Defining Time-Varying Metrics of Task Conflict in Human/Robot Teams Using Simulated Agents

In this paper, we consider a simulated search & rescue application as context for introducing a novel monitoring concept that continually assesses the level of task conflict for a human-robot team. We define task conflict to mean inconsistent mental models of the task, including information about the agents, environment, and the task itself. In order to demonstrate a proof of concept, we used an agent-based modeling approach that simulates information fusion using a Bayesian algorithm. To represent nominal differences in the inferences made by each agent, we randomly perturbed the inputs to the Bayesian algorithm, with levels of randomization chosen to reflect the relevant existing literature regarding human performance. Using simulated nominal data, we generated a time-dependent conflict threshold. Then, this threshold was tested by injecting simulated anomalies and evaluating how often conflict was detected. The high resulting detection rate and the evidenced robustness of the simulation to parameter variation suggest the potential of the monitoring approach for future human-subject testing.