See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/366215747

Non-invasive brain-computer interface for prosthetic limbs control in

Military

Conference Paper · December 2022

CITATIONS READS

0 190

3 authors:

Thendral Dhanabalan Neethu Chittilapilly Jose

Bournemouth University Bournemouth University
1 PUBLICATION 0 CITATIONS 1 PUBLICATION 0 CITATIONS

SEE PROFILE SEE PROFILE

Onyekachukwu Chukwuka
Bournemouth University
1 PUBLICATION 0 CITATIONS

SEE PROFILE

All content following this page was uploaded by Thendral Dhanabalan on 13 December 2022.

The user has requested enhancement of the downloaded file.

Non-invasive brain-computer interface for prosthetic limbs control in Military
Thendral Dhanabalan, Neethu Chittilapilly Jose, Onyekachukwu Chukwuka,
Department of Science and Department of Science and Department of Science and
Technology, Technology, Technology,
Bournemouth University, Bournemouth University, Bournemouth University,
ORCID- 0000-0002-0359-9376 ORCID- 0000-0001-9832-6751 ORCID- 0000-0003-0652-342X

Abstract-Brain-computer interface (BCI) has been The diversification in the drop of higher frontier function was
developed to gain a deep understanding of subsequent remarkable between military personnel [10]. Therefore, the
users for any kind of assistance or reliable tools for the controlling interfaces had improved by providing various
military by using electroencephalogram (EEG) features such a person could select the options concerning their
signals. Brain-Computer interface technology is needs that will lead to personalizing their medicine [16].
playing a major role in representing many significate
fields of research with application systems. Its The innovations in direct brain control of Robotic limbs
contributions in medical fields ranging from permit the users to think regarding the movements in similar
prevention to neuronal rehabilitation for serious ways they would prefer to move the arm to trigger the arm
systems [2] [4]. This development involves live recording,
injuries and towards the military its features have been
reporting, and decoding of motor neural signals to control the
trying to implement for warfighters, like controlling
prosthetic limbs. The EEG electrode for recording brain
prosthetic limbs in war zones and drones as auto-pilot,
waves creates appropriate algorithms to convert the neural
etc. Use of assistive prosthetic limbs to perform their activities into control commands for the motors throughout
tasks without any human interaction but these robots the robotic limb.
may require guidance and support from the user
which may not always be a possibility. A robotic system A brain-computer interface (BCI), sometimes known as a
that can be controlled merely by the brain signals of brain-machine interface (BMI), is a direct line of
the user would be remarkable. This paper illustrates communication between the electrical activity of the brain
the deep understanding of Robotic assistance with a and an outside device, most frequently a computer or robotic
non-invasive brain-computer interface. limb [17]. The Brain-Computer Interface is often used to
support, improve, map, study, sensory-motor processes or
Keywords – BCI, Electroencephalogram, Prosthetic
restore human cognitive [3]. Based on electrode placement to
limb, Neural Network, Brain signal.
brain tissue, BCI implementations range from non-invasive
(EEG, MEG, EOG, MRI) to invasive (microelectrode array)
I.INTRODUCTION to partially invasive (ECoG and endovascular).

The Fundamental application of the Brain- It is important to note that, even though BCI prototypes were
Computer interface of digital numeric values will be very recently developed, the fundamental concepts were
suitable for systems to manipulate accordingly [7]. A proposed as early as the 1970s. Initial efforts that were
data loader is usually known as a discrete data effective were based on the examination of the visual evoked
acquisition system. EEG signal concerns the visual potential in the brain triggered by variations in gaze direction.
perception of the subject or a patient as well as the pre-
Electroencephalography is most widely used to study
processing procedure used to evaluate the signal. The
interfaces in BCI for its easily evolving application,
interface is to dispense a technique for communication
portability and low cost regardless of the before mentioned
or movement by users who are unable to communicate
BCI system [19][1]. EEG is referred to as postsynaptic
and move under any circumstances [4][7]. For those
potentials with a bio-physiological signal that can be
with physical disabilities, there is significant interest in
obtained as a potential difference of oscillatory spikes on
the use of brain signals for communication and device
specific points on the scalp.
operation. Brain states can be recognized and translated
into activities like choosing a letter from an electronic In addition to providing sensory feedback through the direct
keyboard, engaging in video games, or actuating a robot stimulation of the nervous system, BCI closed-loop systems
arm [2]. also provide feedback to the user on system performance
through the presentation of sensory (mainly visual)
Thus, BCI mechanism approaches have comprised information [2].
decoding the recorded neural signals which are
correlated with sensory mechanisms into selection Thereby, EEG signals were measured with electrodes, and
commands or navigation, permitting the user to move they were placed on the scalp of the subject [1]. The sampling
across a real or virtual environment [3]. rate should be kept at 1,000 Hz and the notch filter of 60 Hz
to be applied. Before the acquisition of EEG data, all
electrodes should be below 10 Kohm's by
applying conductive gel [5][1]. classification using artificial intelligence approaches, EEG
signals obtained during control procedures must be
EEG signals can be used to representant the activity in processed using effective signal processing techniques [4].
cranial of a person related to a psychological task [4]. The To convert the identified task signals into the proper robot
electrodes which are placed on the surface of the brain are motions, an interface must be created. Appropriate
mostly used to record the electrical signals with reflections protocols and algorithms for EEG signals are important to
of inhibitory or excitatory postsynaptic potentials in controlling prosthetic limb movements. These protocols
dendrites in the layers of the cortex. and algorithms are to be formulated to allow users to
Electroencephalography [EEG] is a process that detects process their control over the system and then learn to
thought impulses, amplifies them and translates them into obtain flexible control in the number of trials of freedoms
commands. This procedure to be considered while we of operations [6]. The developed algorithms need to
record the EEG signal from the scalp, feature extraction undergo testing and validation. An efficient system for
methods should then be designed and building an operating a robotic assistant utilizing thought signals will
algorithm for the classifier of the EEG signal as statistical be the result of this research [2].
data to understand the changeover of time [11].
III. OBJECTIVE
EEG signals evoked by the thought to control robots are to The main objectives of the Non-invasive brain-computer
be analyzed to identify the feature pattern, classification interface for prosthetic limb control in the Military are of
and subsequent translation into control signals to control following. Initially, we need to detect and thereby
the prosthetic limbs [7]. Develop, diagnose, model, and interface in systems to
formulate the structured tasks.

