You are on page 1of 5

Proceedings of the 20th World Congress

The International
Proceedings Federation
of the 20th Worldof Congress
Automatic Control
The International
Proceedings Federation
of the 20th9-14,
Worldof Congress
Automatic Control
Toulouse,
The France,
International July
Federation 2017
of Automatic Control
Available online at www.sciencedirect.com
Toulouse, France, July 9-14, 2017
The International Federation of Automatic Control
Toulouse, France, July 9-14, 2017
Toulouse, France, July 9-14, 2017
ScienceDirect
IFAC PapersOnLine 50-1 (2017) 9480–9484
Immersion
Immersion of
of Nonlinear
Nonlinear Systems
Systems into
into
Immersion
ImmersionHigher of
of Nonlinear
Nonlinear
Order Systems
Systems
Systems into
into
Higher
Higher Order
Order Systems
Systems
Higher Order ∗
Systems
∗∗
E.
E. Aranda-Bricaire
Aranda-Bricaire ∗∗ C. C. Califano
Califano ∗∗ C.H. Moog ∗∗∗ ∗∗∗
∗∗ C.H. Moog ∗∗∗
E. Aranda-Bricaire ∗ C. Califano ∗∗ C.H. Moog ∗∗∗
E. Aranda-Bricaire C. Califano C.H. Moog

