This self-driving farm robot kills 100,000 weeds an hour by laser: no herbicide required

The Carbon Robotics self-driving farm bot kills weeds with lasers: no fewer than eight 150-watt lasers, to be precise. It images weeds with brighter-than-day flashlights, distinguishes them from good plants with AI and deep learning, and targets them with pinpoint accuracy.

Weeds gone, crops untouched.

That’s pretty cool, but much cooler is that no toxic herbicides are required. That’s safer for the farmer, better for the soil, and produces better, healthier crops for everyone to eat.

Support TechFirst: Become a $SMRT stakeholder

Herbicides not only threaten farmer health, they reduce soil quality over time, making it harder and harder to farm effectively. And crops that haven’t been sprayed with herbicide growth better, faster, healthier.

(Subscribe to my YouTube channel)

In this episode of TechFirst with John Koetsier, we check out the robot, talk to the CEO of Carbon Robotics Paul Mikesell, and see the results on the fields.

Check out the story on Forbes.

Or get the full transcript below …

Subscribe to the TechFirst podcast


Full transcript: Carbon Robotics’ laser-equipped weed-killing farm robot

(This transcript has been lightly edited for length and clarity.)

John Koetsier: The last time you saw farmers using lasers, you might’ve been on the planet Tatooine. However, they could become standard issue for every farmer today if Carbon Robotics continues its trajectory. The company makes a laser-powered weed killer. It’s safe, effective, it doesn’t pollute the earth with herbicidal poisons, and it could be the farmers’ ultimate answer to superweeds that are pesticide resistant. So today, we’re chatting with Paul Mikesell, CEO and founder of Carbon Robotics. Welcome! 

Paul Mikesell: Yeah. Thank you for having me, John.

John Koetsier: Hey, it is a real pleasure. Tell me about the autonomous weeder. What is this thing? 

Paul Mikesell: Yeah, great. So we wanted to figure out how to help farmers with our experience in computer vision, deep learning, robotics, and electro-mechanical controls. We have, our team is made up of folks with deep backgrounds in that area. And we spent a lot of time working with a bunch of farmer friends that we have and kind of learning where the biggest pain points were. And what we discovered pretty early on, is that one of the biggest issues that farmers are having right now is with weed control, particularly coming from a couple different sources. The first problem with weed control is, and maybe people don’t realize this, but if you don’t weed your fields after you’ve planted the crops, you will lose at least half of your field. 

John Koetsier: Wow!

Paul Mikesell: So your productivity is cut in half immediately. You won’t even recover your costs at that point. Farmers don’t have huge margins on these crops, you know, as is. And so the need to do effective weeding is super important, and it’s a thing that everybody winds up spending a bunch of energy on. There’s only a couple methods of dealing with this problem that we have today … before Carbon. So the first one is the thing that I think everybody’s probably aware of, is this continual spray of herbicides. And herbicides are chemicals specifically formulated to tolerate the way that plants grow and receive nutrients and disrupt that cycle. And so what has gone on over the last several decades, or I guess basically the last hundred years, is these herbicides have continued to be sprayed and sprayed onto these fields. And what happens is the weeds are adapting, just like any biological system. We have antibiotic-resistant bacteria that gets created in our bodies in the human biome, partially because of overuse of antibiotics. We had the same problem in the fields, and so these weeds are becoming resistant to the existing herbicides.

What this means is that chemicals are having to become reformulated and they become more toxic, and there’s a lot of different things that happen along the way. And the thing that particularly concerns me is the long-term farmer health. You’ve probably seen a bunch of the research that has come out about some of these chemicals like glyphosate that, depending on the country, they’re either pretty sure it causes cancer or there’s some skepticism about it, but in any case, there’s been multiple billions of dollars in lawsuits in the United States over this. And there seems to be a lot of evidence pointing in that direction. Recently, there have been a lot of press around this chemical called paraquat that they think causes Parkinson’s in folks as they get older. So it’s really kind of a tragic situation, ’cause if you’re a farmer and you’re farming for five decades, trying to provide food for the country, trying to provide a living for your family, and you have this main tool that you have to expose yourself to and your workers over the course of all this time, and then you discover later in life this causes really serious health effects.