A. Acquire the Electroencephalograph signals that

II. PROBLEM STATEMENT constitute the subject’s visual perception.

Currently, there are many BCI that is being used by Electroencephalography signals required to be recorded
thousands of people, such as deep brain simulators. The from the subject whereas distinguishing totally a different
extension of this concept to control robotic assistants will set of visual stimuli. Consequently, experiments towards
be investigated for military purposes [3]. the acquisition protocols are required to be developed so
the recorded signal may be related to the various
There are many disorders of the Motor Nervous System perceptions of pictures that are used for training the model.
(motor action) that occur due to any disruption of electrical Eventually, the obtained signal from the trained model
impulses in the brain and spinal cord that may impair both needs to be validated with pre-processed signals and the
physical and mental function. Such disorders of (MNS) are noise should be filtered by designing algorithm.
Amyotrophic Lateral Sclerosis (ALS), Multiple Sclerosis
(MS), Stroke, Parkinson’s Disease, and Paralysis [3]. B. Design and develop algorithms for extracting the
features from the EEG signal for visual perception.
During battles around the world, hundreds of thousands of
soldiers are exposed to combat and run the risk of dying or The EEG signal needed to be extracted where the
becoming hurt [3]. The majority of these survivors are frequency domain attributes are hidden. The feature
young, they must endure years of suffering from their extraction method regards the procedure of intensifying
wounds. This substantially reduces the quality of life for the feature in different dimensions of the subject’s EEG
those affected and places a tremendous burden on the signal by visual perception that involves the feature
military's medical and rehabilitation system. Military transmissions. Eventually, obtained good features are
service members can now survive serious battlefield chosen to constitute the distinguish between different tasks
wounds that necessitate traumatic limb amputation using of features of the perception.
Non-invasive BCI using EEG technology. This gives
C. Design subsequent ANN models for narrowing the
people who are living without a limb near-natural hand and user’s movement plans and visual perception.
arm control to overcome these difficulties [15][19].
Amputation or paralysis patients could regain function and The classification of various types of visual stimuli are to
independence and injured soldiers might one day be able be designed with classifiers. To function a proper result in
to return to active duty. classifier’s performance and classification we should
identify perfect model to equip the algorithm.
The signals from the brain can be acquired by extracting
the EEG signals using noninvasive electrodes on the scalp.
For feature extraction and subsequent task
D. Develop and implement optimization algorithms for It also requires developing algorithms that can accurately
real-time implementation. and reliably decode the brain's activity, even in noisy or
challenging conditions.
The command signal is the function for the movement of
the prosthetic limb with the use of Raspberry pi with the Recent advances in machine learning and neuroimaging
use of the best classifier so it’s easy for predictions of the have made it possible to develop more sophisticated BCI
subject’s visual perceptions. algorithms that can decode complex cognitive processes,
such as motor intention or emotional state. For example,
one study used fMRI to decode brain activity related to
hand movement, and then used this information to control
IV. BACKGROUND STUDY a robotic arm [20]. Another study used EEG to decode
Brain-computer interfaces (BCI) are a rapidly growing brain activity related to emotional states, and then used this
field of study that aims to develop technologies that allow information to control a virtual character [13].
direct communication between the brain and external In addition to decoding cognitive processes, BCI also has
devices, such as computers, robots, or prosthetic limbs the potential to enhance or augment these processes. For
[8][9]. These technologies have the potential to example, one study used transcranial magnetic stimulation
revolutionize many aspects of our lives, from medical (TMS) to modulate brain activity, and then used this to
rehabilitation to gaming and entertainment. improve memory performance in healthy adults. Another
BCIs are normally classified based on the EEG feature study used TMS to modulate brain activity in patients with
extractions [3] which are Oscillatory and Event-related. depression and found that this improved their symptoms
Event-Related Potential (ERP) in BCIs is developed to [25].
identify an EEG response to a trained superficial stimulus. Overall, the field of BCI is rapidly advancing, with many
They are efficiently robust to enable accurate time of the exciting developments and applications. However, there
ERP to the efficiency [4]. Oscillatory BCIs is used to are also challenges and ethical considerations that need to
control external signal with single power specified by EEG be addressed, such as the potential for misuse or abuse of
band frequency [5]. these technologies [12]. Further research is needed to fully
The P300 potential is a standardized neural analysis to understand the potential and limitations of BCI, and to
acquire visual stimuli [6]. The waveform is a large positive develop safe and effective applications.
deviation of electrical activity monitored at 300ms in post-
stimulus, the strength of the observed deflection being
inversely proportional to the frequency of observed targets V. NEURAL NETWORK
[8]. Out of other various ERP methods used in a human-
computer interface like Movement Related Cortical An artificial neuron is a human-made model motivated by
Potential (MRCP), Error Related Negativity (ERN), the real neurons. The real neurons receive signals and
P300 ERP is one of the strongest neural observables by communicate with other neurons from synapses located on
EEG, specifically when data are infrequently [6]. the membrane or dendrites of the neuron [10]. While the
signals received are strong enough (which contains a
The basic principle of BCI is to use sensors, such as certain threshold), the neuron will be activated and emits a
electroencephalography (EEG) or functional magnetic signal through the axon.
resonance imaging (fMRI), to measure the brain's activity,
and then use algorithms to decode this activity and This signal data will be sent to another synapse known as
translate it into commands for external devices. This Artificial Neuron. While modelling artificial neurons, the
process can be thought of as a "brain-to-machine" or complexities of real neurons are a major concern. In
"brain-to-computer" interface [15]. essence, these are made up of inputs like synapses that are
multiplied by weights that represent the intensity of the
One major challenge in developing BCI is to accurately corresponding signals and then calculated by a
decode the brain's activity. This requires understanding the mathematical function that decides whether the neuron
neural mechanisms underlying various cognitive will fire. The other function computes the artificial
processes, such as attention, memory, and decision- neuron's output, which is occasionally based on a
making [1][15]. predetermined threshold [3][16][7]. ANNs typically
incorporate artificial neurons to process data.
 results is be calculated. Backpropagation is used to reduce
errors [10]. From the random weights, the training begins,
and the output is adjusted by minimizing the error. The
backpropagation equation for the algorithm is majorly
inspired by the factor of using sigmoid activation
functions, is given in Equation.1.1.

