before entering means leaving early product definition, usage feedback, and partnerships to the pioneers. What early investment buys includes both technological progress and the time to choose the next business.

How investment returns to the business

When a company advances multiple robot uses at the same time, it must allocate limited R&D manpower and budget. Stores want robots to look friendly and interact smoothly, while factories care more about payload, motion reliability, and continuous operation. A product optimized for reception may not be suitable for entering the production line; capabilities developed for factories may not attract consumers. Management must first determine whom to serve first and which department bears development and trial costs, so that acceptance can be based on the corresponding requirements. Otherwise, a demonstration gains attention while the production department cannot use it, and the project will find it hard to sustain operations on several scorecards that each tell a different story.

Hyundai places Atlas training within manufacturing tasks, providing an observable path. RMAC first conducts validation around parts sequencing and material handling, while formal factory deployment is scheduled for later stages. Investment during training prepares for future production applications. If robots can gradually take on repetitive handling or heavier tasks, factories can rearrange some positions and use people for work requiring judgment, handling anomalies, and maintaining equipment. Changing the workforce structure involves adjustments to positions and skills and cannot be directly equated with how many workers are laid off.

What the production department accepts is the change across the whole job: how much training, maintenance, manual takeover and downtime cost in total, and whether qualified output is more stable. If the same equipment can also adapt to new parts and new workstations through training, there is a chance to reduce dedicated modifications when changing over. When costs remain high and output is unstable, factories have more reason to narrow the task scope, switch to other equipment, or delay purchasing. The operating records Figure accumulated at the BMW site answered how the product should be changed; expanding procurement still requires proving that the robot is more worth investing in for this task than existing solutions.

The group can arrange trials at its own factories, but external customers will compare the robot with manual labor, fixed robotic arms and other solutions on price, deployment and maintenance costs. A single procurement within the group cannot prove that the group as a whole gains new returns. For the second curve to hold, the product must also make external customers willing to keep paying.

The home market is another hurdle: consumers will look at whether the robot can complete everyday tasks such as carrying and picking and placing, whether it is easy to use and repair, and whether it is worth the price. An automaker's experience running durable consumer goods can help it think about delivery and long-term service, but the reason to buy still has to come from the robot itself.

This kind of market expansion also changes product design. A company cannot build a machine from scratch for every customer, otherwise the more it sells, the heavier the burden of dedicated adaptation and service becomes. The robot needs to gradually develop repeatable manufacturing hardware and scalable task capabilities, so that later customers can build on development already completed. The value of manufacturing experience here is to help the new business organize product updates and continuous delivery. Only after orders and designs gradually stabilize does scale have a chance to further reduce costs.

Before this new business matures, robots may also first enter the car sales process. Moja has publicly described its application in overseas automotive marketing service scenarios and disclosed its use of Chery's channel, logistics and service networks. Car showrooms give robots a place to reach consumers, and robots add a way for showrooms to interact.

Much of the investment in intelligent technology is invisible to consumers day to day. A robot that can communicate naturally and complete reception tasks makes part of that technical capability something that can be experienced on the spot, and also gives sales staff more opportunities to introduce cars.

Factory trials, showroom reception and external sales do not have a uniform order of returns. Hyundai first validates around manufacturing tasks, while XPeng and Moja consider human-facing service scenarios earlier. Showroom applications especially need to distinguish a one-time display from long-term operation: an event budget can support a short-term appearance, but whether dealers are willing to continuously bear maintenance and service costs is closer to the real demand of this channel.

Figure 1: Support from the automotive business for robot iteration and return checks. Solid lines indicate investment, use or design feedback already disclosed by the companies in the text; dashed lines indicate returns and subsequent investment relationships that still need verification; the various relationships do not occur within the same company.

Business goals begin to diverge into different routes

Which kind of return a company hopes to get determines how much of the robot business it must take on. Developing products independently, holding a controlling stake in a specialized company, and purchasing solutions from external companies can all bring robots into the automotive industry, but the destination of funds and the operating responsibility differ.

Hyundai's controlling stake in Boston Dynamics, together with Mobis's participation in developing and supplying Atlas actuators, allows it to coordinate product development, component manufacturing and internal applications within the group. It must invest in robot R&D and manufacturing, and also face the external market for this business. BMW's cooperation with Figure, by contrast, allows it to set requirements and participate in validation around automotive production needs, with Figure responsible for robot product development. BMW needs to judge whether the solution is worth using, and Figure still has to figure out how to sell improvements to the next customer.

XPeng chose to develop IRON independently and build dedicated production lines, placing more of the responsibility for product definition, R&D, manufacturing and delivery within its own business system. It can decide for itself how the robot collaborates with existing intelligent technology, and it also needs to organize sales and after-sales service for this product line. According to its published plan, the robot will start from showrooms and campuses. These scenarios can provide interaction and service tasks; after internal trials, it still has to build delivery and maintenance capabilities for external customers.

Moja enters overseas automotive marketing scenarios through Chery's channels, which allows it to use existing outlets to reach customers and rely on logistics and service networks to arrange delivery. But having an automotive sales and service network does not mean it already has robot service capabilities. Robot companies still need to provide support for local languages, tasks and repair needs so that outlets can actually operate such products.

Figure 2: Three business relationships and public progress. BMW is a user and partner; Figure is the robot product developer. Future deployment, mass production and delivery times are all company plans; the Figure 03 logistics task is a demonstration.

This division of labor also leaves room for specialized robot companies and component suppliers. Automotive customers raise requirements and participate in validation, while product and component suppliers turn experience into results that can be sold again. Automakers can benefit from better production solutions, while suppliers can continue doing business with other customers. The role of the automotive industry in driving robot development will also occur through such cooperation and cannot be measured only by how many robots the automakers themselves have sold.

Hyundai has Atlas first practice parts sequencing, XPeng has IRON start from showrooms and campuses, and BMW has Figure find problems on the production line. They have already handed the next round of R&D to different sites: factories continuously expose grasping, failure and downtime problems, while showrooms expose interaction, reception and maintenance problems. The automotive industry can mobilize capital, manufacturing and sales networks; whichever site it first repeatedly solves problems at is the site where it will first produce that type of robot. Channel: Cars and Mobility

22 articles published on Huxiu. This content is authorized by the author for publication; the views represent the author only and do not represent Huxiu's position. If you have objections or complaints about this manuscript, please contact tougao@huxiu.com. Everyone who is changing and wants to change the world is on the Huxiu APP[ ]

Hyundai lets Atlas first practice parts sequencing, XPeng lets IRON start from showrooms and campuses, and BMW lets Figure find problems on the production line. They have already handed the next round of R&D to different sites: factories continuously expose grasping, failure and downtime problems, while showrooms expose interaction, reception and maintenance problems. The automotive industry can mobilize capital, manufacturing and sales networks; whichever site it first repeatedly solves problems at is the site where it will first produce that type of robot. Channel: Cars and Mobility

22 articles published on Huxiu. This content is authorized by the author for publication; the views represent the author only and do not represent Huxiu's position. If you have objections or complaints about this manuscript, please contact tougao@huxiu.com. Everyone who is changing and wants to change the world is on the Huxiu APP[ ]