The Xiaopeng IRON robot has truly become popular worldwide.
Just now, a chat exclusively learned by Sohu Technology showed that Musk commented on the Xiaopeng robot IRON, saying, “Not bad. The entire robot market will be dominated by Tesla and China.” At the same time, Musk also expressed appreciation for the robot’s feminine qualities. He spoke frankly: “Other robot companies in the West are still weak, but I respect China as a competitor because there are too many smart and hardworking people in China
Recently, Xiaopeng Motors officially released the new generation of humanoid robot IRON. Unlike other robots on the market, the new generation IRON has typical female characteristics.
At first, everyone did not have high expectations for this robot, but when the new generation IRON took a “catwalk” on site, the crowd was completely boiling.
Subsequently, the question “Is Xiaopeng Robot a real person pretending to be?” and related entries quickly sparked a lot of discussion on the internet, with many netizens sharing their observations of small details at the press conference, from ear contours to rib arches, attempting to prove that the robot really had a woman under its skin.
Faced with the overwhelming discussion, He Xiaopeng first replied to the comments from netizens: Thank you for your recognition, and then posted a one shot video without editing on his personal account to prove the authenticity of the IRON robot. In the video, the staff unzipped the back of the robot and displayed the internal structure of the robot controller, transport fixator, and other components in all directions.
He Xiaopeng publicly cut open the leg cover of the IRON robot while it was powered on, showcasing its lattice muscles and mechanical structure on site, and expressed hope that this would be the last time he was forced to self verify. This is also the first time in history that robots have proven themselves to be non-human.
On the evening of November 6th, Xiaopeng’s US stock rose over 9%, with the largest intraday increase reaching 14%. Stimulated by the press conference and the sharp rise in the US stock market, the Hong Kong stock market of Xiaopeng Motors opened higher than 4% in early trading today.
Musk likes, mass production will be achieved by 2026
As stated on the official website of Xiaopeng Motors, ‘Committed to creating the most humanoid robot’, the new generation of IRON is more humanoid in all aspects.
IRON adopts a biomimetic structure design of “bone muscle skin”. Visually, the entire body of IRON is covered with a honeycomb shaped 3D printed “muscle” skeleton and flexible tactile skin, greatly reducing mechanical sensation. Its hands are crafted in a 1:1 ratio to human hands, capable of performing delicate movements such as grabbing eggs and twisting bottle caps.
The body has a biomimetic spine that can bend down and pick up objects like a human. The full body flexibility is similar to that of humans, which also allows it to walk smoothly on the “cat step” during display.
IRON has approximately 82 degrees of freedom throughout its body. Its hands have 22 degrees of freedom, only 5 less than humans. The joint drive adopts a micro harmonic reduction motor, combined with Xiaopeng’s three electric system technology, to ensure high-precision movement.
Surface flexible skin with tactile sensors arranged inside, capable of sensing external contact; Visually, the use of multi camera cameras is based on the “Eagle Eye” panoramic vision system of Xiaopeng Motors for environmental perception. The body material is mainly composed of lightweight honeycomb mesh components, which balance strength and flexibility, and is equipped with an all solid state battery.
IRON is equipped with three self-developed Turing AI chips, with a computing power of up to 2250 TOPS. It runs Xiaopeng’s independently developed physical world model, which can achieve end-to-end control from vision to action, making dialogue, walking, interaction, etc. more natural. He Xiaopeng stated that it is not only the world’s most powerful humanoid robot, but will also be one of the smartest robots next year.
According to He Xiaopeng, the IRON will achieve mass production by the end of 2026. This generation of IRON is the last version of pre research, and the mass production version will not exceed 170cm.
Xiaopeng’s Robot Dream
In fact, Xiaopeng started laying out embodied intelligence as early as 2020. According to sources, He Xiaopeng began to pay attention to the field of robotics as early as 2016 and continuously researched related technologies. The real landing action was in 2020, when Xiaopeng acquired Dogotix, a quadruped robot company founded by Zhao Tongyang, and established Pengxing Intelligence, officially entering the robotics industry.
Sohu Technology learned that in 2021, the Xiaopeng robot team expanded to over 300 people, and in September of the same year, the first intelligent machine Ma “Xiaobailong” was launched. The following year, Pengxing Intelligent completed a financing of over 100 million US dollars, setting a record for a single round of financing in the field of biomimetic robots in China at that time.
In 2023, due to strategic adjustments and other reasons, Zhao Tongyang, the team leader of Pengxing, resigned. He and his former Dogotix partners reorganized a team and founded the Zhongqing Robot.
In April 2023, Pengxing Intelligence began accelerating the development of bipedal humanoid robots. In October of the same year, Xiaopeng released the first humanoid robot PX5. In September 2023, Xiaopeng Motors acquired the remaining shareholders of Pengxing Intelligence for $98.96 million and integrated Pengxing Intelligence into an internal business unit to fully take over the robotics business. Since then, Xiaopeng Motors has officially entered the market to make robots.
In November 2024, Xiaopeng Motors released the fourth generation humanoid robot Iron, which was planned to be used for factory production tasks. At that time, it was expected to launch the fifth generation humanoid robot for industrial and commercial scenarios in 2025. In the past few years, due to route and organizational adjustments, the size of the Pengxing team has been reduced to at least less than 100 people.
However, with the booming of the embodied intelligence industry and intensified competition, the Xiaopeng robot team is also constantly recruiting personnel. Xiaopeng Motors has informed Sohu Technology that the current number one position in the robotics team is still held by Mi Liangchuan, who serves as the Vice President of Xiaopeng Motors Robotics and the head of the AI Technology Committee.
