基于改进的主动逻辑与元认知环的机器人常识推理的研究

针对常识推理的非单调和异常问题，构建了基于改进的主动逻辑与元认知环的机器人常识推理框架。首先，针对机器人在执行任务时易受异常情况干扰的问题，引入元认知环对异常进行监视和评估并引导机器人。其次，对主动逻辑进行改进，定义了事实、常识，及它们相互之间的蕴含、否定和无关三种关系，给出了详细的矛盾知识的发生的条件和定义，并给出了对应的矛盾知识的处理方法，提出在主动逻辑中事实包含关系的传递性及推理的非直接传递性以有效检测和处理矛盾。最后，设计的Pr2机器人取书的实验进一步验证了元认知环以及主动逻辑在机器人执行任务时对异常情况和矛盾知识处理的有效性。

Research in Robot Commonsense Reasoning Based on Improved Active Logic and Metacognitive Loop

Non-monotonic problems and perturbations with commonsense reasoning are discussed in this paper, and a structure of robotic commonsense reasoning based on improved active logic and metacognition loop (MCL) is established. First, as to the problem that robots is prone to suffer from perturbations when executing missions, MCL is incorporated to monitor perturbations and to guide robots. Second, active logic is improved by defining three kinds of relationship between facts and commonsense, so called contained, negated, and irrelevant, further, according to these definitions, the precondition and definition of contradicted knowledge and corresponding handling methods are proposed. What is more, two kinds of attributions are presented that the relationship of facts, namely that contained, has transitivity, and that the logic reasoning rules do not have direct transitivity. Facilitated by above improvements, active logic can deal with contradicted knowledge more efficiently. At the end, the experiment that the Pr2 robot gets a book validates the effects of the fact that MCL and improved active logic can help robots deal with perturbations and contradicted knowledge.