∗ Electrical Engineering Department, CINVESTAV, AP 14-740,
∗ Electrical Engineering Department, CINVESTAV, AP 14-740,
07000
∗ Electrical Mexico
Mexico City,
Engineering Mexico,
Mexico, earanda@cinvestav.mx
City,Department, CINVESTAV,
∗∗ 07000
Electrical Engineering Department, CINVESTAV, AP
earanda@cinvestav.mx AP 14-740,
14-740,
07000
∗∗ Dipartimento Mexico
di City,
Ingegneria Mexico, earanda@cinvestav.mx
Informatica Automatica
Automatica ee Gestionale
∗∗ Dipartimento
07000 Mexico di Ingegneria
City, Mexico,
Antonio
Informatica
earanda@cinvestav.mx
Ruberti,
Gestionale
∗∗ Dipartimento di Ingegneria Informatica
Ruberti, Automatica
Dipartimento di Ingegneria Antonio Informatica Automatica ee Gestionale Gestionale
Università di Roma La
Università di Roma La Sapienza, Sapienza,
Antonio Via Ariosto
Ariosto 25,
Ruberti, 25, 00185
00185 Rome, Rome, Italy,
AntonioVia Ruberti, Italy,
Università di califano@diag.uniroma1.it
Università
∗∗∗ di Roma
Roma La La Sapienza,
Sapienza, Via Via Ariosto
califano@diag.uniroma1.it Ariosto 25, 25, 00185
00185 Rome, Rome, Italy, Italy,
∗∗∗ LS2N, UMRcalifano@diag.uniroma1.it
CNRS
∗∗∗ LS2N, CNRS 6004,6004, 1
UMRcalifano@diag.uniroma1.it 1 rue
rue dede lala Noë,
Noë, BP BP 92101, 92101,
∗∗∗ LS2N,44321
44321UMR Nantes CNRSCedex 6004,3,
3, France,
1 moog@ieee.org
LS2N, Nantes
UMR CNRSCedex 6004, 1 rue
rue de
France, la
la Noë,
Noë, BP
demoog@ieee.org BP 92101, 92101,
44321 Nantes Cedex 3, France,
44321 Nantes Cedex 3, France, moog@ieee.org moog@ieee.org
Abstract:
Abstract: Two Two nonlinear
nonlinear systems
systems havinghaving the the same
same number
number of of inputs,
inputs, but but not not the the samesame numbernumber
of state
Abstract: variables,
Two
of state variables, are considered.
nonlinear systems
are considered. The problem
having the of
same the existence
number of of
inputs, two invertible
but not the state
same feedback
number
Abstract: Two nonlinear systemsThe problem
having the same of the existence
number of twobut
of inputs, invertible
not thestate samefeedback
number
of
laws
lawsstate
and
and variables,
a
a surjective
surjectiveare mapping
considered.
mapping from
from Thetheproblem
the higher
higher of the existence
dimensional
dimensional state
state of space
two invertible
space of
of one
one state feedback
system
system into
into the
the
of state variables, are considered. The problem of the existence of two invertible state feedback
lower
laws
lowerand dimensional
and a state
surjective
dimensional state space
mapping of
space offrom the
from second
the
the second system
higher
system is stated,
dimensional
is stated, suchsuch
state that
space the offirst
one system
system dynamics
into the
laws a surjective mapping the higher dimensional statethat space theoffirstonesystem systemdynamics into the
reduces
lower exactly
exactly to
dimensional the
the second
tostate space system dynamics. This problem generalizes the linear feedback
reduces
lower dimensional state space of
second the
the second
ofsystem system
dynamics.
second systemThis is
is stated,
problem
stated, such that
that the
suchgeneralizes the first
firstthesystem linear dynamics
system feedback
dynamics
equivalence
reduces
equivalence problem
exactly
problemto of
the two
second
of two systems
system
systems and is fully
dynamics.
anddynamics. solved
is fully solved This in the
problem special case
generalizes of feedback
the linear linearizable
feedback
reduces exactly to the second system Thisinproblem
the special case of feedback
generalizes the linear linearizable
feedback
systems.
equivalence problem
systems.
equivalence problem of of two
two systems
systems and and isis fully
fully solved
solved in in the
the special
special case case of of feedback
feedback linearizable
linearizable
systems.
© 2017, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
systems.
Keywords:
Keywords: Feedback
Feedback immersion,
immersion, Feedback
Feedback equivalence,
equivalence, Non Non linear
linear systems.
systems.
Keywords: Feedback immersion, Feedback equivalence, Non linear systems.
Keywords: Feedback immersion, Feedback equivalence, Non linear systems.
1.
1. INTRODUCTION
INTRODUCTION In
In order
order to to understand
understand the the importance
importance of of suchsuch a a problem
problem
1. INTRODUCTION In
it order
is
it is to
sufficient
sufficient understand
to note
to note the the
that importance
a
thatimportance practical
a practicalofcase of such
case isa problem
provided
1. INTRODUCTION In order to understand suchisa provided
problem
it
in is
in thesufficient
the area
area of to note robots,
of walking
walking that a where practical the case
the is provided
dynamics of
of the
it is sufficient to note robots, that a where practical dynamics
case is providedthe
in the
elementary
elementaryarea of walking
Acrobot is robots,
embedded wherein the the more dynamics
complex of the
full
in the area Acrobot
