Robots are not very good at picking strawberries yet

https://www.npr.org/sections/thesalt/2018/03/20/592857197/robots-are-trying-to-pick-strawberries-so-far-theyre-not-very-good-at-it

by CrocodileStreet • 8 years ago

98 87 8 years ago

npr.org

synaesthesisx• 8 years ago

The issue seems to be that fruit is often obscured by leaves/plant matter - why not use ultrasound or something to help locate the strawberries?

mythrwy4• 8 years ago

My thoughts exactly. I've always though the whole artificial eye thing was often unnecessarily limiting the range of perception (and thus capabilities) to what humans are accustomed to.

Maybe part of the answer is in the form factor of strawberry plants. Like olives for instance. They grow olives now in windrows of little bushes, like a low hedge and harvest them with a machine that straddles the rows. Which obviously is a lot more efficient then getting them out of a tall tree.

craftyguy4• 8 years ago

You're comparing harvesting from a tree/small tree to harvesting from a vine that, by nature, always grows very close to the ground. Many plants in strawberry fields are maybe a foot high (often less), and very dense.

mc322• 8 years ago

I can't really tell from the video, but they would, in addition to knowing a strawberry was behind a leaf, need to know if they were ripe or not --so they would have to push leaves out of the way to tell visually via color if they were ripe; unless there are other signals that could be used. Basically, "presence" is not equal to "ready for picking".

jonknee• 8 years ago

so they would have to push leaves out of the way to tell visually via color if they were ripe

That's also what people do, seems to work OK (I grew up in Florida and did lots of u-pick strawberries).

orbital-decay• 8 years ago

So a hypothetical strawberry variety that ripens simultaneously would help this by eliminating this step?

Also, ripe strawberries are quite different in mechanical properties from the green ones, I imagine that ultrasonic methods could possibly be useful for this too.

AndrewKemendo2• 8 years ago

"Hey, it could happen! Put a man on the moon, didn't we?"

I think it's funny cause as a ML engineer I can totally understand why it's in many ways easier to get people to the moon and back than to solve dynamic leaf occlusion and object segmentation at scale.

craftyguy4• 8 years ago

In other news, we don't actually have the technology anymore to put a man on the Moon. Hence why we are having to redevelop it all over again.

AndrewKemendo2• 8 years ago

Having grown up around Johnson Space Center and the decommissioned Saturn V on display, I think this is more of a "technically true" misnomer. It's true in the same way we don't have the technology to build a 69 Camaro anymore because we dismantled it and have no desire to, not because we have lost the technical capability.

throwaway84742• 8 years ago

So true. I think of technological capability as of a cultural artifact these days. In the sense that you can’t really just start making something sophisticated right off the bat, you have to build the culture that enables you to do it, slowly, bit by bit. You can certainly try to cargo cult, but you won’t really get anywhere if you don’t have the intangibles.

raverbashing3• 8 years ago

Meh

A big rocket will get you there.

The tricky part is getting back. Still, in some aspects, easier than going

Also easier when you don't have to do all the calculations by hand

V-2• 8 years ago

Textbook Moravec's paradox

ohiovr• 8 years ago

How about a strong fan that blows the leaves around. Something red shows up, pick it.

NathanWilliams• 8 years ago

I find it amazing that with all the progress around us every day, people still think there is an impossible line automation can’t cross, and that their job is safe.

Even as a programmer, I imagine a lot of coding work being at the minimum offset by automation.

bamboozled• 8 years ago

What are people supposed to do though ?

adrianN2• 8 years ago

A lot of coding has already been automated. We don't write programs by manually punching the bits in anymore and programs that once were a major project are now homework exercises for students.

Luckily the demand for programming has so far outstripped any productivity enhancements.

bjelkeman-again2• 8 years ago

Not only is it automated to a degree, but a lot of the tech stack is shared so we can achieve a lot more quicker. But it still outstrips demand by far.

manojlds• 8 years ago

The article actually says that in this case there is a labour shortage.

coldtea5• 8 years ago

I find it amazing that with all the progress around us every day, people still think there is an impossible line automation can’t cross, and that their job is safe.

Why don't have any contracts signed with nature on infinite progress. Or that any desired progress will be timely.

We might never be able to pass some local maxima.

