Is this project an undergraduate, graduate, or faculty project?
Graduate
Project Type
group
Campus
Daytona Beach
Authors' Class Standing
Rocio Jado Puente, Graduate student Michael Budihartono, Graduate student Eduar Cabrera Gaspar, Graduate student
Lead Presenter's Name
Rocio Jado Puente
Lead Presenter's College
DB College of Engineering
Faculty Mentor Name
Dr. Hever Moncayo
Abstract
Assured Learning for Intelligent Dynamic Systems: A Metacognitive Framework Advanced Air Mobility systems, including electric vertical takeoff and landing (eVTOL) aircraft and autonomous drone platforms, require increasingly high levels of autonomy and safety. Meeting these demands calls for intelligent systems that can adapt in real time to uncertainty and changing environmental conditions. However, traditional certification methods are not well suited to rigorously measure or quantitatively verify the performance of online learning components because of their non-deterministic behavior. This work introduces a novel runtime safety assurance method based on a metacognitive architecture (MCA) that supervises and regulates learning-enabled components. The approach uses concept algebra as a mathematical framework for knowledge representation and formalizes the notion of learning power through a cognitive index. This index quantitatively measures how closely current learning behavior aligns with a desired healthy regime, providing a principled and mathematically grounded way to monitor adaptation quality. The MCA continuously evaluates onboard learning modules, such as adaptive health monitoring and intelligent guidance and control systems, and intervenes when necessary by adjusting internal parameters or learning strategies to improve performance and maintain safe operation. To address the high dimensionality of this supervision task, we also propose a quantitative cognitive performance metric that operates within the metacognitive layer, enabling continuous monitoring of both learning behavior and system safety in a reduced subspace. Dimensionality reduction techniques are used to project the original feature space into a compact, interpretable representation while preserving behavioral fidelity. Finally, we present a reduced case study in which the learning power and cognitive index of an intelligent controller are computed and analyzed under both nominal and off-nominal conditions. The results demonstrate the potential of the proposed approach for runtime assurance in safety-critical, learning-enabled systems.
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
Assured Learning for Intelligent Dynamic Systems: A Metacognitive Framework
Assured Learning for Intelligent Dynamic Systems: A Metacognitive Framework Advanced Air Mobility systems, including electric vertical takeoff and landing (eVTOL) aircraft and autonomous drone platforms, require increasingly high levels of autonomy and safety. Meeting these demands calls for intelligent systems that can adapt in real time to uncertainty and changing environmental conditions. However, traditional certification methods are not well suited to rigorously measure or quantitatively verify the performance of online learning components because of their non-deterministic behavior. This work introduces a novel runtime safety assurance method based on a metacognitive architecture (MCA) that supervises and regulates learning-enabled components. The approach uses concept algebra as a mathematical framework for knowledge representation and formalizes the notion of learning power through a cognitive index. This index quantitatively measures how closely current learning behavior aligns with a desired healthy regime, providing a principled and mathematically grounded way to monitor adaptation quality. The MCA continuously evaluates onboard learning modules, such as adaptive health monitoring and intelligent guidance and control systems, and intervenes when necessary by adjusting internal parameters or learning strategies to improve performance and maintain safe operation. To address the high dimensionality of this supervision task, we also propose a quantitative cognitive performance metric that operates within the metacognitive layer, enabling continuous monitoring of both learning behavior and system safety in a reduced subspace. Dimensionality reduction techniques are used to project the original feature space into a compact, interpretable representation while preserving behavioral fidelity. Finally, we present a reduced case study in which the learning power and cognitive index of an intelligent controller are computed and analyzed under both nominal and off-nominal conditions. The results demonstrate the potential of the proposed approach for runtime assurance in safety-critical, learning-enabled systems.