Are autonomous vehicles driving us to safety? - Understanding adversarial attacks on autonomous vehicle's perception modules

Abstract

This paper aims to discuss adversarial attacks on Autonomous Vehi- cles (AVs), and the defence mechanisms that can be utilized to prevent such attacks. The paper first focuses on spoofing multiple cameras with overlapping field of view, then moves on to discuss other various feature squeezing countermeasure techniques that can be used to protect AVs from these adversarial attacks. The paper includes experiments that eval- uate the effectiveness of these countermeasures using different scenarios and datasets. The paper also highlight potential future works, including exploring other types of adversarial attacks and implementing adversarial training of neural networks.