Proceedings of the 3rd IFAC Conference on

Proceedings of
Proceedings of the
the 3rd IFAC
IFAC Conference on
on
Advances
Advances in of the 3rd
3rd IFAC Conference
in Proportional-Integral-Derivative
Proceedings Conference on Control
Available
Proportional-Integral-Derivative online at www.sciencedirect.com
Advances
Ghent, in
Proceedings of the
Belgium,
Advances in May3rd IFAC
9-11, Conference on Control
Proportional-Integral-Derivative
2018
Proportional-Integral-Derivative Control
Control
Ghent,
Ghent, Belgium,
Proceedings
Belgium,
Advances in May
of the
May 9-11,
3rd IFAC
9-11, 2018
Conference on Control
2018
Proportional-Integral-Derivative
Ghent, Belgium, May 9-11, 2018
Advances in Proportional-Integral-Derivative Control
Ghent, Belgium, May 9-11, 2018
Ghent, Belgium, May 9-11, 2018 ScienceDirect
IFAC PapersOnLine 51-4 (2018) 220–225
Robust
Robust Tuning Rules for Series Elastic
Robust Tuning
Tuning Rules
Rules for
for Series
Series Elastic
Elastic
Actuator
Robust TuningPID Cascade
PIDRules for Controllers
Series
Actuator
Robust Tuning Cascade
Rules Series Elastic
for Controllers Elastic
Actuator
Stefano Ghidini
Actuator
Stefano
PID
∗
GhidiniPID
∗
∗
Cascade
Manuel Beschi
Cascade
Manuel
∗ Controllers ∗
Nicola
Beschi ∗∗ Nicola Pedrocchi
Controllers
Pedrocchi ∗∗
Stefano
Stefano Ghidini Ghidini ∗ Antonio Manuel
Manuel Beschi Beschi
Visioli ∗∗∗Nicola Pedrocchi ∗
Nicola Pedrocchi ∗
∗∗∗
Stefano Ghidini ∗∗ Antonio Manuel
Antonio
Antonio
Visioli
Beschi
Visioli
Visioli
∗∗
∗ Nicola Pedrocchi ∗
∗∗
Stefano Ghidini Antonio Manuel Beschi Visioli ∗∗ Nicola Pedrocchi
∗
∗ Institute of Industrial Technologies and ∗∗ Automation, National
∗
Institute of Industrial Antonio
TechnologiesVisioli and Automation, National
∗
Research Institute
Institute Councilof Industrial
of Industrial
of Technologies
Technologies and
and Automation,
Automation, National
National
Research
∗
Research Institute Council
Councilof of Italy,
Industrial
of Italy,
Via
Via Corti
Italy, Technologies
Via Corti 12,
Corti 12,and
12,
20133,
20133, Milan,
Milan, Italy
Automation,
20133, Milan, Italy (e-mail:
(e-mail:
National
Italy (e-mail:
Research Council of Italy, Via Corti
∗ stefano.ghidini,manuel.beschi,nicola.pedrocchi@itia.cnr.it)
Institute of Industrial Technologies 12, 20133,
and Milan,
Automation, Italy (e-mail:
National
∗∗Research stefano.ghidini,manuel.beschi,nicola.pedrocchi@itia.cnr.it)
Council of Italy, Via Corti
stefano.ghidini,manuel.beschi,nicola.pedrocchi@itia.cnr.it) 12, 20133, Milan, Italy (e-mail:of
∗∗Research stefano.ghidini,manuel.beschi,nicola.pedrocchi@itia.cnr.it)
Department Council of Mechanical
of Italy, Via and Industrial
Corti 12, 20133,Engineering,
Milan, Italy University
(e-mail:of
∗∗ Department of Mechanical and Industrial
stefano.ghidini,manuel.beschi,nicola.pedrocchi@itia.cnr.it)
Department of Mechanical and Industrial Engineering,
Engineering, University
University
∗∗
Department Brescia, of Via
Mechanical
Branze and
38, Industrial
25123
stefano.ghidini,manuel.beschi,nicola.pedrocchi@itia.cnr.it) Engineering,
Brescia, Italy University of
(e-mail: of
∗∗
Department Brescia,
Brescia, of Via Branze
Mechanical
Via Branze 38,
and
38, 25123
Industrial Brescia, Italy
Engineering, (e-mail:
University of
∗∗
Department Brescia, of Via Branze
Mechanical 38, 25123
25123
antonio.visioli@unibs.it)
and Industrial
Brescia,
Brescia, Italy
Italy (e-mail:
Engineering, (e-mail:
University of
Brescia, Via Branze antonio.visioli@unibs.it)
38, 25123
antonio.visioli@unibs.it) Brescia, Italy (e-mail:
Brescia, Via Branze antonio.visioli@unibs.it)
38, 25123 Brescia, Italy (e-mail:
antonio.visioli@unibs.it)
Abstract: This paper deals with antonio.visioli@unibs.it)
the control of a collaborative robot manipulator with series
Abstract:
Abstract: This
This paper paper deals
deals with
with the control
thetuning
controlrulesof
of aaafor collaborative
collaborative robot
robot manipulator
manipulator with
with series
series
Abstract:
elastic This
actuators. paper
In deals
particular, with
robust the control of collaborative
cascade robot
control of manipulator
the joints are with series
presented.
elastic
Abstract:
elastic actuators.
This
actuators. In
In particular,
paper deals
particular, robust
with
robust the tuning
control
tuning rules
of
rules a for cascade
collaborative
for cascade control
robot
control of
of the joints
manipulator
the joints are
are presented.
with series
presented.
elastic
Both
Abstract: actuators.
the motor In particular,
This velocity
paper and
deals robust
link tuning
position rules
of afor
control cascade
loops are control
considered.of manipulator
theThejoints arewith
proposed presented.
tuning
Both
elastic
Both allow
Both
rules
the
the
motor
actuators.
the motor
motor
the
velocity
In particular,
velocity
velocity
online and with
and
and
computation
link
robust
link
link
the control
position
tuning
position
position
of robust
control
rules
control
control
control
collaborative
loops
for
loops
loops
are
cascade
are control
are
parameters
robot
considered.
considered.
considered.
to of the
cope
The
joints
The
The
with
proposed
are
proposed
proposed
the different
series
tuning
presented.
tuning
tuning
link
elastic
rules
Both
rules actuators.
allow
the
allow the
motor
the In
online particular,
velocity
online computation
and
computation robust
link of tuning
robust
position
of robust rules
control
control for
controlloops cascade
parameters
are
parameters control
considered.
to of
to cope the
cope with joints
with
The the are
proposed
the presented.
different
different link
tuning
link
rules
reflected
Both allow
the the online
inertia.
motor velocity computation
Experimental and results
link of robust
show
position control
the
control parameters
effectiveness
loops are of theto method
cope with
considered. Thein the
real
proposeddifferent
applications. link
tuning
reflected
rules
reflected allowinertia.
the
inertia. Experimental
online computation
Experimental results
results of show
robust
show the effectiveness
control
the parameters
effectiveness of
of the
theto method
cope
method within
in real
the
real applications.
different
applications. link
reflected
rules allowinertia.
the Experimental
online computation results of show
robust the effectiveness
control parameters of the method in real applications.
© 2018,
reflected IFAC
inertia.(International
Experimental Federation
results of Automatic
show the Control)
effectiveness Hosting bytoElsevier
of the cope with
method Ltd. therights
in All
real different link
reserved.
applications.
Keywords:
reflected
Keywords: Collaborative
inertia.
CollaborativeExperimental robots;
robots; Series
results
Series elastic
show
elasticthe actuator;
effectiveness
actuator; Robust
Robust of control;
the method
control; PID;
PID; in Tuning
real
Tuning rules.
applications.
rules.
Keywords: Collaborative
Keywords: Collaborative robots; robots; Series
Series elasticelastic actuator;
actuator; Robust Robust control;
control; PID; PID; Tuning
Tuning rules. rules.
Keywords: Collaborative robots; Series elastic actuator; Robust control; PID; Tuning rules.
Keywords: 1. Collaborative robots; Series elastic actuator;
1. INTRODUCTION
INTRODUCTION Robust control;
model mismatches:
model mismatches: the PID;
the Tuning
reflected rules.
inertia strongly depends depends
1. INTRODUCTION
1. INTRODUCTION on model
model the mismatches:
mismatches:
robot the reflected
the
configuration reflected
reflected
and on
inertia
inertia
inertia
the
strongly
stronglyunknown)
strongly
(possibly depends
depends
Standard industrial
Standard industrial robots
1. robots
INTRODUCTION are not
are not typically
typically designed designed to to on on on the
model the
robot
the mismatches:
robot
robot
configuration
the spring
configuration
configuration
and
reflected
and
on the
the (possibly
oninertia
and characteristic
on the strongly
(possibly
(possibly
unknown)
depends
unknown)
unknown)
Standardwith industrial robots
1. robots
INTRODUCTION are notnot typically designed In payload.
to payload.
model Further, the can be non-
non-
Standard
co-exist
co-exist
industrial
with humans
humans in the
in the same are
same typically
working
working
designed
environment.
environment.
to
In on the mismatches:
payload.
payload.
linear,
Further,
robot
Further,
Further,
aging can
