Toward enabling iCub to detect lies in everyday life

Social robots will be soon part of human society, where interactions are sometimes characterized by dishonesty. Hence, they will need to detect lies to better understand humans’ behavior, for instance, to assess who is trustworthy and provide better support in professions like teaching, caregiving, and law enforcement. In this manuscript, we present ongoing work, started  3-years ago, aimed at enabling the humanoid robot iCub to autonomously detect lies, in real-time, during an informal interaction. Our approach is based only on assessing the human partners’ cognitive load, through the measure of their pupil dilation. We show our scientific advancements and provide useful insights to both improve our system and to drive future developments in the field of lie detection in HRI.