You are on page 1of 6

2010 IEEE/ASME International Conference on

Advanced Intelligent Mechatronics


Montral, Canada, July 6-9, 2010

A Mechatronics Approach to Legged Locomotion


Victor Ragusila and M. Reza Emami

Abstract This paper describes a mechatronics methodology


that allows the design of systems that are simple and intuitive to
analyze and control. The primary application of this
methodology is the design of a leg mechanism for a running
robot. The approach first divides the system behaviour into
several phases, and then finds Simple to Control Mechanisms
(SCMs) that can achieve the desired performance in each
phase. The mechanical element of the system is designed to
emulate these SCMs. It is shown that simple controllers
designed for the SCMs of the running robot leg can provide the
desired system behaviour.

I. INTRODUCTION

he notion of mechatronics has evolved to a systemic


design paradigm for creating synergy and providing
catalytic impacts on discovering simpler solutions to
traditionally-complex problems. The physical artifacts of
such a design philosophy, often referred to as mechatronic
systems, demonstrate a seamless integration of mechanical,
electrical and software constituents, in a sense that their
characteristics are all specified concurrently during the
design phase. Concurrent engineering of such
multidisciplinary systems, however, is no trivial task, for it
most likely results in a large number of objective and
constraint functions that must be taken into account
simultaneously with a great number of design variables.
Should one follow a formal optimization approach, the
multi-objective constrained optimization problem with large
number of functions and variables can be quite challenging.
Hence, practical multiattribute concurrent design approaches
are required for mechatronic systems.
A number of concurrent engineering approaches have
been suggested for mechatronic systems, which address the
challenges of multi-objective constrained optimization
problems. For instance, a concurrent system evaluation
model is proposed in [1], based on three indexes of
intelligence, flexibility, and complexity, and it is formulated
using t-norm and mean operators. Another evaluation model
is suggested in [2] based on Mechatronics Design Quotient
(MDQ), where a nonlinear fuzzy integral is used for the

Victor R. Ragusila is a graduate student in the Space Mechatronics


Group with the U. Toronto Institute for Aerospace Studies, Toronto,
Ontario M3H5T6 Canada (e-mail: victor.ragusila@utias.utoronto.ca).
M. R. Emami (corresponding author) is the director of the Space
Mechatronics group and the coordinator of the Aerospace Undergraduate
Laboratories at the University of Toronto Institute for Aerospace Studies,
Toronto, Ontario M3H 5T6 Canada (e-mail: emami@utias.utoronto.ca).

978-1-4244-8030-2/10/$26.00 2010 IEEE

aggregation of design criteria. An alternative concurrent


design methodology is presented in [3], introducing the
notion of Linguistic Mechatronics (LM), where the
subjective aspects of multidisciplinary systems design are
captured through fuzzy operators, along with the quantitative
system performances using a holistic system modeling based
on bond graphs. Another notable effort is the Design for
Control (DFC) methodology [4], which prescribes that
control parameters be designed concurrently with the
structure parameters, so that by simplifying the underlying
system dynamics simple controllers can be employed
successfully. In addition, a number of ad hoc techniques for
mechatronics have been reported in the literature, which are
innovative design schemes for specific applications such as
robotics (e.g., see [5]).
This paper details a concurrent design approach to the
development of a new leg mechanism for robotic systems.
Robot locomotion can be divided into wheeled and legged
strategies. While wheeled robots are arguably simpler and
cheaper to build and control, their legged counterparts
present a number of unique advantages, such as ability to
better negotiate obstacles, choose the footholds that offer the
best stability, or navigate tight areas [6]. A wheeled robot
has to be able to negotiate the worst part of a given path,
whereas a legged robot can choose the best footholds on the
same path [7]. Further, legged robots offer the possibility of
studying the walking and running behaviours of animals or
humans. General control strategies and mechanical designs
inspired by biological systems are already implemented in
robotics [8], [9]. Theories developed to study human running
can be tested on robots, offering a simpler and more
repeatable testing environment [8].
Important advances have been made in legged locomotion
amongst the research and industrial communities. To name a
few, Cornell Biped and Denise bipeds walk as efficiently as
humans do [10], and ASIMO [11] and WABIAN [12] can
perform almost the entire range of human motions.
However, no humanoid robot can run nearly as fast as
humans; the fastest running robot (at 5.7 m/s) is the MITs
Monopod, built by Raibert in the 80s, with an efficiency far
from ideal [13]. A fast running gait for legged robots is a
challenging but desirable goal, which is hoped to be
achieved
through
innovative
concurrent
design
methodologies.
Section II of the paper outlines the mechatronics
approach, as well as its application to a running robot named
Sprinter. In section III the proposed leg mechanism is
analyzed and synthesized. Section IV discusses the control

