Task-Based Mixed-Initiative Coordination

Abstract

Interactive teaching is a coordinated activity in many different respects. This includes intra-personal aspects, like the joint production of speech and gestures, as well as inter-personal aspects, like the processing of verbal corrections while an action is performed. Given the experimental paradigm of human-robot interaction research, coordinated activities may be changed, added, or removed in an iterative manner. To keep the system maintainable despite coordination dependencies is an architectural challenge that is systematically analyzed in the following and supported by a toolkit. In many proposed robotic software architectures, coordination of active components that carry out tasks occurs through coupled statemachines that track the shared system state. This represents a generalizable software pattern that we have identified and analyzed in a general manner for the first time.

Furthermore, in mixed-initiative Human-Robot-Interaction, tasks can be initiated by either participant, causing the active and passive roles to change. Such changes have not been addressed before and we have generalized task coordination to encompass them. Last, but not least, distributed state tracking is complex, and previous implementations have thus often placed it entirely in a centralized coordination service that, however, increases coupling. Instead, we have developed a task service toolkit, which can be embedded in components, and demonstrated that this reduces component complexity considerably, without affecting coupling. Based on it, both centralized and de-centralized coordination services are possible.