Exclusive from Intelligence Emergence: Unveiling XPeng Robotics - Recruiting a Big Shot from NVIDIA and Having Hundreds of Units on the Production Line

Text by | Qiu Xiaofen

Edited by | Su Jianxun

In recent exchanges between "Intelligent Emergence" and individuals in the field of embodied intelligence, XPeng Robotics was frequently mentioned.

"In terms of the level of domestic humanoid robots, XPeng can rank among the top three," a person in the embodied industry told us directly.

Compared with the bustling embodied entrepreneurship in China, XPeng Robotics seems a bit low - key. However, at XPeng Motors' earnings conference call on August 19th, He Xiaopeng also revealed that XPeng's fifth - generation humanoid robot will be unveiled at the "XPeng Technology Day" on October 24, 2025, and mass production will be achieved in the second half of 2026.

"Intelligent Emergence" exclusively learned that XPeng Motors' humanoid robot team began preparations around 2020, and currently, the person in charge is Mi Liangchuan. As we understand, Mi Liangchuan joined the NVIDIA team in 2005 and later served as a senior manager of Android software at NVIDIA in Santa Clara, California, while also being responsible for the Android platform department in China.

During his tenure at NVIDIA, Mi Liangchuan reported to Wu Xinzhou, the global vice - president of NVIDIA, and had experience managing a team of hundreds of people. Before joining NVIDIA, Mi Liangchuan was an alumnus of Class 9923 of the University of Science and Technology of China and once served as a researcher at Carnegie Mellon University.

XPeng has been quietly working in the robot track for a long time, even before this wave of embodied intelligence.

A person familiar with the matter told "Intelligent Emergence" that He Xiaopeng's attention to the robot track began as early as 2016. "He conducts serious research on the robot direction almost every year," and he once had a long conversation with Wang Xingxing, the founder of Unitree.

In 2020, He Xiaopeng took action, acquiring "Dogotix" founded by Zhao Tongyang and establishing "XPeng Robotics" to explore XPeng's robot business. The aforementioned person said that He Xiaopeng's contact with this team initially came from the introduction of Lei Jun, the founder of Xiaomi.

For this acquisition, XPeng also paid a considerable price - spending 100 million US dollars to buy out the shares of its investors. At its peak in 2020, the "XPeng Robotics" team had as many as 300 people.

However, due to reasons such as a conflict with He Xiaopeng's strategic direction, the core team of "XPeng Robotics" eventually left. Zhao Tongyang, the former team leader, founded "Zhongqing Robotics".

After 2023, He Xiaopeng introduced Mi Liangchuan to be in charge of the robot business. The aforementioned person told us that in the past two years, XPeng Motors' robot team has undergone a continuous adjustment. "The team size was once reduced to 70 people, and then rebuilt to the current scale of about 200 people." In the domestic field of embodied intelligence, this team size can be considered a luxurious configuration.

A person familiar with the matter told "Intelligent Emergence" that in terms of robot strategy, He Xiaopeng highly follows Elon Musk. He Xiaopeng once said that for XPeng Motors, robots are its "third growth curve" after AI + intelligent driving and globalization.

Actually, entering the robot field from the automotive industry is a logical choice.

On the one hand, intelligent driving is the smallest implementation unit of embodied intelligence. The technology chain of "perception - decision - execution" in the automotive field is highly homologous to that of robots.

Taking Tesla's Optimus as an example, it reuses about 60% of the algorithms of Tesla's FSD system. The robot Iron previously released by XPeng also reuses XPeng Motors' 720° Eagle Eye system.

He Xiaopeng previously said that humanoid robots and automotive intelligent driving can promote each other's evolution in a homologous architecture.

On the other hand, there is also room for synergy between the automotive and robot industries in terms of the supply chain.

Most of the core components in the automotive field, such as power batteries, sensors, chips, lidars, and cooling systems, can be applied to the robot field, reducing the cost of robots. On Tesla's Optimus, the homologous battery pack of Model Y was directly used before.

In terms of products, after years of hard work, XPeng Motors has released several robot products - including the first bipedal humanoid robot PX5 in October 2023 and the fourth - generation humanoid robot Iron released in November last year.

According to He Xiaopeng's introduction in the conference call, XPeng's fifth - generation humanoid robot highly overlaps with XPeng Motors' existing technology stack.

For example, in terms of the core processor, the fifth - generation robot will be equipped with XPeng's self - developed Turing AI chip to enhance edge computing power and support small reinforcement learning models and segmented end - to - end architectures.

In terms of models, its robots will also be equipped with XPeng's physical world base large - model VLA architecture, sharing cloud - based AI infrastructure with cars.

In addition, XPeng's fifth - generation humanoid robot also reuses a large amount of basic technology from XPeng Motors.

Specifically, the EEA architecture of the robot is derived from XPeng Motors' EEA architecture; the joints of the robot come from XPeng Motors' three - electric system; the "brain" of the robot comes from XPeng Motors' intelligent driving team - overall, the robot and the automotive business are 70% homologous in terms of technology.

With technology and products ready, XPeng Robotics is also exploring the possibility of implementation.

A person familiar with the matter revealed to "Intelligent Emergence" that currently hundreds of humanoid robots have been deployed on XPeng Motors' production line. However, not all of these robots are currently used to assist automotive production and manufacturing. Instead, most of them are used for algorithm tuning, remote operation data collection, etc. "The deployment of hundreds of robots is already a large number among car manufacturers," the aforementioned person said.

end

end