Functional validation and improvement of a low-cost control system for assistive robotic devices

The goal of this research is to increase the affordability of portable and intuitive control systems for assistive robotic devices used by patients and researchers. Current control systems are either expensive or utilize unintuitive inputs such as foot-mounted inertia measurement units (for prostheses) or oral sipand- puff devices (for wheelchairs). These unintuitive control strategies place unnecessary physical and mental strain on users and lead to the abandonment of assistive devices. Here we present a low-cost intuitive control system based on surface electromyography (sEMG). The low-cost control system consists of an sEMG acquisition system and a single board computer. The high-performance single-board computer enables real-time machine learning to determine an individual's intended movements from sEMG. We integrated the low-cost control system with a commercially available prosthesis (i.e., DEKA "LUKE" Arm) and then validated the performance of the low-cost control system against a high-end research-grade control system (Ripple Nomad). We found no significant differences in functional hand dexterity between individuals using the low-cost and researchgrade control systems. The low-cost control system presented here can provide researchers and patients an inexpensive and portable option for more intuitive realtime control of robotic devices.

ABSTRACT The goal of this research is to increase the affordability of portable and intuitive control systems for assistive robotic devices used by patients and researchers. Current control systems are either expensive or utilize unintuitive inputs such as foot-mounted inertia measurement units (for prostheses) or oral sipand-puff devices (for wheelchairs). These unintuitive control strategies place unnecessary physical and mental strain on users and lead to the abandonment of assistive devices. Here we present a low-cost intuitive control system based on surface electromyography (sEMG). The low-cost control system consists of an sEMG acquisition system and a single board computer. The high-performance single-board computer enables real-time machine learning to determine an individual’s intended movements from sEMG. We integrated the low-cost control system with a commercially available prosthesis (i.e., DEKA “LUKE” Arm) and then validated the performance of the low-cost control system against a high-end research-grade control system (Ripple Nomad). We found no significant differences in functional hand dexterity between individuals using the low-cost and researchgrade control systems. The low-cost control system presented here can provide researchers and patients an inexpensive and portable option for more intuitive realtime control of robotic devices. ii TABLE OF CONTENTS ABSTRACT ii INTRODUCTION 1 I. PROTOTYPE LOW-COST CONTROL SYSTEM 2 II. OBJECTIVE 5 METHODS 7 I. VALIDATION 7 II. IMPROVEMENTS 15 RESULTS 19 DISCUSSION 22 ACKNOWLEDGEMENTS 24 REFERENCES 25 iii 1 INTRODUCTION More than 1.6 million individuals in the United States suffer from limb-loss [1], which often results in chronic pain, depression and functional disability [1], [2]. In addition, the high cost of upper-limb prostheses places financial strain on the limb-loss community [3]. Up to 50% of upper-limb amputees abandon or limit prosthesis use [4] due to ineffective control and high cost (e.g., of repairs) [5]. Existing prostheses fail to combine a cost effective price with the dexterous and intuitive control that makes prosthetics highly functional and simple to use. Existing lowcost systems consist of basic hooks often controlled by movement of the upper body (e.g. close the hook by moving your shoulder) [6].These are neither dexterous or natural to control. More advanced clinical systems include the i-Limb Quantum, TASKA hand, and DEKA “LUKE” Arm [7]–[9]. While they provide mobility of the wrist and digits, they often lack intuitive control methods. The i-Limb Quantum, for example, utilizes mobileapplication control [8] which restricts control to preprogrammed grasps and requires the use of an additional hand to input commands on a mobile device. The DEKA “LUKE” Arm often employs inertial measurement units located on the feet [7] which also limits control to preprogrammed grasps and forces the user to be stationary while using the prosthesis. Recent advancements in myoelectric control algorithms leverage residual biopotentials from the body to provide intuitive and dexterous control