ROS2-based robot system with quantitative granular food handling using a regression coefficient estimation-based deep learning model

Le résumé fourni par la source

In food processing environments, robot systems have gained attention as a solution to labor shortages and the need for consistent product quality. In particular, the development of soft robot hands has advanced to enable the safe and efficient handling of delicate food items. However, accurately handling granular foods that are randomly distributed remains a significant technical challenge. Traditional automation methods rely on large-scale dedicated machinery or weigh-and-sell systems, which lack adaptability to diverse food types and conditions. To address this issue, this study proposes a robot system integrated with a deep learning model, enabling high-precision weight prediction even in non-flattened granular food distributions. The proposed system adopts an ROS2-based framework to enhance the diversity and reliability of food handling data. This framework facilitates the integration of diverse food handling datasets across different factories, reducing dataset bias and improving model generalization. In this paper, the proposed system were subjected to experimental validation. The training handling data for chopped green onions has been collected, and the deep learning model has been applied, confirming the overall effectiveness of the system.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.