(Report produced by: Guotai Junan Securities, Wang Hao, Bao Yanxin, Zhang Han)
1. Core Viewpoint: Wheeled Robots Are Expected to Achieve Commercialization Before Bipedal Robots
Humanoid robots are moving towards large-scale production, with rapid development among domestic and international companies. Embodied intelligence is the intersection of artificial intelligence and robotics, with humanoid robots being the most promising track in embodied intelligence, transitioning from proof of concept to large-scale implementation, attracting participation from automotive companies and startups. In the overseas market, Tesla is leading the industry progress, while Figure AI and Agility Robotics focus on logistics, production lines, and warehousing scenarios. In the domestic market, companies like Xiaopeng and Xiaomi are entering the field; startups such as Zhiyuan Robotics and Yushu Technology are making rapid advancements. The humanoid robot industry chain can be divided into upstream components, midstream body enterprises, and downstream applications. Midstream body enterprises can be likened to automotive manufacturers, occupying a dominant position in the industry chain; upstream components involve a wide range of automotive and mechanical industry companies participating across sectors; downstream application scenarios include industrial manufacturing, logistics warehousing, household, and commercial services, currently in the initial exploration stage. Wheeled robots have three advantages: lower training difficulty, lower cost, and longer endurance, making them more suitable for practical needs and easier to commercialize compared to bipedal robots. Based on the lower limb morphology, humanoid robots can be classified into footed, wheeled, and wheeled-foot types. Wheeled robots significantly differ from bipedal robots in balance control, as they do not involve center of gravity transfer; their lower limb movement only requires a mobile platform, making design and manufacturing simpler than that of bipedal robots. Wheeled robots possess the advantages of lower training difficulty, lower cost, and longer endurance, making them more suitable for practical needs. In terms of training difficulty, wheeled robots rely on simple drive wheels combined with support legs, while bipedal robots’ lower limb control depends on complex algorithms, requiring a large amount of training data with low generalization. In terms of cost, the bill of materials (BOM) cost for bipedal robots is ten times higher than that of wheeled chassis, and maintenance and repair costs are significantly higher for bipedal forms. Regarding endurance, wheeled robots generally have a runtime greater than 5 hours, while bipedal forms often fall short of 2 hours, showing a significant gap. In terms of practical needs, most current application scenarios only require flat movement combined with dexterous hands, where lower limbs need stability and speed to complete movement, making wheeled forms more suitable. Companies such as Galaxy General, Zhiyuan Robotics, Qianxun Intelligent, Xinghai Map, and Stardust Intelligent are all laying out wheeled products, exploring the forefront of commercialization.
Wheeled humanoid robots are expected to achieve commercialization in B-end markets such as retail and factories first; Galaxy General and Zhiyuan Robotics are committed to applying wheeled humanoid robots to specific tasks and are likely to lead commercialization progress. Currently, wheeled humanoid robots are mainly piloting in commercial service scenarios such as retail pharmacies, as well as in factory scenarios for handling and logistics sorting, with the potential to gradually achieve commercialization in B-end markets like retail and factories, validating commercial value and increasing penetration. Compared to general-purpose AI, optimizing design for specific scenarios aligns better with the commercialization process, with the skills closest to implementation still being relatively simple tasks such as moving, grasping, and placing. Galaxy General and Zhiyuan Robotics focus on optimizing the application of wheeled humanoid robots for specific tasks, likely leading commercialization progress, and further advancing towards household scenarios and diversified tasks in the future.
2. Humanoid Robots Moving Towards Large-Scale Production, Rapid Development Among Domestic and International Companies
2.1. Humanoid Robots Transitioning from Proof of Concept to Large-Scale Implementation
Humanoid robots have a human-like structure, capable of autonomously perceiving the environment, making decisions, and executing complex tasks. Unlike traditional industrial robots and collaborative robots that are fixed on production lines, rely on precise programming, and can only repeat single actions in structured environments, humanoid robots, with their human-like appearance and embodied intelligence as the underlying technology, significantly expand joint degrees of freedom and operational space, enabling them to walk, grasp, converse, and even collaborate in unstructured environments such as homes, shopping malls, and factories. Due to the high coupling between their “bodies” and the “environment,” humanoid robots have become a typical implementation platform for embodied intelligence—agents continuously interact with the real world through their physical bodies, accumulating knowledge and acquiring skills through a perception-action feedback loop.
