Our manipulation solution is capable of removing leaves around the crops multiple times throughout the season. The computer vision part of this task is completed and two different cutting end-effectors are being developed for tomatoes and grapes. Updates will come in the summer of 2022.

Scene classification

With a normal RGB camera, the robot can classify objects in the scene. We use deep learning extensively to handle very complex backgrounds in outdoor environments. Robustness to fruit variety, lighting conditions, sun direction, and camera location are other challenges we can solve. In addition to the network's ability to distinguish objects, our software can estimate stem thickness and ensure it's not going for the main tomato trunk.

Structural analysis

To cut large leaves, a precise understanding of the plant structure is needed. Our software is trained to detect and analyze complex scenes in poor lighting conditions, often difficult for human eyes too. Each plant type requires a different strategy to de-leaf, and a generic manipulator enables us to provide a versatile solution.