Analyzing the override strategy for collision avoidance functions

Abstract

The automotive industry has been shifting towards leveraging intelligent software solutions to ensure safety and ease of use. However, ensuring safety during execution heavily depends on how the human user interacts with these automated systems. In particular, one of the most commonly used safety features in current vehicles is called Automatic Emergency Braking (AEB). Although this automatic function has been proven effective in practice, there still exists an option for the driver to override the functionality as needed. This motivates the question of understanding the underlying intention of the driver when performing an override, as this knowledge can further improve the system’s safety when encountering edge cases. In this work, we analyze the driver behavior using unsupervised machine learning models and demonstrate an effective overriding strategy for AEB, through which undesired AEB intervention can be overridden faster by an average of 0.5 seconds. If verified, the new strategy would directly impact vehicle safety and enhance the user experience.