Wang Tianmiao: In 2025, embodied intelligence will kick - start the industrial implementation led by "intelligence demarcation" | The Frontline
The world model has not been established, the computing power cost is extremely high, and the operation and safety issues have not been resolved. How to overcome the three challenging problems in the wave of embodied intelligence and promote the industry to shift from "showcasing skills" to industrial implementation has always been the focus of attention in the industry.
At the 2025 Science and Technology Innovation Changemakers Conference hosted by the Zhiyou·Yarui Science and Technology Innovation Platform, Professor Wang Tianmiao, the initiator of the Zhiyou·Yarui Science and Technology Innovation Platform and the honorary director of the Robotics Research Institute of Beihang University, released the "2025 China Embodied Intelligence Industry Star Map" and proposed a new idea of breaking through the current bottleneck of industrial implementation with "intelligent demarcation".
The "2025 China Embodied Intelligence Industry Star Map" divides the industry into five core sectors: base technology, general embodied intelligence, special - purpose embodied intelligence, dexterous hands and end - effectors, and core components, covering more than 300 representative enterprises in the upstream and downstream of the domestic industrial chain. From the perspective of technological evolution, the five core sectors always pull each other, promoting the coordinated development of the overall industry.
Wang Tianmiao reviewed the main driving forces at each stage of the development of embodied intelligence: From the 1950s to the 1980s, artificial intelligence robots started with the technology base. There was not much progress in technology for more than a decade. However, Japan took the lead in promoting the introduction of a hardware platform with a general base. The technology base entered specific industrial scenarios, from heavy industries such as welding, casting, and assembly to daily consumer goods such as automobiles, motorcycles, and household appliances. Relatively general six - axis four - cylinder and parallel technologies were formed. China made great progress in small six - axis and collaborative robots, and half of the world's industrial robots are made in China.
From the 2010s to the 2020s, the breakthrough point of development returned to the technology base. There was a substantial breakthrough in the deep learning of artificial intelligence. ChatGPT and Tesla's humanoid robots set off a wave of embodied intelligence. "Technology" has become the direct variable driving the transformation of the current embodied intelligence industrial chain.
However, although the generalization ability of general technology provides the possibility for application implementation, it still faces multiple challenges such as safety, efficiency, and maintenance cost at present. The next round of transformation of the embodied intelligence industrial chain will be triggered by "applications".
Professor Wang Tianmiao released the "2025 China Embodied Intelligence Industry Star Map"
"In 2025, the industrial implementation led by 'intelligent demarcation' will be launched." Wang Tianmiao proposed the concept of "intelligent demarcation": By defining boundaries in specific scenarios such as manufacturing, consumption, medical care, and business, re - examine the problems of models, computing power, and platforms, understand the diverse needs of embodied intelligence in different scenarios, drive the support of the general base Agent, and re - define core components and special - purpose actuators. Its purpose is to "form a pulling force in the leading vertical sectors first, and at the same time be able to design and manufacture global intelligent hardware products to form an ecosystem of manufacturing services."
Wang Tianmiao further said that a series of sub - fields such as unmanned delivery, unmanned sanitation, companion communication, Agent assistants, supermarkets and pharmacies, hospital applications, and manufacturing logistics, welding, and grinding are expected to be implemented first.
The industrial circle expects to explore the feasibility of "intelligent demarcation" in the second half of the development of embodied intelligence, transforming the difficult problem of constructing a complex, time - varying, and uncertain world model into constructing small vertical models that focus on specific real - world environments and are easier to "solve within a limited scope".
The key to this process is to be led by "operation", systematically driving the integration and adjustment of various nodes such as the "Al Agent model", "multi - modal tools for VLM/VLA model and data training", "adapted computing power domain control", "operating environment", and "process flow". Ultimately, solve problems such as difficult generalization, difficult real - time performance, poor data quality, low operation fault tolerance, and high cost, and accelerate the emergence of leading new - force start - up companies.
Wang Tianmiao pointed out that embodied intelligence is the key technology and the breakthrough point for high - end science and technology industries in China to cope with the "once - in - a - century change". Facing problems such as strategic games among major powers, overcapacity in traditional industries, and over - dependence on real estate, AI technological innovation, physical AI (embodied intelligence), global cooperation, and trade will promote China's future economic recovery and long - term growth.
China's embodied intelligence industry is expected to bid farewell to the chaotic state of pursuing general models and shift to implementation breakthroughs led by "intelligent demarcation" in vertical scenarios.