Experiments of Visual Servoing in Presence of Humans based on Task Priority

This paper presents a visual servoing system for human-robot interaction that emulates mutual gaze and joint attention. A YOLOv8n module detects human faces or objects of a given category, and the image-plane error is regulated to zero via an Image-Based Visual Servoing (IBVS) law implemented within a Singularity-Robust Task-Priority framework. Four tasks are arranged in a strict hierarchy: joint limit avoidance, collision avoidance, camera roll stabilisation and IBVS tracking. This hierarchy guarantees that safety constraints are never sacrificed for tracking performance, while redundant degrees of freedom are used to maintain the image horizon level and keep the detected entity to the center of the image plane. Experimental results demonstrate robust tracking, strict enforcement of all constraints and the qualitative benefits of roll stabilization for achieving natural gaze behavior.