Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

11th IFAC Symposium on Robot Control

11th IFAC
August Symposium
26-28, on Robot
2015. Salvador, Control
BA, Brazil
11th IFAC
11th IFAC Symposium
Symposium on
on Robot
Robot Control
Control
August
11th 26-28,
IFAC 2015. Salvador,
Symposium on RobotBA, Brazil
Control
Available online at www.sciencedirect.com
11th IFAC
August Symposium
26-28,
August 26-28, 2015. on Robot
Salvador, Control
Salvador, BA, Brazil
26-28, 2015. Salvador, BA, Brazil
August
August 26-28, 2015.
2015. Salvador, BA,
BA, Brazil
Brazil

ScienceDirect
IFAC-PapersOnLine 48-19 (2015) 033–038
Stable
Stable Model-Based
Model-Based Predictive
Predictive Control
Control for
for Wheeled
Wheeled Mobile
Mobile Robots
Robots using
using Linear
Linear
Stable
Stable
Stable Model-Based
Model-Based
Model-Based Predictive
Predictive
Predictive Control
Control
Matrix
Control for
for Wheeled
Wheeled
Inequalities
for Wheeled Mobile
Mobile
Mobile Robots
Robots
Robots using
using
using Linear
Linear
Linear
Stable Model-Based PredictiveMatrix Control
Matrix for Wheeled
Inequalities
Inequalities Mobile Robots using Linear
Matrix
Matrix Inequalities
Inequalities
Jonatas R. Pitanga*, HumbertoMatrix Inequalities
X. Araújo*, André G. S. Conceição*, Gustavo H. C. Oliveira**
Jonatas
Jonatas R. R. Pitanga*,
Pitanga*, HumbertoHumberto X. X. Araújo*,
Araújo*, André 
André G.
G. S.S. Conceição*,
Conceição*, Gustavo Gustavo H. H. C.
C. Oliveira**
Oliveira**
Jonatas
Jonatas R.
R. Pitanga*,
Pitanga*, Humberto
Humberto X.
X. Araújo*,
Araújo*, André

