Dual Arm and Multi-segment Spine Motion Control

for Assistive Humanoid Robots

Branislav Borovac
1
, Mirko Rakovi
1
, Sran Savi
1
and Milutin Nikoli
1
1
Faculty of Technical Sciences, University of Novi Sad, Novi Sad, Serbia
rakovicm@uns.ac.rs, savics@uns.ac.rs, milutinn@uns.ac.rs, borovac@uns.ac.rs


Abstract This paper presents the problem of simultaneous
motion control of multi-segment spine and dual arm robot.
Firstly, the mechanical design of SARA robot (Socially
Acceptable Robot Assistant) is described. SARA is designed to
have a spine with 6 degrees of freedom (DOFs) and two arms
with 7 DOFs each. The proposed mechanical design ensures good
mobility and anthropomorphic motion of robot. Afterwards,
kinematic parameters and dynamic model of SARA robot are
introduced and simultaneous motion control of spine and arms is
proposed. During the design of the control law nonlinearity and
coupling of robot system has been taken into consideration.
Therefore, we proposed the robust nonlinear control law that can
compensate external disturbances. The control law is verified for
reaching and moving of object with unknown mass located at the
table in front of the robot.
Keywords-assistive humanoid robots; object reaching;
nonlinear control
I. INTRODUCTION
Nowadays number of robots and their presence in human
environment is constantly increasing. Modern industrial robots
have been designed to cooperate with humans in factories [1].
Significant results have been also achieved in fields of service
robotics [2] and assistive technologies [3, 4]. In all these
applications it is important to make robots safe for people in
their surrounding and to provide design and functionality of
robots that would make people feel comfortable in their
presence. For successful human-robot cooperation it is
necessary to fit robot design and its functionality into
established human social frames and protocols. Robot
prototype, presented in this paper, is being developed at the
University of Novi Sad, Faculty of Technical Sciences, and it is
conceived as a mobile platform for research of socially
acceptable robot behavior. This research includes bilateral
interaction between humans and robots, speech recognition and
synthesis, gesticulation, emotion recognition and expression,
and robot behavior in unstructured environment. Since the
humanoid robot is not manufactured yet, the experiments of
integrated system including industrial robot, stereo vision
system and speech recognition and synthesis system with
cognitive dialog management system have been realized [5].
Manipulation with various objects is one of the basic tasks
for robots working in humans environment. Since human arms
are among the most advanced manipulation systems in nature,
they were used in our work as inspiration for mechanical
design of robot arms. Therefore robot arm design with 7 DOFs,
inspired by human arm anatomy, is proposed in this paper in
order to make robot arm movements as much as possible
efficient and similar to human arms movements.
There are a lot of successful realizations of anthropomor-
phic robot arms. Some state-of-the-art solutions can be seen in
robots DLR-Lightweight-Robot-III [6], ARMAR [2] and
ROBONAUT [7]. These robot arms are actuated with electrical
DC motors and have anthropomorphic kinematic structure with
7 DOFs, which makes them kinematicaly redundant. In some
solutions of robot arm design pneumatic actuators are used [8].
Artificial pneumatic muscles are intrinsically compliant, due to
the compressibility of air, and have great mass/payload ratio
[9] but they need compressed air.
Most of humanoid robots are developed with rigid torso,
which makes their movement unnatural and constrained [10].
There are only few realizations of flexible robot spines. Some
of them are robots KENTA [10], UCL [11] and KOTARO [12].
These solutions are biologically inspired and copy the anatomy
of human spine. Spines of these robots are tendon driven which
makes them more compliant and increases safety in case of
physical contact with humans. They also have very low
backlash. Flexible spine is able to absorb mechanical shock or
impact. Disadvantages of these solutions are lower accuracy
and repeatability. Tendons are susceptible to stretching due to
material aging and creep. This can badly affect system
performances. Another common solution of robot spine design
is based on 2 DOFs tendon driven differential drive. This
design can be seen in robots iCUB [13] and ARMAR [14]. It is
very compact and has high stiffness and accuracy. Since the
whole trunk motion is realized in a single joint it does not look
natural and anthropomorphic.
In this paper a lumbar spine structure with 6 DOFs,
proposed in [15], will be used. Spine consists of three serially
connected segments, each having 2 DOFs, rotation and lateral
flexion. Such movable spine structure increases workspace of
robot arms and significantly contributes to anthropomorphic
appearance of robot upper-body motion.
Since every model is only approximation of real system it is
expected that the real robot will have slightly different
parameters compared with corresponding CAD model. Besides
parameters uncertainties robot will be exposed to external
disturbances caused by collisions or unknown payload masses.
Therefore it is necessary to design control algorithm which is
robust to parameter uncertainties and disturbances and which is
capable of achieving accurate trajectory tracking.
The authors would like to thank the company Dunkermotoren for its
support and equipment donation.
Paper is organized as follows. Basic description of robot
upper-body mechanical design is given in chapter II. In chapter
III kinematic and dynamic model of robot are presented
followed by description of control algorithm. The simulation
results for trajectory tracking problem in presence of parameter
uncertainties and external disturbance is presented in chapter
IV. The paper concludes with final remarks in chapter V.
II. MECHANICAL DESIGN
Mechanical design of biologically inspired anthropomor-
phic robot arms and multi-segment lumbar spine structure is
described in this chapter. In Fig. 1. CAD model of robot upper-
body is shown.
Kinematic structure with 7 DOFs, closely approximating
human arm, has been adopted for robot arm. Shoulder has 3
DOFs and consists of three revolute joints whose axes of
rotation intersect in one point.