Embodied intelligence is the intersection of artificial intelligence and robotics, with humanoid robots being the most promising track in embodied intelligence. Embodied intelligence refers to AI intelligence with a body, where the machine brain can make decisions to control the body in response to changes in the external environment. The core lies in the deep integration of physical devices and intelligent decision-making, achieving a closed loop of perception-decision-execution. Autonomous vehicles, quadruped robotic dogs, and humanoid robots are different manifestations of embodied intelligence, with humanoid robots being the closest to human body forms, naturally adapting to human social production and lifestyle, making them the most application-potential key track in embodied intelligence. Humanoid robots can use two hands to operate tools, mapping visual, tactile, and auditory information to motion control through multimodal sensors, and transforming each fall, grasp, and conversation into transferable knowledge through end-to-end learning. Therefore, humanoid robots are both an extension of industrial robot technology stacks towards high degrees of freedom and adaptability and a key bridge for embodied intelligence to transition from algorithms to the physical world.
Humanoid robots are transitioning from proof of concept to large-scale implementation, with a significant increase in the number of related companies registered. In 2022, Tesla, as an industry pioneer, launched the Optimus prototype, marking the beginning of a rapid development period for the global humanoid robot industry. From 2022 to 2024, following Tesla, numerous domestic and international companies entered the humanoid robot field, forming a diversified competitive landscape. According to data from Qichacha, the number of registered domestic humanoid robot companies reached 104 in 2024, with a year-on-year growth rate of 104%, indicating a significant increase in the number of companies related to humanoid robots.
In the capital market, humanoid robots have become a hot spot for investment and financing. From January to July 2025, the domestic humanoid robot industry saw 101 financing events, with financing amounts exceeding 26 billion yuan, surpassing the total for the entire year of 2024, indicating unprecedented heat in the capital market. Among them, leading automotive industry players have become significant forces in humanoid robot investments, with BAIC Investment making multiple investments in companies like Galaxy General, Zhiyuan Robotics, and Pashini Perception, which are key body and component enterprises; SAIC Group’s industrial capital and private equity institutions have also participated in multiple rounds of financing for companies like Zhiyuan Robotics, Zhujidi Power, and Nuo Shi Robotics.
2.2. Tesla as an Industry Pioneer Leading Technological Iteration
Tesla’s Optimus, as a pioneer in humanoid robot development, continues to lead industry technological iterations. From the initial announcement of the humanoid robot plan at AI Day in 2021 to the first appearance of the Gen 1 prototype in 2022, Tesla integrates the visual perception, AI training, and hardware technology accumulated in the autonomous driving field into the embodied platform. Subsequently, Tesla has continuously promoted version iterations, from the release of Gen 2 in 2023 achieving lightweight and sensor upgrades to the trial production of Gen 3 at the Fremont factory in 2025, keeping its technology at the forefront of the humanoid robot industry. In 2021, Tesla first announced the Tesla Bot plan at AI Day. According to this plan, Optimus will rely on Tesla’s key technologies in the autonomous driving field, including the FSD system, Dojo supercomputing platform, and AI chips, to build the underlying technical architecture for robot visual perception, environmental modeling, and action learning. In September 2022, the Optimus prototype was unveiled at the second AI Day. The Optimus prototype was presented without exterior decoration, exposing its internal joints, circuit boards, cables, and other devices, allowing the audience to intuitively see the details of each action. During the live demonstration, the prototype slowly walked out from the back of the stage, completing actions such as walking, stopping, waving, and turning, and performed a simple dance. In March 2023, Tesla released Optimus Gen1. Gen1 integrated a complete FSD system and self-developed D1 chip, possessing initial environmental perception and action control capabilities, able to automatically recognize, sort, and classify objects through deep neural networks, constructing a “perception-cognition-action” closed-loop intelligent chain.
