AI-driven Visualizations for Performance Monitoring and Anomaly Detection in Robots
Nuño Basurto, Carlos Cambra, Álvaro Herrero · 2020
Smart robotics is one of the fields that are been greatly enriched by the development of Cyber-Physical Systems (CPS). Under this frame, the software facet of robots is as important as the physical one. As it is widely known, the ever-increasing complexity of robots usually implies a parallel increase in the number of failures of such systems. Due to this, system monitoring and anomaly detection play a key role in the implementation of smart robotics and Artificial Intelligence (AI) can significantly contribute to this task. Accordingly, some Exploratory Projection Pursuit techniques, mainly implemented through neural networks, are applied an compared in the present paper to monitor the performance of a component-based robotic software. Thanks to the intuitive projections obtained by these techniques, anomaly detection can be also visually carried out. The visualizations are validated on an open and up-to-date dataset containing information about software anomalies that affect the middleware (a crucial component in CPS) of the analysed robot.