But I do agree with you that identifying weeds isn't that big a deal if it's mass produced. Sure the software will be difficult to design. But once it's done it's done.
But the problem was never the software; given a few weeks I'd probably be able to come up with something that could do a good enough job.
The problem, as outlined earlier, is the issue of computing power - the kind of analysis you would need to analyze plant morphology on the fly is going to take a computer a little bit more powerful than a cell phone. If you want the robot to be able to tell weeds from crops at a reasonable rate of speed (i.e. so you could pick them faster than they grow back) you're going to need a fair amount of processing power; several thousand dollars worth at least - per robot. Plus GPS (or something else to keep them in your fields, and to ensure they go everywhere they need to), plus the cost of the robot itself, plus whatever fuel it uses, plus maintenance, etc.
Hell my cheapy digital camera can recognize poeple's faces.
Yes, but facial recognition is about as easy as morphological analysis gets - you're simply seeking a fairly uniform object (despite our preconceptions, the human face is fairly uniform, with eyes, nose and mouth set in a concrete spatial relationship). And keep in mind that your camera only determines if a face is there. It's not identifying what race that face is, if its a boy/girl face, etc - and those kinds of detailed determinations are exactly what would be needed to ID weeds verses crops.
And plants are much more diverse, morphologically speaking, than are people - even when you're talking about plants of the same species - making an even more difficult issue.
And don't forget, you do a lot of the work for your camera before it recognizes a face - like pointing it at a human, and having that human orientated in a fashion where the face is front-on to the camera. A weed seeking robot doesn't have that advantage; it needs to ID the weeds regardless of the preconditions; even if they are sandwiched between two other non-weed plants.
If it makes a few mistakes it doesn't matter. The existing use of humans for weeding is pretty unreliable too.
We don't use humans for weeding much at the industrial scale, so I don't think that comparison can be made. If plant production is your livelihood, you're not going to be willing to have much leniency. A 1% error rate is a 1% crop loss (ontop of other losses). Given the small margins on farming I don't think many farmers would willingly invest in equipment they know will increase the loss over that of the alternatives. As much as we hate them, pesticides don't cause crop loss, which is why they are so very popular.
Also no need for an expensive robot grabber to dig them out, just spray them, burn them, mow them or whatever.
Fire = bad; I've seen fields accidentally lit by a careless cigarette; a blow torch is out of the question unless you're farming something really wet. Spraying kind of misses the point - the whole purpose of the robot is to avoid chemicals (a laudable goal, IMO). Mowing doesn't get rid of the weed, only slows its growth...
Like I said, its a great idea, but there are a lot of hurdles in the way. And unless you can implement it in a way which is equally efficacious to current practices, and in a format which is financially competitive with the current practices, no farmer on earth is going to adopt the technology.
After all, who is going to buy dozens of robots, at hundreds to thousands of dollars a pop, when they can spend less on pesticide, spray it using their existing equipment, and experience less crop loss as a result?
Its a great idea, and I'd love for someone to prove me wrong on this, but its a far more complex issue that you guys seem to appreciate --AND-- you seem to be completely ignoring the economic angle.
Its like inventing a better mouse trap - sure you can invent better ones, but unless they work as good as, and are as cheap as, the wood/wire ones (3 for a buck at the local dollar store), no one is going to buy them.