Quacky: Design of an Inspection Robot for High-Risk Peruvian Buildings

With the rising rate of unlicensed constructions in Peru due to informal expansion and the high likelihood of earthquakes, regular inspection of walls and structures in high-risk buildings is essential. Conventional manual inspection methods, which heavily rely on the specialist's experience, lack objectivity and precision. This paper proposes the design of an inspection robot for high-risk buildings in Peru that can be remotely operated by a specialist to ensure their safety. The robot captures images, which are then processed for crack detection and sent to the specialist for analyzing the habitability of the buildings. A Convolutional Neural Network (CNN) model, using the YOLO V9 architecture, was trained with images of concrete cracks, employing the lightest version YOLOv9-t. Model validation yielded promising results, with a precision of 95.12% and an F1-Score of 96.51% using the Adamax optimizer, demonstrating its effectiveness in identifying structural anomalies. Additionally, a Simultaneous Localization and Mapping (SLAM) algorithm was integrated into the test robot to generate a map of the inspected terrain, achieving the best performance with an adaptive proportional controller and a Mean Squared Error (MSE) of 0.72 and 0.15 for the x and y axes, respectively.