Online control adaptation for safe and secure autonomous vehicle operations

Modern cyber-physical systems, like automotive systems and aerial vehicles, are not built with cyber security in mind. Several techniques have been developed recently to overcome cyber-attacks on cyber-physical systems both at the software and the control levels. Adding such cyber security techniques to protect a system against malicious attacks, however, can incur runtime overheads that, in the case of autonomous systems, results in performance degradation and may lead the system into unsafe states. In this paper, we propose a framework for online control performance adaptation for secure and safe navigation of autonomous vehicles. Our approach leverages model predictive control (MPC) and knowledge about the system dynamics and the maximum performance degradation that cyber security techniques can impose at every time step to compute the control input that guarantees a safe operation of the system at all times. We validate the proposed approach both with simulations and experiments for an unmanned ground vehicle (UGV) motion planning case study in a cluttered environment.