CAREER: Enabling Human-Aware and Responsive Automation through Cognitive State Modeling and Estimation

Purdue University, West Lafayette IN

Investigators

Abstract

This Faculty Early Career Development Program (CAREER) grant will fund research that enables autonomous systems such as machines, robots, and vehicles, to respond safely and collaboratively to human interactions, thereby promoting the progress of science, and advancing the national prosperity and welfare. Significant increases in automation are expected across the healthcare, manufacturing, and transportation sectors. Major challenges to the safe integration of automation in situations involving human participants are the lack of accurate mathematical models of human behavior, intent, and cognitive state, as well as of reliable ways of gathering relevant real-time data from humans. This project overcomes these challenges by building a new modeling framework that accounts for human cognitive constructs established within the social sciences, such as trust, workload, perceived risk, and self-confidence, while being amenable to rigorous mathematical analysis. It demonstrates how this framework can facilitate the design of sensing and control algorithms that can allow an autonomous system to determine, for example, how confident the human is in taking over a task, say during driving of an autonomous vehicle. Through close integration of research and education, the project will train engineering students to tackle questions surrounding socio-political and ethical challenges associated with the rapid expansion of automation in society. This will be achieved through new interdisciplinary coursework and recurrent immersive learning experiences that bring students together with state and federal policymakers. This research aims to develop the foundations of a control-theoretic framework for human-automation interaction that can use real-time data to continually improve prediction accuracy. It achieves this aim by defining a human cognitive state space and characterizing its dynamics in a model formulation that is compatible with standard tools of control design. A unique feature of this formulation is its interpretability, maintained by grounding the model in established conceptual frameworks governing human decision-making. A principled methodology for real-time parameter and cognitive state estimation will be created that blends information from multiple sensors with different capabilities and costs of querying, and enables adaptation to different individuals. A generalizable technique for exciting human cognitive dynamics, as well as guidelines for choosing the cognitive state estimation algorithm best suited for a particular human-automation interaction context, will be established. Laboratory and field experiments, including tests using trucks equipped with Level 1 driver assistance features, are planned for validation of the modeling framework and estimation algorithms. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.