Figure.1.1 Artificial Neural Network

Equation.1.1 Backpropagation Equation
VIII. METHODOLOGY OVERVIEW
VI. ARTIFICIAL NEURONS

In artificial neurons, the greater weight of an artificial A. Development of suitable protocols for the acquisition of
neuron is the power of the input. Weights can be negative thought signals for controlling a robotic assistant.
as well; on the other hand, we can say that the negative
weight inhibits the signal. The artificial signal sent to the Protocols for EEG signal acquisition are to be developed,
neuron will differ depending on the weight [4]. We can get suitable robot tasks are to be analyzed and chosen for
the desired output for certain inputs by adjusting the developing protocols for acquisition. In addition, a
weights of the neurons. However, it would be quite paradigm is to be developed for each acquisition protocol.
difficult to determine all the necessary weights manually B. Acquisition of EEG thought signals using noninvasive
when we are developing an Artificial Neural Network electrodes.
(ANN) with hundreds or thousands of neurons [14].
However, we can create an algorithm to modify and Subjects are to be trained to perform mental tasks as per
choose the ANN's weights in order to provide the desired the protocols developed; EEG signal acquisition is to be
results. This way of weight adjustment is known as repeated for several sessions.
Training or Learning. The number of uses and types of
ANN is very huge. From the first neural model, there was C. Development of algorithms for feature extraction from
a huge development of different models for ANN, as EEG thought signals.
shown in Figure.1.1. The differences between them might Suitable algorithms are to be developed using digital signal
be the topology, the functions, the learning algorithms, the processing techniques such as filtering, and band
accepted values, etc. [3][9]. Due to space considerations, frequency extraction to extract feature patterns from the
an ANN uses the backpropagation technique to learn thought signals of the individuals, as shown in Figure.1.2.
significant weights. It is one of the models that ANNs
utilize the most frequently, and numerous additional D. Development of algorithms for recognition of thought
algorithms are built on it. The process of information is signals for specific tasks.
used in the ANN function, which helps to relate to the
fields with it [21]. There is a wide variety of ANN that is Suitable algorithms are to be developed using neural
used to control animals, model real neural networks, study network models for the classification of features extracted
behavior, and used for engineering purposes and from the thought signals to recognize the mental task
machines, such as forecasting, data compression, and performed by the individual.
pattern recognition. E. Development of a thought-controlled device interface
VII. BACKPROPAGATION ALGORITHM for the robotic assistant.

The backpropagation algorithm is used as a layer with An interface is to be developed to translate the thought
feed-forward ANN, which means that the ANN is signals to control signals using microcontrollers.
classified in layers to send their signals as a forward
F. Experimentation and testing of the developed
network and then the propagated backwards are errors [8].
algorithms and interface.
The neurons receive the inputs in the input layer, and the
A Series of experiments and tests are to be conducted on
output is given by the neurons from the output layer. Next
the data set to validate the developed algorithms and
to it, there may be more than one hidden layer [17]. The
interface.
Backpropagation algorithm is used in supervised learning,
which provides the algorithm with examples of
the inputs and outputs that want network need to be
computed, and then the errors of actual and expected
G. Real-time verification of the developed thought MATLAB with the control module. With the help of
controlled robotic assistant. MATLAB, the features were retrieved from the recorded
EEG data, and an algorithm was created to categorize the
All the procedures mentioned above deal with recorded different tasks as shown in Fig.1.3.
(off-line) thought signals. To validate the possible real-life
implementation of the developed algorithm, real-time
testing will be investigated and verified.

