Abstract
Human-autonomy teaming is an increasingly integral component of operational environments, including crewed and remotely operated space missions, military settings, and public safety. The performance of such teams relies on proper trust in the autonomous system, thus creating an urgent need to capture the dynamic nature of trust and devise objective, non-disruptive means of precisely modeling trust. This paper describes the use of bio-signals and embedded measures to create a model capable of inferring and predicting trust. Data (2304 observations) was collected via human subject testing (n = 12, 7M/5F) during which participants interacted with a simulated autonomous system in an operationally relevant, human-on-the-loop, remote monitoring task and reported their subjective trust via visual analog scales. Electrocardiogram, respiration, electrodermal activity, electroencephalogram, functional near-infrared spectroscopy, eye-tracking, and button click data were collected during each trial. Operator background information were collected prior to the experiment. Features were extracted and algorithmically down-selected, then ordinary least squares regression was used to fit the model, and predictive capabilities were assessed on unseen trials. Model predictions achieved a high level of accuracy with a Q(2) of 0.64 and captured rapid changes in trust during an operationally relevant human-autonomy teaming task. The model advances the field of non-disruptive means of inferring trust by incorporating a broad suite of physiological signals into a model that is predictive, while many current models are purely descriptive. Future work should assess model performance on unseen participants.