Plastering Location Recognition System with a Robot Arm

Traditional Japanese plastering is a craft that requires extensive training and significant dedication to master [1]. However, due to a declining population, fewer individuals are pursuing this profession. To address this challenge, two tasks need to be considered: digitizing traditional Japanese plastering techniques to ensure their preservation, and enabling robots to perform traditional Japanese plastering. To simplify the problem, this study focuses on arm and wrist movements. While some robot arms have been proposed [14] [15], this study aims to address not only the arm and wrist movements but also the interaction between the wrist and the trowel. Improper use of the trowel can result in mud failing to adhere to the wall or detaching immediately after application. Therefore, addressing these challenges is critical to ensuring successful plastering using traditional tools and materials. This paper proposes a recognition system to accurately determine the plastering locations with a multi-channel deep learning model. The model is trained on a diverse dataset that includes images with varying resolutions and lighting conditions. Based on the identified plastering locations, the robot arm can decide whether to stop or continue plastering. The proposed platering location recognition system evalutes two conditions under wich platering should contiue:(1) parts of the wall lack mud, and (2) mud has detached from the wall. If the wall is fully covered with mud, the system signals the robot arm stop platering. Experimental results demonstrate that the system achieves an accuracy of almost $97 \%$ in determining whether a location requires plastering.