Gasgoo Munich- With the 2026 BAAI and WAIC conferences in the rearview mirror, humanoid robots stole the show without question. From somersaults and serving drinks to twisting caps, brewing tea, and tossing salads, the demonstrations were relentless. The narrative that robots are about to enter "every household" was ubiquitous, creating the illusion that a robot butler might knock on your door at any moment.
Yet beyond the buzz of the exhibition floor, the frontline industry view is markedly sober. An insider at an embodied intelligence firm focused on industrial and commercial service robots told Gasgoo that the home is "relatively late, perhaps even the final core scenario" for robotics. Wang Feng, vice president of JAKA Robotics, was equally clear: the industry's core priority remains landing in industrial scenarios. Liu Jinyu, co-founder and COO of Julingshi Panse (BrainStone), described the home sector as the "ultimate horizon" for embodied intelligence—but one that requires completing a long, gradual commercialization path first.
While audiences marvel at these "acrobatic performances," frontline practitioners have reached a consensus: it will take at least five years before humanoid robots enter homes in any significant numbers—or reach a reasonably functional state in a domestic setting.
Image source: Gasgoo
The Home: The Ultimate, Hardest Test for Robots
Why does deployment in the home rank last among all scenarios?
The insider offered the most direct explanation: the home is a typical fully unstructured environment, brimming with uncertainty. Scattered remotes on the coffee table, tangled cables on the floor, darting pets, and randomly placed water cups—no two homes share the exact same layout, and no single day is a replica of the last. For a robot, this means no preset paths and no standardized tasks; every operation risks encountering a novel, unforeseen situation.
The level of intelligence required for such an environment far exceeds that of industrial or commercial settings. In the insider's view, discussing entry into homes only becomes realistic when robots are intelligent enough to respond quickly and handle various emergencies with flexibility. Current embodied intelligence is clearly nowhere near that stage.
Liu offered a more intuitive benchmark using an "age analogy": placed in a home environment where generalization is critical, today's embodied brain models possess roughly the intelligence of a two- or three-year-old child. They can handle simple interactions, demonstrate basic learning capabilities, and execute the most elementary command tasks.
This stands in stark contrast to industrial scenarios. The same model, when deployed in factory sorting, transport, or loading tasks, already demonstrates operational capability on par with an adult. The reason is simple: industrial settings are typically semi-structured environments. Floors are flat and organized, temperature and humidity are controlled, lighting is consistent, and workflows are highly standardized—robots rarely need to contend with unexpected variables.
Many are misled by the smooth demonstrations at trade shows, assuming that if a robot can perform complex movements on a booth, entering a home will be a breeze. But fundamentally, exhibitions are carefully curated, structured environments where lighting, flooring, and object placement are repeatedly tuned. Their complexity is orders of magnitude lower than that of a real household.
Beyond environmental complexity, cost and engineering readiness stand as two major barriers blocking the path to home deployment.
Liu admits that both the robot hardware and the embodied large models are still in a process of iteration and evolution, far from fully mature. A robot designed for home service must balance agility, human safety, battery life, and an affordable price point—each is a thorny engineering challenge. The more immediate problem is that robots aimed at ordinary households still require massive investment, making it unlikely that average consumers will pay a premium for a novelty product that "isn't quite useful enough" yet.
The challenge at the data level is equally severe. Large language models can iterate rapidly by scraping internet text, but embodied intelligence requires interaction data from the physical world—data from every movement, every grasp, and every environmental adaptation must be collected by deployed hardware. The highly fragmented nature of home scenarios causes the cost and difficulty of data collection to rise exponentially, making it difficult to quickly achieve scale or establish a data flywheel.
In other words, the home scenario is the ultimate exam for embodied intelligence, testing comprehensive ability across all subjects without a defined scope. The industry, however, is still making up basic coursework in individual subjects—far from the stage of entering the exam hall.
A Gradual Landing: From Industry to Commercial, Then Home
With the home as a long-term ultimate goal, the industry's deployment path has become highly unified: start easy, move hard, advance stepwise. Go from industrial to commercial, gradually refining capabilities, then permeate into home scenarios.
The first stop for deployment is, unquestionably, the industrial scenario.
In the insider's roadmap, industry is the optimal starting point. Semi-structured environments are friendlier to robots, with clear tasks and transparent processes—ideal for honing basic capabilities. Starting with the most fundamental movements like mobilization and grasping, and establishing stability and reliability before gradually expanding to more complex operations, is the most pragmatic growth path.
JAKA Robotics is a deep practitioner of this path. Wang Feng noted that JAKA's core customer base and primary business direction remain rooted in industrial deployment, with over 30,000 robotic arms already operating in real industrial settings worldwide. This massive industrial customer base and vast amount of real-world operational data provide solid, realistic feedback to support new hardware releases and software algorithm iterations.
Image source: JAKA Robotics
Currently, JAKA is engaged in joint R&D related to humanoid robots with dozens of key end-users. Applications in scenarios such as bin handling, automotive parts processing, and motor housing production have been validated and are entering the phase of batch deployment. "We are confident we will be the fastest to successfully implement industrial application scenarios," Wang said.
The value of the industrial scenario extends far beyond commercial monetization. For embodied intelligence, industry is the most efficient training ground: standardized tasks can quickly validate model capabilities, stable operating environments accumulate high-quality physical interaction data, and large-scale deployment creates a data flywheel that drives continuous model iteration. More importantly, the B2B business model generates healthy cash flow to support sustained R&D investment—an advantage the B2C sector currently lacks entirely.
Once the foundation is laid in industrial scenarios, the next stop is public service and commercial settings.
