Although general artificial intelligence is the ultimate goal, the path to industrial implementation may be more like laying eggs along the way in practical scenarios, gradually accumulating data, optimizing models, and reducing costs, in order to ultimately unleash productivity and life potential. From technological breakthroughs to commercial implementation, the robotics industry is standing at the starting point of a new round of explosion. Under the multiple drivers of capital, policies, and scenarios, have robots ushered in their own “ChatGPT moment”? Is humanoid a mandatory or optional?
Behind these issues, macro data provides a certain reference for the industry.
IDC predicted in July this year that the global robotics market will exceed $400 billion by 2029, with China accounting for nearly half of the market. At the same time, the global shipment of commercial service robots has exceeded 100000 units in 2024, with delivery robots and cleaning robots ranking among the top with a share of 38.4% and 33.3% respectively, while Chinese manufacturers account for 84.7% of shipments in this field.
For the humanoid robot market, IDC predicts that the shipment volume of commercial humanoid robots in China will be about 5000 units by 2025, and will increase to nearly 60000 units by 2030, with a compound annual growth rate of over 95%. Various signs indicate that the robot market is rapidly expanding, from labor shortages to technological innovations.
During ROSCon China 2025, Interface News reporters chatted with several industry professionals on multiple topics to discuss several highly relevant issues in the field of robotics.
Topic 1: When will embodied intelligence usher in its ChatGPT moment?
Several industry insiders have different opinions on the ChatGPT moment with embodied intelligence.
Hu Chunxu, Vice President of Sweet Potato Robot Developer Ecology, stated in an interview with Interface News that he is confident in the future of embodied intelligence and the robotics industry.
From the perspective of big models and AI driven, we are entering the era of intelligence, and robots will inevitably be reshaped by AI. I am very optimistic about embodied development and firmly believe that there will be robots landing on a large scale in the future. ”Although he also acknowledges that there is currently a lack of universality, with robots performing well in one scenario but potentially experiencing a skyrocketing failure rate in another, in his view, this is a necessary process.
Tan Weijia, Secretary General of the Shenzhen Robot Association, also pointed out that the penetration rate of robots has been very low in the past decade, “only in the single digits”, because every new scenario requires expensive secondary development, and enterprises cannot afford the cost. And embodied intelligence has “opened the door” for the industry, shortening the development and implementation cycle, allowing more basic performance improvements to be made using AI algorithms.
She believes that embodied intelligence may give rise to phenomena similar to ChatGPT, or it may accumulate data in specific scenarios and “lay eggs along the way” to create business value opportunities.
Correspondingly, there are some dialectical voices.
Shi Fengming, the head of innovation business technology at Feixi Technology, reminded in an interview with Interface News that embodied intelligence is one of the potential paths to achieving general artificial intelligence, and technological bottlenecks and business difficulties are real. “We should be cautious about excessive promotion in the short term, and maintain rational optimism in the long term
He emphasized that it is more important to solve the fundamental problem of how “intelligence” can effectively and reliably interact with the real physical world.
So, under the attention of policies and capital, does it mean waiting for a turning point? Several industry insiders expressed a more biased view towards “blooming along the way” when interviewed by Interface News.
Yao Jiajun, a visiting scholar from the School of Information at the University of the Greater Bay Area, believes that “long-term optimism” and “short-term pragmatism” should be parallel. On the one hand, a true breakthrough in embodied intelligence requires a reconstruction of the underlying architecture. Currently, mainstream VLA has strong coupling between information flow and control flow, simple design, and limitations in ontology communication and computing power, making it difficult to maintain stable generalization in non-standard environments; On the other hand, the acquisition of real data itself also faces human resistance.
Therefore, instead of pursuing universal robots from the beginning, it is better to first do ‘scene universality’ in difficult to recruit and high-risk workstations, and accumulate high-value data and process knowledge along the way. ”Yao Jiajun stated.
Topic 2: Human form, is it a necessary form or an optional solution?
The future path of humanoid robots has always been controversial in the industry. McKinsey’s analysis released in June this year pointed out that general-purpose robots come in various forms and do not necessarily have to imitate humans, but humanoid robots do have advantages in adapting to existing environments. They can move in spaces designed for humans without the need for large-scale modifications to the work environment, which is the unique selling point of humanoid robots.
