A clip of Unitree's humanoid robots boxing each other has been doing the rounds, and most people I know laughed at it. I get it. Two robots squaring up, throwing slightly stiff punches, occasionally toppling over like a toddler who's had too much sugar. It's funny.

I watched it more times than I'd like to admit, and I wasn't laughing. I was watching their feet.

They were strafing. Circling out when they got pressured, cutting angles instead of just marching forward. When they threw, you could see the weight moving through the technique, from the floor up through the hips into the punch, not just an arm swinging on its own. And the bit that genuinely gave me goosebumps - when one of them took a hit, it rebalanced. It staggered, found its base, and came back. That's not a trivial thing. I've seen plenty of humans on a mat who can't do that yet.

Were they good fighters? No. Positionally they were lost. They walked into range with their hands down and threw punches at where the other robot used to be. If you've ever watched a beginner's first sparring round, it looked a lot like that - all the physical pieces there, none of the reading.

That gap is the interesting part to me.

The body is mostly solved, or getting there faster than anyone expected. Balance, footwork, getting weight behind a movement - that's the stuff people assumed would take decades, and here it is, a bit clumsy, on a viral clip. What's missing is the eyes. Not cameras, they've got cameras. The ability to look at another body and understand what it's about to do. To see a shoulder dip and know a hook is coming. To notice the other guy's base is gone and step in.

Add some real-time vision that can read an opponent's technique as it's happening and we might have Real Steel sooner than anyone thinks. I'm only half joking.

And it's not a thinking problem so much as a timing problem. A robot that needs three seconds to reason about a punch has already been punched. It needs something fast and reflexive, the part of you that flinches before you've decided to. But reflexes alone just make a very twitchy robot. It also needs something slower underneath, that understands what's been happening over the last minute, what this opponent likes to do, when to be patient. Fast and slow, both at once, handing off to each other without dropping anything.

That's roughly what I spend my days on right now, just pointed at people on a mat instead of robots in a ring. The hardest case is always the same one - two bodies close together, limbs everywhere, and a machine trying to keep track of whose is whose.

The boxing is a gimmick, obviously. Nobody needs robots punching each other. But the same skills are what you'd need for a robot to spar with you, or catch someone who's about to fall, or hand you a cup without knocking it out of your grip. All of it comes down to a machine watching a person closely enough to move with them instead of against them.

People laughed at the boxing robots. I think they'll stop laughing sooner than they expect.

I'll be watching the feet.