In December 2023, Tesla online released Optimus Gen2 and continues to iterate on products and technologies. Gen 2 achieved significant improvements in mechanical structure lightweighting, sensor system upgrades, and fine motor control capabilities. In May 2024, Tesla released a video of the robot sorting batteries on social media, achieving end-to-end neural network-driven autonomous sorting, dynamic error correction, and adaptive adjustment capabilities. In May 2025, Tesla released two videos of Optimus performing dance moves, showcasing its motion control algorithms and body coordination capabilities. Tesla plans to officially launch Optimus Gen3 by the end of 2025, with mass production expected to begin in 2026. Gen3 was showcased in April 2025 during trial production at the Fremont factory and is still undergoing continuous iteration and improvement, with plans for release by the end of 2025 and achieving mass production of 1 million units annually within five years. In terms of downstream applications, Tesla showcased the trial production line for Optimus humanoid robots located at the Fremont factory in its Q1 2025 report. This move marks a key step for Tesla in transitioning robots from demonstration to large-scale internal use and may accelerate the transition to external commercialization services in the coming years.
2.3. Policies Leading Industry Development, Automotive Companies and Startups Entering the Field
Central and local governments have issued industrial planning and guidance programs to lead the development of the humanoid robot industry. In 2023, the Ministry of Industry and Information Technology released the “Guiding Opinions on the Innovative Development of Humanoid Robots,” providing an overall plan for related industry development, proposing to preliminarily establish an innovative system for humanoid robots by 2025, significantly enhancing technological innovation capabilities and achieving comprehensive strength at the world advanced level by 2027. Since 2025, many local governments have introduced industrial guidance programs. The Beijing Municipal Science and Technology Commission, in conjunction with the Zhongguancun Management Committee, issued the “Beijing Embodied Intelligence Technology Innovation and Industry Cultivation Action Plan (2025-2027) (Draft for Comments),” laying out the development of the embodied intelligence industry in Beijing; the Guangdong Provincial Party Committee and the Provincial People’s Government issued the “Guangdong Province Action Plan for Building a Modern Industrial System by 2025,” proposing to vigorously develop humanoid robots and promote innovation in application scenarios.
Automotive companies such as Xiaopeng and Xiaomi are entering the field. Xiaopeng Motors established Pengxing Intelligent in December 2021, specifically responsible for humanoid robot research and development, and released its first bipedal humanoid robot PX5 in October 2023, with a new generation humanoid robot IRON expected to enter production line testing at Xiaopeng’s factory in Guangzhou in 2025. Xiaomi’s CyberOne was officially unveiled in August 2022 and entered phased production on its manufacturing line in February 2025. Startups like Zhiyuan Robotics and Yushu Technology are making rapid progress. Zhiyuan Robotics was established in February 2023 and launched the “Expedition A1” humanoid robot in August 2023, which has autonomous path planning and grasping capabilities. Currently, it has a product layout including the Expedition series, Lingxi series, and Elf series. Yushu Technology was established in 2016, initially focusing on quadruped robotic dogs, and released its first general-purpose humanoid robot H1 in 2023, with the second-generation G1 humanoid robot launched in 2024, featuring 23-43 degrees of freedom and a peak torque of 120 N·m at the knee joint, supporting complex actions such as fast walking, jumping, and squatting, balancing flexibility and motion performance. UBTECH, as an early participant in the domestic humanoid robot field, has seen its products enter factory training. UBTECH was established in 2012 and has grown into an important player in China’s humanoid robot field after more than a decade of development. In terms of overall capabilities, the Walker S1 released by UBTECH has entered BYD’s factory training, capable of intelligent handling, sorting, and quality inspection, with a maximum torque of 250 N·m, and equipped with six tactile sensors in its hands, providing high-precision industrial operation capabilities, and achieving system collaboration with unmanned forklifts and L4 level unmanned logistics vehicles.
