With a GPS/INS system, the multi-rotor robot from the Automation Technology Center (ATC) of Hong Kong University of Science and Technology (HKUST) is able to localize itself and fly according to a preprogrammed flight path. However, the robot often meets scenarios, such as canyons or between tall buildings, where GPS signals are partially blocked, or fully denied. Without position and velocity information from the GPS, the robot is not able to localize itself, and thus risks crashing when the robot is flying autonomously, especially for a mission beyond the visual line of sight of the pilot. This research is motivated by the demands for accurate navigation for a more extensive flight area: from outdoor to semi-blocked regions, and then to indoor regions.

A stand-alone Global Na...[ Read more ]

With a GPS/INS system, the multi-rotor robot from the Automation Technology Center (ATC) of Hong Kong University of Science and Technology (HKUST) is able to localize itself and fly according to a preprogrammed flight path. However, the robot often meets scenarios, such as canyons or between tall buildings, where GPS signals are partially blocked, or fully denied. Without position and velocity information from the GPS, the robot is not able to localize itself, and thus risks crashing when the robot is flying autonomously, especially for a mission beyond the visual line of sight of the pilot. This research is motivated by the demands for accurate navigation for a more extensive flight area: from outdoor to semi-blocked regions, and then to indoor regions.

A stand-alone Global Navigation Satellite System (GNSS) receiver has many limitations for UAV navigation, including long delay, low sample rate and low accuracy. This prompted the navigation community to augment it with additional sensors, in particular the Inertial Navigation System(INS), which consists of accelerometers and gyroscopes. In this thesis, the GNSS and INS have been integrated in two architectures: namely loosely-coupled (LC) and tightly-coupled (TC). An LC system is easier to implement and meets the navigation demand in most cases, but it has the disadvantage of requiring to track a minimum of four satellites to provide the navigation output, which will not be easily met in a semi-blocked region, such as a big canyon. TC architecture, on the other hand, is able to relax the minimum satellite requirements and extend the flight area to these semi-blocked districts.

To extend the flight area further to the indoors, we explore the use of laser sensors and camera sensors for these GPS denied areas. For the laser-based system, a 2D laser scanner was set up on a wheeled robot, and achieved Simultaneous Localization And Mapping (SLAM) results based on a particle filter. The localization algorithm was then extended to 3D by rotating the 2D laser scanner to provide 3D environment measurements. However, the laser sensors and its processor are too heavy to be carried onboard our flying robot. On our multi-rotor platform, light-weight camera sensors were adopted. A mono-vision navigation system was built on a quad-rotor robot, and autonomous flight of the robot was achieved together with the inertial measurement unit (IMU). In this system, images were sent via a wireless data link back to the ground station, where image processing, control feedback, and path planning algorithms were performed. To improve the system’s performance and robustness, two tightly-coupled vision/INS systems were built. The first system was a mono-vision/INS system set up on a quad-rotor robot, and the other was a stereo-vision/INS system equipped on a hex-rotor robot. For both systems, image processing, fusion and control algorithm were all implemented on board with microprocessors. Experiments and field-tests demonstrated that both systems were able to achieve accurate state estimation and good control effects in both hovering and dynamic flight.