In this thesis we aim to bridge the gap between high quality map reconstruction and Unmanned Aerial Vehicles (UAVs) SE(3) motion planning in challenging environments with narrow openings, such as disaster areas, which requires attitude to be considered. We propose an efficient system that leverages the concept of adaptive-resolution volumetric mapping, which naturally integrates with the hierarchical decomposition of space in an octree data structure. Instead of a Truncated Signed Distance Function (TSDF), we adopt mapping of occupancy probabilities in log-odds representation, which allows representation of both surfaces, as well as the entire free, i.e.\ observed space, as opposed to unobserved space. We introduce a method for choosing resolution -on the fly- in real-time by means of a multi-scale max-min pooling of the input depth image. The notion of explicit free space mapping paired with the spatial hierarchy in the data structure, as well as map resolution, allows for collision queries, as needed for robot motion planning, at unprecedented speed. Our mapping strategy supports pinhole cameras as well as spherical sensor models. Additionally, we introduce a first-of-a-kind global minimum cost path search method based on A* that considers attitude along the path. State-of-the-art methods incorporate attitude only in the refinement stage. To make the problem tractable, our method exploits an adaptive and coarse-to-fine approach using global and local A* runs, plus an efficient method to introduce the UAV attitude in the process. We integrate our method with an SE(3) trajectory optimisation method based on a safe-flight-corridor, yielding a complete path planning pipeline.
We quantitatively evaluate our mapping strategy in terms of mapping accuracy, memory, runtime performance, and planning performance showing improvements over the state-of-the-art, particularly in cases requiring high resolution maps. Furthermore, extensive evaluation is undertaken using the AirSim flight simulator under closed loop control in a set of randomised maps, allowing us to quantitatively assess our path initialisation method. We show that it achieves significantly higher success rates than the baselines, at a reduced computational burden.