To be fair, they also started working on the problem before Google even existed, with technology more primitive than Google's first servers.

Yes, I know of professors who were doing field trials of self-driving tractors 30 years ago, and I'm sure they weren't the first.

The mining industry has dreamed of self driving vehicles to replace humans in dangerous work conditions for ages.

Nothing new here : http://mining.sandvik.com/en/products/equipment/mine-automat...

It is, although there are issues that self-driving cars don't have to deal with, like muddy conditions on hills, lining up rows (tricky on head rows), speed-matching during picker unloading (e.g., tractor and truck working together to evenly-distribute the unload), etc.

Self driving cars absolutely do have to deal with varying friction of different road surfaces and conditions, lane alignment, speed matching (or rather, not bunching), overtaking. There is a lot of overlap in the tasks. The only one you mention that is unique is unloading while picking.

Yes, but traction on pavement is quite a bit different than traction in mud on a 20deg slope with a Really Heavy Vehicle. There are varying degrees of "uniqueness". My point, which I think you deliberately "missed", is that looking at it from the "layman's perspective" is flawed.

(And lane tracking is generally more forgiving than row tracking, (except for on the Garden State Parkway) in the sense that a few inch deviation while driving is acceptable, but cumulative parallel row error can add up to quite a bit, both time and money.)

On the other hand, roads are designed for cars. In fields, it's hard for a computer vision system to distinguish between a thicket of brown grass (can easily drive over) or a pipe (will damage you if you try to drive over.) They're different problems, I don't think one is obviously easier than the other. And there are certainly people in fields.