Figure.1.3 Methodology

A. SIGNAL GENERATION

The human brain system is separated into three important

regions based on function the cerebellum or stem, which
corresponds with fundamental physical functions, the limb
movements, which helps in finding the flight reaction,
emotions and the fight expression, and the cortex that is
responsible for critical thinking, processing, and analysis
[12]. The cortex is additionally subdivided into parietal,
temporal lobes occipital, and frontal. These lobes' primary
tasks include visual data processing, movement, critical
thinking, sensation and memory storage. According to
neurophysiology theory, nerve cells are separate entities
that connect with one another via specialized contacts
known as synapses [6][19]. The nervous system is divided
into two basic parts the central nervous system that
includes the brain, spinal cord and peripheral nervous
system, which includes sensory neurons that connect sense
neuron receptors located throughout the body system to the
spinal Cord. Correspondingly, nervous system cells are
classified into two kinds of supportive glial cells and
Figure.1.2 Signal Processing & Control Action Flow
neurons capable of electrical signaling which do not have
chart
the potential to electrical signaling but execute various
IX. METHODOLOGY other principal roles [23]. A neuron is a kind of unique cell
that, despite certain similarities, diverges from a
Firstly, analyze the fundamentals of EEG. The channels stereotypical cell in many ways.
should be selected to get the data. Further, the tasks are all
chosen to be performed by the subject. Here EEG signal is There are numerous distinct types of neurons available in
used as Visual Evoked Potential (montages) for obtaining the human brain that are committed to diverse purposes
data [13]. nonetheless, a neuron generally consists of three parts
which are axon, cell body, and dendrites [1]. Dendrites,
The signals from arrays recognize for the spontaneous also known as dendritic processes, are the principal input
recording and training of over certain single units and then source from other neurons at synapses. These dendrites
processed in live acquisition to send commands for the include a high concentration of ribosomes and cytoskeletal
movements of the prosthetic limb [23]. Live visual protein, which aid in detail processing and reception.
feedback from the degree of the Robotic limb to the However, there are only a few varieties of neurons that
participant enables the closed-loop control. absence dendritic segment and hence have restricted input
capacities [17].
We now build and interface the artificial limb with
software with ROS (Robotic Operating System), which
enables the communication between an algorithm in
The synapse is devised of postsynaptic and presynaptic, affordable and portable. They lack spatial resolution
and components known as postsynaptic specialization and, because the signal must go through several physical
presynaptic terminal, respectively. Most synapses, obstacles, such as the skull and scalp, but they offer
nonetheless, lack cellular continuity linking these parts [3]. exceptional temporal resolution because they measure
Instead, an extracellular gap called the synaptic cleft brain activity directly [17][18]. Using commercially
occurs in between, which is used by neurons to interact available amplifier systems with electrodes positioned in
utilizing restricted chemicals known as neurotransmitters. elastic caps in accordance with international standards,
In human brains, each neuron may receive up to 100,000 EEG signals can be noninvasively monitored. Dry,
synaptic inputs [1]. After travelling between the cell body, capacitance-based electrodes, for example, have been
the integrated signal is observed at the axon origin, which created more recently, which could lessen the demands on
is the component of the cell in control of signal conduction. BCI users for long-term usage [19][20].
These axons can reach in length from a micron, which are
frequently found in interneurons, to nearly a meter, which
is located in the transmits impulses and spinal cord to C. EEG HEADBAND
remote portions of the human body [5]. Action potential
mentions the total activity of this electrochemical signaling EEG Headband is designed for scalable and is also
within a neuron. So, in general, the axon, body and, proposed for human brain research. It is quick and easy to
dendrites, and the element of a neuron are in charge of access brain data with this design. There are 14 channels
conveying information, processing, and receiving, and 2 references [18]. The designed confluence was able
respectively. to attain an actuation accuracy of 94 percent of the data
over 3-5 min the data can be acquired.
Glial cells, on the other hand, outnumber neurons three
times more and do not actively participate in synaptic D. EEG ELECTRODES PLACEMENT
connections [14]. The name glia is of Greek origin and
means glue, founded on the notion that these cells retain Electrodes were placed on locations like F3, F4, FP1, FP2,
the neurons simultaneously, however, there are very few C3, C2, OZ, T7, T8, CP4, P3, P2, P4, C4, CP3, as shown
evidence to support this. Glial cells have several vital in Figure.1.4. Whereas Fpz was mentioned here as ground
activities, including maintaining ionic homeostasis within electrode and near to the earlobes the electrodes placed
the brain, assisting in damage repair, providing a were reference electrodes [14]. The actuation accuracy of
framework for neuronal growth, and controlling over 94% was able to attain while interfacing the design.
neurotransmitters around synaptic clefts [17]. The CNS
contains four kinds of glial cells are ependymal cells,
microglia, oligodendrocytes, and astrocytes.

The astrocytes with star-like intricate activities are in

charge of preserving an applicable chemical domain for
brain communication, as depicted in Fig 1.2.
Oligodendrocytes aid in signal transmission through axon
myelination in the CNS. Schwann cells of the PNS
accomplish the same thing [11][6]. Microglia removes
cellular detritus from damaged sites. It also produces a
field of cytokines to reign local tenderness. In determined
ways, they act similarly to macrophages or immune cells.
Finally, ependymal cells are in charge of the formulation
of cerebrospinal fluid. The optimal signal set down
distance for a single neuron, whereas the degeneration of
neurons foreign object placement in the brain [17].
Figure.1.4. Electrode placement
B. SIGNAL ACQUISITION
E. AMPLIFIER
In the case of EEG-BCI, which is one of the most popular
methods for neural recording, the electrodes are merely The signal that the electrodes pick up is far away and
applied to the scalp at particular locations to capture weaker since it must travel through multiple layers.
averaged neuronal impulses from various intracortical Because of this, an amplifier is required to convert
regions [16]. Systems based on EEG are typically microvolts into a range suitable for digitization. Through a
cable that is 1-2 meters long, the signal is transmitted

.
to an amplifier. Sadly, the cables can serve as antennae and • The electrodes and leads
pick-up signals, interfering with the EEG signal and • The subject: heartbeat electrical activity, eye
amplifying noise. To prevent this noise interference, some blinking, eye movements, and general muscle
"active" electrodes have a tiny internal pre-amplifier [20]. contractions.
F. A/D CONVERTER The main aim of signal pre-processing is to extract
adequate details from the obtained channel signal. The
The amplified signal will be transformed from analogue to
option of selecting signal over pre-processing is complex
digital form by the A/D converter [4]. Since EEG
and can have a significant impact on the performance of
transmissions only have a bandwidth of about 100Hz,
the assistive prosthetic limbs. Signal processing concerned
200Hz is more than enough to sample EEG signals.
with the analysis of data is obtained by EEG [19]. The
X. DATA ACQUISITION signal pre-processing is meant for the process of noise
removal and filtering [8]. The filter that can be used here
Firstly, the data is obtained by an examination regulate to is a bandpass filter for fast response and at appropriate
govern the Prosthetic limb employing bandwidth, window sizing. Finally, the signals are plotted
Electroencephalogram signals [6]. At the commence of the in the graph for the best record, as given in Figure.1.5. The
data acquisition procedure, the subjects are precisely following are the methods for processing the signal,
literate about the objectives, aims of the written consent,
and project is taken.[4] The observation is managed in an • Noise Removal -Cleaning up unwanted data.
outlying room to avert any external distractions. • Signal Processing -shaping the data characters.
• Multiple Removal -Removing unwanted coherent
The process of sampling signals that are calculated based
energy.
on actual physical conditions is known as data acquisition.
It translates the acquired samples into digital numeric After the sensor array data has been collected and stored in
values that will be suitable for systems to manipulate computer memory, the first computational stage begins
accordingly [7]. A data loader is usually known as a [11]. Signal pre-processing is to draw out pertinent data
discrete data acquisition system. EEG signal concerns the from the acquired channel signal. The effectiveness of
visual perception of the subject or a patient as well as pre- prosthetic limbs can be significantly impacted by the
processing procedure used to evaluate the signal. signal preprocessing method that is selected [10]. The
signal pre-processing is mainly meant for filtering and
The sampling and trail rate and values of the obtained
noise removal. The signals are plotted as graphs using
signal is 256 Hertz [4] and the result of the cognitive
Matlibplot in python. Chebyshev 2 bandpass filter is used
software is remitted to the Raspberry pi board to command
because of its best accuracy. The raw data obtained from
the Prosthetic limb [3]. The Raspberry pi microcontroller
the EEG headset is sent to the python platform for signal
is normally used in reign projects and is selected since its
preprocessing. The objective of choosing python to
open-source character. Eventually, a progressing signal is
perform the device for real-time operation.
consigned to the position of servo motors within the
Prosthetic limb to perform the movement of picking and
placing an object [13].
XI. SIGNAL PREPROCESSING

