The Robot Half-Marathon Wasn't About Winning. It Was About Showing Up.

On April 19th, in Beijing’s Yizhuang district, a somewhat unusual half-marathon concluded.

Unusual because some of the runners had two legs and stood about human height. They moved along the track—or more accurately, shuffle-walked, stumbled, and occasionally tumbled their way through 21 kilometers.

My social feeds were full of it. Videos of robots eating pavement at the starting line and getting back up. One robot losing an arm mid-race. The unexpected moment when a robot overtook a human runner, prompting a collective gasp from the crowd.

Then came the headlines: “Robots Beat Humans,” “Embodied Intelligence Year One,” “China’s Robotics Overtaking.”

After reading through all of it, I have just one thing to say: let’s not get ahead of ourselves.

What the Numbers Actually Tell Us

The fastest robot finished in about two hours. That’s roughly equivalent to a below-average amateur human runner. The completion rate was just over 50%. Given the energy constraints, mechanical complexity, and environmental uncertainty—actually, the fact that any robot finished at all surprised a lot of people in the industry.

So what went wrong? Or more precisely, what revealed itself?

First, balance control still isn’t there. I watched multiple videos of robots wobbling badly through turns or over uneven ground. One went face-first into the track. Boston Dynamics’ Atlas can do backflips, yes—but in a controlled lab environment. Outdoor terrain with real crowd noise, temperature variation, and unpredictable surface conditions is a completely different problem.

Second, energy consumption is brutal. One engineer posted afterward that their robot hit battery warning at kilometer 12 and basically coasted the rest of the way. Mechanical systems are just fundamentally less efficient than biological muscle for sustained aerobic activity.

Third, outdoor environmental perception is still lagging. Lighting conditions, ground friction coefficient, unexpected obstacles—these are all things humans process essentially for free, but robots need extensive sensor fusion and real-time computation to handle. At one turn, I watched a robot hesitate for a disturbingly long time, clearly struggling to process the scene.

The Industry Signal Was More Interesting Than the Tech

The teams that showed up weren’t running academic experiments. Several were robotics companies testing their commercial-grade humanoid robots in a real, high-pressure environment with actual crowds, media coverage, and competitive stakes.

Results: mixed but revealing. Yes, some teams finished with a commercial unit on a single charge—impressive. But the finishing robots mostly walked the course at a brisk-walking pace, not ran. And roughly half the field didn’t finish at all.

This suggests a more honest commercial path than the optimistic narrative: embodied intelligence will probably find its feet in specific, controlled environments first—not by matching human general capabilities, but by being dramatically better than humans at narrow, well-defined tasks. Factory robotic arms have been outperforming humans at specific jobs for decades. The reason? Factories don’t need two legs.

My Take

One comment I saw on social media: “After watching this race, I think it might take 20 years before robots can replace humans at marathon running.”

I think that’s still overoptimistic. Or rather, the wrong framing.

The actual value of this event wasn’t “can robots beat humans at a marathon?” It was that the entire industry got to see their products under real conditions, not lab demos. Real crowds, real cameras, real competitive pressure. That’s a different kind of catalyst than any internal benchmark.

I keep thinking about the iPhone analogy. When the first iPhone launched in 2007, nobody thought it would replace feature phones. Worse phone functionality, worse battery life, and it shattered if you dropped it. We all know what happened next.

Embodied intelligence might be on a similar trajectory. Maybe 10 years from now, April 19, 2026 will be remembered as the iPhone moment—not because the product was good, but because it proved the category existed.

Until then, we probably need to accept that today’s humanoids are far from replacing humans at anything general. But they’re genuinely starting to find their place at the edges of human capability.