Proceedings of the 2015

IEEE Conference on Robotics and Biomimetics

Zhuhai, China, December 6-9, 2015

Education-Oriented Portable Brain-Controlled Robot System

Danyang Zhi, Zhengping Wu, Wei Li, Jing Zhao, Xiaoqiang Mao, Mengfan Li, and Meifang Ma

Abstract In this paper, we develop an education-oriented

portable brain-controlled robot system, consisting of a
multi-parameter bioelectric signal acquisition device and a Lego
NXT Mindstorms robot. The education-oriented portable
brain-controlled robot system is only 45 in volume and
weighs 70g. The signal acquisition section wirelessly acquires
brain signals and the control section remotely navigates a Lego
robot via brainwaves through a Bluetooth. Teenagers may use
the low-cost brain-controlled robot to enjoy the pleasure of
learning, to inspire their innovative ability and creativity, and to
raise their mental concentration. Moreover, the proposed
brain-controlled robot system might be used to improve the
quality of adolescent education and autism recovery. The system
modules include the brain signal acquisition, signal processing,
and the robot control, which are developed in LabVIEW-based
programming environment and easy to be expanded and
upgraded. The comparative study of the education-oriented
system and Cerebot equipped with the high precision Cerebus
system widely used in laboratory research shows that the
low-cost system delivers comparable accuracy for the robot
control. An operation task is presented to train users attentions
and concentrations.

I. INTRODUCTION
Brain robot interaction (BRI) technology allows an
individual, such as a disable, to control the robotic devices via
brain signals [1][2][3]. Brain-controlled systems, based on
brain-computer interface (BCI), are classified into two
categories: the invasive technology implanting electrodes over
brain cortex through surgery to record signals of neurons, such
as Electrocorticogram (ECoG); the non-invasive technology
acquiring brain signals through electrodes placed on scalp,
Manuscript received July 24, 2015. The work is supported in part by The
National Natural Science Foundation of China (No. 61473207) and State Key
Laboratory of Robotics at Shenyang Institute of Automation (Grant No.
2014-Z03).
Danyang Zhi is with the School of Electrical Engineering and
Automation, Tianjin University, Tianjin 300072, China. (e-mail:
zhidanyang@tju.edu.cn).
Zhengping Wu is with the School of engineering education, Sanjiang
University, Nanjing 210000, China. (e-mail: wzp_959@163.com )
Wei Li is with the School of Electrical Engineering and Automation,
Tianjin University, Tianjin 300072, China. Shenyang Institute of Automation,
Chinese Academy of Sciences, Shenyang 110016, China. Department of
Computer & Electrical Engineering and Science, California State University,
Bakersfield, CA 93311, USA. Wei Li is the author to whom correspondence
should be addressed. (phone: +661-654-6747; fax: +661-654-6960; e-mail:
wli@csub.edu).
Jing Zhao is with the School of Electrical Engineering and Automation,
Tianjin
University,
Tianjin
300072,
China.
(e-mail:
zhaoj379779967@163.com).
Xiaoqiang Mao is with the School of Electrical Engineering and
Automation, Tianjin University, Tianjin 300072, China. (email:
maoxiaoqian@tju.edu.cn)
Mengfan Li is with the School of Electrical Engineering and Automation,
Tianjin University, Tianjin 300072, China. (e-mail: shelldream@ tju.edu.cn ).
Meifang Ma is with the Brain-Medical Ltd, Nantong 226300, China.
(e-mail: ediotmmf@gmail.com )

