Open Source Ecology Development Team, 6 Sep 2017

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Open Source Robotic Tractor

Description

http://opensourceecology.org/wiki/Open_Source_Robotic_Tractor
=====
What you see here at Open Source Ecology is an ambitious program based on a volunteer effort. To help us reach the goals - please consider joining as an OSE Developer in 2017-

http://opensourceecology.org/wiki/OSE_Developers

Take a minute to subscribe to our email newsletter (updates, workshops, etc): http://bit.ly/1LtcM44

A

Okay: let's go right into this meeting. Other fellas we're gonna talk about computer vision for the tractor, so you can feel free to listen to us, but we got to do this right now. So yes, I'll I'm. So so tell me more about your thoughts, so so the update for what we've talked about so far is we've talked about simply adding computer vision, recognition of markers on a field. So our light, the micro tract, can go on a field with solar power, so we would have. We would have, for example, the use case of.

A

Solar panel, small power cube a controller that has solenoids on it, and the solenoids would drive the the tracks at one side or the other tracks, so you can go forward backward and turn and so forth, but but there's logic to that, so it'll be a Raspberry Pi, that's got computer vision, that's actually recognizing markers on the field, so you can either pre program or even well. Ideally, you would want to also have like pre, predetermined paths or both combination of predetermined paths and visual feedback of the path.

A

So, but maybe you can fill me in more on what you're thinking and what the possibilities with your skillset would allow you to do at present.

B

A

Right, yeah, the track: the micro track right now is 41 inches wide and it's it's about 52 inches long. That's the current size and it's about 38 inches high, so 41 by 52.

B

A

B

A

About a meter.

B

B

So the accuracy of that.

B

Size, so basically we we're done about the branch. So yes, that's.

A

Correct: okay,.

B

So this will make it very easy for the turning point, so we can do the turning very accurately with the dropper scope and with the z-axis, and we can make it follow a certain trajectory with with that pointer being on the town. So it shouldn't be a big big issue to find that to find that mine, especially if we get it to be. Oh, maybe discover so that would be lets say which is already done with the TV, so we can just take it out of pocket there as verify and work like a charm.

B

So it's nothing of a huge group that we need to do over there. We just people identify anything I, don't know pretty much nothing. This correctly. What you mean by water is like a straight line on the ground, working that this is where it's real.

B

This is give me the free take off. Is that true.

A

Yeah there's different ways to do that, so let's actually start a document for how what would be the best strategy to do it. One is steaks that are put in every three meters or so say, PVC steaks that are in a long line of maybe.

C

A

Three meters or every five meters and the rows would be say about a hundred meters or so and then the field. So let's, let's actually document and talk about what would be the best strategy, an easy way to do. Accurate motion.

B

Just to make sure that that is doing the mapping in the right way now we're looking the combination that you talked about, that we need to determine.

B

What I believe is yes, it should have a rough estimation of where these markers should be learning to be verifying. Those coordinates, as it is, moving so trying to match the accuracy of the map. That has a point with the actual which.

A

A

Yeah, okay, take please click on the document that I shared their the presentation, template.

A

So I think the prototype field that we have available for testing is a hundred meters by 20 meters. So it's long long field and it would turn around say at the ends, but there's different ways to go about it. So one way as the markers and others other ways could be. Even things like you pull a string or wire across long distances, but maybe you can tell me more: would the markers be acceptable like every five five meters or even ten meters.

B

A

And what happens there?.

B

A

A

Okay, if you go to the chat box, click on the last link that I put in there. It's a Google, Google Doc.

B

A

A

You're not able to open the chat box, it's a button on the left hand, side the thing that looks like a window on the left hand side. Can you look at my screen.

B

A

If you want to look at my screen, it's a thing on the left.

B

A

I can okay looks like you're in there, so the opencv tractor motion strategy so tell me more. How would we do that? What's the preferable route for it, make it making it easy so say: there's there's no markers.

B

A

So distance between two rows would be there's two ways. One is if the tractor is straddling the plants, so so it could be every two feet. So one option is every two feet or you can make it that it's every four feet. If the tractor is going in between the rows but I'm thinking, if we can get the accurate accuracy, it should just be going over the plants.

A

B

A

So we'd be talking about going over the plants until there, maybe like half a meter tall. We can make the tractor like.

