You are on page 1of 10

See

discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/280050492

CONTROL OF CHAOS IN PASSIVE DYNAMIC


WALKING OF A COMPASS-GAIT BIPED ROBOT
CONFERENCE PAPER JULY 2012

READS

19

3 AUTHORS:
Hassne Gritli

Khraief Nahla

cole Nationale d'Ingnieurs de Tunis

Universit de Montpellier

19 PUBLICATIONS 68 CITATIONS

22 PUBLICATIONS 71 CITATIONS

SEE PROFILE

SEE PROFILE

Safya Belghith
cole Nationale d'Ingnieurs de Tunis
102 PUBLICATIONS 505 CITATIONS
SEE PROFILE

Available from: Hassne Gritli


Retrieved on: 29 September 2015

CONTROL OF CHAOS IN PASSIVE DYNAMIC WALKING


OF A COMPASS-GAIT BIPED ROBOT

Hassne Gritli1, Nahla Khraeif2 and Safya Belghith1


1

National School of Engineers of Tunis, University of Tunis El Manar, BP. 37, Le Belvdre, 1002 Tunis, Tunisia
2

Ecole Suprieure de Technologie et dInformatique, Universit de Carthage,


45 Rue des Entrepreneurs, 2035 Charguia II, Tunis, Tunisia
{ grhass, nahla_khraief }@yahoo.fr, safya.belghith@enit.rnu.tn

Keywords:

compass-gait biped robot, passive dynamic walking, chaos, hybrid limit cycle, controlled Poincar map,
chaos control.

Abstract:

This paper deals with the control of chaos in the walking dynamics of a compass-gait biped robot. This
biped robot has an impulsive hybrid nonlinear model and is capable of displaying passive dynamic walking.
Its passive gait model is known to possess hybrid limit cycles and to exhibit chaotic behaviors. The
primordial contribution in this paper is to determine a linear model by linearizing the nonlinear model
around a desired passive hybrid limit cycle. From characteristic point of view, we show that the passive
hybrid limit cycle of the linear model is fairly close to that of the nonlinear model and so the two models are
almost identical. Thus, basing on the linear model, we have established an explicit expression of the
controlled Poincar map. Using the linearized controlled Poincar map, we have designed a state feedback
control law which is applied for the original nonlinear system. We show via numerical simulations the
effectiveness of our designed controller for the control of chaos in the passive dynamic walking of the
compass-gait biped robot.

INTRODUCTION

In the last decades, the unusual behavior of


nonlinear dynamics and chaos has attracted the
attention of several different scientific communities.
An interesting and challenging research subject
arisen in the field of nonlinear dynamical systems is
the control of chaos (Ott, Grebogi and Yorke, 1990;
Andrievskii and Fradkov, 2003; Alexander and
Robin, 2005), namely, the investigation of bringing
order to chaos. It has now been recognized that
chaos occurs commonly in many fields of science
and engineering and so in a large variety of
dissipative physical systems such as mechanical
systems with and without impact, legged walking
robots, etc. In the last two decades, many legged
robots have been developed. An excellent database
of biped robots built all over the world can be found
in (Beck, 2009).
In dynamic walking of biped robots, two
important walking methods have been well-

explored. The first method is the active walking, by


which some joints of the biped walking robot are
driven by some actuators. It is shown so far that this
method requires mainly the study of the second one
which is known as the passive dynamic walking
(PDW) method. This method was first proposed by
McGeer (1990) by studying an entirely passive
compass-type biped robot. This biped robot can
perform a stable and efficient walk as it goes down a
shallow slope. Interest of the design and the use of a
passive biped robot is its low energy consumption.
The robots of research realized in the framework of
the PDW constiute mainly in an analysis of
properties of the bipedal walk (dynamics, stability,
limit cycles, etc.). In this same context, extensive
studies have been proposed.
The mathematical model of the PDW of the
compass-gait biped robot is defined by an impulsive
hybrid nonlinear dynamics (Grizzle, Abba, Plestan,
2001). It is shown that the compass gait model is
simple enough to be amendable to analysis, yet
complex enough to display a wealth of interesting

