Proceedings of the 2000 IEEE/RSJ

International Conference on
Intelligent Robots and Systems

Development of a Light-Weight Biped Humanoid Robot

Atsushi Konno Noriyoshi Kat0 Satoshi Shirata Tomoyuki Furuta
Masaru Uchiyama

Department of Aeronautics and Space Engineering

Tohoku University
Aramaki-aza-Aoba 01, Aoba-ku, Sendai 980-8579, JAPAN
[email protected] p

Abstract In this work, a light-weight biped humanoid robot

named Saika-3 is developed. Maximum weight of Saika-
I n order t o meet the demand of a research plat- 3 will be a little over fifty kilograms (it weighs 46.9 [kg]
f o r m f o r intelligent robotics, a humanoid robot named without arms at present). Since the total weight of in-
Saika-3 i s developed. It is a self-sustaining biped robot ternal batteries is twelve kilograms, twelve kilogram-
with a dual a r m and a robotic head. It i s equipped s can be reduced when external batteries are used.
with a PC ( I B M PC/AT clone), batteries, a wireless Saika-3 can be considered as a PC equipped with the
ethernet modem, a gyroscope and motor drivers inside biped legs and the dual arms. Since the RT-Linux [4]
i t s body. Saika-3 can be teleoperated via wireless eth- is chosen as the operating system of Saika-3, users can
ernet communication. A graphic dynamic simulator log in Saika-3 via wireless ethernet communication.
i s also developed t o study control strategy. This paper This paper is organized as follows. An outline of
presents a n outline of the hardware of Saika-3 and a the hardware design of Saika-3 is given in Section 2.
simulation result. The way to measure the reactive force from the ground
and the way to compute Zero Moment Point (ZMP)
are described in Section 3. In Section 4, the devel-
oped graphic simulator and the simulation results are
1 Introduction presented. Section 5 is the conclusion.
Nearly thirty years have passed since the first biped
humanoid robot in the world, WABOT-1, was devel-
oped in 1973 by Kat0 and others [l]. Despite the long
history of development, a practical biped humanoid 2 Hardware Design of Saika-3
robot had been considered as a dream because of the
restrictions of hardware.
However, the progress in the hardware design of
humanoid robots has been notable in the recent few Saika-3 in the final state will have 2-DOFs at eyes,
years. Practical biped humanoid robots are becoming 7-DOFs at each arm, 1-DOF at each gripper and 6-
feasible [2, 31. Component technologies in robotics DOFs at each leg. Total degrees of freedom are 30
such as vision, hand-eye coordination, walking and (Figure 1 (a)). Saika-3 in the present state has the
artificial intelligence are going to be integrated onto head, the biped legs and the body. The dual arms
humanoid robots. Robotics seems to be at a turning and grippers are under development.
point. Its focus has been shifting from the develop- The height of Saika-3 from the sole to the tip of
ment of component technologies to the integration of the antenna is 1271 [mm] (Figure 1 (b)). The body,
component technologies. the leg and the head weigh 18.5 [kg], 7.7 [kg] and 1.0
Although the progress in the hardware design was [kg], respectively. As mentimed in the introduction,
remarkable, humanoid robots developed so far were the total weight of Saika-3 in the present state is 46.9
too heavy and bulky to be handled by a person. A [kgl.
massive robot would be quite dangerous if it walked The details of each part are described in the follow-
around the human society. ing subsections.

- 1565-
0-7803-6348-5/00/$10.0002000 EEE.
 Table 1: Movable limits of the joints (["I).
Adduction Abduction Extension Flexion
Ankle 100.1 45.2 45.7 103.5
Knee 0 0 0 151.2
HiD 99.8 48.2 48.2 103.5

(a) Location of DOFs. (b) CAD model.

(a) Layers of the sole. (b) Allocation of the FSR.
Figure 1: Saika-3 in the final state.
Figure 2: Structure of the sole.

2.1 Robotic Head

Since the motor drivers and the P C generate con-
Two N ~ CCD cameras ~ are used for~ siderable- heat, a heat ~sink is needed.
~ Especially, 12
~ ~
stereo vision. Each of the camera has a resolution of motor drivers for biped legs are supposed to generate
38 mega-phels. The camera arrangementhas 2-DOFs an intense heat, and thus two heat pipes and a fan are
for panning and tilting, The movable limit around used for the heat sink. For the Pc and other motor
tilt axis is *g2 ["I, while the limit around pan axis is drivers, four fans are attached to the rear panel of the
f 1 8 0 I"]. The robotic head consists of 9 mechanical body.
component parts. Power switches and a connector for external power
supply are attached to the ceiling of the body.

