Intelligent Control of Robotic Arm using Brain Computer Interface and Neural Networks

Project Title

Intelligent Control of Robotic Arm using Brain Computer Interface and Neural Networks

Project Area of Specialization NeuroTechProject Summary

We are developing intelligent control of robotic arms using a brain-computer interface and neural networks as part of our research project. Based on the required input, the microcontroller determines the degree of freedom of the actuators, and the robotic arm will act accordingly. Electroencephalography (EEG) is a test that uses electrodes attached to the scalp to measure the electrical activity of the brain. The data is collected using EEG headsets. To translate EEG data into movement patterns, a headset is used, which measures brain activity via the scalp of the head. The brain signals will be recorded using a Neurosky EEG headset. The robotic Arm is an extremely versatile robot that can be used in a broad variety of applications. The Degree of Freedom (DOF) is the single most significant factor in the design of a robotic arm. The mathematical complexity of a robotic arm is unquestionably the highest. Each degree of freedom corresponds to each arm joint, a point on the arm at which it may bend, rotate or translate. The robotic arm will be controlled by a micro-controller using brain signals as input. Unless input data is preprocessed, machine learning algorithms behave irregularly and improperly. Several popular algorithms can be applied for classification, including:
Random Forest, K-Nearest Neighbor (KNN), Logistic Regression, Support Vector Machine (SVM), Naïve Bayes, Gradient Boosting, and Decision Tree classifiers. Afterwards, we will apply this algorithm to our microcontroller, which will control our robotic arm accordingly.

Project Objectives

Final deliverables(Objectives):

• To create a prosthetic device that accumulates signals directly from the brain scalp.
• To create a robotic arm with defined functions.
• To develop a translation mechanism to drive a robotic arm with the help of BCI.

Internal deliverables:

• Acquisition of data set E.E.G. signals
• Amplifying and noise reduction system
• Feature extraction techniques and classification methods in Anaconda.
• Neural Network and Active Function

Project Implementation Method

The data in this project is acquired by the EEG headset which provides the alpha and beta rhythms that are used as features for machine learning models. The machine learning model intelligently predicts the classified class based on dominant features for the desired moment of the hardware.
The machine learning and artificial intelligence algorithms are divided into two sub-categories i.e., supervised, and unsupervised machine learning. Supervised learning requires the data for processing and learning whereas, unsupervised learning doesn't require the data for learning purposes. The choice of learning is straight away supervised machine learning because the data is acquired from the EEG headset that is to be used in the learning model.

Class	Thinking of Activity	Output
0	Movement of the left arm	Move Servo Left
1	Movement of the right arm	Move Servo Right
2	No moment	No Movement

The table above describes the categorization of the pre-processed data that is to be used for the learning purpose. Classification models are trained to classify the data into categories of the same types. The categorical data is useful in the classification algorithms whether the categorical data is to be fed or it is the output of the classification model. There are various types of classifier models, therefore the best model in terms of speed and accuracy is to be determined. The classifier models are further divided into binary, multi-class and multi-label classification. The multiclass classification technique is used to predict one or more than one classes.
Popular algorithms that can are used for multi-class classification in the project are:

1. k-Nearest Neighbors.
2. Decision Trees.
3. Naive Bayes.
4. Random Forest.
5. Gradient Boosting.

The models are prioritized based on their scores. The score is the accuracy of the model to predict the accurate target. The input data is unseen for the model to predict the classes.

Following is the pictorial representation of the scores of the models.

The best model to deploy in this project are Random Forest, KNN and Gradient boosting with an overall recommendation of 18%, 17% and 17% among all the deployed models.

Pyfirmata will be used to drive the controller according to class predicted

Class

0

1

2

Benefits of the Project

• The technology has the potential to assist with such conditions by either passing electrical signals onto the muscles, bypassing the damaged connection, or allowing the person to move again.
• The technology enables patients to use their thoughts to control robotics or prosthetic limbs so that they can move independently.
• The advantage of this solution is that the EEG pulse is located on the user's head and the user can wirelessly control the robotic arm. 
• Enhanced reliability of machine learning in Neuroscience.
• Longer lifespan, high dynamic response and efficiency.

Technical Details of Final Deliverable

This system has three final deliverables the first is to create a prosthetic device that accumulates signals directly from the brain scalp. The technical detail of that deliverable is that we make a prosthetic robotic arm that will move after getting signals from the brain with the help of the Neurosky headset. The second deliverable is to make a robotic arm with a defined function. A Robotic Arm is a highly adaptable robot that may be used for a wide variety of tasks. Using a Robotic Arm, you may do a wide range of jobs. A robotic arm is probably definitely the most complex robot that can be built. An arm's design is influenced by several factors, but the most fundamental of them is its degree of freedom (DOF). Increasingly, miniature robotic arms are benchtop-mounted and operated by electrical signals. They can be found working in a wide range of fields and environments. Although floor-mounted models are more prevalent for larger models, solid metal construction (often steel or cast iron) and 4-6 articulating joints are standard across the board for all models. To put it another way, the shoulder, elbow, forearm, and wrist of a robot are all modelled after the primary mechanical components of a human arm. A microprocessor and stepper motors power the great majority, if not all, of robotic arms, which can have up to six joints connecting to seven parts. For most industrial applications, an end effector is a specialized tool or attachment designed to execute a specific action or a recurring sequence of articulations. This allows the arm to be positioned exceedingly precisely. This enables incredibly accurate placement. For a proper movement with a defined function, it is necessary to accurately define its degree of freedom. Each degree of freedom corresponds to one arm joint, a point on which the arm can bend, spin, or translate. The number of degrees of freedom (DOF) will be equal to the number of actuators (a device that converts a control signal to mechanical motion) on the robot arm. As we know Robotic arms are capable of a broad variety of industrial production, processing, and manufacturing jobs, that is, an operation that demands exceptionally accurate, quick, and repeatable movement. The third final deliverable of the project is to develop a translation mechanism to drive the robotic arm with the help of BCI. We can achieve it by using pyfirmata which act as the linkage between signals and our robotic arm. So whole work is completed as firstly, we get signals from the brain, then apply signal processing on them, then apply feature extraction techniques on them then classify these signals, then apply neural network, and then convert the signal into device commands using pyfirmata and provide these signals to controller and then the movement of robotic arm occur with these signals.

Final Deliverable of the Project HW/SW integrated systemCore Industry HealthOther Industries IT , Others Core Technology NeuroTechOther Technologies Artificial Intelligence(AI), Robotics, Big DataSustainable Development Goals Good Health and Well-Being for People, Industry, Innovation and Infrastructure, Sustainable Cities and Communities, Life on LandRequired Resources

Elapsed time in (days or weeks or month or quarter) since start of the project	Milestone	Deliverable
Month 1	Literature Review	Report
Month 2	Robotic Arm Implementation and Interfacing of PWM-controller, robotic arm servos Servos and MCU	Robotic arm
Month 3	Deploying Machine learning algorithms for classification task.	Best approach for software implementation with insights
Month 4	Python-Controlled robotic arm interfacing	BCI-controller robotic arm
Month 5	Testing and bug fixing	Final Deliverable