824

strategy, and section V presents simulation results


supporting the system design and control scheme. Some
concluding remarks are made in section VI.
II. THE UNDERLYING MECHATRONICS APPROACH
The proposed mechatronics approach aims at designing
systems that are easy to control and function in an intuitive
and transparent way. The approach consists of seven stages,
as seen in Figure 1, on the left-hand side. On the right-hand
side of Figure 1, the specific stages for the design of a planar
biped robot, called Sprinter, are shown as an example on
how the mechatronics approach is applied to a particular
system.
The first stage is to define the desired system behaviour.
For legged locomotion, a good discussion of various gaits
can be found in [14]. The running gait of Sprinter is defined
such that it results in significant energy conservation from
step to step, similarly to that of ostriches [15].
The second stage is to break down system behaviour into
several phases. For instance, legged locomotion is an
alternation between stance phase and swing phase. For

Fig. 1 The proposed mechatronics approach.

Sprinter the swing phase is defined as the leg motion while


the leg is not in contact with the ground, and the stance
phase is defined as the leg motion while the leg is in contact
with the ground.
The third stage is to find Simple to Control Mechanisms
(SCMs) that can achieve the desired performance during
each individual phase. These mechanisms are, ideally, well
studied mechanisms with well understood dynamics and
available viable control strategies. In case of Sprinter, two
SCMs are needed, one for the swing and one for the stance
phase. For the stance phase, the Spring Loaded Inverted
Pendulum (SLIP) method is adopted, which has been used
by other researchers to successfully model both the human
and robotic stance phase of running motion [14]. Most
existing control strategies for running robots are based on
SLIP model legs, thus making it an ideal SCM [16], [17].
For the swing phase, it has been observed that human and
bird legs contract their knees mostly passively, assuring
ground clearance and minimizing the moment of inertia of
the leg [15], [18], [19]. Hence, the simplest mechanism that
achieves passive knee retraction is the double pendulum,
which is chosen to be the swing phase SCM for Sprinter.
The fourth stage of the mechatronics approach is to
develop a single mechanical design which can emulate all
SCMs for each of the system phases. For the particular case
of Sprinter, the desired leg mechanism needs to behave as a
SLIP model during the stance phase and as a double
pendulum during the swing phase. The linkage leg presented
in Figure 2, left-hand side, is proposed as such a mechanism.
The leg knee joint, , is able to lock elastically during the
stance phase, and unlock during the swing phase. The
Analysis section will detail how the parameters of the
linkage leg will be selected such that the linkage leg
approximates the SLIP and the double pendulum models, as
well as the circumstances that affect the degree of
approximation.
In the fifth stage Simple Traceable Control Systems
(STCSs) must be developed to control the SCMs to achieve
the desired system performance during each phase. The
advantages resulting from having simple and analytically
traceable control systems for walking robots, as well as other
systems, have been previously addressed [7]. One of the key
features of the proposed approach is that the SCMs, which
the controllers are based upon, are not linear approximations
of a leg design, but non-linear, real-life mechanisms whose
behaviour is similar to that of the real robotic leg but they
are simpler to analyze. This allows simple and traceable
control to be developed for systems where linearization
cannot be used, such as running robots. Some iterations
between this stage and stage three may be required to
achieve the desired performance.
The sixth stage of the approach is to create a control
algorithm that can properly switch between the STCSs for
different system phases. Switching control has been studied
in depth (e.g., see [20]). For Sprinter a suitable switching

