Author | Wu Yuqing, Editor | Tian Zhe
"From automobiles to robots, what are automakers truly racing to seize behind this cross-border layout? "
BYD's (Build Your Dreams) humanoid robot may soon step out from behind the scenes and into the spotlight.
Recently, the official WeChat account of "BYD Di Space | Zhengzhou Pavilion," operated by BYD, hinted that the robot product will make its first public appearance in early August. Subsequently, Cailian Press reported that BYD has confirmed the humanoid robot product will officially debut in August.
This means that BYD's previously ongoing robot project may be publicly showcased for the first time in the form of a physical product.
In fact, BYD's foray into robotics is not a sudden impulse.
Public information shows that BYD's Embodied AI research team was established as early as 2022. Focusing on factory and industrial application scenarios, the team has been engaged in the R&D of process robots, intelligent collaborative robots, intelligent mobile robots, and humanoid-like robots. Previously, BYD was also reported to be advancing an internal humanoid robot project and investing in Agibot to step up its efforts in the embodied AI industry chain.
Regarding the application scenarios for robots, BYD has also unleashed greater room for imagination. Li Ke, Executive Vice President of BYD, previously stated the hope that every BYD dealership will deploy two to three robots in the future for vehicle model explanations, function demonstrations, and customer interactions.
However, BYD is not the only automaker moving closer to the robotics field.
Over the past few years, XPeng has released the humanoid robot IRON, while automakers such as GAC (GAC Group) and Dongfeng (Dongfeng Motor Corporation) have explored bringing robots into auto factories. A group of automotive industry talents has also started robotics companies.
Overseas, Hyundai Motor fully acquired Boston Dynamics, hoping to integrate robotics technology with its own manufacturing system. Toyota, on the other hand, has expanded its partnership with NVIDIA to combine autonomous driving, robotics, and smart factories.
For automakers, is making robots a distraction from their main business, or the next ticket to compete in the smart vehicle arena?
01. The Automotive Industry is Shifting Gears
BYD's frequent moves in the robotics track are a representative microcosm of automakers entering the robotics field in recent years.
Currently, there are several main paths for automakers to layout in robotics: one is to choose to develop robot products independently, hoping to extend the intelligent capabilities accumulated in the automotive field to robots; another is to secure a position in advance by investing in robotics companies; the third is to introduce robots into auto factories first to explore practical applications, while providing robots with real-world application scenarios and training data.
In terms of self-developed robots, XPeng is one of the earliest and most clear-targeted among domestic automakers.
In 2024, XPeng showcased the humanoid robot IRON at its Technology Day, discussing it alongside automobiles and autonomous driving under the same physical AI system. He Xiaopeng, founder of XPeng, once stated that in the future, AI (Artificial Intelligence) will not only exist in smartphones and computers but also needs to enter the real world, with automobiles and robots serving as crucial carriers for AI to connect with the physical world.
Unlike some companies that prioritize robot applications in factories, XPeng is more eager to explore the possibility of robots as future smart terminals. Previously, XPeng also revealed its hope to push robots into factories, commercial scenarios, and explore scaled production.
Li Auto also chose the self-development route, restructuring its intelligent department in February 2026 to form three teams: humanoid robots, software ontology, and foundation models, with the humanoid robot project codenamed "Nexus". Li Auto has planned two robot products, "Dual-Wheel" and "Bipedal", among which the "Dual-Wheel" robot, aimed at factory scenarios, is planned for public release within the year.
In 2026, Li Auto's R&D investment is expected to reach 12 billion RMB, with AI-related investments accounting for about 50%. Li Xiang, founder of Li Auto, once emphasized internally that 2026 is the last year for all companies aspiring to become top AI players to get on board.
If XPeng and Li Auto represent the extension of automotive AI capabilities to robots, then BYD's path leans more towards the manufacturing route.
BYD's early robotics layout focused more on its own manufacturing system. After the establishment of the company's Embodied AI research team, the focus has been on exploring directions such as process robots, intelligent collaborative robots, and intelligent mobile robots, which are naturally related to BYD's massive automobile and battery manufacturing system.