of prostheses [10]. This more advanced control strategy often employs high-density surface electromyography (HD-sEMG) [11] and utilizes more computationally expensive algorithms such as Kalman filters [10], [12], and neural networks [13]. Traditionally, 2 these algorithms are deployed on desktop computers or portable research-grade systems that can cost up to $64,000. Attempts at replicating this at a low cost have recently been developed, often using the Myo Armband [14] or the OpenBCI Cyton [15]. However, low-cost control systems so far have only utilized a single channel of sEMG [16], controlled only a single degree of freedom (DOF) [17], or provided only binary control [18]. A further motivation for developing low-cost intuitive control systems based on sEMG is that they can also aid individuals affected by severe accidents such as spinal cord injuries. Spinal cord injuries that cause extensive paralysis result in higher levels of depression, a struggle to retain independence, and a reliance on care-givers [19], [20]. For patients suffering from paralysis and other disabilities, it has been found that when participating in adaptive sports they experience increased mental, physical, and social wellness [21]. However, those with disabilities requiring the use of mobility equipment like wheelchairs often lack adequate control options for these devices. Of those using current control methods including joysticks, sip-and-puff, or chin control interfaces, around half have difficulty steering and between five and nine percent found such tasks impossible without assistance [22]. The availability of a low-cost sEMG control will provide a proportional and simpler control interface to enable greater independence and control of adaptive sports equipment. I. PROTOTYPE LOW-COST CONTROL SYSTEM To overcome the challenges of expensive sEMG control systems, the Center for Neural Interfaces at the University of Utah recently developed a prototype low-cost 3 control system under $675 [23]. The system works by first acquiring sEMG data via surface electrodes attached to residual limbs for amputees. This sEMG data is amplified with hardware circuits, sampled using a microcontroller, and then sent to a computer for processing and control. The prototype low-cost control system uses a portable singleboard computer (SBC) to filter the incoming sEMG and run a Kalman filter control algorithm that correlates sEMG to motor control of a 3D-printed prosthetic hand. The result is simultaneous, proportional, and intuitive control of the prosthetic hand through a portable and inexpensive system [23]. The prototype low-cost control system consists of three main hardware components (Figure 1) used for sensing, data acquisition, data processing and control, and actuation: • Sensing sEMG data used sticky electrodes connected to a SpikerShield Pro for amplification and filtering of EMG data. • An Arduino Mega (attached to the SpikerShield Pro) acquired the sEMG data samples and packaged them for serial transmission over USB to the SBC. • A Raspberry Pi 4 4gb SBC dealt with data processing, control, and actuation with software to run the complex control algorithm and command the attached protheses. 4 Figure 1: An overview of the prototype low-cost control system components and specifications. 5 The prototype low-cost control system had some deficiencies that inhibited its ability to be properly validated against the lab’s research-grade control system (Ripple Nomad). Primarily, the connected 3D-printed hand was not robust, resulting in servomotor gearbox failures, breaking of cable ties that actuate joints, and overall little strength and mechanical robustness compared to commercially available prosthetics [24]. Other hardware limitations of the prototype low-cost control system, such as bulky wired connections, few sEMG inputs, and limited processing speed, restricted the system’s ability to accurately control more than two DOF and to directly compare its performance with current research-grade control systems. In addition, the prototype low-cost control system did not have the capability to interface with other devices beyond the original 3Dprinted hand and recent partnership with the Technology Recreation Access Independence Lifestyle Sports (TRAILS) program at the Neilson Rehabilitation Hospital [25] raised interest in using the low-cost control system with adaptive sports equipment, giving participants more proportional control over current sip-and-puff control methods while still maintaining low cost. These developments