International Journal of Advanced Robotic Systems

Design and Real-Time Control

of a 4-DOF Biped Robot

Regular Paper



Jose Alejandro Vazquez
1,*
and Martin Velasco-Villa
1


1 CINVESTAV-IPN, Departamento de Ingeniera Elctrica, Seccin de Mecatrnica, Mxico D.F., Mxico
* Corresponding author E-mail: [email protected]

Received 24 Oct 2012; Accepted 7 Jun 2013

DOI: 10.5772/56738

2013 Vzquez and Velasco-Villa; licensee InTech. This is an open access article distributed under the terms of the Creative
Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use,
distribution, and reproduction in any medium, provided the original work is properly cited.


AbstractThisworkisfocusedonthedesign,construction
and model based control of a biped robot during the
walking cycle on the sagittal plane. For the analysis, the
single support phase is considered to be the dominating
dynamics,byassuminganinstantaneousdoublesupport
phase which is only described by the impact
phenomenon. The joint tracking problem is analyzed by
means of a model based control strategy, which
incorporates a reformulation of the Coriolis matrix that
allows the cancellation of non antisymmetric terms in
order to formally proof the asymptotic stability of the
coordinate error system representation in a local sense.
Some experiments are carried out for a predefined
reference trajectory for the walking cycle of a 4DOF
bipedrobot.

KeywordsBipedRobot,Dynamics,WalkingCycle

1.Introduction
It is clear how the legged locomotion systems has had an
increasing interest in the last decades; in particular biped
robots have gained attention because of their special
characteristics regarding their performance on difficult
environments. There exists a large variety of works in the
literature addressing several approaches of biped
locomotion, some of them deal fundamentally with
dynamic models; for instance, in [1] and [2] a model of a
biped robot is developed and experimentally validated; in
[3] it is obtained a dynamic model which is only
implemented in a simulation platform; or in [4], the
dynamics of the robot is analyzed via systems with
impulseeffects.Regardingthewalkingcycle,atypicalgait
is divided into two phases, the single and the double
support [5], [3]; although sometimes it is considered only
thesinglesupportphasewithboundaryconditions[2],[4].
In some situations, it is incorporated a third phase [6],
which corresponds to a flight phase in a running mode. It
isalsowidelyaddressedthedynamicwalkingstabilityfor
particularfeedbackcontrollers;forexamplein[7],whereit
isconsideredthedoublesupporteventasaperturbationof
the single support dynamics, the stability is analyzed in
terms of Lyapunov strategies. The use of Poincare maps
and limit cycles has been discussed in [8], [4], [9]. The
designofreferencetrajectoriesisalsoanimportanttopicin
terms, for example, of optimal performance [5], [10];
natural inspired evolution [11]; or time dependent
polynomial design [12]. Additionally, the mechanical
design of a biped robot structure has been addressed in
order to get special characteristics of mobility; this fact
however represents an important energy consumption
1 Jose Alejandro Vzquez and Martin Velasco-Villa: Design and Real-Time Control of a 4-DOF Biped Robot
www.intechopen.com
ARTICLE
www.intechopen.com
Int. j. adv. robot. syst., 2013, Vol. 10, 312:2013
issue. Many structural designs has been proposed; for
instance,in[13]abipedrobotisdesignedwitharickshaw
based mechanism and, in [14], a biped robot is actuated
only with one motor by taking advantage of passive
characteristics of its structure. In [15], the energy
consumption is addressed by means of a special joint and
drive mechanism. The use of cables, screwnut systems,
gears and springs have been also widely considered; see
forinstance[16],[17],[18]and[19].

Concerning control strategies, biped locomotion has been

studied with different feedback schemes such as passivity
basedtechniques[20];adaptivecontrol[21]orslidingmode
strategies[22]. In [23],itis presented a dynamic modelofa
biped robot, and a computedtorque control feedback has
been experimentally implemented, taking into account the
nonmodeled forces on the motor gear on each union. In
[24], it is assumed that the mass inertia of the legs is
sufficiently small; the design is complemented with a
feedback linearization strategy which is experimentally
implementedinaphysicalplatform.Theclassicalcomputed
torquecontrolisalsoimplementedin[25]fora7linkbiped
robotbyincorporatingagravitycompensationscheme.

Inthisworkit isassumedthatthebipedrobotdynamics
istotallydefinedbythesinglesupportphase,takinginto
account the double support as a discontinuous
phenomenon,duetotheimpactoftheswinglegwiththe
ground and the swapping of coordinates such that the
stance leg becomes the swing leg and vice versa. The
control strategy is developed by considering the
structural characteristics of the single support phase
model and by means of a new representation of the
Coriolis matrix. This fact allows to show the asymptotic
stability of the error coordinates for the closedloop
system.Also,inthephysicalplatform,thecontroltorque
signalsareappliedbymeansofscrewnutarraysinorder
toobtainalowtorquedemandattheactuators.

