Robotic Arm for Mimicking Human Arm Movements
Primary Faculty Mentor’s Name
Dr. Fernando Rios-Gutierrez
Proposal Track
Student
Session Format
Poster
Abstract
In this study, a robotic arm will mimic human arm movements using 3D motion tracking sensors. Robotic mimicking is a popular method to instruct robots to perform specific operations. The advantage of this implementation is instead of using a complex control system, a human can perform operations they want the robot to emulate.The robotic system will be capable of helping individuals who have disabilities or assist individuals who may need physical therapy to repair specific motor skills.
The 3D motion tracking sensors will be placed on a human arm at two locations - the wrist and the upper arm. At a sampling rate of 100 Hz, the sensors track the xyz-coordinate inertial data and the euler angles of the subject's arm as they perform the movement . The raw data will then be averaged by calculating the Root Mean Square (RMS) value and the Rectified average. The normalized data will then be used to train a multi-layer, feed-forward Neural Network.
The Neural Network is trained utilizing backpropagation learning by which the network will process the data by pattern recognition. The trained Neural Network will process outside data and assign an output based on the pattern. Each output will represent a specific movement that has been performed.
To minimize processing time, the trained Neural Network will be generated as a C code and placed on a microcontroller. Once uploaded to the microcontroller, raw data may be fed into the network via the pin inputs of the microcontroller board. From there, the trained Neural Network will generate an output that will be assigned to a specific function that, via the pin output of the board, will prompt the robot to execute the proper movement.
Currently, there are 8-10 movements that have been chosen for the arm to emulate. Due to limitations of non-continuous rotation servos, some movements are executed as an analogue to the human arm movement. Going forward there may be room for a broadening of accomplishable movements.
Keywords
Robotics, Robotic Mimicking, Neural Networks, Machine Learning, Electrical Engineering
Award Consideration
1
Location
Concourse and Atrium
Presentation Year
2015
Start Date
11-7-2015 10:10 AM
End Date
11-7-2015 11:20 AM
Publication Type and Release Option
Presentation (Open Access)
Recommended Citation
Stroud, Joshua L. and Augusma, Imani, "Robotic Arm for Mimicking Human Arm Movements" (2015). Georgia Undergraduate Research Conference (2014-2015). 24.
https://digitalcommons.georgiasouthern.edu/gurc/2015/2015/24
Robotic Arm for Mimicking Human Arm Movements
Concourse and Atrium
In this study, a robotic arm will mimic human arm movements using 3D motion tracking sensors. Robotic mimicking is a popular method to instruct robots to perform specific operations. The advantage of this implementation is instead of using a complex control system, a human can perform operations they want the robot to emulate.The robotic system will be capable of helping individuals who have disabilities or assist individuals who may need physical therapy to repair specific motor skills.
The 3D motion tracking sensors will be placed on a human arm at two locations - the wrist and the upper arm. At a sampling rate of 100 Hz, the sensors track the xyz-coordinate inertial data and the euler angles of the subject's arm as they perform the movement . The raw data will then be averaged by calculating the Root Mean Square (RMS) value and the Rectified average. The normalized data will then be used to train a multi-layer, feed-forward Neural Network.
The Neural Network is trained utilizing backpropagation learning by which the network will process the data by pattern recognition. The trained Neural Network will process outside data and assign an output based on the pattern. Each output will represent a specific movement that has been performed.
To minimize processing time, the trained Neural Network will be generated as a C code and placed on a microcontroller. Once uploaded to the microcontroller, raw data may be fed into the network via the pin inputs of the microcontroller board. From there, the trained Neural Network will generate an output that will be assigned to a specific function that, via the pin output of the board, will prompt the robot to execute the proper movement.
Currently, there are 8-10 movements that have been chosen for the arm to emulate. Due to limitations of non-continuous rotation servos, some movements are executed as an analogue to the human arm movement. Going forward there may be room for a broadening of accomplishable movements.