B

A

Could make it such that yeah it's kind of raised raised up, but I mean we don't have to worry about that. Yet, let's just say that we've got the basically the initial establishment phase where the plants are just sprouting and that's when you need to do all the weeding and then, after some time potentially even this tractor spreads out straw like this yeah I'm, seeing that initially the the weeding happens by the tractor going bare over bare ground.

A

But the other strategy is also that that we could either spread straw or have the tractor autonomously spread straw like right in its tracks right behind its tracks. After it say after it goes over and does the weeding, we can add a functionality where it's mashing up a hay bale and it's putting that straw right in its in its on its tracks. I mean that's more advanced I mean we don't have to talk about that yet.

A

But but the simple simplest thing is that it's going through the rows and it's got either its tracks or little little implement basically like a little cultivator, that's hitting right by behind the tracks. So it could be both the tracks and a tiny cultivator behind the tracks.

A

So in detail- let's- let's let's draw that so the tractor would be like this so so tractor. It's got the tracks and the body of the tractor. So that's how the tractor looks.

A

And the rows the rows of plants are like this.

A

So the plants are growing like this and an attractor could be straddling them in one case, if they're about two feet apart.

A

Or the other cases, so that's case. One case two is where you have them just like that, except every other.

B

A

About that here would be like yeah yeah, something like that, specifically if the tractor with, if that is.

B

B

B

B

A

Like that, yeah.

B

A

That's fine, so here it's about 34, 3/4, meter,.

A

And here would be about a meter, meter, 0.5 1.5 meter.

A

The first is preferable. The second one would probably be probably be more easy.

A

It seems like just a technical challenge, because you've got you don't have to worry about. The spacing between between the tracks can be a little larger here like we could make these as needed so that we know the.

B

A

How heavy is the processing? Is it take time or is it pretty quick with.

B

A

The response is, what like a half a second or something like to to get oriented or tenth of a second or.

B

B

A

And when it shows so how are is the turning addressed.

A

Okay, so then, let's put turning slide number two.

A

A

At both ends, so let's take a look at case. One.

A

And you're you're also able to edit this document. If, if you let's see it's, um let me see: no, you can't advanced anyone can can edit this document, so you can edit this document. You can do things there, so basically you're going and at the end.

A

Yes, so here we're going like this and then we're going like that- and we actually would be going into this this one here with a second over.

A

B

It should go there.

B

A

A

Yeah yeah, that's it.

A

Hardware, can you send me a link to the the camera? What is it that we want to use?

A

Can you send a link to the computer camera.

B

A

B

A

Hiro is the first one.

A

A

$35 for a Raspberry Pi 3.

B

A

B

A

You want to Internet, can. Can you just refresh.

A

Ok, Raspberry Pi zero is like $10 less or something.

B

A

B

A

Five dollars they'll be for cheap clothes, or is that, like the actual fashion.

B

A

Five dollars or six dollars would be for clones or that original Raspberry Pi zero.

A

Okay, camera ready.

B

A

Wow yeah, it's only like $15 on eBay here.

B

B

Applications of the tactic with a bit of use must communicate with the base of the.

B

B

Wi-Fi HDMI four USB ports for deal with.

A

Yep so I put a link to the solid I, can't put a link to the camera.

A

Okay, camera module.

A

B

Yeah price of the controls just vanishes right.

A

Can't take a look at the Amazon link that I put in there does that would that work.

A

Five megapixels I mean does this? Does this get you something comparable to smartphone quality pictures.

B

A

Would that one work.

B

A

And the connectors on them are they all the same as far as the record, the plug-in? What's the plug-in like.

B

A

I see the one you liked.

B

A

One that you link to works with a zero as well.

B

A

I, don't see a difference in how they look might be.

B

The same thing.

A

Are you in the vex robotics program, you're working with with those guys?

A

That's is that, let's see is that what do you do for a living right now? Is that what you're doing the.

A

B

A

A

A

Excellent regarding the the turning have you looked at the relay in solenoids I'm talking about.

A

Yeah, no back to the slide number two: have you clicked on a link to the solenoids in a relay.

B

A

B

A

B

A

A

Yeah so they're the one you'd link to that's: they actually are selling a kit for making your own little camera. Well, that is really nice. That's that's! Actually, that's great.

A

That's nice, we could use I mean I. We take a lot of different documentation around here, so to have a low-cost replaceable camera would actually be a really good thing for us, like here, we'd love to set up what they're, showing there with for just time-lapse photography that we do a lot of might actually look into doing that. Also, okay, excellent.

