Chinese robotics makers pivot from flashy humanoids to practical, efficient designs

Walk through the robotics conference and you'll see two different visions of the future competing for attention. At Unitree Robotics' booth, the crowds gather to watch bipedal robots play table tennis with precision, their movements fluid and eerily human-like. These machines balance on two legs, coordinate complex movements, and execute feats of dexterity that seemed impossible just a few years ago. It's the kind of display that makes headlines and captures the imagination—robots that move like us, think like us, exist in our world on our terms.

But step away from the marquee attractions and a quieter revolution is unfolding. In other corners of the same exhibition halls, robots are doing the actual work. They sort parcels. They load production lines. They move materials from one place to another. Most of them roll on wheels. Some have conventional robotic arms bolted to their frames. They are not trying to be human. They are trying to be useful.

This tension—between the spectacular and the practical—is reshaping how Chinese robotics companies think about their future. Yang Li, a senior engineer at the start-up X Square Robot, posed the question plainly: if wheels can accomplish the task, why build legs? It's a challenge to an entire industry narrative that has treated bipedal humanoid design as the inevitable endpoint of robotics development.

The answer came in the form of numbers. During a controlled test this month, X Square Robot's two-armed WALL-B model sorted 1,816 packages in an hour. The company compared this directly to Figure AI's Figure 03, a bipedal humanoid from a U.S. start-up that has drawn significant investment and attention. The Figure 03 handled an average of 1,248 parcels per hour. The wheeled robot, built for a specific job rather than designed to mimic human form, outperformed the humanoid by more than 600 packages an hour.

This is not a marginal difference. In logistics and manufacturing, where speed and consistency determine profitability, the gap between 1,248 and 1,816 parcels per hour translates directly to cost savings and competitive advantage. It suggests that the industry's long obsession with humanoid form—the two legs, the torso, the arms positioned like a person's—may have been solving the wrong problem.

AI2 Robotics, another Chinese start-up, demonstrated a different approach at the conference. Their wheeled robot made coffee and ice cream for visitors, swapping between different end effectors depending on the task. One arm carried a gripper for handling objects; the other featured a dexterous hand for more delicate work. The machine was not trying to be a humanoid barista. It was trying to be a flexible tool that could adapt to multiple jobs without requiring a complete redesign.

What's emerging is a pragmatic reorientation. The flashy demonstrations—the kickboxing matches, the dancing routines—still happen. They still draw crowds. But the actual deployment, the real money, the genuine productivity gains are happening with machines that look nothing like humans. They are optimized for their specific tasks, built to be efficient rather than impressive, designed around what works rather than what looks right.

For investors and manufacturers watching this shift, the implications are significant. If a wheeled robot can sort packages faster and cheaper than a humanoid, the question of which design to fund becomes a business question, not a philosophical one. The industry may be entering a phase where the most successful robotics companies are those willing to abandon the humanoid dream in favor of something far more mundane: machines that simply do the job better.