motivated the validation and further improvement of the low-cost control system. II. OBJECTIVE Here, we first describe software modifications to the prototype low-cost control system to readily enable communication with various external devices, in particular the DEKA “LUKE” Arm. Then, we provide the first functional validation of the prototype low-cost control system in tasks of daily living by conducting a direct comparison to a research-grade control system. Lastly, we introduce a second-generation version of the 6 low-cost control system (upgraded low-cost control system) that provides major improvements to portability, versatility, and cost. Altogether this work constitutes a major step towards more intuitive, dexterous, and affordable control of assistive robotic devices for both patients and researchers alike. 7 METHODS Software and hardware innovations were required to validate the prototype lowcost control system and improve it through subsequent upgrades. These innovations allowed us to conduct functional comparisons between the low-cost control system and a common research-grade control system and improve upon the low-cost system’s initial design. I. VALIDATION The prototype low-cost control system could only control a 3D-printed hand that severely lacked in robustness. Thus, the primary goal of this research project was to integrate the low-cost control system with the highly advanced DEKA “LUKE” Arm and therefore eliminate variability previously observed with the lower quality 3D-printed prosthetic hand. This enabled a direct functional comparison of the prototype low-cost control system and the research-grade control system with fewer confounds. Along with hardware integration of the DEKA “LUKE” Arm, extensive software development in both C++ and Python was needed to provide communication, testing, and control of the DEKA “LUKE” Arm drivers. Once integrated, we conducted trials using the modified box and blocks test [26] to compare the capabilities of the systems. The first step was to identify the software solutions necessary to connect the DEKA “LUKE” Arm to the prototypes low-cost control system’s Kalman filter control algorithm. Device drivers for CAN communication with the DEKA “LUKE” Arm existed separately on the lab’s desktop control system in C++. We determined that modifications to this existing code with the addition of software to facilitate the communication of 8 prosthesis control commands between the Kalman filter control algorithm and DEKA “LUKE” Arm driver was the best solution. A breakdown of the main software that was modified, developed, and installed on the low-cost control system is visualized in figure 2. This would allow for control of the DEKA “LUKE” Arm as needed to conduct functional tests against the lab’s research-grade system and avoid the time-consuming alternative of recoding the DEKA “LUKE” Arm device drivers from scratch. Figure 2: An overview of key software for DEKA “LUKE” Arm integration with prototype low-cost control system. 9 Due to incompatibilities of the DEKA “LUKE” Arm’s preexisting device drivers and wired interface, modification of the driver and installation of additional proprietary libraries were conducted. Originally developed for use on computers running the Windows operating system, macros, keywords, and Window specific function calls had to be replaced with their Linux counterparts. General sanitization of the code was also performed to ensure it met modern C++ standards; in particular, include statements, syntax, and type casting were updated. The DEKA “LUKE” Arm communicates over wired CAN bus. A PEAK Systems CAN-USB cable was used to connect the arm to the low-cost control system’s Raspberry Pi as the single board computer lacked a dedicated CAN port. To make use of this cable, PEAK System’s Linux CAN driver and PCANBasics API library had to be installed and linked to the modified DEKA “LUKE” Arm driver (low_cost_lkdriver.cpp). At this point, a C++ program could control the DEKA “LUKE” Arm by using the API provided by the driver. To enable the low-cost control system’s Python-based control algorithm (lktraining.py or lktesting.py) to access the API of the DEKA “LUKE” Arm driver, C/C++ extensions were implemented to build a Python module. The Python distribution (CPython) we used supports the creation of wrapper modules which allows existing C/C++ code to expose any wrapped methods to Python programs. The DEKA “LUKE” Arm driver contained methods to start communication, stop communication, get arm sensor data, send