Development of a 3D Interactive Virtual Market System with Adaptive Treadmill Control

Abstract—As the computer vision and virtual reality (VR)
technology become mature nowadays, there is a huge potential to
connect VR with the real world. In this paper, a 3D interactive
VR system with adaptive treadmill control is developed. The
subject walks on the treadmill at will. Interaction force between
feet and treadmill are measured to estimate the subject’s
intended walking speed. The treadmill belt velocity and 3D
display adjust to the walking pace according to the subject’s
walking intention, which makes the subject feel that he (or her)
is walking in the real market. The experiment shows that the
control results are smooth, which verifies the validity of the
whole system.
I. INTRODUCTION
ITH the development of computer science, especially
computer vision and virtual reality (VR) technology,
more and more VR systems have come out. However, most of
them were designed for computer games and special trainings,
like medical training [1-2], social skill training [3]. In fact, the
VR system can be much more valuable and useful when it is
connected with real-world. Take service for the elderly people
for example, the traditional service for the elderly people is
non-interactive. For instance, since the physical strength of the
elderly person is weak, their goods are delivered by express in
normal cases. If things continue this way, the elderly people
are lack of mental and physical exercises. Compared with the
mentioned non-interactive service, VR interactive system can
be designed in the following way that an elderly person takes
walks in the VR world by taking steps on the real-world
treadmill. Furthermore, he (or she) can buy things directly in
the VR shops which connect with the selling system of the real
shops. Such interactive service provides the elderly people
with convenience and health, which is the final objective of
our research. This paper is the very first step of the whole idea
which develops a 3D interactive VR system with adaptive
treadmill.
Actually, this paper concerns on the issue noted as
locomotion interface which is defined as travelling through a
virtual environment by subject’s self-propulsion [4-5]. Here
we list the researches relating with treadmill-style locomotion
interface. The early work was done by Noma et al. who built a
treadmill locomotion interface system ATLAS [6] where
stance duration and body position are applied in PI control for
Manuscript received March 5, 2010.
Haiwei Dong, Tatuo Oshiumi, Akinori Nagano and Zhiwei Luo are with
the Graduate School of Engineering, Kobe University, Kobe, 657-8501,
Japan (e-mail: haiwei@stu.kobe-u.ac.jp; oshiumi@cs11.cs.kobe-u.ac.jp;
aknr-ngn@phoenix.kobe-u.ac.jp; luo@gold.kobe-u.ac.jp).
driving the treadmill. In [7], Darken and Carmein. developed
the first omini-directional treadmill to facilitate turning. In [8],
Iwata and Yoshida built another two dimentional treadmill
which employs twelve small treadmills to form a large belt. In
[9-10], Hollerbach et al. designed two generaltions of
treadmill-style locomotion interfaces named Sarcos Treadport
which imitates slope by pushing or pulling a tether located at
the back of the subject.
In this paper, when one subject walks on the treadmill, the
force plate measures the forces in x, y, z directions exerted by
the subject. Based on the sensor data, we estimate the
subject’s intended walking speed and adjust the velocity of the
treadmill belts by it. Meanwhile the 3D projector goes on with
the walking pace. Compared with previous treadmill-style
locomotion systems, there are two characters of the system
built in this paper. The first is that we use the human’s
intended walking speed to drive the treadmill. Specifically, we
find a linear relation between force measurement and intended
walking speed. Based on it, the speed control is simple as a
linear computation. The second is that our system not only
gives natural feeling of walking, but also gives natural sense of
vision and hearing. The 3D display based on virtual reality
and the ambient sounds of the environment make the subject
have an immersed sense. In addition, the layout of shops is
based on a real market located in Hokkaido Japan, which
makes the whole system be much more realistic.
This paper is organized as follows. Section II illustrates the
system construction and working mechanism. Section III
proposes a velocity control scheme of the treadmill based on
the skeleton model simulation results. Section IV presents the
software structure and parameter settings of 3D display.
Section V shows that the proposed approach is capable to
control the treadmill velocity and 3D display well, which
verifies the validity of the whole system. Section VI concludes
the whole paper.
II. SYSTEM CONSTRUCTION AND PROBLEM FORMATION
The system consists of three modules as centurial
processing PC, Bertec treadmill and 3D stereoscopic display
(Fig.1). When the subject walks on the treadmill, the dual
force plates (placed under the treadmill) measure the force and
moment signals in x, y, z directions as xF , yF , zF , xM , yM ,
zM and output them as analog signals. After amplifying them
and doing analog-to-digital (A/D) conversion by LABVIEW
data acquisition, the digital signals are transferred to centurial
Development of a 3D Interactive Virtual Market System with
Adaptive Treadmill Control
Haiwei Dong, Tatuo Oshiumi, Akinori Nagano and Zhiwei Luo, Member, IEEE
W
The 2010 IEEE/RSJ International Conference on
Intelligent Robots and Systems
October 18-22, 2010, Taipei, Taiwan
978-1-4244-6676-4/10/$25.00 ©2010 IEEE 5238

