- Unitree is putting mass production ahead of the robot-racing spotlight
- Its latest setbacks expose the harder challenge: making robots truly intelligent
Unitree is stepping back from the robot-racing spotlight to focus on mass production — a strategic shift underscored by its decision to scale back events at this week’s World Humanoid Robot Games in Beijing
Moreover, a spectacular crash involving its latest “Superman” robot just days earlier may have reinforced the case for the pivot.
On August 22, as the games opened, Unitree unexpectedly announced that it was withdrawing from several events it had previously entered, citing limited time, manpower, new robot availability and testing capacity.
The company said its resources had historically been focused on mass-production products and apologized for not devoting enough attention to competition preparations.
That explanation might have sounded reasonable on its own. But on August 20, the newly unveiled “Superman” crashed into a power distribution box during a high-speed turn.
Two days later, Unitree’s robot finished last in its 100-meter preliminary heat in 12.41 seconds, while rival Tiangong Ultra clocked 9.39 seconds, faster than the human world record.
The irony is hard to miss: the company that helped turn robot athleticism into a global spectacle is now effectively telling the market that it has more important things to do.
From body to brain
This is less a story about losing a race than about changing priorities.
Unitree built much of its early reputation by making robots run faster, jump higher and move more dynamically.
Its machines could backflip, sprint and dance — capabilities that gave the company enormous visibility and helped distinguish it from a crowded field of humanoid robot makers.
But Unitree’s IPO prospectus shows where its ambitions are heading. Of the 4.2 billion yuan it plans to raise, 2.02 billion yuan is earmarked for embodied AI foundation-model development, nearly half the proceeds.
That is significant because the hardest problem in robotics is no longer necessarily getting a machine to move.
Instead, it is getting the machine to understand what is happening around it and generalize what it has learned to unfamiliar situations.
Founder Wang Xingxing has acknowledged as much, calling generalization the industry’s biggest bottleneck.
Robots can perform almost perfectly in fixed environments, he said, but their success rates fall sharply when objects or surroundings change.
His estimate for the industry’s critical point is also strikingly cautious: two to three years at the earliest, or five to 10 years at the latest.
The cost of being athletic
This creates a dilemma for Unitree.
Its athleticism is precisely what made the company famous. But the more the industry moves toward general-purpose embodied intelligence, the less impressive a robot simply running faster or jumping higher becomes.
The “Superman” crash illustrates the problem perfectly. A machine can have extraordinary motors, actuators and control systems and still make a remarkably poor decision in the real world.
That is why Unitree’s latest strategy deserves more attention than its race results. While competitors are increasingly pouring resources into foundation models and general-purpose robot intelligence, Unitree is emphasizing mass production and commercialization.
The market has valued the company at nearly 244 billion yuan ($36 billion), as of August 25, providing enormous resources but also raising the pressure to turn technical spectacle into commercial reality.
Unitree may therefore be making a sensible trade-off: less time proving that its robots can run, and more time making sure they can actually work.
The next race is no longer about who can run fastest or jump highest. It is about who can make a robot understand, adapt and work reliably — at scale.