Or we might go extinct, or go back to barbarism after a few good wars of famines, before we get to solve some problems.

Or it might take 50 or 100 years in the future to do so.

We still don't have personal robot (with bodies) assistants, flying cars, space colonies, or general AI -- all things that were considered (by experts nonetheless) "just around the corner" in the 60s.

And don't get me started on mass market holographic storage and memsistors...

oldcynic• 8 years ago

For at least some of those things just around the corner, they could have been here, tax cuts and the occasional vastly expensive defence gewgaw were chosen instead.

onion2k• 8 years ago

We still don't have personal robot (with bodies) assistants...

The vision of a having a robot assistant just shows the lack of imagination that people in the 1960s who were making these predictions. They couldn't see further than a mechanised version of what they already had. On the face of it a robot is a replacement for a physical person, but the reality is that the physical person bit is what's unnecessary. Automation of processes is largely about removing as much of the 'physical' bit as possible. What was needed was to distill the tasks down to the absolute minimum and then automate that, and that's what we've got now. We automated the physical side of posting a letter and now we have email. We automated filing by using databases instead. Scheduling meetings is now done using shared calendars.

Automation of jobs is not about making a machine that does the job of a person. It's about removing the person and building systems to replace the processes that person did.

We do have robot assistants. They just don't look like robots.

crusso3• 8 years ago

As you point out, the future is a strange beast to pin down.

It's amazing how people are ready to throw capitalism and scientific advancement out of the window because of where advancement might be headed at some distant point in the future.

Let me know when unemployment shows a trend of rising over the long term because people can't find new lines of work. Then let's talk about putting UBI or other draconian measures in place.

pjc50• 8 years ago

There were a lot of abortive claims of this round about the start of the 1980s with the "4GL" revolution. e.g. https://stackoverflow.com/questions/1293278/what-became-of-t...

tabtab2• 8 years ago

Re: Even as a programmer, I imagine a lot of coding work being at the minimum offset by automation.

Most CRUD applications could be about 90% defined by filling in data dictionaries (tables describing tables). However, trends & fads, mostly UI fads, keep mucking up the details of the implementation. Sometimes UI-related trends are good, but a good many are just wasteful fads. Bosses, users, and customers like shiny fancy dancy new things. Vulcans would puke.

Fads are great for creating jobs because they keep moving the goal-post: often being different JUST to be different; but fads are lousy from a pure efficiency standpoint. Humans are still apes who follow the shiny red things.

forkLding• 8 years ago

Automation on farms would be a necessary move, its back-breaking work that you don't get to experience that is totally different from the u-pick experiences, they grow bushes very small (at about 1/8th the size of bushes you see in movies) with lots of fruits so that the bushes mature very early and need minimal space, all of which make picking it up very tiring, generally speaking the only people picking with me are either very nice "immigrants" or people I wouldnt want to walk home in the night with.

I once did it for some quick cash during my pre-highschool yrs, I quit very fast cos it makes much more sense just to wait tables for the same amount of money.

westmeal• 8 years ago

Why not just probe around the plants and look for red against the green and brown backdrop

coldtea5• 8 years ago

I'm pretty sure that's the very first thing that came to their mind...

ksk• 8 years ago

Why do you think someone can't solve fundamental domain problems by thinking about it for 10 seconds? :)

Animats• 8 years ago

Agrobot has a better strawberry picking robot in test. [1] Abundant Robotics has an apple picking robot in test. Not yet in production. Getting close, though.

Robotic weeding is now commercially available. Vision systems target the weeds, skipping the wanted plants. The weed is hit with an overdose of fertilizer, a laser, or just a powered hammer. Deere and Siemens both have machines for this.

It's coming soon. A cell phone has enough compute power for this.

[1] https://www.wsj.com/video/agrobot-automates-the-work-of-berr...

Qworg2• 8 years ago

Does Agrobot do well in standard fields? It looks like the testbed is a very fancy raised bed system.

Is the Abundant Robotics robot faster now?

jdmichal2• 8 years ago

Strawberries are typically grown in somewhat raised beds, even if it's not as fancy as the pictures I'm seeing with the Agrobot. This is what a typically strawberry field around Plant City, FL looks like:

https://www.adrianasbestrecipes.com/wp-content/uploads/2013/...

