The rise of embodied intelligence and robots has heated up this year’s supply chain.
On May 26th, at Alibaba Center T2 in Shanghai, Tuesday’s “Shanghai Tailored Intelligent Core Components · Exhibition and Exchange Day” can be said to be the most informative day of the entire muShanghai Robot Theme Week.
In the morning, seven Chinese companies took turns taking the stage, and in the afternoon, there were several keynote speeches and roundtable discussions. There are many geeks, entrepreneurs, and investors sitting in the audience, with mixed Chinese and English questions emerging one after another.
Among the companies present, there are three that specialize in tactile sensors, one in joint modules, one in IMU and motion capture, one in whole machines, one in binocular vision, and one in brain computer interfaces. On the surface, it looks a bit like a parts company roadshow. And when they make a concentrated appearance, we will find that China’s robot supply chain is upgrading from simple hardware components to data, tool chains, and infrastructure.
The robot supply chain is also joining the story of Chinese enterprises going global. It is reported that the North American team of Yuejiang will expand threefold by the end of the year. Dai Shi also said that the motion capture system will be targeted at international customers in the future, and many companies will conduct on-site roadshows in English, which shows their willingness to go global.
36Kr “Workplace Bonus” (ID: ZhiChangHongLi)
[Touch: Three Ways of Living for Three Companies]
Touch is the most difficult link in the robot industry chain. In the morning session, three companies were simultaneously discussing touch, but their methods of operation were completely different.
The ‘Daimeng Robot’ has chosen the route of ‘monochromatic light+pattern encoding’. By using black and white cameras, the system bandwidth and power consumption structure are greatly optimized while reducing computing power requirements, achieving millimeter level thickness design and a lifespan of up to 5 million times. At present, the effective sensing units per square centimeter of the sensor have reached 40000, balancing high resolution and full modality, and have achieved IP67 waterproof and dustproof levels, meeting the long-term reliable operation requirements under complex working conditions.
Daimon has also launched the world’s largest tactile full modal physical world embodied dataset, “Daimon Infinity”. The first batch of 10000 hours of real-world data has been launched on the Alibaba Magic Community, including over 2000 human skills and more than 300 real items, covering 16 industries and 80 real-life scenarios. According to Dai Meng’s introduction, this dataset currently ranks first in the download volume of the embodied dataset in the world of magic physics. Dai Meng plans to expand its scale to millions of hours within this year.
At present, Daimeng has established a multi-level data collection system. In terms of remote-controlled data acquisition equipment, the system integrates force/tactile feedback and range extension functions, supporting heterogeneous control architectures – not only can it control Daimeng’s own body, but it can also adapt to mainstream humanoid robots such as Yushu and Zhiyuan.
In terms of non physical data collection equipment, Daimeng provides a highly portable solution, where data collectors are not limited by fixed locations and can flexibly carry out collection work in various indoor and outdoor environments, even extending to complex outdoor scenes. All collected data comes with tactile tags, achieving multimodal high-precision alignment and providing a solid foundation for building large-scale tactile data bases.
In the afternoon, Zhang Dong, the head of commercialization at Daimeng, further supplemented the logic of the data layer. The biggest bottleneck in the current robotics industry is the lack of tactile rich information, which is essential for both large and world models. ”Daimeng is breaking through the data loop through scenario based data cognition, defining workflows, and edge deployment, and has launched a series of products including remote control data acquisition devices with force/touch feedback, two finger grippers, and five finger data acquisition gloves.
This means that in Dai Meng’s business model, hardware is the entrance, and data closed-loop is the barrier.
36Kr “Workplace Bonus” (ID: ZhiChangHongLi)
The ‘Vision Tower Robot’ emphasizes that it is a ‘full solution’ tactile supplier. Their team comes from universities in Hong Kong, Singapore, and Shenzhen, and their technical route is to use special photosensitive materials for surfaces with high linearity, and solve problems through optical imaging. The five finger fingertip model displayed on site has a calculation board of only 5 × 5 centimeters, which can be directly embedded in the robot’s palm to complete end-to-end calculations without the need to transmit data back to the robot’s main control.
