Embodied Intelligence Revisited: From Unitree to Emerging Startups — Capital Frenzy or a Modern Tower of Babel?

Posted by on

 


Embodied Intelligence Revisited:

From Unitree to Emerging Startups — Capital Frenzy or a Modern Tower of Babel?

A few days ago, I shared some thoughts on humanoid robots. After listening to several podcasts and interviews in February, I’d like to expand on that discussion with a few additional perspectives.

Do Humanoid Robots Even Need to Exist?

The answer is yes—without much debate.

A general-purpose platform always carries more long-term value than specialized systems.

Take the “big four” industrial robot manufacturers as an example. Their success is largely built on generalization:

  • Strong adaptability
  • Wide application range
  • Scalable deployment

Without general-purpose capability, scaling becomes inefficient, as engineering effort gets fragmented across too many use cases.

A useful analogy is the evolution of mobile phones:

  • Before smartphones, feature phones had fragmented platforms
  • Limited applications were not just due to technological constraints, but also ecosystem fragmentation
  • Too many platforms diluted developer effort and reduced ROI

A unified platform—like smartphones—unlocked the modern app ecosystem.

Similarly, a general-purpose robot platform will significantly lower the barrier for future software and application development.

As for whether that platform should be humanoid:
for now, it appears to be the most practical choice.

Some argue that quadrupeds are more stable. That may be true—but stability is an engineering problem:

  • Add deployable support structures when stationary
  • Reduce energy consumption for balance

Form factors may evolve in the future, but today, the humanoid form remains the closest approximation to the human-centric world.

Even in nature, primates often stand upright during complex interactions like combat. The humanoid structure is not arbitrary—it is a functional equilibrium shaped by evolution.

So when people ask, “Why not build specialized robots instead?”
they are actually mixing two different discussions.

Skepticism about humanoid robots is reasonable.
Rejecting the underlying business logic of general-purpose platforms is not.

Will Embodied Intelligence Succeed?

The honest answer: no one knows.

But one thing is clear—its current appeal to investors comes from the convergence of:

  • Humanoid robotics
  • Artificial Intelligence

At present, most “humanoid robots” are still at the level of:

  • Advanced automation
  • Adaptive motion control

They are far from what we would call true intelligence.

In 1950, Alan Turing proposed what is now known as the Turing Test:
If a human cannot distinguish whether they are interacting with a machine or another human, the machine can be considered intelligent.

This raises an interesting question:

Will embodied intelligence have its own “Turing moment”?

Is Embodied Intelligence Computer Science—or Biology?

The “embodied” part is clearly rooted in mechatronics.

But what about “intelligence”?

Is it:

  • A problem of computer science?
  • Or something closer to biological systems?

Consider this:

Modern AI can:

  • Write code
  • Generate images

These are learned, acquired capabilities.

But a human infant:

  • Learns to roll, crawl, sit, and stand
  • Without explicit instruction or modeling

Humans do not calculate motion trajectories consciously.
Even when lifting an object, we only estimate feasibility—not compute precise control.

Motor intelligence is innate.

Which leads to a deeper question:

Can something that is inherently biological and emergent
be fully replicated through computational learning?

From Autonomous Driving to Embodied Intelligence

Some companies are attempting to transfer methodologies from autonomous driving into embodied intelligence.

But the objectives are fundamentally different:

  • Autonomous driving: avoid collisions
  • Embodied intelligence: complete tasks

These are not equivalent.

The key question becomes:

Will computing power evolve to a point where robots can perform tasks with human-level adaptability?

Domestic vs. Overseas Talent Paths

There is also an interesting contrast in founder backgrounds.

Take Wang Xingxing as an example:

  • His academic background is not considered elite
  • Yet he built a profitable company

This reflects a common pattern in grassroots entrepreneurship:

  • Focus on cash flow first
  • Prioritize product-market fit over theoretical perfection

Unitree has reportedly:

  • Achieved revenue exceeding RMB 1 billion
  • Maintained profitability

With increasing visibility (e.g., national exposure events), its growth is likely to continue, and an IPO seems plausible.

In contrast:

  • Founders with overseas education and stronger academic credentials
  • Often secure more venture funding early

But higher funding does not necessarily translate into faster commercialization.

Interestingly, much of the current investment enthusiasm is built on the validation created by early commercial success stories like Unitree.

Is This Just a Capital Play?

Some argue that humanoid robotics is simply a fundraising narrative.

But capital always seeks an outlet.

Consider the ride-hailing sector between 2014–2017:

  • Over RMB 30 billion burned
  • Peak daily subsidies exceeding RMB 100 million

Compared to that, the capital flowing into humanoid robotics may just be getting started.

The real question is not whether money is being spent—
but whether it creates lasting technological value.

Historically, even failed experiments can yield breakthroughs.

For example:

  • During World War II
  • Metallurgical research driven by military needs contributed to the development and scaling of stainless steel

Technology investments rarely go entirely to waste.

What Does This Mean for the Automation Industry?

The automation sector is already benefiting:

  • Public companies supplying components have seen valuation gains

But the more important question is:

Will embodied intelligence feed back into traditional automation?

The answer is likely yes—but not immediately.

Potential spillover effects include:

  • More advanced joint motors
  • Higher integration density in mechatronic systems
  • Smaller, more efficient components

Over time, these innovations could reshape the automation landscape.

Final Thoughts

Humanoid robotics may:

  • Struggle before succeeding
  • Or discover entirely new applications along the way

It is still too early to draw conclusions.

But one thing is certain:

Money spent on technology rarely disappears without a trace.

Even if the original vision fails,
the byproducts often reshape entire industries.