Award for TUM Research on Cybersecurity of Autonomous Vehicles

Michael Kühr, Dr. Mohammad Hamad and Professor Sebastian Steinhorst from the Professorship of Embedded Systems and Internet of Things received the coveted award for their paper “SoK: Security of the Image Processing Pipeline for Camera-based Sensing in Autonomous Vehicles.” The ASIA Conference on Computer and Communications Security of the Association for Computing Machinery (ACM ASIACCS), with a CORE ranking of A, is one of the world’s leading academic events in the field of cybersecurity and brings together top researchers from around the globe every year.

Cameras as a Key Technology for Autonomous Vehicles

Autonomous vehicles rely heavily on cameras to sense their surroundings and navigate safely. However, the cameras themselves are only one part of a complex overall system: Between the real-world environment and the AI-based decision-making algorithms of modern vehicles lies a multi-layered processing pipeline that ranges from optical sensing through sensor technology and image processing to software and AI components. Security vulnerabilities or malfunctions at any of these points can significantly compromise the reliability of the entire system.

Systematic Analysis of the Entire Image Processing Pipeline

In their SoK paper (“Systematization of Knowledge”), the researchers systematically examine, for the first time, the security and robustness risks along the entire processing pipeline. The aim of the paper is to synthesize and structure the current state of research and to highlight existing knowledge gaps.

The analysis shows that much of the existing research has focused on individual levels of the pipeline – particularly on software- and AI-based attacks. This often results in an incomplete and distorted assessment of the actual risks. While attacks at the software or AI level can have serious consequences, they often require considerable technical effort. Attacks at the physical or sensor level, on the other hand, are often easier to carry out and can still have security-critical consequences.

Foundations and Tools for a Holistic Security Concept

A key finding of this work is the integration of two research areas that had previously been considered largely separate: While cybersecurity research focuses on targeted attacks, robustness research examines the effects of disruptions and environmental conditions. The award-winning paper combines these two perspectives, thereby enabling a realistic assessment of the risks to camera-based sensor systems.

To make this approach practical, the researchers also introduce two tools: TARA-CAM, a method for threat and risk analysis of camera systems in accordance with the international standard ISO 21434 for cybersecurity in the automotive industry, and PICT, an open test environment for investigating camera systems under realistic conditions. The study thus lays an important foundation for a holistic safety concept for cameras in autonomous vehicle systems.