A Constraints-aware Antagonistic Controller with Disturbance-adaptive Attacks

This paper proposes an extension to attack strategies in cyber physical systems, based on adopting input sequences which can maximize an objective function typically designed for minimization. In detail, the proposed method prioritizes the violation of state constraints over cost maximization to directly driving the system into an unsafe region. To achieve this, we reformulate the cost function by introducing a slack variable into the optimization problem, explicitly encouraging constraint violations. The framework also accounts for external disturbances that may counteract the controller's objectives. To guarantee a certain level of damage despite such disturbances, the problem is formulated as max-min optimization where the attacker optimizes for the worst-case scenario. Given that the maximization subproblem is NP-hard, we address this computational challenge using a vertex enumeration method. The effectiveness of the proposed approach is validated in a simulation scenario based on an autonomous aerial vehicle using a Model Predictive Controller (MPC), showing that constraint violations can be achieved at a reduced cost compared to existing methods.