The robot that truly does household chores is here. Zi Wei Wang Qian: In the next 3 to 5 years, home robots will become a standard part of everyday life.

Ask AI · How can variable robots break through technical bottlenecks in home scenarios?

Text | Yuan Xiaoli

Editor | Liu Peng

By 2026, the robotics industry is becoming increasingly lively.

From martial arts performances on the Spring Festival Gala stage to the globally watched backflip, the robotics industry continues to heat up in technological breakthroughs, capital enthusiasm, and public opinion. But for ordinary families, the most concerned question remains: when will robots that can truly enter homes and practically help with chores arrive?

This question seems simple but is actually extremely difficult to answer. Because among all practical scenarios, the home is almost universally recognized as the most complex, hardest to standardize, and most difficult to scale and replicate.

Compared to standardized environments like factories, warehouses, and logistics, the challenge in homes lies in the fact that there is almost no truly definitive standard answer. Different layouts, storage habits, and lifestyles mean that robots face an open environment full of randomness and complex problems. They must not only recognize objects and understand commands but also judge which items can be touched, which cannot, when to move, when to stop, and ensure safety in environments shared with the elderly, children, and even pets.

It is precisely because home scenarios are so complex that, despite global efforts by robotics companies to develop general intelligence and humanoid robots, few enterprises focus on “entering homes and taking on household chores” as their main battlefield.

From overseas, Tesla’s Optimus is still primarily being validated in factory settings, with home applications seen as a long-term vision; Figure’s commercialization focus remains on manufacturing and enterprise scenarios; even the 1X, which explicitly targets home scenarios, emphasizes safety and home assistance in its NEO series, and plans early testing in homes, but overall remains in pilot validation and prototype iteration stages, still far from stable, widespread service.

The situation is similar domestically. Whether it’s the home companionship-oriented Zhiyuan Lingxi X2 or the exploration of smart home ecosystems in collaboration with Haier and Leju, the overall stage is still at demonstration, training, joint debugging, mass production preparation, or small-scale pilot phases.

In other words, the robots that the public currently sees—such as those folding clothes or wiping tables—are only shown in demo videos. Truly entering homes and reliably performing chores in real service processes remains a blank space.

Because of this, despite the unprecedented enthusiasm in the robotics industry, until now, almost no company worldwide has entered real homes and participated in actual household labor.

And now, this blank space is being filled for the first time.

In March this year, China’s first—and globally the first—“robot cleaning staff” to truly enter homes officially started work in Shenzhen. Users can book via the 58.com app and experience home cleaning services performed collaboratively by cleaning ladies and embodied intelligent robots.

This means that embodied intelligent robots doing chores are beginning to move from video demonstrations and conceptual validation into real homes and real services.

“This time, the robot cleaner is the first in the world to truly enter homes and genuinely help humans with chores,” said Wang Qian, founder of Variable Robots, to Tencent Finance. In his view, “this is also the first time in human history. Embodied intelligence technology has first stepped out of the lab and truly entered the realm of public service.”

01 The robot cleaner officially enters home services

From the publicly available videos, this “robot cleaner” not only covers floor cleaning but also participates in tidying clutter, cleaning dining tables, organizing sofas and beds, and can even handle tasks like packing trash and taking it away. It can assist in pet care as well. It is not a single-point cleaning device or a fixed-action automation tool but has been integrated into a complete home cleaning process, beginning to undertake some real household chores.

Currently, this service adopts a “cleaning lady + robot” collaborative mode rather than the robot working alone. Wang Qian sees this not as a compromise but as a more practical current approach. “Our core consideration isn’t whether humans will be replaced, but that human labor is simply insufficient,” he said. Compared to having robots work independently, the human-robot collaboration mode is more acceptable to the market and better fits the current pace of domestic service work. More importantly, this mode allows robots to quickly enter real homes, collect long-tail issues through repeated services, gather high-quality data, and continuously iterate.

“This is truly groundbreaking; we couldn’t have predicted what situations robots would encounter once they entered real homes,” Wang Qian admitted. “Home service robots are still far from mature. Because they are immature, they need to be tested and refined in real homes, collecting problems and data through practice, step by step moving toward maturity.”

For the robotics industry, the greatest danger isn’t imperfect capabilities but remaining stuck in labs and unable to enter the real world.

02 Why Variable

While global giants push forward in industrial scenarios, why can a Chinese startup founded just over two years ago be the first to crack the home scenario? It all starts with a strategic choice made from the beginning.

