Not the least of which is velocity. Tractors generally move under 12mph. And traffic density. The tractor is the only vehicle in the field, as a rule. And obstacles - there is absolutely nobody out in the field, usually. And noisy environment - a field is often table-top flat and empty of all obstructions. Kind of the definition of a field.

So the John Deere problem space was mostly dead-reckoning navigation to a matter of inches, to get crop spacing right. A very different problem from auto-driving cars.

> And noisy environment - a field is often table-top flat and empty of all obstructions.

Perhaps this is true in certain parts of the country, but certainly wasn't my experience growing up in the Southeast. The land there tended to be mostly rolling hills, basically no field was totally table-top flat. (And the 3rd paragraph in TFA mentions that the land there even in Kansas is hilly so he does the first pass).

Obviously you want few obstacles to maximize the useful land area, but fields often had ponds, creeks, patches of trees that broke things up, not to mention unknown obstacles like fallen trees or fences, wild animals, etc. that happen 'randomly'. I'm actually curious how much of this the software can handle on its own.

So it's certainly a different challenge, but to imagine most fields as perfectly empty, flat, continuous, homogenous spaces is to over simplify things.

Pardon my ignorance. I'm here in Iowa where many fields are most definitely empty, flat, continuous and homogenous. Not to mention very fertile, well watered and yielding among the best of any place in the world. Also the home of agricultural research including, I'm guessing, John Deere pilot projects.

Which biased my view. Interesting to know the technology can handle Kansas, which would be a challenge.