Northeastern University. Boston, Massachusetts. August, 2013

Transcription

1 Design and Control of a 2 Degree of Freedom Upper Limb Robotic Rehabilitation Device A Thesis Presented by Patrick Joseph Murphy to The Department of Mechanical and Industrial Engineering in partial fulfillment of the requirements for the degree of Master of Science in Mechanical Engineering Northeastern University Boston, Massachusetts August, 2013

2 Abstract Robotic neurorehabilitation is a rapidly growing field in both research and industry. Robotics offer the ability to create less labor-intensive rehabilitation for therapists, while providing an interactive experience for patients. Robotic therapy also provides the advantage of object data collection for therapists to track patient progress; however, there is still both a clinical and market need for a low-cost, assistive hand rehabilitation system. Therefore, the comprehensive design, control, and initial testing of an actuated, assistive 2 degree of freedom hand rehabilitation system with a virtual environment is presented. The 2 degree of freedom hand rehabilitation system, named the Navigator hand rehabilitation system, can provide assistive or resistive mode exercise for flexion and extension of the fingers, as well as pronation and supination of the wrist. The system incorporates a rack and pinion into a series elastic actuator to provide assistive exercise for flexion and extension of the fingers. A belt drive is used to provide actuation to pronation and supination of the wrist. The design, implementation of an impedance control system utilizing position and load feedback is also presented. Both automated control results and preliminary pilot data of resistive mode exercises are presented. A virtual environment to interact with the Navigator system was designed and implemented. The virtual environment incorporates both degrees of freedom, allowing the user to combine the motions and associate the movements with virtual tasks. The impedance controller and virtual environment interact via serial communication. The Navigator device was designed to provide a low-cost solution to providing assistive exercise in 2 DOF with an accompanying virtual environment for a home and clinical setting. 2

3 Table of Contents Acknowledgments Introduction Project Significance Key Terms Robotic Terms: Human Anatomical Terms: Assessment Method in rehabilitation Contributions Background Previous Work Devices on Market Research Devices Mechanical Design Specifications Priority Priority Priority Priority Priority Priority Priority Priority Handle Design Anthropometric Design Improvements to Handle Design Linear Actuation Motor Selection Series Elastic Actuation Sensor Integration Finite Element Analysis Rotational Actuation Motor Selection

4 3.4.2 Sensor Integration Finite Element Analysis System Control Introduction to Impedance Control Derivation of Impedance Control for Device Effect of Force Feedback Loop Implementation of Impedance Controller Arduino Motor amplifiers Sensors Implementation of Controller Controller Results Linear Actuator Results Rotational Actuator Results Virtual Reality Interface Virtual Reality in Rehabilitation Game Concept Development Virtual Environment Design Game Design Components Collision Detection Serial Communication Virtual Reality Result Conclusions and Future Work Conclusions Future Work Assistive Mode Control Implementation Customization - Handle and Impedance Control Virtual Environment Improvement Works Cited Appendix A: Impedance Control Arduino Software Appendix B: Serial Communication between Arduino and Unity

5 List of Figures Figure 1: Finger Motion ranging from fully extended (left) to fully flexed (right) [6] Figure 2: A fully pronated (left) and fully supinated (right) wrist Figure 3: Joints of fingers and numbering of fingers Figure 4: Three pinch types Figure 5: A patient using the Amadeo System [10] Figure 6: InMotion Arm Robot [11] Figure 7: InMotion Wrist Robot [11] Figure 8: CyberGrasp haptic interface for virtual environments [14] Figure 9: Kinetic Meastra Hand and Wrist Tabletop CPM Rehabilitation Device [15] Figure 10: ReJoyce Rehabilitation Robot [17] Figure 11: EMG-Driven Exoskeleton Hand Robotic Training Device [19] Figure 12: A hand rehabilitation support system with improvements based on clinical practices [20] Figure 13: Actuated Finger Exoskeleton for hand rehabilitation following stroke [21] Figure 14: HEXORR Device attached to a hand [22] Figure 15: Hand rehabilitation device using a wire-driven control mechanism and control glove [23] Figure 16: Means of extension and flexion in the device described in [23] Figure 17: The HEXOSYS portable hand rehabilitation device with two active DOF [24] Figure 18: Multi-finger twist motion robotic device with computer interface [25] Figure 19: ReHapticKnob. Left: the full system with all DOF. Right: both knobs which are controlled by a linear actuator [26] Figure 20: Robohab in pronation and supination position (left) and tension and flexion position (right) [28] Figure 21: The three systems of the MR-Tech and how they interact together [26] Figure 22: Two views of the NJIT-RAVR system [31] Figure 23: Pneumatic rehabilitation robot for PNF therapy [32] Figure 24: Whole Arm Exoskeleton Robot with 8 active and 2 passive DOF [33] Figure 25: Photo of MIT Manus Device [34] Figure 26: 1st Generation HERRI System [35] Figure 27: Cross section of linear and rotational actuators of HERRI System [35] Figure 28: Example of Maze game running with HERRI system [36] Figure 29: Graphical User Interface for practitioner [36] Figure 30: HERRI System fully implemented with both user and practitioner screens [36] Figure 31: Final Design of Navigator Device Figure 32: Index finger length [39] Figure 33: Overall Handle Design Figure 34: Anthropometric Data for Finger Length Figure 35: Elbow-Fingertip Length Figure 36: Elbow- Grip Length During Finger Flexion Figure 37: Index Finger Width Measurement Figure 38: 2 Point Pinch exercise

6 Figure 39: 3 Point Pinch Exercise Figure 40: Final Handle Design- Grasp and Release Configuration Figure 41: Linear Actuation System Figure 42: Linear Actuation System Sensor Layout Figure 43: Deformation of finger flexion force simulation Figure 44: Stress analysis of system at rack and pinion interface Figure 45: Stress analysis at tooth interface, point of max stress Figure 46: Point of minimum safety factor (tooth/rack interface) Figure 47: Deformation of finger extension force simulation Figure 48: Finger extension exercise stress analysis of system at rack and pinion interface Figure 49: Finger extension exercise stress analysis at tooth interface, point of max stress Figure 50: Finger extension exercise point of minimum safety factor (tooth/rack interface) Figure 51: Rotational Actuator Layout Figure 52: Free body diagram of pulley Figure 53: Deformation of rotational subassembly Figure 54: Stress analysis of rotational subassembly Figure 55: Safety factor of rotational assembly Figure 56: General Impedance Control Loop Figure 57: Dynamic Interaction of Forces on End Effector Figure 58: Arduino Mega Microcontroller Board [46] Figure 59: Example of Pulse Width Modulation at a)10% duty cycle, b) 50% duty cycle, and c) 90% duty cycle [48] Figure 60: Junus and Escon 50/5 Motor Amplifiers [49] [50] Figure 61: Potentiometer Accuracy Test Figure 62: HEDS-5605 Encoder [51] Figure 63: SGAU Universal Strain Gauge Amplifier [52] Figure 64: Grounded Faraday Cage Around Torque Sensor Figure 65: Impedance Control Loop Implemented on Arduino Figure 66: Linear Step Input Figure 67: Position Sine Input Figure 68: Grasp Exercise Figure 69: Grasp Exercise Force Response Figure 70: Rotational Step Input Figure 71: Rotational Position Sine Input Figure 72: Pronation and Supination Exercise Figure 73: Pronation and Supination Torque Response Figure 74: Virtual Environment for Navigator System Figure 75: Capsule Collider Figure 76: Sphere Collider Figure 77: Top-down view of terrain

