Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Proceeding of the 2nd

RSI/ISM International Conference on Robotics and Mechatronics


October 15-17, 2014, Tehran, Iran

Pattern Generation for Humanoid Robot with Natural


ZMP Trajectory

Farshad Samadi and Hamid Moghadam-Fard


Faculty of Electrical and Computer Engineering
University of Tabriz
Tabriz, Iran
{f_samadi89, h_moghadam89}@ms.tabrizu.ac.ir

Abstract—Walking pattern generation is one of the key of free leg dynamics in robots locomotion. So biped is
problems in control of humanoid robots. Zero Moment Point considered as two different masses instead of one mass in
(ZMP)-based approaches such as Linear Inverted Pendulum LIPM. One mass is for both the upper body and supporting leg,
Mode (LIPM) and Gravity Compensated Inverted Pendulum and the other mass shows the effect of swinging leg. In the
Mode (GCIPM) were proposed to generate the walking pattern. methods explained above, the reference trajectories are
In proposed methods, the ZMP is considered fixed in the middle generated based on the fixed ZMP under the robot foot during
of supporting foot sole during stepping. But, in the natural its stepping. This kind of reference generation reduces the
human walking, ZMP is not fixed and moves forward under the naturalness of the generated pattern. Investigations about
supporting foot during stepping. This paper proposes a pattern
human walking show that the ZMP doesn’t stay fixed under the
generation method using GCIPM with considering moving ZMP
foot during his/her stepping and it moves forward from the heel
under the robot supporting foot. It is supposed that the ZMP
follows a first-order function and moves forward from the heel in to the toe direction [7, 8]. Zhu et al. [9] proposed a method to
the direction of toe. Simulation results show that the reference generate the walking pattern based on LIPM and considering
trajectories which are generated considering the moving ZMP the moving ZMP under the supporting foot. They consider that
cause the smooth movement for COM. the ZMP moves forward, from the heel to the toe of the
supporting foot, and follows a first order function. Erbatur and
Keywords— GCIPM; humanoid robots; pattern generation; Kurt [10] have offered a reference generation algorithm using
natural ZMP LIPM with the moving ZMP under the supporting foot during
stepping. They have used the Fourier series approximation to
I. INTRODUCTION simplify the solution and smooth ZMP reference generation.
In recent years, robot technology is being developed in Because of the more precise dynamics of GCIPM, this
various fields. One of the important topics in robotics field is model is investigated in this paper to generate the walking
bipedal robots. This kind of robots can be substituted for pattern, considering the moving ZMP. It is considered that the
human workers to do several tasks in different conditions. ZMP follows a first-order function and moves forward from
Therefore, bipedal robots should be able to walk stably on the heel in the direction of toe. Robot’s COM reference is
various surfaces. One of the important issues about bipedal obtained from the desired ZMP trajectory. For comparison,
robots is generating walking pattern and studying their stability simulations are carried out for both the pattern generation using
during walking. One of the methods of generating walking fixed ZMP and moving ZMP.
pattern of bipedal robot was proposed based on ZMP equation
[1, 2]. In this method, the precise knowledge about robot II. GRAVITY COMPENSATED INVERTED PENDULUM MODE
dynamics including mass, location of center of mass and inertia (GCIPM)
of each link is required [5]. Also, very complicated dynamic
equations should be solved. Pattern generation for real biped using LIPM causes the
moving of ZMP away from the presumed point during its
Kajita and Tani [3] proposed a simpler model called LIPM. locomotion. This problem occurs because of considering that
In their proposed method the robot is considered as an inverted all of the masses are concentrated at the point of the base link
pendulum and it is assumed that all of the robot’s masses are and neglecting the dynamics of free leg in robot’s locomotion.
concentrated in its center of mass (COM). But real biped leg But the links masses and moments of inertia are not zero in real
have nonzero masses, so when the trajectory generated by biped. Specially, free leg dynamics influence the biped
LIPM is applied to a real biped, the ZMP moves away from locomotion. By using more precise dynamic model for biped in
presumed fixed point and the stability of the biped is pattern generation, more proper trajectory can be generated in
influenced badly because of neglected dynamics of free leg order to apply to the real biped. In GCIPM, biped is considered
motions[6]. Park and Kim [4] proposed another method for as two different masses instead of one mass as in LIPM. One
biped modeling called GCIPM. In their proposed method, the mass is for upper body and supporting foot and the other mass
trajectory is generated based on LIPM by including the effect for free leg. It’s assumed that the mass of free leg is

