This article was downloaded by: [New York University]

On: 19 October 2014, At: 02:08

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954
Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Advanced Robotics
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/tadr20

Soft Interaction Between Body

Weight Support System and
Human Using Impedance Control
Based on Fractional Calculus
a a a
Yo Kobayashi , Takao Watanabe , Masatoshi Seki ,
a a
Takeshi Ando & Masakatsu G. Fujie
a
Faculty of Science and Engineering , Waseda
University , Tokyo , Japan
Published online: 12 Jul 2012.

To cite this article: Yo Kobayashi , Takao Watanabe , Masatoshi Seki , Takeshi Ando &
Masakatsu G. Fujie (2012) Soft Interaction Between Body Weight Support System and
Human Using Impedance Control Based on Fractional Calculus, Advanced Robotics,
26:11-12, 1253-1269, DOI: 10.1080/01691864.2012.689724

To link to this article: http://dx.doi.org/10.1080/01691864.2012.689724

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information
(the “Content”) contained in the publications on our platform. However, Taylor
& Francis, our agents, and our licensors make no representations or warranties
whatsoever as to the accuracy, completeness, or suitability for any purpose
of the Content. Any opinions and views expressed in this publication are the
opinions and views of the authors, and are not the views of or endorsed by
Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and
Francis shall not be liable for any losses, actions, claims, proceedings, demands,
costs, expenses, damages, and other liabilities whatsoever or howsoever caused
arising directly or indirectly in connection with, in relation to or arising out of the
use of the Content.

This article may be used for research, teaching, and private study purposes.
Any substantial or systematic reproduction, redistribution, reselling, loan, sub-
licensing, systematic supply, or distribution in any form to anyone is expressly
 forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Downloaded by [New York University] at 02:08 19 October 2014
 Advanced Robotics 26 (2012) 1253–1269
tandfonline.com/tadr

Full paper

Soft Interaction Between Body Weight Support System and

Human Using Impedance Control Based on Fractional
Calculus

Yo Kobayashi*, Takao Watanabe, Masatoshi Seki, Takeshi Ando and Masakatsu G.

Downloaded by [New York University] at 02:08 19 October 2014

Fujie
Faculty of Science and Engineering, Waseda University, Tokyo, Japan
Received 26 May 2011; accepted 26 January 2012

Abstract
In recent years, research and development have been conducted on robots designed to assist people with
disabilities in daily activities. There is a great demand for control technology for realizing flexible con-
tact and cooperative behavior. We here report a novel impedance control method based on a fractional
calculation inspired by the viscoelastic properties of biomaterials such as muscle. This paper presents an
evaluation of this concept by simulation and by experiment using a robotic system for body weight sup-
port. The experimental results demonstrated that the fractional impedance controller has superior contact
force absorption performance compared with a conventional controller, especially for high-stiffness
objects and high-velocity movement. This fractional impedance controller may be useful especially for
the purpose of flexible contact for assistive and rehabilitation robots for people.
Ó 2012 Taylor & Francis and The Robotics Society of Japan

Keywords
body weight support, impedance control, viscoelasticity, fractional calculus, flexible contact

1. Introduction

1.1. Background
In recent years, research and development have been conducted on robots
designed to assist people with disabilities in daily activities. However, these reha-
bilitation robots have not yet become popular. In contrast, industrial robots that
specialize in accurately positioning solid targets with well-known mechanical
properties are already prevalent. Unlike industrial robots, rehabilitation robots
have to physically interact with people to assist them in their activities. If solid
*
To whom correspondence should be addressed. E-mail: you-k@fuji.waseda.jp

Ó 2012 Taylor & Francis and The Robotics Society of Japan DOI:10.1080/01691864.2012.689724
 1254 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269

robots that only use position control are used, it is possible to damage the robot
and/or injure the person. Thus, rehabilitation robots require the ability to maxi-
mize a safe user–robot interaction by evaluating the response forces and adjust-
ing accordingly. There is a great demand for control technology for realizing
flexible contact and cooperative behavior.
1.2. Related Work
Force control is generally used to realize flexible contact and cooperative behav-
ior in rehabilitation robots. Force control enables the virtual stiffness of the robot
to be easily changed. This ability is crucial, because the appropriate stiffness var-
ies from person to person. One of the most popular approaches is impedance [1]
Downloaded by [New York University] at 02:08 19 October 2014