phenomena. The attractive phenomenon exhibited in


the PDW is the scenario of period-doubling
bifurcation route to chaos (Goswami, Thuilot, and
Espiau, 1998; Kajita and Espiau, 2008; Moon and
Spong, 2011; Gritli, Belghith and Khraeif, 2012).
Furthermore, Gritli, Khraeif and Belghith (2012)
showed that a cyclic-fold bifurcation is generated in
the PDW of the compass-type biped giving rise to a
period-three route to chaos. They showed also in
(Gritli, Khraeif and Belghith, 2011) that the fall of
the compass-gait biped robot is caused mainly by a
boundary crisis.
The design of an effective and efficient control
scheme of chaos is one of central focuses in the field
of nonlinear science. Chaos control is based on the
richness of response of chaotic behavior. It is known
so far that a chaotic attractor has an infinite number
of unstable periodic orbits (UPOs). UPOs have been
rightly called as the skeleton of chaos. Control of
chaos can be divided in two main tasks: the first one
is the detection of an UPO, and the second one is the
stabilization of this detected UPO.
In literature, some papers have focussed on
control of chaos in robotics. The OGY method (Ott,
Grebogi and Yorke, 1990) and the delayed feedback
control (Pyragas, K., 1992) are the major approaches
to control or stabilize chaos in dynamic walking of
biped robots (Suzuki and Furuta, 2002; Sugimoto
and Osuka, 2004; Harata et al., 2009). While the
biped dynamics is highly nonlinear, the stability
analysis, if done, is usually based on a linearized
model. Morris and Grizzle (2005) used a restricted
Poincar map in order to determine exponential ly
statble periodic orbits. However, Manchester (2011)
and Manchester et al. (2011) used a transverse
dynamics of the nonlinear model for computing
inner estimates of the regions of attraction of limit
cycles. Several authors have used the linearized
model in order to investigate the stability of the
dynamic walking of biped robots (Sugimoto and
Osuka, 2005, 2007; Leines and Yang, 2011; Asano,
2011). Sugimoto and Osuka (2005, 2007) linearized
the nonlinear model around some given point (the
two legs are perpendicular to the walking surface).
They used an approximated Poincar map in order to
analysis stability of the PDW of the linear model.
Using the linearized Poincar map, they designed
also an optimal-control-based controller to stabilize
the fixed point of the PDW. However, the designed
controller was found to be valid for the nonlinear
model only for very shallow slopes. However, the
controller is not valid for the nonlinear model for
steeper slopes. The main cause is that the
characteristics (limit cycle, step period, etc.) of the

linear model are completely different to those of the


nonlinear model. As a result, we can emphasize that
the validity of the linearized model may become
questionable if the walking involves states that are
too far away from the operating point.
Motivated by interesting demonstrations of
chaos, we conduct this paper on the design of a
controller for the compass-gait biped robot in order
to control chaos in its PDW. Our developped
controller is based on the determination of an
analytical expression of the controlled Poincar map.
This is established from a close linear model
obtained by linearizing the nonlinear model around a
passive hybrid limit cycle for a given bifurcation
parameter. From characterictics point of view, the
close linear model is found to be almost identical to
the nonlinear model. Thus, we will show that the
degined controller is valid for the control of chaos in
the PDW of the nonlinear model.

