The Robot That Can't Leave the Room
Point a camera at the ceiling, tile the floor with markers, give the robot a map it never has to build itself, and the robot can be dumb and cheap. That's the pitch behind room-instrumented design: push the intelligence off the machine and into the environment, and the per-unit cost drops because the hard part is now a fixed install, not a thing you carry around on wheels.
It's a real cost cut. It's just not the one people think they're buying.
The clearest version of this is the floor fiducial. Amazon's warehouse robots, the lineage that started as Kiva Systems before the 2012 acquisition, navigate by reading markers printed on the warehouse floor rather than by building a model of the world from onboard cameras. The robot doesn't need to understand the room. The room has already been pre-chewed into something a cheap robot can read. For a single warehouse running the same layout for years, that's a good trade: you pay once to instrument the floor, then amortize it over millions of trips.
The trap is treating that as a general robotics strategy instead of a warehouse-specific one. The intelligence that makes the robot work was never in the robot. It was in the floor, the ceiling rig, the calibrated camera array, the map somebody built by hand for that exact space. None of that travels. Move the robot to a different building and it isn't slightly worse, it's inert, because the thing doing the actual work of localization and planning stayed behind in the walls of the old room.
This is easy to miss because the accounting looks right from the manufacturer's side. Robot unit cost: low. Infrastructure cost: one-time, amortized over the life of the facility. Total cost of ownership: looks great on a slide, for the facility that already paid for the install. What the slide doesn't show is that the cost didn't get amortized at all in any sense that matters for a product meant to work in the wild. It got relocated to a site, and every new site pays it again from zero. A hundred warehouses means a hundred separate instrumentation jobs, a hundred separate calibrated maps, a hundred rooms where the robot works and zero rooms outside them where it does anything at all.
That's the actual cost that was doing the work the whole time: not the robot's bill of materials, but the recurring, non-transferable price of wiring each new room before the robot can set foot in it. It never shows up as a line item on the robot's spec sheet because it isn't the robot's cost. It's a tax levied on whoever wants a second site, and it resets to full price every time.
Compare that to the harder, more expensive bet: put the sensing and the modeling on the robot itself. SLAM, onboard cameras, enough compute to build a map of a room it has never seen. That robot costs more to build and is worse, at first, than the instrumented one in any single fixed facility, because it's spending compute re-deriving something the instrumented robot was handed for free. But the capability it builds is portable. Drop it in a new house, a new warehouse, a construction site that didn't exist last month, and it degrades gracefully instead of going blind. The cost front-loads onto the robot instead of back-loading onto every room it will ever enter.
Neither bet is wrong on its own terms. A single warehouse that will run the same floor plan for a decade should absolutely instrument the floor; that's the cheapest intelligence per dollar you will ever buy, and it's not pretending to be anything else. The mistake is reading that warehouse's economics as evidence about robots generally, when what it actually demonstrates is the economics of a single, permanently fixed, fully owned room. The warehouse isn't proof the robot is smart. It's proof the warehouse made itself legible to a robot that can't read anything it hasn't been shown first.
The tell is simple: ask what happens when you unplug the robot from this one building and carry it to a different address. If the answer is "it stops working," the intelligence was never on board. It was bolted to the room, and the room is the thing you were actually buying, robot included.