Available online at www.sciencedirect.

com
Mechanism
and
Mechanism and Machine Theory 43 (2008) 33–49
Machine Theory
www.elsevier.com/locate/mechmt

Modeling, motion planning, and control of one-legged

hopping robot actuated by two arms
a,*
Guang-Ping He , Xiao-Lan Tan a, Xiang-Hui Zhang a, Zhen Lu b

a
School of Mechanical and Electrical Engineering, North China University of Technology, Beijing 100041, China
b
School of Automation Science and Electrical Engineering, Beijing University of Aeronautics and Astronautics, Beijing 100083, China

Received 22 May 2006; received in revised form 12 January 2007; accepted 16 January 2007
Available online 23 March 2007

Abstract

A new underactuated one-legged hoping robot model is proposed for researching the utilization of the elastic energy of
underactuated flexible mechanical system repeatedly, and researching the motion control method for underactuated hop-
ping robots with dynamic balance. Only two arms actuate the hopping robot, the single elastic leg of the robot has no
actuator, thus the coupling between the arms and the unactuated leg controls the hopping motion. The modeling, motion
planning, and control method are investigated for this kind of hopping robot. A new time-varying feedback control algo-
rithm is suggested based on the nonlinear transformation of inputs locally. It is shown that controlling the orientation and
vibration of the leg would be essential for hoping stably, and controlling the motion of center of mass of the system can
control the moving speed in horizontal direction. Some numerical simulations verified some aspects of the feasibility of the
proposed model and control method.
Ó 2007 Elsevier Ltd. All rights reserved.

Keywords: Hopping robot; Underactuated; Nonholonomic; Dynamic control

1. Introduction

Hopping robot has better mobility in a natural environment for the motion style, with intermittent stance
and flight. It can move with great speed and maneuverability even on rough terrain. A great capability in evad-
ing obstacles indicates that the hopping robot may be suitable for reconnaissance tasks [1]. The dynamic bal-
ance and special motion style of the hopping robot can provide a fundamental research bed for passive
walking robot or running robot [2–6]. The simple mechanical construct of a one-legged robot also provides
a perfect test bed for studying on the control theory of nonholonomic mechanical systems. The hopping robot
holds many scholars interests in recent years, and had been reported in many publications.
One-legged system is the main research focus for hopping robot in the past three decades. Raibert and his
coworkers have significant contribution in the development of hopping robot. They had fabricated several

*
Corresponding author. Tel.: +86 010 88802835; fax: +86 010 88803005.
E-mail address: hegp55@126.com (G.-P. He).

0094-114X/$ - see front matter Ó 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mechmachtheory.2007.01.001
34 G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49

experimental robot systems such as 2D [2,3], 3D [4] one-legged hopping robot and biped gymnastic robot [5,6]
in order to understand the legged systems that hop and run. Since then, many new hopping and running
robots with complex kinematics have been proposed based on the Raibert’s research. For instance, Hyon
and Mita designed a biologically inspired hopping robot-‘‘Kenken’’ that is actuated by hydraulics [7]. Leavitt
and his coworkers designed an acrobot-like hopping robot that is actuated pneumatically [8]. Taghirad
designed a hopper that is actuated electrically [9]. Iida et al. designed a pendulum driven hopping robot
[10]. Hyon and Emura designed a passive one-legged hopper [11]. Shimoda designed a hopping robot for
microgravity environment [12]. Gregorio et al. designed an electrically actuated hopping robot ‘‘ARL Mono-
pod’’ [13]. Zeglin fabricated a bow leg hopping robot [14]. Stoeter and Papanikolopoulos fabricated a small
cylindrical jumping scout robots [15]. Most these studies have used Raibert’s controller based on the Spring
Loaded Inverted Pendulum (SLIP) model [16]. The SLIP model, which consists of a point-mass on a springy
leg, is the minimum model for running. The dynamic balance principle and motion control method can be
investigated more thoroughly for the simple mechanical system. For instance, Vakakis and Burdick et al.
investigated the chaotic motions in the dynamics of point mass hopping robot, and a strange attractor was
developed [17,18]. They find the strange attractor can be controlled and eliminated by tuning an appropriate
parameter corresponding to the duration of applied hopping thrust. Burdick and Fiorini also proposed a
novel approach to the design and deployment of small and minimally actuated jumping or hopping robots
suitable for exploring the unstructured terrains of celestial bodies [19]. For increasing energy autonomy of
the hopping robot, Papadopoulos and Cherouvim find that there exists a particular passive gait which leads
to the least dissipated energy [20,21]. The result is based on the research of Ahmadi and Buehler [22]. In recent
years, Collins, Ruina and their coworkers [23] built the first three-dimensional, kneed, two-legged, passive
dynamic walking machine that extended the work of McGeer [24], in which McGeer introduced a new concept
of passive dynamics firstly. The passive dynamic walker has no actuator but can walk downhill with human-
like gait, and can walk on level-ground when one substitutes gravitational power with a suitable small actuator
[25]. However, neither a full passive dynamic walker [24] nor a passive hopping robot is an animalized robot
system, because of their limited movement capability.
Since the SLIP is represented by a point mass at center of mass (COM), the dynamic coupling of multi-
degree of freedoms (DOFs) mechanical system is not considered. Nevertheless, the dynamic coupling is helpful
for improving the dynamic property of jumping or running robot in some circumstances. For example, swing
of arm of human being can help balance in walking and running, and augment the stride in walking, running
and jumping. In track and field sports such as high jump, long jump, trampoline, diving, discus, shot, javelin,
etc., swing the arms affects the result effectively. Thus researching on the dynamic coupling of hopping or run-
ning robot with non-SLIP model can get better understanding the dynamic balance and motion control
method for legged systems.
In this paper, we proposed a new hoping robot model, which moves in vertical plane. Two arms of the hop-
ping robot are actuated, but the elastic telescopic leg is not equipped with an actuator. The dynamic coupling
between the arms and the leg controls the motion of the system. Thus the hopping robot is underactuated and
the main aim is to find a method that can control the system by internal coupling. Works, similar to that which
lead to this paper, is reported in Ref. [10], in which a pendulum driven hopping robot was studied. Since the
pendulum driven hopping robot has four feet, the system can stable stand statically. However, differing from
work introduced in [10], the hopping robot in this paper is unstable statically. The two arms must control the
motion of the system with dynamic balance simultaneously.