or admittance control, which generates a specific end-of-arm stiffness which

takes the form of impedance. For example, Krebs et al. [2] proposed a rehabilita-
tion robot that performs spatial extensions using impedance control. Pledgie
et al. [3] developed an algorithm for suppressing tremors. Blaya et al. [4] present
a variable impedance method for treating drop-foot gait. Jezernik et al. [5] devel-
oped algorithms for automatically adapting the motion of a robotic rehabilitation
device. Veneman et al. [6] present an exoskeleton robot for interactive gait reha-
bilitation. Frey et al. [7] present a novel mechatronic body weight support
(BWS) system with a passive spring and an active electric drive. Watanabe et al.
developed a force control method for a BWS system that provides support to the
pelvis [8,9].
However, there are several technical challenges that need to be overcome for
robots to realize a safe user–robot interaction. A rehabilitation robot could
behave unstably when in contact with a person. This problem arises from the
fact that human movement support requires a large increase (or decrease) in
force in certain situations. Conventional impedance control may have limitations
to behave cooperatively with a person’s fast movements, causing a robot to
impart a large contact force to a person in some situations.
Human arms and legs are better able to contact flexibly and absorb contact
force than robotic limbs, for example, in movement direction of the joints, which
cause a sudden lengthening or shorting of a muscle–tendon structure. Many
explanations have been proposed for why human limbs have this ability to
absorb contact force; these include the ability to adjust stiffness, feed forward
control, and/or learning of the nervous system, and the presence of redundancies
and polyarticular muscles. However, the presence of redundancies and polyartic-
ular muscles affect endpoint stiffness properties of limbs but do not affect their
dynamic characteristics. The large delay in the human neural system also restricts
the effectiveness of stiffness adjustment and control systems for absorbing con-
tact force (loop transmission delays are typically in the range of 100
150 ms).
We consider that the material viscoelasticity of muscle is one of the key fac-
tors accounting for the ability of muscle to absorb contact force. We expect that
 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269 1255

these abilities stem from ‘special’ dynamic properties of human muscle. In fact,
researchers (including our group) have reported that the viscoelasticity of a bio-
material has different properties from conventional serial and/or parallel arrange-
ments of springs (stiffness) and dashpots (viscosity) considered in models such
as Hill’s model [10,11]. These studies suggest that a ‘springpot,’ which is based
on fractional calculus, accurately represents the viscoelastic properties of bioma-
terials, whereas the conventional springs and dashpots viscoelastic model does
not. (Fractional calculus is a branch of mathematical analysis concerned with tak-
ing real or complex number powers of differential operators.)
1.3. Scope of This Study
Downloaded by [New York University] at 02:08 19 October 2014

Based on the above discussion, we hypothesized that a controller that reproduces

the viscoelastic properties of muscle will have a superior ability to realize flexi-
ble contact than a conventional controller. In a previous study [12], we proposed
a novel impedance control based on fractional calculus (the springpot model) to
provide the flexibility of muscles for a robot. We call this method ‘fractional
impedance control,’. This paper presents an evaluation of this concept based on
simulation and experimentation through application in an actual robotic system
for BWS, which partially unloads human body weight during walking (Fig. 1).

2. Approach and Problem Definition

In this paper, we focus on walking movement in the vertical direction, which is
affected by the force generated from a BWS system [8,9]. Figure 2 shows the
developed BWS system with pelvic support, which allows for natural pelvic
motions while walking. The system consists of the BWS system and the pelvic
support mechanism. The BWS system unloads body weight from below on a

Figure 1. Overview of this study.

 1256 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269

Figure 2. BWS system with pelvic support mechanism [8].

Downloaded by [New York University] at 02:08 19 October 2014

pedestal attached to a motor-actuated device (Fig. 2(b)). The DC motor rotates a

ball screw via reduction gears and moves the plates connected to the pedestal.
The mounted load cell measures the unloading force from the pedestal that sup-
ports the body weight. Figure 2(c) shows the pelvic support mechanism, which
can be mounted on the BWS system. This mechanism supports the pelvis by
holding the left and right parts of the anterior superior iliac spine and the ischial
bone and has three passive rotational Degree of Freedoms, as defined in Fig. 2
(c). This mechanism allows passive rotation about the yaw axis, bending move-
ment about the pitch axis, and hip rotation about the roll axis, which allows for
natural pelvic motion while walking.
Figure 3 shows the diagram which defines the vertical unloaded force from
the BWS system. To achieve soft contact between the BWS system and the
human subject during walking, the following points have to be considered. The
BWS system generates a force to partially unload the subject’s body weight. The
target unloaded force is normally set as constant, but the actual measured force
fluctuates because of the interaction between the subject and the BWS system
during walking (Fig. 3, right). Therefore, flexible contact is achieved by

BWS system
Measured Center
Vertical position mm

Time s

Figure 3. Definition of vertical unloaded force in the BWS system.

