Cooperative UAS Forest Navigation With Feature Based SLAM

Within forest applications, Unmanned Aircraft Systems (UAS) are highly demanded. However, in forest environments conventional navigation systems that rely on a Global Navigation Satellite System (GNSS) are exposed to navigation performance degradation due to the forest canopy. Within this work, 2D Light Detection and Ranging (LiDAR) scanner equipped UAS swarms explore an unknown forest environment. Each UAS generates its own map estimate based on tree features, that are detected within the LiDAR point cloud. Using an Inertial Navigation System (INS) mechanization, an attitude estimate is calculated that is then used to project the features into the horizontal plane. While one UAS has the reference role and shares its map information, all other UAS are, initially, in the discovery role. These UAS make use of a bootstrap particle filter to localize themselves within the reference map. Once converged, they switch to an exploration role and can add or update features of the reference map. Thereby, the uncertainty of map feature positions is characterized and updated. Simulation and experimental test scenarios are presented, where the performance of the proposed method is demonstrated for different speed scenarios up to 16m/s. It is shown that the cooperative exploration of the forest environment yields a faster and more confident map of the forest. Additionally, the navigation accuracy is found to be 40cm at a maximum over a 225m long track while the noise is smaller than 3cm. Even though drift is present, relative navigation and separation can be ensured if UAS operate close to each other enabling a collision avoidance functionality.