GPS field striping robots - How do they do it?

My use case for Rover is to paint stripes on grass. I’ve done a lot of research and have a prototype built. I’m not sure how it took me this long to find them, but there are commercially available units on the market with centimeter accuracy for at least 4 years now.
Has anyone seen these? Or have any information on how they are getting repeatable hash marks on a football field, for instance?
See:
https://pioneerathletics.com/tinylinemarker

GNSS with RTK is the same kind of setup that is used for surveying. With the correct set up the day to day repeatable accuracy is less than the size of a dime.

That’s what I was thinking as well, and what I’m working towards with my robot. It’s just interesting to me that these guys have been painting repeatable lines on football and soccer pitches for years, and we’re still trying to figure out PID tuning and overlap 1/3 of a mower deck per pass.
I wonder if it’s fully closed source for the commercial units…

Are we sure he is not manually plotting the course?

Probably is but w/o RTK it wouldn’t make any difference. I have plotted Auto Missions using Save Waypoint from actual positions and the next day, or the next mission, it will run into something. I do have free Ntrip RTK reference data here in Michigan but haven’t taken advantage of it yet.

I’m sure they are. Sports fields are standardized, so there’s probably some sort of preloaded waypoint list for football or soccer. One would only need to orient the program to the field.
I’m more curious about the hardware: What they use for motor control and guidance. The reliability has to be spot on to sell these things to any random high school grounds keeper.
Edit: found a press release on the Turf Tank. $35k for the robot, tablet and “ground station”. I’m still curious if it’s built on open source code or developed in house.
Sure would be nice to look under the hood…

With low speed and weighty units you could use wheel odometry to get enough accuracy. Using some sort of beacon (landmark, etc…) and you will have a quite accurate system. But that work only because the mission terrain is known !