the spring
the
configuration
the
the
increase
reflected
spring
spring oninertia
and characteristic
theand
characteristic
characteristic
hysteresis
strongly
(possibly can
canunknown)
can
backlash,
depends
be
be static
be non-
non-
Standard
co-exist
co-exist
fact, industrial
with
with
industrial humans
humans
robots robots
in
in
are the
the are not
same
same
usually typically
working
working
employed designed
environment.
environment.
in several to linear,
In
In
tasks on the robot
aging configuration
can increase and characteristic
on theand
hysteresis (possibly
backlash, unknown) static
Standard industrial robots are not typically designed to payload.
linear, Further,
aging can the
increase spring hysteresis and can
backlash, be non-
static
fact, industrial
co-exist
fact, with humans
industrial robots
robots in are
arethe usually
same employed
usually working
employed in several
wallsenvironment.
in fences.
several They In linear,
tasks
tasks and
payload. aging
viscous can increase
friction
Further, terms
the spring hysteresis
stronglycharacteristicand backlash,
depend on cantemperaturebe static
non-
fact,
in
in anindustrial
co-exist
an
fact,
in an
isolated
with humans
isolated
industrial
isolated
robots
workspace
workspace
robots
workspace
are
in
are
by
the
by
by
usually
means
means
usually
means
employed
same employed of walls
working
of
of walls
in
or
or
in
or
several
environment.
fences.
several
fences. They In and
tasks
tasks
They and and viscous
linear,
(Simoni aging
viscous
viscous
et
friction
can
friction
friction
al., 2017),
terms
increase
terms
terms
and
strongly
hysteresis
strongly
strongly
the rigid
depend
depend
body
on
on temperature
andassumption
depend backlash,
on temperature
temperature static
could
in
are an isolated
designed workspace
to provide by
high means
accuracy of walls
and or fences.
precision, They
but linear,
(Simoni
and
(Simoni aging
et
viscous
et al.,can
friction
al., increase
2017),
2017), and
terms
and hysteresis
the rigid
strongly
the rigid bodyand
depend
body backlash,
assumption
on temperature
assumption static
could
could
fact,
are
in industrial
areandesigned
isolated to
designed robots
provide
workspace
to provide arehigh
byusually
high employed
accuracy
means
accuracy and
of walls
andor inprecision,
several They
fences.
precision, tasks
but (Simoni
but not be anet accurate
al., 2017), and the
hypothesis rigid body assumption
for lightweight
lightweight links. could
are
the designed
achievement to provide
suchhigh
of such a high
highaccuracy
level and precision,
performance but
makes and
not not be
(Simoni viscous
be an
an friction
et accurate
al., 2017),
accurate terms
hypothesis
and the
hypothesis strongly for
rigid depend
body
for lightweight on temperature
assumption
lightweight links.
links. could
in
the
are an isolated
achievement
the designed
achievement workspace
of
to provide
of bysuch bya
high means of
level
accuracy walls or
performance
and fences.
precision, They
makes
but Thosenot
Those be an
modelaccurate hypothesis
mismatches can for
imply significant links.
detriment
the
them
are achievement
characterized
designed to of such
provide a aa high
rigid
high
high and level
level
accuracy heavy performance
performance
andstructure.
precision,
makes
makes
These
but
(Simoni
not
Those be anet accurate
model
model al.,mismatches
2017), and the
hypothesis
mismatches can
can rigid
imply
for
imply body assumption
significant
lightweight
significant detriment
links.
detriment could
them
the
them
features
characterized
themachievement
characterized
characterized
do not of by
allow such
by
by
aa rigid
a
robots a high
rigid
rigid to
and
and
and
reach
heavy
level
heavy
heavy
the
structure.
performance
structure.
structure.
safety These Those
These
makes
These
conditions
in
innot
Those be model
model-based
an
model-based
mismatches
accurate control
model mismatches hypothesis
control
can imply
strategies
strategies for
canitimply
whichsignificant
do not
lightweight
which do notlinks.
significant
detriment
take
take
detriment
them
them
the
featuresachievement
do not of by
allow such
robots a high
to level
reach performance
the safety makes
conditions in in model-based
into model-based
account. Incontrol
control
this strategies
strategies
case, is which
which do not
do
reasonable not to take
take them
them
employ
them
features
features
that characterized
are do
do not allow
not allow
necessary to a
robots
robots
sharerigid to
to and
reach
reach
a and
working heavy
the structure.
safety
theenvironment
safety These
conditions
conditions with into Those model
account. mismatches
In this case,can itimplyis significant
reasonable detriment
to employ
them
that characterized
are necessary by
to a
sharerigid a working heavy structure.
environment These
with in
into
a intomodel-based
account.
account.
robust In
In
PID-basedcontrol
this strategies
thiscontrol
case,
case, it iswhich
itarchitecture
is do not
reasonable
reasonable instead take
to
to them
employ
employ
of an
features
that are do not allow
are without
necessary torobots
sharetools to reach
aa working
working theenvironment
safetyskin.
environment conditions with ain model-based control strategies which do not take them
that
humans
features
humans
necessary
do not
without allow to
expensiveshare
robots
expensive to
tools like sensitive
reach
like sensitive
the safety conditions
skin.
with
a a robust
into account.
robust
robust
adaptive
PID-based
In
PID-based
PID-based
model-based thiscontrol
case,
control
control
algorithm. itarchitecture
isOne reasonable
architecture
architecture limit
instead
instead
instead
of thisto of
employ
of
of
solution
an
an
an
that
humans
humans
Due toare necessary
without
without
the increased to
expensive
expensiveshare
demand a
tools
tools working
of like
human-robot environment
like sensitive
sensitive skin. skin.
interaction, with into
adaptive account. In thiscontrol
model-based case, itarchitecture
algorithm. isOne reasonable
limit of thisto solution
employ
that
Due toare necessary to share a working environment with a
adaptiverobust
adaptive PID-based
model-based
model-based algorithm.
algorithm. One limit
Onead-hoc
limit of instead
of this
this solution of
solution an
humans
Due to the increased
without
the shift expensive
increased demand
demand tools of
of human-robot
like sensitivedesign
human-robot interaction,
skin.
interaction, is the computational
aadaptive computational burden of an robust tuning
Due
aahumans
to
paradigm
paradigm
the increased
without
shift
demand
is changing
changing
expensive
is tools of
the
the
human-robot
structural
like sensitivedesign
structural
interaction,
skin. of the is
of the is is robust
the
the
optimization,
PID-based burden
model-based
the computational
computational
which is
control
algorithm.
burden
burden
not
architecture
of
of
of
compatible
an
Onead-hoc
an
an ad-hoc
ad-hoc
with
instead
robust
limit modular
of this solution
robust
robust
of an
tuning
tuning
tuning
design
Due
a
arobot to
paradigm
paradigm
tothe increased
reduceshift
shift theis
is demand
changing
changing
impact of
the
the
forces human-robot
structural
structural
during possible interaction,
design
design of
of the
the
collision. adaptive
optimization, model-based
which is algorithm.
not compatible One limit
with of this
modular solution
design
Due toto is the computational
optimization, which is burden
not of
compatible an ad-hoc
with robust
modular tuning
design
robot
arobot
robot
In tothe
paradigm
to
recent
increased
reduce
reduce
reduceshiftthe
years, a is
the
the
demand
impact
changing
impact
impact
class
forces
forces
forces
of lightweight
of human-robot
the
lightweight
during
structural
during
during
possible
possible
possible
industrial
interaction,
design the optimization,
collision.
of has
collision.
collision.
robot
(Maurtua
is
(Maurtua et which
the computational
optimization, et al.,
is burden
al., 2016).
2016). not compatible
In fact,
In fact,
of an the thewith
ad-hoc modular
robot
robot robust presents
presents
design
tuning aaa
Ina paradigm
recent shift
years, a is changing
class of the structural
industrial design robot of the
has (Maurtua
(Maurtua
modular et which
et
design, al.,thus
al., is its
2016).
2016). notoverall
compatible
In fact,
In fact,
inertia the
thewith modular
robot
robot
