Determining and verifying good policies for cloned teleo-reactive agents

This paper presents a new method for finding and evaluating policies for cloned multiple teleo-reactive agents operating in the context of exogenous events. The method is based upon discounted reward evaluation of policy restricted situation graphs. A teleo-reactive agent behaves autonomously under the control of a policy and is predisposed by that policy to achieve some goal. Our framework plans for a set of conjoint agents by focusing upon one representative of them and introduces generic situations in order to reduce the number of policies. Simulation results reported here indicate that our method affords a good degree of predictive power and scalability.