André G.
G. S.
S. Conceição*,
Conceição*, Gustavo
Gustavo H.
H. C. Oliveira**
Jonatas R. Pitanga*, * DepartmentHumberto X. Araújo*,
of Electrical 
André
Engineering, G.Federal
S. Conceição*,
University Gustavo
of Bahia H. C. Oliveira**
C. Oliveira**
** Department of Electrical 
Engineering, Federal University of Bahia
Salvador, ** Department
BA, Brazil of
Department
Department of
of
Electrical
(e-mail:
Electrical
Electrical
Engineering,
Engineering,
Engineering,
Federal
Federal University
humberto.araujo@ufba.br,andre.gustavo@ufba.br).
University of
of Bahia
Bahia
Salvador,
Salvador, BA,
* Department
BA, Brazil
Brazil (e-mail:
of Electrical
(e-mail: Engineering, Federal University of Bahia
Federal University
humberto.araujo@ufba.br,andre.gustavo@ufba.br).
humberto.araujo@ufba.br,andre.gustavo@ufba.br).
of Bahia
** Department
Salvador,
Salvador, BA, Brazil of
BA, Brazil Electrical
(e-mail:
(e-mail: Engineering, Federal University of Paraná
humberto.araujo@ufba.br,andre.gustavo@ufba.br).
humberto.araujo@ufba.br,andre.gustavo@ufba.br).
**
Salvador,
** Department
BA,
DepartmentBrazil of Electrical
(e-mail:
of Engineering, Federal University
humberto.araujo@ufba.br,andre.gustavo@ufba.br).
Electrical Engineering, Federal University of
of Paraná
Paraná
**
** Curitiba,
Department of PR, Brazil(e-mail:
Electrical Engineering, gustavo@eletrica.ufpr.br)
Federal University of Paraná
** Department Curitiba,
Department
Curitiba,
of
of Electrical
PR,
Electrical
PR,
Engineering,
Brazil(e-mail:
Engineering,
Brazil(e-mail:
Federal
Federal University
gustavo@eletrica.ufpr.br)
University
gustavo@eletrica.ufpr.br)
of
of Paraná
Paraná
Curitiba,
Curitiba, PR,
PR, Brazil(e-mail:
Brazil(e-mail: gustavo@eletrica.ufpr.br)
gustavo@eletrica.ufpr.br)
Curitiba, PR, Brazil(e-mail: gustavo@eletrica.ufpr.br)
Abstract: This paper presents practical results from the application of a new synthesis methodology based
Abstract: This
Abstract: This paper
paper presents
presents practical
practical results
results from
from the the application
application of of aa new
new synthesis
synthesis methodology
methodology based based
on model This
Abstract:
Abstract: predictive
This paper
paper control (MPC)
presents
presents practical
practical applied
results
results tofrom
a three-wheeled
from the
the application
application omnidirectional
of
of aa new
new mobile
synthesis
synthesis robot seeking
methodology
methodology based
based to
onon model
Abstract:
model predictive
This
predictive control
paper presents
control (MPC)
practical
(MPC) applied
results
applied to
to a
from
a three-wheeled
the
three-wheeledapplication omnidirectional
of a new of
omnidirectional mobile
synthesis
mobile robot
methodology
robot seeking
based
seeking to
to
on
onfollow
model
model pre-established
predictive control
predictive trajectories.
control (MPC)The
(MPC) The approach
applied
applied is based
to aa three-wheeled
to three-wheeled on the definition
omnidirectional an objective
mobile function
robot seekingwith to
onfollow
model pre-established
predictive trajectories.
control (MPC) approach
applied is basedbased on
to a three-wheeled theomnidirectional
on the definition of
omnidirectional anmobile
of an objective
mobile robot seeking
function with to
follow
a finite
follow
follow
pre-established
future time horizon.
pre-established
pre-established
trajectories.
The
trajectories.
trajectories.
The
resulting
The
The
approach
approach
approach
is
optimization
is
is based
based problem
on
on the
the
definition
is declaredof
definition
definition ofinan
anthe form robot
objective
objective
objective
seeking
of function
linear
function
function
with
matrix
with
with
to
aa finite
finite pre-established
follow future
future time
time horizon.
horizon. The
trajectories.
The resulting
The
resulting optimization
approach is
optimization based problem
on
problem the is declared
definition
is declared ofin
inanthe
the form
objective
form of
of linear
function
linear matrix
with
matrix
inequalities
aainequalities
finite
finite future
future (LMIs).
time
time The closed
horizon. The loop stability
resulting of the system
optimization problemis guaranteed
is through
declaredthroughin the constraints
form of related
linear matrix to
ainequalities
finite
the future time horizon.
(LMIs).
(LMIs).
non-increasing
The closed
horizon.
The
monotonicity
The
closed
Theloopresulting
loop
resulting
of the
optimization
stability of the
the system
optimization
stability of
objective
problem
system
problem
function. is
is
is declared
is guaranteed
guaranteed
declaredare
Constraints
in the
the form
in also
through form of
of linear
constraints
constraintslinear
implemented
matrix
related
matrix
related
in
to
to
the
inequalities
inequalities (LMIs).
(LMIs). The
The closed
closed loop
loop stability
stability of
of the system is guaranteed through constraints related to
the non-increasing
the non-increasing
inequalities
manipulated (LMIs).
variables
monotonicity
The closed
monotonicity
with
of the
loop
of
the objective
the of the
objective
stability
ofobjective
adapting
system
function.
thefunction.
system
the
is guaranteed
Constraints
isConstraints
control guaranteed
system to the
through
are
through
are also constraints
also
physical
implemented
constraints
implemented
limitations
related
in the
related
in
of the
to
to
the
the non-increasing
non-increasing
manipulated variables monotonicity
monotonicity
with the of
of
objective the
the objective
objective
of adapting function.
function.
the control Constraints
Constraints
system to are
are
the also
also
physical implemented
implemented
limitations in
in
of the
the non-increasing
manipulated variables
robot. Additionally, monotonicity
with
methods the are of
objective the
described objective
of adapting
to handlethe function.
the control
thecontrol Constraints
system
computational to are
the
delay also
physical implemented
in the robot limitations
model andin
of the
the
to
manipulated
manipulated variables
variables with
with the
the objective
objective of
of adapting
handlethe
adapting thecontrol system
system to the physical limitations of the
robot. Additionally,
manipulated
robot.
adjust
Additionally,
the variables
control law
methods
with
methods
to improve
are
theare described
objective
described
the global
to
of performance
adapting
to handle the
the control system to
computational
computational
of the system. to the
the physical
delay
delay in the
the robot
physical
in
limitations
robot model and
limitations
model
of
of the
and to
the
to
robot.
robot. Additionally,
Additionally, methods
lawmethods are
are described
described to
to handle the computational delay in the robot model and to
adjust the
robot.
adjust the control law
Additionally,
control to improve
improve
methods
to the
are the to handle
handle the
global performance
described
global performance theofofcomputational
system. delay
the system.
computational
the delay in in the
the robot
robot model
model andand to to
adjust
©
adjust
2015,
Keywords:the
the control
Mobile
control
IFAC law
law to
Robots,
to
(International improve
Model
improve the
the
Federation global
Predictive
global
of performance
Control
performance
Automatic of
(MPC),
of
Control) the
the system.
Linear
system.
Hosting Matrix
by ElsevierInequalities
Ltd. All (LMIs).
rights reserved.
adjust
Keywords:the control
Mobile law to improve
Robots, Model the global performance
Predictive Control of theLinear
(MPC), system. Matrix Inequalities (LMIs).
Keywords:
Keywords: Mobile
Mobile Robots,
Robots, ModelModel Predictive
Predictive Control
Control (MPC), (MPC), Linear
Linear Matrix
Matrix Inequalities
Inequalities (LMIs).
(LMIs).
Keywords: Mobile
Keywords: Mobile Robots,
Robots, ModelModel Predictive
Predictive Control
Control 