 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269 1257

suppressing the contact force, which accumulates on the target force, between
the subject and robot.

3. Methods
3.1. Measurement and Modeling of Viscoelastic Properties of Muscle
We have studied biomechanical modeling of the viscoelastic properties of liver
[11]. We present here a viscoelastic muscle model based on this modeling
method. In this part, we evaluate whether the model is suitable for muscle tissue.
Our previous paper [11] gave detailed descriptions of this modeling method.
Thus, this paper gives only a brief description of the method and result.
Downloaded by [New York University] at 02:08 19 October 2014

The following experiments were individually conducted to measure the physi-

cal properties of hog interior muscle tissue (tender loin) using a rheometer (TA
Instruments, Japan: AR-G2). The shear modulus, shear stress, and shear strain
were then calculated based on these measurements.
A dynamic viscoelastic test was performed to measure the frequency response
of the muscle tissue. Sinusoidal stress with angular frequencies in the range 0.1–
10 rad/s (giving a strain amplitude of 3%) was applied to the tissue. From the
results of this test, we obtained the mechanical impedance, the complex shear
modulus G⁄, the storage elastic modulus Gs, and the loss elastic modulus Gl.
Figure 4(a) shows the experimental results of the dynamic viscoelastic tests.
In a previous study using a hog liver [11], we showed that the viscoelastic prop-
erties of liver can be modeled using the fractional derivative expressed in (1),
which considers only the low-frequency characteristics:

dr c
G ¼ s; ð1Þ
dt r

10000 10000
Shear modulus Pa

Shear modulus Pa

1000 1000
Gs Model Gs Model Gs Model
Gl Model Gl Model Gl Model
100 100
0.1 1 10 0.1 1 10
Anguler frequency rad/s Angular frequency rad/s
(a) Experimental result and fractional model (b) spring-dashpot (Voight) model

Figure 4. Experimental result of mechanical impedance. Gs is the storage elastic modulus and Gl
is the loss elastic modulus. The rhombus and square plots are the experimental results, and the red
and blue lines are the responses of our model.
 1258 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269

where G is the viscoelasticity, t is the time, γ is the shear strain, r is the order of
the derivative, and τ is the shear stress.
The parameters in (1) were adjusted to fit the experimental results shown in
Fig. 4(a). Figure 4(a) also shows the viscoelastic properties of the model that
were obtained using (1). For reference, Fig. 4(b) shows the viscoelastic proper-
ties of the conventional spring–dashpot model as modeled using (2):

dc
ac þ g ¼ s; ð2Þ
dt

where α is the elasticity and η is the viscosity. Table 1 shows the values of all
Downloaded by [New York University] at 02:08 19 October 2014

the parameters in (1) and (2).

These results reveal that the viscoelastic properties of the hog muscle tissue
and those of the fractional model have similar tendencies. Therefore, the model
based on (1) can accurately reproduce the viscoelastic responses of muscle tissue.
The fractional order r is equal to 0.15 based on the slopes of Gs and Gl in Fig. 4
(a) (this value for the fractional order 0.15 was used for the fractional impedance
controller in Section 3.2). In contrast, the conventional spring-dashpot model
cannot fit the viscoelastic properties of muscle (Fig. 4(b)).
The fractional and conventional modeling results differ mainly in the loss elas-
tic modulus Gl. Compared with Gl for the fractional model, the Gl for the con-
ventional spring-dashpot model increases rapidly as the frequency increases. It is
clear from (2) that the slope of the loss modulus predicted by the spring-dashpot
model will be 1.0 in a log–log plot. In contrast, the measured slope for a muscle
was 0.15. Thus, the spring-dashpot models are practically useless for modeling
the viscoelastic properties when the loss elastic modulus Gl exhibits a weak fre-
quency dependence, such as for these experimental results. We expect that this
difference in the frequency dependencies will affect the capacity for flexible
contact.
3.2. Formulation and Governing Equation
The fractional controller is based on the impedance controller proposed by
Hogan et al. [1], except for the form of impedance. The form of impedance in
our controller is represented by the spring-dashpot model to reproduce the visco-

Table 1.
Value of each parameter in (1) and (2)