978-1-4799-6743-8/14/$31.00 ©2014 IEEE


570
concentrated in its foot and the mass of rest of the body is x(t ) = − L cos(ω f t ) (0 ≤ t ≤ T ) (6)
concentrated at the base link [4]. Fig. 1 shows the biped model
in GCIPM that consist of two masses. In this figure m and M
represent the mass of free leg and the rest of body,
respectively. P and p show the hip and free leg position, hf
z (t ) = ⎡1 − cos(2ω f t ) ⎦⎤ (0 ≤ t ≤ T ) (7)
respectively and S represents the ZMP. 2 ⎣
Using this model, the moment equation about ZMP can be
written as: Where L, T, h f and ω f = π are the stride, one step
T
period, maximum height of free leg and stride frequency,
) + q × (mq) = Q × (mg ) + q × (mg )
Q × ( MQ (1) respectively. By solving (2) and (3) the COM trajectory in
pitch and roll direction can be obtained, respectively. In these
equations, ZMP is considered in the fix point during stepping,
Where Q = P−S , q = p−S , P = [ X , Y , Z ]T , to make the solving of differential equations of (2) and (3),
T T easier. The best choice for ZMP is in the middle of supporting
p = [ x, y, z ] , g = [0, 0, − g ] . g represents the gravity vector foot. Fixed ZMP reference in biped locomotion is illustrated in
and the ZMP is located at S = [ x0 , y0 , z0 ]T . It is assumed that Fig. 2. In the trajectory shown in Fig. 2, one step size is 0.2 m,
the distance between foot centers in y-direction is 0.2 m, and
the height of hip is maintained at constant level H z during one step time is 1s.
locomotion, so Z = H z = const and the COM moves in a
constant height. By neglecting the effect of free legs in the III. GCIPM WITH MOVING ZMP DURING STEPPING
motion of biped in lateral direction, (1) can be simplified to:
GCIPM proposed by Park and Kim [4], is simple and
proper model for biped. By considering the fixed ZMP under
X − ω 2 X = − x0ω 2 + H (t ) (2) the supporting foot, the naturalness of the locomotion is
reduced. In real human walking, the ZMP moves forward from
the heel to the toe of the supporting foot. Fig. 3, shows the
ZMP trajectory during human walking.
Y − ω 2Y = − y0ω 2 (3)

Where (2) and (3) represent the ZMP equations in pitch and
roll directions, respectively. Term H(t) represent the effect of
free leg dynamics.

H (t )  λ ( ( x − x0 )(  x ( z − z0 ) )
z + g ) −  (4)

g m
ω and λ  (5)
H z − z0 M ( H z − z0 )
(a)
To obtain the effect of free leg, i.e. H(t), the swing leg
trajectory should be decided in x and z directions. The swing
foot trajectory is considered as (6) and (7) [4].

(b)
Fig. 2. Fixed ZMP references. (a) ZMP reference in x-direction. (b) ZMP
reference in y-direction.
Fig. 1. The biped model in GCIPM with two different masses.

571
In this section the pattern generation using GCIPM Where,
considering moving ZMP is proposed. Only the robot
locomotion in single support phase (SSP) is considered. Also,
1⎛ 1  ⎞
⎜ X (0) + ( X (0) − β ) + α − ρ ⎟
it is supposed that the ZMP is moving only in x-direction A1 = (11)
during stepping and it is fixed in y-direction. So the ZMP 2⎝ ω ⎠
trajectory in x-direction will be a time-variant function. Here,
linear trajectory is considered for ZMP as (8).
1⎛ 1 ⎞
A2 = ⎜ X (0) − ( X (0) − β ) + α − ρ ⎟ (12)
2b ⎛ T ⎞ 2⎝ ω ⎠
x0 (t ) = ⎜t − ⎟ (0 ≤ t ≤ T ) (8)
T ⎝ 2⎠