(MPC),
(MPC), Linear
Linear Matrix
Matrix Inequalities
Inequalities (LMIs).
(LMIs).
 increasing monotonicity of the objective function. The
1. INTRODUCTION  increasing monotonicity of the objective function. The
1. INTRODUCTION increasing
control
increasing law is monotonicity
developed inof
monotonicity state
of the
the objective
space function.
and minimizes
objective function. The
a finite
The
1. INTRODUCTION
1. INTRODUCTION
INTRODUCTION increasing
control
increasing law monotonicity
is developed
monotonicity in of
state
of the objective
space
thespace and
objective function.
minimizes
function. a The
finite
The
Advanced control techniques 1.
1. INTRODUCTION applied in mobile robots have control horizon
control
control
law
law
law
is
objective
is
is
developed
function
developed
developed
in
in
in
state
subject
state
state to
space
space
and
and
and
minimizes
constraints
minimizes
minimizesin aa finite
thea input
finite
finite
Advanced control techniques applied in mobile robots have horizon
control
horizon objective
law is
objective function
developed
function in subject
state
subject to
space
to constraints
and minimizes
constraints in
in the
thea input
finite
input
Advanced
become
Advanced control
increasingly
control techniques
widespread
techniques applied in
in recent
applied in mobile robots
years.robots
mobile have
Therehaveare signal. horizon The development
objective function of the robot
subject to model is presented
constraints in the in
input
Advanced
become
Advanced control
increasingly
control techniques
widespread
techniques applied
in
applied in
inasmobile
recent years.
mobile robots
There
robots have
are
have horizon
signal.
horizon objective
The
objective function
development
functionof subject
the
subject to
to constraints
robot model
constraintsis in the
presented
in input
the LMIs in
input
become
many
become increasingly
types of
increasingly mobile widespread
robots,
widespread in
such
in recent
recent years.
humanoid
years. There are
robots,
There are signal.
Section
signal. The
2.
The An development
introduction
development of
to
of the
the
the robot
MPC
robot model
technique
model is
is presented
using
presented in
in
become
many increasingly
types
becometypes of
increasingly mobile widespread
robots,
widespread in
such recent
as years.
humanoid
inomnidirectional
recent There
years. There are
robots,
are Section signal.
Section
signal. The
2.
The An development
introduction
development of
to
of the
the
the robot
MPC
robot model
technique
model is
is presented
using
presented LMIsin
in
many
trident
many typessnake of
of mobile
robots
mobile robots,
and
robots, such
such as
as humanoid
humanoid robots,
robots.
robots, is provided
Section 2.
2. An
An inintroduction
Section
introduction 3. to
The
to the
the MPC
results
MPC technique
from the
technique using LMIs
experiments
using LMIs
many types
trident
many typessnake of mobile
of mobile
robots robots,
and
robots, such as humanoid
humanoid robots.
omnidirectional
such as robots, is
robots, Section
provided
Section 2.
2. An
An inintroduction
Section
introduction 3. to
The
to the
the MPC
results
MPC technique
from the
technique using LMIs
experiments
using LMIs
trident snake
Omnidirectional
trident snake robots robots
robots have and
andbeen omnidirectional
highlighted among
omnidirectional the is
robots.
robots. and
is
is
provided
provided
provided
in
in Section
their analysis
in Section 3.
3. The
are presented The results
in Section
results from
from4. the
The experiments
the conclusions
experiments
trident
trident
mobile
snake
Omnidirectional
snake because
Omnidirectional
robots
robots
robots
robots have
robots have
they
and
andbeen
been
can
omnidirectional
highlighted
omnidirectional
highlighted
move in any
among
among
robots.
the
robots.
the
direction
and
is
and
of the
their
provided
their
study in Section
analysis
analysis
are
are
Section
are
3.
3. inThe
presented
presented
presented The results
in
in
Section
Section
results
Section
5.
from
from4.
4.
the
The
the
The
experiments
conclusions
experiments
conclusions
Omnidirectional
Omnidirectional
mobile robots robots have
robots
because have
they been
been
can highlighted
highlighted
move in any among
among the and
the
direction and
of thetheir
their
study analysis
analysis
are are
are presented
presented
presented in in
in
Section Section
Section
5. 4.
4. The
The conclusions
conclusions
Omnidirectional
mobile
without robots
the need robots
because
for have
they can
reorientation. been
can Thishighlighted
moveclass in any
any
of among the
direction
robots has and
of the their
the study analysis
study are are
are presented presented
presented in in
in Section Section
Section 5. 5. 4. The conclusions
mobile
mobile
without robots
robots
the need because
because
for they
they
reorientation. can move
move
This in
in
class any
of direction
direction
robots has of
of the study are presented
2. in
ROBOT Section
MODEL 5.
mobile robots
withoutmanoeuvrability
great the need because
need for they
for reorientation.
reorientation.
and is used can move
Thiswidely in
class ofany direction
ofinrobots
robots
dynamic has of the study are presented
2. in Section
ROBOT MODEL 5.
without the This class has 2.
without
great
without
great
the
the need
manoeuvrability
environments. need
manoeuvrability
Their
for reorientation.