2. The model

Fig. 1 shows the model of the hopping robot, of which the single telescopic leg consists of two segments.
One segment of the leg is nonzero mass and another is massless. The length of the segment with nonzero mass
is l1, the mass of it is m1, and the center of mass lies in the middle of the length. The massless segment has
length, l2, is connected to the former by linear spring serially with same axis line. The stiffness of the linear
spring is k. The length of the two arms is the same as r, and the mass is m2, respectively. The center of mass
of the arm lies at its end. The two arms are hinged at the top of the nonzero mass segment of leg. Defining the
general coordinates of the model are ðx0 ; z0 ; l2 ; u; h1 ; h2 Þ, of which ðx0 ; z0 Þ is the position of the foot toe in the
 G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49 35

θ2 θ1
z
r

r
m2
l1 m1
2

m2
l2

ϕ
z0

o x0 x
Fig. 1. The model of a one-legged robot with two actuated arms.

vertical plane, l2 is the length of the massless leg (l2 ¼ l0 when the spring is free, l2 ¼ l20 when leg is stance
vertically with static balance.), u is the angle between leg’s axis line and horizontal plane, h1 ; h2 are angles
of the two arm in respect to the leg, respectively. Positive direction of all angles are defined as anticlockwise.
Based on these definitions, the kinematics formulation of the COM of the robot system can be expressed as
X C ¼ f ðx0 ; z0 ; l2 ; u; h1 ; h2 Þ ð1Þ

in which XC, position of COM in vertical plane.

The motion of a hopping robot can be divided into two phases, stance and flight (referring to Fig. 2). In
stance phase, the ground supports the leg. In flight phase, ally parts of the robot system are unsupported
by the ground. Hopping motion consists of the two phases alternately. Since there are different constraint con-
ditions, the kinematics and dynamics of the hopping robot are different in different motion phase.

2.1. Formulations in stance phase

In the stance phase, the single leg is supported thus the position of the foot is unchanged. The kinematics of
the COM is expressed by
X_ C ¼ J 1 q_ 1 ð2Þ
T
in which q1 ¼ ½ l2 u h1 h2  , J 1 is Jacobian matrix in stance phase, and is given in Appendix A.
The dynamics formulation can be written as
q1 þ C 1 ¼ Q 1
M 1€ ð3Þ

x
stance flight
phase phase

Fig. 2. Two phases of hopping motion.

36 G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49

where M1 is inertia matrix, C1 includes centrifugal and Coriolis torque, gravity, friction, and spring force.
T
Q1 ¼ ½ 0 0 s1 s2  , the generalized driver force vector with two zero elements, which means the leg unac-
tuated and no external torque applied to the foot. s1 ; s2 are drivers torque in the two arms, respectively. For-
mulation (3) is listed in Appendix A.

2.2. Formulations in flight phase

In the flight phase, there is no support in the leg. The deformation of linear spring is zero. Thus l2 ¼ l0 is a
constant. The kinematics formulation in flight phase is similar to (2), and can be written as
X_ C ¼ J 2 q_ 2 ð4Þ
T
where q2 ¼ ½ x0 z0 u h1 h2  , and J 2 is the Jacobian in flight phase.
The dynamics formulation in flight phase is given by
q2 þ C 2 ¼ Q 2
M 2€ ð5Þ
where M2 also a inertia matrix, C2 is similar to C1 in Eq. (3) but does not include friction and spring forces in
T
leg. Q2 ¼ ½ 0 0 0 s1 s2  , the generalized non-conservative driver force vector, in which the former three
zero elements mean no non-conservative force corresponding to generalized coordinates x0 ; z0 , and u. s1 ; s2 ,
drivers torque in the two arms, respectively. A detailed form of Eq. (5) is given in Appendix B.
According to the formulations listed above, the model of hopping robot systems is varying, and the DOFs
are more than the number of independent control inputs. In Eq. (3), the portion corresponding to zero gen-
eralized driver force is not integrable, represents a second-order nonholonomic constraint, which is similar to
an Acrobot system. In Eq. (5), the portion corresponding to zero generalized driver force has the first integral,
which is represented by the generalized momentum conservation. The linear momentum portion can be inte-
grated once more to show holonomic constraints, but the angular momentum portion is not integrable show-
ing a nonholonomic constraint. A nonholonomic system is underactuated generally, reported by plenty of
research reports [26–34]. The motion plan and control of a nonholonomic system are challengeable for smooth
time-invariant state feedback to be disabled [26], thus time-varying [29,31–33] or discontinuous [34] feedback
need to be utilized.

3. Motion planning

A motion plan for a robot system includes two aspects generally, one aspect is that relates to environment,
and another is that considers the motion of the robot itself, where the environment does no restrict the motion.
The second kind of motion plan is considered in this section for the hopping robot. The purpose of motion
planning is to realize dynamic balance and hopping motion with hypothesizes which include flat terrain, no
slipping, and no obstacles. A draft hopping motion cycle is shown in Fig. 2 that also can be found in some
references such as [2].
Considering a continuous hopping, three tasks should be realized in stance phase. First, balance control. Sec-
ond, control the resonance of elastic leg actively. Third, control the orientation of leg preparing for the next
flight phase. In flight phase, the re-orientation of the body must be carried out using the internal motion. Rai-
bert and his coworkers [2–6] proposed a control algorithm with decomposing control of running into a height
control part, a forward velocity control part, and an attitude control part, based on the PD control method.
Raibert’s controller depends on an assumption of decoupling of the rotary motion of the body from the tele-
scopic motion of the leg. For his controller applicable to one-legged hopping, the mechanical model itself
should be carefully designed to satisfy the assumption that the leg should be sufficiently light compared to
the body, and the COM of the body should be positioned closely to the hip joint [16]. Nevertheless, the assump-
tion restricts Raibert’s controller to be used to a hopping system with strong nonlinear coupling in dynamics.
The two arms hopping robot in Fig. 1 does not satisfy the assumption since the position of COM of the system
varies respecting to that of the hip joints, therefore, the motion of arms can not be decoupled from the telescopic
motion of leg, and the telescopic motion can not be controlled directly for no actuator in leg.
For the hopping control of the system shown in Fig. 1, a simple motion plan can be depicted as following.
 G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49 37

3.1. Motion planning in stance phase

The motion plan in stance phase can be given by

(a) Control the position of COM respecting to that of foot toe, viz. ðxC  x0 Þ ! ðDxÞD .
(b) Control the resonance of the spring leg. When the elastic energy in the system is sufficiently large to over-
come gravity, the foot leaves the ground and hopping begins.
(c) Control the orientation of leg to given position, viz. u ! uD .

