Multimodal Plan Representation for Adaptable BML Scheduling.

In this paper we show how behavior scheduling for Virtual Humans can be viewed as a constraint optimization problem, and how Elckerlyc uses this view to maintain a extensible behavior plan representation that allows one to make micro-adjustments to behaviors while keeping constraints between them intact. These capabilities make it possible to implement tight mutual behavior coordination between a Virtual Human and a user, without requiring to re-schedule every time an adjustment needs to be made.