Solving zero-sum one-sided partially observable stochastic games

Many security and other real-world situations are dynamic in nature can be modelled as strictly competitive (or zero-sum) games. In these domains, agents perform actions to affect the environment receive observations -- possibly imperfect about situation effects of opponent's actions. Moreover, there is no limitation on total number an agent that is, fixed horizon. These settings partially observable stochastic games (POSGs). However, solving general POSGs computationally intractable, so we focus a broad subclass called one-sided POSGs. games, only one has information while their opponent full knowledge current situation. We provide picture for POSGs: (1) give theoretical analysis value functions, (2) show variant value-iteration algorithm converges this setting, (3) adapt heuristic search POSGs, (4) describe how use approximate functions derive strategies game, (5) demonstrate our solve non-trivial sizes analyze scalability three different domains: pursuit-evasion, patrolling,