A

A

Yeah, that's it that actually do you know like we have a revenue model of hosting workshops. I mean that would actually be a great build workshop for a 3d printed item. You know, say we run a 3d printer workshop. We can say okay and we're gonna make a 3d printed camera. On the second day. That's I really like that.

A

Okay, excellent.

B

A

Right, okay, so.

A

Let's get you the information on that select the information there is now, since you have.

A

I'll, send you the file yeah, you you're comfortable, opening up things in a freak out right since you passed the test, so I'm just gonna point you to a link to the file itself.

B

A

Do you run Linux on a separate computer or or USB.

B

A

Virtual machine yeah I mean we're supposed to be running off of a USB, because it's I think it's slower on the virtual machine.

B

B

A

Cuz we need four of those so more like $500. If we want to do full track. Yeah yeah, 520,.

A

We can change that.

A

Yeah, it's gonna, that's gonna, be the big big cost for just the tracks. We can do just two of those.

A

Well, that's that's the big power cube, but that big power cube is the gasoline version. The other part is we will we can switch? We will have a secondary power cube with just a solar panel on top of that, so solar power cube, that's not built yet solar power cube. So we don't have much on the solar power cube, but what what we can do is those solenoids can drive the output of that 16 horsepower power cube 10 gallons per minute.

A

They can also drive the very tiny, very small flow out of just like a 200 watt, solar panel, 200 or 300 watt. Solar panels will be going very very slowly in a day time. Like the expected rate, we did a 1 kilowatt power cube before and it gave us so the so expected speed there is I mean just give you some numbers there.

A

We did and let's translate, that to meters per second and stuff or kilometers per hour, so 14 feet per minute at 1 kilowatt. We expect.

A

1/5 of that, which is, let's say, 2 feet per minute at 200 watts, which means over a day the the important part, is: how much do we travel in a given day? How much can we weed in a given day- and we expect- let's say six hours of full Sun, so let's say.

A

We've got sixty minutes times six hours, 360 minutes times two feet per minute, would get us 720 feet per day or, like 250 meters per day,.

A

So that's like 250 meters per day, so about two rolls per day. If there are a hundred meter rows, so this is the field. Initially is a hundred times. 100 meters by 20 meters.

A

Two rows like that means that in a week, well, it's really four rows.

A

So in a week let's say this thing goes constantly twenty-four. You know every time the Sun shines and you need to weed about every week to two weeks and the intensive weeding period. So let's say we weed. Every week per week we can travel about a kilometer or two kilometers seven times. You know 1.5 kilo, seven times 250 is 1750 kilo, 1750 meters, let's say.

A

So that would be 17 row passes, let's see 17 times back and forth and every time you're getting 2 rows. So he can have 34 rows or 34 row weekly reading capacity.

A

Which is which translates about to an acre two acres about a hectare which would be excellent if that is done on totally unsolo power autonomous? Does that make sense.

B

B

B

B

A

In the current spacing that we have 34 rows is about if they're 3/4 meter rows.

A

It's about 25 meters, yeah,.

B

A

B

A

A

20 re 25 rose.

B

A

Yes per day, yes, 25, no, no.

B

A

B

A

B

And let's assume we're going.

B

B

Someone 1,300 meters.

A

B

A

Yeah, the little robotic wieder yeah.

B

B

A

Yeah yeah we're much bigger than that, though yeah mm-hmm.

B

B

A

I mean because this tractor has 7,000 pounds of pull ha ha 7,000. We can put extender implements if needed. We can actually double this speed because we can put cultivators behind the tractor and even spanning to the rows to the side. So we can do four rows at a time. I mean we have plenty of pull, so we don't have to worry about that. Yet will will optimize this later, but for now. Let's just do this very simple right behind it. No big deal. That's the proof of concept here. This is great yeah.

A

I think this is gonna, be a great application.

A

Okay, so so, right now we're gonna. Do the simple idea of seeing the role of plants that are in front of us and like actually recognizing the plants? Is there going to be actual, like recognition of like image, recognition.

A

Okay, so the poles? Okay, let's talk about the pole, so are we talking about every ten meters? We've got poles like little or too much.

A

Okay but little posts that the tractor goes over: no markers, okay,.

A

Okay, yeah I mean that would be very easy, so we can just basically cut little so how many meters, like every two meters or three meters.

