Location is infrastructure

Robots fail in space before they fail in intelligence. A sidewalk delivery unit that is two metres off thinks the driveway is the road. A tractor that is half a metre off takes the crop, not the weed. Indoor machines lose satellite lock at the door and still need to know which aisle they are in.

Three different “wheres”

• Global position, improved by correction networks. Centimetre-level GNSS, the kind of service a decentralized antenna network such as GEODNET is built to sell, is about making satellite navigation sharp enough for machines.
• Local position: lidar, cameras, wheel odometry, and maps the robot builds as it moves. This is SLAM, and it lives on the machine.
• Shared spatial memory: a map of a mall, a station or a venue that many robots and phones can use, instead of each one rediscovering the doors. Networks such as Auki sit in this lane — machine perception as a commons, not as one vendor’s private map.

These are not interchangeable. A correction service does not know your kitchen. An indoor map does not help a drone between cities. When a robotics post and a DePIN post look unrelated, check which “where” they are selling. If a machine would actually buy it, it belongs on this desk.