PASSIVE COMPASS-GAIT
BIPED ROBOT

2.1 Compass-Gait Biped


In this paper, we model the passive compass-gait
biped robot following Goswami, Thuilot, and Espiau
(1998). Figure 1 shows the model. By setting
suitable physical parameters and initials conditions,
the compass-gait biped robot can exhibit PDW on a
gentle slope. The physical parameters are chosen as
listed in Table 1. In Figure 1, is the half-interleg
1
angle and is defined by : = ( s ns ) .
2

2.2 Impulsive Hybrid Dynamics


The compass gait model consists of two sets of
dynamics: a continuous dynamics of the swing
stage, and an impulsive transition dynamics that
occurs at the event of touchdown of the swing leg
with the ground. Let = [ ns s ] be the vector of
generalized coordinates. The impulsive hybrid
dynamics of the compass-gait biped robot is
described as follows:
T

( )

( )
( )
, ( ) = 0, ( , & ) < 0}

J ( )&& + H , & & + G( ) = Bu , if , &


+ = Re , & + = S e ( )& , if , &

= 21

(1)
(2)
(3)

Khraeif and Belghith, 2012) for further details of


each term. Expression (1) describes the dynamics of
the swing stage. However, relations in (2) express
the algebraic equations during the impulsive impact
stage. This occurs when the two impact conditions in
(3) are satisfied. is the impact surface and will be
defined as the Poincar section. u is the control
input applied at the hip: u = u H (see figure 1) .
The impulsive hybrid nonlinear dynamics (1)-(3)
can be equivalently rewritten in concise way like so:

x& = f (x) + g( x)u ,


x = h(x ) ,
+

Figure 1: Compass-gait biped robot on a slope.

= {x

with x = T
Table 1: Specification of simulation model
Symbol
Description
Value
a
Lower leg segment
0.5 m
b
Upper leg segment
0.5 m
l=a+b
Leg segment
1.0 m
m
Mass of leg
5.0 kg
mH
Mass of hip
10.0 kg
g
Gravitational constant
9.8m/s2

with J is the inertia matrix, H includes Coriolis


and centrifugal terms, G includes gravity forces, B
is the input matrix, Re is the renaming matrix of the
angular positions, S e is the reset matrix of the
angular velocities of the two legs, and represents
the impact section which is defined by the two
expressions in (3). Expressions of these matrices are
given in Appendix. See (Goswami, Thuilot and
Espiau, 1998; Spong and Bullo, 2005;, Gritli,

(a)

if x

(4)

(5)

if x

41

, 1 ( x) = 0, 2 ( x) < 0}

& T

] is the state vector.

(6)

2.3 Passive Walking Patterns


It is well known so far that the PDW of the compassgait biped robot exhibits a cascade of perioddoubling bifurcations from period-one gaits leading
to the formation of chaos. Recently, Gritli, Khraeif
and Belghith (2012) identified some new walking
patterns in the passive gait model of the compass
biped. They showed also that a cyclic-fold
bifurcation is generated. Figure 2 reveals the step
period of the bipedal locomotion with respect to the
slope angle . In Figure 2a, the blue bifurcation
diagram shows the period-doubling route to chaos
from period-one gaits. However, the red bifurcation
diagram reveals the period-three route to chaos
induced by a cyclic-fold bifurcation. Pink dashed
curve reveals the unstable period-one gaits. They are
born form the first period-doubling bifurcation.

(b)

Figure 2: Passive walking patterns of the compass-gait biped robot as the slope angle varies: (a) shows bifurcation
diagrams, and (b) reveals symmetric gaits.

However, green dashed curves express the unstable


period-three passive gaits. These gaits are born from
the cyclic-fold bifurcation. In the PDW of the
compass-gait biped robot, only the period-1 gaits are
symmetric. However, all other gaits are asymmetric.
In particular, a chaotic gait is asymmetric. Figure 2b
reveals the symmetric gaits with respect to the slope
angle. The blue gaits are stable, whereas the gaits
depicted in pink are unstable. These unstable
symmetric gaits are those created from the first
period-doubling bifurcation shown in Figure 2a.
In this paper, we interest only in the symmetric
gaits. Our main goal is to control chaos exhibited for
a given bifurcation parameter in the PDW of the
compass robot. The objective is then to obtain a
symmetric stable gait. For this subject, our major
contribution is to define an analytical expression of a
controlled Poincar map by establishing a linear
model that should be fairly close to the nonlinear
model. Thus, using the linearized controlled
Poincar map, we will design a controller in order to
stabilize an unstable fixed point of a chaotic gait.
Then, control of chaos can be treated in two key
steps. The first step is consists in identifying of a
symmetric unstable limit cycle (or symmetrically the
detection of the fixed point). The second step of the
control process of chaos lies in the stabilization of
this identified unstable symmetric limit cycle.
To solve this problem of stabilization of the
chaotic PDW, we will define a linear model for the
swing stage of the compass robot instead of the
nonlinear model (4). This allows us to determine an
analytical expression of the solution for the linear
differential equations contrary to the nonlinear case.
By comparing with the linear model linearized
around a given point by some authors (Sugimoto and
Osuka, 2005, 2007) and its drawbacks, we will
develop a linear model linearized around a passive
hybrid limit cycle of the nonlinear model. The next
section deals with this problem.