978-1-4673-9675-2/15/$31.00 2015 IEEE

such as electroencephalogram (EEG). The non-invasive

technology has been widely applied due to its easy to use,
inexpensive cost, acceptable temporal resolution, and low
clinical risk [4][5][6].
Brain-controlled systems were initially applied to medical
rehabilitation of the patients who suffer from Palmer's
syndrome and aphasia or handicaps or limited mobility caused
by innate factor or accident. The systems help them navigate a
wheelchair or control a robot to accomplish a specific task
which assist their daily life and rehabilitation [7][8][9].
Recently, the brain-controlled systems have been applied to
new fields such as entertainment, smart home and military
reconnaissance. In November 2012, the PuzzleBox Company
released a novel toy helicopter Orbit controlled by brain
signals for experiencers to adjust their emotional states [10].
Nowadays, the teenagers education has become a hot topic
because schools and societies have committed great
importance of not only imparting the professional knowledge,
but also developing a healthy mind and good psychological
quality. Consequently, a variety of programs for developing
innovative ability have been developed, e.g., courses and
competitions of Lego robots training on teenagers ability to
think innovatively and skills to co-operate with partners. A
serious problem in the modern society is that the scatterbrains
and teenagers with hyperactivity are unable to behave or to
study like normal students. A psychology team from university
of Tennessee offered a biofeedback therapy to hundreds of
teenagers with attention deficit disorders (ADD) or
hyperactivity and has significantly improved their mental state
and intelligence [11]. This initiative opens a new direction for
applying the brain-controlled robot technology to the teenagers
education. The education-oriented brain-control system
enables students to experience the pleasure of learning and to
inspire their imagination and creativity. Control of a robot
system via brainwaves needs the mind concentration, so
developing brain-controlled robot systems will significantly

1018

Figure 1. The structure of the brain-controlled robot system: The

acquisition module connects electrode cap, preprocesses collected EEG
signals, and transmits the signals wirelessly via a USB port. The control
module off-line analyzes the brain signals and on-line controls of a Lego
robot.

contribute to the education and rehabilitation of the autism.

The existing brain-controlled products have several
limitations to education applications because the existing
low-cost portable devices, which deliver low precision and
limited patterns of brain signals, may easily cause subjects to
be frustrated because their intentions are unable to be correctly
recognized and thus gradually lost interest in using the
brain-controlled systems. In addition, the robotic devices
controlled by these products are not easy to be freely
developed. On the other hand, high-precision brain-controlled
equipment used in research laboratories is too expensive and
too complicated to be commonly adopted for teenagers
education.
In this paper, we developed the education-oriented
brain-controlled robot system, as shown in Fig. 1. The system
consists of a multi-parameter biological signal acquisition
device and a Lego Mindstorms robot [12]. It is very low-cost
and portable since its acquisition device is only 453 in
volume, and weighs 70g. Lego Mindstorms, which can be
programmed in a variety of languages and used to create a
customizable robot, such as a mobile robot, a manipulator, an
animal robot with many legs, and a humanoid robot, has been
widely used for education. In the proposed brain-controlled
robot system, users can construct different types of robots,
write programs for their applications, and control robot
behavior via brain signals. Especially, we conduct the
comparative study of the education-oriented system with the
Cerebot [13][14][15] that is equipped with a high precision
Cerebus system widely used in laboratory research. The results
show that the low-cost system delivers compatible accuracy
for the robot control.
This paper is organized as follows. Section II presents the
hardware structure of the proposed system. Section III
describes the brain signal acquisition, and signal processing
and robot control modules. Section IV introduces the model of
steady-state visual evoked potentials (SSVEP). Section V
compares both off-line and on-line performances of the
proposed system with the ones yielded by a Cerebus data
acquisition device, a professional high-quality system for
acquiring reliable EEG signals. Section VI shows an operation
example designed for teenagers education.