7 List of Tables Table 1: Summary of mechanical specifications and priority Table 2: Range of Pronation/Supination in Population (HFDS) [39] Table 3: Hand, Wrist and Forearm Calculations Table 4: Linear System Material Properties Table 5: Finger flexion force simulation constraints and results Table 6: Finger extension force simulation constraints and results Table 7: Arduino Mega performance characteristics [46] Table 8: Specifications of Motor Amplifiers [49] [50] Table 9: Controller Gains of Step Input Test Table 10: Controller Gains of Sine Input Test Table 11: Phase Shift Analysis Table 12: Positional Analysis of Sine Input Test Table 13: Controller Gains of Grasp Exercise Table 14: Controller Gains of Rotational Step Input Table 15: Step Input Analysis Table 16: Controller Gains of Rotational Sine Input Table 17: Phase Shift Analysis Table 18: Position Analysis of Sine Input Test Table 19: Controller Gains of Pronation and Supination Exercise

8 Acknowledgments First, I must thank my advisor, Dr. Dinos Mavroidis. His knowledge and guidance throughout the last year and a half has been invaluable. I owe Dr. Maureen Holden, Dr. Paolo Bonato, and Patricia Massey a great debt of gratitude. They offered their experience and advice to help define the focus and goals of the project. I cannot begin to express the support I've received from my labmates in my time in the Biomedical Mechatronics Laboratory. To the senior members, Dr. Rich Ranky, Dr. Maciej Pietrusinski, Dr. Mark Sivak, and Dr. Ozer Unluhisarcikli, and the current students, Mani Ahmadia, Yu Pu, Qing Chao 'Andy' Kong, Lexi Carver, Sean Suri, Amir Farjadian, and Elias Brassitos, thank you. I have learned so much from all of you. This work would not have been possible without the capstone design team: Katie Bausemer, Ray Adler, Joe Gonsalves, and Kevin Thompson. Thank you all so much for the countless hours you have put into this project. And most importantly, I must thank my family. Thank you Mom, Dad, Jim and Ted. I would never have gotten this far without your unconditional support. 8

9 1 Introduction 1.1 Project Significance Annually, approximately 795,000 people in the United States suffer from stroke, making it the leading cause of long-term disability in the nation [1]. Of these stroke victims, 85% have some kind of arm impairment, and 55-75% of victims retain the impairment after 3-6 months [2]. In 2008, the direct and indirect cost of stroke in the United States totaled $8.8 billion. Many stroke victims suffer motor impairments, speech impedimentals, and emotional problems long after stroke [1]. In 2010, robot assisted therapy was been identified by the American Heart Association as a method of rehabilitation which can provide assistance or resistance of movement, accurate feedback, and also as a method to provide rehabilitation to the patient with less assistance from the therapist. However, it also mentions that "Current robots tend to exercise only the proximal arm, and thus they improve motor skills at the should and elbow but not those of the unexercised wrist and hand; consequently, robots that only train the shoulder and elbow are limited in their ability to improve completion of ADLs (Activities of Daily Living) [3]". Because of this shortcoming in current rehabilitation robotics, robotic therapy of the hand and wrist is necessary research. Experts of the rehabilitation robotics field agreed with this assessment of clinical need in rehabilitation [4]. The technology used in hand and wrist rehabilitation has not progressed very far in quite some time. Currently, the primary tools used in hand and wrist rehabilitation are still low cost, low tech devices [5]. Because time with a therapist is often limited, creating a robotic device to exercise the hand and wrist that can be used in both a clinical and home environment could be greatly beneficial for stroke victim rehabilitation. These improvements would be most noticeable in quantitative tracking of improvement, and time spent doing rehabilitation exercises. 9

10 Due to these clinical needs, a new rehabilitation device was developed in the Northeastern Biomedical Mechatronics Laboratory, called the Navigator 2 Degree of Freedom (DOF) hand rehabilitation system. The Navigator device is related to the Herri device, a similar 2 DOF hand rehabilitation system previously developed in the same laboratory. The Navigator hand rehabilitation device's actuation system was developed by the author as a member of a team of undergraduate students at Northeastern University. The following work deals with the design and implementation of the device's handle, impedance control programming, and virtual reality software. These are three of the key aspects in delivering assistive rehabilitation exercises in the hand and wrist. 1.2 Key Terms These words and definitions are necessary to understanding the following work: Robotic Terms End effector: An end-effector is the point of interface between a human and a robotic device. It can also refer to a robotic device with an end effector that can interact with a patient. Exoskeleton: Exoskeleton devices are gloves that fit over the hand, and use actuators to move fingers either individually or simultaneously. Advantages of these devices are that they can provide individual rehabilitation for each finger, and good exercise for flexion and extension of fingers. However, these devices cannot provide wrist support for low-functioning patients, require many actuators, and are generally not portable because of the number of actuators involved. They are also generally limited to only flexion and extension exercises. Continuous Passive Motion (CPM): CPM devices move patients through a set range of motion repeatedly. The patient does not actively perform the movement. These devices are commonly used after surgery to ensure that the patient s range of motion is preserved. 10

11 Assistive Motion: A device which is assistive actuates the motion of the patient to assist them in completing the motion. These are useful for lower-functioning patients. Resistive Motion: A device which is resistive uses actuators to apply force or torque to resist the motion of the patient, which can be used for strengthening exercises. Electrorheological (ER) fluid: ER fluid is a liquid that experiences a change in viscosity in the presence of an electric field. Magnetorheological (MR) fluid: MR fluid is a liquid that experiences a change in viscosity in the presence of a magnetic field. Backdrivable: An actuation system is backdrivable when it has low friction of energy transmission. A system that can be moved when the motor is off is considered backdrivable Human Anatomical Terms Stroke: Stroke occurs when a portion of the brain is deprived of oxygen either due to a blocked artery or burst blood vessel which causes brain damage. Flexion: Flexion is the full grasping motion of the hand; fingers will end parallel to the hand, adjacent to the palm, as shown in Figure 1. Extension: The extension of the fingers is the full releasing motion of the hand; fingers will end parallel to hand, pointing away from the palm, as shown in Figure 1. Figure 1: Finger Motion ranging from fully extended (left) to fully flexed (right) [6] 11

