Cockroach bait, Robinson Crusoe and the roomba!

Nobody Ships the Floor Plan

The pest control pamphlet says: apply gel to cockroach travel routes.

That's a strange sentence to sit with for more than a second, and worth sitting with anyway, because the manufacturer isn't telling you where cockroaches *could* go. Every gap in a kitchen leads somewhere. The pamphlet is telling you where they actually do go — which is a completely different, much narrower, much more useful piece of information. It's the difference between a floor plan and a route. The floor plan is every path physics allows. The route is the handful of paths something has already decided, through repetition, are worth using.

Which means a route was never really a fact about geography. It's a claim about payoff. A gap under the cabinet isn't a route because it connects two points — it's a route because something has already found that going through it, repeatedly, reliably gets it somewhere worth going, cheaper than the alternatives. Geometry just says a path exists. Repetition is what turns a path into infrastructure.

And the moment a path becomes predictable enough to count as infrastructure, it becomes something else too: interceptable.

That's the darker branch of the same fact, and it's worth taking seriously rather than skipping past, because it's the same mechanism the pest control gel exploits, just aimed at something with more stakes. A predator doesn't need to outrun prey across open ground if it can simply wait somewhere the prey reliably passes. Hunting by ambush isn't a separate skill from route-reading — it's route-reading with a different payoff attached. Scale the same logic up and it stops being about individual animals and starts being about armies: a military force has to move supplies along predictable corridors, and every predictable corridor is a place an enemy can choose to be waiting instead of chasing. War, at the level of pure movement, is largely a contest over who controls the routes that both sides are forced to depend on. Scale it differently and you get colonization — explorers and settlers don't wander a new continent uniformly, they follow rivers, coastlines, and gaps in mountain ranges, because those are the paths that already make movement cheap, and cheap movement is what turns unfamiliar territory into a supply chain. And once a route is well-established enough that everyone relies on it — a mountain pass, a trade road, a river crossing — it becomes something even simpler and older than any of that: a good place to wait for someone. Banditry isn't a separate invention from infrastructure. It's a parasite that can only exist once infrastructure exists to parasitize. You cannot ambush a route nobody uses yet.

All of which makes Robinson Crusoe the clean control case, because he's the one character in this whole set who has no routes yet for anyone to intercept. Defoe hands his character an experimental condition almost nobody gets to run cleanly — one organism, one completely unfamiliar board, no inherited map, no prior information, real survival pressure. Crusoe doesn't arrive on the island and consult a mental floor plan. He has to build one, from nothing, through the same mechanism a cockroach uses without any of the narration: survey, identify resources, identify hazards, mark safe locations, travel between useful points enough times that the traveling stops being exploration and starts being routine. By the time routine sets in, he's not moving through unknown territory anymore. He's moving through infrastructure he built himself, one repeated trip at a time. The island didn't have his routes waiting in it. He seeded every one of them, the only way routes ever get seeded — by finding out, through motion, what pays. Nobody is waiting to ambush him yet, precisely because nothing about his movement is predictable to anyone but himself. Predictability is the whole precondition for everything in the paragraph above this one. Crusoe simply hasn't gotten there yet.

Which is also, unglamorously, what's happening under your kitchen cabinet. A route doesn't require a mind sophisticated enough to plan one. It just requires an organism that can move, notice what worked, and repeat what worked. The mechanism doesn't care whether it's running in a stranded sailor or an insect with no plans at all.

Roomba is the case that shows what happens when that mechanism is missing entirely, and then what happens once you bolt it back on. The earliest Roombas, running reactive navigation, don't build anything that deserves to be called a route. They bump into an obstacle, turn, keep moving, bump into the next one — pure improvisation, with no memory carried from one collision to the next. Every session starts from zero. There's no infrastructure because there's no persistence: nothing about today's wandering makes tomorrow's wandering any cheaper. Later Roombas, running visual or laser mapping, are a completely different animal. They build a persistent representation of the space, localize themselves inside it, and plan coverage instead of reacting to it — which is, structurally, the same move Crusoe makes the first time repeated travel turns into a route he trusts instead of a path he's testing.

But a mapped Roomba is quietly solving two different problems that only look like one problem, and separating them matters. There's coverage — did I reach every part of the required area — and there's transit — how much wasted movement did it cost me to get between the useful parts. Those can genuinely conflict: the route that minimizes distance can require awkward, costly turns, while a slightly longer route with fewer turns can be cheaper overall. This isn't a robotics-specific problem. Agriculture solved a version of it long before anyone had a name for coverage path planning, and solved it with nothing more sophisticated than accumulated practical knowledge. The boustrophedon pattern — strip, turn, strip, turn, named after the way an ox turns while ploughing — is the low-tech ancestor of every modern coverage algorithm, and it exists because a farmer with a draft animal has exactly the same two costs to balance that a robot vacuum does: don't leave any of the field uncovered, and don't waste the animal's energy or your own time recrossing ground you already worked. The farmer's version has extra constraints a Roomba never has to think about — implement width, turning radius, slope, where the harvested material goes — but the underlying shape of the problem is identical, discovered independently, centuries apart, by oxen and by lidar.