20V. 2. The analog front end uses a flexible multi-channel

multiplexing circuit, which is controlled by a low-noise 7-level
programmable gain controller (PGA) to improve the range and
flexibility of the input, to choose any channel or multiple
channels as a reference. 3. Each channel has a 24-bit ADC
and a digital filter. 4. The microprocessor unit adopts
STM32F103VET6: a Cortex-M3-based ARM microprocessor.
5. The wireless signal transmission circuit uses NRF24L01 RF
chip, which works at 2.4 GHz ISM frequency band with a
transmission speed of up to 2 MBPS. 6. The embedded
software monitors the status of the acquired signals and
transmits the sampled EEG data to a host computer.
B. Lego Robot
Lego Mindstorms is a new generation of NXT Bluetooth
robot launched in 2006. As a customizable and programmable
robot product, it has spread in most households due to its
moderate price and become the favorite toys of users at
different ages. Lego Mindstorms uses an intelligent brick,
which is equipped with a 32bits ARM7 microprocessor, 256K
FLASH ROM, 8bits AVR microprocessor and a monochrome
lattice LCD screen, and simultaneously support USB2.0 and
Bluetooth. NXTs three outputs are for stepper motors, and
four inputs are for a variety of sensors: color sensors, touch
sensors and ultrasonic wave sensors, etc. Users can expand
other sensor components for their own requirements. The most
attraction for people is that Lego robot kits, just like traditional
Lego blocks, allow users to create and assemble a variety of
robot models.
Lego Mindstorms packages support multiple programming
languages including NXT-G, ROBOLAB, and LabVIEW.
Through LabVIEW Mindstorms kits, users can create and
download VIs to control Lego Mindstorms robot, and interact
with it while the program is running. When using the third
party sensors, we can create local VIs in LabVIEW to control
hardware devices. Through a USB port or wireless Bluetooth
device, the VIs programmed on the host computer can be
transferred to the NXT brick to run the Lego robot. The
creative learning process of constructing and programming on
Lego NXT Bluetooth robots enable users to develop their
learning skills, stimulate imagination and creativity. It is a very
meaningful experience for users to explore science,
technology, engineering and mathematics (STEM).

II. HARDWARE STRUCTURE

A. Multi-Parameter Bioelectric Signal Acquisition Device
This paper developed a novel multi-parameter bioelectric
signal acquisition device with a multi-channel bioelectric
signal amplifier, which is capable of recording at most 8
channels of electrophysiological signals simultaneously at a
sampling rate of 500 Hz. It consists of a multi-channel analog
front end with high common mode rejection ratio (CMRR), a
24-bit sigma delta analog-to-digital converter (ADC), a digital
filter, an ARM processor based on Cortex-M3, and a wireless
signal transceiver. This system is compact through reducing
lots of peripheral circuits and extends its distance by using a
radio frequency (RF) transceiver.
The considerations of designing the device as follows: 1.
The power supply module uses LT1763 and LT1614 [16],
which only needs a 0.01F capacitor to reduce the output noise
between 10Hz and 100kHz to the root mean square (RMS) of
1019

III. SYSTEM MODULES

The brain-controlled system modules, including the brain

(a)

(b)

Figure 2. (a) Diagram for acquisition module (b) Diagram for data
processing module

signal acquisition, and signal processing, and the robot control

module, are developed under the LabVIEW environment.
They acquire brain signals from bioelectric signal acquisition
device, extract characteristic frequency from the brainwaves
and generate the corresponding commands to control the Lego
robot.
A. LabVIEW Programming Environment
LabVIEW is a general graphical compilation platform,
whose process flows adopt the approach of data stream. This
graphical programming language is different from script-based
programming languages [17]. In LabVIEW, we build the
program by choosing VIs from the function panel, drag them to
programming windows, and link them up following the data
flow direction. Different from traditional programs, the
LabVIEW-based programs improve the program efficiency,
shorten the developing cycle of prototype, facilitate software
later maintenance, and connect to a variety of hardware
devices. Due to more beautiful graphic look and easier to be
accepted and understood than scripts, LabVIEW has been very
popular in industries, research institutions, and universities.
Even teenagers without strong programming background can
quickly learn to use LabVIEW to develop projects by putting
different VIs together to implement functions they want [18].
LabVIEW offers a variety of functions to implement any
complicated programming task, including data collection,
general purpose interface bus (GPIB), serial port control, data
analysis, data display and data storage etc., and convenient
debugging tools, such as setting breakpoints, running single
step [19]. The brain-controlled robot system uses the functions
of serial port control and data collection to receive the brain
signals from multi-parameter acquisition device and uses the
functions of data display, storage, and analysis to analyze and
process the brain signals. We also use LabVIEW Lego module
to control the Lego robot. Our LabVIEW-based modules are
able to invoke existing MATLAB scripts to process brain
signals.
B. Module Design
Fig. 2(a) shows the diagram for the brain signal acquisition
module. The module uses the serial port to connect the
multi-parameter bioelectric signal acquisition device,
configures the parameters as 256000-bps baud rate, 8-bit data
bits and 1-bit stop bit, and establishes the data communication.
The data acquisition process starts when pressing the start
button. Then, the module displays the real-time data acquired

