The Robots Know Where They Are. The Site Doesn't.

A robot that works in a lab is operating in a space built for it. Put the same robot on a factory floor, an airport ramp, or a warehouse aisle, and it has to operate in a space that was not. One full of people, vehicles, equipment, and other machines; none of it holding still. That is where the demo breaks.

The robots themselves are improving fast. NVIDIA has announced the Isaac GR00T robot, an open humanoid reference design aimed at research labs. Japan Airlines has started a demonstration using humanoids in ground handling at Tokyo's Haneda airport. Bessemer calls it robotics' GPT-2.5 moment: real capability, with a still-wide gap between what works in the lab and what works in the field.

Plenty of that gap lives inside the robot. Perception, manipulation, planning, safety, and training data are all still hard, and better coordinates will not save a robot that cannot recognize a blocked aisle, recover from a dropped part, or notice a person about to step into its path. But as robots move from one-off demos to fleets working real sites, a different gap appears, and it is not inside any single machine.

A single robot can often get by on its own sensors and a map it builds or loads for itself. Even a couple of robots can coordinate directly off each other, the way Figure recently showed two humanoids clearing a room by reading each other's movement. A fleet cannot rely on each machine carrying its own private, inconsistent view of the site. Put 50 robots in a building, and the question is no longer whether each one knows its own position. It is whether the whole site shares one consistent, real-time picture of where the robots, the people, the vehicles, and the equipment are. The site needs a common frame of reference that its robots, and the people running them, can trust.

Outdoors, GPS gives everything a common coordinate reference. Indoors, underground, and in the dense, metal-heavy spaces where this work actually happens, GPS does not reach. The places robots are most useful are the places the default location layer is weakest.

The airport ramp makes it concrete. A humanoid working ground handling still needs its own perception to avoid the crew member who steps around a cart. But the operation also has to know where that robot is relative to the aircraft, the baggage carts, the ground vehicles, and the crews, with enough accuracy and little enough delay to keep safety zones, routing, and handoffs working. That picture cannot live inside one robot's camera feed. Without it, a few feet of uncertainty becomes a robot that stops, reroutes, blocks a crew, or moves into a space it should not.

That shared layer is what ZaiNar builds. We use the wireless signals already moving through a site to place the connected things on it, the robots, vehicles, equipment, and people carrying radios, in one shared, real-time coordinate system, accurate to centimeters in the indoor and underground spaces where GPS fails. It works from the signals those machines already send to stay connected, without adding a location tag to each one. It does not replace a robot's onboard perception or its ability to handle what is in front of it. It gives the whole site a common frame those systems can build on. The same layer locates people and equipment on industrial sites today. Robots are what make it urgent.

The models will keep getting better and the demos will keep getting more impressive. But a fleet only works on a real site when the site knows where its moving parts are. Shared location is not a feature on the robot. It is infrastructure for the floor, and the more crowded and dynamic that floor gets, the less it can live inside any single machine.