G R
Fractional model (1) 3200 0.15
K C
Spring–dashpot model (2) 3200 480(=3200 × 0.15)
 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269 1259

elasticity of muscle. The objective of this study is to compare the contact force
absorption abilities of the fractional controller and a conventional controller,
which has the spring-dashpot properties represented in (2). Conventional imped-
ance controllers generally include a mass term. For simplicity, we omit the mass
term in this analysis to facilitate comparison with the fractional controller (1),
which does not have a mass term.
Figures 5 and 6 present the analysis model and its block diagram, respectively.
The analysis model consists of the ‘impedance controller’ and the ‘contact object’.
For simplicity, we assume the following that the system is linear, time invariant,
and one dimensional. The robot is controlled by position-based impedance control
(admittance control). The contact object collides with the robot. The impedance
Downloaded by [New York University] at 02:08 19 October 2014

controller generates motion to reduce the contact force between the end-effector
and the object. We evaluate the force between the object and the robot.
(1) Impedance controller. We implemented position-based impedance control
(admittance control), which is actually a position controller nested within a force
feedback loop. Force feedback is employed in the target impedance model to
modify the reference position. We used two impedance controllers in the evalua-
tion. One impedance controller is a conventional controller that consists of a
spring and a dashpot connected in parallel. The differential equation for this
impedance controller is given by:

dxref
Kxref þ C ¼
ðft
fext ðtÞÞ; ð3Þ
dt

C P, k Kext
K ext
x ext
x ext
K

Impedance Impedance Contact object

Robot arm Contact object Robot arm
(spring and dashpot) (springpot)

(a) Conventional impedance control with (b) Fractional impedance control with springpot
spring and dashpot

Figure 5. Conceptual scheme of the analysis model.

Conventional Fractional
Impedance Object Impedance Object
Xt (=0) controller Robot model xext model Xt (=0) controller Robot model xext model
+ xref x + - Fext + xref x + - Fext
Ft (=0) + 1 Closed loop Ft (=0) + 1 Closed loop
vertical -Kext vertical -Kext
- K+Cs + - Psr + BWS axis
BWS axis

(a) Conventional impedance control with (b) Fractional impedance control with springpot
spring and dashpot

Figure 6. Block diagram of the analysis model.

 1260 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269

where C is the virtual viscosity, K is the virtual stiffness, xref is the reference
position of the end-effector, x is the actual position of the end-effector, and fext is
the target force (fext is the force between the contact object and the end-effector).
The other impedance controller is designated to reproduce the viscoelastic
properties of muscle using a springpot model, as expressed by (1). The differen-
tial equation of this impedance controller is respectively given by:

d r xref
P ¼
ðft
fext ðtÞÞ; ð4Þ
dt r

where P represent viscoelasticity and r represent fractional parameter of the

Downloaded by [New York University] at 02:08 19 October 2014

springpot.
(2) Contact object. The elastic contact object collides with the end-effector of
the robot. We regard walking as a periodic movement and assume that the
position of a contact object is given as the following equation:

xext ðtÞ ¼ xs ðsin xtÞ; ð5Þ

where xext is the (external) position of the contact object, xs is its steady state
position, and ω is the angular frequency of the contact object movement. This
modeling is based on the vertical motion of pelvis during walking.
The force generated between the end effector and the contact object is
described by:

fext ðtÞ ¼
Kext ðx
xext Þ; ð6Þ

where fext is the external force, Kext is the stiffness of the contact object, x is the
position of the end-effector, and xext is the position of the contact object.
(3) Spring-dashpot impedance controller. From (3) to (6), the differential
equation between xext and xref is given by:
 
K K C dxref
1þ xref þ ¼ xext ðtÞ: ð7Þ
Kext Kext K dt

The normalized equation of (7) is given as

dxref
ð1 þ AÞxref þ AR ¼ xext ðtÞ; ð8Þ
dt

K C
A¼ R¼ ; ð9Þ
Kext K
 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269 1261

where A is the ratio of the stiffness of the impedance controller to that of the
contact object and R is the ratio of the viscosity to the stiffness.
(4) Fractional impedance control. From (4) to (6), the differential equation
between xext and xref is given by:

d r xr
Kext xr þ P ¼ Kext xext ðtÞ; ð10Þ
dt r

The normalized equations of (10) are given by:

d r xref
xref þ a r ¼ xext ðtÞ; ð11Þ
Downloaded by [New York University] at 02:08 19 October 2014

