As someone who does farming (part time), I don't see these being practical. For low value bulk crops (corn, wheat, soybeans) the robot is too expensive and existing mechanization is too good. Even for high value stuff (grapes and other fruit) it will be too expensive. Maybe inside a highly controlled greenhouse this will be practical, but there are no food crops that are economical to grow inside a greenhouse. (It is getting there as high-tunnel becomes cheaper to install.)
As an aside, mechanical harvesters for grapes are hilarious. You still need people to run them. One person's job is to find and remove possums from the fruit bin before the critter is buried by a thousand pounds of fruit.
It is a distraction to talk about the gantry, XY or radial. Any gantry based system is going to be a headache from the start because you've got those rails cluttering up the space. Eventually you'll find something the robot cannot do and then you'll be trying to figure out how to get your general purpose tractor around without crushing the gantry.
The radial system is cute, except now your installation and operating costs scale linearly with the amount of crops. With an XY system, one can plant more and extend the rails and have the existing robots do more work. Not possible with the radial system, unless you are going to physically pick up and move the units between circles few times each day. If time is tight (plague of locusts coming in tomorrow, need more robots to squish them) you can increase the robot density by sliding more robots onto the XY gantry. Not possible with radial.
Ultimately it will be about the vision processing and manipulators. And then you might as well mount the vision/manipulator unit on a tractor and have a robot-tractor drive up and down the fields, just like any other piece of equipment. Vastly cheaper and more flexible than gantries.
Eventually we will get there, but it probably won't be until vision processing and lidar gets a whole lot cheaper. We might be on the cusp though, based on the computer vision advances in video gaming.
User roryaronson, I've done a lot with robotics and outdoor electronics too. If you want I can give you a fairly lengthy piece of constructive critism about your whitepaper. You've got a lot of good ideas in there, but just as much misinformation.
Your post is the epitome of how domain-knowledge trumps idealism.
I look at this and instantly think "Dirt. And water." Dirt and water will destroy their "machine." Whoever conceived it doesn't respect that farming involves dirt. And water.
As you remind us above, THE TRACTOR IS THE ROBOT. The tractor is safe from Dirt and water. Unless this is some desktop thing. Even as a gantry system, their rendering is naive. ISHYGDDT.
As someone who knows nothing about farming or gardening, I'll defer to the points you made as they sound very sensible. But as a software guy with an interest in salt-of-the-earth hobbies, I'd love to play with one (if I had a backyard that is). The thought of a robot that runs on a seasonal calendar that automatically harvests fruits and vegetables for my salad is too cool. To peer out your kitchen window and see a robot till the earth and harvest your small garden to feed you and your family is straight out of science fiction, the optimistic type.
That said, having a high tech piece of equipment with a bunch of moving parts outdoors does seem quite problematic and headache inducing. Here's to reaching the cusp fast....
> But as a software guy with an interest in salt-of-the-earth hobbies, I'd love to play with one.
Like you said, high-tech outdoors is troublesome. If you want to get a head start on this sort of thing buy an outdoor robot chasis (or make one if you can hold a screwdriver, it'll save you a grand or two). Throw on a kinect, all the cameras you can fit and all the lidar you can afford.
First, teach it what rain is and how to quickly get back to the safety of the charging port. Second, make it navigate around the garden without destroying the plants. Finally, teach it what to look for (pests, damage, ripeness). Don't bother giving it manipulators, just have it log findings and generate reports. When you get home, it can tell you exactly which plants need attention.
If this device was remotely managed and you had access to the sensors mentioned in the white paper then the tasks above may not be that hard for smaller plots of land.
Teaching it what rain is? How about a query to Yahoo Weather.
Teach it what ripeness is? Imagine this device taking pictures of each plant every day and emailing individual images to you. Start out by manually training on the corpus based on what you think is ripe or not ripe. From their you would have a good stepping stone to start machine learning based on color and size of the plant. Same goes for unidentified plants in the garden that may be weeds.
I think the idea of total automation and remote sensors could make managing a farm a one person job.
You should read PG's column about doing things that don't scale.
Yeah, this will never be practical. But most of the building blocks that go into it are thinks you would need to automate farming. And it is interesting that someone is taking a stab at it.
"there are no food crops that are economical to grow inside a greenhouse"
The Netherlands has so many greenhouses they can seen from the space station without a telescope, and many of them grow food crops (tomatoes and paprika being the largest two)
I can't read the whitepaper due to work network restrictions (scribd? really?) but superficially this seems more directed towards DIY-er gardener types rather than industrial farming. That is to say, people who have no need for a giant combine harvester.
For that purpose it seems to have the capability to automate some tedious jobs quite nicely. I would pay for a machine that grows vegetables in my backyard with little to no input from me.
That said, the possum remover must be the most hilariously bad job in farming.
They've probably fixed the possum thing by now :-)
Your comment makes me wonder what a good pricepoint would be for such a system. How much money and how much space would you be willing to give up, for example.