The raw data from the EEG signal is necessary for

preprocessing before feature extraction. This
preprocessing contains processes of removing noisy
frequency bands, median the current brain level activity,
then transforming the signals which were measured from
the scalp potentials to cortex potential and removing the
Figure.1.5 Filtered EEG Signal (Alpha Band)
noise from the frequency [5]. The four primary sources of
noise and artifacts are as follows: The rhythm used was alpha and its frequency is over 8-
12Hz.In general, seven frequency bands are used, and they
• EEG equipment are delta with the range of 0.1-0.4Hz, theta with a range of
• External electrical interference to the recording 4-8Hz, beta1 with a range of 13-15Hz, beta2
system and the subject
with the range of 15- 18Hz, beta3 with a range above A collective ERP of the sinusoid signal can be obtained as
18Hz, gamma with the range above 36Hz [11]. These are different negative and positive spikes in an EEG signal.
shown in Table.1.1 and it also shows some consequences Even-Related Potentials are generally labeled based on the
of those frequency bands. sequence at which the spikes trigger followed by an
enumerative suffix that denotes the spike’s polarity [21].
Table.1.1 Band Frequencies
B. FAST FOURIER TRANSFORM (FFT)

Fourier Transform (FFT) is used for signal analysis when

the signal breaks down into constituent sinusoids of
different frequencies. Discrete Fourier Transform (DFT) is
used to accomplish Fourier Transform (FT). The fast
Fourier transform (FFT), which is not a separate transform,
is an approach for calculating the DFT of a series [3]. It is
particularly utilized in the processing of EEG signals,
where techniques including filtering, frequency analysis,
and power spectrum estimates are all utilized. Each level's
computation using FFT reveals the frequencies that each
band contains. The most real-time pulses like
Electroencephalography signals which are moveable
signals. On the other hand, the FFT transformation can
particularly be applied by predicting the signal data is
static [17]. This is due to amplitude modifications in a
signal that are meant to be encrypted under an individual
signal remodeling of Fourier Transform for a dynamical
system that was nearly invariable in their structure by
choosing a very short-sized segmentation while window
A. VISUAL EVOKED POTENTIAL (P300) sizing [19]. Hence, the low-frequency component
disregarding where the oscillation of time frame is greater
Visual Evoked Potential (VEP) will ultimately lead to the than the length and size of the time frame that able to
research of the rhythm of oscillatory for monitoring brain obtain a time-frequency analysis towards the signal. Every
activity [7]. In our project, as the first step, we are taking period of interval the observation takes place to ensure the
the signal about P300 of eye movement to articulate the component frequency changes in the signal. However, the
visual movements and signals. On mapping the visual uncertainty principle limits the Shift Fourier Transform
feedback responses of the subject’s brain activity to the remodeling the algorithm with the frequency and time
types of tasks like visual observation methods which are components which should not be recorded under both high
monitored to analyze the signal [9][21]. resolutions [24].
To establish, there is much research interconnected to EEG C. FEATURE EXTRACTION
signals that have been revealed by many previous
researchers which involve the oscillatory activity of Succeeding the preprocessing stage, the patronage step is
dynamics, Event-Related Potentials, Evoked Potentials, to extricate attribute from the obtained signal value. There
and slow cortical potentials [21]. However, the subject of are two ordinary kinds of characteristics that can be
interest in this participant is concentrated on Visual extracted, that is frequency and time domain [13].
Evoked Potential which will finally lead eventually to the Frequency-based characteristics have been frequently
research of the brain activity towards oscillatory rhythm. utilized because of how simple they are to use, how
efficient they are computationally, and how simple they
An ERP will be examined as taking turns negative, and are to comprehend. In the particular examination, at most
positive triggers in an Electroencephalography signal. frequency state attributes are used, therefore the estimation
Event-related potentials are normally marked in relation to time necessary to reckon this attribute is reduced [18].
the progression at which the spikes simulate followed by
an enumerating suffix that denotes the spike’s polarity [7]. D. FEATURE CLASSIFICATION
For example, the peak value from P3 would indicate the Applying classification algorithms to the signal, which is
third simulation of positive spikes in the signal. The N1
now largely free of artifacts, is the next step that has to be
channel would denote the first occurring negative spike. taken.
The channel’s name depends on the latency of the of the
peak and its information. As an instance, post-stimulus of
the evoked potential would be renowned at 300ms [13].
It is possible to train a classifier to recognize which XIII. SINGLE BOARD COMPUTER (SBC)
features, for example, belong to one or more classes using
machine learning techniques. Once more, categorization Raspberry Pi is used as Single Board Computer; Raspberry
aids in determining the type of mental task the person is pi has the capability to process the signal and control the
engaged in [22]. Feature classification is a way of arm on the same board. If we chose other controllers such
classifying the band frequency and analyzing the particular as Arduino and Beagle bone black, they won’t perform
frequency.[11] The classification is most important to train both operations, so Raspberry Pi is chosen to do both
the controller and make sure what kind of frequency it operations on the same board itself [2]. Raspberry pi is
should take for the movement of the arm, that is when the used to process the signal and also used to perform feature
frequency is over 10 - 15 Hz the controller must be trained extraction and classification. The control of the Prosthetic
that it should have the movement at right arm.[1] limb is trained in Raspberry Pi by the data configuration of
subject signals [23]. Those signals are from Arduino which
To make sure of analyzing the signal to perform the gives a triggered pulse and consequentially with respect to
selective task the respective data configuration must be those signal servo movement is trained for the pick and
taken into account.[3][12] By concluding with all the place of an object.
tasks, the signal selection for a frequency is analyzed
perfectly. XIV. PROSTHETIC LIMB

