The capacity to apply knowledge in a context different than the one in which it was learned is still an open research question within the area of learning agents. This paper specifically addresses the issue of transfer of knowledge acquired through online learning in a partially observable environment characterized by its 2D geographical configuration. We propose an autonomous agent architecture combining an agent-centered representation and the supervised and unsupervised learning of discriminating concepts to achieve efficient cross-map transfer. Our preliminary experiments on a grid-world environment where two agents duel each other show that the agent’s performances are improved through learning, including when it is tested on a map it has not yet seen.