825

method should be designed for stance and swing phases, but


this is beyond the scope of this paper.
The seventh and final stage is the evaluation of overall
system behaviour. The switching algorithm from stage 6
may need to be revised to fine-tune the system behaviour.
Further, depending on the final outcome of design, one may
have to reconsider the specifications of the mechanism that
was designed in stage four.
III. ANALYSIS AND SYNTHESIS OF THE LEG MECHANISM
This section analyzes a mechanism that is selected as the
Sprinters leg, such that it emulates the SCMs for the stance
and swing phases. The mechanism, named linkage leg, is
shown in Figure 2, left-hand side. The leg consists of five
joints ( and ) and four links. Joints
and form a four-bar linkage configuration. The toe point
is the only ground contact point of the leg. The
mechanism has two degrees of freedom, which can be
utilized by the hip joint , to change the angle between the
leg mechanism and the body of the robot (B), and the knee
joint , to control the configuration of the four bar
linkage and thus the length of the leg. The knee
joint can be either free during the stance phase or can be
locked to spring system during the stance phase.
A. Stance phase
During the stance phase, the linkage leg design is required
to perform similarly to its corresponding SCM, the SLIP
model. The main feature of the SLIP model, shown in the
middle of Fig. 2, is that the force at the knee joint 
(prismatic) has no effect on the torque at the hip joint
[16], [21].
For the linkage leg, the length of the MA segment
determines the coupling between the hip and the knee joints.

This is because the torque at the knee joint generates an


instantaneous ground reaction force at the toe point ,
which is always perpendicular to the line [22]. This
reaction force acted at the moment arm MA creates a torque
at the hip joint. Therefore, by reducing MA the coupling
between the hip and knee joints will be reduced. The length
of MA changes as the knee joint moves. However, knee joint
displacement during the stance phase is small, and thus by
choosing proper lengths for the links it is feasible to
maintain MA approximately constant. To achieve SLIP-like
performance during the stance phase, the length of MA must
be as close to zero as possible, which can be accomplished
through an optimization. The equations necessary to
determine the kinematics of the linkage leg, and thus the
location of points and , given the angles and , are
derived in [22].
B. Swing phase
During the swing phase, the linkage leg must perform
similarly to the chosen SCM, the double pendulum. First, the
equations of motion are derived for the linkage leg.
Secondly, the linkage leg dynamic performance will be
compared with that of the double pendulum.
The dynamic model of the leg mechanism is represented
by (1). The two degrees of freedom are the hip and knee
angles and , respectively.
   


   
   


   




(1)





Fig. 2 The linkage leg on the left-hand side, the equivalent stance phase SCM in the centre and the swing phase SCM on
the right-hand side.

826

The, and parameters vary as functions of the


parameters and , where


(2)

and and are function of the knee angle 


The goal is to achieve an element-by-element match
between the dynamic equations of the linkage leg (1) and
those of the double pendulum, shown in Fig. 2 right-hand
side, as follows:
   


 


   

 










(3)

where are constant parameters dependent on the masses


and inertias of the double pendulum links, ,  and  ,
 and the locations of centres of mass relative to the link
origins and .
If such a match between (1) and (3) can be achieved, the
linkage leg will behave exactly as a double pendulum during
the swing phase. How to achieve a match is not trivial, due
to the non-constant parameters of (1). If the linkage leg
mechanism is parallel, i.e., the link pairs and
remain parallel for all knee angles, and
are constantly zero, resulting in constant parameters,
and . The dynamic model of the parallel linkage leg is
shown in (4), with constant and parameters.
   


 


   

 






(4)





Table 1 Optimization Results



      
 














  

 




 
 

       





systems the theory of System Similarity is well established


in controls community [27]. The relevance of these
alternative approaches to the mechatronic system design is
still to be investigated.
C. Linkage leg optimization
The optimization is used to find linkage leg parameters
such that its behaviour matches those of the two SCMs
during each motion phase. The parameters to be optimized
are the lengths of the leg links and the angles
and
.

For the stance phase optimization, the length of the MA


segment must be minimized. Ideally MA is zero for a perfect
match with the SLIP model. To match the dynamic model of
the linkage leg during the swing phase, the sum of
squared and squared is minimized. This would practically
make the linkage leg dynamic model similar to that of a
parallel linkage leg, for which a double pendulum match can
be found. As an example, for a linkage leg length (h) of 1m
and a compression of 0.1m during the stance phase and 0.4m
during the swing phase, Table 1 summarizes the results of
the optimization. The range for each of the optimization
variables is also shown in the table.
IV. CONTROL STRATEGY