and
for characteristics
reorientation.
and is
is used
used
This
give
class
Thiswidely
class an
widely
them
of
ofin
in
robots
dynamic
robots
dynamic
advantage,
has
has The mathematical model 2. ROBOT
2. ROBOT
ROBOT
MODEL
MODEL
MODEL
of an omnidirectional mobile robot,
great
great manoeuvrability
manoeuvrability
environments. Their and
and
characteristicsis
is used
usedgive widely
widely
them an in
in dynamic
dynamic
advantage, The mathematical 2.
model ROBOT
of an MODEL
omnidirectional mobile robot,
great
e.g., manoeuvrability
environments.
relative toTheir
robots inand
characteristics
the is used
Ackerman give widely
them
and an
the in dynamic
advantage,
differential The
as mathematical
shown in Figure model
1, is of an
described omnidirectional
in this section. mobile robot,
environments.
environments.
e.g., relative toTheir
Their
robots characteristics
characteristics
in the Ackerman give
give them
them
and an
an
the advantage,
advantage,
differential The
The
as mathematical
mathematical
shown in Figure model
model
1, is of
of an
an
described omnidirectional
omnidirectional
in this section. mobile
mobile robot,
robot,
environments.
relative
configurations
e.g., toTheir
robots
(Li and characteristics
in
Zell,the
2007).Ackerman give them
and an differential
the advantage, as The mathematical
shown in Figure model
1, is of an
described omnidirectional
in this section. mobile robot,
e.g., relative
relative
configurations
e.g., to
to robots
robots
(Li and in
in
Zell,the
the
2007).Ackerman
Ackerman and
and the
the differential
differential as
as shown
shown in
in Figure
Figure 1,
1, is
is described
described in
in this
this section.
section.
e.g., relative to(Li
configurations robots
and in the
Zell, 2007).Ackerman and the differential as shown in Figure 1, is described in this section.
configurations
configurations
However, (Li
(Li and
nonlinearities,and Zell,
Zell,
such2007).
2007).
as actuator saturation (from the
configurations
However, (Li
nonlinearities,and Zell,
such2007).
as actuator saturation (from the
However,
voltage
However, nonlinearities,
feed of the
nonlinearities, robot such
motors)
such as
as actuator
and the saturation
actuator (from
effect of friction
saturation (from the
on
the
However,
voltage
However, nonlinearities,
feed of the
nonlinearities, robot such
motors)
such as actuator
and
as actuator the saturation
effect
saturation of (from
friction
(from the
on
the
voltage
the
voltage feed
dynamics
feed of
of the
of
the robot
the motors)
robot,
robot motors) can and the effect
significantly
and the effect of friction
affect
of system
friction on
on
voltage
the
voltage feed
dynamics of the
of
feed(Conceicao
ofofthetherobot
the motors)
robot,
robot motors) can and the effect
significantly
and the effect of friction
affect system
of friction on
on
the
the dynamics
behaviour
dynamics of the robot,
et
robot,al., can
can significantly
2015). The
significantly model affect
affect system
predictive
system
the dynamics
behaviour
the dynamics of
(Conceicaothe
oftechniquerobot,
et
the robot, al., can
can significantly
2015). The
significantly model affect
affect system
predictive
system
behaviour
control
behaviour (MPC) (Conceicao
(Conceicao et
et has
al., 2015).
shown
2015). The
good
The model predictive
performance
model predictive in
behaviour
control
behaviour (MPC) (Conceicao
technique
(Conceicao et al.,
et has
al.,
al., 2015).
shown
2015). The
good
The model predictive
performance
model predictive in
control
practical
control (MPC)
applications
(MPC) technique
technique has
because
has shownshown
it
showncan good
handle performance
nonlinearities,
good performance
performance in
in
control
practical
control (MPC)
applications
(MPC) technique
technique has
because
has it
showncan good
handle
good nonlinearities,
performance in
in
practical
control
practical applications
signal constraints
applications because
becauseand it
it can handle
multivariable
can handle nonlinearities,
systems.
nonlinearities, For
practical
control
practical applications
signal constraints
applications because
becauseand it
it can handle
multivariable
can handle nonlinearities,
systems.
nonlinearities, For
control
example,
control signal
Oliveira
signal constraints
and Carvalho
constraints and multivariable
and multivariable
have successfully systems.
systems. For
applied
For
control
example,
control signal
Oliveira
signal constraints
and
constraints and
Carvalho
and and multivariable
have systems.
successfully
multivariable systems. For
applied
For Figure 1: Mobile robot Figure 2: Coordinate system
example,
MPC
example,to Oliveira
mobile
Oliveira robotsand
and Carvalho
(Oliveira
Carvalho have
have successfully
Carvalho, 1999;
successfully applied
Oliveira
applied Figure 1: Mobile robot Figure 2: Coordinate system
example,
MPC
example,to Oliveira
mobile
Oliveira robotsand
and Carvalho
(Oliveira
Carvalho and have
have successfully
Carvalho, 1999;
successfully applied
Oliveira
applied Figure
Figure 1:
1: Mobile
Mobile robot
robot Figure
Figure 2:
2: Coordinate
Coordinate system
system
MPC
and
MPC to
to mobile
Carvalho,
mobile robots
2003).
robots (Oliveira
(Oliveira and
and Carvalho,
Carvalho, 1999;
1999; Oliveira
Oliveira Figure
Figure 1:
1: Mobile
Mobile robot
robot Figure
Figure 2:
2: Coordinate
Coordinate system
system
MPC