Therestofthepaperisorganizedasfollows:InSection2,
itispresentedabriefdescriptionoftheconsideredbiped
robot and its dynamic model is obtained by an Euler
Lagrange formulation describing also their main
structural properties. In Section 3, it is addressed the
physical laboratory prototype. It is also stated the
mobilitycharacteristicsandthetypeofactuationforeach
joint. Section 4 describes the design of the feedback
control law by taking into account the obtained dynamic
model.Inthissection,astabilityanalysisisdevelopedby
meansofaLyapunovtechnique.Theexperimentalresults
for the closedloop system are presented in Section 5,
where the reference trajectory is also defined. Finally,
someconclusionsaregiveninSection6.
2.Classofbipedrobot
The structure of the biped robot is designed with two
degrees of freedom per leg, corresponding to the knee
andhipjoints.Sincetorsoandanklearenotconsidered,it
is obtained a 4DOF system. The dynamic analysis is
carried out on the sagittal plane with a punctual contact
with the ground. In the literature, more complex robot
designs have been addressed; however the apparent
simplicityoftheproposedconfigurationfollowsfromthe
attempt to successfully implement, in real time, a
particular actuation mechanism with a specific feedback
control. The walking cycle is considered as a periodic
event defined by a single support phase dynamics,
restarted by an instantaneous double support. This
restarting phenomenon is described by means of the
impact dynamics that produces an instantaneous change
onthejointvelocitieswithoutaffectingtherobotposture.
In addition, it is assumed that the robot does not slip at
either the support or the impact point. This fact implies
new initial conditions at the beginning of each step. This
generalcontrolstrategyhasbeenimplementedfollowing
[7].
2.1Bipedrobotdynamics
TheclassofbipedrobotisdepictedinFigure1.

Figure1.Bipedrobotconfiguration.
The robot dynamics is obtained by means of the Euler
Lagrange formulation [26], under the hypothesis of
concentrated mass at each link and the assumption of
neglectedfrictionforceatjointsandactuators.Therefore,
the standard representation of the single support phase
modelisstatedas,

( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) D q t q t C q t ,q t q t G q t B t , t + + = (1)
where
( ) ( ) ( ) ( ) ( )
T
31 32 41 42
q t q t q t q t q t ( =

is the
generalized coordinates vector. As usual, ( )
D q is the
inertiamatrix,boundedandpositivedefinite,and ( )
C q,q
is the matrix of Coriolis and centripetal forces. ( )
G q
represents a matrix of gravitational effects and B defines
2 Int. j. adv. robot. syst., 2013, Vol. 10, 312:2013 www.intechopen.com
the input matrix. The vector
( ) ( ) ( ) ( ) ( )
T
31 32 41 42
t t t t t t t t t t ( =

definestheappliedjoint
torquesoftherobot.
After a straightforward computation, matrices
D(q),C(q,q),G(q) and B involved on the single support
model(1),canbedefinedas,
( ) ( )
( )
1 2 3
31
1 4 5
32
2 4 6
41
42
3 5 6
1 2 3
1 4 5
1 2 q 1 q 2 q
1 q
2 q 3 1 q 5 q
2 q
1 q 1 q 4 6 q
3 q
6
4 q
2 q 5 q 6 q
2 q 32 1 q 41 2 q 32
2 q 31 1 q 41 5 q
a a c k c k c
g s
a c k z c k c
g s
D q , G q
k c z c k k c
g s
k
g s
k c k c k c
2
0 a s q k s q k s q
a s q 0 z s q k s q
C q, q
(
(
(
(
(
(
(
= = (
(
(
(
(
(

(

(



=

2 4 6
3 5 6
42
1 q 31 1 q 32 6 q 42
2 q 31 5 q 32 6 q 41
k s q z s q 0 k s q
k s q k s q k s q 0
1 0 1 0
0 1 0 1
B ,
0 0 1 0
0 0 0 1
(
(
(
(

(
(
(

(
(

(
=
(
(
(

where, for simplicity of notation, it has been defined

( )
( )
( )
( )

s sin ,c cos , = = and

1 32 31 2 41 31 3 42 31
4 41 32 5 42 32 6 42 41
q q q , q q q , q q q
q q q , q q q , q q q .
= = =
= = =
(2)
Being g the gravity constant, parameters
i i i
a , k , z and
i
g
depend on the structural configuration of the robot and
theyaregiveninTable1.

1 3 4 H
2 5
a L M M M
3 4
| |
= + +
|
\ .
1
1 4 3 3
3
a 1
k L L M
L 4
| |
= + |
|
\ .

2 3 4
2 1
a L M M
3 2
| |
= +
|
\ .

2 4 3 4
1
k M L L
2
=
1 3 3 4 H
3
g L g M M M
2
| |
= + +
|
\ .

3 3 4
2 1
k L M M
3 4
| |
= +
|
\ .

2 3 3 4
1
g L g M M
2
| |
= +
|
\ .
4 4 3 H
2 5
k L M 2M M
4 4
| |
= + +
|
\ .

3 4 3 4 H
3
g L g 2M M M
2
| |
= + +
|
\ .