(a)

from 8 channels at a sampling rate of 500 Hz in the front panel

and records these data in a file.
Fig. 2(b) shows the diagram for data processing module.
The data processing module processes in real-time the brain
signals from channels O1, O2, Oz and POz in every 3 seconds
and adopts canonical correlation analysis (CCA) to construct
feature vector with the greatest relativity with the reference
data [20]. The module uses Fast Fourier Transform Algorithm
(FFT) to calculate the frequency spectra of the brain signals
and to display them in the front panel. Finally, the module
classifies the brain signals by Fisher Linear Discriminant to
identify subjects intentions and converts them into commands
for the Lego robot.
Fig. 3 shows the diagram for robot control module. The
robot control module sends the commands to the Lego robot to
control three basic motions: moving forward, turning left and
turning right. The robot motors can be controlled by a constant
power or a constant speed. Currently, the module chooses the
constant power approach to control the three motions.
According to a command from the processing module, the
control module sends the corresponding power level to the
Lego port B and port C which connect the two wheels motors
by cables. Each motion lasts for 500ms, and then the input
power is changed to zero for the next command. The VI in Fig.
4(b) implements that the robot is controlled to turn right
because it receives a command with the power level of 100 and
power level of 0 respectively to port B and port C to run the
right wheel. The VI in Fig. 4(a) controls the robot to move
forward with the full speed because the robot receives a
command with the same power levels to the left and right
wheels.
IV. SSVEP-BASED MODEL
SSVEPs are induced by brain activities in response to
visual stimuli of the retina [21]. In 1966, Regan [22] recorded
the harmonics of the visual evoked potential to a flickering
light modulated sinusoidally through an analogue Fourier
series analyzer and found that the brain attained a steady-state
neural response in which the phase and amplitude of the

(a)
Figure 3.

(b)

Figure 4. (a) VI for moving forward. (b) VI for turning left

(b)

Figure 5. (a) User Interface. (b) Frequency spectrogram

Diagram for control module

1020

harmonics were approximately constant over time. Those

types of brain signals responding to a visual stimulus
modulated at a fixed frequency are known as steady-state
visual evoked potentials (SSVEPs). In most studies, SSVEPs
can be recorded from the scalp over the visual cortex with
maximum amplitude at the occipital region [23]. Some
electrophysiological experiments have shown that the neurons
in brain visual cortex synchronize their firing to the flicker
frequency, which leads to SSVEPs containing the same
fundamental and harmonic frequencies as that of visual
stimulus [24].
Up to now, we have investigated a variety of
brainwave-based models to build the brain-controlled system,
while owning to the complexity of human brain thinking and
brain signals, the most of brainwave-based models have more
or less shortcomings in signal stability, information transfer
rate, etc. [15]. In this study, we choose the SSVEP model for
our experiments, because SSVEP-based BCI usually delivers
higher information transmission rate (ITR) and demands on
less training.
Our SSVEP model chooses 4.61Hz, 20Hz and 15Hz as
stimulus frequencies, as shown in Fig. 5(a). The user interface
flickers three robot images on the computer monitor
representing the behaviors: moving forward, turning left and
turning right at the fixed frequency. Fig. 5(b) shows SSVEP
signals induced through brain signals. We define the SSVEP
signal () evoked by the stimulus at time t by [25]
() =

