Robotic Farming, Aussie Style

Australian roboticists from the Queensland University of Technology have developed a prototype agricultural robot that uses machine vision to identify both weed and crop plants before either uprooting or poisoning the weeds or applying fertiliser to the crop.

The machine is a wide platform designed to straddle a strip of the field upon which it is working, with electric wheel motors for propulsion. It is solar-powered, and it is envisaged that a farm could have several of them continuously at work.

At a superficial level there is nothing new in the robot, its propulsion, or even the plant husbandry and weeding equipment. The really clever technology lies in the identification and classification of the plants it will encounter. It is on the success or failure of this in real farm environments that the robot’s future will hinge. The university’s next step will be to take it on-farm, and the ABC report linked above has a wonderfully pithy quote from a farmer on the subject. You can see the machine in action in the video below the break.

Farming robots have a significant following among the hardware hacker community, but it is possible that the machine-vision and plant-identifying abilities of this one would be beyond most hackers. However it is still an interesting project to watch, marking as it does a determined attempt to take the robot out of the lab and into real farm settings.

We’ve covered farming robots a few times in the past here at Hackaday. Gantry-based robots like Farmbot, the walking hexapod Prospero robot farmer, or a further six-legged robot also called Farmbot. Practical farming robots have been a little more rare though, with a Canadian autonomous John Deere tractor made not by engineers but by a farmer being one of the most promising.

Thanks [muA] for the tip.

41 thoughts on “Robotic Farming, Aussie Style

    1. Oh that was a project I got stalled on way back… it wasn’t going to hunt the dandelions to dig out so much though, just dehead them…

      The premise behind mine was an autonomous self powering lawnmower/care system, and it was going to have a solar concentrator on top to bake terpenes out of grass clippings and lecithin out of dandelion heads which would then act as an emulsifier to combine terpenes and excess moisture, to run an average-ish commodity internal combustion motor. (Terpenes would be desperately low octane so you’d need a very low efficiency low compression spark ignition engine, whereas adding water slows the burn and acts like an octane modifier.) Probably you could use terpenes more directly in a small diesel, but adding control to those seems like more of a bastard, plus they don’t seem to come much smaller than about 9HP.

      Anyhoo… I got bogged down in that side of it and didn’t tackle much of the machine vision, it was probably going to be a rather stupid path follower at first.

      1. I now want to do this with grass clippings. It’ll have to wait ’til next summer. Do you have the chemistry worked out for this? Lab notes? Etc? That you’d be willing to share, I mean. I’d love to test just the combustion portion of the engine. I think it would be a lot of fun figuring it out.

        1. Well for background on emulsified hydrocarbon/water fuels there’s this..
          https://sites.google.com/site/braxpeace2/waterinfuelblends

          Otherwise, that’s where I got stuck, there was a lot of empirical work to do to characterize how much combustible hydrocarbon stuff you could pyrolise out of the clippings, from a solar concentrator, some of it might have been methanol and other things than terpenes. So I had the whole summer free for once and got all gung ho about it and what happens, rainiest. cloudiest. summer. ever. also on the cool side, so the grass wasn’t even growing much to clip a load to cook up on a hob to get some idea… got lecithin out of dandelion heads by boiling them, and made mayo with that and cooking oil to prove the emulsification. Got to be late August and I’m thinking screw it, gotta get somewhere, I’ll buy some turpentine as a stand-in to mess with and maybe then at least I’ll know what ballpark mixture I need for the motor to run….. yeahno, nowhere around here has real turpentine any more, just substitute… apart from an art store where I can get 100ml for $7 or something stupid. … then family SHTF again and that was the end of any significant free time for months.

          1. Actually terpenes may be more of a misnomer, it’s the term I latched onto for the pyrolysis products when booting this round in possible 2008 or 09, since then the sexy term for such has become… “‘Fast Pyrolysis Bio-oil’ FPBO

            Which you’ll see has had a crap load of action in the 2010s

          2. Well, that’s the way life goes, I guess. I’m assuming any sort of plant matter will do as far as pyrolysis goes (sort of like wood gas burners). I’ll collect what links you’ve provided and see what I can get up to this summer. I’m assuming that I’ll be able to get through the basic maths, and maybe even burn something at some point. Adaptation to an actual IC engine would be an entire other summer’s work (I’m guessing).

