Natural communication with robots

Abstract

We have developed a system which mediates between an unmodified reactive mobile robot architecture and human natural language I/O. We introduce reactive-odometric plans and demonstrate their use in plan execution, plan recognition, and learning to associate human terms with perceptually unremarkable locations in the environment. The communication component of our architecture supports typewritten natural language discourse with people. It lets users name places either immediately or in relation to other known places, ask questions about the robot’s plans and the spatial relationships of known places, and give the robot short and long term goals. This thesis presents results obtained with our implementation of this architecture on a physical mobile robot system designed by[Connell, 1992a] and in simulation. These results reflect experiments performed by the author and