Date of Award

12-2019

Document Type

Thesis - Open Access

Degree Name

Master of Science in Aerospace Engineering

Department

Aerospace Engineering

Committee Chair

Dr. Richard Prazenica

First Committee Member

Dr. Troy Henderson

Second Committee Member

Dr. Hever Moncayo

Abstract

For the past decade, terrain mapping research has focused on ground robots using occupancy grids and tree-like data structures, like Octomap and Quadtrees. Since flight vehicles have different constraints, ground-based terrain mapping research may not be directly applicable to the aerospace industry. To address this issue, Adaptive Learning Terrain Estimation algorithms have been developed with an aim towards aerospace applications. This thesis develops and tests Adaptive Learning Terrain Estimation algorithms using a custom test benchmark on representative aerospace cases: autonomous UAV landing and UAV flight through 3D urban environments. The fundamental objective of this thesis is to investigate the use of Adaptive Learning Terrain Estimation algorithms for aerospace applications and compare their performance to commonly used mapping techniques such as Quadtree and Octomap. To test the algorithms, point clouds were collected and registered in simulation and real environments. Then, the Adaptive Learning, Quadtree, and Octomap algorithms were applied to the data sets, both in real-time and offline. Finally, metrics of map size, accuracy, and running time were developed and implemented to quantify and compare the performance of the algorithms. The results show that Quadtree yields the computationally lightest maps, but it is not suitable for real-time implementation due to its lack of recursiveness. Adaptive Learning maps are computationally efficient due to the use of multiresolution grids. Octomap yields the most detailed maps, but it produces a high computational load. The results of the research show that Adaptive Learning algorithms have significant potential for real-time implementation in aerospace applications. Their low memory load and variable-sized grids make them viable candidates for future research and development.

Share

COinS