Figure 1. CAD model of robot upper-body
All DOFs of shoulder are actuated with brushless DC
motors (Dunkermotoren BG series motors), with integrated
multi-turn digital absolute and incremental encoders to
measure joint positions and velocities. Actuators for first and
second DOF are placed in robot torso to reduce masses and
moments of inertia of arm movable parts. Elbow joint has one
DOF while wrist has 3 DOFs whose axes of rotation intersect
in one point. Elbow flexion and wrist rotation are also actuated
with Dunkermotoren brushless DC motors with integrated
encoders, same as in the shoulder. Wrist flexion and abduction
are actuated with linear actuators. Axes of rotation of these two
DOFs intersect in the center of universal joint which is fixed on
the central rod [16]. These linear actuators have integrated
linear potentiometers for position measurement. Details about
arm mechanical design and mechanical parameters can be
found in [17].
Proposed lumbar spine design has three segments with 6
DOFs in total. This kinematic structure is able to realize
movements of lateral flexion and rotation in each segment.
Humans perform flexion/extension movements in hip up to
45, not in spine. Therefore implementation of this movement
was not considered during design phase. For actuation of
lateral flexion joints brushed DC motors are used. These
motors have integrated planetary gearheads and incremental
encoders for angle and angular velocity measurement. Power
transfer from motor shaft to joint shaft is realized with spur and
worm gears. Worm gears are used to achieve compact design
and self-locking in any posture. Thus, increased reliability and
preservation of upper body posture without additional energy
has been achieved. For additional height reduction, specially
profiled worm wheel has been designed. Torsion joints are
realized as specially designed inseparable axial bearings. For
actuation of torsion joints linear actuators are used. These
actuators consist of DC motor with spindle drive and have
embedded linear potentiometers. In Fig. 2. CAD model of
robot lumbar spine is shown.

Figure 2. The lumbar spine CAD model: (left) in upright position; (right)
flexion (30) +torsion (45)
Total height of three incorporated segments is 201 mm,
diameter of the base is 150 mm, and the spines mass is 6 kg (2
kg per segment). The overall lateral flexion range is 30, and
torsion range is 45 (10 and 15 in each joint).
III. MOTION PLANING AND CONTROL
A. Kinematic and Dynamic model of Robot
The software used for modeling [18] is capable of forming
dynamic model of humanoid robots and of simulating robots
with one or more open and closed kinematic chains. Segments
are connected with joints having one rotational degree of
freedom, each. Kinematic structure of robot model is shown in
Fig. 3.
First segment of spine has been adopted as a base link. The
state vector q, consists of the position (x, y, z) and orientation
of the base link (, , ) and the n-elements vector consisting
of relative angles between two adjacent links, where n
represents the number of the mechanism links (base link
excluded). So, the total number of the mechanism DOFs is
6+n:

| |
T
1 2 n
= x, y, z,, , , q , q ,,q q

Figure 3. Kinematic structure of robot upper-body
Kinematic model presented in Fig. 3, has 2 kinematic
chains. The first chain is comprised of the trunk and left arm,
second of the trunk and right arm. The trunk is modeled as
single rigid segment. In this study trunk is not moving,
(position and orientation of trunk is constant) i.e. the first 6
values in vector q are not changing.
To realize the arm movement the appropriate driving
torques should be applied at each DOF. Since driving torques
cannot be applied to the first 6 DOFs the vector of driving
torques is defined in the following way:

| |
T
1 2 n
= 0, 0, 0,0, 0, 0, , ,,
The dynamic model of humanoid robot is (in accordance
with the defined vectors q and ):

( ) ( )
( ) ( ) ( )

= +
0
0
= H q q + h q,q
h q,q C q,q G q
(1)
where is vector of driving torques, q is vector of generalized
coordinates, H(q) is symmetric, semi-definite inertia matrix,
C(q,q) is vector of moments due to Coriolis and centrifugal
forces and G(q) is vector of moments caused by gravitational
forces. The overall number of DOFs is 26 (7 DOFs of each
arm, 6 DOFs for spine and additional 6 DOFs representing
position and orientation of basic link).
B. Motion planning
The motion planning is based on setting the target position
to which the robot should move its hand linearly. When the
target position is outside the arm workspace, the trunk moves
so that robot can reach the target position. The target position is
a 6 components vector that determines the desired position and
orientation of robot end-effector. In the case of object reaching
task the end-effector is associated with the hands. For motion
whose realization encompasses trunk, the end-effector is
attached to the last segment of the trunk.
Let us observe instantaneous position of the end-effector
coordinate frame O
A
attached to the hand and position of target
coordinate frame O
B
to which we want to bring it. To achieve
smooth motion when new target position is set, the velocity
A
s of O
A
has to depend on velocity at initial time
0
p
t . Now, we
will consider situation, when arm movement starts. Initial
velocity
0
A
s of the coordinate frame O
A
can be of arbitrary
intensity and direction. Let
A
r and
B
r denote position vectors
of coordinate frames O
A
and O
B
:
| |
T
T
A A A Ax Ax Ax Ax Ay
Az
p p p u ( = =

r p o
| |
T
T
B B B Bx Bx Bx Bx By
Bz
p p p u ( = =

r p o
Vector
A
r depends on joints coordinates, while
B
r is desired
goal position, while vector ( )
A
t s denotes desired velocity and
have six components:
| |
T
T
A A A Ax Ay Az Ax Ay Az
v v v e e e ( = =

s v
which correspond to desired linear and angular hand velocity.
Also, current intensity of linear
int
v

and angular
int
e

desired
velocities should be determined. To do this, goal velocities
int
kr
v
and
int
kr
e

should be set. By comparing
int
v and
int
e with
int
kr
v
and
int
kr
e

corrections of desired velocities are determined as:

int int int int int
int int int int int int
int
int int int int int
int int int int int int
int




kr
kr
kr
kr
v dv if v v dv
v v dv if v v dv
v otherwise
d if d
d if d
otherwise
e e e e e
e e e e e e
e
+ <

= > +

+ <

= > +

(2)
Values of
int
dv and
int
de are 0.15m/ms and 0.15rad/ms
respectively. This way, gradual change of desired linear and
angular velocities of frame O
A
is ensured. Values
int
kr
v and
int
kr
e
must be specified within predefined boundaries ( |
int max
0
kr
v v e
and ( |
int max
0
kr
e e e . When vector
B
r and velocities
intensities
int
v and
int
are known, vectors
e
ort
p and
e
ort
o
representing orts of vectors
e B A
= p p p and
e B A
= o o o
should be calculated. To ensure continuous and smooth desired
velocity change, the following expression has been introduced:
( ) ( ) ( ) ( ) | |
0
int int
1-
T
A A e e
t b t b t v e = + s s p o (3)
Coefficient b is not constant over time, and has to be
determined to ensure gradual change of ( )
A
t s from initial
velocity
0
A
s . The coefficient b is determined from equation:

( )
0
- /
300
p
t t dt
b satlin
| |
| =
|
\ .

