2025-07-16
Humanoid Robots: Charting a New Course Amidst Opportunities and Challenges
Source:Guangming Daily

  The year 2025 is shaping up to be a “breakthrough year” for humanoid robots. From the robots performing the Yangko dance (a traditional rural folk dance) at the CCTV Spring Festival Gala for the Year of the Snake, to the robot marathon in Beijing’s Yizhuang, and the robot fighting competition in Hangzhou, Zhejiang, humanoid robots are moving beyond the confines of the “laboratory” and into a variety of real-world applications. This shift not only serves as a “stress test” for the technology but also represents a crucial leap from flashy demonstrations to practical, everyday use.

  According to the 2025 Humanoid Robots and Embodied Intelligence Industry Research Report released at the 2nd China Humanoid Robot and Embodied Intelligence Industry Conference, China’s embodied intelligence market is expected to reach 5.295 billion yuan by 2025, accounting for roughly 27% of the global market. Meanwhile, the humanoid robot market in China is projected to grow to 8.239 billion yuan, representing about 50% of the global market.

  In the context of the rapid development of artificial intelligence, why is the humanoid robot industry experiencing such explosive growth? What advantages does China have in the development of humanoid robots? Looking ahead, which areas will serve as key indicators for industry growth? With these questions in mind, the reporter conducted a thorough investigation.

  Embodied Intelligence Drives the “Evolution” of Humanoid Robots

  Embodied intelligence is seen as a “key step towards achieving general artificial intelligence.” It refers to intelligent agents achieving integrated perception, cognition, decision-making, and action through real-time interaction with the environment via physical entities. Unlike traditional programmed robots, which act as “precise executors,” intelligent robots equipped with embodied intelligence can autonomously interact with their surroundings and make dynamic decisions, making them “autonomous decision-makers.”

  A 2023 publication in a sub-journal of Nature suggested that the ultimate challenge for the next generation of AI is to pass the embodied Turing test, which involves replicating the sensory-motor capabilities of biological organisms. This includes interacting with the world, exhibiting flexible behavior, and utilizing energy efficiently. “Embodied intelligence gives AI a physical body and the ability to interact with and learn from the physical world. Unlike large models that ‘read thousands of books’ through text and images, this is more akin to ‘traveling thousands of miles’ in the real sense,” explained Zhang Weimin, a senior engineer at the Artificial Intelligence Research Institute of the China Academy of Information and Communications Technology.

  “Integrating embodied intelligence technology into humanoid robots has led to rapid innovation and iteration of related components such as joints, modules, sensors, and dexterous hands. This boosts their abilities in general grasping and motion control, while enhancing spatial awareness and operational precision, making them more flexible and accurate. As a result, humanoid robots become more versatile and adaptable, reducing development costs and enabling faster integration across various industries and into homes,” said Xu Man, a senior engineer at the Chinese Institute of Electronics.

  While large models currently show significant potential in cognition and decision-making, industry experts highlight that humanoid robots’ performance in dynamic real-world scenarios, such as marathons and combat, still depends on human support. The core bottleneck lies in the limitations of their fundamental hardware—key issues such as battery capacity, mobility, flexibility, and stability have yet to be fundamentally addressed.

  “To achieve a truly transformative shift in the humanoid robot industry, it will take at least another 3 to 4 years,” said Wang Xingxing, founder and CEO of Unitree Robotics. He believes that hardware limitations are not the biggest obstacle in the robotics industry. The real challenge lies in the AI capabilities of current robots, which have yet to experience breakthroughs. Whether it’s in AI models, training data, or the application of AI in real-world scenarios, humanoid robots still have significant room for improvement.

  According to Zhang Weimin, the development of the humanoid robot industry, driven by embodied intelligence, faces three major challenges. First, obtaining high-quality data is a significant hurdle; while simulation data is cost-effective, it often fails to reflect real-world conditions. Next, multi-modal perception fusion technology is still in its early stages, resulting in limited ability to perceive and predict complex environments. Additionally, actuators lack the precision required for fine operations and flexibility, while human-robot interaction still needs improvement in terms of naturalness and accuracy. Finally, intelligent agents have a serious deficit in the ability to transfer and generalize knowledge across different scenarios and tasks, as well as from simulated to real environments, making it difficult for them to adapt to the variability and complexity of the real world. This represents the key bottleneck hindering their large-scale application.

  Multiple Advantages Drive the Application and Implementation of Humanoid Robots

  At the 2024 World Robot Conference, the humanoid bionic robot named “Su Shi” captivated visitors. Developed by Liaoning New Dimension Intelligent Technology Development Co., Ltd., this interactive humanoid robot brings Su Shi’s poetry and thoughts to life through deep learning of ancient literature, historical knowledge, and personal experiences, enabling it to interact with visitors at cultural and historical sites. “The robot brings historical figures back to life, and children’s interest in learning ancient poetry has noticeably increased,” remarked Mr. Zhang, who was visiting with his child.

  The development of humanoid robots is driven by a dual approach of “hardware + intelligence”—large models provide an “engaging soul,” while embodied intelligence creates the “appealing appearance.” Thanks to these technological advancements, humanoid robots are gradually becoming more prominent and taking on important roles across various sectors, including “porters” in lights-out factories, “night assistants” in 24-hour convenience stores, “performers” at cultural and tourism attractions, and “companions and caregivers” for elderly people with disabilities or dementia.

