Regret-Based Automated Decision-Making Aids for Domain Search Tasks Using Human-Agent Collaborative Teams

Abstract

In this paper, we consider domain search tasks using a human-agent collaborative team. An automated Bayesian sequential decision-making strategy is first proposed for the optimal allocation of manual and autonomous sensing modes in the presence of sensing uncertainty. An agent will generate a request for human intervention whenever the manual mode is deemed optimal. However, this type of optimal strategy is not always the style of human decision making under uncertainty. In particular, regret has been shown to play a critical role in human’s rational decision making. Humans experience regret when they perceive that they are better off with another option and tend to make choices to avoid such experience. Furthermore, it has been shown that team members who share the same mental model perform better than teams with a more accurate but less similar mental model. Therefore, to enable more effective human-agent collaboration (HAC) and improve joint performance, we embed regret analysis into the proposed decision-making framework to provide suboptimal but more humanlike automated decision aids. We will reconcile the dominant Bayesian methods with the analysis of human behaviors during decision making. We provide the simulation results to compare the decision-making strategies with and without regret analysis and show that regret-based decision making can integrate human tendency in risk averse and risk seeking for better HAC.