dt

P
a¼ ; ð12Þ
Kext

where a is the ratio of the stiffness of the impedance controller to that of the
contact object and r is the fractional parameter of springpot (i.e. the viscosity
ratio in the springpot model).
3.3. Method of Simulation Analysis
We analyzed the system performances of the conventional and fractional imped-
ance controllers. We performed time-domain analysis in this evaluation. We com-
pared the force response of each controller for the following states of the contact
object and the robot. We compared the fractional and conventional controllers
using the same values for R in (8) and r in (10), because both parameters have a
similar definition (i.e. viscosity ratio) in each controller. Based on the viscoelastic
properties of muscle tissue described in Section 2, R in (8) and r in (10) for the
impedance controllers were set to 0.15 (ref. Section 3.1 and Table 1). We ana-
lyzed the force response with three different stiffness ratios A and a. Specifically,
the stiffness ratios A and a were set to 1.0 and 2.0. We also analyzed the force
response with two different angular frequency ω (1.0, 1.5). As shown, the
parameters that determine the force response are {A (or a), ω} in this investiga-
tion. We repeatedly calculated the external force for each impedance controller
by changing these parameters.
3.4. Method of Experiment
This section describes experiments using the rehabilitation robot to evaluate frac-
tional impedance control. We implemented fractional and conventional control-
lers in the rehabilitation robot of a BWS system with a pelvic support
mechanism (Fig. 2) presented in [8,9], and measured their respective contact
force absorption performances. We performed the following experiment to
 1262 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269

evaluate the fractional impedance controller, for ensuring flexible contact and
cooperative behavior with people.
Figure 7 depicts the conceptual scheme and experimental setup. The BWS
system was placed on the floor, and the subject climbed onto the system and
stood on a separated treadmill, which was developed in our laboratory. The
movement of the pedestal L was impedance controlled, corresponding to the
force perpendicular to the axis of the pedestal (=unloaded force). Both fractional
and conventional impedance controllers were individually implemented in the
system. The respective viscosity ratios R and r in these impedance controllers
were set to 0.15, as in the numerical analysis.
The force response was measured with two different magnitudes of the con-
Downloaded by [New York University] at 02:08 19 October 2014

troller parameters K and P: low (0.5 × 103 N/m) and high (1.0 × 103 N/m). The
1.0 × 103 N/m magnitude of the controllers was chosen based on trial and error,
and we decided on the other magnitude, 0.5 × 103 N/m, to provide softer proper-
ties (the stiffness ratio a and A was relatively high). We also measured the force
response with two different walking speeds: slow (2.0 km/h) and fast (3.0 km/h).
The walking speed was chosen based on the average walking speed of healthy,
elderly people (about 2.0 km/h) and young people (about 3.0 km/h). The walking
gait cycle was controlled by treadmill motion. The subject of this experiment
was a healthy young male, who agreed with the objectives and the risks of the
experiment. The subject’s body weight was 65 kg. The target position Lt was set
to the pedestal position, where the subject stood naturally. The target unloading
force Ft was set as 50 N. We simply set the minimal value that the subject felt to
be unloaded, because the objectives of the experiment were to compare the per-
formance of the fractional and conventional controller. Table 2 compares the
parameters used in the model analysis in Section 3.3 with those of the experi-
ment in this section.

Figure 7. Conceptual schema and setup of the experiment.

 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269 1263

Table 2.
Correspondence table

Model analysis Experiment

Position of end-effector (pedestal) X L
Target position of end-effector xt Lt
Reference position of end-effector xref Lref
Force between contact object and end-effector fext fext

The controller of the robot was made by the authors using the C++ program-
ming language. The operating system (OS) used in the experiments was QNX,
Downloaded by [New York University] at 02:08 19 October 2014

which is a real-time OS. The sampling period was 3 ms in each experiment.

