On the problem of comparing two models for rational decision making in autonomous agents

Abstract

The main objective of this work is to establish a basis for comparing BDI agents with the optimality of decision theory models such as POMDPs. It is argued that a direct comparison is not possible due to the intractability of the algorithms used for solving POMDPs, and therefore an approximation must be used. We propose the reduction of the state space, and the combination of sub-solutions as the main approaches towards this goal. Throughout this work, the tileworld testbed is used as a frame of reference for the discussion of the various concepts.