Where, T is period of SSP and b is the range of ZMP 1⎛ Y (0) ⎞


movement under the supporting foot. It should be noted that B1 = ⎜ Y (0) + ⎟ (13)
the ZMP may move on the line connecting the heel to the toes
2⎝ ω ⎠
without covering this line completely [10]. So b can be smaller
than foot size. The trajectory generated by (8) for period of 4s
is shown in Fig. 4. In this trajectory, b is 0.04 m. The trajectory 1⎛ Y (0) ⎞
generated by (8) is not the most natural ZMP trajectory, but it B2 = ⎜ Y (0) − ⎟ (14)
is similar to ZMP trajectory in human walking and is more 2⎝ ω ⎠
natural than that of fixed ZMP. The ZMP trajectory in y-
direction is the same as Fig. 2(b), because the ZMP is
considered fixed in y-direction. Considering the ZMP
trajectory as (8), the solution to (2) and (3) is: α=
(
b ω2 + λ g ), β=
2
α (15)
2 T
ω
X = A1eωt + A2 e−ωt + β t + ρ cos ω f t − α (9)

ρ=
(
λ L g − ω 2f z0 ) (16)
2
Y = B1e ωt
+ B2 e −ωt
+ y0 (10) ω + ω 2f

ωT ωT
1+ e (2β T − 4α )e β
X (0) =
ωT
(ω X (0) + α − ρ ) + + (17)
1− e 1 − e2ωT ω

ωT
1+ e
Y (0) = ωY (0) (18)
1 − eωT

Fig. 3. Natural ZMP trajectory. IV. SIMULATIONS


In this section, COM trajectories generated by GCIPM for
fixed and moving ZMP are shown and compared with each
other. The trajectory of biped joints for both fixed and moving
ZMP are obtained solving inverse kinematics problem. Fig .5
shows a 6 degree of freedom (DOF) bipedal robot. This robot
is used in simulations to solve the inverse kinematics and
obtaining the joint trajectories during walking. It’s assumed
that the robots swinging foot is parallel with the ground surface
during its walking. Also it is supposed that the upper body is
upright during walking. Also, the simulations are done using
Working Model 4D for fixed and moving ZMP.
Fig .6 shows the biped structure simulated in this software.
It’s supposed that the ZMP moves only in x-direction so the
COM references in y-direction will be the same for both fixed
and moving ZMP. That’s why in this section the simulations
Fig. 4. Moving ZMP trajectory in x-direction. are done only in x-direction to compare the trajectories for

572
fixed and moving ZMP. Simulations are done for T =1s, hf =0.1 The joint trajectories are obtained as shown in Fig. 10.
m, L=0.2 m, b=0.04 m, M=20 kg, Hz =0.55 m and m=1 kg. Simulations are done for li = 0.3m (i = 1,2,…5) and lab = 0.1m.
As illustrated in Fig. 10, the joint trajectories corresponding to
Fig. 7, show the ZMP and COM trajectories generated for moving ZMP have the same pattern with fixed ZMP but with
both fixed and moving ZMP. COM trajectories generated by less variation in joints. To compare the torque needed in joints
moving and fixed ZMP are illustrated in Fig. 8. It is seen that for walking in both fixed and moving ZMP cases, the results
the trajectory generated considering the moving ZMP is obtained using Working Model 4D are shown in Fig. 11. It is
smoother than fixed ZMP and it is near to straight line. Also seen from Fig .11 that the joint torques in moving ZMP is
considering b=0.04 m it can be conclude that the ZMP located decreased. So it can be stated that, the energy consumption for
under the supporting foot. biped locomotion with trajectories generated for moving ZMP
Fig. 9(a) and (b) represents the COM trajectories in is less than the one generated for fixed ZMP and it has optimal
different masses for fixed and moving ZMP, respectively. gaits.
Comparing these figures, it can be concluded that the COM has
less variation and stable motion in moving ZMP. For more
comparison, the mean square errors (MSE) for different masses
with fixed and moving ZMP are shown in Table. 1. It can be
seen that MSE for moving ZMP is less than fixed one. To
compare the joint trajectories for fixed and moving ZMP COM
reference, the inverse kinematics problem is solved for 6-DOF
bipedal robot shown in Fig. 5.

(a)

Fig. 5. The model of bipedal robot.

(b)
Fig. 7. ZMP and COM references in x-direction. (a) ZMP and COM
trajectory for fixed ZMP. (b) ZMP and COM trajectory for moving ZMP.

Fig. 6. The model of biped robot in Working Model 4D.