may presents
presents
change design
based a
robot
In recentto
recent reduce the
years, (Guiochetimpact
aa class
class of forces
of lightweight
lightweight during possible
industrial collision.
robot has modular optimization,
In
been
robot
been
years,
presented
to reduce
presented the impact
(Guiochet et al., during
forces
et al., 2017)
2017)
industrial
where
possible
where the
the
robot
impacthas
collision.
impact (Maurtua
modular
modular
on the et which
design,
design,
design,
setup al.,
of
thus
thus
the
is its
2016).
thus notoverall
its
its
compatible
In fact,
overall
overall
mechanical
inertia
inertia
inertia thewith
may
may
may
structure.
modular
robot change
presents
change
change
Moreover,
design
based
based
basedin a
In
been
been
forcesrecent years,
presented
presented
areyears,
reduced a class
(Guiochet
(Guiochet
by of of lightweight
et al.,
et al., the
limiting 2017)
2017) industrial
where
where
mass, andthe therobot
impact
impact
thereforehas on (Maurtua
the setupet al.,
of 2016).
the In
mechanical fact, the
structure.robot presents
Moreover, in a
In
forces recentare reduced a class
by lightweight
limiting the industrial
mass, and robot
thereforehas on modular
on
many the
the design,
setup
setup of
of
applications, thus
the
the the its overall
mechanical
mechanical
usage of inertia
a may
structure.
structure.
single change
Moreover,
Moreover,
controller based
for in
in
the
been
forces presented
are reduced (Guiochet
by et
limiting al., 2017)
the where
mass, and the impact
therefore modular design, thus its overall inertia may change based
forces
the
been are reduced
momentum
presented of theby robot,
the
(Guiochet limiting
et al.,and the the
2017) mass,
maximum
where impact many
andthetherefore
exerted on
manymany theapplications,of thethe
setupworkspace
applications, usage
themechanical
usage of
of toaa structure.
single
single controller
Moreover,
controller for
for thein
the
the
thethe momentum
forces
forces. are
momentum
momentum reduced
However,
of
ofmany
of by robot,
the
the limiting
robot,
robot,
tasks
and
and
and
do the
not
the
the
the
maximum
mass,
maximum
maximum
require and
the
exerted
therefore
exerted
exerted
precision
entire
on
entire
many
entiretheapplications,
robot
robot of thethe
setupworkspace
applications,
robot workspace the
usage
can
can
usage
can
lead
mechanical
lead
lead
of
of
a structure.
to
toa
single
very poor
very
single
very
controller
poor performance,
Moreover,
performance,
controller
poor performance,
for
for
the
in
the
forces
forces.
the are
momentum
forces. reduced
However,
However, ofmany
many by limiting
tasks
the robot,
tasks do
and
do the
not
notthe mass,
require
maximum
require and
the therefore
the precisionexerted
precision entire
requiring robot theworkspace
retuning canof lead
theof to very poor
controllers. performance,
In thisforcase,
case,
forces.
provided However,
by an many
industrial tasks
robot doand not require
therefore the
impact precision
forces many
requiring
entire
requiring applications,
robot the
the retuning
workspace
retuningthe usage
canof the
of lead
the toa single
controllers.
very to
controllers. controller
poor In this
performance,
In this the
case,
the
providedmomentum by an ofmany
the robot,
industrial robot and
and notthe maximum
therefore impact exerted
forces requiring
forces the retuning of the controllers. beIn anthis case,
forces.
provided
provided
reduction
forces.
However,
by
by
cananbe
an
However,
industrial
industrial
obtained
many
tasks by do
robot
robot
tasks do
and
and
introducing
not
require
therefore
therefore
require
the precision
impact
impact
elastic
the forces
elements
precision aaentire
gain-scheduling
requiring robot
a gain-scheduling
the
gain-scheduling workspace approach
approach
retuning
approachcanof lead
is proven
is
the
proven
to very to
controllers.
is proven
proven poor
to be
beInperformance,
an
anthis
effective
effective
case,
effective
reduction
provided
reduction can
by
canan be obtained
industrial
be obtained by
robot introducing
andal.,
by introducing
introducingtherefore elastic
impact
elastic elements a
forces solution.
elements gain-scheduling
solution.
requiring Also in
the retuning approach
this case,
of the is
effective
controllers. to
robust be In an
tuning
this effective
rules
case,
reduction
in the robot
provided robotcan
by an be
chain obtained
(Verstraten
industrial by
robot et
andal., 2016).
therefore elastic
Series
impact elements
Elastic
forces solution.asolution. Also
gain-scheduling
Also in
in this
approach
this case,
case, effective
is proven
effective robust
to be
robust tuning
an
tuning rules
effective
rules
in the
reduction
in the robotcan chain
be
chain (Verstraten
obtained
(Verstraten by et
introducing
et al., 2016).
2016). Series
elastic
Series Elastic
elements
Elastic can
a be used
usedAlso
gain-scheduling to in
tunethis thecase,
approach effective
controller
is proven robust
withoutto the
be tuning
anneed rules
of
effectivean
in the
Actuators
reduction robotcan chain
(SEAs) (Verstraten
are explicitly
be obtained explicitly et al.,
by introducing 2016).
designedelastic Series
to pursue
pursue Elastic
elementsthis can can be
solution.
can be be used
usedAlso to tune
to in
tunethis the controller
thecase, effective
controller without
robust
without the
the need
tuning
need of of an
rules
of an
Actuators
in the
Actuators robot (SEAs)
chain
(SEAs) are
(Verstraten
are explicitly et designed
al.,
designed2016). to
to Series
pursue this
Elastic
this external
solution. Also to
dedicated tune
in this the
computer controller
case, (Misgeld
effective without
et
robust the
al., need
2017).
tuning an
rules
Actuators
idea.
in
idea. the Inrobot
In
(SEAs)
fact,
fact, an elastic
chain
an
are explicitly
elastic elementetis
(Verstraten
element isdesigned
inserted
al., 2016).
inserted
toSeries
in
in
pursue
series
series the external
to this
Elastic
to the can be
external
external
dedicated
used to
dedicated
dedicated tune computer
the
computer
computer
(Misgeld
controller
(Misgeld
(Misgeld
et
without
et
et
al.,
the
al.,
al.,
2017).
need
2017).
2017). of an
Actuators
idea. In
idea. In fact,(SEAs)
fact, an
an are explicitly
elastic
elastic element
element isofdesigned
inserted tothat
in pursue
series to this
the The The
can approximation
be used of
to tunecomputer
approximation oftherobust problem
controller
robust problem solutions
without the
solutions withoftun-
need
with tun-
an
actuator
Actuators
actuator thus
(SEAs)
thus reducing
reducing the part
are explicitly
the part is inserted
the mass
designed
of the mass in series
tothat pursue to
directlythe
this
directly Theexternal
Therules dedicated
approximation
approximation of
of robust
robust (Misgeld
problem
problem et al.,
solutions
solutions 2017).with
with tun-
tun-
idea.
actuator
actuator In fact,
thus
thus an elastic
reducing
reducing element
the
the part isof
part isofthe inserted
the mass
theoverall
mass in series
that
that to
directlythe ing
ingexternal
rules is a common
dedicated
is a common strategy
computer
strategy in
(Misgeld
in process
process et control
al.,
control2017). (( Åström
Åström
impacts
idea.
impacts
actuator
impacts
the
In fact,
the
thus
the
human,
an
human,
reducing
human,
andelement
elastic
and
andthe
therefore
therefore
part ofthe
therefore inserted
thetheoverall
mass
overall seriesdirectly
impact
inimpact
that
impact
force.
to the ing
force.
directly
force.
The
and approximation
ing rules
rules is aa common
is
Hägglund, common
2004),of strategy
robust
strategy
but there problem
there in are
in nosolutions
process
process controlrules
control
tuning with tun-
((Åström
Åström
that
impacts
Moreover, the human,
springs and
allow therefore
estimating the overall
the impact
exerted force.
forces andThe
ing
and approximation
Hägglund,
rules is
Hägglund,a 2004),
common
2004),of strategy
robust
but
but problem
there in are no
process
are no solutions
tuning
control
tuning with
rules
rules( tun-
that
Åström
that
actuator
Moreover,
impacts
Moreover, thus
the reducing
springs
human,
springs allow the
and therefore
allow part
estimating
estimating of the mass
the
the overall that
exerted directly
impact forces
the sensors.
exerted force. and
forces address Hägglund,
the control2004), but there
of elastic
elastic mechanicalare no tuning
systems, rules as that