John Koetsier: Yeah.

Paul Mikesell: And so that is kind of the devastating impact that we’ve been experiencing, you know, learning about. There’s the farmer health issue. There’s the quality of what happens to our vegetables and produce over time as we’re using these chemicals to keep the fields clean for the produce. The nutrient content in the vegetables that we’re eating is down by a huge amount — as much as 40% over the last two decades because of this stuff. 

John Koetsier: Wow.

Paul Mikesell: And what happens to the land over time, where we wind up stripping out a lot of the essential micro bacteria that’s down there in the ground, we’re changing the way that things are composting in the ground and it’s causing a bunch of longer term issues with soil health. So what are farmers really leaving for the next generation? There’s a lot of questions and problems with what’s going on. So that’s first and foremost. The alternative to using herbicides has been human labor, just people out in the field manually picking and pulling these weeds. And that, you can probably imagine, is a very difficult job. It’s a very dangerous job. With things like global warming, as it gets hotter in the field, they’ve had fatalities out there as people are in the fields working and they get overheated, and it’s a very difficult job. So, that’s not a great thing for humans to be doing with their time just generally. And then there’s all the immigration issues we have around even being able to get these folks into the country, take care of them, make sure that they’re well taken care of, that they have good places to live, access to healthcare, access to reasonable nutrition and things like that. So, we became super concerned about what was going on. And I really feel for the farmers, because they have these limited tools and they’re really trying to provide food and just make a living, and a lot of this stuff winds up coming down on them. 

And so what we wanted to do was figure out if there’s a better way we could do this. And what we discovered relatively early on is that through the use of high-powered energy systems — so, lasers, which is essentially a way of delivering targeted energy — we can kill these weeds. And we can do it with the use of our computer vision and deep learning expertise that we’ve got from our tech backgrounds, which allows us to in real time identify what’s a weed, what’s a crop … and kill the weeds. Get rid of them. You may have seen some videos of farmers doing things called flaming, where they have a series of propane tanks on the back of a farm implement, a tractor implement, and they just burn the whole field down. So that’s a technique you can use before you’ve planted to sort of reset the field. So one way to think about what we’re doing is a micro-targeted version of that flaming. We do the same thing, except we have millimeter control on it, so we can get in between the crops and essentially use that same technique all throughout the season. And so it’s very effective. We’ve done this through several seasons now. The farmers are elated and we’ve been able to save them a lot of money on their weed bills. And what we are doing now — I think you probably saw some of these announcements — we just closed our $27 million Series B, growing sales and support, expanding the engineering team, and we’ll start to put some more effort into marketing. Most of our work has been, most of our sales have just come word of mouth and farmers talking about what they’ve experienced. So that’s kind of where we are, that’s where we’re at.

John Koetsier: Yeah. Super interesting. Talk a little bit about the autonomous weeder, what it looks like, how it works. You talked about deep learning, so I’m assuming you’ve got some vision systems. I’m assuming you’ve got some autonomous driving systems. And obviously you’ve got some learning as to what’s a weed, what’s not a weed. How many plants can you recognize? How many weeds can you recognize? And is there one laser? Are there multiple lasers? How much ground can you cover? Tell us a little bit about that machine.

Paul Mikesell: Yeah, these are all great questions. So the machines that you’ve seen the videos of — and you can find these on our website: — that is an 80 inch wide machine. So, 80 inches is a relatively standard configuration. Some people have, well, they’ll do two 40s in that or some people will do 80 inches as a single row and they plant more densely. And in that system, there are eight lasers going across the row, so horizontally across that row. These are 150 watt CO2 laser tubes. So they’re probably six feet long, the tubes themselves. And the tubes come from other industries, so these are used for metal cutting, typically. So you’ll find one very effective way to take a hardware platform and make it useful and profitable, and have a good ROI in another industry, is to take parts from somewhere that’s already gone to market. And so these laser tubes come from that, which is why we’re able to build these things so cost-effectively. 