TABLE I. MSE FOR COM REFERENCE WITH DIFFERENT MASSES

m=0 vs m=2 m=0 vs m=4

MSE for moving ZMP 0.0012 0.0047

MSE for fixed ZMP 0.004 0.0159


Fig. 8. COM trajectories for fixed and moving ZMP.

573
(a)
(c)

(b) (d)
Fig. 9. COM references for different mass of m. (a) COM references for
fixed ZMP. (b) COM references for moving ZMP.

(e)
Fig. 10. Joint trajectories for fixed and moving ZMP. (a) θ 1 , (b) θ2 , (c)
(a) θ 3 − θ 4 , (d) θ 5 , (e) θ6 .

V. CONCLUSION
In this paper, reference trajectory generation for biped,
using GCIPM with moving ZMP has been proposed. By
choosing a linear time-variant function instead of fixed point
for ZMP reference, more natural trajectory for ZMP was
considered. Applying this ZMP trajectory to GCIPM, the COM
trajectory was obtained. The comparison of moving and fixed
ZMP references indicated that for moving ZMP, COM
trajectory is smoother. Also, solving inverse kinematics, the
joint trajectories for 6-DOF bipedal robot were calculated.
Results showed that trajectories for moving ZMP have the
(b) same pattern as fixed ZMP with less variation in joints. Biped

574
walking simulation using Working Model 4D demonstrated
that in moving ZMP, biped need less torques in joints than
fixed ZMP so the pattern generated considering moving ZMP
is optimal.

(e)
Fig. 11. Joint torques for fixed and moving ZMP. (a) θ 1 , (b) θ2 , (c) θ 3 − θ 4 ,
(d) θ 5 , (e) θ6 .
(a)

REFERENCES
[1] A. Takanish, M. Tochizawa, H. Karaki, and I. Kato, “Dynamic biped
walking stabilized with optimal trun and waist motion,” in 1989 Proc.
IEEE/RSJ Int. Workshop on Intelligent Robotics and Systems, pp. 187-
192.
[2] J. Yamaguchi, A. Takanish, and I. Kato, “Development of a biped
walking robot compensating for three-axis moment by trunk motion,” in
1993 Proc. IEEURSJ Int. Workshop on Intelligent Robotics and
Systems, pp. 561-566.
[3] S. Kajita and K. Tani, “Experimental study of biped dynamic walking in
the linear inverted pendulum mode,” in 1995 Proc. IEEE Int. Conference
on Robotics and Automation, pp. 2885-2891.
[4] J. H. Park and K. D. Kim, “Biped robot walking using gravity
(b) compensated inverted pendulum mode and computed torque control,” in
Proc. IEEE Int. Conf. Robot. Autom, 1998, pp. 3528–3533.
[5] S. Kajita, F. Kanehiro, K. Kaneko, K. Fujiwara, K. Harada, K. Yokoi,
and H. Hirukawa, “Biped Walking Pattern Generation by using Preview
Control of Zero-Moment Point”, in 2003 Proc. of IEEE Int. Conference
on Robotics and Automation, pp.1620-1626.
[6] Y. Fujimoto and A. Kawamura, “Three dimensional digital simulation
and autonomous walking control for eight-axis biped robot,” in 1995
Proc. IEEE Int. Conference on Robotics and Automation, pp. 2877-
2884.
[7] A. Dasgupta and Y. Nakamura, “Making feasible walking motion of
humanoid robots from human motion capture data,” in Proc. IEEE Int.
Conf. Robot. Autom., Detroit, MI, May 1999, pp. 1044–1049.
[8] K. Erbatur, A. Okazaki, K. Obiya, T. Takahashi, and A. Kawamura, “A
study on the zero moment point measurement for biped walking robots,”
in Proc. 7th Int. Workshop Adv. Motion Control, Maribor, Slovenia,
(c) 2002, pp. 431–436.
[9] C. Zhu, Y. Tomizawa, X. Luo, and A. Kawamura, “Biped walking with
variable ZMP, frictional constraint, and inverted pendulum model,” in
Proc. IEEE Int. Conf. Robot. Biomimetics, Shenyang, China, Aug.
2004, pp. 425–430.
[10] K. Erbatur and O. Kurt, “Natural ZMP Trajectories for Biped Robot
Reference Generation,” IEEE Transactions on Industrial Electronics,
vol.56, No.3, March 2009.

(d)

575

You might also like