3.1

A CLOSE LINEAR MODEL AND


ITS CONSTRAINED
POINCAR MAP
Close Linear Model

It is quite obvious that the impulsive hybrid


nonlinear model of the PDW is characterized by a
passive hybrid limit cycle for each bifurcation
parameter such as the slope angle . Then, the
linear model, to look for, must imperatively give a

passive hybrid limit cycle fairly close to that of the


nonlinear model. Thus, the nonlinear model should
be linearized around this passive limit cycle. We
indicate in the next by d the desired passive limit
cycle which is adopted to be symmetric. The desired
passive limit cycle d is taken relying on Figure 2b.
For a given bifurcation parameter , we define then

d . Furthermore, we emphasize that a passive limit


cycle d is characterized by the step period d .
Then, the linearization of the nonlinear model
around d lies in the determination of n different
linear models M i , for i = 1,K, n , around different
points x 0 i of d . Each linear model will be defined
during a certain interval of time [t i1 t i ] with
i
t i = d . The linear model M i will be linearized
n
around the point
x0i
chosen such that

2i 1
x 0 i = x
d .
2n

Using the nonlinear model (4), the linear form


around the point x 0 i is defined by:
x& = Ai x + bi + Bi u

f (x )
Ai =
,
x x0i
Bi = g ( x 0 i ) .

with

for t i 1 t t i

bi = f ( x0i ) Ai x0i

(7)
and

The complete linear model which should be very


close to the nonlinear model is given by the
following system:
x& = A1 x + b1 + B1u

x& = A x + b + B u

i
i
i

x& = A x + b + B u
n
n
n

x + = h(x )

for t0 t t1
M
for t(i 1) t ti
M

(8)

for t( n 1) t
if

x ()

Here, is defined by expression (6). Moreover,


the instants t i , i = 0, K , (n 1) , are well-defined
from the passive limit cycle of the nonlinear model.
However, is the impact time for the linear model.
This impact time must be calculated numerically.
In fact, for an initial condition taken just after the
impact and that belongs to the desired limit cycle
d , this impact time should be (quasi) identical to

d . Furthermore, the two passive hybrid limit cycles


for the nonlinear and linear models should be very
close. However, for a different initial condition, the
impact time will be different to d .
In addition, it is necessary to have t n 1 < . Thus,
from this expression, we can determine the
maximum value of n that permits to reproduce the
behavior of the nonlinear model:
1
n<


(9)
1
d

Relying on bifurcation diagrams in Figure 2, we


have deduce that the possible maximum value of n
must be 6. This number of linear models is sufficient
such that the linear model (8) is fairly close to the
nonlinear model (4)-(6). Figure 3 proves this
conclusion. In fact, for a value of n different from
6, the difference between the two passive hybrid
limit cycles will be more clear with respect to the
decreases of n . Here, we have taken the admissible
maximum value of n in order to have two models
very close. Then, instead of using the nonlinear
model in order to control chaos, we will use the
linear model (8). An analytical expression of the
Poincar map of the linear model will be then
derived in the next subsection.

3.2

Controlled Poincar Map

Our main goal is to determine the analytical


expression of the Poincar map for the linear model
binding the state vectors right before the impact. We
assume that the control law u is constant during a

step of walk, in other word u (t ) = u , for all

t [t 0

] .

