Robots Ran a Half-Marathon. Here's What It Actually Tells Us About Embodied AI

I was skeptical when I first heard about a humanoid robot half-marathon in Beijing. “Gimmick,” I thought. But the more I read, the more I realized this was something more interesting: a genuine stress test.

21 kilometers tests everything — real-time navigation, gait stability, battery management, thermal regulation. For a bipedal robot, covering that distance at a human jogging pace requires serious engineering.

The Glory Robotics team finished in about 50 minutes. That’s 4+ km/h, roughly a slow jog. For context, the first time I tried running a legged robot outdoors, it fell over at 30 meters.

What’s interesting is the two schools of thought on display: fully autonomous robots versus “human-in-the-loop” designs where remote operators intervene during complex sections.

I personally think the hybrid approach wins in the short term. Real-world applications — factory inspection, hazardous environment work, elderly care — don’t require perfect autonomy. They require useful assistance with human oversight.

The full autonomy path matters long-term. But the practical path to market is hybrid systems that work now.

That’s not less exciting. It’s actually more honest about where the technology actually is.