In this case, an exact element-by-element match between


the equations of motion of the parallel linkage leg and
double pendulum can be found. This means that a double
pendulum can be constructed to behave exactly as a chosen
parallel linkage leg. However, for a regular linkage leg (nonparallel) the best match between (1) and (3) can only be
approximated through an optimization.
An exact element-by-element match between the dynamic
equations of two systems is a very restrictive requirement.
Other approaches to determining if the behaviour of two
systems is similar still need to be investigated. One possible
approach is the Theory of Similarity [23], also called
Dimensional Analysis, which has been used to determine
whether the gaits of robots or animals of different sizes are
comparable [24]. It has also been successfully employed to
express the motion of a double-pendulum using only three
dimensionless parameters [25]. The Bisimulation Theory has
been used recently to define the exact or approximate
equivalence between dynamical systems [26]. For linear

Stage five of the new mechatronics approach involves


designing Simple Transparent Control Systems (STCSs) for
each of the SCMs to achieve the desired performance in
their respective phase of motion. In stage six the STCSs are
combined into a uniform controller that will enable the
linkage leg to attain the desired behaviour during all the
system phases.
A. Stance Phase
The stance phase SCM is the SLIP model leg, which was
first used at the MIT Leg Lab in the 80s and 90s to develop
controllers for monopod and biped running robots. The
classic planar SLIP model has two actuators, one controlling
the hip angle and one controlling the length of the leg. A
planar running robot has three degrees of freedom that need
to be controlled, the body pitch, the step height and the
forward velocity [7]. The MIT-developed controller uses the
hip actuator to control the robot pitch and the leg actuator to
control the apex height of the body. The hip touchdown
angle ( in Fig 2, centre) at the beginning of stance phase
controls the forward velocity of the robot [17], [21].

827

A more detailed and realistic SLIP model is used in [16]


and [18] to determine a relationship between the hip
touchdown angle and angular velocity that guarantees both
the conservation of energy at touchdown and the desired gait
period.
In order to finalize the STCS for stance phase, a decision
must be made as to whether or not to include a knee
actuator. Such an actuator adds mass and complexity, but it
can inject a controlled amount of energy into the leg spring
element, which allows one to use the above-mentioned
control strategies [21]. If a knee actuator is not included, the
leg design becomes much simpler, but the control system
must be able to control both the pitch of the robot and the
height achieved during the swing phase using one actuator at
the hip.

1.02
1

Leg Length (m)

0.98
0.96
0.94
0.92
Linkage Leg

0.9

SLIP model
0.88
0

0.5

1.5

2.5

Time (s)

Fig. 3 Open loop stance phase simulation.


-1.2

B. Swing Phase

V. SIMULATION RESULTS
The following simulations are used to check whether the
linkage leg behaves similarly to the SLIP model during the
stance phase and to the double pendulum during the swing
phase.

-1.4
-1.5
-1.6
-1.7
Desired Knee Trajectory
Double Pendulum Knee Trajectory
Linkage Leg Knee Trajectory

-1.8
-1.9
-2
0

0.5

1.5

2.5

3.5

Time (s)

Fig. 4 Controlled knee joint response


1.5
1
0.5
Hip Angle Theta_1 (rad)

The STCS for the swing phase will have to control the
displacement and angular velocity of the hip joint from the
toe-off state to the desired touchdown state, in the predicted
swing time [16]. Additionally, the knee joint is free during
this phase and should be contracted to prevent the toe from
striking the ground.
Currently there is no running robot that uses a passive
knee joint to contract during the swing phase. The walking
robot Spring Flamingo is able to keep its knee actuator near
zero impedance, emulating a passive joint for knee retraction
during one of the studied gaits [7]. The Cornell Biped is
equipped with knee latches that allow its knees to retract
passively. Near the end of the swing phase, when the knee
reaches a predetermined extension angle, the latch is
engaged and locks the knee. The robot is able to walk at only
one predetermined speed, with one step period [10].
Arai et al. developed a strategy of controlling either joint
of a two degree-of-freedom (DOF) underactuated
manipulator with only one actuator by using the dynamic
coupling between the joints [29]. This method can be timescaled and it is applicable regardless of whether or not the
underactuated joint can be locked, but it was developed for
systems not under the influence of gravity.
Another study derives an offline path that can take a two
DOF underactuated manipulator from a stationary initial
state to a stationary final state, in a predetermined time [30].
A time-optimal underactuated controller based on [29],
[30] will be developed for the swing phase STCS. This
controller will be based on the swing SCM, the double
pendulum mechanism. Any difference between the double
pendulum and the actual linkage leg will be considered by
the controller as a disturbance.