The purpose of controlling the position of COM is to control the moving speed of the system in stance
phase. Whereas, controlling the orientation of the leg can augment the stride in one hopping cycle, also it
is helpful for the moving speed of the system. Controlling the resonance of spring leg is essential for stable
hopping. For instance, the vibration of spring leg can be expected as
l2 ¼ l20 þ AðtÞ sin ðxn ðt  T k Þ þ bÞ ð6Þ
where A(t) is the vibration amplitude. Before the first lift-off, A(t) increases monotonously, and T k ¼ 0. During
hopping continuously, A(t) can be any value that the system permits. The permitted minimum of vibration
amplitude A(t) in stable hopping is Am ¼ ðl0  l20 Þ, while the maximum of A(t) is limited to the permitted
deformation of linear spring in the leg. Tk is the time of end of flight phase of the kth times of hopping cycles.
xn is the nature angular frequency. b is the phase angle. The nature angular frequency can be calculated by
rffiffiffiffi
k
xn ¼ ð7Þ
m
in which m ¼ m1 þ 2m2 is the total mass of the system, k is the stiffness of spring in telescopic leg. The nature
frequency fn of the hopping robot should be set to 2–3 Hz that is a common value of stride frequencies found
in nature [16]. Based on fn ¼ 2p=xn and Eq. (7), one can design the stiffness k of spring. The phase angle b is
defined by the initial position of the spring in leg. Given the initial displacement of spring is Dl and current
vibration amplitude of spring is A(t), the phase angle can be calculated by

arcsinðDl=AðtÞÞ Dl < 0
b¼ ð8Þ
 arcsinðDl=AðtÞÞ Dl > 0
Fig. 3 is a graphical interpretation of b assuming the physical parameters of the hopping robot given by
Appendix C. The vibration amplitude is variable (as given in Fig. 3) before the first hopping or when the hop-
ping height needs to be adjusted in stable hopping cycle. Thus the phase angle b is variable generally.
For getting a definite motion of hopping robot in stance phase, the orientation of leg also should be con-
trolled to a given position u ! uD . For instance, uD ¼ p=2 for vertical hopping and uD 6¼ p=2 for horizontal
hopping generally.

0.34

0.32 L0
β (t 2 )
A(t ) sin(ωn t )

0.3
β (t1 )
0.28

L20
0.26
A(t2 )
0.24
A(t1 )
0.22
-0.06 -0.04 -0.02 0 0.02 0.04 0.06
A(t ) cos(ωnt )

Fig. 3. Calculating the phase angle of resonance in stance phase.

38 G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49

3.2. Motion planning in flight phase

Since there is no external non-conservative force acting on the hopping robot in flight phase, the motion of
the COM of the robot is ballistic. For a stable stance motion in the next cycle, the orientation of leg should be
controlled to adapted position. For example: u ! ðp  uD Þ for symmetry to u ! uD in stance phase.

4. Designing the control law

As that discussed in the second section, the hopping robot shown in Fig. 1 is an underactuated nonholo-
nomic system. A special characteristic of this class of mechanical system is that the DOFs of it is more than the
number of the independent inputs thus not satisfying the condition of accurate linearization technique in non-
linear control theory, and even no smooth time-invariant state feedback exist [26]. For the underactuated
manipulators, we proposed a time-varying control method in [32]. This method is extended to control the
one-legged hopping robot in this paper.
During a small time segment ½ ðk  1ÞT kT , where k is integer, define a piecewise constant vector
vðtÞjt¼½ðk1ÞT kT  ¼ cðkÞ ð9Þ

and define two smooth, time-varying matrices H 1 and H 2 , with relationship

d
H2 ¼ ðH 1 Þ ð10Þ
dt  
H1
and the matrix H ¼ satisfying
H2
rankðHÞ ¼ 2na ð11Þ
where na is the DOFs of the actuated joints. For the robot system in Fig. 1, na ¼ 2.

4.1. Controller in stance phase

Decomposing Eq. (2) into

    
x_ C J 11 J 12 q_ p
¼ ð12Þ
z_ C J 21 J 22 q_ a
 
J 11 J 12 T T
where J 1 ¼ , qp ¼ ½ l2 u  , qa ¼ ½ h1 h2  . Similarly, Eq. (3) also can be rewritten to
J 21 J 22
qp þ mpa €
mpp € qa þ c p ¼ 0
ð13Þ
qp þ maa €
map € qa þ c a ¼ s
 
mpp mpa T T  T T
where M 1 ¼ , qp ¼ ½ l2 u  , qa ¼ ½ h 1 h2  , C 1 ¼ cTp cTa , s ¼ ½ s1 s2  .
map maa
Define the inputs u ¼ €
qa , and let
q_ a ¼ H 1 v
ð14Þ
qa ¼ H 2 v
€
Thus the piecewise constant vector can be resolved into
 
_
þ qa
v¼H ð15Þ
€
qa
From the first equation of (12), we obtain
þ
q_ a ¼ ðJ 12 Þ x_ C  J 11 q_ p ð16Þ
 G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49 39

from the first equation of (13), we get

 þ
qa ¼  mpa
€ qp  c p
mpp € ð17Þ
Define exc ¼ xdC  xC , ep ¼ qdp  qp , where xdC ; qdp are given values, the inputs with feedback can be written as
Z
þ d
q_ a ¼ ðJ 12 Þ x_ C þ k 1 exc þ k 2 exc dt  J 11 q_ p ð18Þ
 h i
þ
€qa ¼  mpa mpp €qdp þ k 3 e_ p þ k 4 ep  cp ð19Þ

where k i > 0; i ¼ 1; 2; 3; 4. Substituting (18) and (19) into (15), the piecewise constant virtual input vector v can
be obtained. Substituting the virtual input v into the second equation of (14), the actual inputs can be resolved
to
2 þ R 3
ðJ 12 Þ x_ dC þ k 1 exc þ k 2 exc dt  J 11 q_ p
u ¼ H 2H þ4  h i5 ð20Þ
þ
 mpa mpp €qdp þ k 3 e_ p þ k 4 ep  cp

Since that the passive coordinates (in Eq. (19)) and the motion of COM of the robot (in Eq. (18)) are con-
trolled at the same time, the control law (20) controlled the whole system in stance phase.
The stability of the controller (20) is obvious since the derivation of (20) is reversible when matrix mpa is full
rank. The fundamental of the proposed control law is to find a piecewise constant virtual inputs which satisfies
the first equation in (12) and (13), respectively. Since that the nonlinear transformation (14) is local, and dur-
ing a small time segment ½ ðk  1ÞT kT , the accurate piecewise constant virtual inputs with extended dimen-
sions more than the DOFs of the system can be obtained by (15). In the virtual inputs space the system is
exponential stabilizable locally. By the time-varying inverse transformation (15), the control law (20) is based
upon a time-varying feedback, though the PD controller with time-unvarying feedback is inbuilt locally.