12 Grasp: Grasping is the action of closing the fingers, pulling them tight the hand, in order to hold and object. A grasping motion involves only flexion and not extension. Pronation: The pronation of the wrist is the rotation of the hand about the axis of the forearm. A fully pronated wrist will cause the palm to face downward. It is a counterclockwise rotation with respect to the right forearm, as seen in Figure 2. Supination: Supination is the rotation of the hand about the axis of the forearm. A fully supinated wrist will cause the palm to face upward. It is a clockwise rotation with respect to the right forearm, as seen in Figure 2. Figure 2: A fully pronated (left) and fully supinated (right) wrist Abduction: Abduction is the motion of the hand with the palm facing down and the bending of the wrist towards the first finger or thumb Adduction: Adduction is the motion of the hand with the palm facing down and the bending of the wrist towards the fifth finger or pinky. Metacarpophalangeal (MCP): The MCP joints are located where the fingers connect to the hand, as seen in Figure 3. Proximal interphalangeal (PIP): The PIP joint is the second joint of each finger, as seen in Figure 3 below. Distal inter phalangeal (DIP): The DIP joints are the furthest joints from the hand located before the fingertips, as seen in Figure 3 below. Numbering of fingers: The fingers are numbered from one to five, with one being the thumb and five being the small finger. The numbers are labeled in Figure 3 below. 12

13 Numbering of fingers: The fingers are numbered from one to five, with one being the thumb and five being the small finger. The numbers are labeled in Figure 3 below. Figure 3: Joints of fingers and numbering of fingers Pinch 1: A pinch using tip of the first and second fingers, as seen in Figure 4. Pinch 2: A pinch using the tips of the first, second and third fingers, as seen in Figure 4. Lateral pinch: A pinch using the tip of the 1 st finger and the side of the second finger, as seen in Figure 4. Figure 4: Three pinch types Assessment Method in rehabilitation Fugl-Myers Assessment: The Fugl-Myers assessment evaluates and measures a post-stroke patient s path to recovery. In the test, the patient is required to perform 113 tasks, each of which is scored using the following metric: 0 - patient unable to complete task 1 - patient partially able to complete task 13

14 2 - patient able to complete task The scores of all of the tasks are added, with a maximum total score of 226 [7]. Motricity Index: The motricity index is a manner in which the motor impairment of a patient can be quantitatively assessed based on pinch strength, elbow extension, and shoulder abduction [8]. Proprioceptive neuromuscular facilitation (PNF) stretching: PNF stretching is type of physical therapy where the patient contracts particular muscles while keeping them in a static position. One example of PNF stretching is a patient holding their arm away from the body and resisting a therapist s motion as they press on top of the patient s arm for intervals lasting several seconds [9]. 1.3 Contributions The main contributions of this thesis are: The novel design and implementation of a linear acutation system, which utilizes serieselastic actuation to deliver assistive and resistive forces to the fingers for both flexion and extension The novel design and implementation of a rotational actuation drive system, which utilizes a belt drive to deliver assistive and resistive motion for pronation and supination of the wrist. The novel design and implementation a haptic interface handle, which offers position and load feedback in both DOF's and allows the user to exercise finger flexion and extension, wrist pronation supination, as well as the pinching exercises An impedance control algorithm implemented on the Arduino microcontroller platform, a previously unseen implementation of the algorithm. A rehabilitation game which will allow patients to exercise finger flexion and extension as well as wrist pronation and supination in a virtual reality environment As a results of this work, the following publications are in preparation: 14

15 P. Murphy and C. Mavroidis, "Design and Control of A Two Degree of Freedom Upper Limb Robotic Rehabilitation Device," Robotica: Rehabilitation Robotics and Human- Robot Interaction, Special Issue, P. Murphy and C. Mavroidis, " Design and Control of A Two Degree of Freedom Upper Limb Robotic Rehabilitation Device," in International Conference on Robotics and Automation, Hong Kong, Because of contributions of this work, two provisional patent applications have been filed: Murphy M et al., "Multiple Degree of Freedom Portable Rehabilitation System Having DC Motor-Based, Multi-Mode Actuator." INV Provisional Application Filed November 30, Ranky R et al., "Apparatus with Rolling Contact Handle." INV PCT Application Filed March 25,

16 2 Background 2.1 Previous Work Much work has been done in the area of robotic rehabilitation for the upper limbs. Many previous devices have been developed, and range from research prototypes to fully developed products available on the market. In addition, a great deal of these devices have been patented Devices on Market The Amadeo Rehabilitation System from Tyromotion focuses on exercising the extension and flexion of the fingers. The device is an end-effector system, with interaction between each of the user s fingers and the device. The robot implements its controls by sensing the movement of the patient, and helping the patient to finish the movement. This device provides support of the wrist and forearm, as seen in Figure 5, but does not restrict the movement of the patient. It is also easily adjustable to patients of different sizes [10]. While the device allows flexibility in how the patient rests his forearm, the device does not specifically address pronation and supination exercises. However, the device can exercise each finger individually, which is a design feature not address with the Navigator hand rehabilitation device. Figure 5: A patient using the Amadeo System [10] 16

17 Interactive Motion Technologies, Inc. (Watertown, MA), sells two products based on rehabilitation robotics research at MIT. Firstly, the InMotion Arm Robot is an upper limb rehabilitation robot, intended to improve the function of the shoulder and elbow. The MIT Manus had three DOF (each spatial direction), and the commercially available version of the robot includes a Hand Robot with grasp, release, and pinch features [11]. These products, along with the MIT Manus, are the most widely studied robotic upper limb rehabilitation devices. The device is shown in below. Figure 6: InMotion Arm Robot [11] A 2004 study on the effectiveness of the Manus noted that a key feature of the device was modularity. Even withthe same affliction, each patient will requre different exercises, so it is important that the parts for each exercise are modular [12]. InMotion also produces the InMotion Wrist Robot, shown in Figure 7 below. The robot has three DOF: flexion and extensionn of the wrist, abduction and adduction of the wrist, and pronation and supination of the wrist. The devices targets rehabilitation of the wrist specifically, without dealing with the hand [11]. Figure 7: InMotion Wrist Robot [11] 17

