An annotated database of a skilled behavior: Extraction of patterns that may underlie imitation

Imitation's been widely studied not only to understand how animals and humans learn novel actions and tasks by imitating actions performed by others but also for its application to robotics. This thesis offers a naturalistic study to address the following question: How can an observer, seeing the same task performed in different ways by different performers, nonetheless extract hierarchical parsing of shared features of the behaviors which links actions to goals and sub goals? In this study, 20 videos of the process of making tea were recorded, annotated in detail and analyzed in an attempt to isolate the strategies which an observer can use to organize the observation of diverse performances. The videos were annotated manually and several interesting behaviors were observed in the process. The result is a set of hypotheses about complex imitation, and a database which can be used in future to test parsing algorithms hypothesized to capture the learning process.