and to mobile
Carvalho,
MPCCarvalho, robots
2003).
to mobile2003). (Oliveira and Carvalho,
robots (Oliveira and Carvalho, 1999; Oliveira 1999; Oliveira 2.1 Kinematic Model
and
and Carvalho, 2003). 2.1 Kinematic Model
and
On
and Carvalho,
the other
Carvalho, 2003).
hand, many control approaches
2003).many control approaches have also used have also used 2.1 Kinematic
2.1 Kinematic
Kinematic Model Model
Model
On
On the
the other
other hand,
hand, many Figure
2.1
2.12 Kinematic
illustrates Modelthe placement of the velocity vectors in
the
On
On numerical
the
the other
other properties
hand,
hand, many
many ofcontrol
control
control
approaches
linear matrix
approaches
approaches
have
inequalities
have
have
also
also
also
used
(LMIs).
used
used Figure 2 illustrates the placement of the velocity vectors in
the
On numerical
thecontext
other hand,properties
many of linear
ofcontrol matrix
approachesinequalities
have (LMIs).
also used Figure
the 2 illustrates
coordinate system the placement
the robot,of the velocitythe vectors in
the
In
the
the
In
numerical
the
numerical
numerical
the context
properties
of MPC,
properties
properties
of MPC,
the
of
of
the
linear
use
linear
linear
use
of
of
matrix
LMIs
matrix
matrix
LMIs
inequalities
was first
inequalities
inequalities
was first
(LMIs).
proposed
(LMIs).
(LMIs).
proposed
in
in
Figure
Figure
the
Figure
2 illustrates
2 illustrates
illustrates
coordinate
2 system theof
the
the
placement
placement
of the
placementrobot, ofgenerating
of
of
the
the velocity
the velocity
generating
velocitythe
following
vectors
vectors
following
vectors
in
in
in
the
In
the numerical
the context
1990s properties
of
(KothareMPC, et of
the
al., linear
use
1996). of matrix
LMIs
Since inequalities
was
then,first (LMIs).
proposed
this approach in the
set
the coordinate
of equations:
coordinate system
system of
of the
the robot,
robot, generating
generating the
the following
following
In
In
thethe
the context
context
1990s of
of
(KothareMPC,
MPC, et the
the
al., use
use
1996). of
of LMIs
LMIs
Since was
was
then,first
first proposed
proposed
this approach in
in the
set
the coordinate
of equations:
coordinate system
system of
of the
the robot,
robot, generating
generating the
the following
following
In
thethe context
1990s of MPC,
(Kothare the use of LMIs was first proposed in set set of
of equations:
has
the
the been
1990s widely
(Kothare et
et al., 1996).
explored.
al., Some
1996). Since
proposals
Since then, this
then,follow
this approach
the line
approach set
set of
equations:
of equations:
the 1990s
has
has been
1990s
been
developed
(Kothare
widely
(Kothare
widely
by
explored.
Kothare
et
et al.,
al.,
explored.et
1996).
Some
1996).
Some
al.
Since
Since etthen,
proposals
proposals
(Kothare then,
al.,
this
follow
this
follow
1996)
approach
the line
approach
the line
(Mao, equations:
has
has been
been
developed widely
widely
by explored.
explored.
Kothare et Some
Some
al. proposals
proposals
(Kothare et follow
follow
al., 1996) the
the line
line
(Mao,
has been
developed
2003; widely
Casavola by et explored.
Kothare
al., 2003).
et Some
al. proposals
(Kothare et al.,follow
1996)the(Mao,line
developed
developed
2003; Casavola by
by Kothare
Kothare
et al., et
2003). al. (Kothare
et al. (Kothare et al., 1996)
et al., 1996) (Mao,
(Mao,
developed
2003; Casavola by Kothare
et al., et
2003). al. (Kothare et al., 1996) (Mao, (1)
2003;
2003; Casavola
In thisCasavola
study, the et al.,
et al., 2003).
2003).
application of a new MPC strategy is (1)
2003;
In this Casavola
study, et al.,
the 2003).
application of a new MPC strategy is (1)
(1)
In this
proposed study,
for anthe application
omnidirectional of a new
robot. The MPC strategy
constraints areis (1)
(1)
In
In this
this
proposed study,
study,
for anthe
the application
application
omnidirectional of
of a
a new
new
robot. The MPC
MPC strategy
strategy
constraints areis
is
In this
proposed
based onstudy,
for
LMIs anthe
and application
omnidirectional
the stability of isa new
robot. The
guaranteedMPC bystrategy
constraints
the are
non- is The equations in (1) can be rewritten as follows:
proposed
proposed for
for an
an omnidirectional
omnidirectional robot.
robot. The
The constraints are
based
proposed
based on LMIs
on for
LMIs an and the
the stability
omnidirectional
and is
stability robot. The constraints
is guaranteed
guaranteed by
by the
the non-
constraints are
are The
non-
equations
The equations
equations in in (1)
in (1) can
(1) can
can bebe rewritten
be rewritten
rewritten as as follows:
as follows:
follows:
based
based on
based on LMIs
on LMIs and
LMIs and the
and the stability
the stability is
stability is guaranteed
is guaranteed by
guaranteed by the
by the non- The
the non-
non- The equations
The equations in in (1)
(1) can
can bebe rewritten
rewritten as as follows:
follows:
Copyright
2405-8963 ©© 2015,
2015 IFAC
IFAC (International Federation of Automatic Control)34 Hosting by Elsevier Ltd. All rights reserved.
Copyright © 2015 IFAC 34
Copyright
Peer review
Copyright ©
© 2015
under IFAC
responsibility
2015 IFAC of International Federation of 34
Automatic
34 Control.
Copyright
Copyright ©
© 2015
2015 IFAC
IFAC
10.1016/j.ifacol.2015.12.006 34
34
IFAC SYROCO 2015
August 26-28, 2015. Salvador, BA, BrazilJonatas R. Pitanga et al. / IFAC-PapersOnLine 48-19 (2015) 033–038
34

