Rule-Based Filler on Misidentification of Vision Sensor for Robot Knowledge Instantiation

The intelligent robot recognizes the surrounding environment in order to model expressible objects and spaces, and performs tasks by combining actions that it can perform. To this end, we used a robot knowledge system that expresses objects, spaces, situations, and actions using ontology, and provides various inference methods through Java-based rules for performing specific tasks. The robotic knowledge system used ensures that the instances being created are consistent in the class and attribute values ​​of the data and do not contradict other data. In order to use this robot knowledge system efficiently, the creation of a complete ontology instance must be underpinned. However, in a real environment, there is a problem that a recognition error such as False Positive False Negative occurs when a robot recognizes an object through a Vision Sensor. In order to compensate for this, this paper proposes a rule-based eclipse error detection filter that enables stable instance management even in object recognition errors by considering Spatial Relation, Temporal Relation between objects, and recognition rates and properties of each object.