3. Body Enterprises Occupy the Core of the Industry Chain, Upstream Components Are Diverse, and Downstream Market Space Is Vast
The humanoid robot industry chain can be divided into upstream components, midstream body enterprises, and downstream applications. Midstream body enterprises can be likened to automotive manufacturers, occupying a dominant position in the industry chain; upstream components involve a wide range of automotive and mechanical industry companies participating across sectors; downstream application scenarios include household services, industrial manufacturing, special operations, etc., currently in the initial exploration stage.
3.1. Various Body Forms Develop in Parallel, Automotive Companies and Startups Lead Competition
Among overseas body enterprises, Tesla, Figure AI, and Agility Robotics are at the forefront. Tesla launched the Optimus project in 2021, being the first humanoid robot developed by an automotive company based on autonomous driving technology, integrating the FSD system, Dojo chip, and self-developed perception algorithms, with its technology and industrialization capabilities serving as a benchmark. Figure AI was established in 2022 and released its first product, Figure 01, in 2023, with the second-generation humanoid robot Figure 02 launched in August 2024. In terms of applications, in February 2025, Figure AI announced that its Figure 02 robot had learned to take a second job, organizing packages at a logistics center; previously, in 2024, Figure 02 successfully applied to BMW’s production line, responsible for precision tasks such as sheet metal component assembly. Agility Robotics, established in 2015, focuses on bipedal robot development, launching Cassie in 2016 and continuously iterating the Digit series. In terms of applications, Agility Robotics focuses on handling, unloading, and shelf management in warehouse scenarios, reaching a multi-year agreement with GXO Logistics in 2024 to provide logistics and warehousing solutions; it has collaborated with Amazon to integrate the Digit robot into its distribution centers to optimize processes such as handling turnover boxes.
Domestic body enterprises are developing rapidly, quickly completing the research and development of their first prototypes and technological iterations. Zhiyuan Robotics was established in 2023 and released its first humanoid robot “Expedition A1” in the same year, with plans to build a factory in Shanghai in 2024 and release multiple humanoid robot products such as “Expedition A2” and “Lingxi X1”; in March 2025, it launched the upgraded “Lingxi X2” and the first general-purpose embodied base model “Zhiyuan Qiyuan GO-1”. Xiaomi has been laying out robotics since 2021, releasing the quadruped robotic dog CyberDog and the full-size humanoid robot CyberOne (Iron); Xiaomi regards robotics as part of its “full ecosystem of vehicles, homes, and AI,” focusing on the integration of smart hardware and AI capabilities. UBTECH has been involved in robotics research and development for a long time, launching the Alpha model small humanoid robot in 2014 and participating with 540 robots in the Spring Festival Gala stage in 2016; since 2018, UBTECH has released the Walker series of bipedal humanoid robots and continues to iterate to the industrial-grade Walker S1 model. In terms of applications, UBTECH has achieved product landing since 2022, with the current Walker S series entering multiple automotive factory training. Fourier started with rehabilitation robots, entering the embodied intelligence track from the lower limb exoskeleton X1 in 2017, launching the general-purpose humanoid robot project in 2019, and releasing the first mass-produced humanoid robot GR-1 in 2023, with the iterative product GR-2 launched in 2024. Yushu Technology initially focused on quadruped robotic dogs, launching series products such as Laikago, A1, and Go1, widely applied in industrial, consumer, and research scenarios; in 2023, it released the humanoid robot H1, with a smaller G1 product launched in 2024. Galaxy General released its first-generation product Galbot G1 in 2024, featuring a mobile dual-arm and wheeled lower limb design, capable of 360° movement and covering a wide working area. In terms of applications, G1 has completed tasks such as picking, restocking, and sorting in 24-hour retail pharmacies and industrial logistics scenarios.