18 Although not specifically marketed as a hand rehabilitation device, the CyberGrasp haptic interface globe is similar to many of the exoskeleton devices being researched, which will be discussed in Section The device functions using one actuator per finger, to control the movement of each digit individually. The design intent was to create a glove that could mimic the feeling of real objects in a virtual environment, design mostly with naval applications in mind [13]. While not the original design intent, the glove has been tested for rehabilitation settings. While control of each finger is desirable, the glove is rather heavy, and offers no wrist support to the patients. Figure 8: CyberGrasp haptic interface for virtual environments [14] The Kinetic Maestra Hand and Wrist Tabletop Continuous Pasive Motion (CPM) Device is a rehabilitaiton device designed for post-surgery rehabilitation, shown in Figure 9 below. It exercises finger flexion and extension, wrist pronation and supination, and wrist flexion and extension. However, it can only perform one motion at a time, as it is a 1 DOF device with multiple attachements. Also, the device is CPM, which means the user is simply moved by the device, and does not interact with it [15]. The Maestra is available for patients to rent or purchase. It is available to rent at $475 per two week period [16], or purchased at $4339 [15]. 18

19 Figure 9: Kinetic Meastra Hand and Wrist Tabletop CPM Rehabilitation Device [15] The ReJoyce workstation by Rehabtronics is a table top workstation that focuses on post-stroke rehabilitation, shown in Figure 10 below. The device has a computer interface which allows the patient to play games while exercising with the machine. Patients can perform exercises which simulate a door knob, a key grip, side to side arm movement as well as coordination exercises. The ReJoyce has an adjustable arm, allowing the user to conduct the exercises at a comfortable height [17]. Figure 10: ReJoyce Rehabilitation Robot [17] The makers of the ReJoyce have conducted a clinical study involving 13 patients suffering from chronic spinal cord injury [18]. In the study, 13 people were taken from a group of 21 potential participants. 8 patients were excluded for reasons including having inadequate proximal control in arm, having hand muscles unresponsive to functional electric stimulation (FES), and being unable to commit to the fulll time of the study. The study compared methods of therapy for the upper limbs. The first method, conventional therapy, included strength training, computer games using a track ball, and therapeutic electrical 19

20 stimulation. The other method involved therapy with the ReJoyce system. This entails computer games associated with activities for daily living using the ReJoyce interface. Functional electrical stimulation (FES) was used to stimulate movement if necessary [18]. The study split patients into 2 groups. The first group went through 6 weeks of conventional therapy, 1 month washout, and 6 weeks of ReJoyce therapy. The second group went through 6 weeks of ReJoyce therapy, 1 month washout, and 6 weeks of conventional therapy. The Rejoyce system proved to be significantly better than conventional therapy using multiple performance metrics [18] Research Devices Other upper limb rehabilitation devices have been developed throughout the world. These devices can be classified into 2 main categories. First are exoskeletons, which are devices that are worn by the user, and end effectors, and end-effectors, in which there is an unworn interface between the user and the robot. Upper limb rehabilitation robots that deal with the elbow and shoulder will be discussed as wel Exoskeletons One exoskeleton device being developed, shown in Figure 11 below, is the Interactive Rehabilitation Robot for Hand Function Training, developed at Hong Kong Polytechnic University. This robot uses five actuators, one for each finger, and provides two DOF in the MCP and PIP joints of each finger using a mechanical linkage. The system is controlled by electromyographic (EMG) signals from the hand. Prior to testing, measurements of the EMG signals were taken from the subjects and used as a baseline for the control. With this baseline, the machine can determine whether the patient is trying to perform extension or flexion and assist in the motion. Using a trial of 8 stroke patients, it was found that after a 20-session study the patients displayed a statistically significant improvement in clinical assessments. In total, the device weighs about 500g, and covers many DOF for the hand. However, the device does not have a means of offering wrist support for lower-functioning patients. While the exoskeleton 20

21 design is very good for extension and flexion exercise, the lack of wrist/forearm support could be a hindrance to patients with little functionality in their hand [19]. Figure 11: EMG-Driven Exoskeleton Hand Robotic Training Device [19] Another exoskeleton was developed by Gifu University in Japan. The Hand Rehabilitation Support System, shown in Figure 12, focuses on the challenge of thumb opposability: patients with adduction contracture in the thumb cannot easily separate their thumb from the rest of the hand, making inserting the hand into the system difficult in many cases. This device implements a double parallel link structure to assist the thumb in motion, while ensuring that all motion is directed about the center of rotation [20]. This research also attempted to lower the setup time for the device. Figure 12: A hand rehabilitation support system with improvements based on clinical practices [20] The Actuated Finger Exoskeleton, developed by the Illinois Institute of Technology and Vanderbilt University, uses cables to transmit the torque from the motors to the finger, as seen in Figure 13 below. The current system only controls the index finger, but could be implemented for each finger. Since a cable can only exert force in tension, 2 motors are needed for each joint, 21

22 totaling in 6 motors and cables for the MCP, PIP, and DIP joints. The system uses the stress in the cables as the feedback control, using strain gauges on the wires to measure the force and torque exerted on each joint. Since the system uses cable actuation, it can move at high speeds, is back-drivable, and experiences little friction loss [21]. Although the results show the system to be lightweight and highly responsive, 6 motors and cables are needed per finger, which would make the system expensive and complicated if implemented for the full hand. Figure 13: Actuated Finger Exoskeleton for hand rehabilitation following stroke [21] The Hand Exoskeleton Rehabilitation Robot (HEXORR), shown in Figure 14, was developed at the Catholic University of America and the Center for Applied Biomechanics and Rehabilitation Research. It is a rehabilitation robot which uses mechanical linkages to control the joints, using one for all of the fingers and one for the thumb. All of the MCP joints are controlled by one motor and all of the PIP joints are controlled by another. One strong point of this device is that it replicates grasping patterns accurately, while using only two actuators. One weakness of the current design is that strapping the fifth digit into the linkage is rather difficult due to its size [22]. Figure 14: HEXORR Device attached to a hand [22] 22

23 Researchers from the University of Tokyo developed a finger rehabilitation device to recuperate individual fingers for patients with paralysis or contracture. The device used two four bar linkage assemblies with a pulley system, shown in Figure 15. The linkage system is adaptable to different hand sizes and allows the tension/flexion of the finger to be actuated by one motor, as seen in Figure 16. In this research, the control mechanism for the user is a glove worn on the opposite hand. The advantage of this is that the user has full control of the mechanism, allowing them to avoid discomfort. Another strength of this design is that the pulley performs the flexion and extension, which allows the actuator to be located off of the glove. This allows for a more lightweight design than other exoskeletons [23]. However, this device would require five motors for application to the entire hand. Other devices similar to this have been developed, but this device s pulley system removes the motor s weight from the patient s hand. Other devices have the motor directly connected to the linkages, greatly increasing the weight of the exoskeleton. Figure 15: Hand rehabilitation device using a wire-driven control mechanism and control glove [23] Figure 16: Means of extension and flexion in the device described in [23] 23

