text
stringlengths
0
89.3k
Math Control and InformationComputationally Efficient System Level TubeMPC for
Uncertain Systems
Jerome Siebera Alexandre Didiera Melanie N Zeilingera
aInstitute for Dynamic Systems and Control ETH Zurich 8092 Zurich Switzerland
Abstract
Tubebased model predictive control MPC is the principal robust control technique for constrained linear systems affected
by additive disturbances While tubebased methods that compute the tubes online have been successfully applied to systems
with additive disturbances their application to systems affected by additional model uncertainties is challenging This paper
introduces a new tubebased MPC method named filterbased system level tubeMPC SLTMPC which overapproximates
both uncertainties with an online optimized disturbance set while simultaneously computing the tube controller online
Extending prior work we generalize the method to polytopic disturbance sets and for the first time provide rigorous closedloop
guarantees for the receding horizon controller These guarantees are obtained by virtue of a new terminal controller design and
an online optimized terminal set To reduce the computational complexity of the proposed method we additionally introduce
an asynchronous computation scheme that separates the optimization of the tube controller and the nominal trajectory Finally
we provide a comprehensive numerical evaluation of the proposed methods to demonstrate their effectiveness
Key words Robust model predictive control Linear systems Parametric model uncertainty System level synthesis
1 Introduction
Robust model predictive control MPC is the main ro
bust control technique for constrained linear systems
affected by bounded uncertainties In the presence of
model uncertainty formulating a robust MPC problem is
challenging since the predicted trajectories are affected
by the uncertain model and the computed control inputs
coupling the model uncertainty and the optimization
variables Early approaches to tackle this problem were
based on linear matrix inequality LMI methods 12
which are computationally expensive and scale poorly
with the system dimension Therefore tubebased MPC
methods have recently become popular for these types
of uncertain systems Tubebased MPC is based on sep
arating the system behavior into nominal and error dy
namics which describe the system dynamics neglecting
uncertainties and the deviations from the nominal dy
namics due to uncertainties respectively Since the un
certainties are assumed to be bounded the error dy
namics can be bounded in socalled tubes which are
computed based on the uncertainty bounds and a tube
controller 34 However the challenge of disentangling
Email addresses jsieberethzch Jerome Sieber
adidierethzch Alexandre Didier
mzeilingerethzch Melanie N Zeilingerthe model uncertainty and the optimization variables re
mains which has been addressed by mainly two different
strategies in the literature
One strategy for managing both additive disturbances
and model uncertainties with tubebased MPC in
volves lumping both uncertainties into a single state
and inputdependent additive term which is subse
quently bounded offline resulting in a system with
only additive uncerainty This approach enables the
use of various tubebased MPC methods proposed in
the literature Standard tubeMPC computes the tubes
offline and optimizes the nominal trajectory subject
to tightened constraints online see eg 567 To re
duce conservativeness homothetic tubeMPC 8 and
elastic tubeMPC 9 enable online optimization of the
tubes by fixing the tube shape but allowing dilation
and translation of the tubes online As shown in 10
online optimization of the tube controller enables tube
flexibility beyond dilation and translation The online
optimization is facilitated by parameterizing the tube
controller as a disturbance feedback policy 1112 or via
the system level parameterization SLP 13 which is
part of the system level synthesis SLS framework 14
Beyond tubebased MPC the lumped uncertainty ap
proach is also used in classical disturbance feedback
control synthesis 1516 which tightens the constraints
Preprint submitted to Automatica 19 June 2024arXiv240612573v1 eessSY 18 Jun 2024with offline computed bounds as a function of the online
computed control input However the lumped uncer
tainty approach is typically conservative since it relies
on offline computed global uncertainty bounds
Another strategy treats the uncertainties separately
and explicitly designs the tubes and the tube controller
for the individual uncertainty descriptions Most ap
proaches following such a strategy build on either the ho
mothetic tubeMPC formulation 171819 or the elastic
tubeMPC formulation 2021 Additionally the meth
ods proposed in 181921 incorporate online model up
dates based on setmembership estimation 22 to reduce
conservativeness However there are also tubebased
methods that use configurationconstrained tubes 23
which do not explicitly parameterize a tube controller
but compute a control input for every vertex of the tube
or use ellipsoidal tubes 24 which are computed via a
linear fractional transformation LFTbased formula
tion of the uncertainties More recently 25 proposed
a tubebased MPC method based on the SLP that si
multaneously overapproximates both uncertainties and
computes a tube controller online which we denote as
filterbased system level tubeMPC SLTMPC This
method provides highly flexible tubes and minimal of
fline design but only provides closedloop guarantees
for a shrinking horizon formulation and is computa
tionally expensive The computational issue has been
addressed on an algorithmic level by GPU paralleliza
tions 26 and structureexploiting solvers 27 and on
a system theoretic level by asynchronous computation
schemes 28 which split the computations of the tubes
and the nominal trajectory into different processes In
this paper we propose a filterbased SLTMPC method
with closedloop guarantees in receding horizon and