Authors:

Emanuele Bossi

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Estimating human cognitive states is vital for safe and efficient human- autonomy interaction, especially in safety-critical contexts like automated driving. Cognitive states-such as trust, workload, and perceived risk- affect driver behavior but are not directly observable. They must be inferred from behavioral, physiological, and self-report data. Prior work often relies on static, categorical models with uni-modal inputs, lacking real-time updates, uncertainty quantification, or adaptive querying of user input. We propose a novel modeling framework that treats cognitive states as continuous, time-varying latent variables and automation reliance as a discrete, observable output. Our hybrid dynamical system combines self-reports and psychophysiological signals for real-time state estimation. Critically, the framework quantifies predictive uncertainty, allowing for adaptive self-report scheduling to improve sample efficiency while minimizing participant burden. We address the question: How can online learning and dynamical systems theory be integrated to enable real-time, participant-specific estimation of cognitive states and prediction of automation reliance? Latent states are predicted using an affine dynamical model and mapped to reliance via a tree-based classifier. When self-reports are unavailable but uncertainty is high, a particle filter fuses selected physiological and gaze features using an auxiliary ensemble model to refine estimates. Experimental results demonstrate improved accuracy over static or purely data-driven baselines, highlighting the benefit of integrating dynamical models with machine learning in an online setting. This enables individualized, continuous cognitive state estimation and prediction of reliance on automation, supporting future integration into closed-loop systems for managing driver engagement in dynamic environments. Keywords: Dynamic Modeling; State Estimation; Machine Learning; Human- Autonomy Interaction; Sensor Scheduling

Source:

Purdue University / 2025

Topics:

No topics listed

Co-authors:

Emanuele Bossi