| Resumo: | The end goal of this dissertation is to develop an autonomous exploration robot that is capable of choosing the Next Best View which reveals the most amount of information about a given volume. The exploration solution is based on a robotic manipulator, a RGB-D sensor and ROS. The manipulator provides movement while the sensor evaluates the scene in its Field of View. Using an OcTree implementation to reconstruct the environment, the portions of the de ned exploration volume where no information has been gathered yet are segmented. This segmentation (or clustering) will help on the pose sampling operation in the sense that all generated poses are plausible. Ray casting is performed, either based on the sensor's resolution or the characteristics of the unknown scene, to assess the pose quality. The pose that is estimated to provide the evaluation of the highest amount of unknown space is the one chosen to be visited next, i.e., the Next Best View. The exploration reaches its end when all the unknown voxels have been evaluated or, those who were not, are not possible to be measured by any reachable pose. Two case studies are presented to test the performance and adaptability of this work. The developed system is able to explore a given scene which, initially, it has no information about. The solution provided is, not only, adaptable to changes in the environment during the exploration, but also, portable to other manipualtors rather than the one used in the development.
|
|---|