Instruction and Action in the Three-Tiered Robot Architecture

We present an approach to integrating voice-based natural language instruction and action in the three-tiered robot architecture. Voice inputs activate goals organized in a semantic network. Activated goals enable and disable behaviors. We describe how we implemented and tested our approach on two autonomous agents: a software agent and a mobile robot. Our implementation offers a proof by construction to the claim that instruction interpretation and behavior control are largely domain independent. Thus, the same mechanisms can be used to integrate instruction and action in autonomous agents working in different domains.