The World Artificial Intelligence Conference (WAIC), held every mid-summer, has become the measuring tape for the progress of embodied intelligence. This year, the buzz around embodied intelligence is even louder: more than 150 humanoid robots debuted on the same stage, setting a new record.

Walking through WAIC, the most immediate takeaway is that robots can do far more than before. There are robotic "nannies" that fold clothes and polish shoes, "high-level" machines whose hand speed rivals pro gamers, and "all-round entertainers" that can break-dance or spar in the boxing ring. In the embodied-intelligence zone—now in its third year—almost every mainstream vendor is zeroing in on raw capability: from pure showmanship to real work, scenario expansion is in full swing.

In a sense, "show" and "work" also define the two paths in embodied intelligence. The former centers on humanoids and dynamic performance; the latter is form-agnostic, laser-focused on practical deployment. Neither path is inherently superior, but their popularity differs. Early-year humanoid marathons and boxing matches, for example, quickly pulled public attention back to human-shaped robots.

The inevitable memes followed. Online quips abound: "If life feels too easy, buy a 'robot' and let it take care of you." The joke cuts straight to a key pain point—today's humanoids still face clear, stage-specific hurdles.

Asked about these "roasts," ROBOTERA CEO Chen Jianyu told National Business Daily (NBD) on the afternoon of July 27, "Even though future applications for humanoids will be vast, we must be honest: the technology isn’t fully there yet, especially for complex scenarios like the home. Overall, however, the industry's pace is faster than we expected, driven by an influx of talent and resources that has created a genuine ecosystem."

Below are selected excerpts from the interview, lightly edited for clarity.

Chen Jianyu Photo/Provided to NBD

After technology comes consensus on commercialization and scenario-based deployment

ROBOTERA, the only embodied-intelligence company in which Tsinghua University holds equity, currently fields three product lines: 

- L7 targets industrial logistics. 

- XHAND1 is a core component sold to research institutes, robot makers, and industrial clients. 

- Q5 focuses on retail and cultural-tourism services.

To date, ROBOTERA has delivered more than 200 units; over half of the orders come from overseas, and nine of the world’s ten largest tech companies by market cap are customers.

NBD: What stage is embodied intelligence at, and what trends do you see?

Chen: On this wave, the sector has reached broad technical consensus. Previously cutting-edge methods have proved effective and been widely adopted. The next step is consensus on scenarios. Everyone agrees robots can enter factories, but a factory is enormous. Which tasks and which business models fit best? There’s no consensus yet.

Two trends follow. First, we'll see exploration—then partial consensus—around commercial and scenario-based deployment. Because these are general-purpose robots, the range of possible applications is huge; the consensus space may be even larger than on the technical side. New techniques already show results and can be produced at small scale, so in the coming period we’ll watch companies test promising applications and converge on a few.

Second, we need the next modeling paradigm—an endless pursuit. A new paradigm could elevate today's ceiling, enabling things once impossible. Yet scalability remains an open question: architecture, data pipelines—there are still plenty of challenges.

NBD: Humanoids still struggle with long-duration reliability. What do you think?

Chen: True. Even if future applications are vast, the technology stack isn't complete, especially for homes. We're starting with industry and commercial services, refining tech, building data flywheels, and iterating capabilities before moving into households.

Overall, progress is faster than expected. Talent and capital have created an ecosystem of communication, learning, collision, cooperation, and competition. Two years ago no one talked about robot foundation models; now there's consensus and real capability. Hardware and data have changed almost overnight.

NBD: Many vendors let their robots box. Does winning or losing in the ring prove stability?

Chen: Not necessarily. Boxing has weight classes; our torque is several times higher than some smaller humanoids. It isn't a universal metric. Within the same class you might see whose balance is better.

When will humanoids enter every home?

NBD: ROBOTERA's lineup can both "show" and "work," and you've achieved initial commercialization. How far are we from mass deployment? What are the hurdles and opportunities?

Chen: Early adopters—universities and top tech giants—have already bought our robots and built on them, completing the first commercial loop. We're now exploring broader markets.

We're running two parallel tracks. 

- L7, a full-size biped, is designed to industrial specs: speed, efficiency, dual-arm payload, precision, and force. Modular design lets us detach the upper body as a lighter version for specific tasks. Main use case so far is logistics—sorting, scanning, upper-body manipulation—driven end-to-end by a VLA (Vision-Language-Action) model. Results look very promising. 

- Q5, aimed at services, is slimmer and striking. It has five-finger dexterous hands and full-body articulation, making it ideal for lifelike service and interaction. Pilots are underway in retail and storefront settings.

We'll enter B-end (business) scenarios first: high-value logistics and targeted service deployments.

NBD: Will 2024 be the "commercialization year"? What's the threshold?

Chen: Real commercialization hasn’t started; at the latest it will begin next year. Units are already selling, but large-scale, high-value commercialization is a 2025 milestone.

My yardstick is 10 000 units—not for the entire industry, but for any single player. Leading industrial-robot companies expect to reach that next year. Even if we don’t hit it exactly, we'll be on the path. Once that number is crossed, scaled shipments will follow.

NBD: When will we see humanoids in our homes?

Chen: Today's volumes are still low; costs will drop quickly for factory use. Homes are harder. Entry-level households require bigger value. Technically and economically, the bar is higher.

To justify itself at home, a robot must cook meals, wash all the clothes, and more. Yet as technology advances, household deployment is inevitable. Given current speed, I expect meaningful in-home progress within three to five years. The first versions may not be fully humanoid; once capabilities mature, they'll start handling household tasks.