4. Results and Discussion

4.1. Simulation
Figure 8 displays the results for the external force fext, with the angular fre-
quency ω = 1.0 Hz and the stiffness ratios having the following values: A and
a = 1.0 (Fig. 8(a)), A and a = 2.0 (Fig. 8(b)), and with the angular frequency
ω = 1.5 Hz and the stiffness ratios having the following values: A and a = 1.0
(Fig. 8(c)), A and a = 2.0 (Fig. 8(d)). Figure 8 shows that for each value of A
and a, the maximum external force fext was lower when the fractional impedance
controller was used. The force with conventional impedance control becomes
large as the stiffness ratio increases and the angular frequency increases.
4.2. Experiment
Figure 9 shows the experimental results for the slow walking speed and high-
stiffness parameter in the impedance controller. The unloaded force fext changed
in accordance with the gait cycle. The experimental results revealed that the bot-
tom peaks of the unloaded force fext were about 50 N, which was the target
unloading force Ft. The amplitude of the unloaded force fext for the fractional
controller was lower than that for the conventional controller.
Figure 10 shows the experimental results for the fast walking speed and high-
stiffness parameter in the impedance controller. The bottom peaks of the unloaded
force fext were also about 50 N. These results demonstrate that the unloaded force
for the fractional controller was almost the same as that with the slow walking
speed shown in Fig. 10. In contrast, the unloaded force for the conventional
controller was substantially increased compared with that recorded with the slow
walking speed.
Figure 11 shows the experimental results for the slow walking speed and
low-stiffness parameter in the impedance controller (the stiffness ratio a and A
was relatively high). The bottom peaks of the unloaded force fext were again
 1264 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269

Object Position Object Position

Force (Fractional) Force (Fractional)
Force (Conventional) Force (Conventional)
1.5 1.5 1.5 1.5

External force fext N

External force fext N

Object position xext

Object position xext

1 1 1 1
0.5 0.5 0.5 0.5
0 0 0 0
-0.5 -0.5 -0.5 -0.5
-1 -1 -1 -1
-1.5 -1.5 -1.5 -1.5
0 1000 2000 0 1000 2000
Downloaded by [New York University] at 02:08 19 October 2014

Time t ms Time t ms

Object Position Object Position

Force (Fractional) Force (Fractional)
Force (Conventional) Force (Conventional)
1.5 1.5 1.5 1.5
External force fext N

Object position xext

Object position xext

External force fext N

1 1 1 1
0.5 0.5 0.5 0.5
0 0 0 0
-0.5 -0.5 -0.5 -0.5
-1 -1 -1 -1
-1.5 -1.5 -1.5 -1.5
0 1000 2000 0 1000 2000
Time t ms Time t ms

Figure 8. Simulation result of time-domain analysis for different stiffness ratio a and A, different
angular frequency of contact object ω.

Figure 9. Experimental results for a walking speed of 2.0 km/h and with controller parameters K
and P set at 1.0 N/m.
 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269 1265

Figure 10. Experimental results for a walking speed of 3.0 km/h and with controller parameters K
and P set at 1.0 N/m.
Downloaded by [New York University] at 02:08 19 October 2014

about 50 N. This result shows that the amplitude of the unloaded force for the
fractional controller was almost the same as that with the high-stiffness parame-
ter in Fig. 11. In contrast, use of the conventional controller resulted in a large
increase in the unloaded force when compared with the result with the high-
stiffness parameter in Fig. 11.
4.3. Discussion, Contributions
The simulation results reveal that the fractional controller provides better contact
force absorption than the conventional controller. These results show that the
fractional controller has superior contact force absorption for an elastic object
that has a high stiffness and for rapid input (object position xext).
The experimental results in Fig. 9 suggest that the fractional controller has
superior flexible contact with a person than does the conventional controller,
because the fractional controller did not generate a large contact force. The
experimental results in Fig. 10 demonstrate that the performance of the fractional
controller is superior to that of the conventional controller, especially for rapid
input (high walking velocity). The experimental results in Fig. 11 demonstrate

Figure 11. Experimental results with low-stiffness parameter in the impedance controller: walking
speed was 2.0 km/h and controller parameters K and P were 0.5 N/m. The stiffness ratio a and A
was relatively high.
 1266 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269

that the performance of the fractional controller is superior to that of the conven-
tional controller, especially for low-stiffness BWS. As was the case for the
numerical analysis, the experimental results suggest that the fractional impedance
controller shows superior performance in flexible contact compared with the con-
ventional controller, especially for high-stiffness objects with high-velocity
movement.
These results support the conclusion that the fractional impedance controller
may have advantages for use in assistive and rehabilitation robots where the
manipulation targets are human because human movements have variables
stiffness and velocity.
Downloaded by [New York University] at 02:08 19 October 2014

4.4. Discussion, Limitations

The robot dynamics give rise to a certain delay in the reference position of the
controller in simulation evaluation. However, we analyzed here the force
response with the analysis when the robot’s position is close to the reference
position. The evaluation with robot dynamics will be investigated in future
studies.
We focus here on only the viscoelasticity of muscle and exclude other factors
such as those mentioned above and the contractile element in Hill’s model; these
aspects will be considered in future studies.
We report here the analysis for when the fractional parameter of springpot r is
0.15, which is based on the muscle properties. Obviously, fractional impedance
control using a different value for the fractional parameter r (including variable
impedance control) is conceivable and may generate new insights. This will be
investigated in future studies.