From the system (8), x(t 0 ) = x 0+ is the initial


condition right after the impact. In addition,
x() = x is the final condition of the swing phase
and so it is the condition right before the impact.
Then, we can write the following relation:
x 0+ = h(x 0 ) , where x 0 denotes the initial condition
right before the impact for a walk step. Then, x
will be the initial condition for the next step. In the
next of this paper, we will follow the following
notation:
- x k is the initial condition just before impact
-

for the k th step,


k is the walk period (impact time) of this k th
step,
u k is the controller applied in the k th step, it is
constant between (k 1) and k .

x k+1 will be the initial condition just before


impact for (k + 1) step.
th

Accordingly, we can show that x k+1 is mapped


from x k by means of the following controlled
Poincar map:

x k+1 = (x k , k , u k )

(10)

The analytical expression of the map is


defined by expressions (13)-(16). For a control law
u k , from an initial condition x k , and for an impact
time k , the Poincar map will give the state x k+1 .
Nevertheless, this state must belong to the impact
surface given by (6). Then, the two impact
conditions in (6) must be satisfied. These two
conditions are equivalent to:

1 ((xk , k , uk )) = 0

(11)

2 ((x , k , u k )) < 0

(12)

Expressions (10)-(12) represent the constrained


controlled Poincar map. This constrained map
defines the discrete-time model of the linear model
(8), and equivalently of the nonlinear model (4)-(6),
under the control law u k . We stress that the impact
Figure 3: Hybrid limit cycles for a 3-slope of the
nonlinear model (blue) and the linear model (pink).

time k depends closely on the state x k and also on


the control law u k . Indeed, if any tiny variation in

P(xk , k , uk ) = E(xk , k ) + F( k ) + G( k )uk

(13)

E(xk , k ) = e

(k d ) An+d Ai
n

i=1

h(xk )

(14)

d
d Ai

Ai
n
n
n

1
1
1
i= p
i=1
(Ap bp A( p1)b( p1) ) An1bn
A1 b1 + e
e
p=2

F( k ) = e

( k d ) An

(15)

n
d
d n Ai

Ai
n
n

n
(k d ) An
1
i= p
1
1
i=1
(Ap Bp A( p1) B( p1) ) An1Bn
G( k ) = e
A1 B1 + e
e
p=2

(16)

P(xk , k , uk ) 1 (xk , k , uk ) P(xk , k , uk ) 1 (xk , k , uk ) P(xk , k , uk )

DP(x* , * , u* ) = I 4
P(x , , u )
P(x , , u )

k
k
xk
k
k
k
k
k
k

P(xk , k , uk ) 1 (xk , k , uk ) P(xk , k , uk ) 1 (xk , k , uk ) P(xk , k , uk )

DQ(x , * , u* ) = I 4
P(x , , u )
P(x , , u )

k
k
uk
k
k
k
k
k
k

x k or u k occurs, the value of the impact time k


will change radically. Thus, k plays a fundamental
role in the stability of the compass gait.
Accordingly, we can note the following:
k = k (x , uk )

x k+1 = x k = x* . The

obtaining of the fixed point x * requires solving the


two transcendental equation (11) and (12). Therefore
a root finding algorithm should be applied. Once the
fixed point x * is located, its stability analysis may
be carried out by studying the local behavior of the
map near the fixed point. The next section studies
the stability of the fixed point x * . Then, the
linearized controlled Poincar map will be
established.