And then, yes, there’s a deep learning system. So people are probably familiar with their experience on things like Facebook where if you post a photo, and it will make a suggestion about maybe the person in the background is your cousin or your brother or your friend or whatever. That technique is a technique called deep learning, which is where we have a neural net that is trained from a bunch of example images. And what it’s doing is it’s a system where it somewhat mimics the human visual system in the way that we have a cascade of what’s called neurons, in the brain — and it’s the same thing, you use the same term in deep learning — that learns over time how to build a series of layers. In this case, these layers are called convolutions, but a series of layers that learns higher and higher feature levels about what it’s seeing. In our case, in humans, through your eyes and in these machines, through their cameras. And it’s able to say, in our technique, ‘this is a spinach,’ which is a crop that somebody might grow, and ‘this is a purslane,’ which is a weed that somebody may want to kill. And we know what the crops are and what the weeds are, and this is done through us taking a bunch of pictures, labeling those examples — that means marking out with a tool that we’ve built, what’s what — and over time, the neural net will learn this information. And so it’s important for us to be able to know weeds and crops because these farmers do rotations. So you might do carrots and then, so your field is full of carrots. You grow the carrots, harvest them, send them to market, hopefully make a nice profit. And then, after that you plant onions. So in the first scenario, the carrots were the crop and everything that’s not a carrot you want to kill. In the second scenario, the carrots are now weeds. If there’s any leftover carrots you want to kill them and protect the onions. And so our machines know what it’s actually looking at and can say, ‘Okay, it’s onion time, let’s kill the carrots.’ And so this is, we found that it is very important to understand everything about what we’re looking at. And you asked about the self-driving aspect, so we use a very similar technique. In the farm environment, we have almost a perfect scenario for us in that, as compared to self-driving cars on the public streets, for example, because in the farms they’re what’s called furrows, which is, you just think of it as the tractor tracks in between rows. And so all we have to do is find those furrows and keep our wheels in the furrows, and then when we’re at the end of the row, turn around and come down the other side. Great. So, you kind of imagine the problem, you can imagine what the computer’s trying to do. So we do a similar technique. We use deep learning neural nets in our computer vision system to find those furrows, and then all we have to do is keep the wheel in the furrow. 

John Koetsier: Yep. Yep.

Paul Mikesell: Perfect. So that’s kind of the basics of how it works. 

John Koetsier: Very interesting. Does it look ahead of itself and say, ‘Oh shoot, there’s a human, I need to stop’ or something like that, some basic avoidance?

Paul Mikesell: Yeah. It looks ahead of itself and looks for other things in the path. And our problem is even easier in that we also have lidar-based safety scanners…

John Koetsier: Wow.

Paul Mikesell: …that just say, ‘If I see anything in my way’ — they’re pointed down at an angle, so anything gets in the way between the robot and the top of the field for, I think it’s 15 or 20 feet, the thing will just stop. So we have a backup there. So it’s computer vision looking to see what’s in front of it. It’s a lidar safety system that will emergency stop if anything gets in the way. And then we also have the GPS boundary around the field. So we know if we’re ever approaching the edge of that boundary it’s time to stop, because something has gone wrong. So we can sort of have a three layer safety-redundancy system that prevents us from running into things. 

John Koetsier: Nice. So how quick is this? This depends obviously on the size of your farm and which crops you need taken care of, but how many acres can this do in a day? And how many machines would you need if you’re a large farmer? Those sorts of things.

Paul Mikesell: Yeah, it depends a lot on weed prevalence. So how many weeds we have to shoot dictates the speed of the machine. But you can think of it as doing a sort of one to two acres an hour as that machine you saw there. 

John Koetsier: Okay.

Paul Mikesell: And so, as a farmer, this is the way that I think we’ve been successful with folks in sort of working through the ROI. So it’s, at what point given the savings do our machines pay for themselves? So if you say, ‘I’m just going to add up all the dollars I save’… you know, proveably, we’ve shown this through several seasons. But you say, ‘Okay, how much money am I saving? The cost of the machine, how long will it take me to pay that down?’ So in our case, it’s between one and a half to three years. And that, some people have multiple seasons per year, multiple plantings per year, so that can be through multiple plantings, but … in just raw time, one and a half to three years, you say, ‘Okay, I’m going to pay the cost of this thing down.’ So then it’s just a scaling problem. So how many of them do you want? It depends on how many acres you have and as long as it’s profitable on a per acre basis, then everybody’s happy. 

