A POMDP-Based Framework for Humor-Aware Human-Robot Dialogue

Abstract

Humor plays a crucial role in communication, influencing engagement, emotional state, and information retention. This study investigates the effect of adaptive humor in human-robot interaction (HRI) using a robot implemented within a partially observable Markov decision process (POMDP) framework. The robot dynamically adjusts its humor level based on users' real-time states, inferred from multimodal cues including speech emotion, facial expressions, head position, gaze, and blink rate. 20 participants took part in a withinsubject experiment. Compared to a non-adaptive randomhumor robot, the adaptive robot was perceived as significantly more humorous, better timed, and more contextually appropriate. Participants also reported higher enjoyment of the interaction and higher perceived knowledge retention. These findings suggest that real-time humor adaptation can enhance users' subjective interaction experience and perceived learning outcomes. The results provide insights for designing socially aware conversational robots capable of tailoring humor to users' states.