‘Ten years ago this was science fiction’: the rise of weedkilling robots | Hacker News Reader

‘Ten years ago this was science fiction’: the rise of weedkilling robots

theguardian.com

123 pointsedward39 comments

>Removing pests, such as aphids, thrips and lygus bugs, is a next step for FarmWise

This would be great. Poisons become ineffective against their intended targets pretty quickly after a few generations, but evolving resistance against being zapped is a bigger hurdle to overcome.

I'm guessing they would eventually evolve towards camouflage/hiding from the zapper rather than resistance to zapping.

Assuming it’s using some sort of ML to detect the bugs, that’s a matter of just inputting a new set of data. Easier to update a set of training images than find new poisons that don’t kill absolutely everything.

Or updating the sensors and hardware, in cases where the camouflage is too good

Or they would start widening their diet to include non-agricultural plants as well. That, too, is a way of escaping the killer robots.

> camouflage/hiding from the zapper

Mirrored carapace. The refection could even harm the sensor.

Someone needs to write a SciFi short about this.

Where do you think I got the idea in the first place?

Bug evolves a pattern that rewrites zapper into ai sentient human zapper

there are already reverse-engineered "pathalogical" geometries that can convince ml that a random squiggle is a car or a human or whatever.it would be interesting to see insects evolve dazzle patterns that fall into some local minima in the ml gradient and are clearly bugs to the human eye but seen as nothing at all by the ml implementation.

The method implied in the rest of the paragraph is targeted use of pesticides, only where they are needed.

If these displace pesticides, that is a big win for everyone involved.

I alway wanted to build a flesh-eating robot for my garden, since I saw it back in 2001:

https://web.archive.org/web/20040129031356/http://www.ias.uw...

Nowadays I have a lot more problems with bindweed.

...

I see in Japan, they are further ahead to Make Room! Make Room!

https://www.popsci.com/technology/article/2011-03/six-robots...

Slugpeople.

Uzumaki anyone?

For weeding, there are not just robots. There are wide implements pulled behind a tractor. These have many cameras looking down at the rows and zapping weeds. Deere sells that. Weed-It has been selling it for ten years.

This just applies the herbicide to anything that's green and not dirt.

That's not nothing, but still the last tech improvement is using Blue LED's because they see green better. Also cool.

It's different to attacking only the weeds, which this technology can't do yet because it's still science fiction for a few more years yet as the headline points out. But it will be cool when it can.

Interesting article, but unfortunately light on technical details. How do these systems distinguish weeds from crops? How accurate are they?

Knowing someone working in one of those companies it's mainly computer vision (based on recent machine learning techniques), and they're accurate for particular conditions only (not working on all crops yet).

The difficulty is on handling edge cases (unsurprisingly), and they're tackling that by collecting a lot of in the field feedback; afaik they send technicians on lot of deployments still.

They look very much like the robotic harvesters featured in Interstellar. I guess Christopher Nolan knows the field.

I wonder how weeds and pests will evolve to avoid this technology. Will we end up with laser resistant plants?

That's biologically impossible, but we might end up with weeds which resemble the crop in new and interesting ways.

This has already happened in history for weeds which came under artificial selective pressure due to practices of separating seeds, etc. The fascinating thing is that some of these weeds took on so many of the desirable properties of the primary crop that they became crops in their own right. Rye and oats originally being mimicks of wheat for example: https://en.wikipedia.org/wiki/Vavilovian_mimicry

I tend to not be a fan of replies that are just "thanks" etc, but I must say here: thanks for your comment and link - that is ridiculously fascinating.

Well, deer and wolves haven't evolved bullet-proof hides either.

The best strategy against such a source of massive trauma is probably not to be seen.

Or to not be present. There isn't much that has evolved to live in the hotter parts of the Sahara. The challenge for any solution to these issues is to make sure it is so pervasive and comprehensive that it is too difficult to evolve around.

Realistically better camouflage vs the associated crop, these robots aren’t doing DNA testing or anything.

Right. You probably get weeds that look identical to the crop — at least for the neural network powering the robot. Interestingly similar to research that tries to fool image recognition software, but carried out by Mother Nature.

Natural selection will lead to weeds that cant be seen by the cameras over time.

We do not really know how the robots will look like in 20 or 30 years.

It is entirely possible that the robots will remember positions of individual crops and kill anything that grows outside the grid.

It is even possible that the robots will be capable of seeing individual cells on the plant and distinguish the species by morphology of the stalk.

Or cameras just sense such a wide light spectrum that nothing can hide against them.

Is "natural" the right word when a robot is involved?

We conventionally use artificial selection to mean only cases where we decided what we wanted. e.g. artificial selection is why chickens and cows are how they are today, likewise most types of dog but not why the type of spiders that live in my home are the way they are even though that is an adaptation forced by man-made environmental changes.

So I think it's crucial here that the goals are external, and it's just natural selection if the goal is merely "Survive and reproduce" and the selecting factors happen to be man-made.

Weeds that evolve to hide from an AI weed zapper thus aren't artificial selection. Unless there's some comic book bad guy who decides to engineer such a thing.

A crucial difference in outcome here is that Mother Nature doesn't give a shit. Where there's a natural equilibrium, Ma doesn't ensure neither side wins. The wolves could become perfect predators, kill all their prey and then starve to death wiping out both groups. Mother Nature doesn't give a shit. But because we have an actual goal, outcomes that defy our goal are unacceptable and we'll tweak the selection process. That's what happened with dog breeds where the "preferred" type characteristics were so unhealthy that they're basically a death sentence.

How much energy does this system use? Powering a 10,000 lb machine and running a laser takes a lot of fuel whereas a few human laborers can do the same job with enormously less fuel cost.

> a few human laborers can do the same job with enormously less fuel cost.

and this is where you're wrong. humans have to live in society. the cost isn't just food. you need to pay for housing, their electrical bills, water, food, entertainment, children, their 2,000 lb machine to get to work...

suddenly that 10,000 lb machine with lasers is pretty cheap.

While weeding by mechanical means instead of lasers, FarmDroid is powered by an onboard solar array.

Yes, fiction as a wish list works for future laborers.

I too have seen the. Classic robot dystopia film Runaway.

I too have seen the. Lassic robot dystopia film Runaway.

T-when till the robots classify humans as weeds and go to work?

> At the same time, they are battling a rise in herbicide-resistant weeds, giving extra impetus to the search for new ways to kill weeds.

Great, so now we will be getting laser resistant weeds...

Yes, this was the joke. It was also true it may happen.

Reply on news.ycombinator.com