Xiaopeng is developing its robotics business through Xiaopeng Pengxing, releasing its first humanoid robot prototype in November 2024, with plans to enter Xiaopeng’s factory training in 2025 and achieve mass production in 2026. Qianxun Intelligent was established in 2024, with a technical path of general humanoid robots and embodied large models iterating simultaneously, launching its first robot product Moz1 in March 2025. In the WAIC conference in July 2025, Qianxun Intelligent officially launched its first robot product Moz1. Qianxun Intelligent’s Moz1 is the first high-precision full-body force-controlled embodied intelligent robot in China, with 26 degrees of freedom (excluding dexterous hands), equipped with the highest power density integrated force control joints globally, designed for multi-task assembly line scenarios. In terms of motion control, Moz1 is the first force-controlled robot in China equipped with a high-precision high-speed WBC full-body motion control system, achieving zero-latency remote control, with actions that can be smooth and coordinated. Moz1 is equipped with the self-developed embodied intelligence large model Spirit v1, supporting multi-threaded motion planning and visual closed-loop, capable of efficient perception and precise control in complex production environments. Relying on the Spirit v1 large model, Moz1 can automatically complete tasks such as desktop organization, tissue replacement, garbage collection, chair repositioning, and whiteboard wiping, and has taken the lead in overcoming the long-range operation challenges of flexible objects (such as folding clothes) in China. In the office scenario, Qianxun Moz1 robots have shown strong flexibility, autonomously completing cleaning tasks in meeting rooms, collecting pens, and wiping whiteboards, expected to become a starting point for commercialization. In the future, it will further replicate technology to commercial scenarios such as shopping centers and retail stores, exploring more downstream fields. Qianxun Intelligent has raised over 1.1 billion yuan in financing in the first half of 2025. In July 2025, Qianxun Intelligent completed a new round of approximately 600 million yuan in PreA+ financing, with a post-investment valuation of 4 billion yuan; this round of financing was led by JD.com, with participation from Zhongzhu Network, Jinkong Capital, Huatai Zijin Investment, and other well-known institutions.
Stardust Intelligent officially released its first complete form product Astribot S1 in August 2024. Stardust Intelligent was established in December 2022 and conducted its first product technical demonstration in April 2024, where the humanoid robot performed complex tasks such as ironing clothes, sorting items, flipping a pan to stir-fry, vacuuming, and competitive cup stacking; in August 2024, it officially released the new generation AI robot assistant Astribot S1, based on an AI-oriented integrated hardware and software system architecture, demonstrating industry-leading general operational capabilities across various scenarios. Stardust Intelligent S1 has a “high-value upper body and a deployable lower body,” with key components self-developed, providing significant cost advantages, facilitating commercialization. In June 2025, Stardust Intelligent reached a deep strategic cooperation with a public nursing home in Shenzhen, jointly exploring innovative applications of embodied intelligence technology in life assistance, health monitoring, and emotional companionship in elderly care scenarios. Astribot S1 has become the first wheeled humanoid robot to enter nursing homes in China, capable of delivering mineral water to the elderly, distributing snacks, and accompanying them in Tai Chi and square dancing, marking an important step for Stardust Intelligent towards commercialization.
Stardust Intelligent completed A and A+ rounds of financing totaling several hundred million yuan in April 2025. The latest round of financing was led by Ant Group and Jinqiu Fund, with participation from Yunqi Capital and Daotong Investment. Reviewing Stardust Intelligent’s financing history, the company completed angel round financing in March and October 2023, with investors including Yunqi Capital and Dexun Investment; in June 2024, the company completed several million dollars in pre-A round financing led by Jingwei Venture Capital, with participation from Qinghui Investment, Daotong Investment, Yunqi Capital, Jinqiu Fund, and follow-up investments from existing shareholders such as IDG Capital.