In BYD's view, robots can first serve its own production system. As robotics technology matures, the scope of application may further expand. Li Ke, Executive Vice President of BYD, previously stated that if robots enter household scenarios in the future, BYD also has the capability to promote robots through its own sales network.
In addition to independent R&D, some automakers choose to layout the robotics industry chain in advance through investment and cooperation.
For the humanoid robot industry, which is still in its early stages, building a complete team from scratch is not easy. Therefore, investing in robotics companies has become a way for some automakers to enter this field. BYD's previous investment in Agibot is part of its layout in the robotics ecosystem.
NIO's attitude, however, appears more conservative, preferring to wait and see before entering the market. In April this year, William Li stated in a media interview that NIO has made multiple investments in the embodied AI field through NIO Capital. "We will not rush to the forefront in everything as we did in the smart vehicle industry chain. Please allow us to adopt a strategy of waiting and observing before entering this matter, learning first before entering."
Meanwhile, robots are also entering the automotive manufacturing sites.
Compared to household and consumer scenarios, auto factories possess more standardized environments and clear task requirements, making them one of the easiest scenarios for humanoid robots to land. Currently, automakers such as Dongfeng and GAC are exploring the application of robots in the manufacturing process, hoping to verify robot capabilities through real production environments.
Beyond corporate-level layouts, talents in the automotive industry are also migrating to the embodied AI field.
Over the past few years, the smart vehicle industry has accumulated a group of talents with experience in visual perception, autonomous driving algorithms, motion control, and hardware R&D. These capabilities highly overlap with the technical systems required by robots.
Among them, Peng Zhihui, founder of Agibot, was previously a member of Huawei's "Genius Youth" program. After leaving Huawei, he entered the humanoid robot entrepreneurship field. Liu Fang, former head of autonomous driving at Xiaomi EV, has also entered the robotics startup track to layout intelligent robot business. Chen Yilun, former CTO of the autonomous driving system at Huawei's Intelligent Automotive Solution BU, has also entered the embodied AI field to start a business, founding Tashi Robotics. Yu Yinan, former head of intelligent driving at Horizon Robotics, has also left his original position to devote himself to embodied AI entrepreneurship.
Meanwhile, AI and robotics-related talents cultivated internally by smart vehicle companies such as Li Auto and XPeng are also beginning to flow into the embodied AI field.
According to the latest exclusive news from Leiphone, Shi Xiaoxin, one of the core leaders of XPeng's robotics business, left XPeng Robotics in early June and recently founded a startup company "Aimo Robotics," entering the home robotics direction.
In addition to the robot bodies and related talents, the manufacturing scenarios and technical capabilities accumulated by the automotive industry are also becoming an important foundation for entering the embodied AI field.
From independently developing robots to investing in robotics companies, and then to exploring the integration of robots into production processes, the automotive industry is entering the embodied AI industry chain in various ways.
02. What Gives Automakers the Confidence to Build Robots?
Automakers are flocking to the robotics field not just because humanoid robots have suddenly become a trending topic.
In the view of many automakers, robotics is not a completely unfamiliar new industry. Over the past few years, the technology, supply chains, and application scenarios accumulated by the automotive industry around intelligence are increasingly intersecting with embodied AI.
To some extent, smart vehicles and robots are solving the same problem: how to enable machines to understand the real world and take actions based on the environment.
In the past, automakers competed on engines, batteries, and mechanical manufacturing capabilities. Entering the era of smart vehicles, the competition has gradually shifted to perception, algorithms, and data closed loops. These capabilities are equally critical for robots to achieve intelligence.
In the view of XPeng, which has already released humanoid robot products, automobiles, Robotaxis, and humanoid robots are not three completely independent directions, but different carriers for AI to enter the real world. He Xiaopeng, founder of XPeng, has mentioned the concept of physical AI multiple times, hoping to enable AI to not only stay in the digital world but also possess the ability to understand and manipulate the real environment.
He Xiaopeng once stated in an interview with Xinhua Finance: "Physical AI is about giving things in the physical world a brain, mobility, and manipulation capabilities. Humanoid robots, Robotaxis, and flying cars are essentially different product forms of physical AI."
This is also the reason why XPeng explores the humanoid robot IRON alongside smart vehicles and autonomous driving under the same technical system.