5. Conclusions and Future Work

We reported a novel impedance control method based on a fractional calculation
inspired by the viscoelastic properties of biomaterials such as muscle. This paper
presents an evaluation of this concept using simulations and experiments. The
numerical analysis results demonstrate that a fractional impedance controller has
superior flexible contact performance compared with a conventional controller
for contact with elastic objects, especially for high-stiffness objects and high-
velocity movement. Moreover, the experiment using a robotic system for BWS
demonstrates the effectiveness of a fractional controller to moderate and reduce a
potentially large contact force between subject and robot.
We reported here the analysis when the fractional parameter of the springpot r
is 0.15 and evaluated only the silicone with these properties. The use of frac-
tional impedance control with different values of the fractional parameter r
(including variable impedance control) is conceivable and may generate new
insights.
 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269 1267

References
1. N. Hogan, Impedance control: an approach to manipulation: part i, ii and iii, J. Dyn. Syst.
Meas. Control 107, 1–23 (1985).
2. H. I. Krebs, M. Ferraro, S. Buerger, M. J. Newbery, A. Makiyama and M. Sandmann, Rehabilita-
tion robotics: pilot trial of a spatial extension for MIT-Manus, J. Neuroeng. Rehab. 1, 5 (2004).
3. T. Pledgie, K. E. Barner, S. K. Agrawal and T. Rahman, Tremor suppression through
impedance control, IEEE Trans. Rehab. Eng. 8, 53–59 (2000).
4. J. A. Blaya and H. Herr, Adaptive control of a variable-impedance ankle-foot orthosis to assist
drop-foot gait, IEEE Trans. Neural Syst. Rehab. Eng. 12, 24–31 (2004).
5. S. Jezernik, G. Colombo and M. Morari, Automatic gait-pattern adaptation algorithms for reha-
bilitation with a 4-DOF robotic orthosis, IEEE Trans. Robot. Autom. 20, 574–582 (2004).
Downloaded by [New York University] at 02:08 19 October 2014

6. J. F. Veneman, R. Kruidhof, E. E. G. Hekman, R. Ekkelenkamp, E. H. F. Van Asseldonk and

H. van der Kooij, Design and evaluation of the LOPES exoskeleton robot for interactive gait
rehabilitation, IEEE Trans. Neural Syst. Rehab. Eng. 15, 378–386 (2007).
7. M. Frey, G. Colombo, M. Vaglio, R. Bucher, M. Jörg and R. Riener, A novel mechatronic
body weight support system, IEEE Trans. Neural Syst. Rehab. Eng. 14, 311–321 (2006).
8. T. Watanabe, E. Ohki, Y. Kobayashi and M. G. Fujie, Leg-dependent force control for body
weight support by gait cycle estimation from pelvic movement, in: Proc. 2010 IEEE Int. Conf.
on Robotics and Automation, pp. 2235–2240 (2010).
9. T. Watanabe, E. Ohki, T. Ando and M. G. Fujie, Fundamental study of force control method
for pelvis-supporting body weight support system, in: Proc. 2008 IEEE Int. Conf. on Robotics
and Biomimetics, pp. 1403–1408 (2008).
10. R. L. Magin, Fractional calculus models of complex dynamics in biological tissues, Comput.
Math. Appl, pp. 1586–1593 (2009).
11. Y. Kobayashi, J. Okamoto and M. G. Fujie, Physical properties of the liver and the develop-
ment of an intelligent manipulator for needle insertion, in: Proc. 2005 IEEE Int. Conf. on
Robotics and Automation, pp. 1644–1651 (2005).
12. Y. Kobayashi, T. Ando, T. Watanabe, M. Seki and M. G. Fujie, Fractional impedance control
for reproducing the material properties of muscle, in: Proc. 2010 IEEE Int. Conf. on Intelligent
Robots and Systems, pp. 5498–5504 (2010).