B

A

So so there would have to be about 25 markers or so, if it's 25, if it's four yeah.

A

Yeah, that's that's good. I mean setting that up is only one time so.

B

A

Excellent so now tell me more about soso. So then the strategy is marker. Every four meters machine straddles markers.

A

But the markers have to be every other row. That's fine.

A

Yeah I mean this. This would be a super easy way to do it now. Let's talk briefly about, and maybe we should just restrict ourselves to that as a minimum viable product for now, and definitely I mean that's what we want to start with. But how much more difficult is that if you have posts like say every so the machine is actually looking straight ahead onto the field and there's posts. Let's say every 15 meters or so or 10 meters like a grid of.

A

Posts that are to the side of the machine. How much more difficult is that, in terms of the computer algorithms, to detect the precise position.

B

B

A

And it's not straddling them it's there to the side, they're taller. So basically there is a permanent total posts which are tall means one or two meters say: 1.5 meter, tall and they're in the ground and they're every say: 15 every you say: 10 meters or 5 meter, yeah, 10 meters. Let's say.

A

Because then, you're looking not straddling the the row but you're looking to the side, you're looking.

B

It should not be that hard taking frustrations of they.

B

Knock you out of you. They should be in a brief so that we everybody on the images that.

B

B

A

Yeah I can hear you I just clicked on the paper that Michelle linked, oh yeah, take a look at that paper, see the link in the chat box.

A

Yeah and it's yeah the little different strategy, because they're they're actually detecting weeds here, we're detecting markers which is simpler, so we're using.

A

Yeah yeah different little different, more complicated in what we're doing.

C

A

Paste that into.

A

A

Yeah yeah definitely good place to start.

C

I looked into it a few years ago,.

C

A

So please uh Michel, when you do them, if you find any other papers post them on page three.

C

A

Ly, definitely, can you link that you can upload it to the wiki or something.

C

C

A

Upload it to the wiki, if it's a nice paper, yes,.

C

A

Clearly, on the open-source robotic tractor.

A

C

A

Okay, so I'm gonna put that on I'm gonna put all these documents. I'm gonna start a page on the wiki. If you look at my screen yeah, you can see my screen. I'm just gonna go open-source robotic tractor weeding trip, weeding tractor. Well, it's gonna be more than weeding, so it's called open-source robotic tractor.

A

And I'll embed this document that we are working on right now in there.

A

I'll just go to my log.

A

So Salaam so then, as far as the accuracy positioning accuracy of the tractor itself, so we can get three inch or, like seven point, five centimeter, that it wouldn't sway off the course like plus.

A

What's the what's the maximum resolution that we could get for the motion of the tractor.

A

Well so so the idea is, the concept here is you're, the tracks are driving or whatever implement is drawn behind the tracks, and it wants to be as close to the plant as possible without killing the plant. So what is that acceptable distance? The acceptable distance would be I mean at best. It's like one inch from the plant, but that's like to close I would say two inches or three inches, where you're certain that you're, not gonna accidentally hit plants.

B

B

A

Right but the point is that yes, so say you put, the rose right, I mean it's.

A

Yes, let's say we got the treads of the tractor.

A

But more likely it's we're actually pulling a little little weeding implement behind the tracks. Unless because then we're just limited by the geometry of the tracks for the row spacing. So what we probably want to do is just have a little implement behind the tractor, which is being pulled and that implement just leaves the space for the row in the middle and it and it straddles that with three inches on each side.

A

So it's a six inch gap in the middle and then it's gets to three inches of the plants on this on the other two rows and it clears everything in its way so that three inches adjusts for like say, there's uneven terrain and whatever that the tractor does not shake around that it actually hits plants accidentally because I think the control itself. It's gonna be very clear as far as steering cuz, it's gonna be moving very slowly, at least on solar operation.

A

So the reason I mean it's we're not gonna have any problems controlling it. The question is just the stability in general of the tractor like if there's uneven terrain, it can't wobble and say that the wieder hits the actual plants. So we have to experiment with that, but the idea is I. Guess on the motion side assume we've got perfect Drive. That means we can turn one wheel or the other I mean there are no limits. There I mean the theoretical limit. There would be. What like? How do you calculate the theoretical limit.

A

It's gonna be based on an accuracy of how you're reading the workers right like what is it? Can you show me.