2.2 Biped Legs 2.4 Electronic Devices

In designing legs, attention is given t o standardiz- Motor drivers, a PC, a hard disk, a gyroscope, a
ing parts as much as possible for cost-efficiency and video transmitter, the video field multiplexing device
to increasing the movable limits of each joint. For in- [5], batteries, DC-DC converters and a wireless ether-
stance, since the allocations of DOFs at the hip and net modem are mounted inside the body.
at the ankle are similar, many parts were standard- The P C is an IBM PC/AT clone equipped with the
ized. As a result of standardization, the number of Intel Pentium MMX 233 [Mhz]. Two Fujitsu RIF-
varieties of parts is only 65 although the legs consist 01A interface cards are used for D/A conversion, A/D
of 246 mechanical component parts. conversion and pulse counting.
Movable limits of the joints at the ankle, the knee A pair of batteries connected in serial supplies 24
and the hip are presented in Table 1. Considering the [VI, 4.5 [Ah]. Three pairs of batteries, i.e. six batter-
movable limit of the human knee, the flexion limit at ies, can be mounted in maximum. Electronic devices
knee is set at zero. consume 9.63 [A] for 5 [VI, 2.64 [A] for +12 [VI, and
Each joint consists of a DC servo motor, a harmonic 0.18 [A] for -12 [VI.
drive reduction gear, a rotary encoder, a synchronous The gyroscope measures three translational accel-
belt driver and pulleys. eration and three rotational angles of the body.

2.3 Body 2.5 Sole

The mechanical structure of the body consists of 31 In order to measure the reaction force from the
mechanical component parts. ground, Force Sensing Resistors (FSR) supplied by IN-

-1566-
 VOUI

-
(a) Experimental circuit. (b) Force vs. Resistance.

Figure 4: Preliminary experiment.

(a) Sitting pose. (b) Standing pose.

Figure 3: Appearances of Saika-3 in the present state.

Figure 5: Measurement of ZMP.

TERLINK ELECTRONICS are used.
As shown in Figure 2 (a), the sole consists of sev-
en layers. Twelve FSR devices are attached to the voltage of 5 [VI is supplied as VREF (see Figure 4 (a)).
aluminum plate as shown in Figure 2 (b). FSR de- The force vs. resistance characteristic measured in this
vices and rubber sheets are sandwiched between the experiment is plotted in Figure 4 (b). A linear char-
two sheets of aluminum plate. acteristic can be seen in the range of force between 1
In order to absorb the shock of walking, a styrene and 7 kilograms. In this range, the force F applied to
form and a sorbothane sheet are utilized. the active surface of FSR can be approximated by the
following linear equation.
2.6 Present State of Development
F = - 0 . 7 1 4 3 V o ~+
~ 0.701. (1)
Up to the present, the whole body except arms has
been developed. The appearances of Saika-3 in the
present state are shown in Figure 3. 3.2 Principle of Measurement of ZMP

The FSR devices measure the reaction force from

3 Measurement of Zero Moment Point the ground as shown in Figure 5. p i = [ x i , yiIT de-
notes the places where the FSR devices are attached,
This section addresses the way to measure the Ze- while p Z M P= [ X Z M P , y z ~ p denotes
] ~ the ZMP.
ro Moment Point (ZMP) based on the Force Sensing The ZMP is defined as the point where the sum-
Resistors (FSR) devices. Preliminary experimental re- mation of all the moments affecting the body become
sults are also presented. zero. However, the ZMP is equivalent t o the center
of the reaction force from the ground. Therefore, the
3.1 Measurement of Force ZMP can be computed as follows:
n n
The FSR device is a polymer thick film device which
exhibits a decrease in resistance with an increase in the
force.
In order to investigate the force vs. resistance char- fa
acteristic, a circuit shown in Figure 4 (a) is used. The i=l i=l

- 1567-
 100WI
@A-p1a*
........

Y bxFSR Rubbersheet

(a) The way of measurement.

96 .......... ............. .........

72 ............... ...............

48 ............. ............... ............. ............. ........ Right leg Left leg Right leg Left leg

24 ............. ............. .......... .........

(a) Link-fixed coordinates. (b) Link parameters.

a
Figure 7: Model of the simulator.
0 Y d

0 33.5 67 100.5 134 167.5 201

X [mm1
(b) Measurement error.

Figure 6: Preliminary measurement of ZMP. Table 2: Structural Parameters (Iml).

1Oy ~ O Z 11, 13z 141: l6x
0.082 0.0035 0.149 0.25 0.25 0.106
3.3 ZMP Measurement
A Preliminary measurement of ZMP is performed. a quadratic programming (QP) problem so as to min-
As shown in Figure 6 (a), force of 100 [NI is applied t o imize the energy under the given conditions [6].
each grid point shown in Figure 6 (b). The length of
the solid lines plotted at each grid shows the measure- 4.2 Modeling
ment error. The maximum and average measurement
errors were 8 [mm] and 4 [mm], respectively. The base-link-fixed coordinate system is indicated
by C B , while the ground-fixed coordinate system is
indicated by X I . The link-fixed coordinate system is
4 Dynamic Simulation introduced as shown in Figure 7 (a). Link 13 repre-
sents the upper body.
4.1 Dynamic Simulator The structural parameters are defined in Figure 7
(b) and presented in Table 2. The other link parame-
For the purpose of the evaluation of hardware de- ters used in the simulation are presented in Table 3.
sign and off-line planning of walking patterns, a dy-
namic simulator is developed. The simulator is devel- 4.3 Simulation Result
oped on the basis of MATLAB, which is a language of
technical computing. An inverse dynamic simulation result is presented
A legged robot can be considered as a free-flying in this section. In t h e inverse dynamic simulation, ref-
robot that has interaction with the ground [6]. The de- erence trajectories of the base and each link -are given
veloped simulator partly utilizes the Spacedyn [7] that t o the simulator as inputs. As a result of simulation,
is a MATLAB toolbox for space and mobile robots. the trajectory of ZMP and the joint torque which is
The Spacedyn was developed a t Tohoku University and needed in order t o follow the given motion are ob-
has been distributed from [7]. tained.
The plastic collision model is used to simulate walk- The reference trajectories of the base and each link
ing. The plastic collision can be simulated by solving are previously computed so as t o realize a walking pat-

