A child exploring an unfamiliar apartment crawls from object to object, memorizing what’s visible from each corner. That’s how the brain builds a mental map. The gradCSCG algorithm works the same way: it receives a video stream from a camera and draws a floor plan relying solely on the changing images. Previously, machines needed hints like pre-labeled objects. The new method does without. One part of the program compresses visual scenes into a short code spectrum (a set of distinctive features), and the second, gradCSCG, links these codes to the path taken. Everything learns simultaneously: recognition errors correct the topology, and vice versa. To keep from losing the thread, the system has built-in mechanisms that damp down entropy (confusion) and prevent it from “falling asleep” in monotony. A surprising result: even when the camera showed a handwritten digit “3” or a blob in the same spot, the algorithm confidently identified it as the same location. This is how a standard model (typical layout) of a place is reconstructed. The same principle is used by grid neurons in our brains: their discovery won a Nobel Prize in 2014. The technology promises helper robots that explore unknown floors without GPS, like the BB-8 droid.