Auditing Demographic Bias in Facial Landmark Detection for Fair Human-Robot Interaction
A new paper published on arXiv details a study on demographic bias in facial landmark detection, a crucial component for human-robot interaction. The research found that while biases related to gender and race largely disappear after accounting for visual factors like head pose and face resolution, a significant age-related bias persists, with older individuals experiencing higher error rates. The authors emphasize the importance of auditing and correcting these biases in low-level vision systems to ensure trustworthy and equitable robot perception. AI
IMPACT Highlights potential fairness issues in low-level AI perception systems, crucial for equitable human-robot interaction.