Optimal status sets of heterogeneous agent programs

There are many situations where an agent can perform one of several sets of actions in responses to changes in its environment, and the agent chooses to perform the set of actions that optimizes some objective function. In past work, Eiter et. al. have proposed a rule based framework for programming agents on top of heterogeneous data sources, but they provide no solutions to the above problem. In this paper, we propose a semantics called optimal feasible status set semantics for agents which allows the agent to associate an objective function with feasible status sets and act according to the feasible status set that optimizes this objective function. We provide both an algorithm to compute exact optimal feasible status sets as well as the TierOpt and FastOpt algorithms to find (suboptimal) feasible status set much faster. We report on experiments on a suite of real agent applications showing that the heuristic algorithms works well in practice.