5 4 3 4
1
k M L L
2
=
4 4 4
1
g L M g
2
=
6 4
1 2
k M L
2 4
=
1 4 2 3
z L a / L =

Table1.Structuralmodelparameters.
2.1.1Structuralproperties.
The biped robot model satisfies the following structural
properties that will be used later on the development of
thecontrolstrategy.Mostofthemareinaccordancewith
thedynamicmodelofageneralrobotmanipulator[26].
- Property 1. Matrices D(q)

and ( )
C q,q are such that: i)
D(q) 2C(q, q) N(q, q) =

for a skew symmetric matrix

N(q,q) andii)
T
D(q) C(q,q) C (q,q). =

- Property 2. For any vectors
n
z, w R : e i)
C(q,q)w C(q, w)q = and ii)
C(q, z w) C(q, z) C(q, w). + = +
- Property 3. ( )
C q,q is a bounded matrix, i.e.
C(q,q) k q s forsome k R. e
- Inaddition,theCoriolismatrixsatisfiesthefollowing
property,whichisveryusefulinthesynthesisofthe
feedbackcontrol.
- Property4.Matrix ( )
C q,q canberewrittenintermsofa
skewsymmetric matrix ( )
1
C q,q as
1 2
C(q,q) C (q,q) C (q, q) = + with,
1 2 3
1 4 5
2 4 6
3 5 6
1 2 3
4 5
6
2 q 31 1 q 31 2 q 31
2 q 31 1 q 32 5 q 32
1
1 q 31 1 q 32 6 q 41
2 q 31 5 q 32 6 q 41
2 q 1 1 q 2 2 q 3
1 q 4 5 q 5
2
6 q
0 a s q k s q k s q
a s q 0 z s q k s q
C
k s q z s q 0 k s q
k s q k s q k s q 0
and
0 a s q k s q k s q
0 0 z s q k s q
C
0 0 0 k s
(
(

(
= (

(
(
(



=

6
.
q
0 0 0 0
(
(
(
(
(
(

Tocompletethewalkingcycle,thedoublesupportphase
is defined in terms of an instantaneous impact dynamics
and a swapping strategy, producing a set of initial
conditions for the next step. With this aim, it is assumed
thattheimpactwiththegroundofthecurrentswingleg,
attheendofeachstep,producesaninstantaneouschange
of the joint velocities without implying a change in the
posture of the robot. This approach has been widely
consideredintheliterature;seeforinstance[4]and[27].
For the analysis of the double support phase, it has been
considered two additional coordinates,
T
1 2
[ , ] , , to define
theCartesianpositionoftheendpointofthestancelegin
the X Y plane. In this way, an extended vector of
generalized coordinates
T
e 31 32 41 42 1 2
q [q q q q ] , , = is
alsodefined.

Inordertoobtainananalyticalexpressionforthevelocity
justaftertheimpact q ,
+
considerthekineticenergyofthe
robotfromtheextendedmodelas,

( )
T
e e e e
1
T q D q q
2
= (3)
wheretheextendedinertiamatrix
e
D hastheform,

( )
( ) ( )
( )
T
e e
D q L q
D q
L q M
(
( =
(

(4)
with ( )
D q beingtheoriginalinertiamatrixof(1)and,
3 Jose Alejandro Vzquez and Martin Velasco-Villa: Design and Real-Time Control of a 4-DOF Biped Robot
www.intechopen.com
( )
1 31 2 32 3 41 4 42
1 31 2 32 3 41 4 42
b c b c b c b c
L q
b s b s b s b s
(
=
(


with,
1 3 3 4 H 2 3 3 4
3 4 4 3 H 4 4 4
3 1
b L M M M , b L M M
2 2
3 1
b L M 2M M , b M L
2 2
| | | |
= + + = +
| |
\ . \ .
| |
= + + =
|
\ .

and
4 3 H
4 3 H
2M 2M M 0
M .
0 2M 2M M
( + +
=
(
+ +


Substituting equation (3) into the Lagranges impulsive
equationasin[28],yields,

( ) ( ) ( )
T
2
2
x
D q q D q q L q 0,
y
+
(
=
(

(5)
where q

is the articular velocity just before the impact

and
T
2 2
[x y ] is the position of the endpoint of the swing
leg with respect to the supporting point. Moreover, the
slippingrelationattheendoftheimpactlegisexpressed
as,

2
2
x
Eq
y

(
=
(

(6)
where E istheJacobiandefinedas,

2
P
E
q
=
c

(7)
with
T
2 2 2
P [x y ] . = The post impact velocities q
+
can be
expressedfrom(5)and(6)as,

1 T
q D D L E q .
+
(
=

(8)
Notice that, since matrices D and
T
L E are bounded, the
postimpactvelocitydoesnotincreaseindefinitely.

Once the post impact articular velocities are known, the

supporting and swing leg swapping is defined by the
map,

1 T
q
q
R
D D L E q
q

+

+
(
(
(
= (
( (
(

(9)
where R isatransformationmatrixdefinedas,
n n
n n
T 0
R
0 T

(
=
(


with
0 1 0 0
1 0 0 0
T .
0 0 0 1
0 0 1 0
(
(
(
=
(
(
(

The described velocity discontinuities and the defined

boundedness characteristic, which can be stated with a
particularreference,allowtoconsiderthedoublesupport
as a perturbation of the states in the single support
model.
3.Physicalplatform
The biped robot has been designed in order to be
dynamicallyanalyzedonlyonthesagittalplane.Tocarry
out the experiments, an external restricting frame has
beendesignedbymeansofatranslationalfreemovement
mechanism along the vertical and horizontal axes. This
exostructure avoids the robot to fall down on the lateral
plane and provides a free movement on the forward
direction.Therobotisfixedtotheexostructureatthehip,
allowing a simple noninvasive support. Figure 2 shows
this structural frame. Notice that, apparently, there is an
element that acts as a torso; however, it is not free to
rotateanditsdynamicsisneglected.

