Mr Red: in my experience GPS that is commercially available is barely accurate enough to taunt at the control tower window.

Not at all correct. We use UAVs with RTK GNSS which are accurate to centimetres. We use "Mission Planner" to programme autonomous flights for our 1.5 metre wide drones along flight lines which are only 1 metre apart. They successfully navigate a whole swathe of lines of this separation to, usually, within 20 cm of the pre-plotted lines.

There is a range limitation on this though, of a few km, subject to the radio link range for the RTK correction.

For other applications we use PPP (Precise Point Positioning) DGNSS (I call it GNSS rather than GPS because even many lower-end commercial systems now combine US GPS, Russian Glonass, Chinese Beidou and, soon, EU Galileo into a single "best solution). The corrections for this mode of operation use global networks of stations to calculate corrections and transmit via satellite to the receiver. These systems are typically accurate to within 1-2 metres in a dynamic environment. Much better in a static environment.

As far back as 1986 I assisted in re-triangulating a bunch of points in northern Brazil using the 25 minutes of 4 satellite coverage per day then available with GPS. We were getting precisions of +/- 2mm plus 1 part per million of the baseline distance between stations, even over 30 years ago. That's better than 20 cm relative positioning over a 150km baseline distance.