Sensor Noise
actualV++
+
+
intendV
•
•
Sensor Noise
actualV++
+
+
intendV
•
•
Fig.2 System block diagram. The treadmill has an inner controller with many feedback loops to control the speed of the belts.
When the user walks on the treadmill, the force plate measures the forces and moments in x, y, z directions. By a transfer
function ( )G s , the desired walking speed of the user can be obtained from the sensor data.
processing PC. Based on the force signals, intended walking
speed of the user is estimated. The estimated speed is used in
two ways: one is to drive the two motors corresponding to left
belt control, right belt control; the other is to drive the 3D
virtual reality scene to move on.
Fig.1. System construction. The system is composed of three
modules including centurial PC, treadmill and 3D display.
The system block diagram is shown in Fig.2. From the
viewpoint of control theory, the control plant is Bertec
treadmill which is controlled by treadmill inner controller in
the inner loop feedback. When the user walks on the treadmill,
there is a noise adding to the control signal. It is noted that the
control objective is to control the treadmill by the user’s will.
Hence, we establish an outer loop feedback between the user’s
intended walking speed intendV and the treadmill inner
controller. In this case, the force plate is chosen as the
observer which measures the interactive forces between feet
and treadmill. Besides of the sensor noise, we want to
establish a relation between intendV and the measurement data,
i.e. recognize and identify the transfer function ( )G s .
III. BERTEC TREADMILL CONTROL
In this section, according to the walking simulation data, a
critical feature which corresponds to the intended walking
speed is obtained. By the least-squares regression method, the
relation between the proposed ratio and intended walking
speed is established.
A. Force Plate
First of all, let us define the standard coordinate system as
that the positive y-direction points forward; x-axis points left
when looking in the y-direction; and the z-axis is defined
downwards by the right hand rule. The origin of the coordinate
system is centered at the inner corner of the outer back roller
support block of the corresponding (right or left) treadmill
half.
Each half of Bertec dual belt treadmill incorporates an
independent force plate measuring six load components: the
three orthogonal components of the resultant force and three
components of the resultant moment in the same orthogonal
coordinate system. All the forces acting between the foot and
the ground can be summed to yield a single ground reaction
force F and a free torque vector zT . The point of application
of the ground reaction force on the plate is the center of
pressure (CP). All the small reaction forces collectively exert
on the surface of the plate at the CP. The point of application
of the force and the couple acting can be readily calculated
from the measured force and moment components
independently on each half of the treadmill (Fig.3).
The two force plates are calibrated individually and the
calibration matrix is stored digitally in the force plate. The
voltage output of each channel is in a scaled form of the load
with the units of N and N⋅m for the forces and moments,
respectively. The scale factor for each channel for a gain of
unity is given in the product data sheet supplied with the
transducer. The force and moment values are calculated by
multiplying the signal values with corresponding scale factors,
5239

as given
1 1
2 2
3 3
4 4
5 5
6 6
0
0
x
y
z
x
y
z
C SF
C SF
C SF
C SM
C SM
C SM
    
    
    
    
=     
    
    
    