205guy• 8 years ago

Article did not mention whether the autonomous berry picker crushed a strawberry by accident for the first time.

tabtab2• 8 years ago

It indeed was not clear about the specific problems. It basically implied that it shouldn't pick all strawberries, just a subset, probably the most ripe. But it didn't describe well what the bot keeps getting wrong.

I suspect people use tactile feedback in addition to sight, but that is difficult for bots to get right because it requires lots of contact sensors, lots of processing, and lots of coordination with vision to tune well (at our current stage of AI). But it does sound like it's getting incrementally better such that the bots' time will eventually come. I wonder if a spectragram could reduce the need for tactile sampling? That's something a bot has up on humans.

m-i-l• 8 years ago

Rather that developing robots that fit into existing strawberry farms which are designed to make it easier for humans to harvest, they should be developing new types of strawberry farms which are designed to make it easier for robots to harvest. That could be combined with the indoor farm model too to bring other benefits, such as allowing all-year-round harvests, potentially bringing the food closer to the consumer, possibly assisting pest control, etc.

But yes, I've often thought strawberry harvesting would be a really interesting (and challenging) AI and robotics problem to work on.

pbhjpbhj• 8 years ago

existing strawberry farms which are designed to make it easier for humans to harvest //

All the farms I've seen in the UK have just been fields, making harvesting very arduous work (that farms near where I grew up imported Eastern European labourers for).

What's the design you're talking of here?

jdietrich• 8 years ago

A large proportion of strawberries are now produced in polytunnels using the matted bed or table top system. The former allows strawberries to be picked from a seated position from a wheeled trolley; the latter raises the strawberries to about chest height. Gross yields can be slightly lower and there's a significant capital cost for table top production, but it can work out cheaper overall due to easier organic pest control, higher quality and reduced husbandry and picking costs. We're likely to see a considerable shift towards table top production as brexit starts to bite.

fredley3• 8 years ago

"You could afford to give people a day off, you know. Afternoons off, holidays off. If you have machines behind you."

What total bullshit. No advance in mechanisation has ever resulted in this happening. What happens is that the workers get the days off, but it's not holidays, it's being unemployed.

jstanley• 8 years ago

In the short term it's unemployment, in the long term it's freeing up those people to work on something more useful than picking strawberries.

sandworm1012• 8 years ago

Those out of work for the decade(s) while the economy retools wont see it as short-term. And the 'freeing' of labour in one place drives down the cost/wages elsewhere. The more useful thing in the short term will be things that previously nobody ever wanted/needed to do.

fredley3• 8 years ago

The problem is that the number of jobs that a completely unskilled person can do better than a robot are rapidly diminishing, and there already aren't enough to go around. To achieve this utopia, we need to invest in much better education for everyone, and we need to do it 30 years ago at least. An unskilled labourer made unemployed at 40 or 50 has nowhere to go.

220V_USKettle• 8 years ago

Like the outcome of Eli Whitney's invention of the cotton gin?

sandworm1012• 8 years ago

There are a few areas where automation made the workers life easier without replacing them. Aircraft autopilots reduced workload without eliminating pilots. Same too for many hand tools that a made carpenter's life less painful without any concept of replacing them.

ArekDymalski• 8 years ago

No advance in mechanisation has ever resulted in this happening.

Can you point to some source of this information? I'm asking because when I look at the data I see that for example industrial revolution was correlated with: 1. Reduced length of a work-day. 2. Increased lifetime expectancy as well as overall quality of life. 3. Elimination of child labor.

And all of this happening while general population was growing.

I just wonder why this could be different?

dustinmoorenet• 8 years ago

The 40 hour week was not a gift of tech. It was a hard fought labor struggle. And, if you haven't noticed, we haven't moved that needle again after introducing computers.

1. https://en.wikipedia.org/wiki/Eight-hour_day

ivanhoe• 8 years ago

Those benefits came later, and slowly, after a lot of struggle for worker/female/child rights. Initially you had kids working 10h a day in coal mines because of the industrial revolution, steam machines needed the coal to run. Many people lost their jobs. There was of course also many great benefits, like cheaper textile making clothing much more affordable. With time the work force adapted to the new conditions, found new jobs, factories opened, and slowly the benefits outweighed the horrors of the transition.

stult• 8 years ago

Work day length actually increased dramatically during the early industrial revolution. Early factories frequently required 10-14 hour shifts, seven days per week. It was only the labor movement pushing back that resulted in a shorter work day and week. Even then, it's unlikely that the current work day or week are significantly shorter than they were in pre-industrial Europe. Meaning the labor movement really only brought us back to something resembling a reasonable baseline. Even if pre-industrial agricultural laborers worked longer days or weeks during periods of maximum activity, they did so seasonally, with long periods of relative rest between harvesting and planting seasons.

