Certified Control in autonomous vehicles with visual lane finding and LiDAR

Abstract

Certified Control is a safety architecture for autonomous vehicles in which the controller must provide evidence to a runtime monitor that the actions it takes are safe. If the monitor deems the current state of the vehicle as unsafe, it intervenes by signalling the vehicle actuators to brake. In this work, we demonstrate how Certified Control can be used to increase safety through two implementations: one involving visual lane-finding and one involving LiDAR. Through experiments utilizing real driving data, a robot racecar, and simulation software, we show examples in which these runtime monitors detect and mitigate unsafe scenarios.