          
(1)
where, xF , yF , zF and xM , yM , zM are the force and
moment components in the force transducer coordinate
system, in N and N.m, respectively. 1S , 2S , 3S , 4S , 5S and
6S are the output signals corresponding to the channels
indicated by their subscripts, in volts, divided by the
respective channel gain. The diagonal matrix composed of 1C ,
2C , 3C , 4C , 5C , 6C is the calibration matrix with units N/V
for the force channels 1 2 3( , , )C C C and moment channels
4 5 6( , , )C C C [12].
Fig.3 Force plate. There are four force sensors located at the
bottom of the force plate. Finally, the force plate outputs
forces and moments of xF , yF , zF , xM , yM , zM . The
height of pad, denoted as h , is 0.15m.
B. Characteristic Index yzR
The simulation was done by OpenSim 2.0. The skeleton
model is 72.419 kg weight where red arrow, green arrow and
blue arrow denote y, -z and -x directions, respectively (Fig.4).
The model is composed of as many as 12 parts as torso, pelvis,
femur-r, tibia-r, talus-r, calcn-r, toes-r, femur-l, tibia-l, talus-l,
calcn-l and toes-l. In order to simulate the real human model
better, 28 muscles forces and contact forces, etc. are also
added in the model. It is noted that, because upper limbs are
not be of crucial importance, the model was built without
upper limbs for simplicity.
The simulation results show that in one dynamic circle, the
force zF is a bell-shape signal, and yF is a sine-shape signal
(Fig.5). We can explain the reason for curve shape of yF and
zF as follows. When the foot gets in touch with the surface of
treadmill, zF increases very rapidly. At the same time, the
foot has to make a break to adjust its speed to the velocity of
the belt by friction. After break process, the foot applies a
force with inverse direction to drive the leg to take a step, i.e.
make a preparation for higher speed of leg in the next moment.
It is noted that, compared with break process, yF changes its
direction at this time. Until now, the foot is on the treadmill all
along. Hence, zF maintains large (for a normal person with
70 kg weight, zF is about 700 N). Finally, the body alternates
the other foot to support body and zF decreases very rapidly.
Fig.4 Human skeleton model. The coordinates system is set
that red arrow, green arrow, blue arrow denote y, -z, -x
directions, respectively.
0 0.2 0.4 0.6 0.8 1 1.2
−200
0
200
400
600
800
Fx
,Fy
,Fz
(N)
0 0.2 0.4 0.6 0.8 1 1.2
−0.3
−0.2
−0.1
0
0.1
0.2
Time (s)
R
y,z
F
x
F
y
F
z
Fig.5 Forces between foot and ground. yF is a sine-shape
curve and zF is a bell-shape curve. Hence, yzR is a
sine-shape curve.
Considering the fact that when zF is large enough, one foot
is firmly on the treadmill. We define a force ratio ,y zR as
,
0
y
z
y z z
F
F
R F
others
ξ

