In the rapidly evolving field of artificial intelligence, the development and application of embodied intelligence, which includes humanoid robots, have become a central focus. While breakthroughs in robots playing soccer, boxing, dancing, and running marathons are met with enthusiasm, parallel in-depth considerations are underway regarding their ethical, legal, and societal implications. Recently, at Chengdu’s Tianfu Square, a location with daily foot traffic peaking at 100,000 people, five robot “police” began routine patrols alongside human officers. Typically operating in pairs and rotating every two to three hours, these embodied robots transmit live footage and anomalies to a command center in real-time during patrols, assisting in alerting nearby personnel. This futuristic scene epitomizes China’s accelerated deployment of embodied intelligence across multiple scenarios and its push for industrial growth. Statistics indicate that the domestic embodied intelligence market surpassed 480 billion yuan last year. With continuous advancements in large model technologies, the market’s growth potential is immense, projected to exceed one trillion yuan by 2031. As the global race for embodied intelligence intensifies, the future is arriving—but is humanity truly prepared?
Embodied intelligence represents a critical frontier in artificial intelligence research and application, though it is not a recent concept. As early as the 1950s, AI pioneer Alan Turing envisioned a prototype of embodied intelligence in his paper “Computing Machinery and Intelligence,” proposing machines equipped with superior sensors to interact with humans and learn like infants. Around the same time, Norbert Wiener, the founder of cybernetics, introduced the idea of “behavioral intelligence.” Embodied intelligence can be defined as intelligent agents—such as robots, drones, and smart vehicles—engaging in real-time interaction with their environment through physical entities to achieve integrated perception, cognition, decision-making, and action. Its core philosophy challenges the “disembodied” limitations of traditional AI, emphasizing that intelligence must be shaped and manifested through dynamic interactions between the body and the environment.
“The ‘body’ is key to moving ‘intelligence’ from digital space into the physical world,” explained Professor Jiang Yugang, Vice President of Fudan University and Dean of the Institute of Intelligent Robots and Advanced Manufacturing Innovation. Traditional artificial intelligence often prioritizes symbolic computation and data processing, whereas embodied intelligence emphasizes “learning through interaction.” This shift is quietly redefining the understanding of “intelligence,” as there is growing anticipation for intelligent agents to step out of virtual realms and into complex real-world environments, learning and evolving through continuous perception and engagement. The transition from intelligence that “computes and reasons” to agents that “understand and act” marks a significant milestone in AI’s progression to the next stage.
Although the terms “artificial intelligence,” “embodied intelligence,” and “humanoid robots” are frequently used interchangeably, they each have distinct emphases: artificial intelligence focuses more on algorithmic and computational aspects of information processing and cognitive enhancement; humanoid robots center on achieving physical forms that resemble humans; and embodied intelligence stresses the synergistic evolution of “agent-environment-algorithm” as a trinity, highlighting how physical entities can fully leverage perception, decision-making, and execution mechanisms while continuously evolving through environmental interactions. According to Professor Jiang, enabling AI to perceive the world through a “body” and respond flexibly faces significant technical hurdles beyond hardware or contextual understanding bottlenecks. The current intelligence heavily relies on data, but in the realm of embodied intelligence, acquiring high-quality, diverse training data is extremely costly and difficult to sustain. To genuinely realize “understanding and influencing the world with a body,” it is essential to break away from existing paradigms and develop intelligent agents capable of autonomous learning and collective co-evolution, representing a promising direction for future exploration.
In recent years, a wave of technological convergence has injected strong momentum into the development of embodied intelligence. Computer vision, lidar, and depth cameras have enabled millimeter-level environmental perception, while deep reinforcement learning allows intelligent agents to optimize strategies through trillions of trials in simulated environments. For instance, NVIDIA’s Isaac Gym platform supports parallel training for millions of agents, accelerating this process. OpenAI’s Dactyl robotic arm, utilizing tactile feedback, successfully solved a Rubik’s Cube, demonstrating the potential of embodied intelligence in fine manipulation. The PaLM-E multimodal large model further integrates language understanding with robot control, ushering embodied intelligence into a new era of semantic interaction.
