There is a lot of active research in this domain but it is hard to make something that is as reliable, usually the goal with those systems is not to allow for lower soil compaction but to be able to work in places where wheeled vehicles can't.
The actual driving part is obviously much more trivial than autonomous driving on roads. However there are a lot of hard to automate adjustments and maintenance items going on with many tractor implements.
As you outlined, uptime is critical during these periods. I spent a lot of my time in remote locations, and reliability trumps everything. The cost of a hotshot is significant - however I was far more remote than the farms where.
Paying a person to drive the tractor doesn't seem like it's the large cost. It's the machinery you need 1x per year, at the same time everyone else does.