where is the rotational torque. The geometric and the


(2) dynamic parameters of the motor are shown in Table 1
(Conceição and Correia, 2012).
where , i=1, 2 and 3, (3)
Table 1: Model parameters
Symbol Description Value
is π/6 rad and b is the distance from the wheels to the center
r1; r2; r3 (m) radii of the wheels 0.035
of mass. The robot velocities can be written as functions of l1, l2, l3 motor reduction ratios 19
the angular velocities of the motors, considering the In (kg.m2) inertial moment of the robot 0.025
equations in (3), where ri is the radius of each wheel i and li is M (kg) mass of the robot 1.258
the motor reduction ratio of each motor i. Thus the following Bv (N.s/m) viscous friction related to v 2
model is obtained: Bvn (N.s/m) viscous friction related to vn 1.5
Bω (N.s/m) viscous friction related to ω 0.024
Cv (N) Coulomb friction related to v 1.2
, (4) Cvn (N) Coulomb friction related to vn 0.8
Cω (N) Coulomb friction related to ω 0.0035
Ra1, Ra2, Ra3
armature resistances 1.66
2.2 Dynamic Model (Ω)
Kt1, Kt2, Kt3
emf constants 0.0059
(V.s/rad)
The model of the dynamic behaviour of the robot in Figure 1
distance from the wheels to the center of
is based on a previous study that used a simplified model, b (m)
mass
0.1
i.e., both Coulomb friction and viscous friction are included,
but static friction is neglected (Conceição et al., 2009). Thus, 2.3 Representation in State Space
using Newton’s second law and using the robot as a
reference, the dynamic model is given as follows: A non-linear term appears in the representation of the state
space of the robot model from the Coulomb friction. Thus,
(5) the following equations are obtained:
and (6) (17)
(18)
(7) where:
where: (19)
(8)

and represents the force vector on the robot (20)


body and the torque moment acting on the center of gravity
of the robot. It is assumed that the center of gravity is equal
to the geometric center. The inertial moment is and the
mass of the robot is . Using the relations illustrated in
Figure 2, the following relations are developed between the (21)
forces on the robot and the wheels:
(9)
and (10) The control input corresponding to the motor’s armature
(11) voltages is the vector where u(k) = [u1(k) u2(k) u3(k)]T , the
output and the state-variables of the system are the robot
In matrix form, the equations are as follows: velocities, y(k) = x(k) = [v(k) vn(k) w(k)]T. Assuming the
presence of a zero-order hold, a discrete model is obtained
(12) with the following structure in state space:
The dynamics of each motor can be described by the (22)
following equations: (23)
where the following are defined:
and (13) and (24)
(14) (25)
where is the inductance of the motor armature, is the The following model is then obtained:
armature resistance and is the electromotive force (emf)
(26)
constant. The electrical circuit dynamics of the motor can be
neglected, which yields the following: (27)
(15) The control law is then designed based on this model, with
The traction force of each wheel is as follows: the exception of the nonlinearity from the Coulomb friction.
Control systems in discrete time are subject to time delays
(16) caused by the computational time for processing the control
law and by the data communication time. The specialized

35
IFAC SYROCO 2015
August 26-28, 2015. Salvador, BA, BrazilJonatas R. Pitanga et al. / IFAC-PapersOnLine 48-19 (2015) 033–038 35

literature (Astrom and Wittenmark, 1996) recommends that (33)


these delays be less than 10% of the sampling period to where and are the measured output and variation
assure that their influence on closed loop system performance of the system state. If the following vectors are defined:
can be neglected. In the case of the robot shown in Figure 1
and its control hardware, the computational and
communication processing time can vary between 40 and 60 , (34)
ms. This period is close to the sampling period used here,
which varies from 50 ms to 100 ms. Thus, if a delay is
inserted into the robot model, the resulting model will be then the predicted output can be calculated in the finite
more representative of the dynamics of the robot, resulting in prediction horizon using the following equation:
an improvement of system performance. Including a delay of (35)
one sampling period in the state space model results in the where:
following:

(28) (36)

(29)

and (37)
3. MPC CONTROLLER
The algorithm used in this study was originally proposed by Thus, the objective function shown in equation (30) is
Araújo et al. (Araújo et al., 2011). A sufficient condition to equivalent to the following:
guarantee closed loop stability obtained with LMIs is (38)
imposed on the controller. This condition is defined by the where:
imposition of non-increasing monotonicity of the objective (39)
function at each iteration. Thus, the MPC algorithm is
reduced to a convex optimization problem involving LMIs.
The idea that relates closed loop stability with the non- and (40)
increasing monotonicity of the objective function has been
explored in the literature (Scokaert and Clarke, 1994; Zheng The vector r refers to the reference signals. The system
and Morari, 1993; and Oliveira et al., 2000). states are assumed to be available at each sampling period. In
this study, the system states are the outputs.
3.1 Objective Function
The control law uses a quadratic objective function with a 3.2 Constraints on the Inputs
finite horizon as follows: In this study, constraints on the actuators are designed to
establish the lower and upper limits for the control signal
(30) and its increments as follows:

