Cognitive architecture for mixed human-machine team interactions for space exploration

Abstract

With some exceptions, current robotic systems need to be explicitly commanded for interactive tasks with humans. This constraint leads to a reduction in robot-human system level capabilities due to a lack of common grounding, and in addition, the robotic systems are selfcontained with little or no “social” intelligence. A cognitive architecture embodies robotic behaviors and the logical analysis of the surrounding, dynamic operational environment. JPL has developed a formal mathematical model for behavior-based robotic control, inference of human intent from sensed action, learning, and explanation capabilities all based on process algebras. This framework allows robotic systems to autonomously infer intentions/activities of humans through sensing of their actions. This paper describes the Activity Reasoning Module (ARM) that is responsible for this analysis, reviews the formal process algebra framework, and presents an experimental study in simulation of the determination of astronaut activity from sensory analysis of perceived action.