The product line of Vision Tower covers flat surfaces, curved surfaces, fingertips, industrial grippers, and even fabric solutions. Nearly 1000 sensing points are placed on a piece of cloth, allowing toys or furniture to perceive interactions of different forces such as tapping and caressing. At the same time, we provide a complete set of data collection and scene implementation solutions, including fabric data collection and optical material data collection.
According to Shita, they have currently supplied in bulk to multiple domestic manufacturers of head dexterous hands and grippers.
The perspective of ‘One Eye Technology’ is broader, targeting ‘General Physics AI’. Zhang Wenbin, CMO of Yimu Technology, believes that when a person touches a key in their pocket, they do not need eyes and can quickly model it through touch. Robots should also have this ability. The tactile sensor developed by Yimu is ultra-thin and self calibrated, and the surface gel can be replaced.
In the video played on site, the robot completed laboratory liquid operations. It can determine the position of the test tube, pour the liquid smoothly, transfer the liquid accurately with a pipette, and even achieve dropwise replenishment at 22 weighing positions. Zhang Wenbin mentioned that through the simulation training of Sim to Real and Real to Real, the path of using simulation data to supplement the gap in real tactile data has been achieved. Building an end-to-end ecosystem with touch is its proposed collaborative vision.
Three companies, three paths. Dai Meng sells data in a closed loop, and the engineering of the sales side of the visual tower is integrated. It can be seen that there is no unified standard for the tactile track, but various personnel in the supply chain have already arrived in advance.
36Kr “Workplace Bonus” (ID: ZhiChangHongLi)
[Joint Module: Scene Differentiation]
The joint module is the most fundamental unit of the execution layer. The ‘Titanium Tiger Robot’ attempts to prove that in the harmonic module category, Chinese companies can not only produce standard products, but also customize delivery in extreme scenarios.
Titanium Tiger Robot “was established in 2020, specializing in harmonic joint modules and complete machine platforms. It has independently developed over 50 types of harmonic modules, achieving lightweight under the same torque. We have developed collaborative arms, dexterous hands (6 degrees of freedom), and full machine humanoid robots based on modules. The product line has clear division of labor: T140A is 1.4 meters tall and focuses on education; The T170 series is aimed at unmanned supermarkets and smart kitchens; T230C is used for factory handling, with a maximum load capacity of 40 kilograms on both arms.
The differentiation of Titanium Tiger lies in its deep customization ability. They delivered IP67 underwater modules and extreme environmental modules ranging from minus 40 degrees Celsius to 80 degrees Celsius, with clients including leading enterprises and shipping companies. Its R&D team currently has about 200 people, most of whom are hardware engineers. This year, Titanium Tiger has launched a manufacturing center with an annual production capacity of 100000 joint modules.
[Whole machine: 199000 yuan from Yuejiang]
Yuejiang Robot is the only complete machine factory that has been launched on the market in the morning. In December 2024, Yuejiang was listed on the Hong Kong Stock Exchange with the stock code 2432.HK, known as the “first stock of collaborative robots in China” in the industry. According to data from Zhuoshi Consulting, based on the shipment volume in 2023, Yuejiang ranks among the top two in the global collaborative robot industry and first in China, with a global market share of 13%. As of 2024, the cumulative shipment has exceeded 80000 units, with customers including 80 Fortune 500 companies such as BYD, Tesla, Mercedes Benz, Porsche, CATL, Foxconn, etc.
