Experiments with multiple BDI agents with dynamic learning capabilities

Abstract

In this paper we show how multiple BDI agents, enhanced with temporal difference learning capabilities, learn their utility function, while they are concurrently exploring an uncertain environment. We focus on the programming aspects of the agents using the Jason agentoriented programming language. We also provide experimental results showing the behavior of multiple agents acting in a Markovian grid environment. We consider agents with the perception function affected by the intermittent faults and Gaussian noise, as well as agents for which their action function is not always successful.