arm commands, and check communication status. We wanted to give our Python control algorithm the ability to interact with all of these and therefore needed to wrap these API methods so they could be exported to a Python module. To wrap the 10 C++ methods required the use of C++’s standard Python.h header file that encapsulates the methods in PyObjects and provides functions to interact with Python types (Figure 3). Method declarations and module definitions are also passed to Python using Python.h functions. A Python script (setup.py) was developed that employs CPython’s setup and Extension modules to install the wrapped C++ code into the Python environment. This produces a Linux library (.so file) that when imported into a Python file lets it utilize the API of the wrapped DEKA “LUKE” Arm device driver. Figure 3: Code snippet showing wrapped DEKA “LUKE” Arm API method. An additional piece of software was developed to perform error handling, command range conversions, and proportion control of wrist degrees of freedom (DOF) for DEKA “LUKE” Arm control. The low_cost_lkdriver.py provided methods used to 11 call the API methods of the wrapped DEKA “LUKE” Arm driver with added exception handling and descriptive return values to provide more detailed information to callers. Methods for running demos of the DEKA “LUKE” Arm and testing were also created to allow use without sEMG. These demonstration programs included: • The “LUKE” demo in which the arm moves in response to touching of its integrated sensors. • Standard demo in which the arm moves through its entire range of motion. • Terminal-based user interface that allowed isolated testing of all the API methods. Values the DEKA “LUKE” Arm motors accepted consisted of a number -1024 to 1024 (various DOF were restricted further within this range due to physical limitations of DEKA “LUKE” Arm movement). To offer more simplistic command values, range conversion for all six DOF was implemented to translate simple prothesis commands consisting of six double values from -1 to 1 into the appropriate range values for each individual motor. Finally, with standard control of wrist rotation and pitch done with the control value of -1 to 1 representing a direction and speed, an option was given to choose between directional or proportional control of the wrist’s two DOF. This gave a consistent control scheme for the arm (other DOF were controlled by position) by making the control value -1 to 1 represent the position of the wrist. This was accomplished by translating a single command requesting movement of the wrist into two separate commands to start and then stop the wrist movement after a given timer interval. With this code in place, most of the software needed to run the DEKA “LUKE” Arm from the 12 low-cost control system was done and only minor modification to the control algorithm and serial communication of EMG data was left. The software on the Arduino Mega used for data acquisition (arduino_read_emg.ino) and the Kalman filter used as the systems control algorithm (lktraining.py and lktesting.py) required a handful of small alterations to enable control of the DEKA “LUKE” Arm. Code for controlling the previous 3D-printed hand had to be removed from the Arduino Mega’s code, leaving only the necessary code to read the sEMG data from the SpikerShield Pro and send it to the Raspberry Pi. In the Kalman filter’s training and testing code, calls to the DEKA “LUKE” Arm’s API functions were added to start and end communication with the prosthesis. Most importantly, the DOF that each of the six command values referenced had to be changed because of the difference between the DOF of the 3D-printed hand and the DEKA “LUKE” Arm. The 3D-printed hand had a single DOF for each digit and one for the wrist. The DEKA “LUKE” Arm had DOF for wrist rotation, wrist pitch, thumb flex, thumb abduction, index flex, and MRP (middle ring pinky) flex. Once modified, the Kalman filter could train itself by correlating preprogrammed DEKA “LUKE” Arm movements to collected EMG data and then test itself by using real-time EMG data to predict and command the DEKA “LUKE” Arm. To compare the prototype low-cost control system and the research-grade control system, we used a modified box and blocks test: a simple, low-cost, and efficient test of gross manual dexterity that has standardized instructions along with reliability and validity data reported in many studies [27]. In its most basic form, the test requires the movement of blocks from one bin to another over a barrier. This