Automobiles need to perceive the road environment through sensors such as cameras and LiDAR, and then complete decision-making and control through algorithms. Similarly, robots need to perceive their surroundings and complete actions through motion control. Although the application scenarios are different, the two share certain commonalities at the perception, computation, and execution levels.
Wang Lei, CFO of Horizon Robotics, stated at this year's Caixin Summer Summit: "The embodied AI robot industry chain and the automotive industry chain are highly synergistic. The automotive industry itself has strong manufacturing, supply chain, and quality management capabilities, so expanding its business footprint to the robotics field is a natural progression." He also revealed that many companies that originally supplied core components such as thermal management for Tesla vehicles "have now also become supply chain companies for Tesla's robots."
In addition to technical capabilities, another advantage of automakers lies in manufacturing scenarios.
For humanoid robots, the biggest challenge is not just building the robot, but finding truly viable application scenarios, which is the key to achieving a commercial closed loop. Currently, auto factories may be the most realistic first step for robot commercialization.
Compared to open environments such as homes and shopping malls, automotive manufacturing scenarios are more standardized, and tasks are easier to break down. Robots can start with relatively clear jobs such as inspection, material handling, and quality inspection, continuously training and optimizing in real production environments.
This is why many automakers choose to let robots enter their own factories first.
For example, Dongfeng Liuzhou Auto once cooperated with UBTECH to introduce humanoid robots into automotive manufacturing scenarios for production assistance. GAC is also exploring the integration of robots with its own manufacturing system, hoping to enable robots to play a role in industrial scenarios.
For automakers with large-scale manufacturing systems such as BYD and XPeng, the factory itself is the easiest place to verify the value of robots. Compared to seeking external customers first, utilizing their own production systems to test robot capabilities can reduce uncertainties in the early stages of commercialization.
In addition to manufacturing capabilities, the automotive industry has another advantage that is difficult for robotics companies to replicate: real-world data.
Compared to traditional industrial robots, humanoid robots need to face more complex environments. They not only need to complete mobility but also understand space, recognize people and objects, and perform tasks such as grasping and manipulation. The training of these capabilities is inseparable from a large amount of real-world data.
Han Feng, COO of Ruqi Travel, told Leiphone that autonomous driving itself is rooted in real physical scenarios and is an important branch of broad physical AI. Currently, embodied AI is ushering in rapid development, and such technologies are also highly dependent on high-quality data. "We reviewed the data closed-loop engineering system built over the past two to three years and found that the underlying tool chains possess universality. Whether it is autonomous driving or embodied AI, the underlying layer cannot do without the four major links of data collection, processing, annotation, and model evaluation. The entire set of technical tools can be reused."
However, autonomous driving data cannot be directly copied to robots. Autonomous driving data can support embodied AI in completing navigation capability training in parks and open roads, but embodied AI more urgently needs manipulation data such as robotic arms, robot operation actions, and motion trajectories.
Han Feng told Leiphone that, based on this point, Ruqi is exploring extending its data capabilities to the embodied AI field. "From the perspective of the embodied AI field, the scarcity and value of high-quality real data are more prominent. Ordinary embodied AI companies find it difficult to enter real industrial production lines to collect data, and there is a very high threshold for cooperation here. First, most companies and large manufacturing enterprises do not have a foundation for deep cooperation, and the other party will not easily allow robots to enter the production line to collect data; second, external robots cannot assist in completing production line operations, and there is also a huge risk of data leakage. This cooperation threshold itself is a moat."
From technical capabilities to manufacturing scenarios, and then to real-world data, the resources accumulated by the automotive industry over the past years are becoming an important foundation for automakers to enter the robotics field.
But this does not mean that automakers can naturally become robotics players. What automakers excel at is manufacturing and engineering, while for humanoid robots to truly achieve widespread adoption, issues such as AI model capabilities, generalization in complex environments, and cost control still need to be resolved.
Overall, for automakers, robots represent an extension of intelligent capabilities into a broader physical world.
03. Will Automakers Replicate the Success of Car Manufacturing When Building Robots?
However, automakers entering the robotics field does not mean they can simply replicate the successful path of the NEV (New Energy Vehicle) era.