of walking is embedded
robots, where in the themore complex
dynamics of full
the
The elementary
dynamics
dynamics of ofAcrobot
aa walking is embedded
robot
robot with in the more complex
aa higher numberfull
number of
The equivalence
equivalence of of systems
systems underunder different
different kindkind of of trans-
trans- elementary Acrobot walking is embedded within the higher
more complex of
full
The equivalence
formations is one ofofsystems
the coreunder different
subjects of kind
modern of trans-
control dynamics
degrees of
degrees of of of a
freedom
freedom walking
(Celikovsky
(Celikovsky robot with
et al. a higher
(2013)).
et al. a(2013)). number of
formations
The is oneofofsystems
equivalence the core subjects
under of modern
different kind ofcontrol
trans- dynamics a walking robot with higher number of
formations degrees of freedom (Celikovsky et al. (2013)).
theory.
theory. It isis
It is one of the
sufficient
issufficient to core
to thinksubjects
think to
to the of modern
the wide
wide control
literature
literature on
on degrees
formations
theory.
topics It
such isas
one of the
sufficient
the linear to
core
think
feedback
subjects
to the
of modern
wide
equivalence
control
literature
problem, on
or
In this of
In this freedom
paper,
paper, (Celikovsky
aa general
general formulation
formulation et al. (2013)).of
of the the problem
problem is is
topics such
It isassufficient
the lineartofeedback
think toequivalence problem, on or In this paper, a general formulation of the problem
theory.
topics
the such
equivalenceas the
to linear
the feedback
observer
the wide literature
equivalence
canonical form problem,
up to input or
proposed.
proposed.
In this
It is
It isa shown
paper,
shown
general
that,
that,
formulation
if the
if the problem problem
of the admits is
admits
problem
a
a
is
the equivalence
topics such as the to linear
the observer
feedback canonical form problem,
equivalence up to input or proposed.
solution, It is such
then shown that, can
solution if the also problembe admits by
obtained a
the
and equivalence
output to
injection the observer
both in canonical
continuous form
and up
discreteto input
time. solution,
proposed. then
It is such
shown solution
that, can
if the also problembe obtained
admits bya
and equivalence
output injectionto theboth in continuous
canonicaland discrete time. solution, then
the
and output
observer form up to input feedback
feedback on
time. solution, onthenthe such
the largersolution
larger
such
dimensional
dimensional
solution
can system
also beonly.
can system
also beonly.
obtained
A
A general
obtained
by
general
by
Within
Within
and the injection
the
output nonlinearboth
nonlinear
injection both
in continuous
setting,
setting, the
the feedback
in continuous
and discrete
feedback linearization
linearization
and discrete time. feedbacksolution on
for the larger
case ofdimensional
linear systems system is only.
presented, A general
based
Within
problem, the
was nonlinear
originally setting,
solved the
in feedback
the historical linearization
papers by solution for
feedback on the larger
case ofdimensional
linear systems system is presented,
only. A based
general
problem,the wasnonlinear
originallysetting,
solved in thethe historical papers by solution on for the of case of linear equation systems is(Sylvester presented, based
Within
problem,
feedback linearization on the
by solution the solution
solution
for the of case
aa Sylvester
Sylvester
of linear equation systems is(Sylvester presented,
(1884)).
(1884)).
based
Brockett was
Brockett
problem,
(1978)
(1978)
was
originally
and
originally
solved in the
and Jakubczyk
Jakubczyk
solved in the andhistorical
and Respondek
Respondek
historical
papers
(1980)
(1980)
papers by on the solution of a Sylvester equation (Sylvester (1884)).
Brockett
and further
and further (1978) and
investigated
investigated Jakubczyk
in the and
discrete
in the discrete Respondek
time (1980)
context
time context in on
A the
worked solution
example of a is Sylvester
presented
in A worked example is presented to show that the obtained equation
to show (Sylvester
that the (1884)).
obtained
Brockett (1978) and Jakubczyk and Respondek (1980)
and further(1987),
Jacubczyk
Jacubczyk investigated in the discrete
Aranda-Bricaire et
et al.time context
(1996), Cali-in A worked
results are example
are far more is
far more presented
general thanto
than theshow
the feedback that the obtained
linearization
and further(1987), Aranda-Bricaire
investigated in the discrete al.
time (1996),
context Cali-in A results
worked example is general
presented to feedback
show that thelinearization
obtained
Jacubczyk
fano et al. (1987),
(1999)
fano et al. (1999) andAranda-Bricaire
Aranda-Bricaire et al.
and (1996),
Moog Cali-
(2008). results
problem. are far more
Specifically, general
we than
compute the feedback
explicitely linearization
the
the coordi-
Jacubczyk (1987),and Aranda-Bricaire
