A Consideration on Anomaly Detection of Mobile Robots by One Class Support Vector Machine and Its Performance Evaluation

This paper considers the anomaly detection of mobile robots in the consensus control, whose objective is to that states of all agents in a multi-agents system converge to common state value by exchanging information over a network. In the presented scheme, one class support vector machine, that is one of unsupervised learning approaches, is used and multiple classifiers are designed to deal with two cases: i) a failed robot does not move from the initial position and ii) a failed robot continues to move in a certain direction. Simulation and experiments with omnidirectional mobile robots show that anomaly detection accuracy is improved by using both two classifiers.