Attacking the V:On the resiliency of adaptive-horizon MPC

Abstract

Inspired by the emerging problem of CPS security, we introduce the concept of controller-attacker games. A controller-attacker game is a two-player stochastic game, where the two players, a controller and an attacker, have antagonistic objectives. A controller-attacker game is formulated in terms of a Markov Decision Process (MDP), with the controller and the attacker jointly determining the MDP’s transition probabilities. We also introduce the class of controller-attacker games we call V-formation games, where the goal of the controller is to maneuver the plant (a simple model of flocking dynamics) into a V-formation, and the goal of the attacker is to prevent the controller from doing so. Controllers in V-formation games utilize a new formulation of model-predictive control we have developed called Adaptive-Horizon MPC (AMPC), giving them extraordinary power: we prove that under certain controllability conditions, an AMPC controller can attain V-formation with probability 1. We evaluate AMPC’s performance on V-formation games using statistical model checking. Our experiments demonstrate that (a) as we increase the power of the attacker, the AMPC controller adapts by suitably increasing its horizon, and thus demonstrates resiliency to a variety of attacks; and (b) an intelligent attacker can significantly outperform its naive counterpart.