4.2. Controller in flight phase

The detailed form of dynamic formulation (5) is

€ þ m14 €
m11€x0 þ m12€z0 þ m13 u h1 þ m15 €
h2 þ c 1 ¼ 0
€ þ m24 €
m21€x0 þ m22€z0 þ m23 u h1 þ m25 €
h2 þ c 2 ¼ 0
€ þ m34 h1 þ m35 €
m31€x0 þ m32€z0 þ m33 u € h2 þ c 3 ¼ 0 ð21Þ
€ þ m44 €
m41€x0 þ m42€z0 þ m43 u h1 þ m45 €
h2 þ c 4 ¼ s 1
m51€x0 þ m52€z0 þ m53 u €1 þ m55 €
€ þ m54 h h2 þ c 5 ¼ s 2
where m12 ¼ m21 ¼ 0, x0 and z0 are separated from the coordinates ðu; h1 ; h2 Þ thus the former two equations in
(21) can be integrated to a expression only included configuration variables. These are holonomic constraints
for the hopping system in flight phase. The position of COM does not depend on the inputs €h1 ; €h2 but the ini-
tial speed of COM. In (21), the third equation can be integrated to an expression of angular momentum con-
servation that is not integrable any more, thus a first-order nonholonomic constraint. The total angular
momentum of the hopping robot system is constant and does not depend on the inputs €h1 ; €h2 . Nevertheless,
u can be controlled by the motion of arms when singularity of dynamic coupling does not occur. The main
object of motion control in the flight phase should be to control the orientation of the leg to a planned position
in preparation for the next motion in stance phase.
For designing a control law in flight phase, we rewrite the former two equations in (21) to forms
1
€x0 ¼  € þ m14 €
m13 u h1 þ m15 €
h2 þ c 1
m11
ð22Þ
1
€z0 ¼  € þ m24 €
m23 u h1 þ m25 €
h2 þ c 2
m22
40 G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49

where m11 ¼ m22 ¼ m1 þ 2m2 6¼ 0. Substituting (22) into the third equation in (21), we obtain
€ þ M H 2€
M H 1u h1 þ M H 3 €
h2 þ C H ¼ 0 ð23Þ
in which
m31 m13 m32 m23 m31 m14 m32 m24 m31 m15 m32 m25
M H 1 ¼ m33   ; M H 2 ¼ m34   ; M H 3 ¼ m35   ;
m11 m22 m11 m22 m11 m22
m31 m32
C H ¼ c3  c1  c2
m11 m22
When u is the single variable to be controlled, by the Eq. (23), one can design the controller conveniently.
Nevertheless, the control problem of a nonholonomic system such as the hopping robotin flightphase should
T
be to control the whole motion ðuðtÞ; h1 ðtÞ; h2 ðtÞÞ by reduced dimensional inputs u ¼ €h1 €h2 . Similar to
designing the controller (20), considering the nonlinear transformation (14), the controller in flight phase
can be designed as
"  T R #
_
h d
h_ d þ k 5 e h þ k 6 e h dt
u ¼ H 2H þ 1
þ
2
 d  ð24Þ
½ M H 2 M H 3  M H 1 u € þ k 7 e_ u þ k 8 eu  C H
 T  T
where eh ¼ h_ d1 h_ d2  h_ 1 h_ 2 , eu ¼ ud  u, and k i > 0; i ¼ 5; 6; 7; 8.

5. Simulations

Numerical simulations were used to evaluate the motion plan and control algorithm for the model shown in
Fig. 1. We designed a hopping robot prototype that the parameters of it in physical are listed in Appendix C.
The simulation results in this section are based on the parameters of the prototype. Fig. 4 is the 3D model of
the robot, which is actuated by two 90 W MAXON motors RE35 with reducer. The ratio of the reducers is
113:1. The motor with reducer can provide 25 N m output torque approximately, under maximum turning
speed 10 rad/s instantaneously. The total mass of the robot is 3.2 kg with two 1.0 kg arms and a 1.2 kg leg.
The robot is fabricated by duralumin except a few of parts such as the motors, reducers and spring.
For controlling the hopping robot system to motion stably, the control tasks include:

(1) Control the errors between the position of COM and it of foot tip in x-direction to satisfy
ðxC  x0 Þ ¼ vx t, in which t is time variable, vx is the expected moving speed.
(2) Control the angle of leg to given position.For instance, u ¼ 0:5p, for vertical hopping. u ¼ 0:5p  Du,
for lift-off of horizontal hopping, and u ¼ 0:5p þ Du, for touch-down of horizontal hopping, where Du
is the lift-off or touchdown angle of leg off the vertical axis.
(3) Control the motion of the spring in the leg to satisfy the resonance
l2 ¼ l20 þ AðtÞ sinðxn ðt  T k Þ þ bÞ
Dl l0 l20
where AðtÞ ¼ At, xn ¼ 6p, and b ¼  arcsin AðtÞ
¼  arcsin AðtÞ
are given.

We select the time-varying nonlinear transformation matrices of controller (20) and (24) to be
" # " #
2 3 2
1 Dt1 ðDt1 Þ ðDt1 Þ 0 1 2ðDt1 Þ 3ðDt1 Þ
H1 ¼ 2 3
; H2 ¼ 2
1 Dt2 ðDt2 Þ ðDt2 Þ 0 1 2ðDt2 Þ 3ðDt2 Þ
where Dt1 ¼ t  t1 , Dt2 ¼ t  t2 , in which t1 ; t2 are arbitrary constants and t1 6¼ t2 . It is obviously that
H1
rank ¼ 4 satisfying the condition (11).
H2
Fig. 5 plots some simulation results of achieving to first hopping from initial static configuration (u ¼ 110 ,
h1 ¼ 160 , h2 ¼ 220 , l2  l20 Þ. The expected moving speed vx is zero, and Du ¼ 0, which means a vertical hop-
ping. The resonance amplitude is AðtÞ ¼ 0:05t m in the simulation. In fact A(t) can be any value that the system
permits. A larger amplifying coefficient A results in the first jump being faster, while larger actuation torque is
G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49 41

Fig. 4. The 3D model of the robot prototype.

