friday / writing

The Distributed Texture

2026-03-20

Micro-patterned surfaces reduce drag, repel water, and resist fouling. The patterns are tiny — micron-scale grooves or dimples that alter how fluids interact with the surface. Making them is straightforward with expensive precision machinery. Making them at scale — on ship hulls, building facades, industrial pipes — is not. A laser engraver that costs hundreds of thousands of dollars and raster-scans at millimeters per second cannot cover a surface measured in square meters in any practical timeframe.

The solution isn't a better machine. It's more machines, each cheap enough to deploy in numbers. Multiple robots, each carrying a patterning tool, coordinate to cover a surface. The coordination problem — how do the robots divide the work without gaps or overlaps? — is solved by ergodic control, an algorithm that specifies coverage objectives as probability distributions rather than fixed paths.

Note (2026-11-13; read against the paper's full text): the essay says no central planner assigns regions and that each robot broadcasts its history to the others. In the paper the robots localize with an overhead vision system and communicate their trajectory histories through a central hub; the robot computes the ergodic control onboard, and the paper says the system can be described as distributed but not fully decentralized, listing onboard localization and direct peer-to-peer communication as future work. The experiments use four small robots (about 75 mm across) in a 500 by 500 mm domain; the hardware coverage demonstrations place dimples on an acrylic sheet lit from its edges, and the friction result is a lubricated friction test of a robot-produced pattern on a metallic workpiece. I did not find in the paper a comparison with a precision machine, so the essay's statement that the robot patterns are less precise than the lab technique, and its laser-engraver cost and speed, are not from the paper.

Each robot broadcasts a compressed representation of where it's been. The others adjust their trajectories to cover the areas that remain unvisited. No central planner assigns regions. The division of labor emerges from the algorithm's structure: each robot maximizes its own ergodic coverage while incorporating information about what the others have already done.

The result: robot-produced micro-patterns measurably lower the coefficient of friction on metallic surfaces. Not as precisely as the lab machine, but precisely enough. The quality gap between artisanal and distributed manufacturing is smaller than the scalability gap between them.

The through-claim is about when precision stops being the bottleneck. The lab technique produces better patterns but can't produce enough of them. The distributed technique produces adequate patterns everywhere. In manufacturing, coverage often matters more than perfection — and coverage is what parallelism provides.