B

A

Yeah I would say that smallest step could be I mean we can say you can probably say one centimeter.

A

Like yeah for Pratt for practical approximation, I mean it's continuous, it's really continuous, but because of the mechanics of the tractor, let's say it's it's a say. One centimeter I think that's fair to say.

B

B

A

B

B

A

I see I see so the mechanism. There is your tell me a little bit more about the mechanism. There, yeah cuz, yes, a one side, there's a yaw or a tilt, you can say say the tractor is tilting to one side. How does that affect your measurements?.

A

Yes, Michel, you know about that.

C

A

No okay, so a thing that adjusts for camera angle. So isn't there an open-source project like that open-source camera holder Salaam, is in there an open-source project that has that what you're talking about camera stabilizer.

A

Okay, okay, let me show you this, because I saw this and and now it's all coming together, so open-source camera stabilizer.

C

Well, it's not frozen.

C

A

Sam yeah, oh.

C

A

A

Exactly okay, let me show you this open Sam.

A

Okay, look at page three.

A

Camera stabilizer.

B

A

That's what we want to use so mount the camera so mount the actual, both the Raspberry Pi and the camera on that little apparatus.

B

A

This one is passive. This one is passive, so it would probably allow us to do the same thing with more ease.

B

A

So so the camera say is stabilized to perfect horizontal position so you're reading, both the gyroscope and the computer vision.

A

Can you describe that a little bit so so so controls? Let's, let's write some notes on that.

B

A

So, can you summarize the logic sequence of the of how your what you're doing to measure and to determine a straight course.

B

A

B

A

Well, even just the basics here so you're reading you're looking at markers your.

A

So you look at markers, get them aligned best as you can.

B

B

A

Right, that's right.

A

Okay, where does the gyroscope come in.

A

So you sent a threshold like if it's, if it's what happens, if it's above that angle,.

A

Well, I'm thinking here in our case, if the camera is in front of the tractor on the open Sam, then do you care about the angle of the tractor I?

A

Don't think it would matter, don't have to even do that.

B

We don't have any idea.

A

A

Well, I think the field is gonna, be I'm thinkin zombie, so for under five degrees. We're okay in all cases,.

B

A

Let's do that what our next steps now cuz I, think I'm pretty good for now.

A

Over what distance.

A

So a picture that shows two of these steaks in front of each other.

B

A

Okay, let's, let's write a slide on that part, so.

A

A

Okay, we can change this one here so.

B

The gyroscope itself can give us a pretty critical stabilization for the orientation on its own, so we use the joyful theta.

B

Let's say part of an initial stage: has the calibration more that can can be stressful world.

B

B

A

So you want a picture that is like this. Let's see.

B

A

A

What kind of document you.

A

Can upload to the wiki just click upload on the left hand, side of the wiki.

A

You can upload to the wiki.

A

Okay, so look at slide. Number four: is that what you want to take a picture of this area.

A

Well, it's every other row right, so it'll be twice that so every 1.5.

B

A

It's three by two: here's that okay.

B

A

From what height should I take this picture.

B

A

So like this, this is what you're looking for.

B

A

If we do not go the computer-vision route, but pre-programmed through GPS, that's easier right.

B

B

A

A

B

B

A

Noise is added on purpose.

B

A

A

A

You're saying that the commercial GPS systems they're very accurate but they're without the noise but riding on the same system.

B

A

Saying that are.

C

C

A

That, on a page.

C

A

Two three centimeter accuracy.

C

A

Meters in XY.

A

Four centimeters in XY.

A

Salaam, are you suggesting that.

C

A

For example, in our camera on a cell phone they have the circuits there, they're built with the noise injection.

A

A

Okay, so we'll take you, some pictures, you're gonna, get a data set now.

C

A

A

um Salaam, do you also have the capacity to get yourself some 3/4 inch PVC? Do you guys have 3/4 inch PVC.

A

Can you get yourself a piece and take some pictures as well.

B

A

A

Because you're, detecting by it light intensity.

A

A

Well, well, the thing is here right now in order to expose bare ground, which is what's gonna happen, we would have to do that we would have to plow or something you know till an area, because, right now it would have grass. Is it acceptable on just grass.

B

A

A

All right could do. That is that, would that be the next step to train? You know, get this data sample and then work with that and as far as the computer, can you computer software, the opencv software? Can you send a link to what you're actually using like? Is it just like one program uploaded to the Raspberry Pi? What are you doing here? You're uploading, a whole operating system with this OpenCV software or how does that work.

