In mid-August, the world’s first comprehensive athletic competition for humanoid robots—the 2025 World Humanoid Robot Games—took place in Beijing. Over two days of events, news about the games frequently dominated social media trends, adding further momentum to the already booming humanoid robot industry. The championship results, such as 21.50 seconds for the 100-meter dash, 1 minute 28.03 seconds for the 400-meter race, and 95.641 centimeters for the high jump, demonstrated that while these humanoid robots still lag far behind human athletes in track and field, they have genuinely stepped onto the field and into the public eye. These humanoid robots can run, jump, play soccer, and perform gymnastics. Compared to their limited abilities just over half a year ago, such as simply twirling handkerchiefs in performances, the humanoid robot sector has made visible progress.
Are humanoid robots already close to integrating into our daily lives? How advanced is the current performance of humanoid robots, and what scenarios can they be applied to? How far are they from entering thousands of households and truly “living like humans”? Nanjing is intensifying efforts to build a “Robot City.” Recently, reporters visited several local humanoid robot enterprises to seek answers to these questions.
1. Humanoid and Embodied: Why Must It Be Humanoid?
At the National People’s Congress sessions in March this year, the government work report首次 mentioned the concepts of “embodied intelligence” and “smart robots,” explicitly calling for vigorous development of related industries. Once a sci-fi fantasy, robots have rapidly become a tangible industry targeted for national growth. Post-sessions, various news about the robot industry quickly captured public attention. Amid the flood of information, numerous new terms and concepts emerged, including “embodied robots.” Many reports directly quoted the government work report’s vocabulary. So, what is the difference between “embodied robots” and “humanoid robots”?
“Simply put, ‘humanoid robots’ are a type of ’embodied robot,’ and the most complex one at that. It could even be said that ‘humanoid robots’ represent the highest form of ’embodied robots,'” explained Song Aiguo, Chief Professor at Southeast University and Director of the Institute of Robot Sensing and Control Technology. The concept of embodied robots is broader; for instance, robotic arms, robotic dogs (quadruped robots), and even robotic vacuum cleaners can be considered part of the embodied robot category. Humanoid robots, as the name implies, must have a form similar to humans.
“Creating a human-like appearance is easy, but simultaneously equipping it with practical functions drastically increases the difficulty,” Song Aiguo illustrated. Currently, some humanoid robot products have hands with fingers that either cannot move or can only make simple gestures; more advanced versions can perform tasks like grasping objects or tightening screws. “They are still far from matching the functionality of real human hands, not to mention tactile sensation, which is one of the areas our team is focusing on. Every step a humanoid robot takes toward resembling a real human requires substantial R&D accumulation.”
Given the immense technical challenges, why insist on making robots humanoid? “Because the real world we live in is designed around humans,” said Luo Dongmei, Vice President of iSoftStone Group and Executive Dean of the Innovation and Technology Research Institute. Most life scenarios are built around human needs, so humanoid robots can have the widest applicability and market potential. Song Aiguo also believes that mature humanoid robots will eventually enter the daily lives of the general public. Faced with various life scenarios, the humanoid form is the most universal choice.
When addressing this question, almost every interviewee highlighted applicability. Some industry insiders also pointed out another advantage of the “humanoid” form—affinity. “Imagine a scenario where humanoid robots enter thousands of households in the future. Would you prefer this ‘family member’ to be a cold metallic frame or a robot with a very human-like appearance?” said Zhou Pan, Sales Director of Huaxia Robot (Nanjing) Co., Ltd.
2. Brain and Cerebellum: Not Just Looking Human, but Operating Like Humans Too
Stepping onto the stage calmly alone, waving to the audience, then turning to point at the large screen playing its own promotional video—at the 2025 Intelligent Robot Development Conference held in Nanjing in June, the second-generation humanoid robot Codroid02 from Nanjing Estun Cuzhou Technology Co., Ltd. introduced itself to the world. Standing 170cm tall, weighing 70kg, with a single-arm load capacity of 5kg and 31 degrees of freedom throughout its body, this humanoid robot showcased its capabilities.
“The seemingly simple actions during the release actually fully demonstrated Codroid02’s excellent abilities,” introduced Zang Jiawei, Chief Technology Officer of Nanjing Estun Cuzhou Technology Co., Ltd. Enabling a humanoid robot to walk, run, and jump with a natural human-like gait on two feet requires fine perception and motion control capabilities. “Humanoid robots not only look like humans but also operate like humans. The core controlling them can be analogized to the brain and cerebellum. The brain is responsible for cognitive decision-making, achieving environmental perception, task planning, natural language interaction, etc., while the cerebellum focuses on motion control, handling joint coordination, force control feedback, and other precise operations.”
