plant bot is a time based interactive art installation where the fates of a living plant and a computer are interdependent. Essentially the plant attempts to train a computer using image recognition. Through this process the computer will learn to recognize when the plant needs water based on images it takes of the plant. If the plant appears healthy, the computer will maintain a regular water regiment. If the plant does not appear healthy to the computer it will attempt to aid the plant by adjusting to what it “thinks” the plant needs based on the images gathered. As the computer becomes more intelligent and hence more adept at caring for the plant, the plant will conceivably thrive and grow in proportion. If the computer is unsuccessful, conceivably the opposite will occur.
This installation consists of a live house plant, a computer and a robotic arm with a camera attached. The robotic arm with the camera observes the plant using and artificial intelligence object detection model running on the computer. It is looking for happy or dead leaves as the plant rotates on a turntable. The robot moves relative to the size, location and type of leaf it sees (happy or dead). If 10 or more leaves are detected this image of the plant and it’s happy/dead leaf annotation boxes are recorded by the computer. When 1000 of these images are collected, the computer uses the images and annotations to retrain the artificial intelligence model. At the same time, the system uses the ratio of happy to dead images to determine how much water to give the plant. For example, if more dead leaves are detected, the plant will receive proportionally more or less water.
This work uses: tensorflow object detection API, python, arduino