Figure2.Physicalplatform.
Therobotisbuiltwithaparticularactuatormechanismat
each joint which allows to generate the high torque
requirements from a low torque actuator; however it has
an important drawback that relies on a high velocity
demand. The design is based on [17] and consists
4 Int. j. adv. robot. syst., 2013, Vol. 10, 312:2013 www.intechopen.com
basically on a linkage mechanism, where the main input
is provided by a brushed DC motor, producing a
translational movement along a ball bearing screw,
converting in this way the rotational movement into a
translational one. Complementary, the linkage
mechanism produces the required rotational movement
for the knee and hip joints. This mechanical interface is
depictedinFigure3forthehip(right)andtheknee(left)
joints.

In Figure 4, a schematic diagram is presented with this

concept.

The analysis of the proposed mechanism can be

developed in terms of the dimensional parameters
(Figure 4) and the dynamics of a screw. Basically, and
based on [17], kinematic relations allow to define the
velocity of the nut of the screw with respect to a desired
reference and as a function of the motor velocity. For
instance, from Figure 4, it can be obtained the vertical
position of the nut in the knee (left side), which after a
timederivationcanbeexpressedas,

Figure3.Detailofthelinkagemechanism.

( ) ( )
k
mk k k k mk
k
a p
L sen q q q q
b 2
t 0 t 0
t
(
+ =
(
(

cos (10)
where
k 4w
q q , w 1, 2; = = denotes a general knee
coordinate and
mk
q is the velocity of the corresponding
knee motor. Equation (10) allows to obtain a relation
between the motor and joint velocities of the form
mk k k k
q (q )q = with a gain factor for the knee
k k
(q )
givenas,
( ) ( ) ( )
mk k
k k k k
k
2 L a
q sen q q .
p b
t
t 0 t 0
(
= +
(
(

cos (11)
Equivalently, a gain factor for the hip ( )
h h
q is obtained
as,
( ) ( ) ( )
mh h
h h h h
h
2 L a
q cos q q
p b
t
t o t o
(
= + + +
(

sin (12)
with
h 3w
q q , w 1, 2, = = describing the general hip
coordinate.Inbothcases,itisdefined,
( )
( )
h diag mc h
2 2
h bh h
k mk k hor
2
k bk k
a d L q
b L a
a L q d
b L a .
t o
t 0
= +
=
= +
=
cos
sin

Figure4.Schematicrepresentationofthekneeandhipjoints.
Notice that the parameters from equations (11) and (12)
describe physical dimensions; therefore they rely on the
dimensions of the physical platform that has been
previously designed and constructed in order to satisfy
velocity and torque requirements. These parameters are
numericallyexpressedinTable2.

mk
L 0.04m =
diag
d 0.06m =
T
M 6Kg =
o
h
q 160 =
min
bk
L 0.06m = 0.26rad 0 =
h
M 2.30Kg =
o
h
q 220 =
max
mc
L 0.04m = 0.85 q =
3
M 1.20Kg =
o
k
q 110 =
min
bc
L 0.03m =
hor
d 0.02m =
4
M 0.32Kg =
o
k
q 160 =
max
0.50rad o = screw 0.1m = p 0.002m =
Table2.Physicalparameters.
The mechanical advantage of the linkage mechanism is
determined by the gains factors (11) and (12) for the
articular range of each joint. Those factors describe the
torquegainateachjointfromthemotorand,atthesame
time, the corresponding velocity relation. Figures 5a and
6a,showhowtherequiredtorqueateachjointisreflected
in a quite low torque in the corresponding DC motor.
Also,itispossibletoseehowthevelocityatthemotoris
increased (Figures 5b and 6b) in order to satisfy a
particular reference trajectory. The performance of the
linkage mechanism is evaluated as a linear displacement
of the nut along the screw, which implies a continuous
angularmotionateachjoint.Forthisparticularsituation,
Figure 5a shows how a high torque of 80Nm at the hip
joint is reduced at the motor shaft to 1.5Nm, assuring an
5 Jose Alejandro Vzquez and Martin Velasco-Villa: Design and Real-Time Control of a 4-DOF Biped Robot
www.intechopen.com
adequate behavior of the mechanism. An inverse effect
can be seen from Figure 5b with respect to velocities. A
similar analysis can be also obtained for the knee from
Figure6b.Theinputoutputtorqueandvelocityrelations
canbemanipulatedbychangingtheinvolvedmechanical
parameters.

Figure5.Hiptorqueandvelocityrate.

Figure6.Kneetorqueandvelocityrate.
The proposed mechanical interface implies a working
space that is delimited by the total displacement of the
nutalongthescrew,aswellasthelengthoftheinvolved
elements. This characteristic defines displacement
restrictions at each joint which in general can be
describedas,

h h h
q q q s s
min max
(13)

h k k h k
q q q q q t t + s s +
min min max max
(14)
where
h
q and
k
q are the hip and knee joint angles
respectively.Theworkingrangesaredefinedwithrespect
toanabsoluteverticalreferenceasdefinedinFigure1.

However the limits

h h k
q ,q ,q
min max min
and
k
q
max
are defined
in a local reference frame, this is, the femur is measured
with respect to the hip and the tibia with respect to the
femur. These relations are depicted in Figure 7 and their
numericalvaluesarespecifiedinTable2.

