Turn-taking — managing turns in conversation — is one of the most fundamental abilities in human interaction. We intuitively know when someone “does not wait their turn” or “takes an oddly long time to answer”. These turn violations break conversational flow and signal social incompetence.

The problem in robotics

In social robotics, turn-taking is a non-trivial technical problem. Systems that rely only on detecting the end of speech through silence fail consistently with users who pause often, speak slowly or are in noisy environments.

Multimodal cues for turn-taking

HRI research indicates that socially competent robots use multimodal cues: gaze direction, intonation contours, completion gestures and prosodic patterns specific to questions or statements. Combining these cues significantly improves accuracy, especially with older people and children.

Back-channels and engagement

An often-neglected aspect is the use of back-channels — signals that the robot is listening without taking the turn: “I understand”, a nod, a tilt of the torso. Robots that emit appropriate back-channels are perceived as more attentive and more socially competent.