China has also made notable strides in this domain. By December 2024, the country had filed over 219,900 patent applications for embodied intelligence, ranking first globally and accounting for approximately 26.45% of worldwide applications. This underscores the nation’s commitment to advancing embodied intelligence technologies and their applications.
| Category | Data | Year/Period |
|---|---|---|
| Domestic Market Size (China) | Over 480 billion CNY | 2024 |
| Projected Market Size (China) | Exceeding 1 trillion CNY | 2031 |
| Global Patent Applications (China) | 219,900+ | As of December 2024 |
| Global Share of Patent Applications | Approximately 26.45% | As of December 2024 |
As embodied robots transition from theoretical research to practical applications, they are poised to first disrupt sectors heavily reliant on manual labor, characterized by complex environments and labor shortages. In manufacturing, for example, future intelligent agents are expected to debut in roles with high repetition or lower skill requirements, such as搬运, assembly, and quality inspection. These embodied robots will not only perform basic operations but also perceive equipment status, predict risks, and autonomously optimize processes to some extent. While this does not imply the complete replacement of all manufacturing jobs, the gradual deepening of human-machine collaboration will continually reshape traditional production models, fostering more efficient and intelligent systems.
Currently, in industrial manufacturing, embodied intelligence robots can autonomously adapt to complex production environments, enabling flexible manufacturing and enhancing productivity. Denmark’s Universal Robots maintains leadership in the collaborative robot market, with its latest product, the UR20, offering greater load capacity and improved safety, widely used in precision industrial scenarios like automotive parts manufacturing and electronic assembly. Switzerland’s ANYbotics has introduced the quadruped inspection robot ANYmal, designed for hazardous environments such as petrochemical plants and energy facilities, successfully replacing human workers in inspection tasks.
In logistics services, Boston Dynamics has launched the Stretch logistics robot and the Spot quadruped robot. The former focuses on automated货物搬运 in warehouses, while the latter is extensively applied in industrial inspections and public safety. In intelligent transportation, Waymo’s Robotaxi service has been fully deployed in cities like Phoenix, San Francisco, and Los Angeles, with plans to expand to Miami, Washington, D.C., and other urban areas.
The medical field is another critical application area for embodied intelligence. According to China’s National Health Commission “2024 Survey Report on the Application Status of Medical Rehabilitation Equipment,” rehabilitation exoskeleton robots are already in use across more than 20 provinces, municipalities, and autonomous regions, assisting patients with motor功能障碍 in rehabilitation training. In the United Kingdom, CMR Surgical’s Versius laparoscopic surgical robot has been implemented in hospitals across multiple European countries and Asian markets, facilitating minimally invasive procedures.
Simultaneously, there is growing anticipation for embodied intelligence to accelerate in the domestic service sector. In response to this interest, intelligent home robots capable of tasks like laundry, watering plants, hanging clothes, pouring milk, toasting bread, and even providing companionship and guided learning are gradually becoming a reality and entering commercialization phases. These advancements highlight the expanding role of embodied robots in everyday life.
“The embodied intelligence prototypes we see today have demonstrated impressive technical prowess in motion control and single-task execution, which is truly awe-inspiring. However, several key issues must be addressed before achieving ‘true embodied intelligence,'” Professor Jiang noted. First is environmental adaptability. Current robots largely operate in pre-set scenarios and tend to ‘malfunction’ when environments become complex or change. Compared to human perception, existing AI perception systems remain relatively simplistic and fragmented. Humans naturally integrate visual, auditory, tactile, and force feedback to comprehend their surroundings, whereas current embodied intelligence systems struggle to achieve this level of multisensory fusion. For example, in rescue operations at chemical plant explosions, rescue robots need to simultaneously perceive factors like fire intensity, smoke density, and structural stability of buildings—capabilities that present technology cannot yet match human efficiency in integrating.
