What can a robot dog see beneath potato plants that a drone misses?
A field can look healthy from above while something is going wrong near the soil. That's a lovely reason to give a camera legs: get a view the grower isn't already getting. In a November 16 Mid-West Farm Report interview, UW–Madison researcher Yi Wang describes using a quadruped robot in potato fields to look below the canopy for potential pests and disease signs. The published transcript also describes onion images being taken back to the office for machine-learning analysis. Collecting pictures and delivering an answer are still separate jobs. I'd want to try one very ordinary task with a grower: take this flagged image and find the plant again. Does it lead them to a particular row and spot, or leave them walking around trying to recognise a leaf? Then record whether that visit found something worth acting on, and how long it took. That's a trial I'd propose, not a result the interview establishes. I like a robot whose useful contribution might be a closer look rather than another machine making the treatment decision. Wang explicitly keeps management decisions with growers. If the camera can spare someone an unnecessary round of the field and direct the next visit somewhere useful, that would be help worth having.
Comments
I'd have a crop scout inspect some plants the robot didn't flag, too. Finding a convincing damaged leaf tells us something different from knowing how often a problem goes unnoticed. Wang describes looking for potential beetle and disease problems; that alone doesn't measure the misses. For a field trial, compare those unflagged plants with a scout's inspection and keep ‘not photographed clearly’ separate from ‘photographed, no problem detected’. That's a check I'd propose, not a result reported here. A grower deciding whether to skip a walking round needs to know what a quiet report means, not just whether the flagged pictures are useful.
I'd put those missing views on the field map, rather than leave them buried in an image report. A robot reaching the end of a row doesn't mean its camera got a useful look beneath every plant. For the trial, I'd mark stretches needing another look and let the grower choose whether to send the robot back or walk them. Otherwise someone may have to repeat the whole round just to find out which bits the machine couldn't see.