Atmospherically Aware Aircraft Guidance Using In Situ Observations
Open Access
- Author:
- Bird, John Joseph
- Graduate Program:
- Aerospace Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- May 09, 2019
- Committee Members:
- Jacob Willem Langelaan, Dissertation Advisor/Co-Advisor
Jacob Willem Langelaan, Committee Chair/Co-Chair
Mark David Maughmer, Committee Member
Sean N Brennan, Committee Member
George Spencer Young, Outside Member - Keywords:
- UAV
drone
atmospherically aware
autonomy
reinforcement learning
flight planning
soaring - Abstract:
- A major challenge to widespread use of small electric UAS is their limited energy budgets and sensitivity to environmental conditions. Differences between forecast and realized weather conditions are often large enough that the energy required to complete a mission can differ significantly from that expected by \textit{a priori} flight plans. This makes autonomous awareness of and response to the environmental state necessary in order to enable small UAS to conduct long-range missions. To develop this capability, the uncertain atmospheric state is decomposed into stochastic and systematic uncertainty which can be managed separately by the aircraft using speed and power output commands respectively. Speed variations enable an aircraft to preferentially spend longer in favorable areas, making the atmosphere appear more conducive to UAS flight. A speed command is developed which responds optimally to stochastic conditions while meeting a specified arrival time at a destination. A modeling system is developed which allows the aircraft to build a model of the vertical structure of the atmosphere using \textit{in situ} observations and to transform the atmospheric state into a mission performance cost. This model is then employed in a multi-armed bandit inspired trajectory planner which balances reducing the aircraft's mission performance cost with maintaining a model of the environment. The planner solves the exploration-exploitation problem without employing heuristics to value information and demonstrates good performance and robustness when compared to a collocation-based optimal planner. Together these techniques compose a system which allows small UAS to reduce their energy expenditure and the effect of the uncertain atmospheric state on their performance. Components of this system are tested in simulation and flight experiments and demonstrate an improvement in the ability of small UAS to operate in unknown and uncertain environments