far as
Moreover,
without the springs
use of allow
expensive estimating
force/torque the exerted forces ing
address
and rules
address is
the
Hägglund,a common
control
the control
control2004),of strategy
but
of elastic
elastic there in
mechanical
mechanical process
arepaper control
systems,
no tuning
systems, rules( Åström
as far
as far as
that
far as
impacts
without
Moreover,
without the
the human,
use
springs
use of
of and therefore
expensive
allow
expensive estimating the overall
force/torque
force/torque the impact forces
sensors.
exerted
sensors. force. address
the the
authors know. of
The aim mechanical
of this systems,
is to cope as as
with
without
However, the
thisusetypeof expensive
of structure
structure force/torque
makesthe thesensors.
control of of the the and
address Hägglund,
the authors the
authors know.
control
know.2004),
The but
aim
of elastic
The there
aim of this
mechanical
of are
this paper no
paper tuning
is
is to
systems,
to rules
cope
cope that
with
aselastic
far as
with
Moreover,
However,
without
However, the springs
this
thisusetype
type allow
of
of expensive
of structureestimating makes
force/torque
makesetthe the
the exerted
control
sensors.
control forces
of the
the thisthe
this authors
issue by know.
proposing The aim
robust of this
tuning paper
rules is
for to cope
series with
However,
link position this type
more of structure
difficult (Malekimakes al., control
2016; of Painethe address
the issuethe
issue
this authors by control
proposing of elastic
robust mechanical
tuning rules systems,
for series aselastic
far as
without
link
However,
link
link
the
position
position
position thisusemore
type
more
more
of expensive
difficult
of structure
difficult
difficult
force/torque
(Maleki
(Maleki
(Malekimakes et
et
etthe sensors.
al.,
al.,
al.,
2016;
control
2016;
2016;
Paine
of
Paine
Painethe this
actuators.
the
actuators.
issue
authors Inknow.
by
by
In
proposing
proposing
this way,
know.
this
The
way,
The itaim
robust
robust
itaim
is
of
of
this paper
tuning
tuning
is possible
possible
this
rules
rules
to
paper
to
is to
for
for
guarantee
is
guarantee to
cope
series
series
cope a
with
elastic
elastic
a robust
robust
with
et al., 2014;
However, 2014; Ragonesi
thisRagonesi
type of et al.,
al., 2011;
structure 2011;makes Calanca and Fiorini,
Fiorini, this issue by
actuators. Inproposing way, robust
thislimiting it the tuningto
is possible
possible rules
to for series
guarantee elastic
aa in
robust
et
et et al.,
link
2017; al.,position
al., 2014;
2014;
Calanca more
Ragonesi
et al., 2017;
et
al., difficult
Ragonesi et Dos(Maleki
et al.,
al., 2011;
2011; etthe
Calanca
Calanca
SantosCalancaand al., control
and
2016;Fiorini,
and
and
Siqueira,
of
Painethe actuators.
Fiorini,
2014).
performance
this
performanceissue by In this
by way,
proposing
by limiting
it is
robust
the maximum
tuningto
maximum
guarantee
rulessensitivity
for series
sensitivity
robust
link
elastic
in link
link
2017; position
Calanca more
et difficult
2017; Dos (Maleki
Santos et
and al., 2016;
Siqueira, Paine
2014). actuators.
performance
performance
position andIn this
by
by
motor way,
limiting
limiting it
velocity is
the
the possible
maximum
maximum
controllers guarantee
sensitivity
sensitivity
for different a robust
in
in link
link
values
et al.,
2017; 2014;
Calanca Ragonesi
et al., 2017;et al.,
Dos 2011;
Santos Calanca
and and
Siqueira, Fiorini,
2014). actuators.
the position andIn this way, it is possible to guarantee a robust
2017;
et al.,Calanca
Indeed, 2014; et al., position
the actual
actual
Ragonesi 2017; Dosof2011;
et al., Santos and isSiqueira,
the Calanca
link given
and Fiorini, 2014).
by performance
position and motor
byinertia
motor velocity
limiting
velocity controllers
thedifferent
maximum
controllers for different
sensitivity
for levels.in
different values
link
values
Indeed,
2017;
Indeed,
Indeed,
position
the
Calanca
the
the
of the et
actual
actual al., position
gearbox 2017;
position
position
outputDosof of
the
ofSantos
the plus
the
shaft
link
andthe
link
link
is
is
given
isSiqueira,
given
given
spring
by
by the position
the
2014).
by the
defor-
of
of performance
the
position
and
the reflected
reflected
and
motor
byinertia
motor
velocity
limiting and
and
velocity
controllers
thedifferent
maximum
controllers
for
damping different
sensitivity
damping
for levels.
different
values
in link
values
2017;
position Calanca
of the et al., position
gearbox 2017;outputDosofSantos
shaft andthe
plus Siqueira,
spring 2014).
defor- of
Theof the
the reflected
reflected
paper inertia
inertia
is organized
organized and
and different
different
as follows:
follows: damping
damping
in Section
Section levels.
2 levels.
is described
described
Indeed,
position
position
mation. the
of
of
The the
the actual
gearbox
gearbox
elasticity output
output
impliesofthe the
shaft
shaft
the link
plus
plus
presence is
the
the given
spring
spring
ofgiven by the
defor-
defor- Thepositionpaper and is motor velocity
as controllers in for different
2 is values
Indeed,
mation. the
The actual
elasticity position
implies the link the
presence is of aaa resonance
resonance
by the The of
theThe the reflected
paper
paper
dynamic is inertia
is organized
organized
model of and
as
as different
follows:
thefollows:
the SEA.
SEA. in damping
in Section
Section
In Section 22 levels.
Section is described
is described
position
mation.
mation.
and of the
The
The
antiresonance
position of the
gearbox
elasticity
elasticity
gearbox
outputin
implies
implies
behavior
output
shaft
the
the the
shaft
plus
presence
presence
frequency
plus the
ofspring
of a resonance
spring
defor-
resonance
response,
defor- theof
The
the the reflected
dynamic
paper
dynamic is inertia
model
organized
model of
of and
the
as
the different
the
follows:
the SEA. in damping
In
Section
In Section2 is 333 is
levels. is pro-
described
pro-
is pro-
pro-
and
mation.
and antiresonance
The
antiresonance elasticity behavior
implies
behavior in
the
in the
the frequency
presence
frequency of a response,
resonance
response, the
posed
The dynamic
the
paper is model
optimization
organized of the
as the
problem
follows: SEA.
performed
in In
SectionSection
to2 is 3 is
obtain
described spe-
and
coupled
mation.
coupled
antiresonance
with elasticity
The
with the typical
the
behavior
typical smallin
implies
small the the
damping
damping
frequency
presence factor
factorof a of metallic posed
response,
ofresonance
metallic the
posed
posed
the
the optimization
dynamic
the modelable
optimization
optimization of theproblem
the SEA.
problem
problem
performed
performed
performedIn Sectionto obtain
3 is pro-
toperformance.
to obtain
obtain
spe-
spe-
spe-
and
coupled antiresonance
coupled with the
with the Moreover, behavior
typical
typical small
small in the
damping
damping frequency
factor
factor of of response,
metallic
metallic cific
the
cific tuning
dynamic
tuning rules,
model
rules, ableof to
the
to guarantee
the SEA.
guarantee robust
In Section
robust 3
performance. is pro-
spring
and elements.
antiresonance behavior there
in theare multiple
frequency sources
response, of posed
cific the
tuning optimization
rules, able problem
to guarantee performedrobust to obtain
performance. spe-
spring
spring elements.
coupled
spring with the Moreover,
elements.
elements. typical small
Moreover, there
theredampingare
are multiple
factor of
multiple sources
metallic
sources of cific
of posed tuning rules, able to
the optimization guarantee
problem performedrobusttoperformance.
obtain spe-
coupled
spring with the Moreover,
elements. typical small
Moreover,
there
theredampingare
are
multiple
factor of
multiple
sources
metallic
sources
of
of
cific tuning rules, able to guarantee robust performance.
2405-8963 © 2018, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved. robust performance.
cific tuning rules, able to guarantee
Copyright
spring © 2018 IFAC
elements. Moreover,of there are multiple sources of220
Copyright
Peer review©under
Copyright © 2018 responsibility
IFAC International Federation of Automatic 220Control.
Copyright © 2018
2018 IFAC
IFAC
10.1016/j.ifacol.2018.06.069
220
220
Copyright © 2018 IFAC 220
Copyright © 2018 IFAC 220
 IFAC PID 2018
Ghent, Belgium, May 9-11, 2018 Stefano Ghidini et al. / IFAC PapersOnLine 51-4 (2018) 220–225 221