Aranda-Bricaire et and
al. Moog
(1996),(2008).
Cali- resultsproblem. areSpecifically,
far more general we compute
than theexplicitelyfeedback linearization coordi-
fano et al. (1999)
Input-output
Input-output and Aranda-Bricaire
equivalence
equivalence to
to a a prime
prime system and Moog
system was (2008).
was solved
solved in problem.
in nates
nates in Specifically,
in which
which two
two non nonwe compute systems
linearizable
linearizable explicitely the exactly
display coordi-
fano et al. (1999) and Aranda-Bricaire and Moog (2008). problem. Specifically, we compute systems explicitely display
the exactly
coordi-
Input-output
Aranda-Bricaire
Aranda-Bricaire equivalence
and
and Hirschornto a prime
Hirschorn (1999).
(1999). system was solved in nates the
the same in which
same dynamics.
dynamics.two non linearizable systems display exactly
Input-output equivalence to a prime system was solved in nates in which two non linearizable systems display exactly
Aranda-Bricaire and Hirschorn (1999). the same dynamics.
Aranda-Bricaire
In the present and
paper Hirschorn
a (1999).
generalization
In the present paper a generalization of this problem is of this problem is the same dynamics.
In the present paper 2.
2. PRELIMINARIES
PRELIMINARIES
considered,
considered,
In the present
namely,
namely,
paper thea
the generalization
conditions
a conditions
generalization underof
under this problem
which,
ofwhich,
given
given two
this problem two is
is 2. PRELIMINARIES
considered,
nonlinear namely,
systems,
nonlinear systems, the
after conditions
a possible under
change which,
of given
coordinates two 2. PRELIMINARIES
considered, namely, after a possibleunder
the conditions change of coordinates
which, given two In the present section, the problem statement is given and
nonlinear
and
and a systems,
a regular
regular staticafter
static statea
state possible change
feedback on bothof
on both coordinates
systems, the In the present section, the problem statement is given and
nonlinear systems, after a feedback
possible change ofsystems,
coordinates the In somethepreliminary
present section, results theare problem
developed statement
which is given how
and
and a
smaller
smaller regular
one can
one can static
be state
immersed
be immersed feedback
into
into theon
the both
largest
largest systems,
one. the
Such
one. Such some
In thepreliminary
present section, results theare developed
problem which suggest
statement suggest
is given how
and
and a regular static state feedback on both systems, the some the preliminary
problem should results
be are
tackled. developed which suggest how
smaller
a problem
a problem one can
has be
been immersed
already into
addressed the inlargest one.
Aranda-Bricaire Such the
some problem
preliminary should be
results tackled.
are developed which suggest how
smaller onehascanbeenbe already
immersed addressed
into theinlargest
Aranda-Bricaire
one. Such the problem should be tackled.
aand
problem
Moog has
Moog (2006)beenwhere
already addressed
however in Aranda-Bricaire
feedback laws
laws were
were notnot the problem should be tackled.
aand
problem (2006)
has beenwhere
already however feedback
addressed in Aranda-Bricaire
and Moog
considered.
considered. (2006)
The where
immersion
The immersion however
of a feedback
continuous-time
of a continuous-time laws were
nonlinearnot 2.1 2.1 Problem
Problem statement
statement
and Moog (2006) where however feedback laws nonlinearwere not 2.1 Problem statement
considered.
system into
system intoThe Thea immersion
linear
a linear one of
one ofwas a
was continuous-time
studied for
studied for instanceinstance
nonlinear in
in 2.1 Problem statement
considered. immersion a continuous-time nonlinear
system
Menini
Menini andintoTornambé
and a linear (2011).
one wasThe studied
immersionfor instance
of
of discrete- in Consider
Consider two two multi–input
multi–input nonlinear nonlinear systems systems given given byby
system intoTornambé
a linear (2011).
one wasThe immersion
studied for instance discrete- in Consider twoΣmulti–input nonlinear systems given by
Menini
time
time and
systems
systems Tornambé
into
into linear
linear (2011).
or
or The
affine
affine immersion
polynomial
polynomial of discrete-
systems
systems was
was Consider two : ẋ
multi–input
1 1 = f 1 (x )
nonlinear
1 + g 1 (x 1 )u
systems 1 , given by
Menini and Tornambé (2011). The immersion of discrete- Σ1 : ẋ1 = f1 (x1 ) + g1 (x1 )u1 , (1)
time systems
studied in into linear
in Monaco and or affine polynomial
and Normand-Cyrot systems was
(1983a,b). Σ (1)
studied
time systems Monaco
into linear Normand-Cyrot
or affine polynomial (1983a,b).
systems was Σ2211 :: ẋẋ2211 == ff1221 (x
(x1221 )) +
+ gg1221 (x
(x1221 )u
)u1221 ,, (1)
studied in Monaco and Normand-Cyrot (1983a,b). Σ 2 : ẋ 2 = f 2 (x 2 ) + g 2 (x 2 )u 2 , (1)
studied in Monaco and Normand-Cyrot (1983a,b). Σ2 : ẋ2 = f2 (x2 ) + g2 (x2 )u2 ,