As stated, the contact with the walking surface is

designed to be punctual and a pressure mechanism
consisting of a low cost resistive sensor has been
implementedattheendofeachlegsuchthattheinduced
pressure force at this point can be measured. In this
particular case, the force is used to detect the impact
instant and its magnitude is not relevant. This
implementationisdepictedinFigure8.

Figure7.Mechanicallimitsforthelegs.

Figure8.Pressuremechanismattheendofeachleg.
To compute the articular position, an optical encoder is
installed at each joint. Also, there is an inclinometer at
each tibia to sense an eventual fall of the mechanism.
Limit switches are used to prevent a mechanical damage
as a consequence of displacements out of the ranges
previouslydescribed.

Itisclearthatthecomplexityofthemechanicalprototype
is not totally described by the mathematical model (1)
due to (among others) the fact that this model does not
consider the autolock characteristic that the type of
actuation induces, this is, the model assumes that the
motor acts directly at the joint. However, it is possibleto
incorporate the effects of the actuation mechanism by
meansoftherelation,

( )
j m
q t t = (15)
where
m
t is the motor torque; the subscript j defines the
hip joint, j h, = or the knee joint, j k; = andt is the control
6 Int. j. adv. robot. syst., 2013, Vol. 10, 312:2013 www.intechopen.com
signal for each joint. Notice that
( )
j
q are defined in
equations(11)y(12)andtheyhaveaninverserolewhen
actingasgainfactorswithtorquesignals.
4.Modelbasedcontrolstrategy
In this section it is addressed the proposed control
strategy in order to get a stable walking cycle. Roughly
speaking a global asymptotic stability is not possible
because of the nature of the complete walking cycle in
terms of a resetting event at each end of step; however,
theassumptionofconsideringthesinglesupportphaseas
themaindynamics,allowstolocallyanalyzethewalking
cycle as a perturbed dynamics with a periodic
perturbation described by an instantaneous double
support.Thisapproachhasbeenalreadystudiedin[7].

Taking into account model (1), it can be proposed a

modelbasedfeedbackcontrolas,

( ) ( ) ( ) ( )
( )
j m d d d
d p 2 d
B q D q q C q,q q G q
k q k q C q, q q
t = + +
+



(16)
where
n n
d
k R

e and
n n
p
k R

e are diagonal positive
definite matrices and
d
q q q = is the joint tracking error
with
d
q representing a sufficiently smooth desired joint
trajectory. ( )
2 d
C q,q is an upper triangular matrix
previouslydefinedinProperty4.

Notice that equation (16) is defined in the joint space;

however, the absolute reference and the real dynamics
are first expressedin the Cartesian plane. In this way, a
desired Cartesian performance is indirectly induced by
meansofanexpectedstablefeedbackcontrolthroughan
inversekinematicmapping.

In addition, the term

( )
j
q results in a scalar that is
assigned with a suitable transformation such that it
matches with the corresponding joint, either hip or
knee. Moreover, it is required that
( )
j
q 0, = which
meansthat,

h
h
h
b
q a
a
t o
| |
= + +
|
|
\ .
tan (17)
and

h
k
h
a
q a ,
b
t 0
| |
= +
|
|
\ .
tan (18)
hence, the definition of the desired performance has to
considerthisfact.

Finally, notice that matrix B is invertible, so the

calculationofthemotortorquesisdirectlyobtainedfrom
equation(16)inapracticalimplementation.
Lemma 5. Consider system (1) and the feedback law given in
equation (16). Under these conditions, feedback (16) renders
theclosedloopsystem(1)(16)asymptoticallystable.

Proof.Theclosedloopsystem(1)(16)resultsin,

( ) ( ) ( ) ( )
d d d 2 d p
D q q C q,q q C q,q q k C q,q q k q 0. ( + + + =


(19)
Noticethat,byconsideringProperty2,
( ) ( ) ( ) ( )
d d d
C q,q q C q,q q C q,q C q,q q. ( = +



Then,equation(19)canbewrittenas,

( ) ( ) ( ) ( )
d d p 2 d
D q q C q,q C q,q q k q k q C q,q q 0. ( + + + + =


(20)
The latter equation allows to show the importance of
Property 4, which helps to finally obtain the closedloop
dynamicas,

( ) ( ) ( )
1 d d p
D q q C q, q C q,q k q k q 0. ( + + + + =


(21)
Noting first that the error dynamics given by equation
(21) has an equilibrium point at the origin,

(q, q) (0, 0), =

the closedloop stability analysis can be done by
consideringaLyapunovcandidatefunctionoftheform,

( )
T T
p
1 1
V q, q q Dq q k q.
2 2
= +

(22)
The time derivative of
( )
V q, q

along the solution of (21)

produces,
( ) ( ) ( )
( )
T T
1 d d
T
d 1 d
1
V q C q,q C q,q q k q q Dq
2
q k C q, q q
( = + +

( = +

and the fact that ( )

1 d
C q,q is a skewsymmetric matrix
accordingtoProperty4,allowstofinallyobtain,
( )
T
d
V q, q q k q. =


ItisnowclearthatbyconsideringtheLaSallesinvariance
principle; it is possible to show the required asymptotic
stabilityfortheerrorcoordinate q.

