Game theoretic Golog under partial observability

Alberto Finzi, Thomas Lukasiewicz

Abstract: We present the agent programming language POGTGolog, which combines explicit agent programming in Golog with game-theoretic multi-agent planning in a special kind of partially observable stochastic games (POSGs). The approach allows for partially specifying a high-level control program for a system of multiple agents, and for optimally filling in missing details by viewing it as a generalization of a special POSG and computing a Nash equilibrium.