Copyright
2405-8963 ©© 2017,
2017 IFAC 9888Hosting by Elsevier Ltd. All rights reserved.
IFAC (International Federation of Automatic Control)
Copyright © 2017 IFAC 9888
Copyright
Peer review©under
2017 responsibility
IFAC of International Federation of 9888Control.
Automatic
Copyright © 2017 IFAC 9888
10.1016/j.ifacol.2017.08.1581
Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 E. Aranda-Bricaire et al. / IFAC PapersOnLine 50-1 (2017) 9480–9484 9481

where x1 ∈ n1 , x2 ∈ n2 , u1 , u2 ∈ m and fi , gi , i = 1, 2, ∂ϕ  


are meromorphic vector fields. [f1 (x1 ) + g1 (x1 )α̂1 (x1 )] + g1 (x1 )β̂1 (x1 )v =
∂x1 (5)
= [f2 (ϕ(x1 )) + g2 (ϕ(x1 ))α̂2 (ϕ(x1 ))]+
Consider two regular static state feedbacks
+g2 (ϕ(x1 ))β̂2 (ϕ(x1 ))v.
ui = α̂i (xi ) + β̂i (xi )v,
for i = 1, 2. By regular, it is meant that det(β̂i (xi )) = 0 Now define a new feedback control law for system Σ1 in
for i = 1, 2. Then, in closed–loop, systems Σ1 and Σ2 take the following manner:
the form: u1 = α1 (x1 ) + β1 (x1 )u2
= α̂1 (x1 ) − β̂1 (x1 )β̂2−1 (ϕ(x1 ))α̂2 (ϕ(x1 )) (6)
ẋi = [fi (xi ) + gi (xi )α̂i (xi )] + gi (xi )β̂i (xi )v, i = 1, 2. (2)
+ βˆ1 (x1 )β̂2−1 (ϕ(x1 ))u2
Without loss of generality, assume that n1 ≥ n2 . By
Direct substitution of the control law (6) into (3) produces:
definition, system Σ2 can be immersed into system Σ1
if there exist a regular static state feedback for each ż = f2 (ϕ(x1 )) + g2 (ϕ(x1 ))u2 = f2 (z) + g2 (z)u2 . (7)
system and a mapping ϕ : n1 → n2 , z = ϕ(x1 ) with
∂ϕ Therefore, system Σ2 can be immersed into system Σ1 ,
rank ∂x = n2 such that for the two closed–loop systems
1 without feedback on Σ2 .
the following equation holds true:
∂ϕ   3. MAIN RESULTS
ż = f1 (x1 ) + g1 (x1 )α̂1 (x1 ) + g1 (x1 )β̂1 (x1 )v
∂x
 1  (3)
 3.1 Abstract formulation of the Problem
= f2 (z) + g2 (z)α̂2 (z) + g2 (z)β̂2 (z)v  .
z=ϕ(x1 )
The present section characterizes the conditions which
In plain words, the lower dimensional system is said to be allow to immerse Σ2 into Σ1 in terms of the left annihilator
immersed into the higher dimensional one if, after regular of the given vector fields. Such a characterization is only
feedback on both systems, there is a state immersion such formal, since it assumes the knowledge of the mapping
that the lower order dynamics is reproduced as a part of z = ϕ(x1 ) which allows to get the immersion. However it
the higher dimensional one. gives some more insights on a different characterization of
the necessary and sufficient conditions.
In the case when both systems have the same dimension,
i.e. n1 = n2 we recover the usual definition of equivalence Proposition 2. System Σ2 can be immersed into system
of dynamical systems under state space diffeomorphism Σ1 by regular static state feedback on Σ1 and change of
and regular static state feedback variables if and only if there exists a mapping ϕ : n1 →
∂ϕ
n2 , z = ϕ(x1 ) with rank ∂x 1
= n2 which satisfies:
2.2 Simplification of the Problem ∂ϕ
i) f1 (x1 ) − f2 (ϕ(x1 )) = 0 mod g2 (ϕ(x1 ))
∂x1
In the following we will show that if Σ2 can be immersed
into system Σ1 , then this immersion can be obtained by ∂ϕ
ii) g1 (x1 ) = 0 mod g2 (ϕ(x1 )), (8)
feedback on system Σ1 only. Clearly, such a result notably ∂x1
 
simplifies the problem, and represents the starting point ∂ϕ
of our study. iii) rank g1 (x1 ) = m.
∂x1
Proposition 1. System Σ2 can be immersed into system
Σ1 if and only if there exist a regular static state feedback Proof. Necessity. Suppose that system Σ2 can be im-
u1 = α1 (x1 ) + β1 (x1 )v and a mapping ϕ : n1 → n2 , mersed into system Σ1 . Then equation (4) is satisfied and
z = ϕ(x1 ) with rank ∂x ∂ϕ
= n2 such that for the closed– therefore
1
loop system the following holds: ∂ϕ
([f1 (x1 ) + g1 (x1 )α1 (x1 )] + g1 (x1 )β1 (x1 )u2 ) =
∂x1
∂ϕ
([f1 (x1 ) + g1 (x1 )α1 (x1 )] + g1 (x1 )β1 (x1 )v) = = f2 (ϕ(x1 )) + g2 (ϕ(x1 ))u2 . (9)
∂x1
Equalizing the first and second terms of both sides one
= f2 (ϕ(x1 )) + g2 (ϕ(x1 ))v. (4) obtains
Proof. Of course if equation (4) is satisfied, then Σ2 ∂ϕ
can be immersed in Σ1 through the feedback laws u1 = [f1 (x1 ) + g1 (x1 )α1 (x1 )] = f2 (ϕ(x1 )) (10)
∂x1
α1 (x1 ) + β1 (x1 )v, u2 = v and the mapping z = ϕ(x1 ). ∂ϕ
g1 (x1 )β1 (x1 ) = g2 (ϕ(x1 )). (11)
Conversly, assume that system Σ2 can be immersed into ∂x1
system Σ1 . Then there exist z = ϕ(x1 ) and static state Multiplying (11) on the right hand side by β1−1 (x1 ), substi-
feedback laws u1 = α̂1 (x1 ) + β̂1 (x1 )v and u2 = α̂2 (x2 ) + tuting into (10) and after some straightforward algebraic
β̂2 (x2 )v, such that Eq. (3) is satisfied. This implies that operations, one obtains

9889
Proceedings of the 20th IFAC World Congress
9482
Toulouse, France, July 9-14, 2017 E. Aranda-Bricaire et al. / IFAC PapersOnLine 50-1 (2017) 9480–9484