Founded in December 2023, this company is one of the earliest in China to adopt a fully end-to-end approach to developing general embodied intelligence large models, focusing on the physical world’s general large models and home service robots. Since its founding, the company has raised over 2 billion yuan, recently completing a 1 billion yuan A++ round of funding, backed by top institutions like ByteDance, Sequoia China, and Shenzhen Capital Group.

More critically, the company’s initial bet was on the home scenario itself, especially on the “brain” of the robot, rather than just hardware that moves.

“If we can perfect the home scenario, theoretically, we can adapt to all scenarios,” Wang Qian analogized with large language models: it’s not about gradually becoming smarter but about explosively gaining general capabilities first, then gradually applying them across various scenarios. “When robots learn to handle many different tasks, they will learn the commonalities behind these tasks—logic, thinking methods, physical laws. And home scenarios contain the most diverse and complex tasks. Training robots in the most complex environments naturally helps them learn core abilities and become smarter.”

The logic behind this is that if a robot can operate stably in the home environment—where problems are numerous, tasks are complex, and human-environment interactions are highly unpredictable—then the capabilities it develops are most likely transferable to other public service scenarios like elderly care, property management, and catering.

What determines whether this is feasible isn’t the robot’s “body,” but its “brain.”

Wang Qian believes that the core pain point in the embodied intelligence industry today is clear: hardware is becoming increasingly mature, but the intelligence “brain” of robots still lags far behind market needs and public expectations. On the surface, grasping cups, wiping tables, and tidying clutter seem to be issues of robotic arms and dexterous hands, but in reality, they involve highly integrated processes such as visual perception, language understanding, task decomposition, motion generation, environment feedback, and real-time adjustment.

Therefore, Variable’s core technical route is anchored in native multimodal foundational models for embodied intelligence.

Most VLA models typically treat actions as the only output modality. Variable is building a true Omni architecture, i.e., “native multimodal,” with the key breakthrough being the integration of multiple modalities from the ground up, achieving “multimodal input, multimodal output.”

“Multimodal input” means the model can directly and simultaneously receive visual, language instructions, and proprioceptive inputs; the most critical “multimodal output” is not just physical action commands like traditional VLA models but also predictions across multiple modalities such as actions, language, and vision.

For home robots, the challenge is never whether the robotic arm can lift objects but whether it can understand a cluttered living room, comprehend a vague command, and pre-visualize the physical outcomes before acting. Based on the native Omni model, the robot not only reacts to current visual scenes but also possesses “imagination” and “prediction” abilities similar to humans—before acting, its “brain” predicts the next visual scene (e.g., if it reaches for an object, will nearby items be knocked over?), generates interactive language, and outputs precise physical actions.

Industry progress shows that end-to-end learning has become a consensus; companies like Figure AI, 1X, and Variable have chosen this path. But what sets Variable apart is its deep integration of VLA with World Models, first applying this combined capability in real home service scenarios rather than just in lab demonstrations.

03 Beyond Home, Elderly Care Has the Most Potential

Currently, the robot cleaner is only available for limited experience in Shenzhen. The team’s immediate goal is to refine the product, collect feedback data from real homes, train the robot’s “brain” model, and iteratively improve performance. Wang Qian revealed that they plan to promote nationwide soon. Although full-scale adoption still requires further polishing, he firmly believes that in 3–5 years, household robots could achieve large-scale popularity and become a standard part of daily life.

Looking ahead, besides home cleaning, Wang Qian is more optimistic about the application of embodied intelligence in elderly care. “Elderly care is a highly valuable and large-gap core scenario.” Whether for home-based or institutional elderly care, although there are dedicated caregivers, public satisfaction with service quality remains low. Wang Qian predicts, “Robots will become one of the key technologies supporting the sustainable development of the elderly care industry, and this trend is inevitable.”

Additionally, service scenarios like catering and property management, which face non-standardized service challenges, are also potential key applications for robots.

Reflecting on his entrepreneurial journey, Wang Qian admits that the hardest moments weren’t technical breakthroughs but the early skepticism about the “home general robot” track. But his unwavering belief in the “general intelligence first” approach kept Variable moving forward to today.

He is full of hope for this robot cleaner and the entire industry: “Today marks the beginning of a new era. With continuous technological iteration, embodied intelligence will ultimately transform the landscape of life services, truly benefiting every household.”

For the entire industry, the significance of this robot cleaner may not lie in its current perfection but in its first step of allowing robots to genuinely serve as helpers, entering ordinary people’s homes and taking on some real labor.

In recent years, the robotics industry has lacked nothing but performance; what it lacked most was entry into daily life.

And now, that is beginning to change.