John Koetsier: Amazing.

Paul Mikesell: So that’s kind of how we, that’s how we work through the financial scenarios. 

John Koetsier: Amazing. Do you sell them or do you lease them? 

Paul Mikesell: We sell them, just sell them right out. We do have financing for folks who want to pay this down over a longer period. We have a five-year term program for … and we will finance them, for the farmers, for over the course of those five years. And what we like to do is make sure that the farmers are cash flow positive on this thing for every year they do this, so that’s why we spread it out over a five-year term, even though it’s going to be ROI over one and a half to three years.

John Koetsier: Yep. So I’m assuming this thing runs gas or diesel or something like that, is that correct?

Paul Mikesell: Yeah, it’s just diesel right now. Yeah, that’s right. It’s diesel, that’s what’s in the farm, that’s what’s available. 

John Koetsier: Right.

Paul Mikesell: And I think you’re probably heading in this direction about how do we get to the point where we’re not burning fossil fuels in all of these farms? It’s a great question. You know, it’s a thing that we would like to participate in … the transformation of. We need to get to the point that we have some infrastructure to be able to do that. 

John Koetsier: Right. 

Paul Mikesell: Sort of, you could think of this as supercharging stations for these kinds of robots… 

John Koetsier: Around your farm [laughing].

Paul Mikesell: Sure, yeah. I mean, the power is already there in a lot of places. There’s already other things that are running electrically in the field. You can think of those center pivot arms that you’ve probably seen that make the circles. It’s why a lot of the crop fields are circles is ’cause there is a thing in the middle that will go around, rotate around and put water on. There’s already a bunch of electricity there, so it would be nice if we could build out more of that infrastructure in a cost effective way, but then we could do charging in the field. The issue with electrical right now, any of the electrical machines we’ve seen have this kind of ratio, they’re able to be in the field for six to eight hours, but then they need a 12 to 16 hour charge time. 

John Koetsier: Yes.

Paul Mikesell: And that just doesn’t work for these farmers because when you gotta go, you gotta go. It’s crops are there, you know, weeds are coming up. It’s time to do the job. And we’ve seen farmers use these things 20 hours a day. And so we’d need to get to the point that the charge-to-drive ratio started to make sense, and we’re just, we’re not there yet. 

John Koetsier: Yep. Let’s talk about—

Paul Mikesell: So we’ll focus on our business but we would like to participate in that. 

John Koetsier: Yeah. Let’s talk about that 20 hours a second, ’cause that’s quite interesting actually. You know, if it’s a big enough field it’s staying on the field. Can it move itself between fields? Does a person need to come for there? Does it light up its path so it sees what’s around it?

Paul Mikesell: Yeah. We need people right now to guide it between fields. We do have some software to follow a route between fields, but because that environment is much less protected, we currently want people to help guide it between the fields. Because these are usually places where other trucks might drive and things of that nature.

John Koetsier: Sure.

Paul Mikesell: And some of them are pretty skinny, so you’re driving past fences and poles and stuff, and so we can’t really use our lidar system there. And so we need to be pretty careful about that. Eventually I’d like to get to the point that they can automatically go between fields, but we’re not there yet —infrastructure wise, we’re not there yet. It does light up its path. There are two main lighting systems. The one that’s maybe most obvious for driving is we just have essentially headlights. And that gives us enough of a field of vision that we can see those furrows through our cameras and drive the [inaudible]. But then the other thing that’s really essential is the ability to light up the bed top … brightly, consistently, with good color spectrum. And we’ve had to develop our own lighting system for that. You may have, if you see our videos, there’s a lot of flashing underneath the bed of the robot on the bed tops. That flashing you see is not the lasers. The lasers are what’s producing the little puffs of smoke you see, but the flashing is our lighting system. And so we’ve had to develop our own system for this because there just simply was not anything on the market that had anywhere near the kind of accuracy and response rate that we developed. So this system is synchronized with the cameras and what happens, as a person if you’re standing there looking at it, you get a sort of sense of strobing, but it’s happening so fast you can’t really tell. If you watch it through a camera, through a video camera — which is what you’d probably see on our website and YouTube, etc. — you do see some flashing, and that’s because of the effect of the shutter of the camera synchronizing in timing and then getting off sync again with the speed of the lights flashing. 