And even a Roomba that's solved both of those problems perfectly is still only mapping half of what's actually there, because Roomba's map and a cockroach's map are not the same map even though they're drawn over the identical square footage. Roomba's map says floor, obstacle, room, dock. A cockroach living in that exact space is running a completely different behavioral map on top of the same geometry — food, darkness, edge, hiding place, escape route. Same walls, same floor, same furniture, two entirely different behavioral geographies laid over one physical one, because a route was never really about the space. It's about what the space is being used for. Move the same chair, and Roomba logs a new obstacle to route around. A cockroach might log a new corridor, a new edge to hug, a new shadow to shelter in. The furniture didn't change meaning on its own. It changed meaning because two different payoff structures were reading the identical object.

Which is the actual point this thread kept almost reaching and then wandering away from: a route is never a permanent fact once it's established. It's a live answer to a question that keeps getting re-asked, and the moment the answer changes, the route is already obsolete whether or not anyone announces the update.

Rearrange the kitchen — move the fridge, shift the bin, swap where the containers sit — and you haven't erased the cockroach's underlying problem. It still needs to get from shelter to food and water without crossing too much exposed space. What's changed is the cost and benefit of the specific passages that used to solve that problem. And critically: nobody ships the roach a floor plan update. There's no patch notes, no memo, no map redrawn overnight. The animal discovers the change the only way it ever discovers anything — by moving, hitting a gap that didn't used to matter, and finding out it works. If it keeps working, that gap graduates. Call it a service lane: not yet infrastructure, but no longer just an accident either, already being reused because reuse is cheaper than re-exploring from scratch every time.

That's the same exact mechanism Crusoe ran to seed his first routes, just triggered by a different kind of change. Crusoe's routes appeared because there was no board information yet. The cockroach's new service lane appears because the old board information just went stale. Seeding and rerouting aren't two separate processes that happen to look similar. They're the same process, running twice — the first time because nothing existed yet, the second time because something that used to be true stopped being true.

Elephants run the identical update on a scale that makes the mechanism obvious instead of easy to miss: a highway starts carrying trucks loaded with sugarcane, on a predictable schedule, and elephants that had no reason to care about a strip of asphalt start treating it as part of their working spatial knowledge, because a road that reliably produces food is functionally no different from a game trail that reliably produces water. The road didn't get built for them. They didn't ask permission to route through it. The payoff changed, so the map changed.

Rabbit Island runs the same update, and it's worth doing this one properly rather than gesturing at it, because it's got something the other examples don't: an actual controlled experiment, run by accident, with a clean before-and-after. Okunoshima's rabbit population, before tourism, was working off whatever the island's vegetation could support — a diffuse, unglamorous resource, spread wherever plants happened to be growing, producing whatever movement pattern that spread justified. Tourism changed the resource geography without changing a single blade of grass. Visitors carrying snacks became a new, concentrated, extremely reliable food source, arriving at predictable points — the ferry landing, the paths tourists actually walk — and rabbit movement reorganized around it hard enough that the standing joke about the island is that the most dangerous animal on it isn't a predator, it's a tourist with a bag of pellets, because fifty rabbits converging on the same ten square feet is a real hazard to the rabbits themselves. That's a route forming in real time, in a population with no capacity to plan one, purely because a new, better-paying option appeared and individual rabbits kept finding it by moving toward it.

Then COVID hit, tourist traffic dropped sharply, and the population reportedly declined afterward. That's not a metaphor for what happens when a route stops paying. That's the actual event. The vegetation-based food supply hadn't changed. What had changed was the concentrated, human-supplied option the population had reorganized around — and when that option went away, so did the movement pattern built on it, and so, eventually, did some of the population the movement pattern had been supporting. Nobody rerouted the rabbits back to a purely vegetation-based existence on purpose. They didn't get a memo either. The board changed, the payoff disappeared, and the infrastructure that had been quietly built on top of it didn't survive contact with that fact.

None of this needed humans to invent it, which is the part worth sitting with longest. Ants seed routes to a food source and let them decay the moment the food is gone, no sentiment attached, no memorial for the abandoned trail. The mechanism is older than infrastructure, older than tool use, almost certainly older than the nervous systems sophisticated enough to be surprised by it: a useful path gets discovered, reused while it pays, and abandoned or replaced the moment it stops. Humans didn't invent this. Humans just started noticing the discarded routes and building permanent things on top of them — a worn footpath becomes a paved one, a river ford becomes a bridge, a caravan stop becomes a town, a trade road becomes valuable enough to be worth ambushing — but the underlying logic predates every road ever paved and every war ever fought over one.

Which is why a person leaving a deliberate landmark for their own future self — a bait station taped in place specifically so it can be found again years later, a task reminder propped where tomorrow's foot traffic will trip over it — isn't a clever human trick layered on top of nature. It's the oldest trick there is, just finally being run on purpose, by someone who noticed they were already doing it by accident. Present-you edits the board. Future-you, an organism with incomplete memory moving through a space it half-remembers, encounters the edit and gets the message without needing it spelled out. That's not a metaphor borrowed from ant pheromones. It's the identical mechanism, just given a name and pointed inward.

The pamphlet was never really about cockroaches.

It was about the fact that nothing gets to keep its routes for free. Every path — a kitchen shortcut, an ancient trade road, a rabbit's ten feet of ferry-landing grass — stays a path only for as long as it keeps paying. Predictable enough to be useful is the same thing as predictable enough to be exploited, predictable enough to be waited on, predictable enough to collapse the moment the payoff behind it quietly stops existing. And the moment any of that happens, something is already out there, moving, quietly finding out where the new one is.