(Advisor: Prof. Dimitri Mavris)
will defend a doctoral thesis entitled,
"A Framework for Offline Risk-aware Planning of Low-altitude Aerial Flights During Urban Disaster Response"
Friday, September 23
Join Zoom Meeting
Meeting ID: 963 6697 5152
Disaster response missions are dynamic and dangerous events for first responders. Active situational awareness is critical for first responders’ decision-making, and in recent years unmanned aerial assets have successfully extended the quality and range of data collection from sensors. However, literature and industry lack a systematic investigation of the algorithms and datasets for aerial system trajectory planning that optimizes mission performance and guarantees mission success. This work seeks to develop a framework and software environment to investigate the requirements for offline planning algorithms and flight risk models when applied to aerial assets exploring urban disaster zones.
The modular framework develops rapid urban maps, efficient flight planning algorithms, and formal risk metrics demonstrated in scenario-driven experiments using Monte Carlo simulation. First, an experiment compares rapid urban mapping strategies for efficient data processing and storage by independently investigating the obstacle and terrain layers. The methods use open-source data when available and supplement the data with an urban feature prediction model trained through deep learning on satellite imagery. Second, an experiment evaluates sampling-based planners for efficient and effective trajectory planning of nonlinear aerial dynamic systems. The results measure the ability to find collision-free, feasible trajectories using stochastic open-loop control actions from black-box dynamics models. An iterative algorithm updates the open-loop control commands from nominal trajectories to form a lattice of control maneuvers that improves the planning algorithm’s speed and convergence on small maps. Third, a formal risk analysis is added to the planning algorithm to develop a risk-aware planner that considers energy use, obstacle collision, and data collection uncertainty. The risk metrics combine into a single measure, or cost, representing the likelihood of mission failure in a worst-case scenario. The planner demonstrates success on a flood scenario in Atlanta, Georgia, after adding heuristics, conducting parameter tuning, and leveraging a multi-goal technique.
The three modules integrate into a framework where the rapid urban maps and risk-aware planner perform against benchmarks for mission success, performance, and speed. A series of experiments demonstrate the different planners’ robustness and performance and then outline a playbook of the best algorithms for each scenario. Monte Carlo simulation finds limitations for the offline planning algorithms by adding wind from the environment and noise in the system dynamics. Furthermore, the experiments and demonstrations create unique benchmarks from open-source data and software, which should serve as a foundation for increasing the effectiveness of first responders’ safety in the challenging task of urban disaster response.
- Prof. Dimitri Mavris – School of Aerospace Engineering (advisor)
- Prof. Daniel Schrage – School of Aerospace Engineering
- Prof. Polo Chau – School of Computational Science and Engineering
- Prof. Richard Vuduc – School of Computational Science and Engineering
- Dr. Youngjun Choi – United Parcel Service (UPS), Inc.