=1
,

sin(2 + , ) + ,
= 1,2, , ,

(1)

where is the flicking frequency of the visual stimulus at

4.61Hz, 20Hz, or 15Hz. N=3 is the total number of visual
stimuli. signifies the number of harmonics taken into
consideration. , and , represent the amplitude and phase
of each sinusoid, respectively. , includes noise, artifacts and
other components. The visual stimulus flickering at the given
frequency induces electrophysiological responses which

described by sinusoidal signals =1

, sin(2 + , ).
The data processing module eliminates , and extracts
features , ( = 1, 2, 3), where is the ordinal of the most
responsive harmonics to the visual stimulus and ,
represents three stimulus targets. In order to reduce unexpected
disturbance, such as signals caused by head shaking, we divide
the energy of all frequency spectra by the mean value of
signals between 4.3-40Hz and extract their multiple features. It
is important to find feature parameters , .
We set 1 = 1,1 , 2 = 2,2 , 3 = 3,3 . In this study,
we choose 1 = 1,1 and 2 = 2,1 for the stimuli of target 1
and target 2, i.e., the fundamental harmonics of their brain
signals, but 3 = max{3,1 , 3,2 } for the stimulus of target 3 as
the fundamental harmonics of its brain signals is not very
stable. The neural responses at 4.61Hz are easily disturbed by
noise, so we amend the 1 value for the stimulus of target 1
when the average amplitude of the brain signals induced at
3.0-4.0Hz is higher than 10. We build the feature vector
[1 , 2 , 3 ] for each trail. We train a Fisher classifier by using
the feature vectors for identifying a mental activity to control
the Lego robot.

TABLE I.

TARGET FREQUENCY-DOMAIN CHARACTERISTICS

COMPARISON

subjects

Frequency-domain characteristics

systems
The
portable
system

Cerebot

S1

S2

S3

S4

Average

4.61Hz

31.2

19.2

17.9

19.1

21.9

20Hz

13.6

17.1

4.7

11.9

11.8

15Hz

7.2

17.1

5.4

6.9

9.2

4.61Hz

25.9

16.6

19.8

11.8

18.5

20Hz

17.2

8.7

6.7

16.9

12.4

15Hz

10.4

8.9

6.7

12.2

9.6

V. EXPERIMENTAL RESULT
A. Experimental Preparation
In order to evaluate the system performance, we off-line
analyze the brain signals and on-line control the Lego robot by
comparing the developed brain-controlled robot system with
Cerebot [13][14][15] because our previous studies via a variety
of brain models [15][26], including both the off-line analysis
and on-line control, demonstrated that Cerebot achieved high
accuracy and reliability of controlling a humanoid robot.
The Cerebot system is based on a Cerebus data acquisition
system [27]: a professional multichannel acquisition system
used for bio-potential signals recording, processing and
displaying. The system dimension is approximate 0.123 and
weight is about 7.4kg. The Cerebus system, consisting of the
signal amplifier, the amplifier power and the neural signal
processor, can record simultaneously brain signals from 128
channels at a sampling rate of 30 kHz and with 16-bits
resolution [15].
Four subjects with normal or corrected to normal vision,
aged among 23 to 27 years old, participated in the comparative
studies. The subjects sat in front of the monitor with a distance
away from about 60cm. For the two brain-controlled robot
systems, we used a 64-channels standard electrode cap at the
same sample frequency of 500Hz. The two systems adopted
the same off-line analysis and on-line control modules except
the data collecting and recording procedures. The experimental
process has been examined by Tianjin medical university
general hospital ethical review committee, and all subjects
have signed a written informed consent before the experiment.
B. Off-line Analysis
For the off-line analysis, the subjects were required to
conduct 90 trails to collect samples data using the two systems

1021

TABLE II.

