We introduce an architecture based on interactionist principles which enables autonomous agents to infer the behavioural patterns of other agents, without prior knowledge. Previous works have shown that agents can leverage the relations between entities and sensorimotor abilities afforded by them to build a model of the environment which supports behaviours satisfying innate motivational principles. However, efficient collaborative or competitive behaviours relies on the ability to infer hidden motivational principles of other agents by observing their behaviour. In this paper, we introduce a novel architecture able to identify mobile affordances generated by other agents, infer their behavioural preferences and predict their movements. We validate our proposition though experiments showing that a predator agent can learn to infer that its prey is attracted by specific elements of the environment.