Tez Türü: Doktora
Tezin Yürütüldüğü Kurum: University of Pennsylvania, School of Engineering and Applied Science, Mechanical Engineering and Applied Mechanics , Amerika Birleşik Devletleri
Tez Danışmanı: Vijay Kumar, Camilo J Taylor
Tezin Onay Tarihi: 2020
Tezin Dili: İngilizce
Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu
Özet:
Multi-rotor Micro Aerial Vehicles (MAV) have become popular robotic platforms in the last decade due to their manufacturability, agility and diverse payload options. Amongst the most promising applications areas of MAVs are inspection, air delivery, surveillance, search and rescue, real estate, entertainment and photography to name a few. While GPS offers an easy solution for outdoor autonomy, using onboard sensors is the only solution for autonomy in constrained indoor environments. In this work, we study onboard state estimation, mapping and navigation of a small MAV equipped with a minimal set of sensors inside GPS-denied, axisymmetric, tunnel-like environments such as penstocks. We primarily focus on state estimators formulated for different sensor suits which include 2D/3D lidars, cameras, and Inertial Measurement Units (IMU). Penstocks are pitch dark environments and offer very weak visual texture even with onboard illumination, hence our estimators primarily rely on lidars and IMU. The point cloud data returned by the lidar consists of
either elliptical contours or indiscriminate partial cylindrical patches making localization along the tunnel axis theoretically impossible. Cameras track features on the walls using the onboard illumination to estimate the velocity along the tunnel axis unobservable to range sensors. Information from all sensors are then fused in a central Kalman Filter for 6 Degrees-of-Freedom (DOF) state estimation. These approaches are validated through onsite experiments conducted in four different dams demonstrating state estimation, environment
mapping, autonomous and shared control.