Anomaly modelling in machine learning based navigation system of autonomous vehicles
Hossein Eshraghi, Babak Majidi, Ali Movaghar · 2020
In the past few years, autonomous navigation systems are gradually introduced to the consumer vehicles and the cars using these systems are gaining popularity. The autonomous navigation systems are using machine learning and deep learning models for visual processing of complex road scenes in various scenarios. These machine learning models are trained using numerous test rides and the process of training sometimes continues as the vehicle is navigating the streets. These machine learning models are vulnerable to anomalies which can lead to various issues for the autonomous driving systems. These issues will put the safety of the vehicle in danger. In this paper, the issue of the impact of the anomalies on the deep learning models of smart vehicles is investigated and a method for anomaly modelling in the machine learning based autonomous driving systems is presented. The presented model can increase the robustness of the autonomous vehicles.