Discussing the operational mechanism of humanoid robots, Wang Yulin, Professor at the School of Mechanical Engineering, Nanjing University of Science and Technology, also mentioned the concepts of “brain” and “cerebellum.” “From appearance to internals, humanoid robots imitate humans,” he exemplified. “For instance, the simple action of pouring water requires the coordinated effort of the robot’s ‘brain and cerebellum.’ First, understanding the command to pour water, using its eyes—cameras—to capture images, identifying the positions of the kettle and cup, then planning to walk to the kettle and cup, pick up the kettle, and pour the water into the cup—these are the brain’s tasks. Driving various joints throughout the body to move via motors, especially the fine operations of the hands, such as how much force to use to lift the kettle and how to align it with the cup—these are the cerebellum’s tasks.”
The “brain” and “cerebellum” of humanoid robots essentially refer to a current mainstream R&D direction in the field—the VLA model, i.e., Vision-Language-Action. Through the VLA model, the general capabilities and large-scale learning paradigms of foundational models can be leveraged to process general visual and linguistic information and generate real-time actions, thereby enabling humanoid robots to perform various functions.
“Training the VLA model, like foundational models, requires ‘feeding’ large amounts of data, which are relatively scarce currently. This is one of the main bottlenecks we face now,” Zang Jiawei said. For general large models, massive data available on the internet can be used for training. However, in the humanoid robot R&D field, there is still a lack of public infrastructure similar to the internet and the massive multimodal datasets accumulated on such infrastructure. Training data for the VLA model is best collected through humanoid robots in actual environments, which requires considerable time to accumulate.
Recognizing this industry pain point of data scarcity, Huaxia Robot (Nanjing) Co., Ltd. initiated the construction of the Embodied Intelligent Robot Application Center. Reporters saw at the scene that the center simulates various physical scenarios such as daily households, company offices, and factory warehouses, and also sets up various road environments like gravel, plastic, and slopes, allowing humanoid robot products to conduct tests in diverse real environments. “The center combines data collection and scenario verification functions and is open to the industry,” Zhou Pan said. They welcome products from other companies to test and verify at the center. In the future, the center plans to establish an open and shared data platform to help promote the development of the entire humanoid robot industry.
3. Bonsai and Landscape: A Product Must Be Truly Useful
Although humanoid robots are a new field in the public eye and the entire industry has just gained momentum, related enterprises, including whole-machine companies, have sprung up like mushrooms after rain. Data from Qichacha shows that as of August, there are over 160 humanoid robot whole-machine platforms domestically, accounting for more than 50% globally; core component supply chain enterprises exceed 600. Not only is the quantity large, but the strength is also substantial. Over half of the humanoid robot whole-machine companies have registered capital at the tens of millions level, with nearly 30% exceeding 50 million yuan.
Driven by capital, China’s humanoid robot industry is presenting a scene of a hundred flowers blooming. According to the latest report from Morgan Stanley, over the past five years, China has filed 5,880 humanoid robot patents, far surpassing the United States’ 1,483, Japan’s 1,195, and the World Intellectual Property Organization’s 1,123.
Facing intense competition early in the industry’s inception makes choosing the right track crucial. “A product must be truly useful,” said Liu Danfu, Assistant to the Chairman of Nanjing Tacter Robot Company. Humanoid robot products cannot aim for perfection in one step or be overly ambitious. Instead, based on current technological levels, they must quickly form product strength and enter the market. This, on one hand, allows enterprises to enter a virtuous cycle of development; on the other hand, enables products to iterate and upgrade in broader real-world scenarios, rather than just behind closed doors in laboratories.
Tacter Robot’s choice is to target special industries and specific scenarios. In January, they released the world’s first humanoid robot to obtain explosion-proof certification IIC T6, “Tiankui-1,” filling the gap in the application of humanoid robots in complex industrial scenarios. “Tiankui-1” is a “big guy” and “powerhouse” among humanoid robots, standing 220cm tall, with a single-arm load capacity of 25kg, and 30 degrees of freedom throughout its body. Moreover, the core advantage of “Tiankui-1” lies in its specially modified motion structure, power system, and electrical circuits, allowing it to operate in explosive environments such as those containing hydrogen or methane. “Such characteristics have made ‘Tiankui-1’ highly sought after since its debut, with many units proactively contacting us for discussions,” Liu Danfu said. In these complex special scenarios, explosion-proof humanoid robots can effectively provide unmanned solutions, improve operational efficiency, and reduce personnel casualty accidents.
