Figuratively speaking, a robot is a machine with a brain. They don't really have brains, of course; they have processors and programs, which would be depicted on some low-detail flowchart as lying between sensors and actuators. But, for many purposes, processors and programs are as good as a brain, in some cases even better. If you can accurately describe, at the level of detail a computer requires, what a brain would be doing in some particular circumstance, what factors it would be taking into account and what decisions it would be making, there's a good chance you could craft a program that would serve just as well. Load this program into the processor of an appropriately designed machine, and you have a machine capable of handling the job.
Granted that it's far easier to create such a program for a machine intended to vend beverage cans than for one intended to plant, tend, and harvest a mixture of crops in an uncontrolled, nonuniform environment. The latter is harder, but not impossible, nor even so far beyond some other tasks to which computers have already been put, such as handling a complex mix of financial transactions or handling the control surfaces of an otherwise unstable aircraft.
Plants have numerous characteristics which, if not already known, are at least measurable. Soils likewise can be measured and typed. Weather is the most unpredictable factor, but even weather is regular enough to be categorized as a climate, and statistical approaches allow decisions to be made in the face of unknowables, which on average work out well enough. It's a complex context, but not unmanageably so.
Thankfully, it isn't necessary that any such system perform perfectly. It's only necessary, as an initial condition, that it either perform on par with conventional practice while costing less, or outperform conventional practice while costing about the same.
This becomes easier with each passing year, as the cost of diesel fuel and other petroleum-based agricultural inputs steadily rises, and as the power of computing equipment and the sophistication of programming tools improve.
The rising cost of petroleum works against conventional practice and for a dramatically different approach utilizing robotics because conventional practice is utterly dependent upon huge amounts of fuel, fertilizer, and pesticides, whereas robotics can make possible an alternative approach without those dependencies.
The alternative approach itself is as old as gardening. Robotics can make this approach a contender by duplicating the knowledge, skills, and hands of the gardener many times over, making it applicable to land areas measured in hundreds or thousands of acres instead of square feet. That's where the distinction between a robot and other types of machines comes in. A robot is a machine with a brain of its own. Robots can operate autonomously, meaning there doesn't have to be a human operator present and you don't need as many human operators as you have machines.
That's an essential point. It's the one-to-one correspondence between operator and machine that has shaped conventional practice, exerting upward pressure on the size and power of the machines as economics exerted upward pressure on the amount of land required to earn a decent income. And large, powerful machines are only really useful for conducting operations that are applied uniformly over entire fields. The result is mile after mile of fields, all planted to the same crop at very nearly the same time.
Breaking that one-to-one correspondence is the key to breaking the dependence on oil, while at the same time dramatically increasing the variety of production and relieving the oppressive monotony that has so overtaken the rural American landscape.
And the key to breaking that correspondence is machines with enough brains to operate autonomously.