≥
= 


(2)
x
'x
'z z
y
'y
1F
2F
3F
4FF
zT
plate surface
pad surface
•
•
'P
P
h
5240

where ξ is a threshold. In the normal cases, it is chosen that
max{ } 80%zFξ = × . According to the curve shapes of yF
and zF , the curve shape of ,y zR is a composite signal of
connection of sine-shape signals and zero signals (Fig.5).
Without loss of generality, the zero signals are ignored in
analysis. ,y zR becomes a composite signal of connection of
numerous sine-shape signals with different magnitude. ,y zR
has a direct relation with the intended walking speed. When
the user intends to speed up, ,y zR
∞
becomes large. In other
words, the envelope curve of ,y zR determines the intended
walking speed (Fig.6 (a)). The above illustration is roughly
verified by the walking experiment when the treadmill
velocity increases from 1.0 m/s to 1.6 m/s as shown in Fig.6
(b). Further verification is shown in the section of experiment.
0 1 2 3 4 5 6 7 8 9 10
−10
−5
0
5
10
Velocity
R
y,z
acceleration phase
(a)
0 100 200 300 400 500 600 700
-200
0
200
Fx
(N)
0 100 200 300 400 500 600 700
-200
0
200
Fy
(N)
0 100 200 300 400 500 600 700
0
500
1000
Time (ms)
Fz
(N)
(b)
Fig.6 (a) In the acceleration phase, the origin equilibrium
state with low speed transfers to a new equilibrium state with
high speed. (b) With the increase of treadmill speed (from 1.0
m/s to 1.6 m/s), the magnitudes of xF , yF , zF become large
gradually.
C. Least-squares Regression
First of all, define the local peak value ,y zR+
and local valley
value ,y zR−
of ,y zR as
, ,
0
, ,
0
max{ }
min{ }
y z y z
t T
y z y z
t T
R R
R R
+
≤ ≤
−
≤ ≤
=
=
(3)
where T is the period of ,y zR . Then we can get a sequence of
measurements
1 , ,1 2 , ,2 , ,( , ),( , ), ,( , )y z y z n y z nV R V R V R− − −
L (4)
Assumed that ,y zR−
is predicted as a function of V then one
can model this situation by
, ( , )y zR f V λ ε−
= + (5)
where λ is parameter vector. The random variable ε is
independent of V and on average it is equal to zero, i.e.
( ) 0E ε = . We want to find f that fits the measurement data
best and we define the loss function to measure the quality of
the fit as
, 2
( , )y zL R f V λ ε−
= − − (6)
and we minimize it over all choices of parameter vector λ . To
find the approximation function, we write
( ), , ( , ),
0
y zL R f V λ ε
λ
−
∂
=
∂
(7)
It is noted that, the above estimation solution is unbiased
estimation since
( ) ( ), | ( , ) | ( , ) ( | )
( ) ( ) ( )
y zE R V E f x V f x E V
f x E f x
λ ε λ ε
ε
−
= + = +
= + =
(8)
IV. STEREO DISPLAY
The virtual market display module is connected with other
modules by TCP/IP protocol. Each time, the centurial
processing PC collects the force measurements from the force
plates. After system identification computation, the intended
walking speed of the subject is estimated. Based on the
estimated speed, the 3D display goes on.
The design of the virtual market is based on a real market
located in Hokkaido, Japan. Thus, the environment has
apparent Japanese characters. Right now, the market is
designed as a north-south street with a length of 80 meters,
which includes one traffic corner and numerous cross roads.
There are 20 shops in all with 10 shops distributing on each
side of the road. At the end of the road, there is a train station
by entering which the program terminates (Fig.7).
The program was developed by Visual C++. In general, the
utility toolkit GLUT was used for vision design and SDL
library was used for sound design. For the vision design, the
3D models of shops, post office, etc. were constructed by
Metasequoia which is a polygon modeler for 3DCG and game
development. Specifically, everything is created as an
assembly of several basic geometrical bodies. Each face of the
5241

Fig.7 Binocular vision of 3D virtual shopping street. The market is designed based on a real market located in Hokkaido
Japan. The virtual market is a north-south street with a length of 80 meters including one traffic corner and numerous cross
roads.
composite unit is assigned a special texture. For example, the
surface of ground is mapped with asphalt material; the surface
of sky is mapped with cloud. Finally, the above generated 3D
model data were used by the utility toolkit GLUT OpenGL
directly. For the sound design, we used SDL-mixer library to
create various sounds, like the footstep sound, the market sell
voice, etc. Even at the place near the train station, the subject
can hear the jingle of the train from the distance to the close.
In order to realize stereoscopic display, we have to
configure the height of the viewpoint. Assuming that the
average height of man is about 170 cm and of woman is about
160 cm, respectively, the height of viewpoint is compromised
to be 165 cm. In practical application, the user wears a 3D
eyewear. Actually, the 3D eyewear has two different
eyeglasses which allow different polarized light pass
separately. In such a way, the left and right eyes of the user see
different images from its own viewpoint (Fig.7). In the system,
papillary distance is set as 7 cm. Thus, the left and right optical
lines of sight cross at 1 m before the user.
In addition, the system has many options. For example, the
user can choose whether there is background music at the
beginning of the program. Besides, the user also can choose
different street to walk.
V. EXPERIMENT
In the experiment, the subject walks on the treadmill at will.
Meanwhile, the VR shopping scene goes on with the intended
walking speed. As the projector is a stereoscopic projector, by
wearing a pair of 3D glasses, the subject have a more
immersive and realistic experience (Fig.8). The program was
written by three software packages as Visual C++, LabView
and Matlab. The usage details of the software packages are
shown as follows. The VR-related works, e.g. 3D display, is
developed by Visual C++. The hardware-related works, like
data acquisition, treadmill control, etc. were programmed by
LabView. The computation-related works, e.g. least squares
regression, were accomplished by Matlab.
Fig.8 Experiment environment. The subject walks on the
treadmill and meanwhile watching the 3D scene.
During the process, the communication between three
modules is done by TCP/IP protocol. The sender is localhost
(meaning “this computer”) which translates to the IP address
127.0.0.1 and the port is defined as 4000. In detail, when
transferring speed signal to treadmill, the TCP/IP packet is
shown in Table I.
TABLE I
TCP/IP TRANSFER PACKET
Space Meaning
U8 Format specifier (set to 0)
4*S16 Belt speed in mm/s
4*S16 Belt acceleration in mm/s2
S16 Incline angle in 0.01 deg
4*S16 Belt speed complement (bit-inverted) in mm/s
4*S16 Belt acceleration complement (bit-inverted) in
mm/s2
S16 Incline angle complement (bit-inverted) in
0.01 degree
27*U8 Padding
(where <U|S> - U(nsigned)|S(igned) <8|16> - size in bits)
5242

