Machine Learning Based Hardware Accelerated Cobot Grasping in the Food Industry

Abstract

In food industry automation, the integration of machine learning with robotic systems can introduce improved efficiency and precision in various tasks such as fruit and vegetable handling and manipulation. This paper introduces an innovative approach to robotic grasping, specifically customized for the challenge of identifying and handling apples and oranges within a dynamic and unstructured environment. Utilizing the YOLO (You Only Look Once) object detection algorithm, the system consists of a camera mounted on the robot's end effector, enabling real-time identification and localization of fruits and vegetables. Implemented on the NVIDIA Jetson platform, this solution presents the combination of machine learning techniques with hardware acceleration, ensuring optimal performance in real-time object recognition and manipulation tasks. Our experimental setup tests the robustness of our object recognition and localization approach under varying lighting and orientation scenarios and evaluates the system's overall performance in a real-world, cluttered environment typical of the food industry. The findings of this study underscore the potential of integrating ML-powered vision systems with hardware-accelerated robotics for food handling and manipulation applications. By demonstrating high levels of accuracy in recognition, classification and grasping, our research contributes to the broader field of robotic automation, offering insights into the scalability and adaptability of such systems across different sectors of the food industry.