Around 2010, academic circles witnessed a surge in AI research fueled by breakthroughs such as the ImageNet project, a publicly available large-scale image database. The field reached a tipping point in 2016 when Google’s AlphaGo defeated Go world champion Lee Se-dol and gained widespread public attention with the release of OpenAI’s ChatGPT in November 2022. Just one year after ChatGPT’s debut, Chinese AI firm DeepSeek launched its open-source general large model, a milestone in the evolution of AI technology.
However, a conversational machine is still far from being the ultimate goal of AI. So what level of advancement is necessary for AI to be truly transformative? The answer may lie in the rise of embodied intelligence. Intelligent agents, ranging from robotic dogs and autonomous vehicles to humanoid robots, are breaking free from the confines of data and screens. They are acquiring physical bodies and venturing into the real world.
“This is an evolutionary revolution from mechanical shells to digital life forms,” He Han, a member of the Chinese People’s Political Consultative Conference (CPPCC) National Committee, China’s top political advisory body, and CEO of Tianyu Digital Technology Co. Ltd., told Beijing Review.
Taking shape
Seventy-five years ago, British computer scientist Alan Turing—widely considered the father of AI—posed the bold question: Can machines think? To explore this, he devised the renowned Turing Test to determine whether a machine could simulate human intelligence. At that time, only four years had passed since the creation of ENIAC, the world’s first general-purpose computer, in 1946.
The term “artificial intelligence” was formally adopted at the 1956 Dartmouth Conference. Over the past seven decades, AI has experienced multiple cycles of hype and decline, yet humanity’s enthusiasm for its exploration has never waned.
Turing foresaw two stages in the development of machine intelligence: disembodied intelligence and embodied intelligence. Current systems like DeepSeek or ChatGPT represent disembodied intelligence, while embodied intelligence necessitates a physical entity.
In 1986, Rodney Brooks, former Director of the Massachusetts Institute of Technology’s Computer Science and Artificial Intelligence Laboratory, proposed that intelligence is embodied and contextual, emerging through interaction with real-world environments rather than relying on pre-programmed knowledge or objectives. Recent breakthroughs in neural networks, large models and sensory technologies have reignited interest in the concept of embodied intelligence.
Stepping into the real world
At a group discussion on March 7 at the Third Session of the 14th CPPCC National Committee, there was a debate over whether AI can replace humans in the workforce. One side acknowledged the potential for AI to replace humans, particularly with advancements like embodied intelligence. The other side was skeptical, deeming the likelihood of replacement minuscule, especially in fields like experimental science.
This debate revealed evolving perceptions of AI’s capabilities.
As Chen Tianshi, CEO of Cambricon Technologies and an AI expert, explained, embodied intelligence is akin to a brain in the cloud that grows arms, legs and hands, enabling interaction with the physical world.
Xu Lijin, Chairman of the Anhui Wuhu Robot Industry Development Group, adopted a cautious stance regarding the displacement of human labor by robots and AI. Xu thinks that economic viability serves as a primary factor in the adoption of robots; if the technology fails to meet their standards, people won’t adopt robots.
“We need to solve how machines can achieve perception, cognition, decision-making and execution,” he said.
CPPCC National Committee member He thinks that embodied AI can learn better than traditional AI. He gave an example: Traditional industrial robots require engineers to program parameters, whereas embodied intelligence uses 3D vision and imitation to learn by watching—just like humans. “It enables robots to mimic human movements as if the brain and body are integrated,” he explained.
“Moreover, robots develop generalization capabilities—after learning to grasp a cup, they can likely grasp a water bottle without reprogramming.” However, questions remain: “If a robot crushes a cup, it won’t apologize.”
He noted that currently general AI applications are concentrated in content generation, customer services, programming and other desktop scenarios, with limited industrial use. This is because industrial environments are 3D, while most large models are trained on 2D data, including language, images and video.
“Embodied intelligence deeply integrates AI into 3D physical environments, enhancing perception, comprehension, interaction and decision-making in real-world contexts and redefining relationships between humans and machines in both the physical and virtual worlds,” He told Beijing Review. “It vastly expands AI’s depth and breadth of understanding, enabling more scientific, rational and adaptive decisions.”
Qiao Hong, a member of the Chinese Academy of Sciences (CAS), believes embodied intelligence holds many promising prospects, from facilitating more efficient lifestyles to driving innovation across industries. “As these technologies develop, embodied intelligence will profoundly reshape our daily lives and work,” she said.
Humanoid robots are prime examples of embodied intelligence, acting as ideal carriers for these systems. Today, humanoid robots are utilized in different applications, including rescue drills, exhibition guiding, and factory quality control.
China’s National and Local Co-Built Embodied Artificial Intelligence Robotics Innovation Center (HUMANOID) has developed the Tiangong humanoid robot, which can climb 134 steps, run at speeds of 12 km per hour and navigate complex terrains. The center is based in Beijing and jointly supported by the Central Government and the Beijing Municipal Government.
Moreover, Qiao’s team at the Institute of Automation of the CAS, created the Q-series humanoid robots, establishing a “universal humanoid robot factory.”
“This platform can rapidly produce low-cost, high-performance robots for industrial and agricultural applications,” Qiao said.
Technological breakthroughs
The embodied intelligence sector presents huge market potential, with applications in smart manufacturing, homes, healthcare and other services. However, the industry also faces several obstacles.
He said an absence of a universal algorithm development platform leads to redundant investments and inefficiencies. Additionally, a lack of standardized data restricts the depth of training, while the absence of regulations also hampers the sector’s growth.
To address these issues, He’s company, Tianyu Digital, based in Hangzhou, Zhejiang Province, is developing a universal embodied AI system called Behavision. This system aims to integrate algorithms, data and computing power, creating interoperable “brain” and “cerebellum” platforms to accelerate the next phase of the robot industry.
Open-source platforms present a practical solution for the trials in the field of embodied intelligence. On March 12, Huisi Kaiwu, the world’s first universal embodied AI platform that supports multiple body types and scenarios was launched by the Beijing-based HUMANOID. The platform is expected to accelerate the development of robotics, autonomous driving and human-machine interaction.
On March 21, the Shanghai-based National and Local Co-Built Humanoid Robotics Innovation Center, Humanoid Robot (Shanghai) Co. Ltd., released a new embodied AI simulation platform known as Ge Wu, which is designed to be a high-performance simulation environment for robot development.
Furthermore, the Beijing center released the first standardized dataset for embodied intelligence, which initially contains 100,000 entries this January. That same month, China also launched its first training ground for heterogeneous humanoid robots to overcome issues in data collection and standardizing training protocols.
China has a big market for humanoid robots. According to the 2024 Humanoid Robot Industry Report, China’s humanoid robot market is projected to reach 5.3 billion yuan ($730 million) by 2025 and 75 billion yuan ($10.3 billion) by 2029. Global consulting firm McKinsey & Company has projected that the global embodied intelligence market may exceed tens of trillions of yuan (trillions of U.S. dollars) by 2030. With embodied intelligence now included in this year’s government work report, delivered by Chinese Premier Li Qiang at the annual session of the National People’s Congress, the top legislature, on March 5, the development of the sector has officially become a priority for China. With embodied AI entering factories, streets and homes, a more intelligent future is rapidly approaching. –The Daily Mail-Beijing Review news exchange item
