Comparison of Anomaly Detection and Solution Strategies for Household Service Robotics using Knowledge Graphs

Abstract

The concept of anomaly detection is a majorly investigated problem in the service robotics domain. The motivation of this work is to enable household service robots to detect abnormalities in the environment and solve them. This paper investigates two approaches using knowledge-based systems to detect and solve anomalies in a household environment. Both methods use knowledge graphs as a knowledge representation format. The first approach is a classical approach that records absolute positions of objects and performs clustering to solve positional anomalies. In the second approach, we perform deep learning methods on knowledge graphs using graph neural networks to detect and solve anomalies. These approaches fall at the intersection between anomaly detection and problem-solution strategies for the service robotics domain. Finally, the paper also presents a comparison between the two approaches highlighting their advantages and disadvantages.