In Virtual Reality environments, real humans can meet

virtual humans to collaborate on tasks. The agent Max is

such a virtual human. In construction tasks, he is standing

face-to-face to a human partner as a co-situated companion.

To maintain a social conversation, Max has to perceive

the user’s engagement and to solicit the user’s engagement,

he needs to demonstrate engagement himself. This paper

presents ongoing work on the development of a model that

describes interacting levels of engagement. We present how

the agent’s perceptive, cognitive, and expressive capabilities interact on these levels ranging from observing the human partner, to taking her goals, mental state or emotions into account when making decisions on how to interact and intervene.