When the velocity of treadmill is set as 0.5 m/s, the forces
xF , yF and zF are shown in Fig.9. It shows that yF is a
sine-shape curve and zF is a bell-shape curve, which verifies
the simulation analysis in Section III, Part B.
20 20.5 21 21.5 22 22.5 23
−50
0
50
100
Fx
(N)
20 20.5 21 21.5 22 22.5 23
−100
0
100
F
y
(N)
20 20.5 21 21.5 22 22.5 23
0
500
1000
Fz
(N)
Time (s)
feet left pannel feet left pannel
Fig.9 Measurement of xF , yF , zF . The foot leaves the force
plate in the time interval of [20.4,20.9] s and [21.8,22.3] s.
In order to do least-squares regression, we have to collect
many observation samples. In the experiment, we compute
,y zR ten times when the velocity varies from 0.1 m/s to 1.0 m/s.
The observation results are shown in Table II.
TABLE II
OBSERVATION SAMPLES
Velocity
(m/s) ,y zR+
,y zR−
0.1 0.0080 -0.0213
0.2 0.0310 -0.0353
0.3 0.0716 -0.0542
0.4 0.0622 -0.0695
0.5 0.0957 -0.0884
0.6 0.0872 -0.1172
0.7 0.1078 -0.0834
0.8 0.1126 -0.1224
0.9 0.1153 -0.1187
1.0 0.1841 -0.1421
By using the least-squares regression method in Section III,
Part C, we obtained four types of regression models as linear,
quadratic, cubic and 4th degree polynomial (Table III). The
regressions are also shown in Fig.10 (a) and the residual
statistics is shown in Fig.10 (b).
TABLE III
LEAST-SQUARES REGRESSION RESULTS
Regression
Model
Results
Norm of
Residuals
linear , 0.13 0.011y zR V−
= − − 0.039887
quadratic
2
, 0.067 0.2 0.00072y zR V V−
= − −
0.034701
cubic
3 2
, 0.08 0.19 0.25
0.0022
y zR V V V−
= − + −
+ 0.034094
4th degree
polynomial
4 3 2
, 0.6 1.1 0.56
0.097 0.0021
y zR V V V
V
−
= − + −
− − 0.031844
0 0.2 0.4 0.6 0.8 1
−0.15
−0.1
−0.05
0
0.05
0.1
0.15
0.2
Velocity (m/s)
R
y,z
data 1
data 2
linear
quadratic
cubic
4th degree
(a)
0 0.2 0.4 0.6 0.8 1
−0.25
−0.2
−0.15
−0.1
−0.05
0
0.05
0.1
0.15
0.2
0.25
Velocity (m/s)
Residuals
residuals
linear
quadratic
cubic
4th degree
(b)
Fig.10 Least-squares regression. (a) Regression results of the
four types of model as linear, quadratic, cubic and 4th-degree
polynomial. (b) Residuals results of regression.
It is shown that the residual norms of the four types do not
have big difference. Hence, we choose the simplest linear
regression model to use. The proposed treadmill velocity
control is obtained
( )
,
,
( 1) ( ) 0
( )
( ) 0.011 / 0.13
y z
y z
V k R k
V k
R k others
−
−
 − =
= 
− +
(9)
There are many reasons, e.g. sensor noise, to cause
computed velocity not smooth. In this paper, we choose a
delay-line filter to overcome the problem. The filter is
described by the difference function
5243