elicits repetitive cycles 13 of motion allowing for multiple recorded trials. In the modified box and blocks test, blocks are ordered in a grid and the subject was instructed to proceed in sequence thereby producing more consistent and predictable movements in each test (Figure 4) [26]. The number of blocks successfully transferred in a minute was recorded. A set of instructions was written to ensure setup and orientation of experimental participants was similar from one participant to another. The control system to be used first was randomized by coin flip and the type of control system was not disclosed to the participants. Participants were allowed up to two minutes of practice with each control system before recording trials. The research-grade control system was also limited to using six of its typical 32 sEMG channels to mirror the restrictions of the six-channel SpikerShield Pro of the prototype low-cost control system. Six sticky electrodes plus ground and reference were placed on the forearm muscles used for flexion of the index and thumb. The average number of blocks successfully transferred in one-minute provided a functional metric to compare the prototype low-cost system performance with respect to that of the research-grade control system. If all blocks were moved under one minute, then the amount was extrapolated out to predict how many would have been moved in a minute. 14 Figure 4: An intact participant using the DEKA “LUKE” Arm with the lab’s bypass socket to perform the modified box and block test. Initial pilot experiments were carried out to resolve any problems that occurred during system setup and box and blocks trials. A total of three pilot experiments resulted in two key changes. The first was that the number of trials for each control system was reduced from eight to five. This was deemed necessary to prevent significant fatigue affecting performance. Along with this, participants were allowed to rest between trials. Second, the number of DOF controlled had to be reduced from three to two. Originally, wrist pitch was used to help position the hand over a block. It was discovered that one of 15 the six sEMG channels of the SpikerShield Pro was broken and thus the DOF were reduced to two (thumb and index flex) which yielded better accuracy and proportional control. With these adjustments, volunteers were gathered for subject trials. Subject trials used for analysis were conducted with intact participants. The bypass socket, developed by the Center for Neural Interfaces [28], enabled the intact participants to use the prothesis in similar manner to an amputee. A total of 5 participants were brought in and their results were recorded. Participant demographic consisted of both males and females, right and left-hand dominant, and an age range of 20 to 22 years old. Along with performance in the modified box and blocks test, other observations by experimenters and participants were noted such as perceived prothesis control characteristics (e.g., smoothness, responsiveness). II. IMPROVEMENTS With validation of the prototype low-cost control system complete, upgrades to the system were identified to expand the portability, versatility, and capability. Optimization of case design and the addition of a NeuroRobotics Lab developed wireless sEMG acquisition device aimed at reducing weight, size, and wire clutter were made (Figure 5). Partnership with the Neilson Rehabilitation TRAILS program presented the unique opportunity to integrate the low-cost control system with assistive robotic devices other than prostheses such as adaptive sports equipment. Evolution of other technologies like batteries and single-board computers presented the ability to increase overall system capability while maintaining or decreasing total cost. All original hardware components were replaced, and software revisions were made to obtain these goals. 16 Figure 5: An overview of the upgraded low-cost control system components and specifications. 