Term satlin denotes linear saturation function. In Fig. 4 is
shown change of coefficient b (from 0 to 1) if
0
1
p
t s = .

Figure 4. The change of coefficient b when tp0 is 1s
The coefficient b is 1 after 300 iterations, i.e. since dt=1ms,
after 0.3s. If b=0, from eq. (3) follows that ( )
A
t s is equal to
initial velocity
0
A
s . As time is passing, coefficient b is rising till
value 1 has been reached. When b=1, velocity
A
s is:
( ) | |
int int
T
A e e
t v e = s p o
In order to ensure smooth deceleration of arm, the intensity
of linear and angular velocity
int
v and
int
should be gradually
decreased when O
A
is close enough to the target position O
B
.
Decrease of the intensity of linear and angular velocity occurs
when the conditions
int
3
e
v > p and
int
3
e
e > o are met. For
example, if the linear velocity of hand is
int
0.1m/s v = ,
deceleration of arm begins when the distance between O
A
and
O
B
is less than 0.033m. During deceleration, the intensity of
velocities
int
v and
int
e are determined from the equation:

int
int
3
3
e
e
v
e
=
=
p
o
.
Since it is known how to determine the velocity ( )
A
t s , now
the desired angular velocity of the arm joints
d
q should be
determined. This can be done with the following equation:
( ) ( ) ( )
A A
t t
+
=
d
J q q s (4)
where
A
+
J is the Moore-Penrose pseudo-inverse Jacobian
calculated for O
A
. Vectors q are current joint angles. In
equation (4) the pseudo-inverse Jacobian is used because arms
are redundant. Desired joint angles
d
q and desired angular
accelerations
d
q are obtained by differentiation and integrating
d
q calculated from eq. (4).
C. Feedback linearization and sliding mode control
Humanoid robot described with (1) is multivariable,
coupled and strongly nonlinear system. In order to cancel
system nonlinearities feedback linearization has been used. To
apply feedback linearization it is convenient to represent robot
dynamic equation (1) in state space in controllability canonical
form [19]:
( ) + = = +
(-1) (-1)
0
q f q,q b(q)u H (q)h (q,q) H (q) , (5)
,
d
dt
( ( (
= = =
( ( (

q q
q q
q
X X
f(X)+b(X)u
, (6)
where u= is a vector of driving torques, q is output vector of
joints angles, X is a state vector and f(X) and b(X) are nonlinear
functions of states.
For this method to be successful it is necessary that system
parameters are known and that system model is precise. By
choosing next control law:

1
( )

= b v f , (7)
nonlinearities are cancelled and state space model (5) is
reduced to a system of double integrators:
= q v. (8)
Pole placement method is used to obtain desired dynamics
of the system. If we consider a problem of trajectory tracking,
by choosing vin the form:
=
d 1 0
v q K e K e, (9)
where e(t) is tracking error, exponentially convergent tracking
is obtained [19]. Matrices K
1
and K
0
must be Hurwitz gain
matrices. Gains K
1
and K
0
are chosen to provide desired
dynamic behavior of the system.
In reality system model is never precise and it is only an
approximation of real system. For that reason it is necessary to
modify control law to make it robust to parameter uncertainties
and disturbances. In this paper robustification is done by using
sliding mode control. Time-varying hypersurface S(t) in the
space is defined by equation s(t,x)=0 where:
0 = + = s e e . (10)
In eq. (10) is diagonal Hurwitz matrix. Trajectory tracking
problem x(t)=x
d
(t) can be considered as a problem of re-
maining on surface S during the motion [20]. Control law (9) is
equivalent continuous control (v=u
eq
) that would keep
trajectory on hypersurface defined by eq (10) if the system
model were precise. To keep the zero value of s control law
must satisfy so-called sliding condition:

2
1 d

2 dt
s s s
, (11)
where is strictly positive constant that governs the reaching
time [20]. Expression for sliding mode control is now given as
follows:
( ) sat / =
d 1 0
u q K K e K s e , (12)
( )

( sat / ) = +
d 1 0 0
H q K K e K s h e , (13)
where K is gain of discontinuous term and

H ,

0
h are
calculated on basis of a known model. Saturation function has
been used instead of nonsmooth sign function to eliminate
chattering effect [21]. This is done by approximating sign
function in a thin boundary layer around S surface:
( )
( )
/ , / 1
sat /
sgn / , / 1 / 1
u u
u
u u u
s

< v >

s S
s
s s s
, (14)
where is a thickness of a boundary layer.
In (12) constant sliding mode gain K was used. K must be
large enough to compensate all disturbances and system
uncertainties. This means that most of time, in absence of
disturbance, K will be oversized which reinforces undesirable
chattering effect. So it is desirable to find an expression for
adaptive sliding mode gain K. Proposed solution is based on a
constant disturbance estimator. Feedback linearized system in
presence of disturbance has the following form:
= + q u d, (15)
where d is assumed to be constant disturbance vector. Extended
state space model is defined as:
, , = = =
1 2 3
x q x q x d, (16)

0 1 0 0
d
0 0 1 1
dt
0 0 0 0
( ( ( (
( ( ( (
= +
( ( ( (
( ( ( (

1 1
2 2
3 3
x x
x x u
x x
, (17)
| | 1 0 0
(
(
= =
(
(

1
1 2
3
x
y x x
x
. (18)
State space model of state and disturbance estimator is
given with following expression:

d

( )
dt
= + + X AX Bu L y CX , (19)

0 1 0 0
d
0 0 1 1 u ( )
dt
0 0 0 0
( (
( ( (
( (
( ( (
= + + ( (
( ( (
( (
( ( (

( (

1
2
3
q q
l
v v l q q
l
d d
, (20)
where qis a vector of estimated joint positions, v is a vector
of estimated joint velocities while d is a vector of estimated
disturbances. Gains of corrective factor L are chosen to make
poles of estimator three times faster than poles of process.
Sliding mode control law (12) can now be modified. Estimated
disturbance is added in eq. (12). Estimated disturbance acts as
an integral term and eliminates the steady-state error. Integral
anti-windup is implemented by restricting the derivative of
estimated disturbance as follows:

( ) 3 min max max
min
max min
l , ( d ,d ) ( d ( ) 0 )
d
( d ( ) 0 )
dt
0,( d ( ) 0 ) ( d ( ) 0 )

e v > . <

v s . >

> . > v s . <

q q d d q q
d
d q q
d q q d q q
. (21)
where d
min
and d
max
are allowed boundary values of disturbance
which define the region where anti-windup is inactive. An
expression for adaptive sliding mode gain K is proposed:
= + K d , (22)
which helps in eliminating chattering. Sliding mode control
law with disturbance estimator is now given with the following
expression:
( ) sat / u =
d 1 0
u q K K e K s e d, (23)
( )

( sat / ) u = + 0
d 1 0
e H q K K e K s d h . (24)
IV. SIMULATION RESULTS
The proposed motion planning and control algorithm for
motion control of multi-segment lumbar spine and dual arm
system has been tested through a numerical simulation in
presence of parameter uncertainties and external disturbance.
An example of reaching an object and moving an object to
another position has been simulated. The final position of an
object can not be reached with the rigid trunk. This simulation
shows how movable spine increases robot workspace and
contributes to anthropomorphic look of robot motion.
Initial pose of robot is shown in Fig. 5. where trunk is in
upright position with arms hanging beside the torso. At the
beginning of simulation the robot moves up its left hand in
order to avoid collision with the table. When robot hand is
3.3cm distant from the target position, a new position for gras-

Figure 5. Starting pose of the robot and position of the object on the table in
front of the robot

Figure 6. Robot is reaching an object with its left hand
ping an object is set. After the robot reaches the position for
grasping the motion of left arm is continued with object
attached to robots hand (Fig. 6.). The mass of the object is
near 1kg, but this information is not available to motion control
system. The green line in Fig. 6. represents the trajectory of
end-effector.