∂ϕ where x1 ∈ n1 , x2 ∈ n2 , u1 , u2 ∈ m and Ai , Bi are


f1 (x1 ) − f2 (ϕ(x1 )) = −g2 (ϕ(x1 ))β1−1 α1 (x1 ) (12) matrices of appropriate dimensions.
∂x1
∂ϕ Proposition 4. Let m be the number of inputs of systems
g1 (x1 ) = g2 (ϕ(x1 ))β1−1 . (13) ΣLi , i = 1, 2. System ΣL2 can be immersed into system
∂x1
The previous relations prove the necessity of conditions i) ΣL1 by regular static state feedback on ΣL1 and change of
and ii) in (8). The necessity of iii) is obvious. variables if and only if there exists a linear transformation
T : n1 → n2 , z = T x1 with rank T = n2 and a regular
Sufficiency. Since condition ii) is satisfied then the rank static state feedback u1 = K1 x1 + G1 v which satisfy:
condition ensures that, for some β(x1 ) = 0
∂ϕ T A1 − A2 T = −B2 G−1
1 K1 (15)
g1 (x1 ) = g2 (ϕ(x1 ))β(x1 ),
∂x1 T B1 = B2 G−1
1 , rank(T B1 ) = m. (16)
that is
∂ϕ
g1 (x1 )β −1 (x1 ) = g2 (ϕ(x1 )) Proof. The Proof is similar to that of Propo3rsition 2,
∂x1 noting that ϕ(x1 ) = T x1 , α1 (x1 ) = K1 x1 and β1 (x1 ) =
while condition i) guarantees that for some α(x1 ), G1 .
∂ϕ Corollary 5. Let (ki1 , · · · , kim ), i = 1, 2 be the control-
f1 (x1 ) − f2 (ϕ(x1 )) = g2 (ϕ(x1 ))α(x1 ) lability indices respectively of Σ1 and Σ2 and assume
∂x1
without loss of generality that kij ≤ ki,j+1 for j ∈ [1, m−1]
∂ϕ and i = 1, 2. Then Σ2 can be immersed in Σ1 if and only
= g1 (x1 )β −1 (x1 )α(x1 ).
∂x1 if k2j ≤ k1j for all j ∈ [1, m].
As a consequence
∂ϕ 3.3 The case of single-input systems
f2 (ϕ(x1 )) + g2 (ϕ(x1 ))u2 = f1 (x1 )
∂x1
In this Section we consider the single input systems; that is
∂ϕ ∂ϕ
− g1 (x1 )β −1 (x1 )α(x1 ) + g1 (x1 )β −1 (x1 )u2 systems Σ1 and Σ2 given by (1) with m = 1 and n2 ≤ n1 .
∂x1 ∂x1
which shows that (4) holds true with β1 (x1 ) = β −1 (x1 ) In the case of single input systems, if both systems are
and α1 (x1 ) = −β −1 (x1 )α(x1 ). exactly linearizable by regular state feedback, then both
systems are equivalent to chains of integrators of lengths
Under the assumption of controllability in the first approx- n1 and n2 . Therefore the shorter chain—namely Σ2 — can
imation of system Σ2 , the following result is derived from always be immersed into the larger chain—namely Σ1 —.
Proposition 2. Thus, the only interesting cases arise when at least one of
Proposition 3. Let m be the number of inputs of systems them is not feedback linearizable.
Σi , i = 1, 2, and assume that Σ2 is controllable in the first Proposition 6. Suppose that system Σ2 is controllable and
approximation, that is that
dim span{g2 (x2 )adf2 g2 (x2 ), · · · , adfn22 −1 g2 (x2 )} = n2 (1) span{g2 (x2 ), . . . , adfn22 −2 g2 (x2 )} is constant dimen-
Then system Σ2 can be immersed into system Σ1 by sional and involutive
regular static state feedback on Σ1 and change of variables (2) There exists an integer k, n2 ≤ k < n1 , such that
only if there exists a mapping ϕ : n1 → n2 , z = ϕ(x1 )
with rank ∂x ∂ϕ
= n2 which satisfies: dim span{g1 (x1 ), . . . , adk−2
f1 g1 (x1 )} = k1
1