110

100
ϕ (o)

90

80
0 0. 1 0. 2 0. 3 0. 4 0. 5 0. 6 0. 7 0. 8 0. 9 1
time (s)
250
θ2
θ1 ,θ 2 ( o )

200

150
θ1
100
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time (s)

40
τ 1 ,τ 2 ( Nm )

20
τ2
0
τ1
-20
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time (s)

100
f spring ( N )

50

-50
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time (s)
0.03

0.02
z 0 (m)

0.01

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time (s)

Fig. 5. The first hopping from initial configuration.

42 G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49

needed. Ordinarily, a number of resonant oscillations need to be built in amplitude because of the limited actu-
ation torque. Subfigures (a)–(e) plot the angle of leg u, position of actuated joints h1 ; h2 , torque of actuators,
force of spring, and vertical position of foot toe z0 respectively, during the time 0–1 s. Referring to Fig. 5(c),
the torques of the two arms is acceptable for the model (Fig. 4) with parameters given in Appendix C.
Fig. 6 shows a 30 s simulation results. The expected moving speed vx is 0.9 m/s. The subfigures (a)–(d) in
Fig. 6 show the motion of u; l2 , h1 ; h2 , and z0, respectively. In stance phase, angle of leg u is controlled to 80°,
and in flight phase, u is controlled to 100° (referring to Fig. 6(a)). Rhythmic moving speed vx and swing angle
of leg Du is propitious to stabilize the system. Generally, a larger swing angle of the leg tends to increase the
moving speed. Unsuited swing angle of the leg with regards to the expected moving speed will result in con-
siderable actuation torque and difficulty in stabilizing the system. The resonance of the leg is controlled to rule
(6) with Am ¼ 3ðl0  l20 Þ stably (referring to Fig. 6(b)) so that the robot can hop to a height of about 0.1 m
(referring to Fig. 6(d)), where Am denotes the stable maximum amplitude of the spring vibration for a specific
hopping height. Referring to Fig. 6(c), one can find that one of the two arms balances the system, and another
propels the robot to move in a horizontal direction, approximately. The action of the two arms is exchange-
able stochastically relating to the initial configuration of the robot, according to our many simulations.
Fig. 6(e) plots some configurations of the robot in hopping, and the continuous dash line plots the motion
of foot tip of the robot. Fig. 7 shows the change of configurations for showing the motion of the robot both
in single stance phase (Fig. 7(a)) and in single flight phase (Fig. 7(b)) more clearly.
The last 15 s of data from Fig. 6 are replotted in the phase plane in Fig. 8. The abscissa is the velocity of
COM of the robot in vertical direction, and the ordinate is the position of COM of the robot in vertical direc-
tion. During flight the trajectory of COM is similar to a parabola, which was caused by constant gravitational

120

100
ϕ (o )

80

60
0 5 10 15 20 25 30
time (s)

0.4
l 2 (m)

0.2

0
0 5 10 15 20 25 30
time (s)
400
θ2 θ1
θ1 ,θ 2 ( o )

200

-200
0 5 10 15 20 25 30
time (s)
0.4
z 0 (m )

0.2

0
0 5 10 15 20 25 30
time (s)

2
1
z (m)

0
-1

0 5 10 15 20 25
x (m)

Fig. 6. Hopping sequence during 0–30 s.

 G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49 43

a b

Fig. 7. Some configurations of hopping robot during one hopping cycle. (a) In stance phase; (b) in flight phase.

0.45

0.4
touch lift
top
0.35 down off
z C ( m)

l20
0.3

0.25 bottom

0.2
-2 -1 0 1 2
•
z C (m/s)

Fig. 8. Phase plot for vertical hopping.

300
θ2 ϕ θ1
ϕ , θ1 , θ 2 ( o )

200

100

0
0 1 2 3 4 5 6 7 8 9 10
time (s)
0.4
z 0 , l 2 (m )

0.2
z0 l2

0
0 1 2 3 4 5 6 7 8 9 10
time (s)
1
x Cd
x C , x Cd ( m )

0.5

0
xC
-0 .5
0 1 2 3 4 5 6 7 8 9 10
time (s)

Fig. 9. Hopping sequence during 0–10 s with motion plan parameters given by vx ¼ 0:1 m=s, Du ¼ 2 , Am ¼ 1:5ðl0  l20 Þ.
44 G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49

acceleration, and during stance the trajectory of COM is harmonic, which was caused by the vibration of
spring. Four events depicting the whole motion cycle in vertical direction, lift-off, top, touchdown, and bottom
can be indicated by the figure, which is similar to a full actuated one-legged hopping robot.
Figs. 9–11 plot three simulation results with different motion plan parameters from Fig. 6. Fig. 9 shows a
slow hopping speed with low hopping height, which corresponds to motion plan parameters vx ¼ 0:1 m=s,
Du ¼ 2 , and Am ¼ 1:5ðl0  l20 Þ. Fig. 10 shows a middle hopping speed with middle hopping height, which
corresponds to motion plan parameters vx ¼ 0:3 m=s, Du ¼ 5 , and Am ¼ 2ðl0  l20 Þ. Fig. 11 plots the results
corresponding to given parameters vx ¼ 0:6 m=s, Du ¼ 10 , and Am ¼ 2:5ðl0  l20 Þ. Comparing to the numer-
ical simulation results, the robot system can be controlled in a large range of hopping speeds by regulating the
swing angle and the vibration amplitude of spring in the passive leg. Where the maximum amplitude of res-
onance of the spring leg limits the maximum hopping speed and hopping height. For the model plotted in

400
θ2 ϕ θ1
200
ϕ , θ1 ,θ 2 ( o )

-200
0 1 2 3 4 5 6 7 8 9 10
time (s)

0.4
z 0 ,l 2 (m)

0.2
z0
l2

0
0 1 2 3 4 5 6 7 8 9 10
time (s)
4
xCd
xC ,xCd (m)

0
xC
-2
0 1 2 3 4 5 6 7 8 9 10
time (s)

Fig. 10. Hopping sequence during 0–10 s with motion plan parameters given by vx ¼ 0:3 m=s, Du ¼ 5 , Am ¼ 2ðl0  l20 Þ.

