Is the Inflection Point for Humanoid Robots in Household Scenarios Finally Here?

Beware of the prosperity that only exists in public relations slogans.

When "accelerated implementation" becomes a high - frequency term in the industry, we need to be more vigilant that the "slogan - style prosperity" of public relations narratives may obscure the real "signals" of technological evolution. The real inflection point for the home scenario has never been a "one - time success" on the manufacturers' demonstration platforms, but a systematic breakthrough in technological maturity, cost controllability, and scenario adaptability.

In 2025, the expected picture of the humanoid robot track has undergone a drastic reversal. In the first half of the year, the industry consensus still regarded the home scenario as the "final battlefield." Achieving implementation within 5 to 10 years was considered a "technological explosion - level" breakthrough, and some practitioners even said it might be out of reach within 20 years. Before the end of the year, leading manufacturers have intensively taken actions, collectively demonstrating the technical feasibility of household robots. The home scenario, once labeled as "distant," seems to be emerging from the fog.

Has the implementation process of household robots really reached a historical inflection point?

At the just - concluded 2025 China Mobile Global Partners Conference, Wang Xingxing's statement was quite representative: "From last year to this year, we have been focusing more on the movement level, but the core goal has always been to promote the real integration of robots into life." The "life scenarios" he referred to cover multiple fields such as home, service, and industry, revealing a clear signal of the shift of the technological focus towards practical applications.

The pace of technological advancement across the ocean is even more rapid: the highly anticipated Figure 03 is directly defined as "born for the home scenario," and its functional positioning is aimed at the rigid needs of household chores; Li Feifei, an authority in the AI field, launched the "2025 BEHAVIOR Challenge," constructing a technological arena centered on complex household tasks, targeting the core ability bottleneck of home services.

Image source: Figure

In the domestic track, technological verification and scenario layout are almost advancing simultaneously: Qianjue Technology and Jiasu Jinhua have completed in - depth joint verification. Their robot Booster T1 has achieved long - term autonomous decision - making and execution of multiple complex tasks in the home scenario; earlier, Fourier launched the GR - 3 companion robot, and Ubtech reached a long - term and in - depth strategic cooperation with Easyhome, focusing on the intelligent upgrade of the home furnishing scenario. Behind a series of actions, the home service robot scenario, once regarded as the "final stage," is accelerating from a pure technological concept towards the eve of commercialization.

Analyzing the industry logic, the home scenario is essentially the "ultimate ToC trial" of robot technology. The personalization of its needs and the unstructured nature of the scenario directly reflect the core ability of robots to evolve towards "general intelligence." This also explains why in the first half of 2025, most enterprises were always reluctant to talk about the implementation schedule of the home scenario. However, the trillion - level market space contained in the home scenario is sufficient to drive the industry to continue to increase investment.

Amid the upsurge, rational voices in the industry are not absent. The high threshold of the home scenario has long been an open industry consensus, and the triple tests of technology, cost, and compliance have not been fundamentally solved:

Leju's judgment points directly to the core threshold: "When humanoid robots reach the intelligent level of a doctor and the price is as affordable as household appliances...", which points out the dual bottlenecks of intelligent level and cost control in one sentence;

Ubtech gives a more specific timeline prediction: "By 2028, robots are expected to obtain sufficient task - operation and scenario - generalization abilities and gradually enter some home scenarios. It is expected to start from needs such as home companionship and ultimately expand to comprehensive home services...", revealing the gradual nature of scenario penetration;

Unitree emphasizes the comprehensive challenges of the scenario: "The home scenario has very high comprehensive requirements for robots. Whether it is the technical level, the ethical and moral level, or the safety level, the thresholds are far beyond imagination," highlighting the complexity of the implementation problem.

What does it take to become a robot butler?

The general idea of the scenario evolution path of humanoid robots is from factories to services to households. This is a process in which robots gradually shift from interacting with objects to interacting with people, and the operation scenarios gradually become more unstructured from structured. Accordingly, the requirements for robots gradually change from being a "tool" to being a "partner." This means that robots need to ensure safety and accessibility when contacting humans and be able to autonomously complete more complex operations in open scenarios.

In such an evolutionary process, the first word that robots cannot avoid when entering the home scenario is "safety." If a robot is used to accompany children or take care of the elderly, the requirement for its fault - tolerance rate in terms of personal safety can never be too high. Almost all doubts and fears come from whether it will harm people. We have seen many videos of "robots falling" or "robots stepping on little girls." Even a slight touch of a little child by a metal robot can have unbearable consequences. Therefore, we can see that the first step in the design of most robots for the home scenario is to change their appearance. The NEO of the Norwegian humanoid robot company 1X, the Figure 03 in the United States, and the GR - 3 launched by Fourier in China are all made of soft materials such as fabric or foam cotton.

Image source: 1X

However, these static and passive - contact protections are just the first step. In daily life, when robots do housework, the problem of success rate in the operation process needs to be improved. If a robot grabs a set of dishes but drops them, or if a fragile electronic product or a luxury watch is mishandled, it will cause varying degrees of property and personal damage. However, the reality is that on one hand, robots are required to have a high success rate, and on the other hand, bad cases occur frequently. These bad cases are essentially caused by the "immaturity of technology" at this stage. The current technological bottleneck is the problem of data acquisition and utilization and model architecture.