Figure 7. Robot is moving an object to another position on the table by
simultaniously moving trunk and left arm
The final position of object is set at far left part of the table.
That position can not be reached without moving the trunk.
Thus, while moving an object on destination the trunk is
moving simultaneously with arms, so that the robot is able to
reach the final position. When final position is reached and
object is placed back on the table, the simulation is finished.
V. CONCLUSION AND FUTURE WORK
In this paper mechanical design of humanoid robot upper-
body, consisting of two arms and lumbar spine, is presented.
Further, the motion planning and robust control algorithm is
introduced. Presented control law is a combination of
feedback linearization and sliding mode control with
implemented state and disturbance estimator. Anti-chattering
technique based on constant thickness boundary layer has
been implemented.
Efficiency of proposed algorithm was tested by a numerical
simulation for reaching an object and moving it to another
position. The destination position is such that the robot has to
move the trunk to be able to reach the final target position.
Results have shown that this control law achieves satisfactory
tracking accuracy since the end-effector motion is linear. Also,
the control law successfully deals with parameter uncertainties
and disturbance. Biologically inspired mechanical design
provides operating range and functionality similar to humans.
Presented multi-segment lumbar spine increases workspace of
robot arms and contributes to natural appearance of motion.

In further work, an additional joint will be added in a spine
base to allow flexion and extension of the robot spine. Models
of actuators and gears with actuator saturations will be also
included in overall dynamic model. In this paper, joints are
driven by applying the driving toques. After expanding the
overall model of the robot with model of actuators and robots,
the control algorithm will be adjusted since the joints will by
driven by applying appropriate motor voltage.
ACKNOWLEDGMENT
This work was funded by the Ministry of education and
science of the Republic of Serbia under contract III44008 and
by Provincial secretariat for science and technological
development under contract 114-451-2116/2011.
REFERENCES
[1] http://www.abb.com/cawp/abbzh254/8657f5e05ede6ac5c1257861002c8
ed2.aspx, access time 29.04.2013.
[2] A. Albers, S. Brudniok, J. Ottnad, ARMAR III-Design of the Upper
Body, Proc. IEEE-RAS 6th International Conference on Humanoid
Robots, Genova, Italy, pp. 308-313, December, 2006.
[3] E. Wade, A. Parnandi, R. Mead, M. Matari, Socially Assistive
Robotics for Guiding Motor Task Practice, PALADYN Journal of
Behavioral Robotics, Vol. 2, No. 4, pp. 218-227, Dec, 2011.
[4] H. Iwata, S. Sugano, Design of Human Symbiotic robot TWENDY-
ONE, Proc. IEEE International Conference on Robotics and
Automation 2009, Kobe, Japan, pp. 580-586, 12-17 May, 2009.
[5] J. Tasevski, M. Nikoli, D. Mikovi, Integration of an Industrial Robot
with the System for Image and Voice Recognition, Serbian Jour of
Electrical Engineering, vol. 10, no. 1, pp. 219-230, February 2013.
[6] O. Eiberger, W. Friedl, B. Baml, G. Hirzinger, ,,A Humanoid Two-
Arm System for Dexterous Manipulation, Proc. IEEE-RAS 6th