Knee Angle Theta_2 (rad)

-1.3

0
-0.5
-1
-1.5
-2
-2.5
-3
0

Double Pendulum Hip Trajectory


Linkage Leg Hip Trajectory
0.5

1.5

2.5

3.5

Time (s)

Fig. 5 Hip joint motion resulting from control of the knee


joint
For the stance phase, the linkage leg, attached to a mass
simulating the robot body, is dropped from 0.3m above the
floor, while the toe is vertically aligned with the hip. The
robot is restricted to the sagital plane. With no control the
SLIP model would jump perfectly vertical until the repeated
collisions with the ground drain all the energy, and then it
remains at unstable equilibrium [21]. The SLIP model leg
and the linkage leg are simulated in ADAMS/View, and the
dynamic performance is displayed in Fig. 3. It shows the leg
length for the linkage leg and the SLIP model, and each

828

compression section marks the stance phase of a step. The


linkage leg during the stance phase behaves quite similarly
to the corresponding SLIP model even without any
controller. The leg mechanism, however, starts pitching
forward gradually due to the fact that the moment arm MA is
small but not zero in the optimized configuration. This effect
needs to be corrected by a stance phase controller at the hip
joint.
For the swing phase, the robot was assumed to have no
contact with the ground, and have a projectile motion. In this
situation, gravity affects only the trajectory of the center of
mass of the robot and not the motion of the legs with respect
to the body. As a result the leg was simulated in the
horizontal plane, eliminating gravity. The control scheme
outlined in [29] was used to control the hip actuator such
that the passive knee joint followed a prescribed trajectory.
The exact same controller was used for both the linkage leg
and the corresponding double pendulum, and the Matlab
simulation results can be seen in Fig. 4 and Fig. 5. A good
tracking of the desired knee trajectory was obtained for both
mechanisms, as seen in Fig. 4. The hip displacements
necessary to achieve the desired trajectory for the knee joint
can be seen in Fig. 5. Although this is an initial
investigation, it indicates that a controller developed for the
swing phase SCM, the double pendulum, can adequately
control the linkage leg.

[8]

[9]

[10]

[11]

[12]
[13]

[14]

[15]

[16]
[17]
[18]

[19]

VI. CONCLUSION
A new mechatronics approach was presented to allow the
development of easy to control mechanisms. While further
work is still required for formalizing the methodology, its
primary application to a new mechanism for legged
locomotion indicates that simple and traceable controllers
can be obtained to control a complex mechanical system.

[2]

[3]

[4]

[5]

[6]

[7]

[21]
[22]
[23]
[24]

VII. REFERENCES
[1]

[20]

V. C. Moulianitis, N. A. Aspagathos and A. J. Dentsoras, An


application to mechatronic design of robot gripper," Mechatronics,
vol. 14, no. 6, pp. 599-622, Jul. 2004.
S. Behbahani and C.W. de Silva, Mechatronic design quotient as the
basis
of
a
new
multicriteria
mechatronic
design
methodology," EEE/ASME Transactions on Mechatronic, vol. 12,
no. 2, pp. 227-232, Apr. 2007.
R. Chhabra and M. R. Emami, "Linguistic Mechatronics," in 2008
IEEE/ASME International Conference on Advanced Intelligent
Mechatronics, Xian, China, 2008.
Q. Li, F. X. Wu, Control performance improvement of a parallel
robot via the design for control approach, Mechatronics, vol. 14, no.
8, pp. 947-964, Oct. 2004.
S. Lohmeier, T. Buschmann, and H. Ulbrich, "System design and
control of anthropomorphic walking robot LOLA," IEEE.ASME
Transactions on Mechatronics, vol. 14, no. 6, pp. 658-666, Nov.
2009.
D. Wight, A foot placement strategy for robust bipedal gait control,
Ph.D. dissertation, Faculty of Eng.., Univ. of Waterloo, Waterloo,
ON, CA, 2008.
J. Pratt, Exploiting Inherent Robustness and Natural Dynamics in
the Control of Bipedal Walking Robots, Ph.D. dissertation, Comp.
Science Dept., Massachusetts Institute of Technology, Cambridge,
MA, 2000.

