Successful collaboration on shared-workspace tasks requires that team members (both robot and human) have a similar model of the task.  For systems that learn tasks through common learning-from-demonstration methods, this requirement adds a substantial set of constraints on learning.  In this talk, I will present work on building a shared hierarchical model of the task that uses a simple graph transformation that allows an interactive robot to automatically construct a hierarchical task model from small numbers of linear demonstrations.  This talk will be based heavily on the thesis work of Brad Hayes in my lab.

Video Recording