Clustering Trust Dynamics in a Human-Robot Sequential Decision-Making Task

In this paper, we present a framework for trust-aware sequential decision-making in human-robot team wherein the human agent’s trust robotic agent is dependent on reward obtained by team. We model problem as finite-horizon Markov Decision Process with of robot state variable. develop reward-based performance metric to drive update model, allowing make recommendations. conduct human-subject experiment total 45 participants and analyze how evolves over time. Results show that proposed able accurately capture dynamics. Moreover, cluster participants’ dynamics into three categories, namely, Bayesian decision makers, oscillators, disbelievers, identify personal characteristics could be used predict which type person will belong to. find disbelievers are less extroverted, agreeable, have lower expectations toward agent, compared makers oscillators. The oscillators tend get significantly more frustrated than makers.