CLASSIFICATION ACCURACY /%

subjects
systems

S1

S2

S3

S4

Average

The portable
system
Cerebot

96.7

95.6

92.2

91.1

93.9

Off-line

96.7

95.6

91.1

98.9

95.6

93.3

94.4

92.2

85.6

91.4

On-line

The portable
system
Cerebot

93.3

95.6

90.0

88.9

92.0

respectively. In each trail, the program randomly selected a

flickering image as a target at which the subjects gaze for 4s to
induce SSVEP signals, and recorded the signals and target
order. The subjects took a rest of 3s to relax eye muscles
between two trails. The brain signals recorded by the two
brain-controlled robot systems were used for off-line
analyzing their performances as follows: loading the off-line
data, removing noise by a spatial filter, extracting feature
vectors using FFT, and applying Fisher to classify
frequency-domain characteristics. TABLE I lists the target
frequency-domain characteristics comparison of the two
systems. We adopted the 9*10 cross-validated method to
process 90 sample data acquired in each trial, i.e., we chose 9
sample data of them as the test data and the rest as the train data.
This procedure was repeated until all the data was processed.
Table II lists the classification accuracies.
From TABLE I, it can been seen that, for different stimulus
frequencies, the same subject exhibited the different
frequency-domain characteristics, for example, for S1 the
portable system delivered the characteristic of 31.2 at 4.61Hz,
while it did the one of 7.2 at 15Hz. It also illustrates that the
subjects produced the different responses to the different
stimulus frequencies. For the same system, the different
subjects exhibited the varying characteristics, which stated the
objective individual differences which were validated by the
research findings in physiology and cognitive psychology. We
took the average of the all subjects characteristics to compare
the two systems and concluded that all the average values were
over 9.0 with obvious SSVEP response features. The subjects,
except S1, had no experience on the experiments. The off-line
analysis on TABLE II shows that all the subjects achieved the
high classification accuracies. Furthermore, both of the
portable system and Cerebot achieved nearly the equivalent
classification accuracies because the average accuracy yielded
by Cerebot was only 1.7% higher than the portable system.
The brain data of subject S4 induced by the two systems
exhibited an obvious difference, but they were still acceptable
for education consideration.
C. On-line Control
We conducted the on-line control experiments for
comparative studies. For each trail, the program randomly
selected a robot behavior to be activated, so the subjects had to
gaze at its relevant robot image for 4.0s. The brain-controlled
systems processed the induced brain signals to check if the
activated behavior was correct. The experiment continued until
90 trials were completed to obtain the control accurate rates.
TABLE II also lists the classification accuracies achieved
by both brain-controlled systems during the on-line control
experiments. Generally, the on-line classification accuracies
were slightly lower than the off-line ones. The reason for that
could be: 1. in the on-line experiments, the subjects knew the
classification result on the spot which influenced the subjects
mental activities; 2. the Fisher parameters calculated in off-line
might not be perfect for the online experiments because the
two type of experiments were not conducted in the same day,
the subjects mental states were not under the same condition,
and the wearing of electrode cap could be different. All of
those could have some influence to the experiment result and
lower the classification accuracy in online experiment. The

(a)

(b)

(c)
Figure 6.

(a) Live video feedback (b) Scene (c) The moving of Lego

discussions above implied that the portable system achieved a

comparable performance of the Cerebot system.
VI. OPERATION TASK VIA LIVE VIDEO FEEDBACK
All the subjects achieved the desired performances when
they conducted both in the off-line and on-line experiments
using the portable system. Next, we tested an operation task of
the Lego robot via live video feedback, which could be used
for training concentration, improving psychological
adjustment ability, enhancing quick thinking ability and
direction-sense.
The subjects controlled the Lego robot via brain signals to
follow a path from its start to its end with obstacle avoidance in
an office environment, as shown in Fig. 6(a). Collision
numbers during the operation missions were recorded for
evaluation. Each subject conducted the operation task to gather
the total numbers of commands outputted and repeated the
mission three times. This operation task is much more
challenging, because for each control cycle a subject took 4.0s
to stare at a robot image and used next 2.5s to activate the
behavior corresponding to the command generated from the
data processing module. At the same time, the subject had to
focus on live videos to determine promptly a command based
on robot positions and surrounding information.

1022

TABLE III.