John Koetsier: Yes.

Paul Mikesell: And if you look through one of our cameras which is looking at the field top, it just looks perfectly bright … because it’s synchronized with the shutter of the camera. So, camera opens, lights flash, and it’s actually, it’s brighter than the sun. And so this is important for us because we need to get rid — we need to sort of strip out all of the shadows that happen, particularly in the early evening or early morning where the sun’s just coming up or just coming down and those rays are able to get underneath there. So, that was a really important aspect. You know, everything else we’ve seen in the field has a, what we call a skirt, but it has like a protective barrier around the bottom of the machine and that bumps into crops and causes all kinds of problems and it’s not consistent, so… We do alter our own lighting system, and so what that means is, we can run at any time in the day. We can run all night long. We can run at high noon. We can run early evening. It really was, I think, an underappreciated part of the technology we had to develop to make this work regularly and consistently. 

John Koetsier: I’m guessing not underappreciated by the people who actually use it and the farmers who want it in their fields, because if they can set a machine to go in the afternoon and it runs all night and in the morning that field has no weeds in it and the crops can grow, uh, I think they’ll be pretty bloody happy with that. 

Paul Mikesell: Sure, yeah.

John Koetsier: I’m pretty impressed. I mean, like there are multiple ways of dealing with issues of herbicides and weeds and crop health and other things like that — there’s vertical farming, there’s indoor farming, there’s greenhouses and stuff like that, but flat farming isn’t going away and there’s a lot of land in that. And all the herbicides that we’ve been dumping on there for decades is not doing good things to our soil, as you mentioned, so… 

Paul Mikesell: Yeah.

John Koetsier: I would assume that the farmers, I mean, obviously they need to make the economics work, but are pretty happy to not be spreading around a ton of herbicide.

Paul Mikesell: Yeah, not spreading a ton of herbicide. Leaving their land in a healthy state for the next generation. Producing good produce. And keeping the health and safety of their workers at the forefront of what they’re trying to do.

John Koetsier: Yep. Paul, last question, project out maybe five years or so, as you see your innovation continuing and as you see yourself developing product, making it smarter, all those sorts of things. What do you see? What’s the product you’ve got five years from today?

Paul Mikesell: Yeah, there are a number of other things on the farm that we’re starting to work with folks to do some other automation around. So our specialty is really computer vision, electronic control systems, robotics generally. And one of the great things about farmers is they’re super innovative. And so if you sit down with somebody and talk about, well, what kind of things are giving you a problem? What kind of things do you think could be automated? You’ll get hours of conversation. And some of those things are going to start bearing fruit in the relatively near future, and so we have a number of things that we will start working on. And five years from now, I hope to be able to automate as much of farming as folks want us to do, and as much as we can get in there and really be helpful at. 

John Koetsier: Super interesting. Thank you for your time. 

Paul Mikesell: Sure, yeah. Thank you. Nice to meet you

TechFirst is about smart matter … drones, AI, robots, and other cutting-edge tech

Made it all the way down here? Wow!

The TechFirst with John Koetsier podcast is about tech that is changing the world, including wearable tech, and innovators who are shaping the future. Guests include former Apple CEO John Scully. The head of Facebook gaming. Amazon’s head of robotics. GitHub’s CTO. Twitter’s chief information security officer, and much more. Scientists inventing smart contact lenses. Startup entrepreneurs. Google executives. Former Microsoft CTO Nathan Myhrvold. And much, much more.

Consider supporting TechFirst by becoming a $SMRT stakeholder, and subscribe on your podcast platform of choice: