Toward Genuine Robot Teammates: Coordination Through Dialogue

Abstract

Abstract:

Robots are increasingly needed to serve as equal partners in teams with humans in a variety of domains. This presents a number of human-robot interaction challenges, particularly in the areas of team communication and coordination. In this talk, I describe my work which leverages the extensive behavioral literature on human dialogue and teaming,and applies some of these findings to computational models that support effective human-robot teaming. First, I introduce a framework for fluid turn-taking that enables a robot to make predictions about the meaning of an ongoing turn and use this prediction to achieve a variety of interaction capabilities, including faster turn-taking, speech overlap production, and preemptive action execution. I describe how this framework was integrated into the DIARC cognitive robotic architecture and evaluated on a corpus of human-robot instructions. Next, I introduce a coordination framework for robot teammates based on the concept of shared mental models from the human teaming literature. I describe a virtual space robotics task domain that was created to evaluate this framework, and present the results of a human-subjects experiment that tests the benefit of the proposed framework. Overall, this work advances the field of human-robot teaming and brings robots a step closer to the goal of genuine teammates.

Zoom info: https://tufts.zoom.us/j/960202944

Meeting ID: 960 202 944

Password: defense330