When someone says “robot,” what’s the first image that pops into your head?

Probably a humanoid (at least I used to think that).

Like Tesla’s Optimus. Well, maybe the Terminator.

Two legs, two arms, a face that can look you in the eye.

But if you walk into an actual factory or warehouse and look at the robots clocking in right now, almost none of them look like that. What you’ll see is bolted-down arms swinging around welding stations. Flat boxes gliding across warehouse floors on wheels.

The humanoid form is the most intuitive shape, and it’s the one that capital is chasing hardest right now. Yet the robots making money and showing up to work every day look nothing like it.

Why?

It took me a while of studying to get it: robots don’t have a “correct” shape. What they look like gets dictated by the job they need to do, not by what looks cool on a press release.

There’s an old line in design, form follows function. It fits robotics disturbingly well.

There are lots of ways to classify robots. ISO has its terminology standards; IFR (International Federation of Robots) sorts them into six categories by mechanical structure, and it gets pretty granular.

In the context of today’s article, let’s look at how it gets around.

Some don’t move at all. Industrial robot arms are bolted to the floor or a workbench. Their entire job sits within arm’s reach and they have no reason to go anywhere. A few joints rotating covers welding, painting, assembly, all of it.

Some roll on wheels. The AMRs in warehouses zip around on flat ground hauling goods. The floor is smooth, wheels are fast and cheap. Good enough.

Legs only show up when the ground gets rough, there are stairs to climb, or the space was designed for humans and their narrow doorways.

Legs can climb stairs, step over debris, go places wheels can’t. It just sounds way more versatile. Why not make everything bipedal and call it a day?

I assumed the gap between the wheels and the legs was about 2x, maybe 3x. Enough to matter for battery life, but maybe not enough to kill the concept.

And it’s not because anyone’s doing bad engineering. It’s just purely physics. A leg lifts the entire machine, sets it down, lifts it again, every single step, fighting gravity the whole way.

Bipedal walking is worse still: most of the energy goes into not falling over, not into doing anything useful. If you can get the job done standing still, why burn power learning to walk…?

That number rewired how I look at humanoids. Everything after it is the same story in different currencies.

Take cost. A humanoid’s actuators, the mechanical “muscles” (easier to understand) at every joint, eat 40% to 50% of the entire bill of materials. Tesla has been making their own lead screws to bring that down, cutting per-unit cost from $3,000 down to $800. That they can cut 75% tells you how inflated the starting price was (first principle wins!).

Batteries are simpler to summarize: lithium energy density improves about 5–8% a year. Even with solid-state cells, most loaded humanoids probably won’t last a full 8-hour shift. A worker who clocks out halfway through. Why would a factory hire it? Then again, there are humans with the same problem (sigh).

Energy, cost, batteries.

All three elements are pushing against the humanoid form at once.

That’s why working robots don’t look human. Because physics and economics won’t let them, not yet. Arms and wheels are everywhere for a reason: cheap, efficient, reliable, and just barely good enough.

You don’t have to remember any of the numbers. Once you start looking at robots this way, a lot of headlines get easier to decode. Next time you see a slick humanoid demo, or a headline about some company raising another few hundred million for bipedal development, one question to ask is enough:

The math looks this bad, like I mentioned earlier.

And yet the smartest money in the world keeps pouring into humanoid robots. Hundreds of billions, collectively.

Have they spotted something the rest of us haven’t?

Or are they just betting on a narrative?

I haven’t fully figured that out yet.

That one’s for the next piece.

Cheers.