Nellie, the Weed Picking Robot


Introduction: Nellie, the Weed Picking Robot

About: I am an author and a maker. My current project is Santa's Shop. I'm working on a science fiction type book--more later. @EngineerRigsby

Pigweed, which can grow 3 inches per day, has become resistant to the dominant weed killer, glyphosate, sold by Monsanto as Roundup. This super weed threatens the nation's soybean and corn crops. U.S. Farmers spent 13.7 billion dollars in 2012 on agricultural chemicals, 70% higher than in 2002. In soybean fields, weed killer costs have increased $20. per acre over the last five years. Hand weeding—using a hoe—can cost $150. per acre.

Enter Nellie, the weed picking prototype robot. She sees weeds and plucks them the old fashioned mechanical way, one at a time. Since she's a prototype, she plucks a prototype weed.

She runs on carpet, finds a fake plastic weed, picks it and drops it—ready to find another.

This is an entry in Make's "Pitch Your Prototype" competition.

Step 1:

This requires three Arduino Uno's and two Arduino motor shields. A Pixy camera is used for detection of the fake weed and a Ping ultrasonic detector determines closeness of the weed. The base is a Canakit 4 wheel drive robot base. Eleven NiMh AA batteries are used for power. A Jameco servo gripper and a high torque servo are used for gripping. The gear mechanism which holds the gripper was 3d printed (files available at


Step 2:

Control of the robot is a distributed process using three modules. The Pixy camera feeds data to an Arduino processor and that processor sends "nothing, target left, target right, target center" commands to the motor controller module.

The motor controller module receives commands from the Pixy and from the Ping/grabber platform. This Arduino and motor shield causes the robot to go forward, left, right or stop.

The Ping/grabber platform uses an Arduino and a motor shield. When the Ping ultrasonic detector senses something close, it sends a "stop" command to the motor controller, then it extends the grabber, closes the pincer--releases the pincer at the end of travel, retracts the arm, then releases the "stop" command.

Step 3:

Breaking the tasks into discrete functions greatly simplifies the programming and troubleshooting process (at the expense of more hardware cost and greater power drain).



    • Oil Contest

      Oil Contest
    • Clocks Contest

      Clocks Contest
    • Creative Misuse Contest

      Creative Misuse Contest

    13 Discussions

    I'm making a lot of assumptions, but one or two hours per acre is a respectable target time. I'm not a farmer, so there's a lot I have to learn. How often do you weed your potatoes?

    2 hours per acre? That sounds like a lot to me! How much do you think it would take a human to "un-weed" an acre?

    Yes, very wise! I had thoughts about this kind of project myself and started looking into recognition software. There's various different companies doing it but it looked incredibly expensive to buy a license. No doubt there is some way through it with the right kind of perseverance.

    I'd say it would require a swarm of about 10 Nellies to weed an acre in 2 hours, which would be ok for a farmer. There's definitely a future for these kinds of mini weeders and they'd need to be able to recognise different shapes - think facial recognition. You'd soon start to get into some pretty seriously interesting territory. It could possibly be a kickstarter project in my opinion.

    Thanks for your encouraging response! Right now, I'm mainly trying to figure out just how much of my life and resources I'm willing to spend traveling down this path.

    Love this idea! You might need to make a couple of these if there are too many weeds!

    Love this idea! You might need to make a couple of these if there are too many weeds!

    Love this idea! You might need to make a couple of these if there are too many weeds!

    A weeding robot! Nice! The challenge would be to recognize the weed. Then pests like potato beetles. And on. This would aid in the development of healthy farming.


    For the prototype, I'm just considering height--I would need to get into shape recognition to move forward. It's a long distance from this machine to something that can travel bumpy ground, follow the row and exterminate offenders; but those problems appear to be solvable.