After six months of starting a business and raising 300 million yuan in financing, Guo Renjie said he wants to create "consumer products" for robots.

Text by | Deng Yongyi

Edited by | Su Jianxun

Half a year ago, we had a long conversation with Guo Renjie, the CEO of LeXiang Technology. At that time, he had just left his role as the Executive President of Dreame's China region and founded LeXiang Technology. Even though the products were not yet in shape, the company had already secured financing from a group of star - capital firms including IDG, Matrix Partners, ZhenFund, and Sequoia Capital.

"My only advantage is being hard - working," Guo Renjie told us.

At the end of May, we met Guo Renjie again. This time, LeXiang Technology completed another round of over 100 million yuan in Series Angel + financing. Jinqiu Fund led the investment, while existing shareholders Matrix Partners, Oasis Capital, and Monolith continued to increase their stakes, and Light Source Capital participated in the investment as the exclusive financial advisor.

This is the second round of financing completed by LeXiang Technology within three months, and the total amount of the angel round financing is nearly 300 million yuan.

More importantly, when obtaining the new round of financing, the company officially announced its product direction: to develop general - purpose small embodied intelligent robots for households.

Guo Renjie told 36Kr that after spending half a year in this field and getting familiar with the existing hardware and software technologies, he had a big question: Why, despite the abundance of good technologies, are there only a few sporadic tracks like AI toys that have achieved Product - Market Fit (PMF)?

"Many good technologies have not been applied to AI hardware," he believes. The slow progress of achieving PMF is largely due to the fact that the technologies have not been widely adopted. "For example, the performance of motors, algorithms, and models is already very strong, and the perception and understanding of the environment have reached a high level, but many of these technologies are only used in large humanoid robots."

This is indeed the case in the market. Large robots are mostly used in laboratories, exhibition halls, etc. Most of them are priced from 100,000 to several hundred thousand yuan per unit, making them difficult for ordinary users to access.

Guo Renjie hopes to "downscale" the technology and bring it into every household. He has personally witnessed such an example - the "downscaling" of lidar technology in autonomous driving led to the explosion of the robotic vacuum cleaner market.

During this period, the field of embodied intelligence has remained bustling. Many humanoid robot companies, including Unitree, Leju, and Songyan Power, have said that "orders skyrocketed in the first quarter of 2025."

However, behind the enthusiasm, there are still many challenges and crossroads on the path from productization to true large - scale implementation, and the commercial scenarios are not yet clear. That's why Zhu Xiaohu, a partner at GSR Ventures, caused an unexpected stir when he said, "We are withdrawing from humanoid robot companies in batches."

In a sense, this also shows that this field is still in its very early stage.

Having experienced the growth of the robotic vacuum cleaner market from scratch, Guo Renjie has a more practical perspective.

During the interview, Guo Renjie mentioned the word "consumer product" more than twenty times. He told 36Kr frankly that no matter how grand his plans are, whether the robots are humanoid or multi - legged, they are essentially the most suitable product form and solution at present. "What remains unchanged is that we only want to make consumer products in the robot track."

The "small general - purpose robot" for household scenarios is just the first step.

After the large - scale model learns to "think," it will give birth to the AI hardware version of Manus

When seeking the product direction, Guo Renjie and his team first set their sights on the household scenario.

Even today, the household scenario is still one of the few scenarios where large - scale consumer products have emerged. Both the profit and scale levels have proven that this is a stable and consumption - driven scenario.

Secondly, Guo Renjie and his team noticed that the form of small robots can already meet several core needs and functions in the household scenario: anthropomorphism, mobility, and integration.

LeXiang Technology's first product was developed based on these three criteria.

Currently, the R & D of LeXiang Technology's first - batch product platforms has been completed. One is a household robot called Z - Bot, about 50 centimeters tall, capable of moving throughout the house in an indoor scenario, with an anthropomorphic appearance. The other is an intelligent companion robot called W - Bot, which can cover both outdoor and indoor scenarios.

△Small embodied intelligent robot Z - Bot

Even so, Guo Renjie and his team still went through a process from "looking for nails with a hammer" to re - reflection.

For example, the anthropomorphic form corresponds to a natural product intuition. "At first, we just had an idea that we should make a human - like robot for the household that can move throughout the house, and then we looked for scenarios with the product in hand," Guo Renjie said.

But after a month, Guo Renjie found that the entire organization was more confused about the product functions and scenarios. For example, should they add a robotic arm? Should it have two legs or tracks?

Finally, he had to quickly stop the entire team and start from the actual pain points in the household scenario, looking at what the most practical and solvable pain points in family life were.

The method was very simple. Guo Renjie and his team started from their own lives, listing the pain points in family life one by one and conducting a large amount of user research. "For example, I thought about how I communicate with my family members on a daily basis and considered the safety issues of my grandfather living alone, and how a mobile robot could help him." Guo Renjie even asked employees to bring their pets to work just to better refine the product requirements.