(30)
where is the prediction horizon, is the control horizon,
(41)
>0 and >0 are the weighting matrices, and
are the output predicted at time and the where is the number of inputs from the plant, and
optimal value of the control signal increments at time are the lower and the upper limits, respectively, of the
, respectively, computed at time , and is control signal from input i at the prediction time ,
the reference at time . The control law is obtained by and are the lower and the upper limits, respectively, of
minimizing the objective function relative to the control
signal increment as follows: the variation of the control signal from the input i at the
prediction time . Because the following relations are
(31) true:
The control signal is assumed to be constant at times greater
(42)
than ; i.e., for .
At each iteration, according to the receding horizon the constraints on the control signal can be manipulated using
principle, vector , i.e., the optimal sequence of the control the Schur complement, resulting in the following inequalities:
signal increments, is calculated. Only the first control signal
increment is applied to the system. To synthesize the
controller, a model of the system to be controlled is defined
by the state space equations as follows:
(32) (43)

36
IFAC SYROCO 2015
August 26-28, 2015. Salvador, BA, BrazilJonatas R. Pitanga et al. / IFAC-PapersOnLine 48-19 (2015) 033–038
36

The notation (M)i denotes specifically line i of matrix M. The matrix can be partitioned as follows:
The constraints on the control signal variations in (41) can be (54)
rewritten in a similar manner as follows: The MPC problem is thus presented below with the
guarantee of stability (Araujo et al., 2011).
(44) Theorem 1: The control law, with the guarantee of
The definition of the condition for stability, which will stability, including the constraints on the control signal and
be shown in the sequence, requires the calculation of the on its increments, can be calculated from the following
optimal value of , i.e., the optimal values of the control optimization problem:
signal increments calculated in the previous iteration, (55)
computed at time as follows: where:

(45) (56)

Constraints on the control signal must also be included at (57)


time . Defining the matrix T as follows:
(46)
(58)
and in a manner analogous to the inequalities developed in
(43) and (44), calculated at time , the limits can be (59)
determined at time as follows: and with the added constraints (43), (44), (47) and (48).
Additionally, is the optimal value of calculated at time
. At each iteration, two sequences of variations of the
control signals are calculated, , starting at time , and ,
starting at time . Only the first element of the vector
is actually used at time . In the first iteration, the LMI (58)
is not used because there is no value available for .
This control law, which considers the optimal values of
calculated at previous, present and future times, guarantees
(47) that the upper limit of the objective function is monotonically
non-increasing, leading to bounded-input bounded-output
(BIBO) stability in a closed loop.
(48) The LMI (56) is related to the upper limit of the objective
function. LMI (57) is similar to LMI (56), but it applies to
3.3 Control Law . LMIs (58) and (59) guarantee closed loop stability
through the imposition of non-increasing monotonicity of the
The minimization of the objective function is performed
upper limit of the objective function as follows:
with the aid of an upper limit as follows:
(61)
(49)
Considering (35), (49) can be rewritten as follows: where (*) signifies the optimal value of (55).
During the experiments, the control law given by
(50) Theorem 1 was shown to have aggressive tuning almost
independent of the weighting matrices of the objective
Using the Schur complement, it is possible to rewrite
function. This behaviour is undesirable for robot control
(50) as follows:
because the closed loop system exhibits elevated oscillations
(51) when the tuning is too aggressive, as will be illustrated in
Section 5. The oscillations are caused by delays in the system
and nonlinearities and uncertainties in the model, among
The symbol * is used to indicate a symmetrical structure. other reasons.
To determine the stability condition, the following matrices
are defined: The minimized upper limit is not that of the objective
function at the current time, k; it is that of the objective
function at the following time, , i.e., , described by
(52) inequality:
(62)

(53) The product then yields the control signal increments to


apply according to the receding horizon principle.
Even though these increments are present in the
calculation of the predicted error, they are not weighted by

37
IFAC SYROCO 2015
August 26-28, 2015. Salvador, BA, BrazilJonatas R. Pitanga et al. / IFAC-PapersOnLine 48-19 (2015) 033–038 37

the matrix R, whose objective is to guarantee that the control period and also because the phase margin of the closed loop
force is considered in the control calculation. Thus, the system decreases.
algorithm uses the elements of to minimize the predicted
error without, however, considering the control force
employed in these elements. To solve this problem, a
modification of inequality (62) is proposed as follows:

(63)

where is a new weighting matrix. Applying the Schur


complement, the LMI is obtained as follows:

(64) Figure 3: Response with no consideration of the delay

where: 4.2 Step Response Including the computational delay


(65)
As shown in Figures 4 and 5, the data show an improvement
Thus, by minimizing , both the objective function and in the system response when a delay is included in the control
the control force associated with the elements are law, i.e., applying to model (28) and (29). Figure 4 shows the
minimized. The aggressiveness of the controller can now be system response by tuning to 0.2, and Figure 5 shows the
adjusted, using the weighting matrices , and . system response with at 0.6. The increase in the control
This modification does not alter the non-increasing signal weighting yields a signal with more damped
monotonicity of the objective function. Therefore, Theorem 1 oscillations but with a larger rise time.
still describes the control law, and it will be used with LMI
(57) replaced by LMI (64).

4. RESULTS
To verify the performance of the controller, real-time
experiments were performed using the robotic platform.
Some of the experiments are described below. In the first two
experiments, the tracking capacity of the reference step of the
control system is tested. In two of the tests, the tracking
capacity of a different step trajectory is verified. Finally, the
effect of changing the sampling period from 50 ms to 100 ms
is analyzed. The weighting matrices used in the experiments
Figure 4: Response with equal to 0.2
are as follows:

(66)
where is 0.05, 0.2, 0.6 or 1.0, depending on the test
performed. The computational processing time is critical in
this project. The dimension of the matrices involved in the
controller, and consequently the algorithm processing time,
increases as the prediction horizon increases. Therefore, in
this experiment, the prediction horizon selected is 1 to keep
the processing time low.

