Multi Agent Learning of Relational Action Models
Résumé
A number of recent works have designed algorithms that revise a relational action model online from interactions with their environment and use this model, even a potentially incorrect or incomplete one, for building plans and better exploring their environment. This paper addresses Multi Agent Relational Action Learning : it considers a community of agents, each rationally acting following some relational action model, and such that the observed effect of past actions that led an agent to update its action model can be communicated to other agents of the community, potentially speeding up the learning process of agents in the community. We describe in this paper a framework for collaborative action model revision in which agents are autonomous : they do not have any shared memory but benefit from past observations memorized by the community.