Visual Scene Understanding for Efficient Cooperative Control of Agricultural Dual-Arm Robots

Abstract

This study introduces an system designed to cooperative control for agricultural dual-arm robots through visual scene understanding. Leveraging images, the system constructs scene graphs and assigns weights to various object attributes, enabling the identification of the most efficient sequence for performing harvesting tasks. Key factors, such as object ripeness, size, and accessibility, are considered to enhance the precision and productivity of robotic harvesting operations. Experimental validation demonstrates the potential of the proposed system in robotic harvesting and promoting advancements in smart agriculture. Future work will be focused on integrating deep learning models for improved object detection, coupling the system with control mechanisms for dual-arm robots to enable seamless task execution, and extending its application to various agricultural tasks beyond harvesting.