Second is autonomous learning capability. Humans learn through trial and error, experience accumulation, and intuitive judgment, but current AI lacks genuine “intuition” and “experience,” making it difficult to cope with environmental changes. When confronted with novel scenarios not covered in training data, embodied intelligence systems may fail to make reasonable decisions and actions. Third is deep collaboration with humans and the environment, such as understanding human intentions, emotional shifts, and responding flexibly to environmental changes—all of which are still in nascent stages. Human walking involves millisecond-level coordination of over 200 muscles, whereas existing AI motion systems appear crude in comparison. For robots to achieve agile movement, they must overcome challenges in dynamic balance, fine manipulation, and energy efficiency.
As embodied intelligence becomes more deeply integrated into human life, a series of ethical and social governance issues require proactive attention. Privacy and data security are primary concerns. Embodied intelligence devices collect vast amounts of user data during operation, including personal privacy and behavioral habits, posing risks of leakage during storage, transmission, and usage. For instance, home service robots may gather information on family members’ daily routines and consumption habits; if improperly accessed or misused, this could severely compromise user privacy.
Another issue is the ambiguity in liability for safety incidents. Embodied intelligence systems often possess significant mobility, potentially causing physical harm to people and environments through incidents like robot失控 or operational errors. When accidents occur, determining responsibility among developers, manufacturers, users, or other stakeholders remains challenging.
Emotional and ethical dilemmas are equally pressing. The design of emotional value in embodied intelligence could spark unprecedented debates. As these systems integrate into家庭, education, and other life contexts, societal ethical issues will grow more complex. People might develop dependencies on intelligent agents or form emotional bonds, leading to “emotional misplacement” that interferes with value judgments and behavioral decisions, posing new moral and legal challenges. Some even question whether embodied intelligence agents, if equipped with pain perception (even simulated), would impose moral responsibilities on humans.
In response to this, Professor Jiang emphasized that this direction warrants serious consideration—”Even if ‘pain perception’ is simulated and not a biological ‘experience of suffering,’ we must confront the reality that humans instinctively project emotions and moral judgments onto these human-like intelligent agents. This highlights the need to early contemplate, from a socio-ethical perspective, how to treat embodied intelligence agents, avoiding scenarios of ‘abuse’ towards them and preventing such ‘abuse’ from occurring.”
“In science fiction films and literature, embodied intelligence is often portrayed as an ‘out-of-control threat’ or a ‘new species surpassing humans.’ As an artistic expression, these works amplify human expectations and fears towards unknown technologies. Although these imaginations lack direct real-world basis, they reflect deep-seated anxieties within us,” he stated. From a current development standpoint, embodied intelligence remains far from “out-of-control” or “surpassing humans.” What truly demands attention is not whether they will exceed humanity, but how to establish effective governance mechanisms to guide the technology towards benevolent development.
The journey towards fully realized embodied intelligence involves continuous innovation and collaboration across disciplines. The integration of embodied robots into various sectors underscores their potential to transform industries, but it also necessitates robust frameworks to address accompanying risks. As research progresses, the focus should remain on enhancing the adaptability, learning capabilities, and ethical alignment of these systems to ensure they serve humanity’s best interests. The global community must work together to foster responsible development, ensuring that embodied intelligence becomes a force for positive change rather than a source of conflict.
| Application Domain | Examples of Embodied Robots | Notable Features |
|---|---|---|
| Industrial Manufacturing | Universal Robots UR20, ANYbotics ANYmal | Autonomous adaptation, hazardous environment operation, precision tasks |
| Logistics Services | Boston Dynamics Stretch, Spot | Automated搬运, inspection, public safety |
| Intelligent Transportation | Waymo Robotaxi | Autonomous vehicle services, urban mobility |
| Medical and Rehabilitation | Rehabilitation exoskeletons, CMR Surgical Versius | Motor function assistance, minimally invasive surgery |
| Domestic Services | Home assistant robots | Chores, companionship, educational support |
In conclusion, the rise of embodied intelligence and embodied robots presents both opportunities and challenges that require careful navigation. While technological advancements continue to push boundaries, the human aspect—addressing ethical, legal, and social implications—must remain at the forefront. By fostering interdisciplinary dialogue and implementing proactive policies, society can harness the benefits of embodied intelligence while mitigating potential risks. The future of embodied intelligence is not just about creating smarter machines, but about building a harmonious coexistence between humans and intelligent agents in an increasingly interconnected world.