36Kr “Workplace Bonus” (ID: ZhiChangHongLi)
Sharing person Joe Zhang is responsible for the North American education market. He reviewed the path of Yuejiang from starting with a small desktop robotic arm on Kickstarter in 2015 to entering the industrial field in 2019. Yuejiang has a team of over 20 people in North America and plans to triple it by the end of this year. In addition, the most significant move of the year was the release of the full-size humanoid robot Dobot Atom, with a pre-sale price starting at 199000 yuan – the price anchor for full-size humanoid robots has been redefined. Perhaps the logic of ‘price butcher’ is also applicable in industrial markets.
As the Chinese company with the largest shipment volume in the field of collaborative robots, Yuejiang is shifting its cost control capabilities from industrial arms to humanoid machines. According to official disclosure, Yuejiang has launched scene cooperation with domestic first tier car manufacturers, electronic manufacturing plants, and coffee and milk tea shops, with plans for mid year trial production and mass production.
IMU and motion capture: calibration bottleneck of Dai Shi
Dai Shi Intelligence is headquartered in Jiading, Shanghai, with its factory located in Taicang, Jiangsu. They have been producing IMU (Inertial Measurement Unit) for 11 years. Its key equipment and core calibration process are fully self-developed, breaking through the bottleneck of military IMU calibration process, with an annual production capacity of 5 million sets, and certified by the vehicle specification system.
Dai Shi’s products are directly integrated into mainstream domain controllers such as NVIDIA and Horizon Robotics. Customer coverage includes L4/L5 autonomous driving, robotic dogs AGV、 Consumer grade robots and full body humanoid robots, as well as domestic leading humanoid robot manufacturers such as Zhiyuan, Luming, Yunshen, and Ubiquitous, have all used their IMU modules.
Dai Shi showcased the IMU based motion capture system “Ink Shadow” on site. Compared to optical motion capture, inertial motion capture has a lower price (in the tens of thousands of yuan range, while optical motion capture typically costs millions) and no obstruction restrictions, achieving sub millimeter level accuracy. At present, there are over a thousand domestic orders for robot data collection, remote operation, gaming, film and television, and big health fields.
The speaker Dai Shi mentioned a detail: IMUs on humanoid robots are usually installed in three positions: mouth (integrated with binocular vision), chest (inside the motherboard controller), and large part (lower body rotation posture monitoring). Wheeled robots have one chassis and one upper body. The output frequency of the Dai Shi IMU reaches 500 to 1000 hertz.
Dai Shi’s logic is clear: IMU is the core of robot motion perception, and calibration accuracy determines the reliability of motion data. They first solved the bottleneck of the calibration process, and then extended to downstream motion capture systems, essentially selling the production capacity of “high-precision motion data”.
36Kr “Workplace Bonus” (ID: ZhiChangHongLi)
[Brain Computer Interface: Clinical Pathways and Tailored Integration of Surgical Innovation]
The landing scenarios of embodied intelligence are not only in factories, but also in hospitals.
The concept of brain computer interface was first proposed in 1973, aiming to connect the human brain with external intelligent devices. Musk’s Neuralink has propelled the transition from laboratory to industry, with over 20 participants now able to perform simple daily operations through invasive brain computer interfaces.
We know that embodied intelligence emphasizes that intelligent agents learn and complete tasks through interaction with their bodies and environment, while brain computer interfaces are a technology that reads neural activity in the brain and translates it into instructions that computers or machines can understand. When two technologies are combined, human intent, perception, and intelligence can be directly injected into external devices, achieving “human-machine integration”.
Arfysica Innovation was founded in 2016, focusing on non-invasive brain computer interfaces. As a national high-tech enterprise, Shuli Innovation is also one of the national level specialized and innovative small giant enterprises in the field of brain computer interfaces. The company has established the Shanghai Key Laboratory of Brain Computer Collaboration and Cognitive Expression and the Shanghai General Surgery Intelligent Medical Brain Science Research Institute, and has completed Series C financing. The company participated in the development of standards in the field of brain computer interfaces, with Dr. Wang Wei serving as an expert member of the first Brain Computer Interface Standardization Committee of the Ministry of Industry and Information Technology.