400
θ2 ϕ θ1
200
ϕ ,θ1,θ 2 ( o)

-200
0 1 2 3 4 5 6 7 8 9 10
time (s)
0.4
z0 , l 2 (m)

0.2
z0
l2

0
0 1 2 3 4 5 6 7 8 9 10
time (s)
10
xCd
xC, xCd (m)

0
xC
-5
0 1 2 3 4 5 6 7 8 9 10
time (s)

Fig. 11. Hopping sequence during 0–10 s with motion plan parameters given by vx ¼ 0:6 m=s, Du ¼ 10 , Am ¼ 2:5ðl0  l20 Þ.
 G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49 45

Fig. 4, the maximum deformation of the spring of the leg is limited to 100 mm, which equals to 3:5ðl0  l20 Þ
approximately.
New characteristic of the passive spring one-legged hopping robot is that it is underactuated in all the two
motion phases. The simulation results in Figs. 5–11 show that the passive spring one-legged hopping robot
can be controlled by two arms. It differs from a hopping robot with SLIP model, which can not get to first
jump by itself so far, the two arms actuated one-legged hopping robot can be controlled dexterously, for
instance, stand stably, getting to first jump by itself, vertical hopping, level hopping, etc. This reveals the
importance of the arm in controlling the motion of one-legged hopping robot, since the motion of the system
is actuated by the dynamic coupling only. The dynamic coupling is helpful for improving the dynamic prop-
erty of jumping or running robot in some circumstances. Based on the result given in this paper, one is con-
vinced that a one-legged hopping robot with non-SLIP model has better mobile capability than one with SLIP
model.

6. Conclusions

A new type of non-SLIP model based one-legged hopping robot model is proposed for understanding the
control method through the internal dynamic coupling. The robot system consists of two actuated arms and
single passive, telescopic, spring leg. Based on a time-varying nonlinear transformation to the inputs locally, a
motion control method is proposed for this kind of underactuated nonholonomic system. The essential task in
the control algorithm is to control the vibration of the spring leg actively. Whereas, the orientation of leg and
motion of COM can influence the level hopping speed considerably. When the angle of the leg is controlled to
vertical, it is helpful for the balance control. When the leg is control to a non-vertical position, it is helpful for
moving speed. Numerical simulation results reveal the importance of the arm in controlling the motion of one-
legged hopping robot by the internal dynamic coupling. This will encourage us to research a dynamic walking
or a running anthropomorphic robot as a long-range goal.

Acknowledgements

The authors would like to acknowledge the anonymous reviewers for their valuable suggestions. This work
is supported by National Nature Science Foundation of China (No. 50375007 and 50475177) and Beijing Nat-
ure Science Foundation (No. 3062009).

Appendix A. Equations of motion for model of Fig. 1 in stance phase

(a) The kinematics equations of COM in stance phase

 
1 1
xC ¼ m1 l1 þ l2 cos u þ 2m2 ðl1 þ l2 Þ cos u þ m2 r cosðu þ h1 Þ þ m2 r cosðu þ h2 Þ
m1 þ 2m2 2
 
1 1
zC ¼ m1 l1 þ l2 sin u þ 2m2 ðl1 þ l2 Þ sin u þ m2 r sinðu þ h1 Þ þ m2 r sinðu þ h2 Þ
m1 þ 2m2 2

(b) The speed equations of COM (Eq. (2)) X_ C ¼ J 1 q_ 1 where the Jacobian matrix
 
j11 j12 j13 j14
J1 ¼
j21 j22 j23 j24

In which
j11 ¼ cos u
 
1 1
j12 ¼ m1 l1 þ l2 sin u  2m2 ðl1 þ l2 Þ sin u  m2 r sinðu þ h1 Þ  m2 r sinðu þ h2 Þ
m1 þ 2m2 2
46 G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49

m2 r sinðu þ h1 Þ
j13 ¼ 
m1 þ 2m2
m2 r sinðu þ h2 Þ
j14 ¼
m1 þ 2m2
j21 ¼ sin u
 
1 1
j22 ¼ m1 l1 þ l2 cos u  2m2 ðl1 þ l2 Þ cos u  m2 r cosðu þ h1 Þ  m2 r cosðu þ h2 Þ
m1 þ 2m2 2
m2 r cosðu þ h1 Þ
j23 ¼
m1 þ 2m2
m2 r cosðu þ h2 Þ
j24 ¼
m1 þ 2m2
(c) Dynamic equation (Eq. (2)) in stance phase.
Given the stiffness of the spring leg is k, the friction force coefficient is c, gravity acceleration is g, the dynamic
equation in stance phase is expressed by
q1 þ C 1 ¼ Q 1
M 1€
where
2 3
m11 m12 m13 m14
6 7
6 m21 m22 m23 m24 7
M1 ¼ 6
6
7
7
4 m31 m32 m33 m34 5
m41 m42 m43 m44
m11 ¼ m1 þ 2m2
m12 ¼ m2 rðsin h1 þ sin h2 Þ
m13 ¼ m2 r sin h1
m14 ¼ m2 r sin h2
m21 ¼ m12
" #
1
2
1 2 h i
2
m22 ¼ m1 l1 þ l2 þ l1 þ m2 ðl1 þ l2 Þ þ 4r2 þ 2m2 ðl1 þ l2 Þrðcos h1 þ cos h2 Þ
2 12

m23 ¼ 2m2 r2 þ m2 ðl1 þ l2 Þr cos h1

m24 ¼ 2m2 r2 þ m2 ðl1 þ l2 Þr cos h2
m31 ¼ m13
m32 ¼ m23
m33 ¼ 2m2 r2
m34 ¼ 0
m41 ¼ m14
m42 ¼ m24
m43 ¼ 0
m44 ¼ m33
T
C 1 ¼ ½ c1 c2 c3 c4 
 G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49 47

where
 
1
c1 ¼  m1 l1 þ l2 þ 2m2 ðl1 þ l2 Þ u_ 2 þ m2 rðcos h1 þ cos h2 Þðh_ 21  u_ 2 Þ
2
þ ðm1 þ 2m2 Þg sin u þ kðl2  l20 Þ  cl_ 2
 