4
1
1
( ) ( )
4 i
V k V k i
=
= −∑ (10)
There are also many low-pass filters, such as Butterworth
filter, Chebyshev filter, Bessel filter, etc, which are available
to be used considering the frequency of noise. Fig.11 (a)
shows a complete dynamic walking process with acceleration
motion, deceleration motion and uniform motion (i.e. constant
velocity motion), where the solid line denotes velocity and
dash line denotes yF , respectively. Fig.11 (b) shows part
views of acceleration motion, deceleration motion and
uniform motion in the time interval [28,34] , [55,65] and
[15,24] , respectively.
0 10 20 30 40 50 60 70
0
0.2
0.4
0.6
0.8
Velocity(m/s)
Time (s)
(a)
28 29 30 31 32 33 34
0
0.5
1
accelertation
55 60 65
0
0.2
0.4
Velocity(m/s)
deceleration
15 16 17 18 19 20 21 22 23 24
0.35
0.4
0.45
0.5
Time (s)
equilibrium
(b)
Fig.11 Experiment results. (a) The whole walking dynamic
process (about one minute) with acceleration motion,
deceleration motion and uniform motion. (b) Part views of the
acceleration motion, deceleration motion and uniform motion
are shown from top to bottom.
VI. CONCLUSION
In this paper, a 3D interactive virtual reality shopping
market system was developed. The subject walks on the
treadmill at will. The treadmill and 3D display adjust to the
subject’s intended walking speed, correspondingly. There are
two main characters of the developed virtual market system.
The first is treadmill velocity control which only uses force
measurements. Compared with the previous treadmill-style
locomotion methods, the proposed speed control in this paper
shows great effectiveness and real-world applicability. The
second is 3D display by which the subject gets a more
immersive and realistic feeling. The future work is to connect
the virtual shops with the real ones. In this way, the elderly
person can healthily order goods in VR worlds at home.
ACKNOWLEDGMENT
This paper received constructive suggestions from Hisahito
Noritake, Yusuke Taki and Shouichi Katou. The authors also
would like to thank Haifeng Dong of the Arizona State
University USA for proofreading.
REFERENCES
[1] B. C. B. Yip and D. W. K. Man, “Virtual reality (VR) based community
living skills training for people with acquired brain injury: a pilot
study,” Brain Injury, vol. 23, pp. 1017-1026, 2009.
[2] M. Farber, F. Hummel, C. Gerloff et al., “Virtual reality simulator for
the training of lumbar punctures,” Methods of Information in Medicine,
vol. 48, pp. 493-501, 2009.
[3] D. Van, R. Hunneman and S. Wildlevuur, “Self city: training social
skills in a game,” in 2nd European Conf. Games-Bases Learning, 2008,
pp. 481-488.
[4] J. M. Hollerbach, “Locomotion interfaces and rendering,” in Haptic
Rendering: Foundations, Algorithms, and Applications, A. K. Peters,
pp. 83-92, 2008.
[5] J. M. Hollerbach, “Locomotion interfaces,” in Handbook of virtual
environments: design, implementation, and application, Lawrence
Erlbaum Associates, pp. 239-254, 2002.
[6] H. Noma and T. Miyasato, “Design for locomotion interface in a large
scale virtual environment ATLAS: ATR locomotion interface for active
self motion,” in 7th Annual Symposium on Haptic Interface for Virtual
Environments and Teleoperated Systems, 1998, pp. 111-118.
[7] R. P. Darken and W. R. Cockayne, “The omni-directional treadmill: a
locomotion device for virtual worlds,” in Proceedings of the 10th
Annual ACM Symposium on User Interface Software and Technology,
1997, pp. 213-221.
[8] H. Iwata and Y. Yoshida, “Path reproduction tests using a Torus
treadmill,” Teleoperators and Virtual Environment, vol 8, pp. 587-597,
1999.
[9] R. Christensen, J. M. Hollerbach, Y. Xu, et al., “Inertial force feedback
for the Treadport locomotion interface,” Teleoperators and Virtual
Environments, vol. 9, pp. 1-14, 2000.
[10] J. M. Hollerbach, Y. Xu, R. Christensen, et al., “Design specifications
for the second generation Sarcos Treadport locomotion interface,” in
Proceedings of ASME Dynamic Systems and Control Division,
DSC-vol. 69-2, 2000, pp. 1293-1298.
[11] Instrumented Treadmill User Manual (Version 3.3), Bertec
Corporation, 2008.
5244

Development of a 3D Interactive Virtual Market System with Adaptive Treadmill Control

More Related Content

What's hot (19)

Viewers also liked (7)

Similar to Development of a 3D Interactive Virtual Market System with Adaptive Treadmill Control (20)

More from toukaigi (20)

Recently uploaded (20)

Development of a 3D Interactive Virtual Market System with Adaptive Treadmill Control