.
qt

Motion
planner + .
qt + Link position + Θt Motor velocity τ SEA actuator
control control
− −
q .
Θ

Fig. 2. Cascade control scheme.

A robust cascade controller is therefore used to cope with
Fig. 1. Illustrative scheme of the SEA model.
mismatches without complex model adaptivity algorithms.
The experimental results are described in Section 4 , they The primary loop regulates the link position by acting on
have been obtained by means of two trajectories on a three the motor velocity set-point θ̇sp , that is the input of the
degree-of-freedom series elastic actuated robot. secondary loop, which computes the desired torque. The
2. SERIES ELASTIC ACTUATOR MODEL current loop is considered fast enough to be neglected.
Figure 2 shows the overall scheme.
A series elastic actuator manipulator can be modeled
2.1 Reduction of model parameters
as an open kinematic chain consisting of a rigid link
and lumped elastic part coaxial with the joint rotational In order to simplify the robust tuning rules expression, a
axis. The proposed approach considers a decentralized model similarity is used to reduce the number of model
model, where the coupling effects are considered as model parameters.
mismatches. This model takes into account two inertias: Rewriting (1) in the Laplace domain it is possible to
the first one represents the motor and the gearbox, while obtain:
the second one represents the inertia of the link and of Jm s2 θ = hsδ + kδ − fm sθ + τ (4)
the downstream part of the kinematic chain for the actual 2
Jl s (θ + δ) = −hsδ − kδ (5)
joint configuration. Thus, the state variables are the motor
Then, by defining the following quantities
position θ, the motor velocity θ̇, the spring deflection δ and  
the spring velocity δ̇. Other interesting variables are the k Jm
s= ŝ ⇔ ŝ = s,
link position q = θ + δ and its velocity q̇ = θ̇ + δ̇, which Jm k
J l
represent the real motion of the link. A joint model can be θ̂ = kθ δ̂ = kδ, Jˆl = ,
derived by dynamic laws as: Jm (6)
  
Jm θ̈ = hδ̇ + kδ − fm θ̇ + τ − Cm sign θ̇ fm k h k
(1) fˆm = , ĥ =
Jl (θ̈ + δ̈) = −hδ̇ − kδ k Jm k Jm
where Jm is the motor inertia, Jl is the link (and the
downstream part of the kinematic chain) inertia, τ is the it is possible to rewrite (4) as
motor torque, fm is the motor viscous friction (between ŝ2 θ̂ = ĥŝδ̂ + δ̂ − fˆm ŝθ̂ + τ (7)
motor and stator), Cm is the motor Coulomb friction
(between motor and stator), k is the elasticity (between and (5) as
link and motor), and h is the viscosity (between link and Jˆl ŝ2 (θ̂ + δ̂) = −ĥŝδ̂ − δ̂ (8)
motor). An illustrative scheme of the model is shown in The ratio Jˆl is the most important parameter, in fact,
Figure 1. it causes a decrement of the resonance and antiresonance
Equation (1) can be linearized by neglecting the Coulomb frequencies as shown in Figure 3 and, at the same time, it
friction terms, so that the resulting transfer function increases their peaks. The effect of ĥ, shown in Figure 4, is
between the motor torque and the motor velocity is: mainly on resonance and antiresonance peaks, while their
θ̇(s) frequencies present a slight shift. Finally, the motor viscous
Pθ̇,τ (s) = τ (s) =
(2) friction parameter changes the low frequency behavior and
Jl s2 +hs+k
Jl Jm s3 +(Jl fm +(Jm +Jl )h)s2 +((Jl +Jm )k+fm h)s+fm k it reduces the resonance peak.
while the transfer function between the motor torque and It is worth stressing that ĥ has normally a small value
the link position is: for mechanical elastic elements (like torsional, coned-disc
q(s) and coil springs) without dampers, while fˆm has not
Pq,τ (s) = = a significant impact on the crossover frequency range.
τ (s) (3)
hs+k
Finally, it is important to note that the motor velocity
(Jl Jm s3 +(Jl fm +(Jm +Jl )h)s2 +((Jl +Jm )k+fm h)s+fm k)s frequency response is less sensible to parameter changes
It is worth stressing that the link inertia Jl strongly for frequencies higher than the resonance frequency.
depends on the robot configuration and on the (possi- 3. ROBUST OPTIMAL TUNING
bly unknown) payload, while several possible model mis-
matches can occur in real setup. For example, the spring The tuning procedure starts by considering the inner loop
characteristic can be nonlinear, and aging can increase and continues with the outer one. The process of the inner
hysteresis and backlash. Finally, static and viscous fric- loop has a transfer function defined between τ and θ̇. Once
tion terms might strongly depend on temperature (Simoni the tuning parameters of this loop have been detected, the
et al., 2017). Those model mismatches can imply signifi- velocity closed loop obtained will be seen as the process
cant detriment in model-based control strategies which do of the outer loop, which has a transfer function defined
not take them into account. between θ̇ and q.