Xinghai Map focuses on creating “one brain, multiple forms” embodied intelligent robots. Xinghai Map was established in September 2023, releasing its first product, the mechanical arm A1, in May 2024, and launching the full-size wheeled dual-arm humanoid robot R1 in October 2024. In January 2025, it officially launched three models of the R1 series, starting at 199,000 yuan, equipped with the NVIDIA Jetson AGX Orin 32GB platform, featuring an 8-core CPU and 200 TOPS GPU computing power. Xinghai Map’s company positioning is to create embodied intelligent robots with “one brain and multiple forms,” with its technological moat primarily based on two key points: the EFM-1 model, which integrates a hundred billion parameter VLM model and a billion parameter VLA model, capable of breaking down commands like “move the box to workstation A3” into a closed-loop decision of “navigation-grasp-assembly”; and the RSR engine, which scans factories with consumer-grade cameras to generate centimeter-level precision digital twins, deriving thousands of simulation samples from real data, solving the industry’s “data desert” problem. Xinghai Map R1 is currently mainly aimed at research applications. Xinghai Map is developing rapidly in the research field, with over a hundred research clients globally, such as Fei-Fei Li’s team at Stanford University, which used the Xinghai Map platform to develop the BEHAVIOR Robot Suite system, allowing the robot to learn tasks like cleaning toilets, folding clothes, and picking up trash for just $500; Google’s Physical Intelligence company has also trained a general strategy model π0.5 using its hardware. In terms of industrial landing, pilot projects are currently being conducted. Xinghai Map has completed multiple rounds of A and A+ financing totaling over 1.3 billion yuan in the first half of 2025. In July 2025, Xinghai Map announced the completion of A4 and A5 rounds of financing led by Meituan and Today Capital, raising over $100 million, with a post-investment valuation of $800 million. Previously, in April 2025, the company completed a series of A2 and A3 rounds of financing led by KKR, with participation from Lenovo Capital, Haier Capital, and other industrial capital, and follow-up investments from existing shareholders such as IDG Capital; in February 2025, the company completed A round financing totaling nearly 300 million yuan.
5. Industry Outlook
Wheeled humanoid robots are expected to achieve commercialization in B-end markets such as retail and factories first. Wheeled forms possess three advantages: lower training difficulty, lower cost, and longer endurance, making them more suitable for application needs in practical scenarios, and easier to commercialize compared to bipedal robots. Currently, wheeled humanoid robots are mainly piloting in commercial service scenarios such as retail pharmacies, as well as in factory scenarios for handling and logistics sorting, with the potential to gradually achieve commercialization in B-end markets like retail and factories, validating commercial value and increasing penetration. Galaxy General and Zhiyuan Robotics are committed to applying wheeled forms to specific tasks and are likely to lead commercialization progress. Compared to general-purpose AI, optimizing design for specific scenarios aligns better with the commercialization process, with the skills closest to implementation still being relatively simple tasks such as moving, grasping, and placing. Galaxy General is conducting model training for specific tasks, releasing large models such as GraspVLA for grasping, GroceryVLA for retail, and TrackVLA for navigation, and is committed to applying moving, grasping, and placing skills in pharmacies and retail scenarios first. Galaxy General has already achieved market operation in retail pharmacies, helping 24-hour stores efficiently pick goods from shelves and operate pharmacies, with plans to cover 100 smart pharmacies in Beijing, Shenzhen, and Shanghai by 2025; in August 2025, Galaxy General will launch the world’s first fully autonomous robot service pop-up store, marking a solid step in the commercialization of embodied intelligence technology in the retail field. Zhiyuan’s wheeled robots are mainly used in handling and sorting tasks, having made initial progress in commercialization. In terms of handling tasks, Zhiyuan’s Expedition A2-W wheeled robot can perform continuous handling operations in high-temperature environments, having received dozens of purchase orders from automotive companies to replace high-risk handling jobs. In terms of sorting tasks, Zhiyuan’s Elf G1 demonstrated end-to-end operations in logistics scenarios at the WAIC conference in 2025, capable of flexibly and skillfully sorting clothing packages, accurately placing them in the correct positions, and quickly adjusting the orientation of barcodes for scanning. Additionally, in the commercial service field, Zhiyuan’s Expedition A2 has received an order for 200 units from China Mobile, to be deployed in nationwide service halls for guiding work. Galaxy General and Zhiyuan Robotics focus on optimizing the application of wheeled humanoid robots for specific tasks, likely leading commercialization progress, and further advancing towards household scenarios and diversified tasks in the future.