Towards a Simulation−based Framework for the Security Testing of Autonomous Vehicles

Abstract

Autonomous cars depend on data collection and processing in order to operate safely. Attacks against sensors can lead to an incorrect perception of the environment. Testing the impact of sensor attacks on the road is costly and dangerous. Simulation tools for autonomous driving research enable the simulation of the urban environment, thus providing a cheap and risk-free platform for the testing of perception systems. In this paper, we extend an open-source autonomous driving simulator with models of attacks to camera sensors. These attacks are modelled as computer graphics effects, which are subsequently added to the simulated environment. We also present a formal model, developed in the process alge- bra CSP, by which systematic testing of all possible sensor- attack combinations may be yielded. Although our results are at a preliminary stage, we are mindful of its contributions to the security — and therefore, safety — of autonomous driving.