Goog produced many videos vaguely describing how their self driving car worked, in 2013. Then, they stopped discussing how it worked & focused more on philosophy. No matter who the employee is, whether it's a marketing guy, middle manager, or an engineer, they all show the same slides, show the same videos, read the same lines. It's a perfect corporate machine.
They reveal hardly anything about how it works. The mane nuggets are it uses GPS to get a rough position fix. Then it matches the current LIDAR image with a 3D map of the world to get the exact position. LIDAR detects landmarks & pedestrians. Radar detects metal objects like cars. They have to manually drive every route the computer drives, to make a 3D map of the world, manually entering all the traffic rules, all the lanes, all the traffic light positions.
It doesn't use vision for anything besides traffic light color. They made no mention of vision detecting lane markings, unlike lane keeping systems. Lane keeping systems must have proven too unreliable for a fully autonomous car.
A path with no discernible lane markings probably needs a similar algorithm involving GPS & a database of known features. It might work with a database of scanline intensity profiles, since LIDAR is too expensive.