About the Authors

Yo Kobayashi was born in 1981. He received the BS and MS degrees from the
Graduate School of Science and Engineering, Waseda University, Tokyo, Japan, in
2004 and 2005, and the PhD degree in engineering from Waseda University in
2008. He was a Visiting Research Associate in the Graduate School of Science
and Engineering, Waseda University, from 2005 to 2006. He was a recipient of a
Japan Society for the Promotion of Science (JSPS) Research Fellowship for Young
Scientists during 2007. During 2008, he was a Research Associate in the Institute
for Biomedical Engineering, Waseda University. He has been a Research Associate
during 2009 and Lecturer (junior researcher) with the Faculty of Science and
Engineering since 2010. His current research interests include medical robots and rehabilitation
robots. He is a member of IEEE, the Japan Society of Mechanical Engineers, the Society of
Instrument and Control Engineers, the Robotics Society of Japan, and the Japan Society of
Computer-Aided Surgery.
 1268 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269

Takao Watanabe received the BS and MS degrees from the Graduate School of
Science and Engineering, Waseda University, Tokyo, Japan, in 2006 and 2009
respectively. He is currently with the Department of Advanced Science and
Engineering, Waseda University, where he is involved in studies of medical robots
and rehabilitation robots.

Masatoshi Seki was born in 1985. He received the BS degrees from The System
Downloaded by [New York University] at 02:08 19 October 2014

Design Engineering Department, Keio University, Tokyo, Japan, in 2007. And, he

received the MS degrees from the Graduate School of Science and Engineering,
Waseda University, Tokyo, Japan, 2009. He is Research Fellow of Japan Society for
the Promotion of Science and PhD Student of the Graduate School of Science and
Engineering, Waseda University. His current research interests include welfare
robots and orthotic robots. He is a member of the Japan Society of Mechanical
Engineering, and Robotics Society of Japan.

Takeshi Ando received the BS and MS degrees from the Graduate School of
Science and Engineering, Waseda University, Tokyo, Japan, in 2006 and 2008,
and the PhD degree in engineering from Waseda University in 2011. He was a
Visiting Research Associate in the Graduate School of Science and Engineering,
Waseda University, from 2008 to 2011. He has been an appointed Fellow,
Department of Robotics and Design for Innovative Healthcare (Panasonic),
Graduate School of Medicine, Osaka University since 2011. His current research
interests include medical robots, rehabilitation robots, and nursing engineering. Dr
Ando is a member of the IEEE, JSME, and RSJ. He received the First Prize in the
IEEE Engineering in Medicine and Biology Society (EMBS) Japan Young Investigators
Competition at EMBS2008, the 2008 IEEE RAS Japan Chapter Young Member Award, JSME
ROBOMEC Award (2010) and was a Finalist for the Best Conference Paper and the Best Student
Paper at the Second IEEE/RAS-EMBS BIOROB 2008.

Masakatsu G. Fujie received the MS degree from the Graduate School of

Science and Engineering, Waseda University, Tokyo, Japan, in 1971, and the PhD
degree in engineering from Waseda University in 1999. From 1971 to 2000, he
was with the Mechanical Engineering Research Laboratory, Hitachi Ltd., where he
was a Senior Researcher during 1984 and a Principal Researcher and Project
Leader for the Medical and Welfare Apparatus Development Project during 1995.
He was also the Head of Researchers in the Mechanical Engineering Research
Laboratory, Hitachi Ltd., and the Director of the Medical and Welfare Apparatus
Development Research Laboratory during 1999. Since 2001, he has been a
Professor with the Faculty of Science and Engineering, Waseda University, where he is the
Director of The Grobal Robot Academia for MEXT Grobal COE Program. His current research
interests include surgical robots, image-guided surgery, endoscopic surgery, and assistive and
rehabilitation robots. He is a Fellow of the Japan Society of Mechanical Engineering (JSME) and
 Y. Kobayashi et al. / Advanced Robotics 26 (2012) 1253–1269 1269

a member of the International Society of Computer-Aided Surgery. He received the Atomic Energy
Society of Japan’s Technology Development Award (1991), the Notable Invention Award from the
Science and Technology Agency Director General (1994), the Technical Innovations Awards from
the Robotics Society of Japan (1999), the JSME Medal for New Technology from the Japan
Society of Mechanical Engineers (2000), the Nikkei BP Technology Award, Medical/
Biotechnology Division (2000), the Japan Society of Mechanical Engineers, Robotics and
Mechatronics Division, Technical Achievement Award (2000), the Japan Society of Mechanical
Engineers ROBOMEC Award (2006), and the IEEE Harashima Award (2008).
Downloaded by [New York University] at 02:08 19 October 2014