It is evident that the above lemma holds only when the

single support phase is defined as the main dynamics.
This means that, in a complete analysis of the walking
cycle, the effect of the impact avoids to obtain an
asymptotic stability property, because of the expected
and periodic discontinuity of the error coordinates. In
ordertoovercomethisproblem,apossiblesolutioncould
be the definition of a reference path profile that includes
theimpactdynamicsdescribedbyequation(9)attheend
of the step. However, this is not practical because of the
switching reference requirement that in general does not
7 Jose Alejandro Vzquez and Martin Velasco-Villa: Design and Real-Time Control of a 4-DOF Biped Robot
www.intechopen.com
allow to produce a perfect tracking and the asymptotic
propertydoesnotevenhold.

The approach that has been considered in this work

consists on the definition of a reference trajectory such
that at the impact, the velocities are sufficiently low and
the impulse effects are not significant. Even with the
adopted approach, the asymptotic stability cannot be
stated, fulfilling only a simple condition of stability. In
order to analyze the impact into the complete walking
cycle,anaturalapproachcorrespondstoahybridsystem
that involves impulse effects; studied in [4] for biped
robots.

It is important to clarify that the above result is a

theoretical one. This means that in order to guarantee
dynamical stability, the robot has to perfectly follow the
proposed reference. In practice, and due to non modeled
dynamicsorexternaldisturbances,itisalmostimpossible
to achieve; moreover, the robot could naturally tend to
fall when its posture is such that the center of mass CM
does not remain exactly over the vertical line defined by
the supporting point. This fact could allow to define a
non falling condition, just by assuring an immediate
supportwiththeswingleg,thisis,

2 x x
x CM when CM 0, (23)
which means that, when the center of mass CM is out of
the vertical line defined by the supporting point, the
swing leg should act as an eventual support if the robot
tendstofall.Thiscouldbesimilartoadynamicalstability
indexforrobotswithsuperficialcontactwiththeground,
wherethenonfallingconditionisassuringbymeansofa
ZMPcondition.
5.Realtimeevaluation
5.1ReferenceTrajectory
ThereferencetrajectoryisdesignedintheCartesianspace
and translated to the articular one, where the feedback
control is implemented, via inverse kinematics. The
trajectory is designed taking into account two points on
therobot:thehipandtheendpointoftheswingleg.

It is required that the hip follows a constant reference hh

along the vertical axis Y and, a reference around the
origin on the X axis; this latter condition is in order to
maintainthecenterofmassoftherobotaroundtheorigin
during the first part of the step for stability purposes. At
the end of step, the center of mass is moved forward on
the X axis as a consequence of the passive dynamics
produced by the effect of the swing leg when it almost
touchtheground.

In addition, the trajectory of the end point

2 2
(x , y ) of
the swing leg defines the length of the step and the
maximalelevationofthefoot.Thistrajectoryisdefined
by a time dependent, smooth polynomial function of
theform,

3 4 5
2 0 1 2 3

x t c c t c t c t . (24)
For the vertical displacement, the trajectory is designed
withamaximalheight
s
h andasteplength s as,

2 s
2 s 2
2
h

y t h x t .
s
(25)
Notice that the vertical trajectory is a function of the
horizontalone.Also,theuseofthehighorderpolynomial
(24) allows to obtain a sufficiently smooth desired
trajectory that could reduce the velocity discontinuity at
the impact point. Time

t is restarted periodically in order

to redefine (24) and (25) for the beginning of each step.
Table 3 shows the required parameters involved in the
referencetrajectoryfunctions.

s 0.030m
h
h 0.44m
0
c s
2
4
p
s
c 30

t

s
h 0.01m
p

t 2.2s
1
3
p
s
c 20

t

3
5
p
s
c 12

t

Table3.Referenceparameters.
p

t istheinitialreferencetimeofthestepduration,which
allows to calculate the coefficients
i
c in (24) by means of
the interpolation of a specific desired performance;
howeveritdoesnotnecessaryimpliesthateachsteptakes
exactly this time since the real time duration is
determined by the contact of the swing leg with the
ground, which eventually could be slightly larger or
shorter than
p

t . This phenomenon defines a new time

parameterwhichisusedontheonlinecomputationofthe
new coefficients
i
c . Symbolically, those coefficients are
definedasintheTable3.
5.2Experimentalresults
To carry out the experiments, feedback law (16) is
implementedinaMatlabSimulinkplatform,bymeansofa
DSP system DS1104 from Dspace. This board acts as real
time interface for the overall system: 1000 cpr optical
encoders; two FlexiForce resistive force sensors; two
9000cpr inclinometers at each tibia and the limit switches.
Also,theDSPsystemcomputestherequiredfeedbacklaw
and provides the analog output torque signal for driving
the DC motors. A general scheme of the experimental
platform is depicted in Figure 9. For all the experiments it
wasconsideredasamplingtimeof0.001seconds.
8 Int. j. adv. robot. syst., 2013, Vol. 10, 312:2013 www.intechopen.com

Figure9.Experimentalplatform.
In the experiments, the robot is attempted to follow the
referencetrajectorypreviouslydescribedwhichproduces
thestabledynamicalperformanceofthewalkingcycle.In
Figure 10 it is shown the evolution of the joint
coordinates as a result of the robot displacement. This
figure depicts the first 10 seconds of the experiment
wheretherobotdevelopsfourcompletestepsdividedby
dashed vertical lines. As expected, a periodic
performanceisevidentandthefeedbackcontrolshowsa
stableresult.

