text
stringlengths
0
89.3k
adapt it to an asynchronous computation scheme to
reduce its computational demand
Contributions Based on the filterbased SLTMPC ap
proach introduced in 25 we propose a tubebased
MPC method that overapproximates the combined ef
fect of additive disturbances and model uncertainties
with an online optimized disturbance set and simul
taneously computes a tube controller Different from
previous results the proposed approach can handle
general polytopic disturbance sets and guarantees re
cursive feasibility in receding horizon This is achieved
by introducing a new terminal controller and an online
optimized terminal set These results extend the usage
of filterbased SLTMPC to possibly infinite task hori
zons and avoids a shrinking horizon implementation via
controller switching as in 25 Section 53 We further
propose a new asynchronous computation scheme for
filterbased SLTMPC that reduces the computational
complexity by splitting the tube and nominal trajec
tory optimizations into different processes Lastly we
provide a comprehensive numerical evaluation of the
proposed methods on a double integrator and a vertical
takeoff and landing VTOL vehicleThe paper is organized as follows Section 2 introduces
the notation the problem formulation and some ba
sic definitions Section 3 discusses the filterbased sys
tem level tubeMPC SLTMPC generalized from 25
before Section 4 introduces the recursively feasible ver
sion and Section 5 details the asynchronous computa
tion scheme Finally Section 6 showcases the proposed
filterbased SLTMPC variants on two numerical exam
ples and Section 7 concludes the paper
2 Preliminaries
21 Notation Definitions
We indicate column vectors eg a and matrices eg A
with lowercase and uppercase letters respectively We
denote stacked column vectors eg a and block matri
ces eg A with bold letters and index their block ele
ments with subscripts eg ai Aij We use diagiA to
denote the blockdiagonal matrix whose diagonal con
sists of itimes the Amatrix We define the operator
shiftAB for a blocklowertriangular matrix Awith
NNblock elements and a block row matrix Bwith
Nblock elements ie B B0BN1 as
shiftAB
A11
AN11AN1N1
B0 BN1
which removes the first row and column of Aand in
sertsBin the bottom block row We use Into denote
the identity matrix of size nn We distinguish between
the states of a dynamical system xk and the states pre
dicted by an MPC algorithm xi We use andto de
note Minkowski set addition and Pontryagin set subtrac
tion respectively which are both formally defined in 3
Definition 310 For a sequence of sets A0Aj we
use the convention thatL1
i0Ai0 which is the set
containing only the zero vector We define Kfunctions
and uniform continuity of a function below
Definition 1 Kfunction 29 A function γR0
R0is called a Kfunction if it is continuous strictly
increasing and γ0 0
Definition 2 Uniform continuity 29 A func
tionfx u X U Rwith sets X RnU Rm
is uniformly continuous in xfor all x X andu U if
there exists a Kfunction βsuch that
fx1 ufx2 uβx1x2x1 x2XuU
This also covers functions with a single argument ie
fx1fx2 βx1x2x1 x2 X
222 Problem Formulation
We consider linear timeinvariant LTI dynamical sys
tems with parametric and additive uncertainties of the
form
xk1 A Axk B Buk wk1
with ARnnBRnmwk W Rn and
AB D where WandDare convex polytopes
defined as
WwRnHwwhw 2a
D coi
Aj
B i 1 nA j 1 nB2b
where HwRnwnhwRnw and nAandnBare the
number of generators for Aand B respectively The
setDis thus defined by the vertices Dd d
Ad
B
d 1 nD with nDnAnB The system is subject
to compact polytopic state and input constraints
XxRnHxxhx 3a
UuRmHuuhu 3b
where HxRnxnhxRnxHuRnumhuRnu
with both sets containing the origin in their interior
To safely control system 1 we consider the following
robust constraint satisfaction problem CSP over the
task horizon N
Problem 1 Robust CSP Find a sequence of in
puts uii 0N1 such that the following con
straints are satisfied for all i 0N1