Our software is expert at detecting and locating multiple thin and small objects in a very complex scene. This can be peduncle, petiole, stem, ropes, or even the trellis system. Massive data collection and augmentation are the key to achieving a robust performance.
Soft grippers are widely proposed for picking different crops. In agriculture, soft robotics is a buzzing word for farmers. However, deep learning today is powerful enough to detect peduncles directly. Harvesting without touching avoids bruising the skin and spread of diseases. It often makes harvesting faster and some crops last longer with their peduncles attached.
Our software can track and pick targets repeatedly. This can be made even faster by using a more agile robotic arm. The algorithm works for moving objects as well, but in reality, crops are static. We use this feature to speed up the harvest process when the trolley is progressing forward.
Our software is not limited by crop shape as it goes for the peduncle directly. Crop detection is indeed part of the process as well to estimate the ripeness level. We support different types of tomatoes and grapes today as cutting leaves is a common practice for most of these types. However, our next-generation software will look behind the leaves and thus unlocks harvest automation with a high success rate for many other crops including cucumber, pepper bell, and eggplants.