ASCENT

What this is

This machine has no sensors, no feedback, and no computer. One motor turns one crankshaft at constant speed. Everything else is twelve link lengths, found by evolution (~one million candidate designs), that turn a crank circle into a stepping motion matched exactly to the standard staircase of the building code: 170 mm rise, 280 mm going. One crank revolution = four stairs. What you see above is a replay of the actual rigid-body physics simulation that gates the claim: a 14-step flight, climbed at exact geared speed, with spring feet and real friction.

The machine even fixes its own mistakes without knowing it made them: every stride overshoots by 1.5% and presses the landing foot against the next riser. The staircase itself keeps the machine in phase — placed wrong by half a tread, it converges back onto the same track within a few stairs. The staircase is the program.

The leg lab

Theo Jansen's Strandbeest numbers walk on flat sand. Drag the slider and watch the same linkage become a stair-climber as its twelve lengths morph to the evolved genome. Amber curve: the path the foot traces.

The genome

The unglamorous truth

It took five failed machines to get here. Rigid feet locked the mechanism solid (8,000 N·m of self-fight). Slippery feet ratcheted downhill at every support handoff. The optimizer learned to lie to my support model by gliding its feet in low. A two-steps-per-revolution gearing turned out to be geometrically impossible (alternating feet land on lines half a riser apart). And the first balanced design was a cantilever that dove face-first up the stairs until it got a ballast tail. Every fix is in the physics you're watching. What I still can't verify: whether a real, buildable machine with bearing slop and 3D dynamics would forgive as much as a clean 2D simulation does.