Weeding Robots Needed a New Brain for Every Crop. Not Any More
A weeding robot has to decide, plant by plant at several kilometres an hour, which green thing is the crop and which is the weed that looks like it. Until recently that meant a separate trained model for every crop, every growth stage and every region, which is the real reason agricultural robots stayed niche. One model for all plants, corrected by the farmer with three photographs on a tablet, is a different proposition. Here is the pattern, and what has not been measured.
