Multimodal Predictors of Team Trust and Task Cohesion Measured During an Experiment Examining Dynamic Task Allocation Technologies in a Next Generation Combat Vehicle Simulator

Author(s):

David Chhan, Murat Kucukosmanoglu, Catherine Neubauer, and Andrea Krausman

Abstract:

This report summarizes recent analyses aimed at identifying indicators of team trust and cohesion by examining relationships between subjective survey data and physiological and behavioral measures in support of the Scalable Cross-Echelon Command and Control Program. These analyses are part of a broader effort to develop metrics and predictive models for continuous assessment of human– autonomy teams and understand how to apply these models within the context of future Soldier and AI-enabled command and control teams. Data for the analyses were collected during a simulation involving 30 Soldiers conducting 26 operational missions in a vehicle simulator. The goal of the study was to test and validate the dynamic re-tasking and task allocation technology developed to assist Soldiers in effectively managing resources and improving situation awareness. Key findings revealed significant correlations between individual self-reported trust and cohesion responses (trust-in-team, trust-in-technology, and task cohesion) and variables, such as communication sentiment, cardiac activities (heart rate and heart rate variability), and oculomotor behaviors (pupil and gaze measures). We found eye-based metrics aggregated at the team level provided the most consistent and strongest correlation. Using the features derived from physiological and behavioral measures, machine learning models were developed to predict individual team trust, trust-intechnology, and task cohesion. The models performed moderately well in predicting the self-reported trust and cohesion responses with good correlation between the predicted and true values where the difference between the two values falls within a 15% range. Although based on a relatively small sample size, these results can provide useful insights and inform the design of systems to enhance the performance of both human–human and human–autonomy teams in future manned–unmanned teaming operations.