A Context-based Approach to Robot-human Interaction

Abstract

CARIL (Context-Augmented Robotic Interaction Layer) is a human-robot interaction system that leverages cognitive representations of shared context as a basis for a fundamentally new approach to human-robotic interaction. CARIL gives a robot a human-like representation of context and an ability to reason about context in order to adapt its behavior to that of the humans around it. This capability is “action compliance.” A prototype CARIL implementation focuses on a fundamental form of action compliance called non-interference -- “not being underfoot or in a human's way”. Non-interference is key for the safety of human-co-workers, and is also foundational to more complex interactive and teamwork skills. CARIL is tested via simulation in a space-exploration use-case. The live CARIL prototype directs a single simulated robot in a simulated space station where four simulated astronauts are engaging in a variety of tightly-scheduled work activities. The robot is scheduled to perform background tasks away from the astronauts, but must quickly adapt and not be underfoot as astronaut activities diverge from plan and encroach on the robot's space. The robot, driven by CARIL, demonstrates non-interference action compliance in three benchmarks situations, demonstrating the viability of the CARIL technology and concept.