From technological breakthroughs to commercial applications, the robotics industry is poised for a new wave of explosive growth. Driven by capital, policy, and various application scenarios, has robotics ushered in its "ChatGPT moment"? Is humanoid design a necessity or an option?
Behind these issues, macroeconomic data provides some reference for the industry.
In July of this year, IDC predicted that the global robotics market would exceed $400 billion by 2029, with China accounting for nearly half of that. Meanwhile, global shipments of commercial service robots exceeded 100,000 units in 2024, with delivery robots and cleaning robots leading the way at 38.4% and 33.3% market share respectively. Chinese manufacturers accounted for a staggering 84.7% of shipments in this sector.
Regarding the humanoid robot market, IDC predicts that China's commercial humanoid robot shipments will reach approximately 5,000 units in 2025, increasing to nearly 60,000 units by 2030, with a compound annual growth rate exceeding 95%. All indications suggest that the robot market is expanding rapidly, driven by factors ranging from labor shortages to technological innovation.
During ROSCon China 2025, reporters from Jiemian News spoke with several industry professionals on various topics to discuss several issues of great interest in the robotics field.
Topic 1: When will embodied intelligence usher in its own ChatGPT moment?
Several industry professionals hold differing views on the ChatGPT moment of embodied intelligence.
Hu Chunxu, Vice President of the Developer Ecosystem at Digua Robotics, told Jiemian News that he is full of confidence in the future of embodied intelligence and the robotics industry.
"From the perspective of large-scale models and AI-driven development, we are entering an era of intelligence, and robots will inevitably be reshaped by AI. I am very optimistic about the development of embodied robots and firmly believe that robots will be deployed on a large scale in the future." Although he also admits that there is currently a problem of insufficient versatility, "A robot may perform well in one scenario, but its failure rate may soar in another scenario." But in his view, this is an inevitable process.
Tan Weijia, Secretary-General of the Shenzhen Robotics Association, also pointed out that the penetration rate of robots remained very low for the past decade, "only in the single digits," because each new application required expensive secondary development, which was difficult for companies to bear. Embodied intelligence, however, has given the industry a new lease on life, shortening the development and implementation cycle and allowing more basic performance improvements to utilize AI algorithms.
She believes that embodied intelligence may give rise to emergent phenomena similar to ChatGPT, or it may first accumulate data in specific scenarios and then create business value opportunities "along the way".
In contrast, there are some dialectical voices.
In an interview with Jiemian News, Shi Fengming, head of technology innovation at FEXI, cautioned that while embodied intelligence is one of the potential paths to achieving general artificial intelligence, technological bottlenecks and commercial difficulties are real. "We should be cautious about excessive hype in the short term, but remain rationally optimistic in the long term."
He emphasized that it is more important to solve the fundamental problem of how "intelligence" can interact effectively and reliably with the real physical world.
So, with policy and capital attention, does this mean we are "waiting for a turning point"? Many industry insiders interviewed by Jiemian News tend to think that things are "blooming along the way."
Yao Jiajun, a visiting scholar at the School of Information Science and Technology of the Greater Bay Area University, believes that "long-term optimism" and "short-term pragmatism" should be pursued in parallel: On the one hand, a true breakthrough in embodied intelligence requires the reconstruction of the underlying architecture. The current mainstream VLA has strong coupling between information flow and control flow, and its design is relatively simple. Coupled with the limitations of ontology communication and computing power, it is difficult to maintain stable generalization in non-standard environments. On the other hand, the acquisition of real data itself also faces human factors resistance.
"Therefore, instead of pursuing general-purpose robots from the outset, it is better to first implement 'scenario-wide' robots in work positions where recruitment is difficult and risks are high, starting with specific examples and gradually expanding to other areas, accumulating high-value data and process know-how along the way," Yao Jiajun said.
Topic 2: Humanoid form – a necessary form or an optional option?
The future of humanoid robots has been a subject of debate within the industry. A McKinsey analysis released in June of this year pointed out that general-purpose robots come in various forms and do not necessarily have to mimic humans, but the shape of humanoid robots does have an advantage in adapting to existing environments. They can move in spaces designed for humans without requiring large-scale modifications to the work environment, which is a unique selling point of humanoid robots.
However, from the perspective of industrial application, the current commercialization path is more flexible and diverse. Liu Yizhang, head of the Embodied Creation Business Unit of the Beijing Humanoid Robot Innovation Center, mentioned that the domestic humanoid robot market is still in its infancy, with sales of only a few hundred units last year, and an estimated increase to about 20,000 units this year. "Most of these robots are being deployed in scientific research and education, and their actual entry into industrial or service scenarios is still being verified."
