Modeling Engagement in a Multi-Party Human-Robot Dialog

Giving a robot the ability to interact naturally with humans using natural language dialog is an important goal of current research in human-robot interaction. Since most existing dialog systems target telephone-based, single-user dialogs, they are not sufficient to solve this challenge.

When a robot is situated in an environment containing multiple possible interaction partners, it has to make decisions about when to engage specific users and how to detect and react appropriately to actions of the users that might signal the intention to interact.

This thesis tries to adress this by extending an existing system for the management of human-robot dialogs with new capabilities. This includes functions for managing multiple interactions, each with one or more distinct participants, and for making engagement decisions based on the actions and intentions of the present users and the current internal state of the dialog system.