E. NETWORK TRAINING Better resolution and sensitivity are provided by machine

vision. It frequently relates to identifying, examining, and
The data set is obtained with the specific feature trained in directing the robot controllers for small parts. The most
Raspberry pi. Raspberry pi, the classified signal of band crucial factor in the automation sector is vision [7].
frequency is directed for the movement of the servo in a Numerous industrial applications, including painting,
Prosthetic limb [17]. In this training process, an algorithm welding, undersea operations, automatic pick-and-place
is used for band frequency to perform respective arm objects, military operations in space, agricultural activities,
movements. The specific data is collected from the subject etc., can be carried out with it [20].
by undergoing tasks which are data configuration.
The most time-consuming and expensive task is palletizing
XII. DATA CONFIGURATION or arranging items in a precise way. Palletizing is robotics'
key contribution. These objects are aligned and random; a
Data is collected for the session, each session has five trials vision system must recognize them and put them in the
per task, and each trial lasts for 10s. Subjects are seated in proper order [2].
a comfortable chair, the room environment must be less
noisy as patients are expected to perform selected tasks Five Degrees of freedom Prosthetic limb was chosen for
[17][2]. The tasks referred to the trial are as follows, better accuracy to pick and place the object from one point
to another point which can be done by Solid Works
Task 1- Neutral: Software, as given in the Figure.1.6. The lightweight
material was chosen to avoid the weight of its own. The
Task 1 involves relaxation, so the subject is advised not
servo motors are used to control the robot's action and to
to perform specific tasks. Relaxation is possible only
pick the object using a gripper [12]. The base rotation is
when the subject cooperates not to think of anything in
controlled by two motors to maintain stability and also to
particular.
withstand critical overload conditions.
Task 2- Left:
For vision-based grasping, numerous strategies have been
This task is where the subject is requested to monitor the put forth. The robot requires prior scene analysis training
left object. And to imagine moving his left hand in the in order to perform grasping tasks [3]. It necessitates
direction of an object. completing two processes, such as grasp planning and
object reconstruction. Techniques like fuzzy logic [3],
Task 3- Right: neural networks, image and position-based visual serving
[2], etc. are needed to grasp an object. The concept behind
The subject is requested to monitor the right arrow.
the work is interacting and gripping items. Vision-based
Further, imagine moving his right hand in the arrow
grasp planning and vision-based pick-and-place control
direction.
are the driving forces behind this study [11].
 XVI. LIMITATION

Although EEG-based BCI will provide able-bodied people

with a convenient way of living, there are still certain
constraints to take into account. The challenges for BCI
applications can be distinguished as technology-related
and user-related; the former include technical issues and
the system's usability, while the latter deal with a person's
ability to learn how to use the BCI application and their
subjective ability to interpret the cues that produce or
modify the acquired EEG signals [5].

