Robotics in Real-Time Applications Using Bayesian Hyper-Tuned Artificial Neural Network

摘要

Cybersecurity threats to robotic systems are increasing. Real-time location systems (RtLSs) are crucial for successful navigation and safe operation of many mobile robots. Therefore, RtLSs have become attack vectors for robotics and automated systems, a phenomenon that has not been adequately investigated. This chapter demonstrates that a supervised learning system can identify cyberattacks on RtLSs. Moreover, it demonstrates that a system developed with machine learning approaches is capable of detecting some forms of cyberattack on RtLSs, notably denial of service (DoS) and spoofing. Bayesian hyper-tuned artificial neural network (BH-ANN) is proposed and tested with a dataset of real data captured by a wheeled robot and a commercial RtLS, based on ultra-wideband beacons, to build algorithms capable of identifying these attacks. The proposed BH-ANN achieved the best possible test score and the lowest possible validation error, as evidenced by experimental findings with a cross-validation analysis. In addition, it is the model with the highest sensitivity for identifying DoS and spoofing cyberattacks on RtLSs, and the lowest overfitting.