ThecorrespondingjointvelocitiesareshowninFigure11;
notice that at the end of each step it is possible to
appreciate a discontinuity velocity of small magnitude
duetotheeffectoftheimpactphenomenon.Theobtained
velocity change is small due to the particular design of
thereferencetrajectoryinaccordancewiththetheoretical
result.

Figure10.Articularperformanceofthebipedrobot.
TheevolutionofthecenterofmassintheCartesianspace
is depicted in Figure 12. Notice that its position remains
aroundtheorigininthehorizontalaxisasrequired,being
this fact more evident at the beginning of the step. The
centerofmassismovedforwardwhenthetimeincreases
inordertotakeadvantageofthepassivecharacteristicof
therobotandfinallythestepends.

Figure11.Jointvelocitiesofthebipedrobot.

Figure12.Positionofthecenterofmass.
The two Cartesian components of the end point of the
swing leg
2 2
(x , y ) are shown in Figure 13 and the
evolution of the hip point
H H
(x , y ) is depicted in Figure
14. Notice that the swing leg develops steps with
amplitude of 6 cm as required, with a maximal elevation
of 1cm from the surface. The hip has a quasi constant
evolutionintheverticalaxis,similartothecenterofmass
of the robot; however, in the horizontal axis the hip
evolvesaroundtheorigin.Thisisbecausethehipelement
concentrates a larger mass with respect to the rest of the
elementsanditspositioniscriticalforstabilitypurposes.

Figure13.Evolutionoftheendpointoftheswingleg.
9 Jose Alejandro Vzquez and Martin Velasco-Villa: Design and Real-Time Control of a 4-DOF Biped Robot
www.intechopen.com

Figure14.Positionofthehip.
Thejointpositionerrorsderivedfromtheexperimentare
shown in Figure 15. Notice that their maximal values
appear at the beginning of each step and they decrease
with time to a neighborhood of the origin at the end of
the step. The performance is apparently erratic and the
zero error is not completely attained, this is because the
design of the reference trajectory does not directly
consider the passive characteristic of the robot, which
appears at the end of the step and produce a slight
deviationfromtheexpectedreference.Inspiteofthisfact,
the performance is quite acceptable to satisfy the stable
walkingcycle.

Figure15.Trackingerrors.
The control torque signal at each DC motor shaft is
shown in Figure 16 and, in order to show the role of the
actuation mechanism, in Figure 17, the torque applied at
eachjointisalsoshown.

Figure16.Controlsignalatthemotors.
Noticethatthereexistsahighgainfactorfromtheoutput
torqueofeachmotor,asdefinedin(11)and(12).Ithasto
be clear that the refered gain factors depend on the
performedtrajectoryinanonlinearwayandtheyarenot
totally achieved in the current experiment, where a
maximal gain is around 60. However it can be increased
bymodifyingthepostureoftherobot,takingintoaccount
thatthevelocitywilldecreaseinaproportionalrate.

