Cyber Intrusion Resilience for Control System of Intelligent Vehicle (ResDrive)

To be resilient, autonomous vehicles must not only detect failures and react to failing software and hardware, but also be able to absorb such failures.

The project at a glance

Start date:

10 Dec 2020
Duration in months:

18
Funding:

Huawei Technologies Deutschland GmbH
Principal Investigator(s):

Marcus Völp

About

In-vehicular resilience against cyber-attacks is essential for maintaining the safety of passengers, pedestrians and other road users, especially if the attacks are mounted by highly-skilled and well-equipped adversaries. To achieve the necessary resilience, autonomous vehicles must not only detect failures and react to failing software and hardware components, but also be able to absorb such failures, tolerating them long enough until failing components can be recovered. It is thereby irrelevant whether the failure originated from accidental faults or was intentionally triggered as a result of adversaries compromising the system. In this partnership project, Huawei and SnT’s CritiX research group joined forces to investigate resilience techniques at all levels of autonomous driving hardware/software stacks. Leveraging Apollo, the team explored how vision and prediction failures can be tamed by replicating the control component that is responsible for monitoring and validating the trajectories a vehicle is meant to follow. Replication thereby ensures fault tolerance of this critical component and buys time to recover failing vision subsystem and control components. By designing new custom components, and in combination with the team’s extremely fast recovery technique, the prediction subcomponent is allowed to restart extremely fast while the vision module slowly reboots, without causing the vehicle to break at its perception horizon, and continues operating seamlessly. Last but not least, the team investigated securing the trajectory communication to the ECUs executing it, by means of eliminating single points of failures in the car’s Brain driver infrastructure. Building in resilience to accidental faults and targeted attacks will make our interactions with increasingly more autonomous cars more reliable and trustworthy.