∂ϕ dim span{g1 (x1 ), . . . , adk−1


f1 g1 (x1 )} > k1
rank( g1 (x1 )) = m,
∂x1 Then, system Σ2 is linearizable through a change of
∂ϕ coordinates and a regular static state feedback. Moreover,
g1 (x1 ) = 0 mod g2 (ϕ(x1 )), system Σ2 can be immersed into system Σ1 through state
∂x1
immersion and regular static state feedback.
∂ϕ
adf1 g1 (x1 ) = 0 mod {g2 (ϕ(x1 )), adf2 g2 (ϕ(x1 ))},
∂x1 Proof. Condition 1 implies that system Σ2 is linearizable
and for k ≤ n2 − 1 by change of coordinates and regular static state feedback.
Therefore, in suitable coordinates, it can be written as
∂ϕ k
ad g1 (x1 ) = 0 mod{g2 (ϕ(x1 )), · · · , adkf2 g2 (ϕ(x1 ))}. x̃˙ 2 = A2 x̃2 + B2 v, (17)
∂x1 f1
3.2 The case of Linear Systems with the pair (A2 , B2 ) is Brunovsky canonical form. By
Condition 2, there exists a function, say λ(x1 ) such that
Consider now two linear systems given by dλ(x1 ) ⊥ adif1 g1 (x1 ), i ∈ [0, k − 2]. That is, λ(x1 ) has
ΣL1 : ẋ1 = A1 x1 + B1 u1 , relative degree k ≥ n2 . Therefore, the functions z1 =
(14) (λ(x1 ), . . . , λ(k−1) (x1 )) can always be completed by n1 − k
ΣL2 : ẋ2 = A2 x2 + B2 u2 ,

9890
Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 E. Aranda-Bricaire et al. / IFAC PapersOnLine 50-1 (2017) 9480–9484 9483

functions z2 = (λ̃k (x1 ), . . . , λ̃n1 (x1 )) such that after a [ge1 , [ge1 , ge2 ]] = 0 and [ge2 , [ge1 , ge2 ]] = ge1 . Then, the
control feedback the closed–loop system reads accessibility distribution is given by
ż1 = A1 z1 + B1 v Le = span{ge1 , ge2 , [ge1 , ge2 ]} (21)
(18)
ż2 = η(z1 , z2 )
and dim Le = 3. Therefore, system Σ1 and Σ2 could be
with (A1 , B1 ) a controllable pair. The proof follows ac- equivalent through change of coordinates. As a matter of
cording to Corollary 5. fact, this is the case.

The previous arguments could be used to face with the The left-annihilator of Le is spanned by the one-forms
problem removing the assumption of feedback linearizabil- ω1 = −dx3 − dz1
ity. The idea which can be used in this case, is to refer to ω2 = dx2 − (z2 cos x3 + z3 sin x3 )dx3 −
the infinitesimal Brunovsky form, introduced in Aranda- − sin x3 dz2 + cos x3 dz3
Bricaire et al. (1995). Therefore, there exist two differential ω3 = dx1 + (z3 cos x3 − z2 sin x3 )dx3 +
one-forms ω1 and ω2 which satisfy Hni i = span{ωi }. Note + cos x3 dz2 + sin x3 dz3
that the choice of ω1 and ω2 is unique up to multiplication
by a non zero function. Moreover, since Σ1 and Σ2 are not Integration of these one-forms and solving for the variables
feedback linearizable, these forms are not integrable, that zi yields
is dωi ∧ ωi = 0.
z 1 = x3
z2 = x2 sin x3 − x1 cos x3 (22)
4. EXAMPLE z3 = x2 cos x3 + x1 sin x3
4.1 Equivalence of two nonlinear systems by change of In these coordinates, system (19) takes exactly the form
coordinates (20).
The following example shows a possible generalization of It is interesting to show that the conjecture supported by
the early results presented in Aranda-Bricaire and Moog this Example is in full agreement with the theory presented
(2006). In that paper, it was shown that—for single input in the paper.
systems—if the accessibility distribution of the composite First, note that, for this example, f1 (x) = 0 and f2 (z) =
system formed by Σ1 and Σ2 has dimension exactly equal 0. Therefore, Condition i) of Proposition 2 is trivially
to n2 , then the dynamics of Σ2 can be immersed into the satisfied. Second, lengthy but staightforward computations
dynamics of Σ1 . show that the mapping z = ϕ(x) defined by (22) satisfies
In the example worked below, the composite system has ∂ϕ
dimension 6, while its accessibility distribution has di- g1 (x) ≡ g2 (ϕ(x)).
∂x
mension exactly equal to 3. Therefore, it is tempting to
conjecture that the two systems are equivalent by change Therefore, Conditions ii) and iii) of Proposition 2 are
of coordinates. satisfied as well.
Consider the two driftless systems:
ẋ1 = v cos x3 5. CONCLUSIONS
Σ1 : ẋ2 = v sin x3 (19)
ẋ3 = ω, Motivated by some practical systems whose control prob-
ż1 = ω lem is essentially solved by means of the control solution
Σ2 : ż2 = v − z3 ω (20) for a smaller embedded system, the general problem of
ż3 = z2 ω. the immersion of one system into another was stated and
solved. It generalizes the well known problem of equiv-
System Σ1 corresponds to the kinematic model of the alence of nonlinear systems, as well as their feedback
unicycle; (x1 , x2 ), x3 , v and ω correspond, respectively, linearization.
to the Cartesian coordinates, the orientation with respect Natural outlooks of this work would be the explicit com-
to the horizontal axis, the linear velocity and the angular putation of the required control laws, as well as the study
velocity of the unicycle. of its possible generalization to discrete-time systems.
For the composite system, we have that fe = 0, As a final comment, it is interesting to point out that
ge1 = (cos x3 , sin x3 , 0, 0, 1, 0) T the immersion of lower dimensional dynamics into higer
dimensional one has been used to solve a variety of
ge2 = (0, 0, 1, 1, −z3 , z2 )T .
problems. To mention a few: stabilization problems Carr
(1982); nonlinear regulation Isidori and Byrnes (1990)
It can be easily checked that and, more recently, adaptive control problems Astolfi and
[ge1 , ge2 ] = (− sin x3 , cos x3 , 0, 0, 0, −1)T , Ortega (2003).