Less hypothetically, why the fixation on machines? We've got all sorts of crazy services these days. How much would you be willing to pay for a gardening service that maintains your garden for you and occasionally leaves a crate of produce at your backdoor? It would be like a lawn service, but without all the terrible effects on the local ecology.
I let kale and chard go to seed last year. I've had more kale and chard this year than I can harvest. Almost all of it volunteer. Also some other random greens that make a quite nice salad.
On the other hand, I've put in a ton of effort on tomatoes, and gotten very little to show for it. The climate just isn't friendly to them.
As someone who lived and worked on farms, I agree that they're not practical right now. That said, I'll take on some of your other points.
"there are no food crops that are economical to grow inside a greenhouse" - there are many organic lettuce/greens growers that put lie to this statement.
I don't like the gantry system, mostly because what you stated. Also because it is _extremely_ expensive vs. other methods. If you're doing that, you might as well just put it on wheels and have it drive around.
The radial system can still run more robots - I don't see any restrictions past weight on the arm. It is the equivalent of a circle sprayer, just with robots instead of water.
As for cheap vision/LIDAR, it is most certainly not driven by games. The big Deere tractors are moving away from LIDAR towards sets of stereo cameras because the algorithms are getting better and the cameras are much cheaper and more robust than LIDAR.
I see this device fitting a niche in "black greenhouses" where there aren't any humans involved at all. Totally recycled, totally hermetically sealed environments, like a clean room for plants.
Regarding greenhouses, any example you cite I can say is way out of economical norms ;-) But the only way to make organic work (without throwing away half the produce or drowning the field in the few allowable chemicals) is to either do it indoors or in a part of the world where the pests don't exist yet. But that is neither here nor there.
How do you get the extra robots installed on the radial systems? With gantry and wheel based systems, you plop them down at one end of the field and then they re-arrange themselves.
I was specifically thinking the advances in time-of-flight LIDAR that seem to be for short range (several feet) gesture recognition. Not for object avoidance (Deere's use case) but on the wrist of a manipulator, where you don't always have the space for a stereo-optical setup.
One day hermetic greenhouses might make sense. Until then it will be cheaper and (arguably) less impactful to grow the veg someplace with reasonable weather and pests and ship it instead.
Okay, that is pretty awesome. Some thoughts... greens are probably optimal for this because they are fast to go from seed to produce. And 100% of the output can be sold, with basically no wastes. Singapore is also optimal for greenhouses. The temperature there is 23C-32C year round, so you don't need HVAC (just air filters on your vent fans) and you can grow continuously.
The waterwheel thing is clever not for the energy efficiency (it'll be the same regardless) but because it greatly reduces the amount of mechanics/electronics that are exposed to the greenhouse humidity, greatly improving reliability. (Installation cost is probably the same as if you had a gearmotor on every tower, just because the funny waterwheel is probably not an OTS part.)
But even in the video they call it a "premium novelty for consumers", which basically means not yet economical. And who knows how much of it was backed by the mentioned incentives. (They show some prices at 5:12 and 5:41, maybe that can give an idea.)
Yup. I went through the site and I still don't know what problem it solves. Modern ag equipment is scarily efficient and fast. The farmer who's just driving into the field when I leave for work in the morning is towing out carts piled full of corn by the time I get home in the afternoon.
As an aside? How to mechanical harvesters for grapes work without breaking/getting tangled in the wires supporting the vines?
Quality is questionable, as it will also harvest the unripe secondary clusters and any rot. (There are probably visual systems for cleaning these out.)
Grapes (and a lot of perennial fruits) are a little oddball in this discussion though, because they are very low impact. For example, you rarely (maybe once when first planting) have to till or fertilize or spray for weeds and we've got plenty of biodiversity in the turf.
As an aside, mechanical harvesters for grapes are hilarious. You still need people to run them. One person's job is to find and remove possums from the fruit bin before the critter is buried by a thousand pounds of fruit.
It is a distraction to talk about the gantry, XY or radial. Any gantry based system is going to be a headache from the start because you've got those rails cluttering up the space. Eventually you'll find something the robot cannot do and then you'll be trying to figure out how to get your general purpose tractor around without crushing the gantry.
The radial system is cute, except now your installation and operating costs scale linearly with the amount of crops. With an XY system, one can plant more and extend the rails and have the existing robots do more work. Not possible with the radial system, unless you are going to physically pick up and move the units between circles few times each day. If time is tight (plague of locusts coming in tomorrow, need more robots to squish them) you can increase the robot density by sliding more robots onto the XY gantry. Not possible with radial.
Ultimately it will be about the vision processing and manipulators. And then you might as well mount the vision/manipulator unit on a tractor and have a robot-tractor drive up and down the fields, just like any other piece of equipment. Vastly cheaper and more flexible than gantries.
Eventually we will get there, but it probably won't be until vision processing and lidar gets a whole lot cheaper. We might be on the cusp though, based on the computer vision advances in video gaming.
User roryaronson, I've done a lot with robotics and outdoor electronics too. If you want I can give you a fairly lengthy piece of constructive critism about your whitepaper. You've got a lot of good ideas in there, but just as much misinformation.