Is this project an undergraduate, graduate, or faculty project?
Graduate
Project Type
group
Campus
Daytona Beach
Authors' Class Standing
Shawn de la Osa, Graduate Student
Lead Presenter's Name
Shawn de la Osa
Lead Presenter's College
DB College of Arts and Sciences
Faculty Mentor Name
Chad Tossell
Abstract
This research explores AI-based decision support in light general aviation (e.g., Cessna 172, Diamond), where onboard hardware is typically too limited for modern AI. In aviation, AI has the potential to act as a “copilot” supporting planning, execution, and assessment, particularly since single pilots must simultaneously aviate, navigate, and communicate under high workload. However, although AI adoption is rapidly expanding, many implementations are introduced without fully incorporating human-centered design processes that account for pilot workflow and workload. We leverage a human factors process in this research. First, we design and evaluate a cloud-based AI copilot . The system leverages Starlink Aviation connectivity and the pilot’s existing tablet interface, which is already widely used for flight planning and navigation. Running the AI in the cloud avoids heavy onboard processing while enabling real-time support. The assistant will follow standard ICAO/VFR radio phraseology and function as an AI teammate for routine tasks such as assisting with ATC interpretation or checklist guidance. The backend will fine-tune a large language model using aircraft-specific knowledge sources, including Cessna 172 and Diamond Pilot Operating Handbooks (POH), flight manuals, and FAA guidance. Domain-specific training allows the model to respond accurately to operational cockpit queries. Frontend development follows a systematic, user-centered design process. Development of the high-fidelity tablet prototype is guided by aviation human-factors principles for cockpit displays. The AI assistant will provide concise visual guidance on the tablet and optional auditory outputs aligned with aviation radio phraseology. Because pilots already experience high visual workload in the cockpit, auditory support may reduce attention demands and improve task management. This will be evaluated in a future test comparing visual-only versus visual+auditory feedback for usability, workload, and response timing. Our project explores how cloud-based AI can be integrated into light aircraft using existing pilot tablets and satellite connectivity and explores a research-based and practical path toward safe human-AI teaming in general aviation.
Did this research project receive funding support (Spark, SURF, Research Abroad, Student Internal Grants, Collaborative, Climbing, or Ignite Grants) from the Office of Undergraduate Research?
No
Prototype AI Assistant for Private and Student Pilots
This research explores AI-based decision support in light general aviation (e.g., Cessna 172, Diamond), where onboard hardware is typically too limited for modern AI. In aviation, AI has the potential to act as a “copilot” supporting planning, execution, and assessment, particularly since single pilots must simultaneously aviate, navigate, and communicate under high workload. However, although AI adoption is rapidly expanding, many implementations are introduced without fully incorporating human-centered design processes that account for pilot workflow and workload. We leverage a human factors process in this research. First, we design and evaluate a cloud-based AI copilot . The system leverages Starlink Aviation connectivity and the pilot’s existing tablet interface, which is already widely used for flight planning and navigation. Running the AI in the cloud avoids heavy onboard processing while enabling real-time support. The assistant will follow standard ICAO/VFR radio phraseology and function as an AI teammate for routine tasks such as assisting with ATC interpretation or checklist guidance. The backend will fine-tune a large language model using aircraft-specific knowledge sources, including Cessna 172 and Diamond Pilot Operating Handbooks (POH), flight manuals, and FAA guidance. Domain-specific training allows the model to respond accurately to operational cockpit queries. Frontend development follows a systematic, user-centered design process. Development of the high-fidelity tablet prototype is guided by aviation human-factors principles for cockpit displays. The AI assistant will provide concise visual guidance on the tablet and optional auditory outputs aligned with aviation radio phraseology. Because pilots already experience high visual workload in the cockpit, auditory support may reduce attention demands and improve task management. This will be evaluated in a future test comparing visual-only versus visual+auditory feedback for usability, workload, and response timing. Our project explores how cloud-based AI can be integrated into light aircraft using existing pilot tablets and satellite connectivity and explores a research-based and practical path toward safe human-AI teaming in general aviation.