Echo, an expert in perception and autonomous system technology at the National-Local Jointly-Built Humanoid Robot Innovation Center, also suggested that there is no need to rush to apply it to all scenarios at once. Instead, it should be like the development of the Internet and aerospace technologies, first investing in some special scenarios supported at the national level, accumulating experience before promoting it.
ZhiNeng Smart Chip, analyzing from a structured application perspective, points out that unstructured scenarios such as home care present significant technical challenges, and in the short term, it's advisable to start with semi-structured scenarios and gradually transition. She also mentioned the Robot as a Service (RaaS) model, which can lower the initial investment threshold, allowing companies to try it out before expanding.
Overall, the industry generally favors a scenario-specific approach. Humanoid robots are not essential for all applications, but they have a natural advantage when seamlessly integrating into human living environments. Alternative solutions could include adapting to the environment or choosing other platform solutions.
Topic 3: Cost and Scenarios, How to Achieve ROI?
Whether robots can truly enter the market depends crucially on the match between cost and application scenarios. Despite the promising market prospects, the actual penetration rate of robots remains very low.
Tan Weijia pointed out that the penetration rate of robots in the manufacturing industry is only in the single digits, and there is little significant breakthrough even in intelligent assistance. The reason is that each new scenario requires expensive secondary development and deployment.
In reality, companies need to clearly define the ROI before large-scale adoption. Otherwise, even if the equipment can work 24 hours a day, it will be difficult to recover costs if the efficiency does not meet expectations. This requires manufacturers to optimize configurations according to scenario requirements.
Yao Jiajun further pointed out that in non-standard scenarios such as welding, workers are reluctant to collect data, fearing they will be replaced. He believes that instead of pursuing a universal robot that can be implemented all at once, it would be better to first achieve universality in specific high-risk or labor-difficult fields, gradually promoting the application of the technology and its benefits.
Gu Qiang, co-founder of Guyueju, drew an analogy to the history of the mobile phone industry, believing that as mass production and technology mature, the cost of robots will eventually decrease, but for now, the core focus should be on effective application scenarios.
Liu Yizhang emphasized that the true value of humanoid robots comes from the added value of "emotion and service," not just hardware costs. He pointed out that many companies are currently engaging in price wars to win orders, with prices approaching losses. "Such infighting is detrimental to the healthy development of the industry."
Several industry insiders generally agree that before a price is finalized, robots need to prove that they can solve problems and that there are real-world use cases before discussing price reductions.
Topic 4: How to resolve data and standards limitations?
Data collection and standardization bottlenecks have long constrained the development of robots.
Hu Chunxu frankly stated that the industry has not yet formed a unified data collection standard, and different companies have different standards for collecting multimodal data such as vision, language, and force feedback. The lack of a unified standard means that the existing data is mostly dirty data with inconsistent quality, making it difficult to feed directly to models such as VLA.
He pointed out that compared with autonomous driving cars, robots lack a massive amount of data samples. "Tens of millions of cars on the road can generate a huge amount of real data, but there are not that many samples in robot scenarios. The data problem is the biggest pain point."
Similarly, Tan Weijia also mentioned that relying on a single robot configuration to collect data in the past was inefficient, and migrating to other structures required a lot of repetitive work. It is necessary to establish a general method or world model to achieve cross-platform migration.
In terms of standardization, the industry is still in its early stages. Liu Yizhang revealed that there is currently no consensus on anything regarding humanoid robots, from manufacturing processes and testing standards to performance indicators and even interfaces for key components. For example, there are no unified standards for what constitutes acceptable motion safety or how to evaluate the reliability and durability of robots. The lack of standards means that each company operates independently, making large-scale promotion difficult.
Furthermore, companies are cautious about data sharing. Sensor manufacturers and algorithm companies worry that core data will become their trade secrets and are unwilling to release it easily. Several industry insiders expressed similar views to Jiemian News: it is difficult for a single institution or country to solve these problems; a more open open-source platform and ecosystem are needed to collaboratively develop standards.
According to many industry insiders, while general artificial intelligence is the ultimate goal, the path to its industrial implementation may be more like "laying eggs along the way" in real-world scenarios. Only by gradually accumulating data, optimizing models, and reducing costs can productivity and life potential be finally unleashed.
This article is from "Jiemian News" , written by Xu Meihui, and published with authorization from 36Kr.
