Privacy-preserving Model Predictive Control

Using Homomorphic Encryption

Master thesis (2020)

Authors

J.A. Govers Mechanical Engineering

Contributors

L. Ferranti Learning & Autonomous Control - Mechanical, Maritime and Materials Engineering (supervisor 1)
J. Alonso-Mora Learning & Autonomous Control - Mechanical, Maritime and Materials Engineering (supervisor 2)
Faculty
Mechanical Engineering
Copyright: © 2020 Jurriaan Govers
Model Predictive Control Homomorphic Encryption Privacy control
More Info
expand_more

Published Date

25-05-2020

Language

English

Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Faculty

Mechanical Engineering

Abstract

This thesis is focused on protecting sensitive data in optimization-based control methods. We propose a novel Privacy-preserving Model Predictive Controller (PMPC) where multiple agents are controlled by an untrusted external coordinator. By using the Paillier additively Homomorphic Encryption (HE) scheme, our PMPC allows the coordinator to solve a Quadratic Programming (QP) problem over encrypted data. The PMPC is based on a Projected Gradient Scheme (PGS) on the Lagrange-dual, which enables the use of quadratic
cost functions with complicated constraints (e.g., constraints on linear combinations of states and inputs). Compared to the state-of-the-art, the proposed controller protects not only the states and inputs of the agents, but also the system models, cost function and constraints. Optimization problems with quadratic cost functions and linear constraints, form the basis of a wide class of Model Predictive Control (MPC) problems. Examples of applications are smart-grids, large industrial plants and robotics. To test our PMPC, we focus on the application in self-driving vehicles. Motivated by the AUTOTRAC 2020 competition, we formulate a controller for multiple vehicles in a platoon. An external coordinator controls the longitudinal velocities of up to ten vehicles, while complicated constraints on the positions prevent collisions. By using our PMPC, the external coordinator does not gain access to the private data of the vehicles, protecting their privacy. Because of the wide application domain of MPC, we would like to extend this research in the future by testing the MPC on other applications as well.