  1. Just a thought, would it be less energy intensive to say microwave a weed in situ than bother digging it up robotically?

    I don’t like the spray option. If you’ve got this tech to avoid application of chemicals directly to the food chain and water table then why the heck even consider it.

        1. If that worked like I thought it did, it wouldn’t work. They could put the muzzle right on top of a weed and let it trigger until the battery was flat, which could be as short as a couple of minutes and it still may not have got more than a solid 5 seconds worth of actual heating in. I’d be thinking that a steady 20 seconds would be more like what you want.

    1. Agreed about the spray. Weeding by hand is just too labor intensive for humans, so we cheat by using chemicals. Removing the human means you can go back to manual weeding handled by a machine where the labor costs over time are extremely low.

    2. Currently they spray weedkiller by the gallon on entire fields. It’s why “Roundup (TM) Resistant” genes have been put into crops. So a little squirt per plant is a massive improvement. If you wanted to go fully chemical-free, it’d just be a small modification, once the machine’s proved it’s worth. Let them get all the other bugs out first, then add a more sophisticated weed killer.

      There are things you can use as weedkillers in small quantities that aren’t so bad, even salt would do.

      1. Except that salt doesn’t degrade, and without sufficient flushing action, tends to stay in the top levels of the soil. We have plenty of salt issues in our croplands already – too much irrigation actually brings salts up from deeper levels and ruins the soil for many crops. https://en.wikipedia.org/wiki/Salinity_in_Australia

        A big enough convex lens under an Australian summer sky, equipped with a shutter mechanism and aiming the focal point at the weed’s main stem or seed head, will take of things without adding additional problems to the soil.

  2. Wonder how closely related this one is to a harvester platform I once worked on in conjunction with Eze Corporation.

    I have a feeling the platform took swimming lessons in the 2011 floods, but essentially it was about the size of a bus with tyres that were ~5′ in diameter.

    All four wheels were independently steerable, and the original design had 4 20kW hub motors. The whole lot ran on 48V. If you turned the wrong two MOSFETs on in the DSP firmware, the resulting flames jumped up 6″.

    I worked on the driver control console. This was responsible for issuing CANbus commands to set the angle and speed of each wheel. The original version was my introduction to microcontroller development, where I ported a Borland C++ application to a MSP430F135 microcontroller (the algorithm was in one C++ method, I ripped that bit out and made it a C function) which talked to CANbus via a Microchip MCP2515.

    It was that project where I learned how crap the MCP2515 was when you drive them hard: there’s a race condition wherein if you happen to receive a message as you’re trying to construct one, it sends a message with the same payload as the one it received. We replaced the whole lot with a Luminary Micro development board in the end.

    1. That problem where software can fry the MOSFETS can be prevented by having some hard logic between the controller and the motor driver to prevent a condition known as shoot through where two transistors on the same side of an H bridge can get turned on causing a short to ground.

  3. Do any farmers still use a Sizz-Weeder? It killed weeds with fire. One setup used thin, fan shaped flames to cook weeds around cornstalks. With the thin flames and fast motion the corn would barely be warmed while the small weeds would be burned.

  4. I have a loose dream of one day building a robot that would be programed, through training, to identify variations in the appearance of desired plants, and destroy all other plants I was thinking of using lasers for the plant destruction. An additional idea was to add the ability to recognize unwanted insects and zap them with the laser too. I figure getting the robot to identify weeds as weeds would be a waste of time.

  5. It may be easier for the system to detect rogue plants if the crop plants had a gene to fluoresce or some other kind of marker that would make them stand out. The system could then characterise weeds as any plant that doesnt fluoresce. The problem would be that the crops would be classed as GM.

Leave a Reply to RWCancel reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.