Intent Recognition on Fixed-Wing Aircraft
Author(s)
Grace, Marie
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Human-robot teams have the potential to achieve performance that humans or robots individually would not be capable of. We propose an architecture to implement intent recognition that can be generalized to many cars, aircraft, and watercraft and explored in our study on fixed-wing aircraft using a custom flight simulation environment and human subjects. Based on Frazzoli’s Maneuver Automata, we created a library of predetermined trims and maneuvers. Using this library with a Hidden Markov Model, we ran Viterbi’s Algorithm to identify the most likely flight path a pilot takes. We ran this using auto-generated flight paths and human flight paths against a ground truth. The results of this research indicate that for the auto-generated data, the algorithm can predict the intended flight path, but will return inaccurate results when two maneuvers are very similar to each other. For the human experiment, we found that although the predicted states do not reflect the ground truths well, the predicted states follow the raw states of the pilot. This research provides a first step to implementing a shared control system consisting of Human-AI teaming correcting the intended flight path, which would overall be contributing to greater accuracy and control during a flight.
Sponsor
Date
2021-05
Extent
Resource Type
Text
Resource Subtype
Undergraduate Thesis