Predictive models of human supervisory control behavioral patterns using hidden semi-Markov models

Abstract

Behavioral models of human operators engaged in complex, time-critical high-risk domains, such as those typical in Human Supervisory Control (HSC) settings, are of great value because of the high cost of operator failure. We propose that Hidden Semi-Markov Models (HSMMs) can be employed to model behaviors of operators in HSC settings where there is some intermittent human interaction with a system via a set of external controls. While regular Hidden Markov Models (HMMs) can be used to model operator behavior, HSMMs are particularly suited to time-critical supervisory control domains due to their explicit representation of state duration. Using HSMMs, we demonstrate in an unmanned vehicle supervisory control environment that such models can accurately predict future operator behavior both in terms of states and durations.