3.3

(18)

noting that by definition

(x , * , u * ) = x , the linearized controlled Poincar


map can be written as:

(20)

where DP(x* , * , u * ) is the Jacobian matrix of

impact time * . This fixed point can be obtained by


periodicity

x , ,u
* * *

(19)

u * = 0 . This fixed point x * depends closely on the


the

u * = 0 , and

(17)

xk+1 = DP(x* , * , u* )xk + DQ(x* , * , u* )uk

The stability of a passive hybrid limit cycle of


the gait model can be studied by analyzing the
stability of the fixed point x * of the map for

enforcing

with

x , ,u
* * *

Linearized controlled Poincar Map

Once the fixed point x* is located, its stability


analysis may be carried out by studying the
eigenvalues of the Jacobian matrix of the Poincar
map (10). Denoting xk = xk x* , u k = u k u * ,

with respect to x k , and DQ(x * , * , u * ) is the


Jacobian matrix of with respect to u k .
Relying on the first impact constraint (11) and
taking into account expression (19), these two
matrices are defined by expressions (17) and (18).
By calculating each term in these two matrices using
Eq. (10) and Eq. (11), DP and DQ can be obtained
in a straightforward manner.
The stability of a passive hybrid limit cycle, i.e.
for u k = 0 , can be studied by analyzing the stability
of the fixed point x* of the Poincar map . The
linearized Poincar map for an uncontrolled biped
robot u k = u k = 0 , is defined by the matrix DP .
The eigenvalues of

DP give the amount of

expansion or contraction near the fixed point x*


when the map is once iterated. Therefore, they
determine the asymptotic stability of the fixed point
and hence of its underlying passive hybrid limit
cycle. A sufficient condition for stability is that all
characteristic multipliers lie inside the unite circle.

4
4.1

CHAOS CONTROL

The

eigenvalues

of

the

Jacobian

matrix

DP(x * , * , u * ) are found to be: -1.2257; -0.1974;

Controller Design

0.1316; 0.0000. These values imply that the fixed


After the detection of the fixed point x* of an
unstable passive hybrid limit cycle, the second step
in the control procedure of chaos lies then in the
stabilization of this unstable fixed point using the
Linearized controlled Poincar map (20). Thus, our
intention is to design a controller u k = u k that
stabilizes the discrete linear system (20). Therefore,
we will adopt the following state feedback
controller:

u k = Kx k

(21)

with K 1 4 is the control gain.


Using a candidate function of Lyapunov:
T
(
V x k ) = x k Sx k , with S = S T > 0 , we have
established the following LMI:

(DP + DQ R )T

with R 1 4 .

DP + DQ R
>0
S

(22)

By solving the LMI (22) in S and R , the gain


matrix K of the control law (21) is defined by the
follwing expression: K = RS 1 . This gain of the
state feedback controller (21) ensures an asymptotic
stability of the fixed point x* of the constrained
Poincar map. Hence, by applying the control law:
u k = K (x k x* ) in the linear model (8), the passive
hybrid limit cycle which corresponds to the
stabilized fixed point x * is controlled. Accordingly,
chaos exhibited for a given bifurcation parameter in
the PDW of the compass-gait biped robot is well
controlled.

4.2

Numerical Simulations

After the design of a state feedback control law, we


will check the effectiveness of our controller
adopted for the control of chaos in the PDW of the
compass-gait biped robot. Basing on the expression
of the constrained Poincar map (10)-(12), and for
the chaotic attractor defined for a 5.2-slope, we
have detected the following unstable fixed point:
x* = [13.4839 - 23.8839 - 136.4288 - 105.692 ]

where the nominal step period is * = 0.76261 (s ) .

point x* is well unstable which will be stabilized by


means of the control law (21).
Using LMI toolbox of Matlab to solve (22), the gain
matrix K of the controller u k is given by:

K = [- 53.7222 8.3858 1.0798 - 9.4259] .


The
eigenvalues of the Jacobian matrix in the close loop
of the system (20) are: 0.1948; -0.1495; 0.0114;
0.0000. Hence, the fixed point x* of the constrained
Poincar map is well stabilized by introducing the
gain matrix K . Then, using this gain K , we have
applied the control law (21) to the original nonlinear
model (4)-(6) in order to control chaos.
Figure 4a shows the passive chaotic attractor.
The detected fixed point x* is for an unstable
hybrid limit cycle embedded into this chaotic
attractor. Fig. 4b reveals the controlled hybrid limit
cycle. Figure 5 shows both the variation of the step
period and the control torque command with respect
to the step number. It is obvious that after 8 steps,
the step period will be constant and then chaos is
well controlled. Furthermore, the control law is
remains constant to the value -0.0471Nm. In
addition, the maximal value of the torque is about
0.5Nm. Thus, we conclude that very low torques are
applied at the hip of the compass-gait biped robot.
Accordingly, these results reveal in fact that a very
low energy is provided for the biped in order to
stabilize its dynamic walking as it goes down an
inlined plan.