  Particularly in the industrial sector, the vigorous development of large models, autonomous driving, and other industries has driven the practical application and deployment of humanoid robots, positioning them to play a key role in industrial manufacturing, smart transportation, home services, and other fields.

  Wu Yiming, a researcher at the Xi’an Institute of Optics and Precision Mechanics of the Chinese Academy of Sciences, China’s high-quality supply chain, developed over years of industrial experience, provides a solid foundation for the advancement of humanoid robots and embodied intelligence. Additionally, China boasts a large pool of engineering talent in fields like artificial intelligence, robotics, and automation, which is essential for the continued growth of the humanoid robot industry.

  “China has a well-established manufacturing system and strong industrial support capabilities, with clear advantages in supply chain integration and large-scale production, enabling the production of highly cost-effective humanoid robots,” said Xu Man. In addition to these industrial and supply chain advantages, China also offers a wide range of application scenarios and a strong market demand. “The diversity of industrial activities has generated a wealth of sector-specific data, providing a vast testing ground, solid data foundation, and ample market space for the development of humanoid robots.”

  Meanwhile, both national and local governments place great emphasis on the development of embodied intelligence, formulating and implementing relevant policies. The Report on the Work of the Government 2025 first listed “embodied intelligence” as a future industry, while also stressing the importance of developing intelligent robots. The Ministry of Industry and Information Technology plans to roll out innovation-driven development policies for embodied intelligence this year. In December, Chongqing introduced the Policy Measures to Support the Innovation and Development of the Embodied Intelligence Robot Industry. Similarly, Beijing issued the Measures to Promote the Innovation and Development of General Artificial Intelligence, which calls for “advancing the research and application of embodied intelligence systems.” Additionally, in October of last year, the National and Local Co-built Embodied AI Robotics Innovation Center was established under the guidance of the Ministry of Industry and Information Technology and the Beijing Municipal People’s Government, focusing on addressing common key challenges in industry development and promoting the practical implementation and application of embodied intelligent products.

  How Can Robots Embrace More Possibilities?

  Are humanoid robots the “optimal solution” for moving from the “laboratory” to real-world applications?

  At the FAW Jiefang Qingdao Automobile Plant, 130 welding robots forge steel frames amid flying sparks, while robotic arms follow pre-set programs to complete the full-vehicle spray-painting process. At the Fengxian branch of Shanghai International Peace Maternity and Child Health Hospital, over 10 KEENON medical robots assist with tasks like drug delivery, specimen transport, and waste collection. In the steep-slope navel orange orchards of Zigui County, Hubei, DJI agricultural drones act as “aerial backpacks,” solving the global challenge of large agricultural machinery being unable to operate in mountainous terrain... These real-world scenarios clearly show that “non-humanoid” robots are carving out their place in the industrial fabric in specialized forms, becoming tangible embodiments of new quality productive forces.

  “Currently, the synergy between the ‘brain’ (AI decision-making) and the ‘cerebellum’ (motion control) of humanoid robots is still underdeveloped. In the early stages of embodied intelligence, the practical path to achieving general intelligence is to invest limited R&D resources in task-specific robots for particular scenarios and accumulate the necessary technology and data by solving concrete problems such as material handling, quality inspection, and picking,” said Gao Feng, Chair Professor at the School of Mechanical Engineering and Power Engineering, Shanghai Jiao Tong University. He believes that robot industrialization should move away from creating human-like forms and instead focus on solving practical tasks, with application-driven R&D, rather than blindly pursuing general-purpose humanoid robots.

  Many industry experts also argue that the infinite potential of the robotics revolution can only be realized when technology is adapted to meet scenario requirements rather than conform to external forms.

  Zhang Weimin provides a conceptual analysis of embodied intelligence: “Embodiment is the foundation of embodied intelligence. It refers to having a body capable of performing tasks through interaction, perception, and action. The form of the embodied entity does not have to be restricted to a ‘humanoid’ appearance.”

  Xu Man believes that, beyond humanoid robots, embodied intelligence can also be applied to a wide range of robots designed for specific scenarios, such as industrial robots, service robots, and specialized machines, as well as intelligent products and equipment like ICVs, intelligent glasses, intelligent home appliances, intelligent security systems, and drones. “However, it is undeniable that humanoid robots are the ‘best platform’ for embodied intelligence.” In Xu Man’s view, humanoid robots, with their human-like posture and body structure, will showcase enhanced capabilities such as autonomous learning, intelligent decision-making, and intelligent interaction when powered by embodied intelligence technologies. Beyond performing precise operations such as grasping, transporting, and assembly, they can also execute various movements, including walking, running, and jumping, while completing diverse tasks in complex environments. Consequently, they will demonstrate tremendous application potential across healthcare, elderly care, service, and manufacturing sectors.

  “Driven by embodied intelligence, humanoid robots will become increasingly capable of ‘observing,’ ‘understanding,’ and ‘imagining,’” said Xu Man. Currently, embodied intelligence is at a critical juncture, poised to usher in widespread commercial application. It is expected to gradually be implemented across industries such as agriculture, manufacturing, and services, with a profound impact on both human productivity and daily life.