A Layered Architecture for Implementing Autonomous Planning Agents

This paper briefly describes an architecture for implementing autonomous agents that embody sophisticated planning capabilities. In particular, we are currently working on a two-pass vertically layered architecture, designed to deal with a complex environment. Such an architecture is currently based on three levels of abstraction (i.e., situated, strategic and deliberative), but has been designed for being easily generalized to a N-levels architecture, depending on the given environment and task complexity. Each level controls the underlying one, so that an agent behavior is supported by a clean hierarchical organization. Our autonomous agents act in a virtual world created for a computer game, and must interact with it by suitably planning and executing complex actions.