CONCLUSIONS AND FUTURE


WORKS

In this paper, we have proposed the control process


of chaos in PDW of the compass-gait biped robot.
Our control process is based mainly on the
derivation of a linear model which was found to be
very close to the original nonlinear model. The
linearization was done around a desired passive
hybrid limit cycle of the nonlinear model for a given
bifurcation parameter. This task conducted to obtain
an analytic expression of the controlled Poincar
map. Basing on this map and the impact conditions
of the compass robot, we have determined a

(a)

(b)

Figure 4: (a) Chaotic attractor for the passive dynamic walking on a 5.2-slope. (b) Controlled hybrid limit cycle.

(a)

(b)

Figure 5: Step period (a) and control law (b) as a function of step number for a 5.2-slope.

linearized controlled Poincar map. It was used in


order to discuss about stability of the fixed point or
systematically the hybrid passive limit cycle. Using
the linearized controlled Poincar map, we have
designed a state feedback control law in order to
stabilize the unstable fixed point. We have shown
the effectiveness of our controller in the control of
chaos using the original nonlinear model.
We hope apply our control strategy of chaos
developed in this paper to general hybrid systems
and then other biped robot in order to control chaos.
Our interest is also to find a control approach that
can stabilize the dynamic walking of the biped robot
for various slopes, and so independently of the
bifurcation parameter.

REFERENCES
Alexander, L.F., Robin, J.E., 2005. Control of chaos:
Methods and applications in engineering. Annual
Reviews in Control, vol. 29, pp. 3356.
Andrievskii, B.R., Fradkov, A.L., 2003. Control of Chaos:
Methods and Applications. I. Methods. Automation
and Remote Control, vol. 64, no. 5, pp. 673713.
Beck, L., 2009. Most famous robots (the humanoid robot).
http://sites.google.com/site/luisbeck007/humanoid.
Asano, F., 2011. Stability analysis of passive compass gait
using linearized model. In Proc. of the IEEE
International
Conference
on
Robotics
and
Automation, pp. 557-562.