24 The Hand Exoskeleton System (HEXOSYS), shown in Figure 17 below, is a portable exoskeleton device with all of the motors and electrical components built onto the glove. This allows the system to be used anywhere, even at home. The device controls the thumb and the forefinger, with one active DOF for each finger. HEXOSYS includes a feedback system which can sense both position and force. Compared to other exoskeletons, this system is designed to be lightweight, weighing only 1 kg in its current form and able to be scaled down to 490 g if a reduction in output force is allowable [24]. However, HEXOSYS can only actuate two fingers. Though rehabilitation is possible, in nearly all cases it would be more useful to rehabilitate all fingers at once. Figure 17: The HEXOSYS portable hand rehabilitation device with two active DOF [24] End Effector Robot University of Washington researchers have developed a haptic device for stroke rehabilitation, shown in Figure 18, which focuses exclusively on a twisting motion, such as for opening a bottle cap or jar lid. The device implements four transducers to measure the force and torque exerted by the index finger, middle finger, and thumb. The system does not include actuation to assist motion, but it allows therapists to measure improvement in individual fingers [25]. Though the motions that can be exercised by the device are limited, it is small and safe without supervision and would be an easy exercise to conduct at home. 24

25 Figure 18: Multi-finger twist motion robotic device with computer interface [25] Another device, the ReHapticKnob, developed by the Engineering Rehabilitation Lab of the Institute of Robotics and Intelligent Systems, is designed to improve flexion and extension of the fingers and pronation and supination of the wrist in stroke patients [26]. The motions exercised by this device are similar to the Navigator device. While the ReHapticKnob exercises the same DOF as the Navigator, the grasp motion of the flexion and extension exercises is less ergonomically accurate and more similar to a pinching motion. As a potential product, this project s strength is the compact design. The product is safe without a therapist, making it a potential home product as well. The system uses two actuators, a linear actuator which moves both knobs, as seen on the right in Figure 19, and a rotational actuator for pronation and supination. One weakness of the system is that assistance in flexion will not be as effective as in an exoskeleton. However, the simple design leads to low friction losses, as well as a safe and compact product [26]. Figure 19: ReHapticKnob. Left: the full system with all DOF. Right: both knobs which are controlled by a linear actuator [26] 25

26 In 2011, The National University of Singapore conducted a feasibility study on an earlier version of this device, the HapticKnob. In this study, 15 chronic stroke patients with impaired upper arm function performed a 6 week rehabilitation regimen, using the device for three hours per week. The patients performed exercises for both flexion and extension of the fingers and pronation and supination of the wrist. In the extension exercise, the robot assisted in opening the patient s hand, held it in an extended position for 3 seconds, and then began the flexion exercise. In the flexion exercise, the patient flexed the fingers against a resistive load of 0-30N. In the pronation and supination exercise, the patient rotated the wrist towards an intended target on the computer screen, against a resistive torque of 0-1 N-m. To train motor control, the patients were asked to perform a task on the computer such as placing a picture in a frame [27]. After removing the data from two patients due to outside circumstances, the remaining 13 patients showed an average increase of 3.00 in the Fugl-Meyers test, and 4.55 on the Motricity Index, with statistically significant results. It was also found that patients grip strength ratio (impaired hand strength/unimpaired hand strength) increased by an average of 12.3% by the end of the study [27]. The results suggest that the robot can be used to advantageously train the hand and fingers of stroke patients and that distal training could benefit the function of the entire arm. A device developed at Sharif University, the Robohab, is designed primarily for patients with spastic muscle function post stroke. The Robohab, shown in Figure 20 below, has two positions, one for working on pronation and supination of the forearms, and another for flexion and extension of the wrist. This machine has seven different working modes for patient treatment, but the most common is the passive mode. Using torque sensors, the machine rotates the handle through a predetermined range of motion, and uses the torque sensors to measure the patient s spasticity. This machine is most effective for lower-functioning patients who are not yet ready for grip exercise [28]. 26

27 Figure 20: Robohab in pronation and supination position (left) and tension and flexion position (right) [28] The MR-Tech is a wrist rehabilitation device developed at the University of Libre Bruxelles in Belgium. Through interchangeable handles [29], the device can achieve three DOF of the wrist (pronation and supination, flexion and extension, abduction and adduction) [30], but only one at a time. The user interacts with the device by grasping the handle of the device and rotating it using the wrist [29]. The MR-Tech utilizes a magnetorheological (MR) fluid brake to apply varying amounts of resistance as the patient rotates the handle to maintain a designated rotational speed. The device can interface with a computer or PDA in order to evaluate, monitor and transfer data recorded during exercise, and is intended to be portable for at home use. The mechanical hardware and other MR-Tech systems are illustrated at a high level below in Figure 21. Figure 21: The three systems of the MR-Tech and how they interact together [26] 27

28 Preliminary clinical studies have been performed to compare the device to a commercial rehabilitation option (from CYBEX) but the MR-Tech was unable to emulate the commercial option already seen on the market [30] Upper Limb Rehabilitation Research Another well-researched area is robotic rehabilitation of the entire upper-limb, including the shoulder and elbow. The Robotic Assisted Virtual Rehabilitation System (shown in Figure 22) was developed using the Haptic Master, a commercially available 6 DOF robotic interface. The NJIT lab designed the virtual reality (VR) rehabilitation simulation. A clinical trial on children with cerebral palsy found that patients using the device improved the upper limb function more than the control group. The difference was not found to be statistically significant [31]. Figure 22: Two views of the NJIT-RAVR system [31] The Pneumatic Rehabilitation Robot was designed by researchers from Universidad Miguel Hernandez de Elche in Elche, Spain. The device focuses on proprioceptive neuromuscular facilitation (PNF) therapy for acute stroke patients, assisting the patient with defined PNF movements. The device also applies a virtual environment to re-teach daily skills. The system has two arms, one to control the patient s hand and one to control the patient s elbow. Both arms have three DOF, and are designed to mimic the way a therapist would help a patient with PNF movements. The robot operates at five levels, ranging from Level 1 where the robot offers assistive therapy, to Level 5 where the robot offers resistive therapy [32]. One strength of this system is the use of pneumatic actuators, which is a relatively lightweight means of controlling the arms. However, the device is very large and could not be released as a home product. 28