- 1568-
 Table 3: Link parameters.
Tensor of inertia [kgmm2] Y
Link Mass [kg]
Ixx I,, I,,
base 2.325 15956 4673 18449
.-
I

link1 1.109 3392 3651 1182 , .... ..

link2 0.648 283 291 283 ~ ..- ... 3

link3 1.649 2390 10345 9226

link4 1.664 2055 10081 9307
Time [SI
link5 0.467 182 182 210
link6 3942 3288 1464 (a) Base trajectory
1.351
link13 20.000 40000 40000 40000 06 . , . , . . . . , .
Right leg
.

iO:
-- Left leg

tern of 0.1 [m/step] and 1.5 [s/step]. The height of the a"
base tg is set constant, i.e. ZB = 0.793 [m].
The given base trajectory and foot trajectories along 0 ' Time [SI
x and z axes are shown in Figure 8. The obtained join-
t torque of the right leg is plotted in Figure 9. The (b) Foot trajectories along x axis
3D animation of the simulation result is shown in Fig-
ure 10.
The maximum torque needed to realize the given
walking pattern is less than 5 [Nm]. Since the recom- Y

mended maximum torque of harmonic reduction gear

used for legs is 84 [Nm], there is an sufficient margin.
In fact, the legs are designed so as to realize the stand-
up motion from the sitting pose shown in Figure 3 (a). 0 ,-' 0 , ' ,
I _ . . I . I . I .
0 4 6
Time [SI
(c) Foot trajectories along z axis
5 Conclusion
Figure 8: Reference trajectories.

This paper has addressed the development of the

biped humanoid robot Saika-3. The main features
of Saika-3 are its light-weight (approximately 50 [kg])
I ' *
- Jomt I
' " " I '
-- Jmnt 4
5 ... . Join, 2 Jam1 5
JOlnt 3 ........ Joint
and the PC-based system. By adopting PC as a main 6

computer, the development of software becomes less

expensive and faster. Moreover, since Saika-3 has a
wireless ethernet modem, it can be teleoperated.
Force Sensing Resistor (FSR) devices were used t o
measure the reaction force from the ground. The cost
of the FSR devices is much lower than the other de-
-51 ' ' ' ' ' ' ' ' ' ' ' ' ' '
Time [SI
vices such as force sensors and load cells.
A graphic dynamic simulator is developed. A walk- Figure 9: Joint torque history of the right leg.
ing pattern was simulated. The simulation result is
also presented in this paper. The simulator will be
used to produce motion patterns. Acknowledgments
The arms are under development at present. The
development of arms will be completed in a few months. This research has been partly supported by grants
of OMRON corporation (No. 991011).

- 1569-
 (h) 7 [SI

Figure 10: Simulation result of a walking.

References [4] http://rtlinux.cs.nmt.edu/ rtlinux/

[l] I. Kato, “Development of WABOT-1,” Biomech- [5] Y. Matsutmoto, et al., “Real-time Color Stereo
anism 2, The University of Tokyo Press, pp. 173- Vision System for a Mobile Robot based on Field
214, 1973 (In Japanese). Multiplexing,” In Proc. of 1997 IEEE Int. Con-
f. on Robotics and Automation (ICRA ’97), p
[2] K. Hirai, “Current and Future Perspective of Hon- p. 1934-1939, 1997.
da Humanoid Robot,” In Proc. of the IEEE/RSJ
Int. Conf. on Intelligent Robots and Systems [6] Y. Fujimoto and A. Kawamura, “Simulation of
(IROS ’97), pp. 500-508, 1997. an Autonomous Biped Walking Robot Including
Environmental Force Interaction,” IEEE Robotics
[3] J. Yamaguchi and A. Takanishi, “Design of Biped and Automation Magazine, Vol. 5 , No. 2, pp. 33-
Walking Robot Having Antagonistic Driven Joints 42, 1998.
Using Nonlinear Spring Mechanism,” In Proc. of
the IEEE/RSJ Int. Conf. on Intelligent Robots and [7] http://www.astro.mech.tohoku.ac.jp/spacedyn/
Systems (IROS ’97), pp. 251-259, 1997.