17 For sEMG acquisition, the Arduino Mega and SpikerShield Pro were exchanged for an Arduino MKR Wifi and TI ADS. To utilize the wireless capabilities of this new data acquisition system, it’s software for packaging and sending data was changed to use Wi-Fi rather than wired serial. To do this, code was added to set up the Arduino MKR Wifi as a wireless access point. This would allow the main single-board computer to connect and stream data from it. A way to send the sEMG data over wireless was needed after a communication channel was established. UDP was determined to be the best protocol to use since fast sEMG data transmission is necessary to provide real-time control. Despite UDP providing simple, fast data communication over Wi-Fi, transmitting one sample of EMG data in each UDP packet was still too slow. This was solved by sending multiple sEMG samples in a single UDP packet and separating each sample by a unique separator. Since sEMG samples were averaged during the data processing phase, dropping singular UDP packets was not a concern and testing proved this to be unlikely. On the receiving end, the single-board computer for data processing and control was switched from the Raspberry Pi 4 to the Nvidia Jetson Nano 2gb. With the Nvidia Jetson Nano running on a Linux OS, much of the software was directly deployable on the new computer. A simple wireless configuration ensured that on boot the Jetson Nano automatically connected to the Arduino MKR Wifi’s wireless access point. The serial reading in the Kalman filter training and testing code was updated to use a UDP socket and sample parsing was altered to split individual sEMG samples out from each UDP packet. The Nvidia Jetson Nano also brought better RAM modules and an onboard Nvidia GPU, enabling exploration of control algorithms other than the Kalman filter. The 18 TI ADS also provided an extra two sEMG channels; thus, the Kalman filter’s input matrix had to be expanded to take in eight inputs. A separate testing and training program for the Kalman filter was developed that output two values instead of six to test the upgraded low-cost control system’s ability to control the turn and wedge angles on the TRAILS Tetraski. Among the final upgrades were the reduction of the low-cost control systems battery capacity and new compact case design. Both were a direct result of the switch to the Arduino MKR Wifi and TI ADS for data acquisition as these no longer need to be mounted with the main single board computer. Rather the Arduino MKR Wifi and TI ADS could easily store on the arm next to the surface electrodes and use a small external battery. 19 RESULTS CAN communication with the DEKA “LUKE” Arm and wrapping of its driver resulted in complete integration of all the DEKA “LUKE” Arm control and data. Figure 6 breaks down the functionality gained from this. lk_start Start CAN channel communication with the DEKA “LUKE” Arm. lk_stop Stop CAN channel communication with the DEKA “LUKE” Arm. lk_running Get status of CAN channel communication with the DEKA “LUKE” Arm. lk_get_sensor Get values of the DEKA “LUKE” Arms 22 embedded sensors. lk_set_command Send values to command the 6 DOF of the DEKA “LUKE” Arm. Figure 6: Table of resulting functionality provided by DEKA “LUKE” Arm integration. The five modified box and blocks experiments with intact-arm participants using five-channel sEMG resulted in 14.88±4.55 and 11.40±5.61 blocks per minute with the research-grade and prototype low-cost control systems, respectively (Figure 7). These results revealed that the difference between functional performance of the research-grade control system and the prototype low-cost control system was not statistically significant (p = 0.524, α = 0.05). Multiple subjects did note that they felt prothesis movement was not as smooth when using the prototype low-cost control system. 20 Figure 7: Average number of blocks moved during the modified box and blocks trials. Upgrades to the hardware of the prototype low-cost control system provided significant improvement to the cost, portability, and capabilities of the low-cost control system. The tables below in figure 8 and 9 break down the key improvements gained by these changes. Specifications of the research-grade control system are included for reference, but it is important to note the additional functionality the research-grade control system has that the upgraded low-cost control system is not currently capable of. 22 DISCUSSION Here we introduce a low-cost control system to increase the affordability of portable and intuitive control systems for assistive robotic devices used by patients and researchers. Here we have shown that at a fraction of the cost, regularly available consumer hardware can be used to construct an sEMG-based control system with select capabilities of research-grade control systems such as the Ripple Nomad. Functional comparison against the research-grade control system displays promising results when considering the low-cost control system’s performance in proportional, real-time control of protheses with multiple DOF. Improvements via