Weekly Briefing: Junsheng Electronics Delivers Brain Backpack to Unitree; Hesai Robot Power Modules Ship Over 35,000 Units
Ningbo Hyperplane Robot Technology, a subsidiary of Junsheng Electronics, has delivered the first high-compute robot brain backpack to Unitree Technology, compatible with Unitree's humanoid, quadruped, and wheeled robots. As of the end of September 2026, Hesai Technology had cumulatively shipped 35,295 robot power modules, with 23,405 units delivered in the third quarter, up 96.8% quarter-on-quarter. Nvidia released the world action model GR00T 2 and Isaac Lab 3.0 during the same period.
Body · AI translation
The major events in the embodied intelligence field this week are as follows: Joyson Electronics delivers the first high-compute brain backpack for robots to Unitree. Recently, Ningbo Hyperplane Robot Technology Co., Ltd., a subsidiary of Joyson Electronics, officially delivered the first high-compute brain backpack for robots to Unitree Robotics. At the same time, Joyson Electronics announced that the two parties are accelerating joint research on key robot components and industrial scenario applications, and are expected to make more positive progress within the year. The high-compute brain backpack delivered this time is the core carrier of Joyson Electronics' embodied intelligence brain software-hardware integrated solution. It is equipped with a high-compute edge-side platform and a high-efficiency energy system, and can be adapted to Unitree's humanoid, quadruped, and wheeled robots in multiple forms. The product can be independently deployed as an external attachment, providing robots with stronger computing support for environmental perception, data collection, autonomous decision-making, and dynamic motion control. It can also efficiently carry real-time inference of edge-side embodied models, self-evolution and iteration of physical Agents, optimize energy management, and significantly extend the continuous operation time of robots. However, neither party disclosed what chip the brain backpack specifically uses, how much computing power it has, or how large the battery capacity is. In addition to the brain backpack delivered this time, the two parties are currently also developing robot industrial scenario application solutions around automotive parts production and intelligent manufacturing scenarios, including scenario data collection, complete machine software and hardware adaptation, industrial large model training, production line pilot implementation, and large-scale replication, to create standardized implementation of industrial embodied robots that can be exported externally. Xiaozhi's comment: Automotive parts companies have begun delivering core robot components, and supply chain spillover has entered a substantive stage. Hesai's robot power modules have shipped more than 35,000 units. On November 10, Hesai Technology (NASDAQ:HSAI; HKEX:2525) announced its unaudited financial data for the third quarter of 2026. In the third quarter, Hesai achieved cumulative revenue of 1.07 billion yuan, a year-on-year increase of 34.3%; GAAP net profit reached 80.79 million yuan, achieving GAAP profitability for six consecutive quarters; the comprehensive gross margin reached 41.0%, remaining stable. Among them, revenue from Hesai's strategic growth initiative (SGI) business was 90.51 million yuan, a quarter-on-quarter increase of 101.4%. In the fields of robotics and physical AI, Hesai is continuing to ramp up volume. In the third quarter, Hesai's robot power module deliveries were 23,405 units, a quarter-on-quarter increase of 96.8%, monthly production capacity was increased to 10,000 units, and full-body joint modules have also entered the mass production stage. As of the end of September 2026, Hesai's robot power modules had cumulatively shipped 35,295 units, and full-year shipments in 2027 are expected to reach six figures. Behind this, Hesai's power modules have been installed in Sharpa's flagship tactile dexterous hand Wave (W01). This dexterous hand has 22 active degrees of freedom, achieves a 1:1 isomorphic structure with the human hand, has entered NVIDIA's Isaac GR00T humanoid robot reference solution, and has been adopted by top universities and research institutions worldwide such as Berkeley, Stanford, and Nanyang Technological University. In addition, Sharpa's first integrated tactile-perception dexterous manipulation robot D01 and the new-generation full-tactile lightweight dexterous hand W02, released at IROS 2026, are also equipped with Hesai power modules, among which key components such as reducers and motors are all self-developed by Hesai. Xiaozhi's comment: Power module shipments reached 23,000 units in a single quarter, and Hesai's second growth curve is beginning to emerge. Yongda Auto purchases 400 Unitree U1 Pro robots. On November 11, UBTECH signed a strategic