Functional Safety Hazards for Machine Learning Components in Autonomous Vehicles

Analyzing functional safety for autonomous vehicles can be a challenging task. The task becomes more challenging as autonomous vehicles highly rely on machine learning components' which are complex to analyze for safety. Moreover, often machine learning engineers involved in the process of development do not have sufficient knowledge on functional safety, and thereby overlook important factors that can result in accidents. In this paper, we investigate and discuss important and necessary aspects of ML components in autonomous vehicles to ensure their functional safety and why we need to consider them. We also discuss the types of hazards we need to consider for each of those aspects. Further, we illustrate the hazard identification using the camera-based pedestrian detection system.