An Approach to Robot Commonsense Reasoning

Abstract

To perform missions assigned by people, the mobile robot needs to learn both itself and the world it is in, to use the learned knowledge to reason and make decisions to resolve problems, and to guide further learning, and to establish its knowledge base which contains huge amounts of common sense. During the process of executing missions by reasoning and learning, it is critical to enable domain-specific robots be more tolerant rather than puzzled when encountering perturbations. In this paper, active logic is improved to deal with contradicted beliefs, metacognitive loop is incorporated to supervise and guide the whole reasoning process. Finally, route crack experiments state that this metacognitive loop and improved active logic based method can handle robot commonsense reasoning in an efficient way and be comparatively robust with perturbations.