Natural Language Processing for Unmanned Aerial Vehicle Guidance Interfaces

In this thesis, I address the opportunities and challenges involved in applying natural language processing techniques to the control of unmanned aerial vehicles (UAVs). The problem of controlling an unmanned aircraft via natural language inputs is formulated as a feedback control problem, and notions of state, controllability, and observability are defined under this model. An implementation of such a system is also described. The phraseology of the existing air traffic control language is used as a base command set; to form this command set, a corpus of air traffic control commands was gathered from recorded exchanges between pilots and controllers at Boston’s Logan Airport, as well as Hanscom Field in Bedford, MA, and these were used as the “target language” for this implementation. Knowledge of air traffic control and airport operations, combined with existing natural language processing techniques, is used to achieve a higher recognition success rate than a traditional natural language processor designed for a more general domain of discourse would. This is the first known attempt at formalizing air traffic control phraseology for use in an unmanned system. Thesis Supervisor: Eric Feron Title: Associate Professor Department of Aeronautics and Astronautics Thesis Supervisor: Robert C. Berwick Title: Professor Department of Electrical Engineering and Computer Science