ExoArm (George Mason University)

ExoArm Device

Scott Oshiro, Rahul Chopra, Luis Asencios Reynoso

The purpose of this project was to develop an active arm brace, which is controlled by the user’s surface electromyography (EMG) signals. This can be used for physical therapy and to assist the user with routine activities. In order to process the EMG signals and eliminate noise, a bandpass filter is used to allow frequencies between 50Hz to 500Hz while rejecting noise. These frequencies are observed in EMG signals produced by the muscles in the arm. The filtered signals then enter the rectification stage, and are finally fed into the microcontroller to generate a control signal for the servo motor. The servo motor controls the brace attached to the arm through a set of Bowden cables, which are connected to a mechanical joint on the user’s elbow. The Bowden cable provides torque to either lift or lower the user’s arm with an additional weight up to 10 lbs. The user will also have the option to manually control the arm through a joystick device. In addition, an OLED display interface will display temperature, battery life and warning messages to alert the user of any system problems. On the joystick device, a button stops the servo motor from running to prevent and avoid hazardous outcomes. The majority of the device’s weight, including the servo motor and circuitry, is located inside a backpack which will be carried by the user. Ultimately, the user should feel as if they are in control of their arm without too much effort while providing smooth movements depending on the direction that is desired.

Electromyography (EMG)
Electromyography is the study of electrical muscle signals. An EMG signal is the summation motor unit action potential (MUAP). It can be modeled by:

where x(n) represents the EMG signal, e(n) represents the processed point, h(r) represents the MUAP, and w(n) represents the white gaussian noise. There are two main issues when trying to detect these signals are the signal-to-noise-ratio, distortion and small amplitude (10mV). In order to compensate the signal needs to be amplified and filtered out where frequencies between 20Hz and 500Hz are passed. There are two types of electrodes that are used for detecting these signals. Invasive and non-invasive electrodes are used to in EMG signal detection. For this project we will be using non-invasive electrodes which lie on top of the surface of the skin rather than invasive electrodes where a needle is placed inside the muscle (Raez, Hussain, Yasin 2006).

To approach the issue of what motors should be used the following formula for torque is used:

Where τ is torque, r is the distance from which torque is measured, and F is the force applied. Torque is a measure of how much force causes an object to rotate relative to a point on the object (Hugh & Roger, 2012). The general torque used for the design will be based off an arm of 0.5m, 3.628kg of arm weight, which will produce a torque for an arm by itself of 17.78Nm.

Servo Motor
A servo motor is a rotary actuator that allows for precise control of angular position, velocity and acceleration. The use of servo motors is essential to our project due to the fact that we need precise control of the movement of the elbow and shoulder joints. Servo motors present a definite advantage over other actuators because they possess an embedded feedback system. Most motors are controlled using Pulse width modulation (PWM) signal provided by a microcontroller.

Figure 1: Duty Cycle of a Servo Motor


Problem Statement
People who suffer from diminished strength capacity have trouble lifting heavy objects. This could be caused by muscle injury, aging, muscle degeneration, and/or weak muscles. They require additional aid, training/rehabilitation to be able to lift objects of certain weights. In some cases, these conditions can be hereditary and chemical treatments can be applied. In other cases, physical therapy and other forms of physical rehabilitation can be applied. For cases like these, a method of controlled movement is needed to enhance strength.

The ExoArm consists of multiple different components which make up the entire system. The components that make up the ExoArm are the exoskeleton arm brace that the user wears, surface electromyography (EMG) electrodes, joystick device, power management system, user interface display on an OLED, 31.38Nm servo motor, and a backpack. The EMG electrodes and joystick device are used to control the servos that move the exoskeleton brace at its joint. The servo motors, power management system, and EMG circuit are all located in the backpack. The servo motors control the exoskeleton brace at the elbow joint through two Bowden cables. When moving either up or down, a Bowden cable rotates the joint to move the exoskeleton brace.

The reason why ExoArm was primarily designed and created was to help users who have difficulties supporting weights in routine activities. Also, this can be used for people who need physical therapy in cases of injury and muscle disorders and diseases.

The ExoArm aims to address these problems by allowing the user to automate their movement through their EMG signals. Thus, this would allow users to feel as if they are in control of their arm without too much effort. Alternatively, the user would have to option to manually control their arm through the manual (joystick) mode for people who require therapy or for rehabilitation purposes. The goal for the user who has therapeutic problems can initially use the manual mode, which will eventually lead up to using the ExoArm with only the automated mode.