29 Figure 23: Pneumatic rehabilitation robot for PNF therapy [32] A robot developed by the University of Chicago and the Rehabilitation Institute of Chicago restores upper limb function and seeks to ensure proper posture of the arm and hand in conducting exercises. Studies found that without proper hand and arm posture, the robotic exercises could have negative effects on flexibility and muscle tone. The device uses eight active DOF and two passive DOF in translational motion of the shoulder, as seen in Figure 24 [33]. Although the scope of the HERRI project does not deal with robotic control of the elbow and shoulder, this does reiterate the importance of incorporating patient posture into the design of the device. Implementing proper supports for the wrist and forearm in the Navigator is crucial to having a robust design for patients to use. Figure 24: Whole Arm Exoskeleton Robot with 8 active and 2 passive DOF [33] The MIT Manus device is a beta prototype capable of two DOF motion used in clinical trials for the rehabilitation and recovery of stroke patients. During a therapy session, a patient sits a table 29

30 with his/her arm secured to the Manus arm and an on-screen prompt would instruct the patient to do a simple task such as connect the dots on screen or mimic a gesture shown. The device would sense motion from the patient and apply minimal force to assist the patient with the motion required to complete the task. In order to complete the tasks described above, the device uses two identical, upper and lower actuator packages, which employ direct drive transmissions. The packages include a brushless DC motor, position sensor, velocity sensor, and torque sensor aligned along the same axis. The patient s arm is strapped to the device in such a way that the hand can grasp a handle and the wrist is held motionless. Shoulder rotation and elbow flexion and extension can then be used to interact with the device. The Manus device is not used for hand rehabilitation; it is used for forearms, elbows and shoulders. The entire system weighs roughly 85lbf, and is costly to manufacture, therefore it is not ideal for home use [34]. Figure 25: Photo of MIT Manus Device [34] HERRI 2 DOF Hand Rehabilitation Device The 2 DOF Robotic Hand Rehabilitation System, developed at the Biomedical Mechatronics Lab at Northeastern University, is a robotic hand rehabilitation system focused on restoring hand and wrist function to stroke patients through task-specific exercises. The system has a linear DOF allowing for flexion and extension of the fingers and a rotational DOF allowing for pronation and supination of the wrist [35]. Through a graphical user interface the user can play virtual reality games that require task-specific exercises that allow the user to rehabilitate wrist and hand 30

31 function at the same time in an interactive manner. The device and its computer interface are shown in Figure 26 below. Figure 26: 1st Generation HERRI System [35] The system uses electro-rheological (ER) fluid for actuation. In both the linear and rotational actuators, the ER fluid flows through 3 concentric aluminum cylinders which act as electrodes. In the presence of an electrical field, the rheological properties of the ER fluid change, causing a change in yield stress [35]. In flow provided by a pump, the change in yield stress causes a change in pressure drop across the flow region. The pressure drop exerts a force on the area of the pistons of the linear and rotational actuators and is seen in Figure 27 below which displays their cross sections. The direct drive of the hydraulic system eliminates much of the friction that would be introduced by gears or cables. To further cut down on friction, precision machining was utilized in the system, which leads to less friction loss than other sealing mechanisms [35]. Figure 27: Cross-section of linear and rotational actuators of HERRI System [35] 31

32 The HERRI system also provides a virtual-reality simulation, allowing the user to increase motor function by creating an interactive therapeutic environment [36]. Figure 28 shows a screen shot of the virtual reality maze game. The HERRI system allows the patient to travel in two directions. Pronation and supination corresponds to left and right motion, and flexion and extension of the fingers corresponds to up and down motion in the game. The therapist can easily customize new mazes for the patient using the game engine. Figure 28: Example of Maze game running with HERRI system [36] The system also includes a graphical user interface for the therapist, which includes controls for customizing the resistive and assistive forces and designing new mazes. Figure 29 below shows the user interface. This allows the patient to play the rehabilitation games on one screen, while the therapist can control the assistive forces on another screen [36]. Figure 29: Graphical User Interface for practitioner [36] 32

33 The device was successfully implemented in healthy human feasibility studies. Figure 30 below depicts a healthy person using the HERRI system, with both the patient- maze screen and the practitioner graphical user interface screen visible. Figure 30: HERRI System fully implemented with both user and practitioner screens [36] 33

34 3 Mechanical Design The primary design goal of the Navigator system was to create a hand rehabilitation system with the same DOF's as the Herri system, but implemented for a home and clinical environment. To accomplish this, a comprehensive design process was implemented to create a device that can interact with patients comfortably and effectively. CAD of the final design can be seen in Figure 31 below. Figure 31: Final Design of Navigator Device 3.1 Specifications Design specifications were created through interviews with experts in engineering, biomechanics research, and physical therapy, as well as research into anthropometric data. While force and torque exertion data is readily available for healthy people, the force and torque which stroke patients are capable of exerting is not well documented. Therefore, engineers who worked on the Herri system conducted an unpublished study to document a force and torque range representative of stroke patients. 6 subjects participated in the study. The overall design goals of the project are to create a hand rehabilitation devicee that could be implemented in the home or clinical setting. Therefore, in addition to the anthropometric design aspects, safety, cost, portability, and modularity were crucial design features. The device also 34

35 needs to interact easily with a variety of PC's, which will be discussed further in Chapters 4 and 5. The final design specifications can be found in Table 1 below. A priority rating of 1-8 was applied to each specification in order to aid the design process, with 1 being crucial. Table 1: Summary of mechanical specifications and priority. Mechanical Design Specifications Category Description Specification Priority Source Force capacity (cyclic Force & Torque loading) 45 N 6 HERRI Report Torque capacity (cyclic loading) 1.6 N-m 6 HERRI Report Linear force capacity (Static Loading) 400 N 6 HERRI Report Sensors & Controller Maximum rotational velocity 600 /s 6 Dependent upon frequency response Maximum linear stroke velocity 7.7 cm/s 6 HERRI Report Minimum force perception 0.05 N 2 ETH Frequency response 5 Hz 1 UHD Ergonomics Linear stroke length 10 cm 3 Human Factor Design Standard Allowed degrees of rotation ±140 3 Anthropometry and Biomechanics Portability Weight <7 kg 1 Desktop area <0.16 m 2 1 Maximum setup time 3 minutes 7 Carrying Handling handle 8 Cost Final cost <$ Electronics PC interface USB Standard power source 15 A, 120V 4 Programming language Arduino 4 Safety Stop function Button 6 Backdrivability Backdrivable 3 Rotational hard stops ±140 5 Linear hard stops 8.6 cm 5 Interview with Dr. Paolo Bonato Human Factor Design Standard Human Factor Design Standard 35

