Adaptive Path Planning for Navigation and Sensing of Micro Aerial Vehicles

Micro Aerial Vehicles (MAVs) have gained much attention as data collection and sensing platforms. Professionals in many fields can access rich sensory data with fast update rates by performing automated aerial surveys using MAVs. However, these robots have limited payloads and short flight times. Therefore, it is useful to perform a task with light and low-powered sensors and as quickly as possible. In this proposal, we consider some of the fundamental tasks performed by MAVs and propose methods by which a MAV can achieve these tasks more efficiently and robustly. In the first part of this thesis, we consider the task of navigation in which a MAV, using visual Simultaneous Localization and Mapping (SLAM) to map the environment and localize itself within it, moves from its current location to a goal location. As SLAM is highly dependent on the visibility of visual features, we propose an adaptive path planning approach that avoids visually-poor regions of the environment and can generate safe trajectories for the MAV to perform the navigation task. In the second part, an aerial coverage task is considered where the MAV must map interesting regions with initially unknown locations. Rather than using an exhaustive ‘lawnmower’ coverage pattern that sweeps the entire region uniformly, we propose non-uniform coverage strategies that adaptively cover all the interesting regions with high resolution and coarsely sweep the rest of the area. The proposed methods generate shorter trajectories compared to a uniform lawnmower pattern. In the last part of the thesis, we assume a time/energy budget for the vehicle and consider specific costs for different manoeuvres of the MAV. We introduce a novel problem of guaranteeing complete coverage of an area at low resolution, while identifying regions of interest (ROIs), and locally surveying as much of the ROIs at high resolution as the battery allows. Three different policies are proposed to decide when to switch between the coarse survey and high resolution imagery data collection.