May is brimming with captivating events to anticipate.

First and foremost, the maiden voyage and recovery of the Changshi B spacecraft stand out, marking an unprecedented challenge in maritime network recovery. We've already delved into the significance of this event in our previous article, "The Maiden Flight of Changshi B Looms: Heading to the South China Sea to Retrieve the Rocket." Another pivotal event for an industry on the cusp of transformation is the debut of household robots in real home settings on May 25th, providing genuine services.

The protagonist of this groundbreaking move is a robotics company named "Independent Variable." Despite its low public profile, it enjoys the backing of four industry behemoths: ByteDance, Alibaba, Meituan, and Xiaomi. These investors, rarely seen together, have surprisingly converged on this occasion.

As Laozi once said, "The greatest sound is barely audible; the greatest form has no shape." While others vie for the spotlight, the quietest contender has amassed the most valuable assets.

01 Body-Centric Approach: Accumulating Data Through Movement

In our previous article, "Stumbling and Then Learning to Stagger: The Critical Point of Embodied AI," we examined the landscape of embodied AI and outlined the three fundamental logics of the Body-Centric Approach: the body as the foundation, motion control as a prerequisite for embodied AI, and the accumulation of data through movement to fuel the brain's learning. China's supply chain presents a formidable barrier.

Yet, do companies pursuing the Body-Centric Approach differ?

Unitree Technology, the most recognizable to the general public, follows a strategy of "ultimate hardware + cost-cutting."

Its core strength lies in the in-house development of over 95% of its critical hardware components, including motors, reducers, sensors, encoders, and batteries. With 60% of parts being interchangeable between humanoid robots and robot dogs, the cost advantages are substantial: the self-developed M107 joint motor costs only half of imported alternatives, and the average machine price has plummeted from 260,000 yuan to 167,000 yuan—a price point unattainable by other manufacturers' bill of materials (BOM) costs.

Zhiyuan Robotics has struck a balance between motion capabilities and intelligence that factories can embrace.

The Expedition A3's deployment in the NIO factory validated that humanoid robots can perform real tasks on automotive production lines without preset programs. It not only showcases "the body's ability to work" but also tests "whether the brain can comprehend the factory's task intentions." With significant shipment volumes, it is projected to exceed 5,100 units by 2025.

UBTECH has led the way in commercialization. The Walker S2 delivered 1,079 units throughout the year, making it the only full-sized humanoid robot globally to achieve thousand-unit-level delivery. It has entered BYD and Airbus factories and has become the "first humanoid robot stock" on the Hong Kong Stock Exchange.

Although these three companies all adhere to the Body-Centric Approach, their paths diverge. Nevertheless, their priority is to stabilize the body before addressing other aspects.

It is crucial to emphasize that the Body-Centric Approach we discuss is merely a focus or priority. The approach is well aware that the brain represents the next frontier. Therefore, they have never neglected the brain; they are simply prioritizing the body first, akin to laying the foundation before constructing the superstructure.

While Unitree advances factory deployments, it also establishes mechanisms for accumulating motion data; Zhiyuan has launched the "Lingxi" platform, attempting to bridge intelligence and motion; UBTECH is exploring possibilities in education and services beyond industrial scenarios.

They are not oblivious to the brain's importance; they have simply chosen to lay the foundation before constructing the building.

The body is the entry ticket; the brain is the main event.

Knowing when to pause leads to determination; determination leads to tranquility; tranquility leads to security; security leads to contemplation; contemplation leads to achievement. The Body-Centric Approach focuses on "pausing" first, ensuring the robot stands firm and moves; the Brain-Centric Approach focuses on "contemplation," enabling it to understand why it moves. The order varies, but neither step can be bypassed.

02 Brain-Centric Approach: Pioneering the Engine

Since the Body-Centric Approach is also evolving towards the brain, why didn't the four giants directly invest in them and instead chose a company without mass-produced robots?

Wang Qian, the founder of Independent Variable, has a frequently cited maxim: "In the long run, hardware has no barriers in China; only the brain does."

This statement is often misconstrued as opposition to the Body-Centric Approach. However, Wang Qian has never denied the body's importance; rather, he emphasizes that China's supply chain capabilities will continue to lower the hardware threshold. The body advantages developed over several years may be eroded by the supply chain in half a year or even less. Yet, the barriers to generalization capabilities, such as enabling a robot to understand tasks, adapt to new environments, and not rely on preset programs, will not disappear due to supply chain maturity.

Therefore, Independent Variable has chosen a path that appears slower and more enigmatic. It does not participate in performance shows or marathons but focuses on building the "foundational model of the physical world," iterating from the VLA architecture to the WUM architecture.

The four giants may each have their own considerations.