International Conference on Humanoid Robots, Genova, Italy, pp. 276-
283, Dec, 2006.
[7] M. A. Diftler, R. O. Ambrose, S. M. Goza, K. S. Tyree, E. L. Huber,
Robonaut Mobile Autonomy: Initial Experiments, Proc. IEEE
International Conference on Robotics and Automation 2005, Barcelona,
Spain, pp. 1437-1442, April, 2005.
[8] I. Boblan, A. Shulz, A Humanoid Muscle Robot Torso with
Biologically Inspired Construction, Proc. 41st International Symposium
on Robotics (ISR) and 6th German Conference on Robotics
(ROBOTIK), Munich, Germany, pp. 1-6, 7-9 June, 2010.
[9] F. Daeden, D. Lefeber, Pneumatic Artificial Muscles: actuators for
robotics and automation, European journal of Mechanical and
Environmental Engineering, vol. 47, pp. 10-21, 2000.
[10] I. Mizuuchi, R. Tajima, T. Yoshikai, D. Sato, K. Nagashima, M. Inaba,
Y. Kuniyoshi and H. Inoue, The Design and Control of the Flexible
Spine of a Fully Tendon-Driven Humanoid Kenta, Proc. of the 2002
IEEE/RSJ International Conference on Intelligent Robots and Systems,
Lausanne, Switzerland, pp. 25272532, 2002.
[11] I. Mizuuchi, M. Inaba, H. Inoue, A Flexible Spine Human-Form Robot
Development and Control of the Posture of the Spine, Proc. of the
2001 IEEE/RSJ International Conference on Intelligent Robots and
Systems, Maui, Hawaii, pp. 2099 2104, 2001.
[12] I. Mizuuchi, T. Yoshikai, Y. Sodeyama, Y. Nakanishi, A. Miyadere, T.
Yamamoto, T. Niemela, M. Hayashi, J. Urata, Y. Namiki, T. Nishino,
M. Inaba, Development of Musculoskeletal Humanoid Kotaro, Proc.
of the 2006 IEEE International Conference on Robotics and Automation,
Orlando, Florida, pp. 8287, 2006.
[13] W. M. Hinojosa, N. G. Tsagarakis, G. Metta, F. Becchi, G. Sandini and
D. G. Caldwell, Performance Assessment of a 3 DOF Differential
Based Waist joint for the iCub Baby Humanoid Robot, Proc. The
15th IEEE International Symposium on Robot and Human Interactive
Communication (RO-MAN06), Hatfield, pp. 195-201, 6-8 Sep, 2006.
[14] C. Sander, T. Soworka, A. Albers, Design of a New Torso-Joint for the
Humanoid Robot ARMAR, Journal of Mechanical Engineering and
Automation, 2(4), pp. 58-64, 2012.
[15] M. Peni, Lumbar Spine Development for Humanoid Robots,
Zbornik radova Fakulteta tehnikih nauka, Vol. 27, No. 14, pp. 3091-
3094, 2012 (in Serbian).
[16] D. Krklje, M. Nikoli, S. Savi, L. Na, K. Babkovi, Humanoid robot
wrist prototype, 48. International Conference on Microelectronics,
Devices and Materials MIDEM, Otoec, Slovenia, pp. 255-260, 19-21
September, 2012.
[17] S. Savi, M. Juroevi, Design of an anthropomorphic robot arm, 56.
ETRAN, Zlatibor, Serbia, pp.1-4, 11-14 June, 2012 (in Serbian).
[18] V. Potkonjak, M. Vukobratovi, K. Babkovi, B. Borovac, General
model of dynamics of human and humanoid motion: feasibility,
potentials and verification, Int. Jour. of Humanoid Robotics, Vol. 3,
No. 2, pp. 21-48, 2006.
[19] J. E. Slotine, W. Li, Applied nonlinear control, 1
st
ed., Englewood
Cliffs, New Jersey, Prentice Hall, 1991.
[20] K. D. Young, V.I. Utkin, . Ozgner, A Control Engineers Guide to
Sliding Mode Control , IEEE Transactions on Control System
Technology, vol. 7, num. 3, pp. 328-342, May 1999.
[21] T. V. M. Nguyen, Q. P. Ha, H. T. Nguyen, A Chattering-Free Variable
Structure Controller for Tracking of Robotic Manipulators, Proc. of
The Australasian Conference on Robotics and Automation (ACRA
2003), Brisbane, Australia, pp. 1-6, 1-3 Dec, 2003.