9891
Proceedings of the 20th IFAC World Congress
9484
Toulouse, France, July 9-14, 2017 E. Aranda-Bricaire et al. / IFAC PapersOnLine 50-1 (2017) 9480–9484

REFERENCES
Aranda-Bricaire, E. and Hirschorn, R.M. (1999). Equiv-
alence of nonlinear systems to prime systems under
generalized output transformations. SIAM Journal on
Control and Optimization, 37(1), 118–130.
Aranda-Bricaire, E., Kotta, U., and Moog, C. (1996).
Linearization of discrete-time systems. SIAM Journal
on Control and Optimization, 34(6), 1999–2023.
Aranda-Bricaire, E. and Moog, C.H. (2006). Embedded
dynamics of continuous time nonlinear single input
systems. In Proc. 45th. IEEE Conf. Decision and
Control, 973–976. San Diego.
Aranda-Bricaire, E. and Moog, C.H. (2008). Linearization
of discrete-time systems by exogenous dynamic feed-
back. Automatica, 44(7), 1707–1717.
Aranda-Bricaire, E., Moog, C.H., and Pomet, J.B. (1995).
A linear algebraic framework for dynamic feedback
linearization. EEE Trans. Autom. Contr., 127–132.
Astolfi, A. and Ortega, R. (2003). Immersion and invari-
ance: A new tool for stabilization and adaptive control of
nonlinear systems. IEEE Trans. Autom. Contr., 48(4),
590–606.
Brockett, R.W. (1978). Feedback invariants for nonlinear
systems. In In Proceedings VII IFAC World Congress,
1357–1368. Helsinki.
Califano, C., Monaco, S., and Normand-Cyrot, D. (1999).
On the problem of feedback linearization. Systems
Control Lett., 36, 61–67.
Carr, J. (1982). Applications of Centre Manifold Theory,
volume 35 of App. Math. Sci. Springer Verlag.
Celikovsky, S., Anderle, M., and Moog, C.H. (2013). Em-
bedding the generalized acrobot into the n-link with
unactuated cyclic variable and its application to walking
design. In Proc. ECC13, 682–689. Zurich, Switzerland.
Isidori, A. and Byrnes, C.I. (1990). Output regulation of
nonlinear systems. IEEE Trans. Automat. Contr., 35(2),
131–140.
Jacubczyk, B. (1987). Feedback linearization of discrete-
time systems. Systems Control Lett, 9, 441–446.
Jakubczyk, B. and Respondek, W. (1980). On lineariza-
tion of control systems. Bull. Acad. Polon. Sci. Sr. Sci.
Math, 28, 517–522.
Menini, L. and Tornambé, A. (2011). Symmetries and
Semi-invariants in the Analysis of Nonlinear Systems.
Springer-Verlag.
Monaco, S. and Normand-Cyrot, D. (1983a). The immer-
sion under feedback of a multidimensional discrete time
nonlinear system into a linear system. Int. J. Contr,
245–261.
Monaco, S. and Normand-Cyrot, D. (1983b). On the
immersion of a discrete time polynomial analytic system
into a polynomial affine one. Syst. & Contr. Lett., 3, 83–
90.
Sylvester, J. (1884). Sur l’équation en matrices px = xq.
C. R. Acad. Sc. Paris, 99(2), 115–116.

9892

You might also like