221
IFAC PID 2018
222
Ghent, Belgium, May 9-11, 2018 Stefano Ghidini et al. / IFAC PapersOnLine 51-4 (2018) 220–225

40 40

30 30

20 20

10 10

0 0

-10 -10

-20 -20

-30 -30

-40 -40
10 -2 10 -1 10 0 10 1 10 2 10 -2 10 -1 10 0 10 1 10 2
Normalized angular frequency [rad/s] Normalized angular frequency [rad/s]
100 100

50 50

0 0

-50 -50

-100 -100

-150 -150
10 -2 10 -1 10 0 10 1 10 2 10 -2 10 -1 10 0 10 1 10 2
Normalized angular frequency [rad/s] Normalized angular frequency [rad/s]

Fig. 3. Effect of the parameter Jˆl on frequency responses Fig. 5. Effect of the parameter fˆm on frequency responses
Pθ̂,τ
˙ (s) (upper plot), Pq̂,τ (s) (lower plot). Blue line: Pθ̂,τ
˙ (s) (upper plot), Pq̂,τ (s) (lower plot). Blue line:

Jˆl = 0.5, ĥ = 0.2, fˆm = 0. Magenta line: Jˆl = 1, ĥ = Jˆl = 1, ĥ = 0.3, fˆm = 0. Magenta line: Jˆl = 1, ĥ =
0.2, fˆm = 0. Ocher line: Jˆl = 2, ĥ = 0.2, fˆm = 0. 0.3, fˆm = 0.1. Ocher line: Jˆl = 1, ĥ = 0.3, fˆm = 0.2.
Violet line: Jˆl = 4, ĥ = 0.2, fˆm = 0. Violet line: Jˆl = 1, ĥ = 0.3, fˆm = 0.3.


40 k/Jm k
30
K̂p,v = Kp,v , Kp,v = K̂p,v  ,
k k/Jm (10)
20
1
10
K̂i,v = Ki,v , Ki,v = K̂i,v k,
0 k
-10
The controller parameters are then computed by solving
-20
the following optimization problem:
-30
2
-40 min(ω̂c,v − ω̂d,v ) − λv φm
10 -2 10 -1 10 0
Normalized angular frequency [rad/s]
10 1 10 2

 φ ≥ φm,min
100
 m (11)
Ms ≤ Ms,max
subject to:
 |Lv (ω̂)| ≤ Lh,v,max , if ω̂ ≥ ω̂h
50

0 |Lv (ω̂)| ≥ Ll,v,min , if ω̂ ≤ ω̂l
-50 where ω̂ is the normalized angular frequency, ω̂c,v and
ω̂d,v are, respectively, the cutoff frequency and its desired
-100
value, λv > 0 is a weighting factor, φm and Ms are
-150
10 -2 10 -1 10 0 10 1 10 2
the phase margin and the maximum sensitivity, Lv (ω̂) =
Normalized angular frequency [rad/s] Cv (ω̂)Pθ̂,τ
˙ (ω̂) is the loop transfer function, Lh,v,max is the

Fig. 4. Effect of the parameter ĥ on frequency responses
Pθ̂,τ
˙ (s) (upper plot), Pq̂,τ (s) (lower plot). Blue line:
Jl = 1, ĥ = 0.1, fˆm = 0. Magenta line: Jˆl = 1, ĥ =
ˆ values has been set equal to:
0.2, fˆm = 0. Ocher line: Jˆl = 1, ĥ = 0.3, fˆm = 0.
Violet line: Jˆl = 1, ĥ = 0.4, fˆm = 0.
3.1 Motor velocity control
A PI controller regulates the motor torque to reduce the
speed trajectory error. The controller is described by the
following transfer function:
Ki,v
Cv (s) = Kp,v + (9)
s utmost importance to guarantee the required level of ro-
where Kp,v is the proportional gain and Ki,v is the
integral gain. The controller parameters can be rewritten
by following (6) as
maximum allowed value of |L(ω̂)| for frequencies greater
than ω̂h , and Ll,v,min is the minimum allowed value for
frequencies lower than ω̂l . In the tuning procedure, those
Lh,v,max = −40dB, ω̂h = 100ω̂d ,
Ll,v,min = 40dB, ω̂h = 0.01ω̂d ,
Ms,max = 1.4, φm,min = 60◦ (12)
−2
λv = 10
It is worth stressing that the presence of the antiresonance
zeros can cause multiple crossing of the 0dB axis, and
therefore the system has more than one phase margin. In
this case, the most critical value is taken into account in
the optimization algorithm. Due to these multiple cutoff
frequencies, the maximum sensitivity Ms is therefore of
bustness. The term −λφm of the cost function discerns
between solutions with the same cutoff frequency by se-
lecting the one with the biggest phase margin.

222
 IFAC PID 2018
Ghent, Belgium, May 9-11, 2018 Stefano Ghidini et al. / IFAC PapersOnLine 51-4 (2018) 220–225 223

3.2 Link position control Table 1. Matrices coeffients of the motor veloc-
ity tuning rules when ω̂d,v = 3
A PD controller addresses the link position control by
computing the target motor speed signal θ̇t . The controller MK̂p 2.531878 -0.187629 0.766849 0.176141 2.369994
is described by the following transfer function: 0.065366 0.590771 -1.589730 0.125980 -2.663157
0.000677 -0.295008 0.763627 -0.124333 0.946668
Kd,p s MK̂ 0.001504 3.001502 -0.006152 -0.000658 -0.009293
Cp (s) = Kp,p + (13) i
Tf,p s + 1 0.051657 -0.256998 0.367284 -0.000003 0.009802
0.010598 -0.028327 0.025340 0.000178 -0.003194
where Kp,p is the proportional gain, Kd,p is the derivative
gain, and Tf,p is the filter time constant. Controller Cp (s)
Table 2. Matrices coeffients of the motor veloc-
acts on the process made by the secondary loop Pθ̇t ,q (ω̂)
ity tuning rules when ω̂d,v = 4
as shown in Figure 2. The target motor speed signal θ̇t is
given by the controller output plus a feedforward action MK̂p 3.708948 -0.108430 0.377811 0.037026 1.437171
equal to the velocity profile provided by the link motion 0.011955 0.260642 -0.653636 0.130305 -1.496274
0.003181 -0.116833 0.292964 -0.071785 0.498541
planner. Integral action is not employed since Pθ̇t ,q (ω̂) has
MK̂ -0.003161 4.029311 -0.046981 -0.000382 -0.009993
a pole at the origin and external disturbances are managed i
0.127712 -0.628493 0.890955 -0.000331 0.009936
by the secondary loop. 0.015017 -0.008518 -0.026994 0.000229 -0.003068
The controller parameters can be rewritten by following
(6) as Table 3. Matrices coeffients of the motor veloc-
  ity tuning rules when ω̂d,v = 5
Jm k
K̂p,p = Kp,p , Kp,p = K̂p,v ,
k Jm MK̂p 4.791588 -0.113563 0.302272 0.032533 0.976789
K̂d,p = Kd,p, Kd,p = K̂d,p, (14) -0.003609 0.189006 -0.436630 0.068670 -0.936631
0.003717 -0.074892 0.180993 -0.036597 0.298819
k Jm MK̂ -0.009775 5.061989 -0.088155 -0.000461 -0.010757
T̂f,p = Tf,p , Tf,v = T̂f,p , i
Jm k 0.249213 -1.201760 1.679001 -0.000247 0.010181
0.014515 0.064108 -0.166778 0.000201 -0.003044
As in Section 3.1, the controller parameters are computed
by solving the following optimization problem:
Table 4. Matrices coeffients of the motor veloc-
2
min(ω̂c,p − ω̂d,p ) − λp φm ity tuning rules when ω̂d,v = 6