Ultimately, the core idea that the team figured out was: First, the robot should be able to move throughout the house. On this basis, it should first meet enough long - tail household needs.

A typical example is the care - giving scenario for the elderly and children. For many elderly people, installing a fixed physical camera first faces the psychological resistance of the elderly, as well as the difficulty that the camera and sensors cannot fully cover the area. However, if there is a mobile robot that can cruise within a fixed time and range, it can provide reliable care - giving.

In addition to care - giving, there are also security and other long - tail needs. Guo Renjie calls this the "mobile version of hardware Manus."

Once the robot can accompany and cruise, it can meet the information - acquisition needs in different scenarios. "For example, when watching TV, you may need an agent that can interact with you via voice. You want to control all the hardware in the house with voice at any time or know about the security cruise inside and outside the house. A mobile terminal can significantly improve the interaction experience."

Many household users may think that these needs are not frequent enough or not essential when paying for them individually. However, if there is a product that integrates enough functions and can effectively solve the pain points in mobile scenarios, it may be able to establish a commercial closed - loop.

Not just "doing somersaults," but being an emotional household terminal

The release of DeepSeek R1 was another important turning point for LeXiang Technology when refining its product direction.

"This is the luckiest thing that happened to me in 2025," Guo Renjie said.

The release of the previous two products represents a huge improvement in model inference capabilities. If applied to the "brain" of a robot, the robot can then learn "how to think" and be able to plan and complete more tasks autonomously. This is of great significance for the robot track, which is still in the early stage of the technology curve.

When defining the product, the LeXiang Technology team also planned a version for geeks, college students, and researchers: a highly open small - scale robot development platform. The original intention was to allow users to highly customize the product in the early stage.

For example, after the robot is adapted to different sensors and software functions, users can develop based on their different household environments, and family members can enjoy corresponding customized functions.

"After adding reinforcement learning, it means that the product will have strong future extensibility, which is the most important difference from traditional robots," Guo Renjie explained.

To a large extent, this is also highly related to the team's pursuit of "anthropomorphism."

In the past, if users wanted to teach a robot to perform a specific action, such as playing with a cat with a cat - teaser, such actions basically had to be programmed in advance by engineers, with a fixed number of actions.

However, after reinforcement learning became the mainstream approach, users can more easily define a set of robot actions, and the robot can even autonomously plan and learn the actions.

This will give users a more personal and anthropomorphic experience - users interact with and educate the robot themselves. Conversely, as long as the robot stays in the user's household scenario long enough and understands more context, it will provide more personalized interaction feedback to the user.

During the interview, Guo Renjie repeatedly emphasized that their pursuit of "anthropomorphism" is not a physical one, like a large humanoid robot with two hands and two feet that can fold clothes and cook. It is more of an "anthropomorphism" in an emotional sense, providing a sense of family companionship.

This is a path of misaligned competition compared with many startups focusing on large humanoid robots.

In the field of embodied intelligence, this state of misaligned competition also exists because a consensus has not been reached, or the consensus is too far away.

LeXiang does not focus on "competing" with large humanoid robots in a specific task, such as folding quilts or doing laundry and cooking. Guo Renjie believes that in scenarios with clear market benchmarks in the real world, even if a robot can achieve 95% of human - level performance, it is still difficult to compete with humans. However, once this direction breaks through the boundary of commercial implementation, it can achieve exponential growth.

LeXiang prefers the more realistic evolution path, that is, following the goal of making the robot "more anthropomorphic and more emotional" and looking for business opportunities in each iteration.

He gave an example - LeXiang's first product may not even have a voice function. Even relying only on the robot's body language and screen display, it can express a considerable degree of emotion. Using the form of companionship, it first enters the user's life scenario and then gradually adds functional attributes.

△W - Bot tracked companion robot

To this end, LeXiang Technology has been developing a local - based emotional model and a consciousness - bionic model by collaborating with large - scale model partners to jointly develop an emotional large - scale model.

Establishing trust and a sense of companionship is the first step. After that, there will be more "context" input - family members are willing to interact with the machine, which makes "autonomous learning" of the machine possible. And the development platform planned by the team leaves more room for functional expansion of the product.

This comes from years of experience in iterative development of consumer products. "Simply put, we must make money while nurturing our dreams," he summarized.

Currently, the R & D of the three product lines planned by LeXiang has been initially completed. According to his schedule, the first product should be launched by the end of 2025.

The company completed the development of the product demo within half a year of its establishment. Both the product and the organization are still in a state of rapid change. "If we think something is right today, we'll do it. But we may also be wrong, so we need to keep iterating and making mistakes at a high speed," Guo Renjie said. The key is how to maintain the ability to iterate and upgrade at a high speed - only then is it possible to gradually approach the vision of making the robot truly become a family member.

Welcome to communicate

Welcome to follow

This article is from the WeChat official account "Intelligent Emergence," author: Deng Yongyi. Republished by 36Kr with permission.