Sensor de escaneamento omnidirecional e mapeamento por janela deslizante

Abstract

LIDAR sensors are commonly used in robotics to perform 2D mapping. Some projects go beyond that and spin the sensor in front of the robot to build a point cloud with which it’s possible to apply 3D mapping techniques. This work tries to improve upon this idea and create a novel mapping strategy by rotating LIDAR for acquisition, scanning the environment and building a surface that represents the mobile robot’s neighborhood with accuracy. A base vertical grid in front of the robot is created to represent the acquired points (local map) and this grid is used to build two elevation maps, one representing the floor (global map) and other representing the ceiling. The resulting map is compatible with common SLAM and mapping methods (occupation map). Several difficulties were faced when dealing with problems intrinsic from the scanning strategy, as using movement compensation to tackle the low update frequency inherent to LIDAR sensors. A prototype was developed to validate the mapping strategy. Navigation experiments were carried out using the new mapping strategy and the results are presented on this work.