POOH: A Robust Hybrid Agent Model

Hybrid agent systems attempt to combine two or more agent philosophies into single agent architectures. In general, these philosophies can be categorized as either producing reactive type behaviors or proactive and goaldirected behaviors. While recent researchers have provided a range of successful models of hybrid agents, they lack in both robustness and autonomy. POOH (Percept Oriented Object Hierarchy) is a hybrid agents model driven by utility functions. These functions encapsulate conceptually economic notions of actual human as opposed to more normatively correct decision making. The utility functions are characterized by two underlying notions. First, “losses loom larger than gains”, individuals are more concerned about changes in utility states than the state itself. Second, the additive nature of utility that is associated with neoclassical economics is dropped for more reference based or “framed” utility functions. Experimental analysis reveals that by incorporating utility functions in this manner into the decision-making module an agent becomes capable of both autonomous behavior and robust enough to handle uncertainties.