Dialogue Policies for Confusion Mitigation in Situated HRI

Confusion is a type of dynamic mental state, which Although our work to date has focused on confusion can not only lead to negative conditions, i.e., frustration, (Li et al., 2021; Li and Ross, 2022), modelling boredom or subsequent disengagement in a and detection, it is also essential that the dialogue task or a conversation, but can also be associated agent is capable of mitigating user confusion and with positive conditions as a user seeks to overcome helping participants reengage in the ongoing taskoriented initial confusion (D'Mello et al., 2014; Li interaction. Our model is based on seven et al., 2021). In mainstream human-computer interaction dialogue act types that are used to implement strategies (HCI) studies, a number of studies have for confusion mitigation. In light of this need, investigated confusion state effects in the context of in the paper, we sketch out our initial approach to online learning and driver assistance (Kumar et al., design a dialogue policy for task-oriented interaction 2019; Grafsgaard et al., 2011; Zhou et al., 2019).