Ultimately it is political conditions, rather than technological, which determine typical working period lengths. For example, despite relatively little difference in technology or economic activity, African-American agricultural workers in the US South shifted to dramatically shorter working periods following abolition. Only the political conditions of slavery enabled slave-owners to compel a longer working day than would normally be tolerated by workers.

And while technology can increase the amount of surplus wealth generated, it can't change the distribution of that surplus (at least not on its own). Factory work increased the value per hour produced by laborers but it took labor organization and mass strikes to shorten the work day. It took women's suffrage and widespread moral outrage to stop child labor (which was far harsher in early factories than it ever was in agriculture). It took government intervention via social security and medicare to extend the life-expectancy of average laborers (in the US, though the equivalent programs in other developed countries had similar effects).

Basically, employers aren't going to give time off or other benefits just because they want to or just because they can afford to. At least not in any great numbers. They're going to do it because political or market conditions compel it, such as when Henry Ford increased the pay of factory workers dramatically to prevent them from working for his competitors. But market conditions are stacked against the laborers getting greater benefits in this situation. Employers are substituting robots for their labor and so their labor is less in demand and thus demands a lower price (benefits + pay). So with increased mechanization and automation, the average laborer can't demand more benefits in the market while still maintaining wages. Barring political/legal pressure, employers will simply pocket the surplus from this situation rather than charitably donating it to their workers in the form of PTO or other benefits.

It isn't necessarily bad that workers get laid off, because that frees them to pursue more remunerative or fulfilling activities. But we can't rely on employers to help workers transition after disruption in the labor market, just as we couldn't rely on employers to set reasonable working days or to not hire child laborers.

travmatt• 8 years ago

How are you reaching those conclusions? If I recall history correctly 1 & 3 were the products of a generations class warfare and more than one massacre at the hands of industrialists. The labor movement rose in hand with the industrial revolution, but it’d seem a perversion to portray those gains as some inevitable beneficence of increased mechanization.

rsuelzer• 8 years ago

Thank you! Even as a Keynesian, I have to laugh when I read him talk about the leisure society.

ThomPete2• 8 years ago

"You could afford to give people a day off, you know. Afternoons off, holidays off. If you have machines behind you."

One of the major issues I have with discussions around automation is that they always look at it from either the point of economics or the point of technology.

Those are important perspectives but we tend to look back in history to determine what will happen in the future.

If technology keeps progressing and economics keep making it cheaper and cheaper to do more then the real question is where does that really leave humans. Our advances as a species is by the invention of new technologies which has allowed us to store more and more knowledge/information/data from trial and error in the past while at the same time allowed us to spend less and less time on surviving. In other words, human ingenuity and advances are based on the technology.

What many seem to forget is that once AI learn to pick strawberries it has learned a lot of other things too which can be applied to other industries/tasks and it's instantly applicable (no need to retrain workforce). In other words each step forward isn't isolated to the area that the AI is learning about it allows for the advance of technology in a number of areas.

And so here is the problem.

Human progress has historically followed the progress of technology which was fine as long as it made many things easier. Human ingenuity helped us keep adapting and venture into new areas. But what we are increasingly facing is technological ingenuity.

A good example of that is: https://www.linkedin.com/feed/update/urn:li:activity:6368815...

In other words, the one thing we had over technology doesn't even seem to be out of it's reach cause while technology is progressing we aren't.

dzdt• 8 years ago

Where does it leave humans? In about the same position as horses when the internal combustion engine started to become popular. Seriously. [1] http://www.cowboyway.com/What/HorsePopulation.htm

ThomPete2• 8 years ago

Exactly. Horses met their mechancial limits in competition with technology just as humans will meet their intellectual limits.

Robots are not very good at picking strawberries yet

Related Stories