AI in Agricultural Robotics and Robotic Weeding

Weeds compete with crops for light, water and nutrients. Farmers have mostly controlled them with broadcast herbicide, spraying the whole field whether a patch has weeds or not. Herbicide-resistant weeds, rising chemical costs and a shortage of workers for hand weeding have pushed growers toward machines that treat only the plants that need it. There are three main approaches: - **Targeted sprayers** mount cameras along a spray boom and open individual nozzles only when a weed is detected. John Deere's See & Spray, built on technology from Blue River Technology (which Deere acquired in 2017), is a well-known example. Others include Bilberry, now part of Trimble, and Ecorobotix, whose ARA sprayer applies very small doses to individual plants. - **Laser weeders**, such as Carbon Robotics' LaserWeeder, use computer vision to find weeds and laser heat to kill them without chemicals. - **Mechanical robots**, from companies like FarmWise and Naïo, use vision to guide blades or tines close to crop plants. The key technical distinction is 'green-on-brown' versus 'green-on-green.' Detecting any green plant on bare soil is relatively easy; older optical sensors did it without deep learning. Telling a weed from a crop plant growing right beside it, often a similar color and size, requires trained neural networks. Harvest robots face a harder problem. Picking strawberries, apples or peppers means finding fruit hidden by leaves, judging ripeness and grasping without bruising. Designs vary widely, including Tevel's flying fruit-picking drones, but harvest automation is less mature than weeding. Three misconceptions are common. Herbicide savings vary widely because they depend on weed density; a heavily infested field saves little. Lasers are not practical for every crop, since throughput and energy use limit where they pay off. And robots usually add to other weed management rather than fully replacing it.