Date of this Version
Thanks to the technology breakthroughs in robot autonomy, autonomous unmanned aerial vehicles (UAVs) have been utilized for remote surveying and inspection of various applications. Nonetheless, fully autonomous aerial inspection still has some important limitations, especially when inspecting complex three-dimensional bodies. This is complicated by the limited battery capacity, restricted onboard perception, and the complexity of the target geometry. One of the key elements to overcome these challenges is to design appropriate aerial inspection paths. Over the last decade, different path planning algorithms have been proposed for finding the shortest paths with a full sensory coverage of the environment, and those paths have been employed on both submarine, ground and aerial robots. Although current algorithms have presented a great performance on reducing the robotic operational cost (e.g. distance, energy consumption, risk level), few existing studies considered the quality of data collection in path generation. Such limitation may induce additional workload of post-processing or may even require excessive flights that voids the claim that UAVs are more effective and efficient than the conventional ground-level methods.
To this end, a new coverage path planning (CPP) algorithm is introduced in this thesis that supports a high-quality vision-based aerial inspection by incorporating the image quality into the path generation. Specially, we pose the problem into an optimization framework to compute the optimal trajectory that balances the flight efficiency against the image quality. The algorithm outputs the resulting trajectory that can be safely and autonomously executed by most commercially available drones for both image-based 3D scanning and vision-based inspection tasks. Simulation test cases are provided to thoroughly evaluate the performance of the proposed algorithm. And the effects of various parameter configurations on the optimization results are also discussed through these test cases.
Advisor: Justin Bradley