According to LinkedIn data, Mi Liangchuan has been serving as the Vice President of Xiaopeng Motors since March 2021, responsible for the Chairman of the Artificial Intelligence Technology Committee and the leadership of the Robotics Division.
Mi Liangchuan graduated from Iowa State University in 2005 with a Master’s degree in Robotics and a Bachelor’s degree in Electrical Engineering from the University of Science and Technology of China.
After graduation, Mi Liangchuan joined NVIDIA as a senior software engineer and was later promoted to Senior Manager of Android Software at NVIDIA, while also responsible for the Android platform department in China. During his time at NVIDIA, Mi Liangchuan reported to Wu Xinzhou, Vice President of NVIDIA Global. In 2015, Mi Liangchuan co founded Miivii Dynamics and served as its Chief Technology Officer, dedicated to the research and development of robot platforms.
From 2016 to 2019, he served as a Senior Algorithm Engineer for Autonomous Driving at NVIDIA Santa Barbara. From 2019 to 2021, Mi Liangchuan went to NVIDIA’s headquarters in San Jose, Santa Clara County, San Francisco Bay Area to serve as the Senior Director of Autonomous Driving.
Several senior headhunters familiar with Xiaopeng have told Sohu Technology that Xiaopeng has long been laying out in the field of embodiment and has been poaching people recently. They dug up Tencent’s youngest T12 and step star big shots. ”A headhunter said. Another headhunter told Sohu Technology, “There will be more people joining in the future, and they are very willing to spend money on it
The headhunter mentioned Ge Yixiao, former head of Tencent Vision&Multimodal Basic Model, and Zhou Shuchang, founding team member of Step Leap Star.
Ge Yixiao graduated from Huazhong University of Science and Technology with a bachelor’s degree and later obtained a doctoral degree from The Chinese University of Hong Kong. After graduation, I joined Tencent and was promoted to Tencent T12 Technical Expert at the age of 28. As the main person in charge of multiple projects in the multimodal direction, I have won the Tencent Technology Breakthrough Award for two consecutive years. Has multiple international patents and has published over 30 papers in top conferences and journals such as NeurIPS, ICLR, CVPR, ICCV, ECCV, ICML, etc. Recently, Ge Yixiao joined Xiaopeng Intelligent Mimicry Department, focusing on multimodal robotics. Her research directions include embodied intelligence native multimodal large models and world models.
Dr. Zhou Shuchang graduated from the Chinese Academy of Sciences, and his master’s degree was from Tsinghua University and the National University of Singapore, respectively. He worked at Google, Broadview Technology and Step Star successively. After joining Xiaopeng Motors, Zhou Shuchang will serve as the Senior Director of Autonomous Driving Algorithms and report to Liu Xianming, the person in charge of autonomous driving at Xiaopeng Motors.
Xiaopeng also told Sohu Technology that although the administrative division is not within the robotics department, the company’s power, chip, and AI teams are all working together on robotics.
Is it difficult for robots to walk cat steps?
The most attention grabbing feature this time is still the elegant gait of the robot, for which Sohu Technology has asked robotics industry professionals about this technology.
The founder of a well-known robotics company told Sohu Technology, “This is not difficult, mainly for gait control, and hardware also needs to add some degrees of freedom
The passive degree of freedom of the toes is indeed a key technology for the lightweight gait of Xiaopeng’s new generation humanoid robot IRON. By increasing the passive range of motion of the toes, the robot’s pace is closer to natural human movement. This design allows the toes to naturally swing without active driving, coupled with a biomimetic structure with 82 degrees of freedom throughout the body, enabling the IRON to complete a gentle walk similar to a “cat step”.
Previously, the well-known use of hardware upgrades to enhance anthropomorphism was the mechanical arm of the Stardust intelligent robot. A Stardust intelligent algorithm engineer told Sohu Technology, “Hardware is a very important part. For example, our robot is rope driven all over its body, and the key to smoothness is to mimic human tendons
The engineer also said, “The mechanical design of Xiaopeng robots is really great, with a very complex screw configuration that can achieve very flexible movements. In addition, the gait should have been specially trained, but personally, compared to the difficulty of robots transitioning from stumbling to humanoid gait, the span from humanoid gait to cat gait is not very large
Another robot business algorithm engineer of an Internet giant told Sohu Technology: “The most difficult thing to do is to set the reward function. The core of good walking is the application of deep reinforcement learning.”
As long as the robust anthropomorphic walking is solved, encoding the walking style into the reward function later is relatively more of a strategic upgrade. It is more difficult to learn the anthropomorphic gait, and it will be easier to achieve what style of gait to walk on this basis
The traditional bipedal gait control, such as the zero moment point (ZMP) based control method, mainly aims to maintain static or quasi-static stability on flat ground. However, this approach is difficult to handle complex nonlinear systems with IRON 82 degrees of freedom in real-time and efficiently. To achieve smooth and robust “humanoid walking”, robots must continuously manage the high-dimensional dynamics of their entire body during motion and adapt to environmental uncertainty without relying on precise physical models.
Just now, Xiaopeng Motors officially released another video, responding to why IRON is so human like. Previously, reinforcement learning required a large amount of reinforcement at each joint, but now with the help of large models, we can see changes very quickly, “said He Xiaopeng
At the same time, he also pointed out that the key to cat walking is “twisting the hips”, because with a humanoid spine and high degrees of freedom, walking is more realistic. Imitation learning algorithms also enable the IRON robot to autonomously learn human action data and continuously adapt to optimization.
暂无评论内容