36Kr “Workplace Bonus” (ID: ZhiChangHongLi)
The morning sharer Cui Jun explained the differences between different technological routes of brain computer interfaces, as well as the enormous value demonstrated by non-invasive brain computer interfaces in the field of motor nerve function remodeling. In the afternoon sharing, Vice President Jin Hui further elaborated on the integration path of brain computer interface technology and embodied intelligence, as well as their application scenarios in the medical field.
Surgical innovation chooses non-invasive routes, capturing motor imagery consciousness through EEG caps, driving exoskeletons to assist patients in executing movements, and reopening neural pathways. The core equipment includes an active dry electrode EEG cap (eliminating the preparation of conductive adhesive for traditional wet electrodes), a central computing system, and a multi joint independent exoskeleton, with a decoding success rate of up to 95%. In terms of clinical implementation, the product has entered top tier tertiary hospitals such as Beijing Temple of Heaven, Peking Union Medical College, Xuanwu, Wuhan Union Medical College, and Shanghai Changhai.
Jin Hui highlighted the fusion demo of brain computer interface and embodied intelligence in the afternoon. For example, by combining non-invasive brain computer interfaces with MR devices, engineers have achieved brain controlled movement of robotic dogs and brain controlled rotation of wheelchairs. This system relies entirely on brainwave control and does not require manual or eye movements.
In the medical application scenario, surgical innovation covers neurological rehabilitation for post-stroke hemiplegic patients, early diagnosis of Alzheimer’s and Parkinson’s diseases, as well as treatment of neurodevelopmental disorders in children, including ADHD, autism, etc.
[binocular vision: a low-key company that did not go on stage]
In addition to the seven companies that shared on stage and the on-site exhibition area, we also discovered a company that focuses on binocular stereo vision.
In the perception layer of embodied intelligence, vision, touch, and IMU are three parallel paths. Robots need both fine tactile feedback and visual depth perception to understand the physical world. The essence of binocular vision is to calculate depth by simulating human eye disparity, providing distance and position information in three-dimensional space for robots.
Yuanxiang Technology has independently developed binocular stereo vision chips and modules, covering from automotive grade to consumer grade. Its products have been applied in scenarios such as lawn mowing robots, drones, and intelligent warehousing. Its core team has over 15 years of experience in developing binocular technology and over 10 years of experience in developing car stereo products. The company’s headquarters and R&D center are located in Beijing, with branch offices in Wuhan, Suzhou, Mianyang, Ji’an, Jiaxing, and Shenzhen.
This also formed an upstream downstream connection with other exhibitors on site. When talking about the installation position of IMU, Dai Shi Intelligence mentioned that its module is integrated with binocular vision at the mouth of the humanoid robot. When vision is responsible for “seeing”, IMU is responsible for “feeling”, and touch is responsible for “touching”, the three perception links are stuck in their respective positions in the embodied intelligence supply chain.
36Kr “Workplace Bonus” (ID: ZhiChangHongLi)
36Kr “Workplace Bonus” (ID: ZhiChangHongLi)
[Finally, our judgment]
While most people are still paying attention to the ontology parameters of robot companies, upstream supply chain enterprises have already started to roll data formats, data loops, and cross ontology compatibility. Whether it’s Dai Meng’s dataset, Dai Shi’s motion capture system, or the end-to-end solution of the Vision Tower… essentially, they are all providing “infrastructure” for the embodied intelligence industry.
This means that the next role in China’s robot supply chain may be the “bottom tier supplier” of global hardware startups. Not only selling parts, but also data and engineering capabilities are valuable links that can generate value in industry commercialization.
Note: The note information is based on the speech content at the May 26th event and the company’s public information. Regarding the clinical data, production capacity planning, financing progress, and future production schedule reported by the enterprise, it is recommended that readers make a comprehensive judgment based on the official disclosure of the enterprise in the future.
相关文章
