HCDStock Code 300278
English
Stock Code 300278

Accelerate iteration, explode! The embodied intelligent robot industry is making dual efforts from breakthroughs in hard technology and scenario based implementation

2025-09-23

Accelerate iteration, explode! The embodied intelligent robot industry is making dual efforts from breakthroughs in hard technology and scenario based implementation

CCTV News: By 2025, artificial intelligence has begun to deeply empower the real economy worldwide, and the global manufacturing landscape is also accelerating its reshaping. As one of the key areas for the development of artificial intelligence, "embodied intelligence" was first included in China's government work report in 2025. And intelligent robots are typical terminal products of "embodied intelligence". From the Spring Festival Gala stage to the sports arena, to the production line and daily life scenes, China's embodied intelligent robots are accelerating iteration at a visible speed to the naked eye. 'embodied intelligence' is about to usher in an unprecedented explosive moment.

China's' embodied intelligence 'robots are accelerating iteration

What is embodied intelligence? Simply put, it means allowing artificial intelligence to truly 'grow bodies'. It not only needs to be able to see, hear, and feel like a human, but also learn to plan and make decisions, and independently complete various complex actions.

In 2025, embodied intelligent robots have achieved more technological breakthroughs. Discard the remote control and equip it with the world's first humanoid robot universal visual perception system. The embodied intelligent robot can achieve autonomous running on sports tracks. From prototype to mass production, embodied intelligent robots have entered real scenarios such as inspection, service halls, factories, and elderly care, winning orders worth billions of yuan. By 2025, the size of China's embodied intelligence market is expected to reach 5.295 billion yuan, accounting for approximately 27% of the global market. From ontology to brain, embodied intelligent robots are accelerating iteration.

How are embodied intelligent robots trained and upgraded?

Have you ever seen a humanoid robot with only two legs that can run and play soccer? How are embodied intelligent robots usually trained and upgraded? Not long ago, the reporter visited a robot company's research and development center in Shenzhen, and even had a big competition with a bipedal robot for human-machine "stability". Who could win?

Inside the swaying carriage, reporters and bipedal robots are undergoing a major stability test. Braking, turning, acceleration and deceleration, the robot exhibits a balance that surpasses that of humans, and can even stand steadily with a kick.

The reporter walked into the R&D center of this robot company located in Shenzhen and immediately saw these bipedal robots with different shapes and unique functions.

A multi joint body with multiple degrees of freedom gives the robot an extremely humanoid pair of legs. To truly enable it to handle any complex environment with ease, it still relies on training and algorithms

Entering the R&D center, you can see that it is more like a real training ground, where robots crawl and roll in these complex terrains. Engineers enable bipedal robots to continuously learn skills in the virtual world while also constantly experimenting and optimizing in the real world.

As a developer oriented R&D application platform, bipedal robots can also be equipped with various software and hardware such as vision, brain, robotic arm, etc., to achieve more functional extensions. Team introduction: With the accumulation of underlying technology and supply chain, they launched a brand new humanoid robot in just 6 months.

How to build intelligent robots from the laboratory to the real world?

2025 is the year when humanoid robots move towards industrial applications. Starting from 2025, humanoid robots will enter real environments such as factories and retail stores to assist humans in completing various tasks. How can robots move from the laboratory to the complex real world? How is a smart robot trained?

Can you imagine what it would be like to have a robot kitchen assistant in the future kitchen? In a super luxurious future kitchen, a journalist and a robot partner work together to make tomato scrambled eggs.

This robot can not only serve as a kitchen assistant, but also as a research assistant.

The robot has completed a series of ultra long and complex tasks, including precise sand excavation, liquid pouring, bottle cap tightening, centrifuge insertion, control of pipette, sample transfer among teammates, and instrument operation. This relies not on a "fixed script", but on the autonomous reasoning and planning abilities that robots possess after learning and training.

Professor Lian Wenzhao from the School of Artificial Intelligence at Shanghai Jiao Tong University and founder of Yuanluo Technology introduced that its difficulties mainly lie in two levels. At the high-level level, its entire task sequence is particularly long, and a task may have dozens of steps. Sometimes robots need to perform very delicate operations, such as twisting the lid, opening the cap of a small test tube, or taking a pipette to suck liquid, all of which require extremely high precision for their real-time motion control.

The team has designed a model architecture that combines big and small brains. The robot will feed back useful item information it sees to the brain, and the large model serves as the brain for task analysis, disassembly, reasoning, and decision-making. The cerebellum is responsible for outputting corresponding actions and results according to the goal. The model also extracts the relationships between target objects, which is like the navigation of robot actions, allowing the robot to make autonomous decisions and achieve goals.

Nowadays, the research robots working in these service laboratories are already working in the gene laboratory. With the development of embodied intelligence, more and more robots are entering various positions such as production lines and service industries, moving from laboratories to more complex real worlds.

Journalists explore the largest embodied intelligent data factory in China

Data is often referred to as the 'fuel' for the development of embodied intelligence. Nowadays, in order for a robot to flexibly complete tasks in real scenarios, it still relies on a large amount of high-quality data to support it. How can we quickly and effectively accumulate this data? In Tianjin, there is a "data factory" dedicated to embodied intelligent robots that is trying to provide answers.

This is the largest robot data collection factory in the country with tactile data. In these small cells, there are data collection workshops one by one. Currently, approximately 500000 high-quality data are collected from here every day.

Just as big language models rely on large-scale text corpora, robots cannot move from "thinking" to "doing" without massive real-world empirical data for training. This specially designed digital glove that can record tactile, force, and motion details in real time is the medium that connects the real world and the data world.

The data collection gloves can fully record the sensations on the hands, and with the synchronous capture of dozens of cameras and other sensors, generate a data fragment containing multidimensional elements such as vision, touch, audio, trajectory, and time. This type of data allows robots to both "see" actions and "feel" details during training.

In the field of robot data training, how to improve data quality, reduce collection costs, and adapt data to more ontology training has become a difficult point in robot data collection. They will further utilize these real data to train various models, thereby accelerating the robots towards the real world.

Multiple regions increase policy resources and accelerate the development of embodied intelligent industries

At present, China's embodied intelligent robot industry is making dual efforts from breakthroughs in hard technology and scenario based implementation, and many places are actively laying out policies and resources to promote it.

In Yizhuang, Beijing, the innovation platform represented by the "National Local Joint Construction Intelligent Robot Innovation Center" has gathered more than 300 robot ecological enterprises, with an industrial chain scale of over 10 billion yuan.

Shanghai has established the "National Local Co built Humanoid Robot Innovation Center", with plans to promote the core industry scale to exceed 50 billion yuan by 2027.

The Guangdong Hong Kong Macao Greater Bay Area, as an important gathering place for the global supply chain of embodied intelligent robots, accounts for about 24% of the global supply chain share. In 2025, Shenzhen announced a special investment of 4.5 billion yuan for the AI and robotics industry, focusing on key areas such as humanoid robots, core components, and intelligent computing chips.

In addition, provincial-level humanoid robot innovation centers have been established in Zhejiang, Anhui, Hubei, Sichuan and other places to gather regional industrial advantages, promote technology sharing and joint research and development.

Editor: Yang Shujie

Responsible Editor: Liu Liang

New source: CCTV.com, September 17, 2025

Original URL: https://news.cctv.com/2025/09/17/ARTI2QSCbbsRH9gQEBh6vxDw250917.shtml