Automating Grape Thinning: Predicting Robotic Arm End-effector Positions Using Depth Sensing Technology and Neural Networks

Abstract

Grape thinning is a labor-intensive task in viticulture that involves selectively removing grape clusters to improve overall fruit quality. The integration of robotics and artificial intelligence (AI) techniques has shown great potential in automating this process, reducing labor costs and increasing efficiency. We propose a novel approach for automating grape thinning using depth sensing and neural networks. The approach was evaluated in both indoor and real-world field environments, achieving a success rate of 97% and 90% in accurately approaching the target berries, respectively. However, the success rate of removing the target berry was lower, at around 60% in real-world field environments. This is primarily due to the gripper not being thin enough to approach the target berry for removal when the bunch is dense. The proposed approach is a promising solution for automating grape thinning operations. However, there are still some challenges that need to be addressed, such as the use of a more reliable depth sensor and the development of a more robust gripper. Future work will focus on real-time implementation, optimization, and further assessments under diverse vineyard conditions.