The insider believes that public service scenarios sit between the factory and the home. They feature standardized spaces and processes yet involve significant interaction with people, making them the perfect intermediate step to further refine a robot's interactive capabilities and adaptability to semi-open environments—a crucial link connecting industry to the home.
Image source: ACE Robotics
Liu defines commercial scenarios as the "core foundation for moving into the home." In his view, commercial settings like retail, hospitality, and elder care are essentially "deconstructed subsets" of the home environment: interacting with shelves and SKUs in retail corresponds to fetching items from cabinets and refrigerators at home; hotel room cleaning and sanitation directly map to bedroom and bathroom cleaning tasks; and human-machine interaction and nursing assistance in elder care heavily overlap with core home service needs.
"If we can serve people well in elder care settings, or even assist high-end nurses in completing full nursing routines, then the next step into the home becomes a natural progression," Liu said.
The logic of this "deconstructed training" is essentially to break down the massive, complex scenario of the home into smaller, relatively simple scenarios and conquer them one by one. Every commercial scenario successfully executed adds a piece of the capability puzzle to the home robot. At the same time, commercial scenarios can achieve a closed commercial loop, allowing companies to gain sustained revenue support while refining their technology.
The commercial conversion cycle for B2C is extremely long, and the business loop for home robots simply cannot close at present; relying solely on burning cash won't sustain companies until technology matures. In B2B commercial scenarios, however, enterprises can refine hardware, iterate models, and improve engineering delivery and after-sales systems while simultaneously achieving commercial cash flow. Using commercialization to fuel technological iteration is far more stable than betting entirely on home deployment.
Moving from industry to commercial to home isn't about a lack of industry will to be fast; rather, deployment in the physical world has its own objective rhythm. Skipping the preliminary steps to rush into the home will likely result in "exhibit pieces" that look good but don't work, rather than products that solve real problems.
The 5-Year Window: What Is the Industry Waiting For?
"It will take at least five years to see robots enter households on a relatively large scale to provide butler-style services," the insider predicted. Liu also believes that achieving general intelligence at an adult level "is certainly not a cycle within five years."
During these five years, the industry is not waiting passively; it is concentrating on making up three core areas of deficiency.
First and foremost is the deep refinement of productization and engineering. Liu assesses that the productization and engineering of the entire embodied intelligence sector are still in an iteration and maturation cycle, requiring another 1 to 3 years of deep work.
"Anyone can tell a grand narrative story, but the core is to truly land in the customer's actual scenario, step by step," Liu said. True deployment isn't a one-time success on a booth; it's stable operation in a real scenario for months or years. It's a full-process loop from installation and debugging to operation and maintenance and after-sales, where customers are willing to continue paying.
This involves improving body reliability, lowering hardware costs, enhancing model generalization, and perfecting deployment and maintenance systems. These cannot be achieved overnight through a single algorithm breakthrough; one must dive into real scenarios and solve problems one by one, validating scenarios one by one. Without this process, even the most advanced models are merely castles in the air.
Second, the industry is waiting for the maturation of a universal intelligence foundation—the true realization of "one brain, many forms."
As mentioned in a Frost & Sullivan report, the robotics industry is evolving from "one machine, one brain" to "one brain, many forms." In the past, every robot required separate algorithm development; data wasn't shared, capabilities didn't migrate, and entering a new form meant starting from scratch, resulting in extremely low efficiency. The future direction is to build a unified intelligence foundation that allows robots of different forms to share perception, planning, and decision-making capabilities, achieving capability reuse across form factors.
Pudu Robotics' Physical Agent three-layer architecture, Mech-Mind's "eye-brain-hand" technology system, and Julingshi Panse's cognitive world model are all essentially doing the same thing: building a universal embodied intelligence brain. Only when this brain is mature enough to quickly adapt to different scenarios and robot forms will the cost of migrating to home scenarios drop significantly, providing the technological foundation for home deployment.
Liu noted that a major current challenge is the sheer number of body manufacturers and a lack of unified underlying protocols, making the cost of cross-body adaptation extremely high. Limited cross-body adaptation is feasible, but unlimited, all-category adaptation is neither realistic nor necessary.
Over the next 3 to 5 years, the industry is expected to gradually form mainstream technical standards and adaptation solutions, bringing the deployment cost of a universal brain down to an acceptable range.
Finally, it must be made clear: five years from now, we won't be welcoming a "do-it-all robot butler."
The insider's assessment is that it will take about five years to see robots enter homes on a relatively large scale to provide butler-style services. However, this does not mean robots will be able to handle all household chores. A more realistic scenario is that they will be able to stably complete specific tasks like cleaning, fetching items, and basic companionship, possess decent human-machine interaction capabilities, and handle common emergencies in the home.
Image source: ACE Robotics
Liu also believes that general intelligence at an adult level will be difficult to achieve within five years. The deployment of home robots will inevitably evolve in layers: starting with single-point tasks, then gradually expanding capability boundaries, moving from "doing one thing well" to "doing several things well," and slowly approaching general capabilities. It cannot happen in one step.
In the past few years, the race in embodied intelligence resembled a "physical fitness competition," focused on who had the flashiest robot movements and could complete the most difficult high-difficulty actions. Now, however, the industry has officially entered a "brainpower competition" stage. The contest is about whose model is more general, whose data flywheel spins faster, and who can continuously achieve successful closed-loop deployments in real-world scenarios.
Five years is not long—enough time for the industry to thoroughly penetrate industrial and commercial scenarios and shape the universal intelligence foundation. Yet five years is not short—enough to filter out conceptual bubbles and leave behind the players truly capable of delivering robots to thousands of homes. For consumers, rather than expecting an omnipotent robot butler immediately, it is better to watch robots step into factories, shopping malls, and hotels first. Only when they have mastered these scenarios will the day they truly walk through our front doors arrive on solid ground.