B

A

And the actual software are you gonna write that from scratch.

B

A

Kind of helper functions.

B

A

What kind of functions, for example, give me an example of a function like.

B

A

You're saying you're gonna have to hunt down these all these functions.

B

Hsp combination.

B

B

A

Yeah, don't they have some samples where people are detecting things like PVC stakes like? Can you find something like that.

A

Yeah seems like this kind of there's something very close to this- that someone must have done already.

A

Yeah, okay! Well, that sounds good. So we'll get this to you, I guess! In the next few days we can.

A

And so the final product will be you upload the libraries and OpenCV libraries to the operating system.

A

You install those libraries on your artwork, Raspberry Pi and then in the program. You just call on some of those libraries.

A

Okay and the libraries are ready.

A

Okay sounds good.

A

What else anything else to cover right now.

B

A

Good for now, so next steps are to get you the data set and then.

A

B

A

Right so the markers you're asking what they are.

A

What height, oh yeah, I mean I'm thinking, not half a meter. Just small like 20 centimeters 23 I mean as far as sticking out of the ground. They're gonna be halfway in the ground, but sticking out you just need, like 20 centimeters,.

A

It would say so: I would expect maybe 20 centimeters underneath and 20 centimeters. Above so yeah. You can say: 20 centimeters, that's yeah, yeah about that.

A

Yeah 1/2 1/2 meter with halfway stuck into the ground.

A

Yeah so that basically they're tall enough stiff enough that they're not gonna be knocked over by anything they're sniffing to the ground, but they're, not that tall, that the tractor wouldn't clear them.

A

Yeah so there have, they actually have to be yeah. I'll, see I, think we have like yeah 20, centimeters Alan is on a high side. It might be as little as 10 centimeters just to make sure like say in muddy conditions. You don't knock over your your posts, your little posts, exits.

A

Well, though, as you can see, and um if you download the tractor off the let's see they put a link on the tractor.

A

Right that should be in a document here, so so tractor.

A

It's got a low, wheelbase, low, low height. That is.

A

Yeah, where is that? Let me see.

A

I'll give you a direct link to that here.

A

Okay and then the condition is upon a startup of a new day when a new day starts. The tractor has to orient itself but say it's gonna be in the middle of the field. So you have to acknowledge that that you'll be in the middle of the field, and you have to orient yourself where you are, but it will. It should be easy because if the tractor should have left off in a good position straddling a row, so you can continue from that point.

B

A

Okay, so I'm page one I put the tractor file: that's in freecad, but you'll see how low the low the axles are to the ground and still without tracks that files being worked on right. Now, it's being updated, okay, so I think anything else. We have everything for now.

B

B

A

I'm putting this so on Selam log, so let's see use your log. That's the first thing is log your hours and log. Your your tasks last one you have is on what Wednesday, our August 16 I'm, just putting a link to the current working document on your log, so you have it for reference. There.

A

Okay, okay, and so you can do you can begin on some things like like starting to look at yeah review the like. Can you start on anything at this point, I, actually starting on a code and getting things ready for that.

B

A

A

Excellent excellent, so anything that you do just please log it on your log like, for example, if you start writing the code, you know it's, the the idea is publish early and often meaning that as soon as you have anything just post it, so it's at least a placeholder, and then you can upload new versions to it. Just like the working document, it's already up there, but we can add to it all right.

B

A

And please free to add to this working document in Google presentation, I mean you start pages there and and so forth. So you're welcome to edit all right. So um thank you very much. So hopefully we get this thing and we build it on October 27, but building the system say sorry say it again: we're gonna! Yes, we're building it.

A

Here we have a workshop organized at the at our facility here we're gonna post that workshop in a cup in like a week to week to two weeks, we'll post that workshop and yeah we'll build it. Then we're gonna build the small version, which is just the 16 horsepower, as well as the 64 horsepower. Now, as far as the small tractor, basically there's GPS I mean it's basically gonna be a module add-on to the tractors tractor, so we can either use that or just use it for normal operation.

C

A

C

A

Yes, so keep going on the research on those things yeah, let's see if there's anything usable for now, we can do this computer vision route, which is good for a first prototype, and that's great so we'll continue doing this. Okay, thank you guys. Take care.

B

A