What attracts Meituan is the last mile of delivery and instant retail. If robots are used by competitors to intercept, Meituan's territory will be compromised. A robot capable of completing the "last ten meters" is more valuable than any food delivery cabinet. Meituan is already developing delivery robots, but that is a "Body-Centric Approach" solution. Betting on Independent Variable is preparing for a "Brain-Centric Approach" solution in advance.

Alibaba may be more interested in the computing entry point of "cloud + embodied." Behind every action in the physical world lies computation, and Alibaba is one of the companies most adept at selling computing power. If the "operating system" of the robot era is not Alibaba's, all its layouts in the AI era will have a gap.

ByteDance is contemplating the next data flywheel beyond content production. Douyin's data flywheel is short videos, but the real physical interaction data generated by robots is a more fundamental production factor than short videos. These data are not on the internet and can only be generated by robots in the real world. Whoever controls these data controls the training resources for the next generation of AI.

Xiaomi has always aimed for the ultimate form of the smart home ecosystem. A robot that can navigate the house is obviously closer to the "center" than any speaker, TV, or phone. Xiaomi has developed CyberOne, but it is also clear that if another robot's brain is more useful, CyberOne's body could easily run someone else's system.

Therefore, what all four companies are focusing on is the monopoly effect of foundational models.

In the era of LLMs (Large Language Models), ChatGPT has already demonstrated this scenario: whoever first creates a foundational model defines the rules of the application layer, and later entrants can only build on top of it. If the physical world follows the same path, then "becoming the operating system of the physical world" is worth betting on by all giants at the earliest stage.

In embodied AI, investment is also a continuation of strategy. Every bet placed by the four giants is not just on Independent Variable but on a future they cannot afford to miss.

Independent Variable has raised over 3 billion yuan in cumulative funding, with a valuation exceeding 10 billion yuan.

This money buys possibility, not certainty.

03 Home Scenarios: The "Ultimate Examination" for Embodied AI

Whether the Brain-Centric Approach can succeed cannot be determined solely by funding and architectural design. Therefore, Independent Variable has chosen one of the most challenging places in the industry to validate itself: home scenarios.

We have already introduced the difficulty levels of different scenarios in the previous article. Factory scenarios offer the highest certainty, logistics scenarios are semi-open, service scenarios require human-robot interaction, and home scenarios are entirely non-standard.

The error tolerance in homes is the lowest. In a factory, if a robot mishandles a part, it can try again; but if it spills a pot of hot soup at home, it's a safety accident, and user trust is lost. Therefore, most companies directly avoid home scenarios, accumulating capabilities in simpler scenarios before considering homes.

But Independent Variable did not. It chose to partner with "58 Daojia" to test the robot's brain in real homes for cleaning tasks.

"The abyss is staring at you." Home scenarios are the abyss of this industry, which everyone avoids, but Independent Variable has chosen to walk through it.

Cleaning tasks are a natural training ground for "fuzzy instructions." For example, if you say, "Clean the kitchen countertop," the robot needs to understand: Where is the countertop? Which items need to be moved, and which can be left? Should it use a wet cloth or a dry cloth? These complex decision chains never appear in factories.

If Independent Variable can make even a small step forward in home scenarios, it proves that "my model can handle non-standard environments," meaning it can be transferred to countless non-standard scenarios, which is significant.

However, the core assumption of the Brain-Centric Approach is that "generalization capabilities can be transferred from simulation and semi-standard scenarios to real homes." If this assumption does not hold in home scenarios, the valuation logic of the entire Brain-Centric Approach may be rewritten.

On May 25th, Independent Variable will submit part of this answer sheet.

This is not just Independent Variable's report card. It is a mid-term validation of the Brain-Centric Approach. If home scenarios can be navigated even a little, the industry's confidence in the "brain-first" path will rise a notch. If not, the "lay the foundation first" logic of the Body-Centric Approach will gain more support.

04 Convergence: The Two Paths Will Eventually Meet

The Body-Centric Approach is laying the foundation; the Brain-Centric Approach is building the engine.

This is not a dispute over paths but a disagreement over construction order. The Body-Centric Approach chooses to make the robot move first and then gradually become smarter; the Brain-Centric Approach chooses to make the robot smarter first and then find a suitable body to carry it.

Both choices are reasonable and come with their own risks.

If the supply chain lowers the hardware threshold sufficiently, the motion control advantages accumulated over three years may be compressed to a few months. This is the anxiety of the Body-Centric Approach. The risk for the Brain-Centric Approach lies in the fact that if generalization capabilities do not reach a usable level, those hefty financings will turn into anxiety on the balance sheet.

But in the end, both paths lead to the same place: a robot capable of autonomous action in the real world.

The leaders of the Body-Centric Approach are aware of this, so they are already moving towards the brain; the Brain-Centric Approach is also aware of this, so they will eventually face the test of physical implementation. The two paths are not parallel lines; they are slowly converging.

All rivers flow into the sea, and we do not know when they will stop and overflow. However, whoever arrives first gets to set the rules.

Looking forward to May 25th, 2026.