cooperation agreement with Yongda Auto, a leading domestic listed luxury automobile dealership service group. According to the agreement, UBTECH's product and technical capabilities will be deeply connected with Yongda's market, customer, and service resources, integrating product promotion, application implementation, and continuous operations. Among them, for the consumer market, the two parties will use UBTECH's hyper-realistic humanoid robot U1 series as the entry point. Yongda will initially purchase 400 hyper-realistic humanoid robots U1 Pro to carry out cooperation in product experience, sales promotion, and customer service, forming a replicable and promotable consumer-side sales and service model for bionic humanoid robots. UBTECH plans to deliver within this year. It is worth noting that, in addition to Yongda, UBTECH also recently signed a comprehensive strategic cooperation agreement with Tianyi Telecom Terminal. In the next three years, the two parties will jointly promote the industry scale of home companion robot terminals to exceed one million units, and promote the industry scale of terminals such as hyper-realistic humanoid robots and wheeled humanoid robots to exceed 100,000 units. Xiaozhi's comment: Stores focus on experience and conversion, and continuous operations after delivery are more worth tracking than order amounts. Zhongding Co., Ltd.'s 1.92 billion yuan convertible bonds are listed, with proceeds to increase investment in robotics, thermal management, and more. On November 12, the convertible bond "Zhongding Convertible 03" (code 127119) issued by Zhongding Co., Ltd. (000887.SZ) was listed on the Shenzhen Stock Exchange. This tranche of bonds totaled 19.2 million units, with a face value of 100 yuan each, a term of 6 years, and total raised funds of 1.92 billion yuan. On the first day of listing, the bond closed at 136.15 yuan, an increase of 36.15%. From the perspective of investment destinations, the funds are mainly invested in production capacity and R&D layout related to robot components, thermal management, and intelligent chassis: 732 million yuan for the intelligent robot core joint manufacturing project, 582 million yuan for the intelligent thermal management system assembly project, 106 million yuan for the new energy vehicle intelligent chassis system R&D and settlement center project, and the remaining 500 million yuan to supplement working capital. According to the disclosed capacity plan, after the aforementioned fundraising projects are completed and reach full production, the company will add annual capacity of 150,000 harmonic reducers, 150,000 sensors, and 280,000 various robot joints; in thermal management products, it will add annual capacity of 15.004 million new energy vehicle thermal management products, 11,200 energy storage thermal management products, and 646,800 data center thermal management products. Xiaozhi's comment: Thermal management and robotics are advancing on two fronts, and Zhongding's automotive chassis capabilities are being horizontally reused. Qiyuan's 10,000th robot officially rolls off the line On November 8, the 10,000th personal robot of Qiyuan Robotics, a subsidiary of Swancor Advanced Materials, officially rolled off the line. This means that, following the rollout of the 5,000th personal robot on October 25, Qiyuan Robotics took only 14 days to leap from 5,000 to 10,000 cumulative units off the line, achieving 10,000-unit-scale mass production. At the mass production site, Liu Dongfang, co-president of Swancor Qiyuan Robotics and president of the Q Business Unit, delivered the Qiyuan Q1 numbered "NO.10000" to representative user Wu Xiaobo. At about the same time, the Swancor Qiyuan Robotics production base, also the country's first "Expert" personal robot manufacturing base project, was signed in Fuyang, Hangzhou, and the Swancor Qiyuan Robotics national headquarters operations center was unveiled at the same time. The project's operating entity was established in the Fuyang Economic and Technological Development Zone by three parties: Swancor Qiyuan, Jinggu Co., Ltd., and Fuyang Industrial Investment Group. Xiaozhi's comment: Personal robots have entered the ramp-up stage, and cost, quality control, and after-sales service will all be tested together. Topstar releases industrial bipedal humanoid robot Star Recently, Topstar unveiled its full-size industrial bipedal humanoid robot Star at the "TOPS DAY 2026" global open day held in Dongguan. The robot is 172 cm tall and weighs 72 kg, with a dual-arm payload capacity of 10 kg, and is mainly aimed at factory material handling, sorting, and other operational scenarios. Unlike most humanoid robots that start with demonstrations, Star's design starting point is existing production lines. The overall dimensions are adapted to factory layouts, the arms use a 7-axis humanoid structure and are equipped