Encrypted MPC based on ADMM real-time iterations
Moritz Schulze Darup
Abstract
Encrypted control enables confidential controller evaluations in cloud-based or networked control systems. Technically, an encrypted controller is a modified control algorithm that is capable of computing encrypted control actions based on encrypted system states without intermediate decryption. The realization of such controllers, e.g., using homomorphic encryption, is non-trivial. Nevertheless, even optimization-based model predictive control (MPC) has already been implemented in an encrypted fashion. However, the existing schemes either require an explicit solution of the parametric optimal control problem (OCP) or they can only consider input constraints. In this paper, we present a novel encrypted MPC that allows to include state and input constraints without the requirement of an explicit solution of the OCP. The approach builds on the encrypted implementation of a single iteration of the alternating direction method of multipliers (ADMM) per time step, i.e., ADMM real-time iterations.