Dennis Perzanowski, Elaine Marsh, William Adams, and Alan C. Schultz
Combined research of the Intelligent Multimodal Multimedia and the Adaptive Systems Groups at the Navy Center for Applied Research in Artificial Intelligence has involved the development of a natural language and gesture interface to a mobile robot. While human communication between individuals occurs on many channels, two of them, natural language and gesture, complement each other fairly regularly in daily human-human communication. Since people feel fairly free to interweave both audial and visual input via these two channels during their interactions, we assumed they might readily do so in their interactions with a mobile robot. The first stage of our interface was built relying on this interaction. The natural language and gestural interface was then enhanced to enable the processing of incomplete and/or fragmentary commands during human-robot interactions. This enhancement has enabled us to keep track of various goals during human-robot interactions by instantiating context predicates, which are basically the topical predicates at various stages of the interaction. By utilizing these context predicates, a discourse component of the interface tracks the goals of the interaction, and records exactly which and to what extent each goal was achieved. With this information and by performing certain logical operations on semantic information of the context predicates, the robot can continue to achieve any unaccomplished goals on its own, no matter at what point or in what state the system is currently. Tracking of the goals, by means of these context predicates, permits the system to work independently on achieving previously stated, but as yet uncompleted, goals. In this sense greater autonomy is achieved, since users can expect the robotic system to be able to continue its performance and accomplish previously stated goals or subsequent logical goals, without the user having to explicitly state or re-state each.