However, from the perspective of industrial application, the current commercial landing path tends to be more flexible and diverse. Liu Yili, the head of the Body Engineering Division at the Beijing Humanoid Robot Innovation Center, mentioned that the domestic humanoid robot market is just starting to take off. Last year, sales were only a few hundred units, but this year they are expected to increase to about 20000 units. “Most of these robots are being used in scientific research and education fields, and their actual entry into industrial or service scenarios is still being verified
Echo, a technical expert in perception and autonomous system of the National and Local Co built Humanoid Robot Innovation Center, also suggested that instead of rushing to eat all the scenes at once, we should, like the development of Internet and aerospace technology, invest in some special and national level support scenes first, accumulate experience and then promote them.
Zhineng Zhixin cigarettes, on the other hand, are analyzed from the perspective of application structure. Unstructured scenarios such as home care are technically difficult, and in the short term, they should start with semi-structured scenarios and gradually transition. She also mentioned the RaaS service model for robot leasing, which can lower the initial investment threshold and allow companies to try it out before expanding.
Overall, the industry recognizes a mindset that varies depending on the scenario. Humanoid robots are not a must-have for all applications, but they have a natural advantage in seamlessly integrating with human living environments. Alternative solutions may include adapting to the environment or choosing other platform solutions.
Topic 3: Cost and Scenario, How to Achieve ROI?
The matching of cost and application scenarios is crucial for robots to truly enter the market. Despite the broad market prospects, the actual penetration rate of robots is still very low at present.
Tan Weijia pointed out that the penetration rate of manufacturing robots is only in single digits, and even in the field of intelligent assistance, it is difficult to make significant breakthroughs, because every new scenario requires expensive secondary development and deployment.
In fact, enterprises need to have a clear ROI in order to adopt it on a large scale. Otherwise, even if the equipment can work 24 hours a day, it will be difficult to recover costs due to low efficiency. This requires manufacturers to optimize configurations based on scenario requirements.
Yao Jiajun also added that in non-standard scenarios such as welding, workers have a psychological aversion to data collection and are afraid of being replaced. He believes that instead of pursuing a one-step universal robot, it is better to first achieve scenario universality in specific high-risk or difficult to recruit fields, gradually promoting technology implementation and benefit returns.
Gu Qiang, co-founder of Gu Yueju, compared the history of the mobile phone industry and believed that with mass production and technological maturity, the cost of robots will eventually decrease. However, for now, the focus is still on effective scenarios.
Liu Yili emphasized that the true value of humanoid robots comes from the added value of “emotions and services”, not just hardware costs. He pointed out that currently many companies are approaching losses in price wars in order to win singles, and “such internal competition is not conducive to the health of the industry”.
Several industry insiders generally agree that before the price is implemented, robots need to first prove that they can solve the problem, and it is more meaningful to discuss price reduction when practical usage scenarios arise.
Topic 4: How to solve the limitations of data and standards?
The bottleneck of data collection and standardization has long constrained the development of robots.
Hu Chunxu admitted that there is currently no unified data collection standard in the industry, and various companies have different standards for collecting multimodal data such as visual, language, and force feedback. The lack of unified standards means that the existing data is mostly dirty data with uneven quality, making it difficult to directly feed into models such as VLA for use.
He pointed out that compared to autonomous driving in cars, robots lack data samples of orders of magnitude. “Running tens of millions of cars on the road can obtain massive amounts of real data, but robot scenarios do not have as many samples, and data problems are 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 was necessary to establish a universal approach or world model to achieve cross platform migration.
In terms of standardization, the industry is still in the early stages of construction. Liu Yili revealed that there is currently no consensus on the process flow, testing standards, performance indicators, and even key component interfaces of humanoid robots. For example, there is no unified standard for what kind of sports safety is considered qualified, and how to evaluate the reliability and durability of robots. The lack of standards means that each company operates independently and it is difficult to promote on a large scale.
In addition, companies are also cautious about data sharing. Sensor manufacturers and algorithm companies are concerned that core data may become their own trade secrets and are unwilling to easily let go. Several industry insiders expressed similar views to Interface News: it is difficult to solve these problems solely by a single institution or country, and more open open-source platforms and ecosystems are needed to collaborate in developing standards.
According to multiple industry insiders, although general artificial intelligence is the ultimate goal, the path to industrial implementation may be more like laying eggs along the way in practical scenarios, gradually accumulating data, optimizing models, and reducing costs, in order to ultimately unleash productivity and life potential.
暂无评论内容