Sistema robótico para inspeção luminotécnica com estimação de deslocamento baseada em aprendizado profundo

Abstract

This work presents the development of Lumen, a mobile robotic system designed for the automated acquisition of illuminance data in outdoor environments, such as sports fields, integrating sensing and deep learning techniques. The system is equipped with lux meters, an inertial measurement unit (IMU), GNSS/RTK, and a spectral sensor, enabling the synchronized collection of spatial and lighting data. For the generation of illuminance maps, the data are projected into an ENU (East-North-Up) coordinate system and interpolated over the area traversed by the robot. Additionally, a deep learning-based approach is proposed, using LSTM and Transformer architectures to estimate the robot displacement from inertial data, aiming to increase the position update rate relative to GNSS, which, although highly accurate, has a low update frequency, limiting its use in real-time control. By leveraging the higher sampling rate of the IMU, it is possible to obtain more frequent motion estimates, improving system responsiveness during navigation. The models are trained and evaluated in real-world scenarios, demonstrating that the use of highresolution temporal data significantly improves displacement estimation accuracy compared to approaches based on aggregated features. Experiments conducted in different environments demonstrate the system’s ability to generate consistent illuminance maps and to capture spatial patterns, as well as the feasibility of using embedded deep learning to support navigation and environmental analysis.