RESULT OF TASK

Subjects
Experiment
Command Num.
First
Collision Num.

S1

S2

S3

S4

17

27

22

23

Command Num.

18

23

22

26

Collision Num.

Command Num.

18

23

28

19

Collision Num.

Second

Third

Fig. 6(a) is a video shot on the experimental environment,

including the given path from the initial position to the
destination, as indicated by a red solid curve, and the white and
gray boxes as obstacles. This path was the only way to allow
the Lego robot to reach its destination from its initial without
possible collisions with obstacles. Fig. 6(b) shows the whole
experiment scene: a subject in front of the computer monitor,
the obstacles arranged in the environment, and the Lego robot.
Through a camera mounted on the celling, the subjects were
able to watch the experimental scene in real-time and
generated a series of proper robot behaviors to complete the
operation task. For each kind of mission, the Lego robot
moved forward about 30cm, turn left and turn right
approximately 30 in situ. During the operation mission, the
subjects had to focus on their mental activities, which could be
disturbed by tension, false command, collision, etc., to
complete the operation task.
TABLE III lists the experimental results indicating the
performances achieved during the three missions. We noted
that all subjects with different levels of accuracies
accomplished these tasks and did improvements after several
experiments, except for S3. In fact, during the experiments, S3
had a trouble to acquire the brain signals because the reference
electrode was unstable. The collision number represented the
control success rate to some extent. The subjects can control
the robot to walk with a longer distance to avoid collisions, but
it results in outputting more command numbers. For example,
S2 had not achieved a high accuracy, so S2 navigated the robot
to track a longer way without obstacle collision. What a
surprising to us was that S4 with the lowest on-line
classification accuracy improved quickly his control
performance in the following-up missions. This may imply
that good psychological reconciliatory ability and proficiency
play an important role like the classification accuracy does.

[3]
[4]

[5]
[6]

[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]

[15]

[16]
[17]

VII. CONCLUSION

[18]

In this paper, we developed a portable low-cost

brain-controlled robot system based on multi-parameter
bioelectric signal acquisition device for the teenagers
education. The signal acquisition device is low cost, easy to
carry, and can transmit wirelessly. The system uses the Lego
Mindstorms robot as the controlled object because Lego
Mindstorms robots are popular among teenagers due to its
moderate price and easy to customize and program. Through
the comparative off-line and on-line studies between the
portable brain-controlled system and Cerebot, the portable
system show us satisfactory performance. The portable
brain-controlled robot system not only plays an entertainment
role but also helps build up our DIY ability and holds
significant values to the teenagers education quality.

[19]
[20]

[2]

[22]
[23]

[24]

[25]

REFERENCES
[1]

[21]

J. R. Wolpaw, N. Birbaumer, W. J. Heetderks, D. J. McFarland, P.H.

Peckham, G. Schalk, E. Donchin, C. J. Robinson, T. M. Vaughan,
Brain-computer interface technology: a review of the first
international meeting, IEEE Trans. Rehabil. Eng., vol. 8, pp. 164
173, June 2000.
M. A. Lebedev, M. A. L. Nicolelis, Brain-machine interfaces: past,
present and future, Trends Neruosci. vol. 29, pp. 536546, July. 2006.

[26]

[27]

