From the Sports Arena to the Market: How Youliqi Uses "Two Golds and One Silver" to Illustrate the Algorithmic Breakthroughs and Commercialization of Humanoid Robots

In most robot competitions, flashy actions tend to become the focus of attention: who can run faster, walk more steadily, or jump higher. However, in the just - concluded 2025 World Humanoid Robot Games, Youliqi's Wanda series of humanoid robots completed a set of seemingly un - "spectacular" procedures: opening the door to enter the room, identifying garbage, picking up waste paper, wiping the floor, and finally cleaning up a messy guest room.

This set of actions involved no leaping, no dancing, and no high - difficulty waving of mechanical arms. Nevertheless, it won Youliqi two gold medals in "Hotel Cleaning Service" and "Hotel Reception Service", as well as a silver medal. It also placed Youliqi in the top three on the medal table and second on the gold medal table.

This was one of the events with the highest requirements for technical stability in this competition. The cleaning task involved handling objects of various materials, shapes, and uncertain positions, requiring the robot to autonomously complete the entire chain of perception, decision - making, and execution. The judges adopted a 360 - degree real - time scoring mechanism, with no manual intervention throughout the process. Youliqi's performance in terms of action coherence, task completion rate, and environmental adaptability was considered that of a stable competitor on the field.

Judging from the competition settings, there was almost no difference between this kind of task and real - world application scenarios. In hotels, property management companies, and elderly care communities, the tasks that robots need to perform are essentially the same: continuous completion, real - time processing, and error - free operation. Youliqi's choices in the competition, to some extent, also reflect its basic judgment on product design and business strategies.

If Unitree's stage is more about "what cool things can be done", then Youliqi's answer leans more towards "what real - world problems can be solved".

Capability System for Handling Highly Complex Scenarios

The hotel cleaning task in the games simulated a highly challenging unstructured environment in reality. The room was filled with various items such as deformed paper balls, empty beverage bottles, tablecloths, and chairs. The robot needed to autonomously complete multiple action steps including identification, path planning, picking up, sorting, cleaning, and disposal starting from opening the door.

Youliqi's Wanda series of general - purpose humanoid robots are equipped with self - developed high - torque - density joint modules and an 8 - axis mechanical arm system, having high dynamic response and trajectory control capabilities. On the control side, Wanda achieved stable movement and task switching throughout the process. The actions were not fast but the rhythm was compact.

Its core support lies in a self - built set of "three - piece algorithm suite":

• UniFlex

It is an imitation learning framework for manipulation tasks, emphasizing rapid generalization with a very small amount of demonstration, suitable for high - contact tasks such as wiping and storage;

• UniTouch

It is a large - model system based on the fusion of vision and touch, used to enhance the robot's understanding of materials and contact feedback, making the manipulation behavior closer to the human way of handling;

• UniCortex

It is a task model for long - sequence planning, used to support path adjustment and priority management in multi - round task execution.

This system does not pursue the complexity of actions but emphasizes whether it can "complete continuously without interruption" in real applications. For scenarios like hotel cleaning with a large variety of objects, strong environmental interference, and a long task chain, stability is more important than dexterity.

What Matters More Than Running Fast Is "Being Ready for Work"

What Youliqi presented on the competition field was a complete and reusable service process. From opening the door to enter the room, identifying the room environment, to completing cleaning, storage, and disposal, this series of actions were performed in a way similar to the working rhythm in a real hotel and required almost no additional customization.

The completeness of this kind of process determines whether it can directly enter the customer's scenario. For most service industries, the prerequisite for deploying robots is not how rich the functions are, but whether they can be smoothly integrated under the existing conditions. Youliqi has made as many preset preparations as possible in this regard, including route planning, accuracy of item identification, and connection control between task links.

Currently, Wanda has entered the stage of small - scale delivery. The company revealed that it has signed orders worth tens of millions of yuan with several hotel groups, property management companies, and elderly care communities. The supporting production capacity system has been established, with an annual mass - production capacity of thousands of units.

In selecting application directions, Youliqi has avoided common short - task - chain scenarios such as handling and reception, and instead chosen the more complex but more customer - demand - concentrated cleaning service. This choice has raised the threshold, but once it is successful, the adaptability of its model, algorithm, and hardware combination can be more easily extended to other task chains.

Currently, the company has extended this idea to service - type scenarios such as elderly care assistance, family companionship, and educational demonstrations. Relying on a unified ontology platform and a modular algorithm architecture, it can cover different demand boundaries through task switching. This product strategy of "taking the task chain as a unit" is building a service robot platform with evolutionary capabilities.

In early 2025, Wanda was listed on JD.com, becoming one of the few humanoid robots to enter the public retail channel in the form of a standard product. This was not just a distribution through channels but an attempt by Youliqi to contact early users, collect market feedback, and establish a reference basis for subsequent product rhythms and function combinations.

The parallel layout of B + C has also given Youliqi two observation entry points: on one hand, the task completion rate and deployment process of service - industry customers; on the other hand, the understanding and acceptance of the product form by end - users. For the company, whether the robots can be sold is not that important for the time being. The key is whether they have started to be used and evaluated as "tools".

Conclusion

Youliqi did not demonstrate high - difficulty actions in the competition, nor did it emphasize the "futuristic" vision of human - robot collaboration. What it did was an almost "clumsy" verification: to verify whether the current algorithms, hardware, and control systems already have the ability to work in specific scenarios.

The boundaries of this set of capabilities are clear and the goals are well - defined, ultimately reflected in the company's basic rhythm of "product delivery and business implementation". Different from robot companies that are more inclined to exploration routes, Youliqi's bet is to start from one scenario and gradually expand to other fields through the combination of a general ontology and modular algorithms.

When embodied intelligence moves from experimental engineering to practical applications, the teams that take the lead are often those that complete the delivery closed - loop first.