Figure.1.6 Prosthetic limb The hardware for signal acquisition is the first issue. The
sensors still have acquisition restrictions for EEG. The
A working table with rectangular objects is in front of a set signal travels a long way before the EEG machine picks it
of six DOF robots in the experimental setup. The subject up, and noise and artefacts of a mechanical,
is positioned in any comfortable area from which all items electromyographic, or electrooculographic character pose
are observed and linked to the computer program. Machine serious issues [24]. Technology must be dependable
programming is carried out with the aid of Python to sort despite the noise that devices make. The creation and
the object on the substratum of the object's region [15]. application of self-paced BCIs, which require the accurate
Machine programming includes object detection, image detection of a variety of different patterns in ongoing brain
preprocessing, pixel-to-world coordinate conversion using activity, is another significant challenge for the future. A
transformation characteristics, and object orientation BCI solution must decrease classification errors and
calculation in order to direct the robot's angle rotation [11]. training time while increasing the information transfer rate
(ITR) and its wide applicability in order to comprehend an
XV. RESULTS AND DISCUSSION operator's brain output [25]. The effectiveness of the signal
that is acquired and the applicability of the AI algorithms
After extricating features from the obtained signal, those
determine reliability. Additionally, some
attributes are divided on the report of definite classifiers
neurophysiological markers should be carefully chosen
and parameters. The classified signal is sent to Raspberry
and have a low latency since they are resistant to
pi and those signals are trained and processed using the
background noise. Additionally, a BCI must also be
Node-Red platform by taking subsequent frequency.
versatile and adjustable in order to be useful and equally
The pick and place movement are trained by taking a few reliable for a variety of users.
trails sets and task which is preferred by many subjects are
For class differentiation, most of the current classification
noted. For real-time operation, each servo in the Prosthetic
techniques demand extensive training. The time-
limb is programmed with different data configurations.
consuming stage of user training greatly affects how
EEG data also needs to be filtered to remove signals not effectively the classifier performs. The number of training
associated with brain activity, including eye movements, assemblies held and apprehension of the intricacy of brain
heart rate, noise, and voluntary muscle activity, from signals are typically trade-offs. Because they need a great
nearby electronic equipment. During specific states of rest deal of attention, and mindfulness of rapid and intermittent
or activity, the frequency ranges can be monitored and input, the majority of BCIs in use today are very draining
removed, and possibly place the electrodes close to the on the user.
scalp surface [8].
Along with the electrode fatigue problem, BCI may not
Suitable algorithms are to be developed using digital signal function if the user is unable to maintain a high level of
processing techniques such as filtering, and band focus. In BCI, the user's attention point changes the input
frequency extraction to extract feature patterns from the while mental activity is continuously tracked. Input and
thought signals of the individuals. The Raspberry pi output are produced as a function of the concentration
receives the control signal and controls the movement of required for stimuli [21][17]. Instead of unwinding, the
the robot according to the type of control signal [15]. From user must focus on one input point before examining the
the uncertainty matrices. of every algorithm, we can results. The interaction eventually takes on a forced quality
monitor that the remaining signal is recognized accurately rather than the natural quality that would exist if the user
by the classifiers with 100 percent precision could select which aspect of the visual output to focus on
[23].
The effectiveness of BCI is strongly affected by The field of BCI is still in its early stages, and there is much
neurological factors, such as functionality and work to be done in order to fully realize its potential. Some
architecture, cognition-related neurophysiology, and key areas for future research include the following, which
mental and emotional mechanisms. This results in are developing more advanced algorithms for decoding
significant inter- and intra-individual variation [21]. The brain activity. This will require a better understanding of
psychological factors of memory load, concentration, the neural mechanisms underlying various cognitive
conflicting cognitive processes, fatigue and the unique processes, as well as the development of new machine
qualities of the user, like gender, lifestyle, and age, all have learning techniques and neuroimaging methods,
an influence on immediate brain dynamics. Participants Improving the reliability and robustness of BCI systems.
that exhibit less empathy exhibit less emotional This will require addressing challenges such as noise,
involvement in a P300-BCI paradigm and produce larger artefacts, and individual differences in brain activity, as
P300 wave amplitudes than those who exhibit more well as developing methods for calibrating and adapting
empathy [15]. BCI systems to individual users and developing new
applications for BCI. This will require identifying new
Usability is another crucial factor and essential domains where BCI can be useful, such as gaming,
prerequisite for end users' successful adoption [23]. For
education, or social communication, and then developing
routine use outside of laboratories, current electrode caps
appropriate technologies and algorithms to support these
are not feasible. More specifically, future recording
applications, addressing ethical, legal, and social issues.
devices will need to be less expensive, easier to wear and
As BCI technologies become more advanced and
compact, quicker to set up, light enough to allow for
widespread, there will be important questions to consider
freedom of movement in various locations, non-wearying
regarding their potential misuse or abuse, as well as their
for the user, multitasking-compatible to enable regular
impact on society and individuals. And overall, the future
tasks without restrictions, and affordable in terms of
of BCI is full of exciting possibilities and challenges. As
training time and specialized resources.
researchers continue to advance the state of the art in this
Finally, there may be ethical concerns with the research field, we can expect to see many new and innovative
applications made for EEG-based BCI. There are various applications of these technologies in the coming years .
ideas about the moral implications of BCI because it
depends on several factors social, psychological, and
physical elements. It has been demonstrated that by XVIII. CONCLUSION
utilizing EEG-based BCI equipment, it is possible in some
situations to monitor users' subconsciously sensitive In the last 20 years, brain engineering has advanced
information [3]. A particularly extensive distribution of significantly and the vast majority of BCI research that has
non-medical gadgets could be used to manipulate user been published uses non-invasive, EEG-based BCIs.
emotions to promote targeted commercial and political Applications for non-invasive EEG-based technology and
objectives [12] if the ethical and security aspects of BCI interfaces are far more varied. The skull dampens signals,
applications are not managed. These factors would need to dispersing and blurring the electromagnetic waves
be considered, and users would need to be made aware of produced by the neurons, making EEG-based interfaces,
any potential dangers and duties. To direct the general despite being simple to wear and not requiring surgery,
development of BCI technology and proactively avoid have a relatively poor spatial resolution. Additionally, they
ethical difficulties, ethical guidelines would need to be are unable to use higher-frequency signals effectively.
created [16]. When compared to non-EEG-based and intrusive
interfaces, those based on EEG also require some time and
XVII. FUTURE WORK preparation before each usage session. Despite having
The uses of non-invasive brain-computer interfaces span advantages over non-invasive BMIs, invasive signal
recording systems still have limitations and disadvantages.
many different industries and are not just restricted to the
One option for combining the benefits of both methods
military sector. With patients who have mobility
challenges, BCI can assist them to regain lost functions while essentially avoiding their limitations is hybrid BMIs.
and mobility levels, or at the very least, help them adjust However, hybrid technology is still in its infancy, and there
to their new limitations. BCI-based prosthetic limbs, also is a need for progress in hybrid BMIs containing three or
more signals. To get these hybrid devices on the
known as neuro-prosthetic devices, can be utilized by
commercial market, significant problems with hybrid
patients who are unable to regain their pre-injury levels of
BMIs such as user burden, system complexity, and high
communication or mobility to achieve normal
functionality. Mobile robots can assist those who are cost, also need to be resolved. Overall, a variety of factors
imprisoned in carrying out regular tasks. determine which BCI is ideal for each user.
 ACKNOWLEDGEMENT [12] G.N. Ranky, S. Adamovich New Jersey Institute of
Technology Biomedical Engineering Department, Fenster
We thank the anonymous referees for their useful Hall Newark, NJ 07103, Analysis of a Commercial EEG
suggestions. And thanks to our university and staff Device for the Control of a Robot Arm.
members.
[13] C. James and S. Wang, “Performance analysis of a
REFERENCES P300 BCI speller through single channel ICA,” Advances in
[1] Jose Principe, “Brain Machine Interfaces: Mind over Medical, Signal and Information Processing, 2008.
Matter”,2005.http://www.ece.ufl.edu/publications/Archives MEDSIP 2008, pp. 1–4, July 2008.
/inthenews/2005/brai nmachine.html Michal Teplan,
“Fundamentals of EEG Measurement”, Measurement [14] Farwell, Lawrence Ashley, and Emanuel Donchin.
Science Review, 2002, Vol. 2. "Talking off the top of your head: toward a mental prosthesis
utilizing event-related brain potentials."
Electroencephalography and clinical Neurophysiology 70,
[2] Jonathan R. Wolpawa, Niels Birbaumerc, Dennis J. no. 6 (1988): 510- 523.k8
McFarlanda, Gert Pfurtschellere and Theresa M.
Vaughana,”Brain-Computer interfaces for communication [15]Hortal, Enrique, Eduardo Iáñez, Andrés Úbeda, Carlos
and control.” Clinical Neurophysiology: Official journal of Perez-Vidal, and José M. Azorín. "Combining a Brain–
the international federation of Clinical Neurophysiology, Machine Interface and an Electrooculography Interface to
vol. 113, no.6, pp.767 91, Jun 2002. perform pick and place tasks with a Prosthetic limb."
Robotics and Autonomous Systems 72 (2015): 181-188
[3] Thurlings ME, van Erp JB, Brouwer AM, Werkhoven
PJ. EEG-based navigation from a human factors [16] Schomer, D. L., and da Silva, F. H. L. (2012).
perspective. In: Tan DS, Nijholt A, editors. Brain–computer Niedermeyer’s Electroencephalography: Basic Principles,
interfaces, human–computer interaction series. London: Clinical Applications, and Related Fields, 6th Edn. Alphen
Springer-Verlag; 2010. p. 71–
aan den Rijn: Wolters Kluwer Health Adis (ESP).
86,http://dx.doi.org/10.1007/978-1-84996-272-8 5.
[4] F. Lotte, “Signal Processing Approaches to Minimize or [17] Wolpaw, J. R. (2007a). “Brain-computer interfaces
Suppress Calibration Time in Oscillatory Activity-Based (BCIs) for communication and control, in ASSETS’07,” in
Brain-Computer Interfaces,” Proceedings of IEEE, vol.103 Proceedings of the Ninth International ACM SIGACCESS
no.6, pp.871 890, Jun 2015. Conference on Computers and Accessibility, (New York,
NY), 1–2. doi: 10.1145/1296843.1296845
[5] Reza Fazel-Rezai, Brendan Z. Allison, Christoph Guger,
Eric W. Sellers, Sonja C. Kleih and Andrea Kübler, “P300 [18] Babiloni, C., Pizzella, V., Del Gratta, C., Ferretti, A.,
brain computer interface: current challenges and emerging and Romani, G. L. (2009). Chapter 5 fundamentals of
trends,” Frontiers in Neuroengineering. Vol. 5, no. 14, 2012. electroencefalography, magnetoencefalography, and
functional magnetic resonance imaging. Int. Rev.
[6] G. Pfurtscheller and C. Neuper, “Motor Imagery and
Neurobiol. 86, 67–80
direct brain-computer communication,” Proceedings of
IEEE, vol. 89, no. 7, pp. 1123 1134, Jul 2001. [19] F. Popescu, S. Fazli, Y. Badower, B. Blankertz and K.
[7] J. Polich, “Updating p300: An interrogative theory of R. Muller, "Single-trial classification of motor imagination
{P3a} and {P3b},” Clinical Neurophysiology, vol. 118, no. using 6 dry EEG electrodes", PLoS One, vol. 2, no. 7, e637,
10, pp. 2128 2148, 2007. 2007.