both software and hardware demonstrate equally exciting advancements for the system’s portability and capability. This presents an encouraging route to increase the use of assistive devices by tackling leading issues of device abandonment such as cost and intuitiveness. The low-cost control system is inexpensive and a versatile platform, which lowers the barrier for patients looking for intuitive control of robotic devices. Readily available low-cost commercial parts also decrease the cost of repairs that prevents those who currently own intuitive control systems from using them in their daily lives. The sEMG based control approach and ability to proportionally command multiple DOF provides dexterous control rarely seen in other low-cost control schemes. Future efforts to expand the low-cost control systems applications and functionally validate the upgraded low-cost control system are necessary to fully prove the system’s versatility and performance. While low-cost control system improvements included modification of the Kalman filter to produce output able to control simple 23 adaptive sports equipment, software to package that command data into the format laid out by Tetraski documentation will have to be developed before testing can begin. Certain hardware improvements such as the Nvidia Jetson Nano’s GPU, which could be utilized to allow onboard training and testing of neural network based control algorithms, have yet to be realized. Further software solutions to improve functional performance include implementation of a sEMG high pass filter to increase quality of sEMG input data. Integration with the NeuroRobotics Lab’s recently developed inexpensive stimulation system would also enable the low-cost control system to provide vibrotactile feedback. Nevertheless, this research succeeds in providing a low-cost and intuitive control system for assistive robotic devices. Functional validation has shown it is comparable to state-of-the-art research-grade control systems for simple prosthesis control. With current and future additions to the system, hopefully other researchers and patients can continue to explore the benefits of low-cost EMG control. 24 ACKNOWLEDGEMENTS First and foremost, I would like to thank all the members of the Center for Neural Interfaces and Neurorobotics Lab for their support throughout my research. In addition, I would like to thank the Office of Undergraduate Research for providing funding to help accomplish this project. Thank you to Dr. Jacob George and Dr. Alan Kuntz for their guidance and mentorship. Thank you to Sridharan Radhakrishnan for his pioneering research on developing the prototype low-cost control system. I also thank Taylor Hansen and Jared Zollinger for their help conducting subject trials and aiding in hardware upgrades for the low-cost control system. 25 REFERENCES [1] K. Ziegler-Graham, E. J. MacKenzie, P. L. Ephraim, T. G. Travison, and R. Brookmeyer, “Estimating the Prevalence of Limb Loss in the United States: 2005 to 2050,” Arch. Phys. Med. Rehabil., vol. 89, no. 3, pp. 422–429, Mar. 2008, doi: 10.1016/j.apmr.2007.11.005. [2] C. G. Bhuvaneswar, L. A. Epstein, and T. A. Stern, “Reactions to Amputation: Recognition and Treatment,” Prim. Care Companion J. Clin. Psychiatry, vol. 9, no. 4, pp. 303–308, 2007. [3] D. K. Blough, S. Hubbard, L. V. McFarland, D. G. . Smith, J. M. Gambel, and G. E. Reiber, “Prosthetic cost projections for servicemembers with major limb loss from Vietnam and OIF/OEF,” J. Rehabil. Res. Dev., vol. 47, no. 4, p. 387, 2010, doi: 10.1682/JRRD.2009.04.0037. [4] E. A. Biddiss and T. T. Chau, “Upper limb prosthesis use and abandonment: a survey of the last 25 years,” Prosthet. Orthot. Int., vol. 31, no. 3, pp. 236–257, Sep. 2007, doi: 10.1080/03093640600994581. [5] E. Biddiss and T. Chau, “Upper-limb prosthetics: critical factors in device abandonment,” Am. J. Phys. Med. Rehabil., vol. 86, no. 12, pp. 977–987, Dec. 2007, doi: 10.1097/PHM.0b013e3181587f6c. [6] A. Manero et al., “Implementation of 3D Printing Technology in the Field of Prosthetics: Past, Present, and Future,” Int. J. Environ. Res. Public. Health, vol. 16, no. 9, p. 1641, May 2019, doi: 10.3390/ijerph16091641. [7] “LUKE Arm Detail Page – Mobius Bionics.” https://www.mobiusbionics.com/lukearm/ (accessed Oct. 26, 2021). [8] “i-Limb® Quantum Bionic Hand. Ossur.com.” https://www.ossur.com/enca/prosthetics/arms/i-limb-quantum (accessed Oct. 26, 2021). [9] “TASKA Hand | TASKA Prosthetics.” https://www.taskaprosthetics.com/products/taska-hand (accessed Oct. 26, 