A simple formalism Artificial intelligence-based to represent knowledge in a multi-agent planning context

Abstract

At the start of the simulation, the agent knows nothing about how the dynamics of interaction with the environment unfold, or what causes his sensations. He does not distinguish obstacles from free paths, and he does not know the consequences implied by his actions. Under these conditions, the CALM mechanism was able to converge steadily towards the expected solution, by building a model of the world adequate to represent the regularities of the environment, the regularities of its bodily sensations, as well as to represent the influence regular actions on both. The agent learns about the consequences of his actions in different situations, which are represented by a reduced number of very general diagrams. From them, the mechanism can build an action policy that allows it to avoid affectively negative situations and to seek those that are affectively positive. This solution manages to describe precisely all the regularities that the agent can perceive without building a complete plan of the environment.