 φ ≥ φm,min
 m (15) MK̂p 5.850311 -0.174389 0.379250 0.000514 0.796468
Ms ≤ Ms,max
subject to: -0.025962 0.249436 -0.541724 0.102136 -0.787528

 |L p (ω̂)| ≤ L h,p,max , if ω̂ ≥ ω̂h 0.010387 -0.094186 0.217908 -0.049836 0.261223
|Lp (ω̂)| ≥ Ll,p,min , if ω̂ ≤ ω̂l MK̂
i
-0.015671 6.076707 -0.089640 -0.000315 -0.012242
0.418010 -1.954373 2.674216 -0.000646 0.011784
where ω̂c,v and ω̂d,v are, respectively, the cutoff frequency
0.007221 0.196555 -0.397502 0.000421 -0.003593
and its desired value, Lp (ω̂) = Cv (ω̂)Pθ̂˙ ,q (ω̂)) is the loop
t
transfer function, Lh,p,max is the maximum allowed value Table 5. Matrices coeffients of the motor veloc-
of |L(ω̂)| for frequencies greater than ω̂h , while Ll,p,min is ity tuning rules when ω̂d,v = 7
the minimum allowed value for frequencies lower than ω̂l
and λv > 0 is a weighting factor. In the tuning procedure, MK̂p 6.897957 -0.263959 0.537100 -0.048779 0.747911
those values have been set equal to: -0.054282 0.386998 -0.835283 0.186529 -0.833085
0.020995 -0.150384 0.344335 -0.088107 0.301129
Lh,p,max = 40dB, ω̂h = 100ω̂d , MK̂ -0.017312 7.046733 -0.007167 0.000191 -0.014580
i
Ll,p,min = 20dB, ω̂h = 0.01ω̂d , 0.631822 -2.839686 3.784653 -0.001803 0.015161
Ms,max = 1.4, φm,min = 60◦ (16) -0.007112 0.383480 -0.702280 0.001026 -0.004937
−2
λp = 10  
 1

3.3 Tuning rules K̂p = 1 fˆm fˆm
2
ĥ ĥ MK̂p  Jˆl 
2
ˆ2
During the robot working activities, the link inertia can  Jl  (17)
significantly change, and therefore a gain scheduling ap-   1
proach is highly recommended. However, solving (11) and K̂i = 1 fˆm fˆm
2
ĥ ĥ2 MK̂i  Jˆl 
(15) online is not possible due to computational burdens. Jˆl2
For this reason, the development of an approximate solu- The resulting values for each ω̂d,v are in Tables 1-8. For
tion by means of tuning rules has been performed, in order all the values of ω̂d,v the maximum sensitivity is much
to allow the online computation of the robust controller for smaller than 1.4, as shown in Figure 6, which demonstrates
a given value of Jl . The motor velocity controller tuning is the effectiveness of the tuning rules to ensure the required
performed by imposing the desired cutoff frequency to be robustness.
part of the set ω̂d,v ∈ {3, 4, 5, 6, 7, 8, 9, 10}. In order to find The link position controller tuning rules are also computed
the tuning rules, the optimization problem (11) has been by solving 2000 times the optimization problem (15) by
repeated 2000 times by varying ĥ in the interval [0.01, 0.5], using the motor velocity tuning rules with ω̂d,v = 3, even
fˆm in the interval [0.0, 0.5], and Jˆm in the interval [0.5, 2] if it can be used also with faster secondary controlled
for each value of cutoff frequencies. systems. The primary loop cutoff frequency is set to
The resulting controllers parameters are approximated by ω̂d,p = 1 but, differently to the secondary loop case, the
means of the following tuning rules: constraint Ms = 1.4 limits the bandwidth in the range [0.3,

223
IFAC PID 2018
224
Ghent, Belgium, May 9-11, 2018 Stefano Ghidini et al. / IFAC PapersOnLine 51-4 (2018) 220–225

Table 6. Matrices coeffients of the motor veloc- Table 9. Matrices coeffients of the link position
ity tuning rules when ω̂d,v = 8 tuning rules when ω̂d,p = 1
MK̂p 7.946372 -0.410546 0.836845 -0.142072 0.832658 MK̂p 0.063902 0.152822 -0.281618 -0.057527 2.330662
-0.101432 0.656619 -1.444846 0.373792 -1.122394 0.035825 -0.212446 0.385927 0.071206 -2.876210
0.040938 -0.269486 0.621562 -0.175203 0.448979 -0.019325 0.069888 -0.132506 0.059622 0.735524
MK̂ -0.011250 7.948294 0.195722 0.001579 -0.018751 MK̂ 0.026688 -0.275392 0.682902 -0.568180 2.708334
i d
0.886549 -3.804287 4.914911 -0.004912 0.022654 -0.042239 0.780534 -1.706131 0.081026 -1.507766
-0.027762 0.613100 -1.055465 0.002601 -0.008212 0.026225 -0.414427 0.850398 0.176679 0.063388

Table 7. Matrices coeffients of the motor veloc-

ity tuning rules when ω̂d,v = 9
MK̂p 8.996517 -0.605634 1.256875 -0.259264 0.988444
-0.163220 1.035979 -2.316475 0.619581 -1.544295
0.068099 -0.440535 1.022770 -0.291915 0.659486
MK̂ 0.005449 8.763461 0.543840 0.003918 -0.024746
i
1.177551 -4.797556 5.980280 -0.010142 0.034360
-0.053453 0.871060 -1.428947 0.005280 -0.013569

Table 8. Matrices coeffients of the motor veloc-

ity tuning rules when ω̂d,v = 10
MK̂p 10.058496 -0.854067 1.833301 -0.489032 1.374578
-0.257929 1.543944 -3.549797 1.106527 -2.438658
0.111149 -0.676386 1.604185 -0.522677 1.091335
MK̂ 0.034812 9.481141 1.048431 0.009575 -0.037039
i
1.500984 -5.778026 6.918657 -0.022352 0.059535
-0.082965 1.143977 -1.799310 0.011330 -0.025308 Fig. 7. Nyquist plot of the link position open loop transfer
function (ω̂d,v = 10 and ω̂d,p = 1). The dashed circle
indicates the value Ms = 1.4.

Fig. 8. FourByThree Series Elastic Actuated robot

Parameter Value Reduced value
Jm 0.883024 [kgm2 ] [-]
k 735.343764 [Nm/rad] [-]
h 7.8928 [Nm/(rad/s)] 0.301
fm 6.966257 [Nm/(rad/s)] 0.27
Jl (no payload) 0.650 [kgm2 ] 0.736
Table 10. Dynamics parameters of Joint 1

