Humans are capable of discovering knowledge through dynamic interaction with the environment, especially guided by the semantics associated with their actions. This study introduces the design of a cognitive framework intended to replicate these human-like capabilities within a robotic system alternating among two processes: (a) exploration to perceive new entities and infer their action-related proprieties; (b) exploitation to plan and execute the most appropriate actions that expand the robot’s knowledge based on what it has learned so far. In addition, a DOLCE-based ontology serves as a bridge between the two processes, characterizing useful “frames” of knowledge.