Specifying Complex Behavior for Computer Agents

Abstract

It is very difficult to specify behaviors for agents that must operate in complex environments. A variety of formalisms have been developed for specifying behaviors for computer agents, including the general paradigms of “classical planning” and “reactive behavior.” These formalisms represent points in a complexity space that has as two of its most important dimensions (1) ease of expression of complex action strategies by the human programmer, and (2) efficiency of execution of formal behavioral specification by the agent. This paper will focus on three different methods for specifying behaviors for agents: direct programming, operator descriptions, and goal reduction rules. These will serve as example formalisms and form a basis for discussion of the ease of human programming, the ease of automatic execution, and the value of compilation.