Estun Cuzhou continues to focus on the industrial field. Statistics show that Nanjing Estun Automation Co., Ltd. has led the Chinese industrial robot market for two consecutive quarters, with a market share of 10.5% in the first half of 2025. “Estun’s advantages in the industrial robot field also give us confidence to grow,” Zang Jiawei said. They will continuously enhance the hand-foot-eye-brain sensing and control coordination capabilities of the humanoid robot Codroid and focus on validating solutions in specific industrial manufacturing scenarios. Starting from simple functions like inspection and handling, they will gradually upgrade to precise operations such as assembly and inspection, ultimately enabling the replacement of existing factory production lines entirely according to specific production needs. Unlike robotic arms, which have relatively single operations, humanoid robots have much stronger generalization capabilities. “Currently, upgrading production line processes or modifying workflows might require replacing robotic arms, but in the future, it would only require ‘telling’ Codroid.”
Huaxia Robot’s chosen track can be discerned from their corporate slogan—”Humanoid Robots with Warmth.” In Huaxia’s exhibition hall, reporters saw the humanoid robot “Xia Lan” with a face almost identical to a real human’s. “We should be among the earliest domestic companies to develop humanoid robots with simulated human faces,” Zhou Pan said. “Xia Lan” has 30 motors in its head alone, capable of making various vivid expressions. The company also has self-developed emotional large models, enabling “Xia Lan” to understand the emotions and feelings conveyed by human language. This characteristic gives Huaxia’s humanoid robot products a stronger sense of affinity during human interactions. Currently, the company’s products have been applied in real scenarios such as tourist attraction guidance and bank lobbies. “Humanoid robots confined to laboratories are ‘bonsai,’ with more ornamental value. Only when more products are truly applied in real life can they form a beautiful landscape for the humanoid robot industry.”
4. Underfoot and Afar: The Development of Humanoid Robots Is Unlikely to Have a ‘ChatGPT Moment’
Humanoid robot enterprises are currently delving deeply into their respective targeted fields. However, they also stated that focusing on segmented fields is due to current technological limitations, and the ultimate goal is undoubtedly to develop towards all-around general-purpose humanoid robots.
Before ChatGPT’s emergence, the public did not expect human-computer interaction to reach such levels overnight; artificial intelligence seemed to have “ascended” in one night. Will the development of the humanoid robot industry similarly experience its own “ChatGPT moment” someday—where the various capabilities of humanoid robots rapidly approach or even surpass humans?
“I think it’s very difficult,” Song Aiguo responded bluntly, again citing the scarcity of training data. “One can’t make bricks without straw. The progress of humanoid robots requires accumulation step by step.”
Regarding the iterative path of the humanoid robot industry, Liu Danfu has clear thoughts. He believes a more realistic path is to start from special scenarios with relatively single environments and simple work content, such as inspection and maintenance in some hazardous scenarios, where current humanoid robots are already competent. Then, as the capabilities of humanoid robots improve, they can gradually enter factories and production lines, engaging in more complex but relatively fixed operations like assembly and quality inspection. In this phase, accompanied by large-scale applications in the industrial field, data collection and industrial upgrade speeds will increase exponentially. Finally, when humanoid robots can perform more complex tasks, they will truly enter the civilian field and thousands of households. “At that stage, the number of robots and data collection volume will again experience explosive growth, and truly general-purpose humanoid robots might not be far away.”
Song Aiguo agrees with this roadmap for industrial development and also mentioned an idea that could accelerate industrial evolution—teleoperation. “Currently, with breakthroughs in hardware technology, the degrees of freedom in the hands of humanoid robots have greatly improved. Manual remote control operation can正好 leverage this hardware advantage,” Song Aiguo said. Teleoperation can rapidly enhance the capabilities of humanoid robots, enabling their application in more scenarios. More importantly, the operation process is also a process of data accumulation, helping to collect data in more scenarios and broader ranges, thereby boosting the R&D and training of humanoid robots.
“Data is a crucial part; the training of humanoid robots must proceed step by step, with no shortcuts,” Zang Jiawei also admitted when discussing the industry’s future, saying “it can’t be rushed.”
Aiming to build a “Robot City,” Nanjing has issued the “Nanjing Robot Industry High-Quality Development Action Plan (2024-2026),” focusing on building a “1+N+1” whole-machine system. The last “1” represents the forward-looking layout of the humanoid robot industry. The “Action Plan” explicitly proposes “carrying out pilot applications of humanoid robots,” and the three application fields are precisely special operations, industrial manufacturing, and service consumption, aligning with practitioners’ visions for industrial development paths.
Today, humanoid robots are very close to us—close enough to be right beside us. Yet, humanoid robots are also far from us—far from truly entering thousands of households. However, whether practitioners or industry experts, all firmly believe that as long as every step underfoot is taken well and accumulation gradually occurs, the distant goal of “general-purpose humanoid robots” will eventually be reached.
The road is long, but walking it will reach the destination; the task is difficult, but doing it will accomplish the goal.