Signal Processing Unit
Being able to analyze and classify the raw EMG signals would be almost impossible without a signal processing unit. Raw EMG signals are approximately 10mV and have a significant signal to noise ratio. As a result, we must amplify and filter the signal to reasonable amplitudes between 20Hz and 500Hz. A signal processing unit must be designed to meet these parameters.

Figure 2a: EMG Circuit Design

Figure 2b: EMG Processing Unit (PCB)


Figure 3: Diagram of Exoskeleton Brace with Bowden Cables

The servo implemented in the design requires a voltage between 12V to 24V. Due to the fact that we needed its maximum torque and speed we chose to input 24 volts to the servo. The maximum torque the servo can provide is 31.38Nm. In other words, it can lift approximately 27 pounds.

Figure 4: The Exoskeleton Arm Brace

We used multiple heavy duty zip ties in order to secure the Bowden cables to the brace structure. The zip ties proved to be strong with a tensile strength of 75 pounds.

Figure 5: Exoskeleton Arm Brace on User

We used 1.5mm steel cable that runs inside the Bowden cable housing in order to move the brace joint which was cut to size in order to fit length between the servo pulley and the brace joint.

Power Design
We purchased a 14.8V – 5AH lithium polymer battery and charger to use with the 320Kg*cm (31.38Nm) servo. In order to produce the maximum torque allowed on the servo, we purchased a 600W boost converter. This converter allows us to convert the 14.8V to 24V. In order to provide a higher voltage, the input current must be increased. The maximum current that the battery will use at any given point from the servo is 11.027A. Although, the power usage is much higher than the previous servo, the amount of weight it can handle is much higher. Therefore, at 11.027A, the battery should last for a minimum of 27 minutes at full power constantly. With no load, the power usage is draining a constant 0.81A, the battery should last for 370 minutes (6.17 hours). A relay was also used in the ExoArm to switch off the power supply going into the high power servo motor.

System Integration

This video shows the two different modes used in the ExoArm device. The first mode as shown in the video, is the automated (EMG) mode. In this mode, when the user flexes their bicep or tricep muscles, the servo motor moves in the appropriate direction. The second mode (which is changed by pressing the “Z” button on the joystick device) shown in the video is the manual (joystick) mode. When the user moves the joystick up, the arm moves up, and vice versa.

For the system integration, all of the internal components were placed on a wooden enclosure, which were screwed down to the board to prevent movement. The internal components include the 14.8V battery, 5V voltage regulator, Arduino ATMega2560, 2 relays, 600W boost converter, and the 320Kg*cm (31.38Nm) servo motor.

Figure 6: Internal Components in the Wooden Enclosure

Figure 7: Servo Motor with Bowden Cable connecting to the Exoskeleton Brace

Ideally, the user would place all of the internal components into a backpack to move while wearing the device.

ExoArm device with internal components placed in the bookbag.

Figure 8: ExoArm Device with Internal Components in Bookbag


Experiment 1: Acquire the EMG Signal
A serial data connection was established between MATLAB and the ATMega2560 in order to record data into graphs. The data recorded into MATLAB was real-time, meaning, the graphs were recording the data points immediately after the sample was taken. The method used to determine which direction the servo rotates was determined by a threshold method.

This method looks at what point on the graph the difference voltage (bicep – tricep) is and sends a signal to the servo indicating the direction of movement.
The graphs shown below, show the bicep, tricep, and the difference between the bicep and the tricep.

Figure 9: EMG Signals

From looking at the “Bicep – Tricep” graph, when the user is flexing his or her arm, the voltages are more positive. However, when the user is straightening his or her arm, the voltages are generally more negative. At rest, the user has a voltage difference closer to zero. Since, there is a ripple voltage, there are some small inconsistencies with how the servo moves.

Figure 10: Original Difference EMG Signal vs. Moving Average Signal

Therefore, from Figure(9), the moving average was taken to get more accurate results. The servo moving graphs were created to indicate what the direction the servo is currently moving. The servo moving graphs are shown below.

Figure 11: Servo Moving Graphs

+1V is when the arm is being flexed, -1V is when the arm is being straightened, and 0V is when the arm is at rest. Comparing the servo moving (original) graph to the servo moving (moving average) graph, the data accurately reflected when the arm was being flexed or straightened through the moving average.

