Row Following with the help of AI

One project that we are currently working on at Werkr Tractors is developing row-detection for our tractors so we can weed autonomously, with high precision.

Right now, this involves collecting as much crop data as possible. We are using our test plots, where we are currently growing onions, salad mix, beets, carrots and green onions, to record images while we finger weed or basket weed. We have also taken our tractor out to a neighboring farm to collect images of their crops, increasing the diversity of the data to strengthen our model.

Below is an example of the images we are recording using Stereolabs' camera mounted on our tractor.

This data is then processed and used to train an AI model, which differentiates the crop from the soil (detecting the rows). The video below shows the models' estimation of where the rows are located. It is based on only the small amount of data we have collected so far, and as we continue to get more images, the model will become more robust.