[25]

[26]

[27]

[28]

[29]

829

P. Manoonpong, T.Geng, and B. Porr, "The RunBot architecture for


adaptive, fast, dynamic walking," in IEEE International Symposium
on Circuits and Systems, New Orleans, LA., 2007, pp. 1181-1184.
H. Witte, H. Hoffmann, R. Hackert, C. Schilling, M. S. Fisher, and
H, Preuschoft, "Biomimetic robotics should be based on functional
morphology," Journal of Anatomy, pp. 331-342, Mar. 2004
S. Collins, A. Ruina, R. Tedrake, and M. Wisse, "Efficient bipedal
robots based on passive dynamic walkers," Science Magazine, pp.
1082-1085, Feb. 2005.
Y. Ogura and K. Shimomura, "Human-like walking with knee
stretched, heel-contact and toe-off motion by a humanoid robot," in
2006 IEEE/RSJ International Conference on Intelligent Robots and
Systems, Beijing, China, 2006, pp. 2976-2981.
M. F. Silva, "A historical perspective of legged robots," Journal of
Vibration and Control, vol. 13, no. 9-10, pp. 1447-1486, Oct. 2007.
J. Koechling and M. Raibert, "How fast can a legged robot run,"
American Society of Mechanical Engineers, Dynamic Systems and
Control Division (Publication) DSC, vol. 11, pp. 241-249, Nov.
1988.
A. Biewener, "Patterns of Mechanical Energy Change in Tetrapod:
pendula, springs and work," Journal of Experimental Zoology Part
A: Comparative Experimental Biology, vol. 305A, no. 11, pp. 899911, 2006.
R.M. Johnston and A. Bekoff, "Energetics of bipedal running. II.
Limb design and running mechanics," Journal of Comparative
Physiology A: Neuroethology, Sensory, Neural, and Behavioral
Physiology, vol. 201, pp. 169-184, Oct. 1998.
S. Hyon, "Energy-preserving control of a passive one-legged
running," Advanced robotics, vol 18, no4, pp. 357-81, Dec. 2004.
M. Raibert, "Running with symmetry," International Journal of
Robotics Research, vol. 5, no. 4, pp. 3-19, Mar. 1986.
T.
McGeer, "Passive walking with knees," in 1990 IEEE
International Conference on Robotics and Automation, Cincinnati
USA, 1990, pp. 1640-1645.
T. Roberts, R. Marsh, and P. Weyand, "Muscular Force in Running
Turkeys : the Economy of Minimizing Work," Science, vol. 275, no.
5303, pp. 1113-1115, Feb. 1997.
M. W. Spong, "Some aspects of switching control in robot
locomotion," Automatisierungstechnik, vol. 48, no. 4, pp. 157-164,
Apr. 2000.
M.H. Raibert, "Legged Robots," Communications of the ACM, pp.
499-514, 1986
W. L. Cleghorn, Mechanics of Machines, 1st ed., New York: Oxford
University Press, 2005.
A. A. Gukhman, Introduction to the theory of similarity, 1st ed., New
York: Academic Press, 1965.
S. R. Bullimore and M. J. Donelan, "Criteria for dynamic similarity
in bouncing gaits," Journal of Theoretical Biology, vol. 250, no. 2,
pp. 339-348, Jan. 2008.
A. Ohlhoff and P. H. Richter, "Forces in the double pendulum,"
Journal of Applied Mathematics and Mechanics, vol. 80, no. 8, pp.
517-534, Jul. 2000.
Y. Mingsheng and M. Wirsing, "Approximate Bisimilarity," in
Proceedings of the 8th International Conference on Algebraic
Methodology and Software Technology, Iowa City, 2000, pp. 309322.
S. Hyon, "Running Control of a Planar Biped Robot based on
Energy-Preserving Strategy," in 2004 IEEE International Conference
on Robotics and Automation, New Orleans, 2004, pp. 3791-3796.
H. Arai and S. Tachi, "Position control of manipulator with passive
joints using dynamic coupling," IEEE Transactions on Robotics and
Automation, vol. 7, no. 4, pp. 528-34, Aug. 1991.
J. A. Rosas-Flores, J. Alvarez-Gallegos, and R. Castro-Linares,
"Robust control of an underactuated planar 2R manipulator," in
Proceedings of 11th Mediterranean Conference on Control and
Automation (MED2003), Rhodes, Greece, 2003, p. 6.

You might also like