1
c2 ¼ 2ðm1 þ m2 Þ l 1 þ l2 þ ðl1 þ l2 Þ l_ 2 u_ þ m2 r cos h1 h_ 1 þ cos h2 h_ 2 l_ 2
2
 m2 ðl1 þ l2 Þr sin h1 h_ 1 ðu_ þ h_ 1 Þ  m2 ðl1 þ l2 Þr sin h2 h_ 2 ðu_ þ h_ 2 Þ
h i
þ m2 rl_ 2 cos h1 u_ þ h_ 1 þ cos h2 ðu_ þ h2 Þ þ m2 rl_ 2 u_ ðcos h1 þ cos h2 Þ
 
1
 m2 rðl1 þ l2 Þu_ sin h1 h_ 1 þ sin h2 h_ 2 þ ðm1 þ 2m2 Þg cos u l1 þ l2 þ ð l1 þ l2 Þ
2
þ m2 gr½cosðu þ h1 Þ þ cos ðu þ h2 Þ
c3 ¼ m2 rðl1 þ l2 Þu_ 2 þ m2 gr cosðu þ h1 Þ
c4 ¼ m2 rðl1 þ l2 Þu_ 2 þ m2 gr cosðu þ h2 Þ

Appendix B. Equations of motion for model of Fig. 1 in flight phase

(a) The kinematics equations of COM in flight phase

 
1 1
xC ¼ x0 þ m1 l1 þ l2 cos u þ 2m2 ðl1 þ l2 Þ cos u þ m2 r cosðu þ h1 Þ þ m2 r cosðu þ h2 Þ
m1 þ 2m2 2
 
1 1
zC ¼ z0 þ m1 l1 þ l2 sin u þ 2m2 ðl1 þ l2 Þ sin u þ m2 r sinðu þ h1 Þ þ m2 r sinðu þ h2 Þ
m1 þ 2m2 2

(b) Speed equations of COM in flight phase.

The speed equations of COM in flight phase are deleted since the motion of COM does not depend on the
inputs.
(c) Dynamics equations in flight phase.
The dynamics equations is written as (21), where
m11 ¼ m1 þ 2m2
m12 ¼ 0
1
m13 ¼ m1 l1 þ l2 sin u  2m2 ðl1 þ l2 Þ sin u  m2 r½sinðu þ h1 Þ þ sinðu þ h2 Þ
2
m14 ¼ m2 r sinðu þ h1 Þ
m15 ¼ m2 r sinðu þ h2 Þ
m21 ¼ m12
m22 ¼ m1 þ 2m2
1
m23 ¼ m1 l1 þ l2 cos u þ 2m2 ðl1 þ l2 Þ cos u þ m2 r½cosðu þ h1 Þ þ cosðu þ h2 Þ
2
m24 ¼ m2 r cosðu þ h1 Þ
m25 ¼ m2 r cosðu þ h2 Þ
m31 ¼ m13
m32 ¼ m23
48 G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49

1
2
1 h i
m33 ¼ m l1 þ l 2 þ m1 l21 þ 2m2 r2 þ 2m2 ðl1 þ l2 Þ2 þ r2 þ 2m2 ðl1 þ l2 Þrðcos h1 þ cos h2 Þ
2 12
m34 ¼ 2m2 r2 þ m2 ðl1 þ l2 Þr cos h1
m35 ¼ 2m2 r2 þ m2 ðl1 þ l2 Þr cos h2
m41 ¼ m14
m42 ¼ m24
m43 ¼ m34
m44 ¼ 2m2 r2
m45 ¼ 0
m51 ¼ m15
m52 ¼ m25
m53 ¼ m35
m54 ¼ m45
m55 ¼ 2m2 r2
1
l1 þ l2 cos uu_ 2  2m2 ðl1 þ l2 Þ cos uu_ 2  m2 r cos ðu þ h1 Þðu_ þ h_ 1 Þ  m2 r cosðu þ h2 Þðu_ þ h_ 2 Þ
2 2
c1 ¼ m1
2
1
c2 ¼ m1 l1 þ l2 sin uu_ 2  2m2 ðl1 þ l2 Þ sin uu_ 2  m2 r sinðu þ h1 Þðu_ þ h_ 1 Þ
2
2
 m2 r sinðu þ h2 Þðu_ þ h_ 2 Þ2 þ ðm1 þ 2m2 Þg
c3 ¼ m2 ðl1 þ l2 Þr sin h1 u_ h_ 1  m2 ðl1 þ l2 Þr sin h1 ðu_ þ h_ 1 Þh_ 1  m2 ðl1 þ l2 Þr sin h2 u_ h_ 2
1
 m2 ðl1 þ l2 Þr sin h2 ðu_ þ h_ 2 Þh_ 2 þ m1 g l1 þ l2 cos u þ 2m2 gðl1 þ l2 Þ cos u
2
þ m2 gr½cosðu þ h1 Þ þ cosðu þ h2 Þ
c4 ¼ m2 ðl1 þ l2 Þr sin h1 u_ h_ 1 þ m2 ðl1 þ l2 Þr sin h1 uð
_ u_ þ h_ 1 Þ þ m2 gr cosðu þ h1 Þ
c5 ¼ m2 ðl1 þ l2 Þr sin h2 u_ h_ 2 þ m2 ðl1 þ l2 Þr sin h2 uð
_ u_ þ h_ 2 Þ þ m2 gr cosðu þ h2 Þ

Appendix C. The parameters of model in Fig. 1 for simulation

The nature frequency: fn ¼ 3:0 ðHzÞ.

The mass of the model: m1 ¼ 1:2 ðkgÞ; m2 ¼ 1:0 ðkgÞ.
The stiffness of spring: k ¼ 4p2 mf 2n  1137 ðN=mÞ.
The length of massless leg segment: l0 ¼ 0:3 ðmÞ.
The length of nonzero mass leg segment: l1 ¼ 0:3 ðmÞ.
The length of arms: r ¼ 0:4 ðmÞ.
The initial length of spring leg: l20  l0  m1 þ2m
k
2
¼ 0:2724 ðmÞ.