1023

A. Ortiz-Rosario and H. Adeli, Brain-computer interface technologies:

from signal to action, Rev. Neurosci., vol. 24, no. 5, pp. 537-552,
2013.
W. Li, Y. Li, G. Chen, D. Shen, E. Blasch, K. Pham, R. Lynch,
Acquiring neural signals for developing a perception and cognition
model, in Proc. Conf. Sensors and Systems for Space Applications V,
Baltimore, MD, 2012, pp. 838501-1-838501-9.
J. R. Millan, F. Renkens, J. Mourio, W. Gerstner, Noninvasive
brain-actuated control of a mobile robot by human EEG, IEEE Trans.
Biomedical Engineering, vol. 51, pp. 1026-1033, June. 2004.
L. A. Farwell, E. Donchin, Talking off the top of your head: toward a
mental prosthesis utilizing event-related brain potentials,
Electroencephalography and Clinical Neurophysiology, vol. 70, pp.
510523, Dec. 1988.
J. J. Vidal, Toward Direct Brain-Computer Communication, Annual
Review of Biophysics and Bioengineering, vol. 2, pp. 157-180, June
1973.
G. Pfurtscheller, C. Guger, G. Mller, G. Krausz, C. Neuper, Brain
oscillations control hand orthosis in a tetraplegic, Neuroscience
Letters, vol. 292, pp. 211-214, Oct. 2000.
G. R. Mller-Putz, R. Scherer, G. Pfurtscheller, R. Rupp, EEG-based
neuroprosthesis control: a step towards clinical practice,
Neuroscience Letters, vol. 38, pp. 169-174, July. 2005.
http://www.puzzlebox.io/
G. Hu, C. Zhu, The application of EEG and evoked potentials- EEG
biofeedback and BCI, The World Medical Devices, vol. 2, Feb. 2002.
http://www.lego.com/zh-cn/mindstorms/?domainredir=mindstorms.le
go.com
W. Li, C. Jaramillo, Y. Li, A brain computer interface based
humanoid robot control system, in Proc. IASTED Int. Conf. Robotics,
pp. 390- 396, Pittsburgh, 2011.
W. Li, C. Jaramillo, Y. Li, Development of mind control system for
humanoid robot through a brain computer interface, in Proc. 2nd Int.
Conf. Intelligent System Design and Engineering Application
(ISDEA), 2012, pp. 679 682.
J. Zhao, Q. Meng, W. Li, M. Li, F. Sun, G. Chen, An
OpenViBE-based brainwave control system for Cerebot, in Proc.
IEEE International Conf. Robotics and Biomimetics, Shenzhen, 2013,
pp. 1169-1174.
http://www.linear.com.cn/
J. Travis, J. Kring, LabVIEW for Everyone: Graphical Programming
Made Easy and Fun, US: prentice hall, 2006, pp. 3-20.
Q. Ruan, I with LabVIEW - ten years programming experience of a NI
engineer, CHN: Beihang University Press, 2012, pp. 1-4.
http://www.ni.com/labview/zhs/
P. Poryzala , A. Materka, Cluster analysis of CCA coefficients for
robust detection of the asynchronous SSVEPs in braincomputer
interfaces, Biomedical Signal Processing and Control, vol. 10, pp.
201208, Mar. 2014.
E. E. Sutter, The brain response interface: communication through
visually-induced electrical brain responses, Microcomputer
Applications, vol. 15, pp. 31-45, Jan. 1992.
D. Regan, Some characteristics of average steady-state and transient
responses evoked by modulated light, Electroencephalography and
Clinical Neurophysiology, vol. 20, pp. 238-248, Mar. 1966.
F. B. Vialatte, M. Maurice, J. Dauwels, A. Cichocki, Steady-state
visually evoked potentials: focus on essential paradigms and future
perspectives, Progress in Neurobiology, vol. 90, pp. 418-438, Apr.
2010.
G. Bin, X. Gao, Z. Yan, B. Hong, S. Gao, An online multi-channel
SSVEP-based brain-computer interface using a canonical correlation
analysis method, Neural Engineering, vol. 6, no. 4, pp. 1-6,
Aug.2009.
I. Volosyak, H. Cecotti, D. Valbuena, A. Graser, Evaluation of the
Bremen SSVEP based BCI in real world conditions, in Proc. IEEE
International Conf. on Rehabilitation Robotics (ICORR), Kyoto
International Conference Center, 2009, pp. 322-331.
J. Zhao, Q. Meng, W. Li, M. Li, G. Chen, SSVEP-based hierarchical
architecture for control of a humanoid robot with mind, in Proc. 11th
World Conf. on Intelligent Control and Automation (WCICA),
Shenyang, 2014, pp. 2401 2406.
http://www.blackrockmicro.com/content.aspx?id=13