Abnormality Detection in Plant Inspection Robots Using Time-Series Differences between Virtual/Real Observations

Abstract

This study aims to develop a unified detection method focusing on ”anomalies” caused by attacks. Security is one of the most concerning issues as robots become more prevalent in society. Recent robots exchange various data via networks and operate in the same space as humans. Therefore, the possibility of damage caused by malicious attacks is increasing, and robot security measures are becoming more critical. However, it is difficult to comprehensively prevent all attack vectors, because robots need to deal with a wide range of attacks from both physical and cyber spaces. In this paper, we propose a unified method for detecting anomalies using a virtual environment that can observe the actual environment. Specifically, anomalies are detected by using the differences between observation information, and the type of anomaly can be identified by analyzing the time series changes in the differences.