[8] Damper RI, Burnett JW, Gray PW, Straus LP, Symes [20] Y. M. Chi, Y. T. Wang, Y. Wang, C. Maier, T. P. Jung
RA. Hand-held text to-speech device for the non-vocal and G. Cauwenberghs, "Dry and noncontact EEG sensors
disabled. J Biomed Eng 1987;9:332–340. for mobile brain-computer interfaces", IEEE Trans. Neural
Syst. Rehabil. Eng., vol. 20, no. 2, pp. 228-235, Mar. 2012.
[9] LaCourse JR, Hladik FC. An eye movement
communication-control system for the disabled. IEEE Trans [21] Tavakoli, P., Campbell, K. The recording and
Biomed Eng 1990;37:1215–1220. quantification of event-related potentials: I. stimulus
[10] Saeed, Sanay Muhammad Umar, et al. "Psychological presentation and data acquisition. June 2015, The
stress measurement using low cost single channel EEG Quantitative Methods for Psychology 11(2):89-97,
headset." Signal Processing and Information Technology DOI:10.20982/tqmp.11.2.p089
(ISSPIT), 2015 IEEE International Symposium on. IEEE, [22] Ochoa Jorge Baztarrica, "EEG Signal Classification for
2015. Brain Computer Interface Applications", 2002.
[11] M. Davies and C. James, “Source separation using
single channel ICA,” Signal Processing, vol. 87, no. 8, pp.
1819–1832, 2007.
[23] Lin, C.-T., Lin, B.-S., Lin, F.-C., and Chang, C.-J.
(2014). Brain computer interface-based smart living
 environmental auto-adjustment control system in UPnP
home networking. IEEE Syst. J. 8, 363–370. doi:
10.1109/JSYST.2012.2192756
[24] Fatourechi, M., Bashashati, A., Ward, R. K., and Birch,
G. E. (2007). EMG and EOG artifacts in brain computer
interface systems: a survey. Clin. Neurophysiol. 118, 480–
494. doi: 10.1016/j.clinph.2006.10.019
[25] Feredoes, E., Silvanto, J., & Pascual-Leone, A. (2011).
Boosting brain function with non-invasive brain stimulation.
Nature Neuroscience, 14(10), 1219-1221.

View publication stats