Figure17.Controlsignalatthejoints.
It is important to state that each DC motor has a power
supplyof24V,andduringtheexperiments,thereexistsa
maximalcurrentconsumptionof1.6Amperespermotor.
It is natural that the maximal consumption is produced
bythesupportingkneemotor,whichloadsalmostallthe
system when developing a step. In contrast to this
situation, the free knee motor consumes no more than
0.2A.Ingeneral,itisobtainedanaverageconsumptionof
2.5A during a single walking cycle which represent no
morethan70Wofelectricalpowerintotal.
6.Conclusions
Inthisworkitispresentedtheanalysisandcontrolofa4
DOF biped robot. It is considered a particular actuator
mechanism with a mechanical advantage which allows to
apply high torques on the articulations from a quite small
torque at the utilized low power DC motors. The walking
cycle is controlled by considering a model based control
strategy which is implemented on the single support
phase. The double support phase is considered as a
perturbation for the robot, producing a change in joint
velocitiesduetotheinstantaneousimpactoftheswingleg
with the ground; however, in order to minimize their
effects,asmoothimpactisdesignedbymeansofasuitable
reference trajectory. It is shown that for the considered
feedback, the resulting closedloop systems is stable. The
evolution of the overall control strategy is evaluated on a
laboratoryprototypeshowinganadequateperformance.
7.References
[1] S. Kajita, T. Yamaura, and A. Kobayashi. Dynamic
walking control of a biped robot along a potential
energy conserving orbits. IEEE Transactions on
RoboticsandAutomation,8(4):431438,1992.
[2] H. Miura and I. Shimoyama. Dynamic walk of a
biped. International Journal of Robotics Research,
3(2):6074,1984.
10 Int. j. adv. robot. syst., 2013, Vol. 10, 312:2013 www.intechopen.com
[3] R. Katoh and M. Mori. Control method of biped
locomotion giving asymptotic stability of trajectory.
Automatica,20(4):405414,1984.
[4] J. W. Grizzle,G.Abba, and F. Plestan.Asymptotically
stablewalkingforbipedrobots:Analysisviasystems
with impulse effects. IEEE Transactions on Automatic
Control,46(1):5164,2001.
[5] D. Djoudi, C. Chevallereau, and Y. Aoustin. Optimal
reference motions for walking of a biped robot. In
IEEE International Conference on Robotics and
Automation,pages20022007,Barcelona,Spain,2005.
[6] C. Chevallereau and P. Sardain. Design and actuation
optimization of a 4 axes biped robot for walking
running. In IEEE International Conference on Robotics
ans Automation, pages 33653370, San Francisco, CA,
USA,2000.
[7] A.ChemoriandA.Loria.Controlofaplanarfivelink
underactuated biped robot on a complete walking
cycle. In IEEE Conference on Decision and Control,
pages20562061,LasVegas,Nevada,USA,2002.
[8] Ambarish Goswami, Bernard Espiau, and Ahmed
Keramane. Limit cycles in a passive compass gait
biped and passivity mimicking control laws.
Autonomous robots, Springer Netherlands, 4(3):273286,
September1997.
[9] F. Plestan, J. W. Grizzle, E.R Westervelt, and G.Abba.
Controlled periodic motion in a nonlinear system
with impulse effects: walking of a biped robot. In
IFACNOLCOS2001,SaintPetersburg,Russia,2001.
[10] Y.Xiang,J.S.Arora,andK.AbdelMalek.Physicsbased
modeling and simulation of human walking: a review
of optimizationbased and other approaches. Structural
andMultidisciplinaryOptimization,42:123,2010.
[11] L. France, A. Girault, and B. Espiau. Generation of
walk transient trajectories for a biped robot. In
International Conference on Advanced Robotics, Tokyo,
Japan,October1999.
[12] Z.Tang,C.Zhou,andZ.Sun.Trajectoryplanningfor
smooth transition of a biped robot. In Proceedings of
IEEE International Conference on Robotics and
Automation,pages24552560,2003.
[13] E. Ottaviano, S. Grande, and M. Ceccarelli. A biped
walking mechanism for a rickshaw robot. Mechanism
Based Design of Structures and Machines, 38(2):227242,
2010.
[14] F.J.BerenguerandF.MonasterioHuelin.Easydesign
andconstructionofabipedwalkingmechanismwith
low power consumption. In Proceedings of the 9th
International Conference on Climbing and Walking
Robots,pages96103,Brussels,Belgium,2006.
[15] M. Okada, T. Shinohara, T. Gotoh, S. Ban, and Y.
Nakamura. Humanoid robot mechanisms for
responsive mobility. In Proceedings of the 2nd
InternationalSymposiumonAdaptiveMotionofAnimals
andMachines,pagesSaPIII3,2003.
[16] I. M. Constantin Olaru, S.Krut, and F. Pierrot. Novel
mechanicaldesignofbipedrobotsherpausing2dof
cable differential modular joints. In The 2009
IEEE/RSJ International Conference on Intelligent Robots
andSystems,pages44634468,St.Louis,USA,2009.
[17] P. Sardain, M. Rostami, and G. Bessonnet. An
anthropomorphic biped robot: Dynamic concepts
and technological design. IEEE Transactions on
Systems, Man and CyberneticsPart A: Systems and
Humans,28(6):823838,1998.
[18] S.PratapSingh,A.Dutta,andA.Saxena.Designofa
biped robot with torsion springs at the joints for
reduced energy consumption during walk. In
ProceedingsoftheASME2009InternationalEngineering
Technical Conference Computers and Information in
EngineeringConference,2009.
[19] J. Yamaguchi, D. Nishino, and A. Takanishi.
Realization of dynamic biped walking varying joint
stiffnessusingantagonisticdrivenjoints.InProceedings
ofthe1998IEEEInternationalConferenceonRoboticsand
Automation,pages20222029,Leuven,Belgium,1998.
[20] M.W. Spong. Passivity based control of the compass
gaitbiped.InIFACTriennialWorlCongress,pages19
23,Beijin,China,1999.
[21] P. K. Vempaty, K.C. Cheok, and R. N. L. K. Loh.
Model reference adaptive control for actuators of a
biped robot locomotion. In Proceedings of the World
Congress on Engineering and Computer Science, pages
983988,SanFrancisco,USA,2009.
[22] S. Tzafestas, M. Raibert, and C. Tzafestas. Robust
slidingmodecontrolappliedtoa5linkbipedrobot.
Journal of Intelligent and Robotic Systems, 15(1):67133,
1996.
[23] N. Chaillet, G. Abba, and E. Ostertag. Double
dynamicmodelingandcomputedtorquecontrolofa
biped robot.InIEEE/RSJ Int.Conf. IntelligentRobotics
Systems,pages11491153,Munich,Germany,1994.
[24] K.Mitobe,N.Mori,K.Aida,andY.Nasu.Nonlinear
feedback control of a biped walking robot. In IEEE
International Conference on Robotics and Automation,
pages28652870,Nagoya,Japan,1995.
[25] J.H.ParkandK.D.Kim.Bipedrobotwalkingusing
gravitycompensated inverted pendulum mode and
computed torque control. In IEEE International
Conference on Robotics and Automation, pages 3528
3533,Leuven,Belgium,1998.
[26] Mark W. Spong and M. Vidyasagar. Robot Dynamics
andControl.JohnWilerandSons,USA,1989.
[27] M.Raibert,S.Tzafestas,andC.Tzafestas.Comparative
simulation study of three techniques applied to a
bipedrobot.InInternationalConferenceonSystems,Man
andCybernetics,volume1,pages494502,1993.
[28] J. Furusho and A. Sano. Sensorbased control of a
ninelink biped. International Journal of Robotics
Research,9(2):8398,April1990.

11 Jose Alejandro Vzquez and Martin Velasco-Villa: Design and Real-Time Control of a 4-DOF Biped Robot
www.intechopen.com