Compared to automobiles, robots face a more complex and also more open competitive environment.
He Xiaopeng, founder of XPeng, once stated that the difficulty of humanoid robots is even higher than that of building cars.
The reason is that compared to mature industrial products like automobiles that have been developed over a century, robots not only need to complete mobility but also possess the ability to understand the environment, plan actions, and execute tasks.
He Xiaopeng also clearly reminded the industry during the Two Sessions in 2026: "Not all automakers can smoothly transition into humanoid robot R&D companies." He pointed out that the R&D difficulty of humanoid robots is tens of times that of automobiles, and truly usable robots need to meet or even exceed automotive-grade standards. At the same time, he called on the industry to focus on the construction of the robot's "brain" (autonomous thinking and decision-making) and "cerebellum" (motion control) capabilities, "rather than rushing to build bodies and joints."
A car only needs to complete driving under relatively clear traffic rules, while the real world faced by robots is more complex. For the same action, such as picking up a cup, a robot not only needs to recognize the position of the object but also needs to judge its weight, material, grasping force, and how to avoid collisions.
Elon Musk also stated during Tesla's Q2 2026 earnings call that the mass production of humanoid robots is "the most difficult product Tesla has ever mass-produced." He explained, "Everything on the robot is new." For electric vehicles, Tesla can procure specific components like wheels, side mirrors, and windshields from existing suppliers, "but for Optimus, there is no supply chain."
This means that the past advantages of automakers in manufacturing, supply chain, and engineering can help robots move towards productization faster, but cannot directly solve all the problems of robot intelligence and general capabilities.
This is also why the first step for most automakers to explore robots is still concentrated in relatively closed scenarios such as factories and logistics.
Compared to home environments, automotive manufacturing scenarios are more standardized, task boundaries are clearer, and it is easier to verify whether robots possess practical value.
Therefore, whether it is Tesla's Optimus, or products like XPeng's IRON and GAC's GoMate, they are currently focusing on exploring scenarios such as industrial manufacturing and commercial services.
But even for companies with advantages in automotive manufacturing, AI, and supply chains, humanoid robots are still a distance away from large-scale application.
Taking Tesla's Optimus as an example, Musk has stated multiple times that he hopes the robots can enter Tesla factories in the future to undertake production tasks and become an important business for the company. However, Optimus is currently still mainly in the R&D and testing stages, and there is still a process before large-scale deployment.
Musk also revealed that Optimus involves tens of thousands of independent components, each of which must be accurate to achieve ramp-up mass production. The initial speed will be very slow because tens of thousands of independent problems need to be solved one by one to achieve large-scale mass production. He admitted that there is an obvious difference between manufacturing a small number of robots and achieving large-scale mass production. Whether Optimus can reach high production capacity will depend on the improvement of production efficiency and the supply capacity of key components. External expectations suggest that the timeline for true large-scale commercialization may fall between 2027 and 2028.
Domestic companies face similar challenges. XPeng has already showcased IRON and is exploring scenarios for robots entering factories and stores. However, from showcasing prototypes to stable operation, and then forming a scaled business model, issues such as reliability, cost, and maintenance still need to be resolved in the middle.
Among them, cost is an unavoidable issue for robot commercialization.
Currently, the pricing of humanoid robot products is still in the exploration stage. Some companies have already launched products priced at the ten-thousand-yuan level. The public price of Unitree G1 is about $16,000. Tesla's Optimus has not yet been officially sold, but Musk once proposed a future target price of about $20,000 to $30,000.
But for enterprise users, purchasing the equipment is just the beginning. Subsequent software training, scenario adaptation, maintenance costs, and long-term operational stability will all affect the final return on investment.
If robots cannot form obvious advantages in efficiency, cost, or safety, it will be difficult for them to truly enter large-scale production processes even if the technology is mature.
Over the past few decades, the automobile has been one of the most complex consumer goods in the industrial era. In the AI era, automakers are trying to extend their capabilities from manufacturing smart cars to manufacturing intelligent machines that can understand and act.
But this road will not be smooth. Whether the robotics industry can ultimately form a scaled market like NEVs still requires time to provide the answer.