References

[1] H.S. Seifert, The lunar pogo stick, Journal of Spacecraft and Rockets 4 (7) (1967) 941–943.
[2] Marc H. Raibert, Hopping in legged systems—modeling and simulation for the two-dimensional one-legged case, IEEE Transactions
on Systems, Man, and Cybernetics SMC-14 (3) (1984) 451–463.
[3] Marc H. Raibert, Francis C. Wimberly, Tabular control of balance in a dynamic legged system, IEEE Transactions on Systems, Man,
and Cybernetics SMC-14 (2) (1984) 334–339.
[4] Marc H. Raibert, H. Benjamin Brown Jr., Michael Chepponis, Experiments in balance with a 3D one-legged hopping machine, The
International Journal of Robotics and Research 3 (2) (1984) 75–92.
 G.-P. He et al. / Mechanism and Machine Theory 43 (2008) 33–49 49

[5] Jessica K. Hodgins, Marc H. Raibert, Adjusting step length for rough terrain locomotion, IEEE Transactions on Robotics and
Automation 7 (3) (1991) 289–298.
[6] Jessica K. Hodgins, Marc H. Raibert, Biped gymnastics, The International Journal of Robotics Research 9 (2) (1990) 115–132.
[7] S.H. Hyon, T. Mita, Development of a biologically inspired hopping robot-‘‘Kenken’’, in: Proceedings of the 2002 IEEE International
Conference on Robotics and Automation, Washington, DC, May 2002, pp. 3984–3991.
[8] John Leavitt, James E. Bobrow, Athanasios Sideris, Robust balance control of a one-legged, pneumatically-actuated, acrobot-like
hopping robot, in: Proceedings of the 2004 IEEE International Conference on Robotics and Automation, New Orieans, April 2004,
pp. 4240–4245.
[9] H.D. Taghirad, Analysis, Design, and Control of Hopping Robot, Master Dissertation of McGill University, Canada, 1993.
[10] Fumiya Iida, Raja Dravid, Chandana Paul, Design and control of a pendulum driven hopping robot, in: Proceedings of the 2002
IEEE/RSJ International Conference on Intelligent Robots and Systems, Lausanne, 2002, pp. 2141–2146.
[11] S.H. Hyon, T. Emura, Quasi-periodic gaits of passive one-legged hopper, in: Proceedings of IEEE/RSJ International Conference on
Intelligent Robots and Systems, 2002, pp. 2625–2630.
[12] S. Shimoda, T. Kubota, I. Nakatani, New mechanism of attitude control for hopping robot, in: Proceedings of IEEE/RSJ
International Conference on Intelligent Robots and Systems, 2002, pp. 2631–2636.
[13] P. Gregorio, M. Ahmadi, M. Buehler, Design, control and energetic of an electrically actuated legged robot, IEEE Transactions on
Systems, Man ad Cybernetics 27B (4) (1997) 626–634.
[14] Garth Zeglin, The Bow Leg Hopping Robot, Doctor Dissertation of Carnegie Mellon University, USA, 1999.
[15] Sascha A. Stoeter, Nikolaos Papanikolopoulos, Kinematic motion model for jumping scout robots, IEEE Transactions on Robotics
22 (2) (2006) 398–403.
[16] Akihiro Sato, Martin Buehler, A planar hopping robot with one actuators: design, simulation, and experimental results, in:
Proceeding of 2004 IEEE/RSJ International Conference on Intelligent Robots and Systems, Sendal, Japan, 2004, 3540–3545.
[17] A.F. Vakakis, J.W. Burdick, Chaotic motions in the dynamics of a hopping robot, in: IEEE International Conference on Robotics
and Automation, Cincinnati, OH, USA, 1990, pp. 1464–1469.
[18] A.F. Vakakis, J.W. Burdick, T.K. Caughey, ‘Interesting’ strange attractor in the dynamics of a hopping robot, International Journal
of Robotics Research 10 (6) (1991) 606–618.
[19] Joel Burdick, Paolo Fiorini, Minimalist jumping robots for celestial exploration, International Journal of Robotics Research 22 (7–8)
(2003) 653–674.
[20] Evangelos Papadopoulos, Nicholas Cherouvim, On increasing energy autonomy for a one-legged hopping robot, in: IEEE
International Conference on Robotics and Automation, New Orleans, LA, 2004, pp. 4645–4650.
[21] Nicholas Cherouvim, Evangelos Papadopoulos, Energy saving passive-dynamic gait for a one-legged hopping robot, Robotica 24 (4)
(2006) 491–498.
[22] M. Ahmadi, M. Buehler, Stable control of a simulated one-legged running robot with hip and leg compliance, IEEE Transactions on
Robotics and Automation 13 (1) (1997) 96–104.
[23] Steven H. Collins, Martijn Wisse, Andy Ruina, A three-dimensional passive-dynamic walking robot with two legs and knees,
International Journal of Robotics Research 20 (7) (2001) 607–615.
[24] Tad McGeer, Passive dynamic walking, International Journal of Robotics Research 9 (2) (1990) 62–68.
[25] Steve Collins, Andy Ruina, Russ Tedrake, Martijn Wisse, Efficient bipedal robots based on passive-dynamic walkers, Science 307
(5712) (2005) 1082–1085.
[26] R.W. Brockett, Asymptotic stability and feedback stabilization, in: R.W. Brockett, R.S. Millman, H.J. Sussmann (Eds.), Differential
Geometric Control Theory, Birkhauser, 1983, pp. 181–208.
[27] Matthew D. Berkemeier, Ronald S. Fearing, Sliding and hopping gaits for the underactuated acrobot, IEEE Transactions on
Robotics and Automation 14 (4) (1998) 629–634.
[28] Matthew D. Berkemeier, Ronald S. Fearing, Tracking fast inverted trajectories of the underactuated acrobot, IEEE Transactions on
Robotics and Automation 15 (4) (1999) 740–750.
[29] Hirohiko Arai, Kazuo Tanie, Naoji Shiroma, Time-scaling control of an underactuated manipulator, Journal of Robotics Systems 15
(9) (1998) 525–536.
[30] Mark W. Spong, Partial feedback linearization of underactuated mechanical systems, in: Proceedings of IROS’94, Munich, Germany,
vol. 1, 1994, pp. 314–321.
[31] F.U. Rehman, H. Michalska, Geometric approach to feedback stabilization of a hopping robot in the flight phase, in: ICRA’97,
Monterey, CA, 1997, pp. 551–556.
[32] Guangping He, Zhen Lu, Self-reconfiguration of underactuated redundant manipulators with optimizing the flexibility ellipsoid,
Chinese Journal of Mechanical Engineering 18 (1) (2005) 92–97.
[33] H.J. Sussmann, Wensheng Liu. Lie bracket extensions and averaging: the single-bracket case, in: Z. Li, J.F. Canny (Eds.),
Nonholonomic Motion Planning, Boston, 1993, pp. 109–148.
[34] F.U. Rehman, Discontinuous feedback stabilization of a hopping robot in flight phase, IEEE Conference on Robotics and
Automation (2001) 190–194.