4. EXPERIMENTAL RESULTS
The proposed tuning rules are tested on a three degree-of-
freedom series elastic actuated robot, designed as a part of
the FourByThree European project (FourByThree, 2018).
Fig. 6. Nyquist plot of the motor velocity open loop The actuators are composed by a brushless DC motor,
transfer function (ω̂d,v = 10). The dashed circle an harmonic driver and an elastic element made by a
indicates the value Ms = 1.4. series of coil springs. The actuator has two encoders: one
measures the motor position after the harmonic drive while
0.6] depending on the values of ĥ, fˆm , and Jˆl . Thus, the the second measures the angle after the spring element.
proposed tuning rules provide the fastest primary loop. The current loop is implemented inside a FPGA board
The tuning rules are expressed by means of the following and it is considered as a unitary gain for the position
equation: and velocity controllers. The torque reference signal is sent
  to the FPGA board by using the ROS operating system
  1
K̂p = 1 fˆm fˆm 2
ĥ ĥ2 MK̂p  Jˆl  framework (ROS, 2018).
ˆ2 Joint 1 has been used to show the effectiveness of the
 Jl  proposed method during a trajectory tracking task. The
  1 (18) dynamics parameters of the robot are in Table 10.
K̂d = 1 fˆm fˆm 2
ĥ ĥ2 MK̂d  Jˆl  Moreover, there is a static friction of Cm = 2.3 [Nm]
Jˆl2 and a backlash of 0.015 [rad] (around 1 [deg]) which are
1 considered as model mismatches, as well as the coupling
T̂f =
5 effect caused by the other links.
The resulting values are in Table 9. The control performance are tested by means of two
Figure 7 shows that the maximum sensitivity is limited to trajectories. In particular, a trapezoidal motion profile
1.4 demonstrating the effectiveness of the tuning rules to from position 0 [rad] to 1.5708 [rad] followed by another
ensure the required robustness. trapezoidal motion profile from 1.5708 [rad] to 0 [rad]

224
 IFAC PID 2018
Ghent, Belgium, May 9-11, 2018 Stefano Ghidini et al. / IFAC PapersOnLine 51-4 (2018) 220–225 225

Parameter Value Reduced value REFERENCES

Jl (no payload) 0.650 [kgm2 ] 0.736
Jl (m= 0.5 [kg]) 0.740 [kgm2 ] 0.8385 Åström, K.J. and Hägglund, T. (2004). Revisiting the
Jl (m= 1 [kg]) 0.786 [kgm2 ] 0.890 Ziegler-Nichols step response method for PID con-
Jl (m= 2 [kg]) 0.924 [kgm2 ] 1.047 trol. Journal of Process Control, 14(6), 635–650. doi:
Jl (m= 5 [kg]) 1.514 [kgm2 ] 1.714 10.1016/j.jprocont.2004.01.002.
Table 11. Link inertia values with disk pay- Calanca, A. and Fiorini, P. (2017). Impedance control
loads. of series elastic actuators based on well-defined force
dynamics. Robotics and Autonomous Systems, 96, 81–
Payload |ē|p |ē|v ce 92. doi:10.1016/j.robot.2017.06.013.
Jl (no payload) 0.0014 [rad] 0.0548 [rad/s] 0.3055 [Nm] Calanca, A., Muradore, R., and Fiorini, P. (2017).
Jl (m= 0.5 [kg]) 0.0019 [rad] 0.0471 [rad/s] 0.3073 [Nm] Impedance control of series elastic actuators: Passivity
Jl (m= 1 [kg]) 0.0018 [rad] 0.0720 [rad/s] 0.4243 [Nm] and acceleration-based control. Mechatronics, 47, 37–48.
Jl (m= 2 [kg]) 0.0012 [rad] 0.0498 [rad/s] 0.3466 [Nm] doi:10.1016/j.mechatronics.2017.08.010.
Jl (m= 5 [kg]) 0.0020 [rad] 0.0648 [rad/s] 0.3902 [Nm] Dos Santos, W.M. and Siqueira, A.A.G. (2014).
Table 12. performance indexes for the different Impedance control of a rotary series elastic actua-
Jl values. tor for knee rehabilitation, volume 19. IFAC. doi:
1.6
10.1109/ICAR.2013.6766567.
1.4

1.2
FourByThree (2018). Fourbythree project homepage.
Link position [rad]

http://http://fourbythree.eu/. Accessed: 2018-01-

1

0.8

0.6

0.4

0.2

0
31.
-0.2
0 0.5 1 1.5 2 2.5
Time [s]
3 3.5 4 4.5 5
Guiochet, J., Machin, M., and Waeselynck, H.
0.015 (2017). Safety-critical advanced robots: A survey.
0.005
0.01

Robotics and Autonomous Systems, 94, 43–52. doi:

Link error [rad]

-0.005
0

https://doi.org/10.1016/j.robot.2017.04.004.
-0.015
-0.01
Maleki, S., Parsa, A., and Ahmadabadi, M.N. (2016).
0 0.5 1 1.5 2 2.5
Time [s]
3 3.5 4 4.5 5

Feed-forward learning with frequency adaptation to-

wards the control of series elastic actuators. In 2016 4th
Fig. 9. Link position and error trends with different Jl International Conference on Robotics and Mechatronics
values. Blue line: no payload. Magenta line: 0.5 [kg] (ICROM), 196–201. doi:10.1109/ICRoM.2016.7886846.
payload. Ocher line: 1 [kg] payload. Violet line: 2 [kg] Maurtua, I., Pedrocchi, N., Orlandini, A., d. G. Fernndez,
payload. Green line: 5 [kg] payload. J., Vogel, C., Geenen, A., Althoefer, K., and Shafti, A.
has been considered. The motor velocity cutoff frequency (2016). Fourbythree: Imagine humans and robots work-
has been chosen equal to ω̂d,p = 3, which corresponds ing hand in hand. In 2016 IEEE 21st International Con-
to ωd,p = 86.5735 [rad/s]. In order to demonstrate the ference on Emerging Technologies and Factory Automa-
robustness of the controller, four different disk payloads tion (ETFA), 1–8. doi:10.1109/ETFA.2016.7733583.
have been attached to the robot flange, changing Jl as Misgeld, B.J., Hewing, L., Liu, L., and Leonhardt, S.
shown in Table 11. (2017). Robust gain-scheduled control of variable stiff-
For each test, the performance are evaluated by computing ness actuators. IFAC-PapersOnLine, 50(1), 8804–8809.
the link position and the motor velocity mean absolute doi:10.1016/j.ifacol.2017.08.1535.
errors (respectively, |ē|p and |ē|v ) and the control effort ce Paine, N., Oh, S., and Sentis, L. (2014). Design and con-
defined as  trol considerations for high-performance series elastic
1 T actuators. IEEE/ASME Transactions on Mechatronics,
ce = |∆τ |dt 19(3), 1080–1091. doi:10.1109/TMECH.2013.2270435.
T t=0
The obtained values are in Table 12, while Figure 9 shows Ragonesi, D., Agrawal, S., Sample, W., and Rahman, T.
the position and the torque trends for each Jl values. It is (2011). Series elastic actuator control of a powered ex-
possible to note the effectiveness of the proposed method oskeleton. In 2011 Annual International Conference of
to follow the required set-point signal in presence of an the IEEE Engineering in Medicine and Biology Society,
unknown payload. In fact, the mean absolute errors are 3515–3518. doi:10.1109/IEMBS.2011.6090583.
limited also in the most critical situation (namely, the 5 ROS (2018). ROS homepage. http://wwww.ros.org.
[kg] payload case). Accessed: 2018-01-31.
Simoni, L., Beschi, M., Legnani, G., and Visioli, A.
5. CONCLUSIONS (2017). Modelling the temperature in joint fric-
Robust tuning rules for Series Elastic Actuator control tion of industrial manipulators. Robotica, 122. doi:
have been presented. The tuning rules are designed for a 10.1017/S0263574717000509.
PD-PI cascade control and allow the online computation Verstraten, T., Beckerle, P., Furnémont, R., Mathijssen,
of the robust control parameters for each value of the link G., Vanderborght, B., and Lefeber, D. (2016).
inertia. Experimental results show the effectiveness of the Series and Parallel Elastic Actuation: Impact of
methodology on a three degree-of-freedom series elastic natural dynamics on power and energy consumption.
actuated robot. Mechanism and Machine Theory, 102, 232–246. doi:
https://doi.org/10.1016/j.mechmachtheory.2016.04.004.
ACKNOWLEDGEMENTS
The work is partially supported by FourByThree Project
H2020-FoF-06-2014-737095.

225