An autonomous agent that explores and acts in a rich world needs knowledge to act effectively. This agent can use knowledge that is available in Commonsense Knowledge Bases (CSKBs), when the agent designer cannot encode all the information the agent might need. CSKBs include general-purpose information about the everyday world in a formallanguage, but this information is not always correct, relevant, or useful for the agent’s purpose. In this paper we present an approach to retrieving commonsense knowledge for autonomous decision making. We consider agents whose formal language is different from that of the CSKB, and can use multiple CSKBs of various expressivity and coverage. We present a complete retrieval framework with algorithms for mapping languages and selection of knowledge. We report on preliminary experimental results of these algorithms for the ConceptNet CSKB.