Handling Semantic Inconsistencies in Commonsense Knowledge for Autonomous Service Robots

Abstract

The necessary amount of commonsense knowledge for autonomous domestic service robots is enormous and avoiding inconsistencies is consequently almost impossible. This is especially true for inconsistencies on a semantic level, that is, inconsistencies as interpreted by human beings where the properties of a concept may contradict each other. In contrast to Boolean algebra (¬A and A), these inconsistencies will not lead to a trivial solution like false, since they are not captured in a logic formalism. Instead, they will result in semantic contradictions when interpreting the solution. In order to tackle this problem, we propose a way to handle semantic inconsistencies, by utilising a commonsense knowledge database and implementing a dynamic knowledge base that is capable of resolving contradicting parts of knowledge. The applied non-monotonic reasoning mechanism is Answer Set Programming, and ConceptNet5 is the commonsense knowledge database. Finally, we evaluate the performance of our approach and show its advantages in a complete household scenario.