Strength Tests

We ran strength tests on the system. It seems after a certain weight the maximum angle that the brace can reach decreases. The graph below illustrates that the mechanical part of the system begins to degrade by 15 degrees when more than 1Kg is applied to the brace. Also after two hours of testing the system began to degrade and the brace’s max angle goes from 90 degrees to 75 degrees. The system begins to degrade due to mechanical issues. Thus the mechanical part of the system needs to be maintained by tightening the wires on the pulley system.

Figure 12: Strength Test Results

The system is required to lift at least 5 Kg. According to the strength test we performed, the system is able to lift up to 1 Kg to the maximum angle (90 degrees). Anything heavier can be lifted to 75 degrees.
The automatic mode was proven to work 100 percent. The experiments we performed in MATLAB (for data acquisition) showed that the servo responds appropriately to the muscle signals of the user, which is indicated by the servo moving graphs in Figure(10). When the user’s EMG signal is increasing, the software correctly recognizes that the servo should be moving so that the brace is pulled up. When the EMG signal is decreasing, the software correctly recognizes that the servo should be moving so that the brace is pulled in a downward direction. This test was done on two different users and the servo correctly recognized the movement of the arm and moved accordingly despite the differences in the muscle signals of the users.

For power management, the desired goal is that the system should be operational for at least 1 hour. A lithium polymer battery was implemented into the system. It provides 5 amp/hr and 14.8V. A test was conducted to measure the current being provided to the system during different instances of use. When the system is connected to the battery while it is running idle, there is approximately 300 mA of current being consumed. Also, when the servo is moving without any load we found that the current being consumed was approximately 300 mA. If the device were to run continuously without any activity, it would run for approximately 8 hours. When a weight of 3.36 Kg was applied to the system (while the user was wearing it) the current consumed raised to about 4 A. If the user were continuously lifting 3.36 Kg, the system would last approximately 1.25 hours. As a result, it can be concluded that for anything weighing 3.36 Kg, if it were to run continuously would not meet the requirement of operation for one hour. However, this is an extreme case. Normally, the user would not be lifting weights the whole time while wearing the device. The operation time of one hour can be met by the system if it is not in constant use for lifting an object weighing more than 3.36 Kg.

Total Funds Spent

Figure 13: Total Funds Spent



Dempster, W. T., & Gaughran, G. R. (1967). Properties Of Body Segments Based On Size And Weight. American Journal of Anatomy, 120(1), 33-54.

Hugh, Y., & Roger, F. (2012). University physics with modern physics. (13th ed., p. 310). San Francisco, CA: Addison-Wesley.

Hugh, Y., & Roger, F. (2012). University physics with modern physics. (13th ed., p. 359). San Francisco, CA: Addison-Wesley.

NG, H.K, The Chinese University of Hong Kong EE1130 Analog Circuits. Retrieved from www.ee.cityu.edu.hk/…08/L05%20Active%20Filter%20(Part%20II)

Papadopoulos, Evangelos , and Georgios  Patsianis. “Design of an Exoskeleton Mechanism for the Shoulder Joint.” IFtoMM World Congress 12 (2007): 1-6. Print.

Paolo de Leva (1996) Adjustments to Zatsiorsky-Seluyanov’s Segment Inertia Parameters. Journal of Biomechanics 29 (9), pp. 1223-1230.

Raez, M.B.I., & Hussain, M.S., & Mohd-Yasin, F. (2006). Techniques of EMG Signal Analysis: Detection, Processing, Classification, and Application. Biol Proced Online., 8: 11–35. Retrieved from: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1455479/

Schiele, Andre, and Pierre Letier. “Bowden Cable Actuator for Force-Feedback Exoskeletons.”International Conference on Inteligent Robots and Systems na (2006): 2-6. IEEE Xplore. Web. 12 Oct. 2013.

Shuang Wang; Jiting Li; Ruoyin Zheng, “Active and Passive control Algorithm for an Exoskeleton with Bowden cable Transmission for Hand Rehabilitation,” Robotics and Biomimetics (ROBIO), 2010 IEEE International Conference on Robotics and Biometrics, vol. no., pp.75,79, 14-18 Dec. 2010

Powered by WordPress. Designed by WooThemes

Skip to toolbar