36 Handle Priority 1 Mechanical Design Specifications (Cont'd) Category Description Specification Priority Source Hand width accommodation cm 8 Sensors & Controllers: frequency response (5 Hz) Human Factor Design Standard A frequency response of 5 Hz was selected because humans can achieve a maximum frequency response of 5-10 Hz [37]. This will allow patients to achieve the low end of the maximum frequency range of the human spectrum by using the device for rehabilitation. Portability: weight (<7 kg) Because the system is intended for home use, it must be highly transportable, and easy to set up and manipulate. It is essential to minimize the weight when the product goes to market to satisfy these requirements. 7 kg was identified as a maximum weight that could still be easily lifted and transported. Portability: desktop area (<0.16 m 2 ) A desktop area of less than 0.16m 2 was selected because it approximates the unoccupied area of a medium sized desk with a laptop or monitor and keyboard placed on it. Cost: final cost (<$3000) The final cost of the system was specified as <$3000 because the maximum available budget for the project is roughly $ Priority 2 Sensors & Controllers: minimum force perception (0.05 N) The lowest force able to be felt by a human was found to be N [38]. The upper limit of this range (0.05 N) was selected as the minimum force able to be sensed by the system because it is the lowest tolerance that is still acceptable within the human range. 36

37 3.1.3 Priority 3 Ergonomics: linear stroke length (10cm) The Human Factors Design Standard states that the 99th percentile of men has an index finger length of 8.6 mm. This is shown below in Figure cm was selected for the linear stroke length specification in order to accommodate this maximum finger length and provide appropriate mechanical clearance [39]. Figure 32: Index finger length [39] Ergonomics: Allowed degrees of rotation (±140 ) The human factors design gives the range of pronation/supination for men and women, as seen in Table 2 below. From the table, it can see that ±140 will cover range of motion of up to the 95 th percentile for adults. Table 2: Range of Pronation/Supination in Population (HFDS) [39] Safety: backdrivability For the device to interact effectively with users, the device must be able to sense the motion and force of the user. While the device would be functional without backdrivability, to increase effectiveness and safety, backdrivability could be implemented. 37

38 3.1.4 Priority 4 Electronics: PC Interface (USB 2.0) For the device to be effective, it should be able to interact with as many computers as possible. To be effective, the device should be able to interact with USB 2.0 cable, which can interact with a wide range of PC s. However, interacting with every PC is not crucial to the device s functionality. Electronics: Standard Power Source (15 A, 120 V) For the device to be easy to use and setup, it should be powered by a standard wall outlet. Therefore, the device should be powered by a standard 15A, 120 V wall outlet. However, this is not a crucial specification for the device to be functional. Software: Programming language (Arduino) Because the design team does not have an advanced skillset in programming and circuitry, the Arduino programming language will be used to control the device. While a number of languages could be used, Arduino is an open-source and intuitive programming language with extensive programming libraries available to draw from Priority 5 Safety: Rotational Hard Stops (±140 ) Based on the human factor design standard, the 95 th percentile of women can supinate to 140. Therefore, to accommodate left and right handed people, the device should be able to rotate to 140 in both directions. However, the range of motion will not be an issue for any of the possible motor systems. Safety: Linear Hard Stops (8.6 cm) Based on the Human Factor Design Standard, the 99 th percentile of men had an index finger length of 8.6 cm, which is a good approximation of flexion and extension stroke length. Therefore, to keep the system safe for a wide range of users, the system should have hard stops 38

39 that are adjustable up to 8.6 cm. However, the hard stops have minimal effect on the actuation design, and are therefore given a design priority of Priority 6 Force & Torque: Force capacity under cyclic loading (45 N) 45 N was calculated based on data from the HERRI Report. A sample of stroke patients (n=16) was tested to determine the average 5 repetition maximum for flexion and extension of the fingers. This 5 repetition average was found to be 27.8 ± 4.6 N; in order to accommodate a wide range of patients, three standard deviations were included [40]. Force & Torque: Torque capacity under cyclic loading (1.6 N-m) 1.6 N-m was calculated based on data from the HERRI Report. A sample of stroke patients (n=4) was tested to determine the average 5 repetition maximum for pronation and supination of the wrist. In the report, the maximum was found to be 63 N applied to a rotating handle. The shaft of the handle had a 2.54 cm radius which gives a 1.6 N-m torque response [40]. Force & Torque: Force Capacity in linear direction under static loading (400 N) N was found to be the maximum grip strength in 16 patients tested at Spaulding rehabilitation hospital [40]. As a result, 400 N was selected as the maximum force to withstand without damaging the system. Sensors & Controller: Maximum rotational velocity (600 /s) In order to satisfy a frequency response of 5 Hz, the system must be able to rotate at a maximum speed of 600 /s. This will ensure that the system has low impedance and that it responds faster than the patient can perceive. Sensors & Controller: Maximum linear stroke velocity (21 cm/s) A report written by researchers at Spaulding Rehabilitation Hospital showed that patients were able to carry out a grasp and release action at 21mm/s [40]. As a result, this speed was selected for the maximum linear stroke velocity of the device. 39

40 Safety: Stop function (Button) For the device to be safe, a stop function must be implemented. To accomplish this, a mechanical button will be used to provide a hard stop for the device. Since this could be accomplished by means of hardware or software, the button will be given a design priority of Priority 7 Portability: Maximum setup time (3 minutes) 3 minutes was selected as the maximum setup time for the system based on interviews with experts. This includes adjusting the system to the patient if necessary [41] Priority 8 Portability: Handling (Carrying Handle) During an expert interview with Dr. Paolo Bonato, it was determined that a handle should be incorporated into the design to enhance the portability of the system [41]. Handle: Hand width accommodation ( cm) The device needs to be able to accommodate a high range of hand sizes. For the Navigator to accommodate a wide range of users, the device should be able to accommodate hand widths of cm [39]. This hand width range covers the 5th percentile of females up to the 95th percentile of males. 3.2 Handle Design Anthropometric Design In designing a prototype for the handle, a modular design was implemented to allow interchange between multiple exercise configurations. From the previous work and background search, the key design aspects to test on the handle are: Exercise of entire hand or individual fingers adding a roller contact to the proximal portion of the handle to improve force sensing for finger extension 40

41 curved thenar grip for increased user comfort angled bar for increased comfort in grip Below the final configuration for the handle can be seen in its grasp/release configuration. To ensure that the handle meets the needs of a wide range of patients, the handle was designed to accommodate rnaging from the 5 th to 95 th percentile. The initial design of the handle can be seen below. Figure 33: Overall handle design One of the most basic necessities of accommodating the human is an adjustable length of grip for the user. To design the device for human use, anthropometric data for the hand, wrist, and arm were considered, as shown in Figures Figure 34, Figure 35, and Figure 36 below. Figure 34: Anthropometric data for finger length 41