Goswami. A., Thuilot, B., Espiau, B., 1998. A study of the


passive gait of a compass-like biped robot: symmetry
and chaos. International Journal of Robotics
Research, vol. 17, no. 12, pp. 12821301.
Gritli, H., Khraeif, N., Belghith, S., 2011. Falling of a
passive compass-gait biped robot caused by a
boundary crisis. In Proc. of the 4th Chaotic Modeling
and Simulation International Conference, pp. 155
162.
Gritli, H., Belghith, S., Khraeif, N., 2012. Intermittency
and interior crisis as route to chaos in dynamic
walking of two biped robots. International Journal of
Bifurcation and Chaos, vol. 22, no. 3.
Gritli, H., Khraeif, N., Belghith, S., 2012. Period-three
route to chaos induced by a cyclic-fold bifurcation in
passive dynamic walking of a compass-gait biped
robot. Communications in Nonlinear Science and
Numerical
Simulation,
DOI:
10.1016/j.cnsns.2012.02.034..
Grizzle, J., Abba, G., Plestan, F., 2001. Asymptotically
stable walking for biped robots: Analysis via systems
with impulse effects. IEEE Transactions on Automatic
Control, vol. 46, vol. 1, pp. 51-64.
Harata, Y., Asano, F., Taji, K., Uno, Y., 2009. Efficient
parametric excitation walking with delayed feedback
control. In Proc. of the IEEE/RSJ International
Conference on Intelligent Robots and Systems, pp.
29342939.
Kajita, S., Espiau, B., 2008. Legged robots. In B. Siciliano
& O. Khatib (Eds.), Springer handbook of robotics,
pp. 361389. Berlin: Springer.
Leines, M.T., Yang, J.-S., 2011. LQR control of an under
actuated planar biped robot. In Proc. of the IEEE
Conference
on
Industrial
Electronics
and
Applications, pp. 16841689.
Manchester, I.R., 2011. Transverse Dynamics and Regions
of Stability for Nonlinear Hybrid Limit Cycles. In
Proc. of the 18th IFAC World Congress, pp. 62856290.
Manchester, I.R., Tobenkin, M.M., Levashov, M, Tedrake,
R., 2011. Regions of Attraction for Hybrid Limit
Cycles of Walking Robots. In Proc. of the 18th IFAC
World Congress, pp. 5801-5806.
McGeer, T., 1990. Passive dynamic walking. International
Journal of Robotics Research, vol. 9, no. 2, pp. 6268.
Moon, J.S., Spong, M.W., 2011. Classification of periodic
and chaotic passive limit cycles for a compass-gait
biped with gait asymmetries. Robotica, vol. 29, no. 7,
pp. 967-974.
Morris, B., Grizzle, J.W., 2005. A restricted Poincar map
for determining exponentially stable periodic orbits in
systems with impulse effects: Application to Bipedal
Robots. In Proc. of the IEEE Conference on Decision
and Control, and the European Control Conference,
pp. 4199 4206.
Ott, E., Grebogi, C., Yorke, J.A., 1990. Controlling chaos.
Physical Review Letters, vol. 64, no. 11, pp. 1196
1199.

Pyragas, K., 1992. Continuous control of chaos by selfcontrolling feedback. Physics Letters A, vol. 170, no.
6, pp. 421428.
Spong, M. W., Bullo, F., 2005. Controlled symmetries and
passive walking. IEEE Transactions on Automatic
Control, vol. 50, no. 7, pp. 10251031.
Sugimoto, Y., Osuka, K., 2004. Walking control of quasi
passive dynamic walking robot "Quartet III" based on
continuous delayed feedback control. In Proc. of the
IEEE International Conference on Robotics and
Biomimetics, pp. 606611.
Sugimoto, Y., Osuka, K., 2005. Stability analysis of
passive dynamic walking focusing on the inner
structure of Poincare map. In Proc. of the
International Conference on Advanced Robotics, vol.
5, pp. 236241.
Sugimoto, Y., Osuka, K., 2007. Hierarchical implicit
feedback structure in passive dynamic walking. In
Proc. of the IEEE/RSJ International Conference on
Intelligent Robots and Systems, pp. 22172222.
Suzuki, S., Furuta, K., 2002. Enhancement of stabilization
for passive walking by chaos control approach. In
Proc. of the Triennial World Congress of the IFAC,
vol. 15.

APPENDIX

mb 2
mlb cos( s ns )
J ( ) =
,

mlb
cos
(

)
m H l 2 + m(l 2 + a 2 )
s
ns

0
mlb&s sin(s ns )
H ,& =
,
&
0
mlbns sin(s ns )

mb sin ( ns )

1
G ( ) = g
, B= ,

1
(m H l + m(a + l )) sin ( s )
0 1
1
Re =
, S e ( ) = Q p ( ) Qm ( ) ,
1 0

( )

p4
p p2
Q m ( ) = 1
, Q p ( ) =

p6
0 p3
p1 = p 3 = mab ,

p5
,
p 7

p2 = p1 + (mH l 2 + 2mal ) cos( s ns ) ,

p4 = p6 p7 , p5 = (m + m H )l 2 + ma 2 + p 7 ,
p6 = mb 2 , p 7 = mbl cos( s ns ) ,

1 ( ) = l (cos( s + ) cos( ns + )) ,
2 , & = l &s sin( s + ) + &ns sin( ns + ) , where

( ) (

l = a+b .

You might also like