Date of Award
Doctor of Philosophy
The use of electrical vertical takeoff and landing (eVTOL) aircraft to provide efficient, high-speed, on-demand air transportation within a metropolitan area is a topic of increasing interest, which is expected to bring fundamental changes to the city infrastructures and daily commutes. NASA, Uber, and Airbus have been exploring this exciting concept of Urban Air Mobility (UAM), which has the potential to provide meaningful door-to-door trip time savings compared with automobiles. However, successfully bringing such vehicles and airspace operations to fruition will require introducing orders-of-magnitude more aircraft to a given airspace volume, and the ability to manage many of these eVTOL aircraft safely in a congested urban area presents a challenge unprecedented in air traffic management. Although there are existing solutions for communication technology, onboard computing capability, and sensor technology, the computation guidance algorithm to enable safe, efficient, and scalable flight operations for dense self-organizing air traffic still remains an open question. In order to enable safe and efficient autonomous on-demand free flight operations in this UAM concept, a suite of tools in learning-based perception and control systems with stress testing for safe autonomous air mobility is proposed in this dissertation.
First, a key component for the safe autonomous operation of unmanned aircraft is an effective onboard perception system, which will support sense-and-avoid functions. For example, in a package delivery mission, or an emergency landing event, pedestrian detection could help unmanned aircraft with safe landing zone identification. In this dissertation, we developed a deep-learning-based onboard computer vision algorithm on unmanned aircraft for pedestrian detection and tracking. In contrast with existing research with ground-level pedestrian detection, the developed algorithm achieves highly accurate multiple pedestrian detection from a bird-eye view, when both the pedestrians and the aircraft platform are moving.
Second, for the aircraft guidance, a message-based decentralized computational guidance algorithm with separation assurance capability for single aircraft case and multiple cooperative aircraft case is designed and analyzed in this dissertation. The algorithm proposed in this work is to formulate this problem as a Markov Decision Process (MDP) and solve it using an online algorithm Monte Carlo Tree Search (MCTS). For the multiple cooperative aircraft case, a novel coordination strategy is introduced by using the logit level-$k$ model in behavioral game theory. To achieve higher scalability, we introduce the airspace sector concept into the UAM environment by dividing the airspace into sectors, so that each aircraft only needs to coordinate with aircraft in the same sector. At each decision step, all of the aircraft will run the proposed computational guidance algorithm onboard, which can guide all the aircraft to their respective destinations while avoiding potential conflicts among them. In addition, to make the proposed algorithm more practical, we also consider the communication constraints and communication loss among the aircraft by modifying our computational guidance algorithms given certain communication constraints (time, bandwidth, and communication loss) and designing air-to-air and air-to-ground communication frameworks to facilitate the computational guidance algorithm.
To demonstrate the performance of the proposed computational guidance algorithm, a free-flight airspace simulator that incorporates environment uncertainty is built in an OpenAI Gym environment. Numerical experiment results over several case studies including the roundabout test problem show that the proposed computational guidance algorithm has promising performance even with the high-density air traffic case.
Third, to ensure the developed autonomous systems meet the high safety standards of aviation, we propose a novel, simulation driven approach for validation that can automatically discover the failure modes of a decision-making system, and optimize the parameters that configure the system to improve its safety performance. Using simulation, we demonstrate that the proposed validation algorithm is able to discover failure modes in the system that would be challenging for humans to find and fix, and we show how the algorithm can learn from these failure modes to improve the performance of the decision-making system under test.
Yang, Xuxi, "Learning-based perception and control with adaptive stress testing for safe autonomous air mobility" (2020). Graduate Theses and Dissertations. 17884.