with cross-axis wrists to replicate human working postures, and the joints carry high-precision joint modules. Topstar said that during actual operations the robot can simultaneously collect physical scene data for subsequent capability optimization, consistent with the company's previously proposed closed-loop logic of "scenario + robot + data + AI." What Star fills in is the bipedal form in Topstar's complete robot matrix. Before this, the company already had wheeled products, and in January this year it launched the quadruped robot "Xingzai" for inspection, forest firefighting, and other scenarios. From wheeled to quadruped to bipedal, the product forms are expanding, while the scenarios they target remain industrial sites. Xiaozhi's comment: Wheeled, quadruped, and bipedal in parallel, Topstar is using different form combinations to cover differentiated scenario needs. NVIDIA releases GR00T 2 and Isaac Lab 3.0 On November 9, during CoRL 2026, NVIDIA released the world action model GR00T 2, the robot learning framework Isaac Lab 3.0, and a number of simulation and evaluation tools. GR00T 2 is built on NVIDIA Cosmos, and its training data includes 100,000 hours of human first-person data. NVIDIA's published test results show that, without task-specific demonstrations, GR00T 2 succeeded at least once in 40 of 50 evaluation tasks; after providing one demonstration per task, the average success rate was 71%, and after five demonstrations it reached 95%. The model also supports learning from human demonstrations and attempts to transfer skills to different robot bodies. Isaac Lab 3.0 further integrates the Newton physics engine and can simulate complex objects such as cables, cloth, and dexterous hands. In the visual dexterous manipulation benchmark test, its training throughput reached up to 2.5 times that of the combination of Isaac Lab 2.3 and PhysX. NVIDIA simultaneously released industrial evaluation benchmarks such as DexBench and VABAR, with test coverage spanning retail, logistics, manufacturing, and hotel services. Xiaozhi's comment: With the model, simulation, and evaluation released simultaneously, NVIDIA is defining the development foundation for embodied intelligence. Google upgrades robot model, achieving a 96% success rate in Mercedes-Benz factory tests Recently, Google DeepMind released a new-generation vision-language-action (VLA) model, Gemini Robotics 2.5, focusing on improving robots' dexterous manipulation, task execution reliability, and safety in complex environments. The model is currently being gradually opened to robot partners. According to Google DeepMind, Gemini Robotics 2.5 has completed tests on more than 200 different tasks, covering scenarios such as tool use, flexible object manipulation, and fine operations. Compared with previous models, the success rate on different tasks increased by 30% to 100%, and execution speed increased by 20% to 30%. Automobile manufacturing became an important scenario for validating this model. Google DeepMind, together with Apptronik and Mercedes-Benz, used the Apollo 2 humanoid robot to carry out parts sorting and kitting tests on Mercedes-Benz production lines in Kecskemét, Hungary, and Sindelfingen, Germany. During the tests, the robot managed the workflow through the Gemini Robotics embodied reasoning model, while Gemini Robotics 2.5 controlled the arms and dexterous hands to grasp, sort, and place bulk automotive metal brackets into designated containers, achieving a 96% task success rate. The project is still in the proof-of-concept stage. Xiaozhi's comment: From laboratory tasks to Mercedes-Benz sorting, the value of physical AI is converging toward the industrial side. South Korea supports sensors, actuators, and humanoid robot batteries On November 11, the South Korean government announced a support plan for core components of physical AI, focusing on next-generation sensors, robot actuators, and humanoid robot batteries. In next year's budget proposal, the three categories of projects are allocated 9 billion, 30 billion, and 13.3 billion won respectively, totaling 52.3 billion won. Actuator R&D goals include increasing output capacity by about 5 times at the same weight and raising the number of repeated uses to more than 1 million times, with plans to start mass production of domestically produced actuators in 2029. Humanoid robot batteries target 8 hours of continuous operation and instantaneous high-power output, with related R&D covering commercial high-performance batteries, all-solid-state batteries, and battery standards, and plans to advance commercialization in 2030. These amounts belong to the budget proposal and R&D plans, and are not completed investments or procurements. Xiaozhi's comment: Budgets and plans are not the same as implementation, and South Korea still needs to build up its local supply chain and mass production capabilities.