4.1 Step Response with no computational delay


In this experiment, the velocity reference vector is Figure 5: Response with equal to 0.6
. From 0 to 3.3 seconds, ; after 3.3 In all cases, the performance with is worse than
seconds, . Figure 3 shows the system response with the performance with , resulting in larger response
the controller given by Theorem 1, without taking into times, larger rise times and more signal overshooting. This
consideration the stability condition, with a sampling period results from the non-linearity caused by the Coulomb friction.
of 50 ms. The value of is 0.2. The data show many high- Figure 6 shows the results from the experiment with at
frequency oscillations in the response. This occurs because of 0.05. With the increase in the sampling period to 100 ms, the
the computational delay of approximately one sampling rise time is lower, resulting in better performance.

38
IFAC SYROCO 2015
August 26-28, 2015. Salvador, BA, BrazilJonatas R. Pitanga et al. / IFAC-PapersOnLine 48-19 (2015) 033–038
38

4.3 Including the stability guarantee based on non-increasing monotonicity of the objective
function. Additionally, constraints in the control signals are
Figure 6 shows the system response with LMI (57), using
included, satisfying the voltage and current limitations of the
a 100ms sampling period. When the reference is modified,
actuators. An undesirable behaviour is observed in the con-
constraint (58) is liberated, maintaining the other constraints.
trol technique, showing aggressive tuning nearly in-
The change of the reference implies an increase in the
dependent of the weighting matrices of the objective
predicted error and, therefore, in the objective function,
function. Therefore, a modification is proposed in the control
according to (38). Then constraint (58) is applied, forcing the
law that permits greater sensitivity of the controller tuning
non-increasing monotonicity of the objective function, as
relative to the weighting matrices. During execution of the
shown in Figure 8a. The controller shows smoother tuning
design, there is a significant delay in the system from the
with fewer oscillations only with elevated values. Even so,
long processing time and the long communication time.
the controller performance is not very good. This result
Including a delay in the model of the robot to compensate for
illustrates the need for the modification represented by LMI
these delays improves the performance of the closed loop
(57). Figures 7 and 8b show the results of this modification,
system.
with harmonized at 0.5.
REFERENCES
Araújo, H., Conceição, G., Oliveira, G. and Pitanga, J.
(2011). Model Predictive Control based on LMIs
Applied to an Omni-directional Mobile Robot, Proc. of
the 18th IFAC World Congress, 18(1):8171-8176.
Åström, K. and Wittenmark, B. (1996). Computer-controlled
systems: theory and design.
Casavola, A., Famularo, D. and Franze, G. (2004). Robust
constrained predictive control of uncertain norm-
bounded linear systems. Automatica, 40:1865–1876.
Conceição, A. G. S., Correia, M. D., Martinez, L. (2015).
Modeling and friction estimation for wheeled
omnidirectional mobile robots. Robotica, available on
Figure 6: Response with the stability guarantee
CJO2015. doi:10.1017/S0263574715000065.
Conceição, A., Moreira, A. and Costa, P. (2009). Practical
approach of modeling and parameters estimation for
omnidirectional mobile robots. Mechatronics,
IEEE/ASME Transactions on, 14(3):377-381.
Conceição, A. G. S., Correia, M. D. (2012). Modeling of a
Three Wheeled Omnidirectional Robot Including
Friction Models. 10th International IFAC Symposium on
Robot Control, 10(1):7-14.
Kothare, M., Balakrishnan, V. and Morari, M. (1996). Robust
constrained model predictive control using linear matrix
inequalities. Automatica, 32(10):1361–1379.
Li, X. and Zell, A. (2007). Motion control of an
Figure 7: Response with an adjustment in the objective function omnidirectional mobile robot. ICINCO-RA, 125-132.
Mao, W. (2003). Robust stabilization of uncertain
timevarying discrete systems and comments on "an
improved approach for constrained robust model
predictive control". Automatica, 39(6):1109–1112.
Oliveira, G., Amaral, W., Favier, G. and Dumont, G. (2000).
Constrained robust predictive controller for uncertain
processes modeled by orthonormal series functions.
Automatica, 36(4):563–572.
(a) Without adjustment b) After adjustment Oliveira, G. and Carvalho, J. (2003). A non-linear predictive
control scheme for nonholonomic mobile robots. IFAC
Figure 8: Upper limit of the objective function,
Symposium on Robot Control (SYROCO), Proc. of the
7th IFAC Symposium on Robot Control.
5. CONCLUSIONS Scokaert, O. and Clarke, D. (1994). Stabilizing properties of
constrained predictive control. IEE Proceeding: Control
In this study, practical results obtained from the application Theory, 141(5):295–304.
of a new MPC synthesis method are shown. The objective Zheng, Z. and Morari., M. (1993). Robust stability of
function is developed on a finite horizon, formulated with constrained model predictive control. Proc. of American
LMIs, for an omnidirectional robot. A peculiarity of this Control Conference, 379–383.
method is the existence of a sufficient stability condition

39

You might also like