2021). [10] J. A. George, T. S. Davis, M. R. Brinton, and G. A. Clark, “Intuitive neuromyoelectric control of a dexterous bionic arm using a modified Kalman filter,” J. Neurosci. Methods, vol. 330, p. 108462, Jan. 2020, doi: 10.1016/j.jneumeth.2019.108462. [11] A. Matran-Fernandez, I. J. Rodríguez Martínez, R. Poli, C. Cipriani, and L. Citi, “SEEDS, simultaneous recordings of high-density EMG and finger joint angles during multiple hand movements,” Sci. Data, vol. 6, no. 1, p. 186, Sep. 2019, doi: 10.1038/s41597-019-0200-9. [12] J. Nieveen et al., “Polynomial Kalman filter for myoelectric prosthetics using efficient kernel ridge regression,” in 2017 8th International IEEE/EMBS Conference on Neural Engineering (NER), May 2017, pp. 432–435. doi: 10.1109/NER.2017.8008382. [13] J. A. George, M. R. Brinton, C. C. Duncan, D. T. Hutchinson, and G. A. Clark, “Improved Training Paradigms and Motor-decode Algorithms: Results from Intact Individuals and a Recent Transradial Amputee with Prior Complex Regional Pain Syndrome,” Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. IEEE Eng. Med. Biol. Soc. 26 [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] Annu. Int. Conf., vol. 2018, pp. 3782–3787, Jul. 2018, doi: 10.1109/EMBC.2018.8513342. S. Pizzolato, L. Tagliapietra, M. Cognolato, M. Reggiani, H. Müller, and M. Atzori, “Comparison of six electromyography acquisition setups on hand movement classification tasks,” PLOS ONE, vol. 12, no. 10, p. e0186132, Oct. 2017, doi: 10.1371/journal.pone.0186132. “Cyton Biosensing Board (8-channels),” OpenBCI Online Store. https://shop.openbci.com/products/cyton-biosensing-board-8-channel (accessed Oct. 25, 2021). I. Ku, G. K. Lee, C. Y. Park, J. Lee, and E. Jeong, “Clinical outcomes of a low-cost single-channel myoelectric-interface three-dimensional hand prosthesis,” Arch. Plast. Surg., vol. 46, no. 4, pp. 303–310, Jul. 2019, doi: 10.5999/aps.2018.01375. N. Sreenivasan, D. F. Ulloa Gutierrez, P. Bifulco, M. Cesarelli, U. Gunawardana, and G. D. Gargiulo, “Towards Ultra Low-Cost Myoactivated Prostheses,” BioMed Res. Int., vol. 2018, p. 9634184, 2018, doi: 10.1155/2018/9634184. M. M. Atique and S. Rabbani, “A Cost-Effective Myoelectric Prosthetic Hand,” JPO J. Prosthet. Orthot., vol. 30, no. 4, pp. 231–235, Oct. 2018, doi: 10.1097/JPO.0000000000000211. R. Williams and A. Murray, “Prevalence of depression after spinal cord injury: a meta-analysis,” Arch. Phys. Med. Rehabil., vol. 96, no. 1, pp. 133–140, Jan. 2015, doi: 10.1016/j.apmr.2014.08.016. A. Alizadeh, S. M. Dyck, and S. Karimi-Abdolrezaee, “Traumatic Spinal Cord Injury: An Overview of Pathophysiology, Models and Acute Injury Mechanisms,” Front. Neurol., vol. 10, p. 282, Mar. 2019, doi: 10.3389/fneur.2019.00282. V. Reljin, “Effects of Adaptive Sports on Quality of Life in Individuals with Disability,” Williams Honors Coll. Honors Res. Proj., Jan. 2019, [Online]. Available: https://ideaexchange.uakron.edu/honors_research_projects/822 L. Fehr, W. E. Langbein, and S. B. Skaar, “Adequacy of power wheelchair control interfaces for persons with severe disabilities: a clinical survey,” J. Rehabil. Res. Dev., vol. 37, no. 3, pp. 353–360, Jun. 2000. J. A. George, S. Radhakrishnan, M. Brinton, and G. A. Clark, “Inexpensive and Portable System for Dexterous High-Density Myoelectric Control of Multiarticulate Prostheses,” in 2020 IEEE International Conference on Systems, Man, and Cybernetics (SMC), Oct. 2020, pp. 3441–3446. doi: 10.1109/SMC42975.2020.9283086. M. Cognolato et al., “Multifunction control and evaluation of a 3D printed hand prosthesis with the Myo armband by hand amputees,” Oct. 2018. doi: 10.1101/445460. “TRAILS: Technology Recreation Access Independence Lifestyle Sports.” https://healthcare.utah.edu/neilsen-physical-rehab-hospital/supportservices/trails/index.php (accessed Oct. 26, 2021). J. S. Hebert and J. Lewicke, “Case report of modified Box and Blocks test with motion capture to measure prosthetic function,” J. Rehabil. Res. Dev., vol. 49, no. 8, p. 1163, 2012, doi: 10.1682/JRRD.2011.10.0207. 27 [27] V. Mathiowetz, G. Volland, N. Kashman, and K. Weber, “Adult Norms for the Box and Block Test of Manual Dexterity,” Am. J. Occup. Ther., vol. 39, no. 6, pp. 386– 391, Jun. 1985, doi: 10.5014/ajot.39.6.386. [28] M. D. Paskett et al., “A Modular Transradial Bypass Socket for Surface Myoelectric Prosthetic Control in Non-Amputees,” IEEE Trans. Neural Syst. Rehabil. Eng., vol. 27, no. 10, pp. 2070–2076, 2019, doi: 10.1109/TNSRE.2019.2941109. Name of Candidate: Aidan Lethaby Date of Submission: May 3, 2022