When the industry is shocked by the home - use potential of Figure 03, it is not difficult to find by analyzing its recent actions that in the Series C financing recently completed by Figure, it deeply bound a capital provider, Brookfield. This is an asset management company with 100,000 residential properties. Subsequently, Figure announced a cooperation with it. Brookfield can provide Figure with a huge amount of real - world data on human - like navigation and operation in various home environments. These data are the key for its robots to be able to move and operate freely in the home scenario in the future. Li Feifei's recent "Housework Challenge" has a similar logic. It is called the "ImageNet moment" in the field of embodied intelligence. The reason is that it also tries to solve the problem of the lack of training data in the field of computer vision.

However, in addition to data, there is also the problem of how the model can release the potential of data. As Karol Hausman, the co - founder and CEO of Physical Intelligence, said, "The biggest bottleneck is that the success rate of these models has not reached the necessary level. This is not just a problem of data collection. It seems that some improvements in algorithms are also needed. I think that even if we have an unlimited amount of data, we cannot achieve a 100% success rate in complex, long - term, and highly dexterous tasks with the existing algorithms."

Now, let's temporarily put aside these technical differences. The third thing for robots to do when entering the home is product detail improvement, which involves the product completion degree when it is actually delivered. We can see that when robots enter the home, their volume is generally reduced, whether they are quadrupedal or humanoid. It is reported that the weight of Figure 03 is reduced by 9%, and the volume is significantly reduced, which is convenient for movement in the home space, and its simultaneous wireless charging ensures the battery life. It is reported that 1X's NEO was designed to ensure gentle and quiet movements from the beginning. These are the improvements needed to enter the home. Of course, we can also imagine whether the thermal management technology of robots can be organically combined with the body surface temperature. This can make robots "warm." Moreover, can robots provide life records and remember them firmly, similar to a mobile camera device, and deeply participate in life instead of being a bystander? These qualities are the warm - hearted aspects of robots in addition to the hardcore skills required to adapt to the home scenario.

It's still very early for robots to enter households

Robots that can enter households are still in training. The biggest problem with current robots is that the acceptance threshold for users is too high. Technically, the success rate of robots in performing tasks should be infinitely close to 100%. However, the current view in the industry is that there are deficiencies in both data and model architecture, so it cannot be achieved in the short term. Wang He, the founder and CTO of Galaxy Universal, once said at a small - scale media meeting during the WRC that the core reason why multi - modal models such as VLA are slightly weaker than language models at present is that there is very little text - image data and data containing actions, and the visual - based action operation ability is still insufficient.

In addition to traditional data, Lu Cewu, the co - founder of Qiongche Intelligence, also said that VLA is like a clear - soup hot pot, which can only serve a small number of people now. To serve more people, we need to keep adding things to it, that is, additional information. This information may be force - feedback data or other perception data. In short, data has become an important bottleneck for robots to enter households.

Of course, some people also think that data is only one aspect. As Wang Xingxing said, "The architecture of the model itself is not good enough," and "even if there is a lot of good data for training, it cannot be used."

In addition to the fact that the technology itself has not reached the household - use threshold, safety still cannot be guaranteed. The guarantee here not only refers to whether the robot can ensure no errors as mentioned above, but also to the social problem of how to protect the rights and interests of users after an error occurs. It involves the responsibility attribution in robot safety accidents, and the relevant laws have not been introduced yet. However, this part can refer to the grading standards and responsibility division of autonomous driving.

In addition, there are two inexpressible obstacles to user acceptance that need to be emphasized. One is the usage skills of ordinary users, and the other is the uncanny valley effect.

On the one hand, in Elon Musk's view, the complexity of robots is even greater than that of cars, and using robots is a process that requires training. Just like drones can be purchased or "accessed," but there is a significant difference in skills between ordinary pilots and trained, licensed pilots or geeks. At present, the "usage gap" that once occurred with computers and smartphones has emerged among different groups of people using the same model of robots. If such products are extended to the home scenario, this gap needs to be bridged. In addition, if we compare robots to cars, the current situation is similar to the fact that it is easy to sell a car, but using the car requires training, and users may need to obtain a "robot driver's license." At present, this field is still blank. Considering the safe use of robots, this should not be a problem for users themselves. Can enterprises develop robots that are easier to use? Should regulations require grading and standardization for the use of robots? On the other hand, the uncanny valley effect cannot be avoided, and it is difficult to grasp this delicate balance. In the process of robots entering households, they are bound to intervene in a more approachable way. If robots always maintain a difference from humans and are restrained, there will always be a "barrier"; while products like those of the Polish humanoid company Clone Robotics, which pursue extreme bionics and are very similar to those in "Westworld," will eventually make people fall into the panic of "excessive human - likeness." This paradox always exists.

In short, the robots in the home scenario like Figure 03 that we see now are more like students who have "made some progress" but "have not graduated." In terms of learning to integrate into human life and do housework, in many aspects such as models and data, safety and norms, usage ability and psychological construction, including the cost and substitution problems not mentioned in this article, they have not crossed the user acceptance threshold. The technological illusion has obscured the reality of the scenario. Perhaps the current "inflection - point wave" is more like a signal of industrial acceleration rather than the end for enterprises. For household robots to truly enter ordinary families, they not only need to break through the technical barriers of data, algorithms, and safety, but also need to cross the gap from "machine thinking" to "humanistic thinking."

At this moment, the track has more enthusiasm than precipitation. After all, the scenario that humanoid robots are about to enter is the final stage.

This article is from the WeChat public account "Embodied Learning Club," author: Peng Kunfang. It is published by 36Kr with authorization.

该文观点仅代表作者本人，36氪平台仅提供信息存储空间服务。

Has the inflection point for humanoid robots in household scenarios finally arrived?

What does it take to become a robot butler?

It's still very early for robots to enter households