Navegação de robôs terrestres em culturas agrícolas utilizando redes neurais em dados LiDAR

Abstract

Population growth disproportionate to the increase in agricultural production requires the modernization of technologies used in agriculture. In this field, terrestrial robotics stands out for its versatility of applications to increase the sector's productivity. Featuring versatility and accessibility in a single platform, the TerraSentia robot is a solution for under-canopy navigation, an environment in which the lack of confidence in sensors and the irregularity of the scenery make navigation difficult, and robot control. Thus, the use of a 2D LiDAR (Light Detection and Ranging) emerges as a resource to overcome the described adversities, in addition to enabling data collection in low-light environments. That said, this Scientific Initiation project proposes the use of a Convolutional Neural Network in data from a 2D LiDAR sensor for navigation of the TerraSentia platform. Thus, it is proposed to create 2 systems (data processing Pre Neural Network and the Neural Network itself). Therefore, TerraSentia's performance is expected to increase in under-canopy experiments, compared to the state of art through algorithms using only heuristics in place of the Neural Network. (AU)