42 Figure 35: Elbow-fingertip length Figure 36: Elbow- grip length during finger flexion As seen above, the overall range of finger length in the population is 2.3 in (1st percentile women) to 3.4 in (99th percentile men). To approximate the total stroke length of the hand during grasp, the elbow-grip lenth and the elbow-fingertip measurements are subtracted, as seen in Table 3 below. It should be noted that these can only be used as an approximation, as the difference in length was not measured at an individual level. Table 3: Hand, wrist and forearm calculations Women- 5th percentile (in.) Men- 95th percentile (in.) Elbow- Fingertip Elbow- Grip Difference It is seen that the distance between the fingertip and the center of grip will range from approximately 4.3 in to 5.9 in. However, these measurements also do not account for the width of 42

43 the thenar support in the handle. Therefore, the design specifications for the handle were based on the length of the index finger. The extension contact bar was redesigned with a compressible EVA foam pad, a typical material used for foam grips of many types of handles, including fishing rods. The gap between the grasp bar and extension bar was designed to the Human Factor Design Standard's range of finger diameters, as seen in Figure 37 below. Figure 37: Index finger width measurement To cover the 5th percentile of women to the 95th percentile of men, the device must be able to accommodate finger widths of.6 to.9 in. However, some extra spacing is needed, as constant contact on both sides of the fingers would create discomfort in the user. Therefore, the grasp and release bars were designed so that the spacing is 1 in. With this design the Navigator device will accommodate the vast majority of user hand sizes. Another configuration of the system was designed to accommodate pinching exercises, which are extremely important for injury related orthopedic rehabilitation. For this configuration, a groove was placed into the thenar support to accommodate the thumb. An attachment was designed to interact with the index finger for a two point pinch exercise. 43

44 Figure 38: 2 point pinch exercise Another attachment allows for a 3 point pinch exercise, in which the thumb meets both the index finger and middle finger, as seen in Figure 39 below. Both the 2 point and 3 point pinch exercises can be adapted to allow individual finger extension exercises. Figure 39: 3 point pinch exercise Improvements to Handle Design After manufacturing the handle, it was found the some redesign efforts were needed to increased comfort of use. The final design of the handle, shown in the grasp and release configuration, can be seen in Figure 40 below. 44

45 Figure 40: Final handle design- grasp and release configuration First, it was found that the flexion and extension bars should be moved farther apart. Although the 1 inch separation was found from anthropometric data, it was found that having both rolling contacts touching the hand was uncomfortable for use. Therefore, the distance used for between the flexion and extension bar was increased to 1.75 in. Next, it was found that the angled bar created a negligible difference in the comfort of use. Therefore, to increase the ease of assembly and manufacturability, the rolling contact bars were straightened to a vertical configuration, rather than the 5 angle applied to the bar as seen in Figure 40 above. 3.3 Linear Actuation The linear drive system uses a rotational motor with a rack and pinion and series elastic actuation to drive the system. A diagram of the major components of the system is shown in Figure 41: Linear actuation systembelow. This diagram is shown from the bottom in order to improve visibility of all major components. The potentiometer is not shown in the diagram because it is place in the proximal portion of the system near the handle. 45

46 Figure 41: Linear actuation system As seen in Figure 41 above, the system is driven by a geared motor. The motor, along with most components of the system, are secured in place using ABS base plate and mounting brackets. The motor is coupled to a rack and pinion system which translates the torque into linear force. The rack is secured on a mounting block with 3 shafts: 1 shaft exerts the force on the handle, and 2 separate shafts are used for stability. Each component of the linear actuation system is discussed in further detail in the following sections Motor Selection The motor was selected by using the force and velocity specifications and the equations of a rack and pinion. The following assumptions were made for selecting a motor: the pinion of the rack and pinion has a pitch diameter of 2 cm (r = 1 cm) The pitch angle of the pinion is 20 (Ø) the max linear velocity of the rack should be 10 cm/s the max linear force of the system should be about 50 N The pinion radius was arbitrarily selected, as any pinion between.8 cm and 1.5 cm would have worked. Therefore, the pitch radius of 1 cm was used. From here, an estimate of the output force was found through the following equation: 46

47 cos Ø From this equation a torque of.4698 N*m was found to be the approximate output needed from the motor. Next, the required angular velocity of motor was found through the following equation: Using a max velocity of 10 cm/s and a radius of 1 cm, a rotational velocity about 95.5 rpm was found. With these values, the final motor selected was a Maxon A-Max 22 motor, with a gear ratio of 53, and a final total output of 2.5 N*m and 300 rpm Series Elastic Actuation To increase backdrivability of the system, a series elastic actuator (SEA) was implemented for the linear drive system. One of the most difficult aspects of SEA design is choosing a spring which will sufficiently deflect under user input, but will not yield under the maximum input force to the system. To select a spring, the following criteria are most crucial to design Spring deflects while not yielding under rated loads Spring deflects in a manner which allows force resolution to be maintained The spring constant was found by needed to maintain a 10 Hz bandwidth while maintaining the necessary force measuring resolution. Therefore, the following criteria for the spring constant were used: 1 max 10 Where the maximum force is 50 N, and the max speed used is 7.7 cm/s. With a bandwidth of 10 Hz, the spring constant comes to N/mm, or lb/in. A spring constant of 37 lb/in is used in the system. 47

48 3.3.3 Sensor Integration To implement a system capable of impedance control, both force and position sensing are needed. Therefore, the linear acutation system was outfitted with both a tension/compression load cell, as well as a linear potentiometer. To guarantee that the system maintains the necessary force resolution, a tension and compression load cell was implemented in series between the linear actuation assembly and the handle. This will measure the interaction force between the user input to the handle, and the force exerted by the actuator. An Omega LC105 tension and compression load cell was implemented into the system. To measure the position, a linear potentiometer was implemented between the actuation system and the handle. For the position of the user's hand to be recorded accurately, the deflection of the spring cannot alter the measurement. For this reason, the potentiometer was placed distal to the actuation system, between the spring and the handle. The sensor configuration of the linear system can be seen in Figure 42 below. Figure 42: Linear actuation system sensor layout Finite Element Analysis To ensure that the linear actuation system is mechanically safe for use, a finite element analysis was conducted on the system to ensure that the rated loads do not cause the system to fail. The 48

49 analysis was conducted using Ansys Workbench software. The material properties used in the system, and their physical properties, can be found in XX below: Material Density (kg/m^3) Table 4: Linear System Material Properties Elastic Modulus (Pa) Poisson Ratio Tensile Yield Strength (Pa) Compressive Yield Strength (Pa) ABS [42] E E E+07 Acetal Plastic [42] E E E+07 Aluminum Alloy [43] E E E+08 Nylon [42] E E E+07 Stainless Steel [43] E E E+08 Structural Steel [43] E E E+08 Titanium Alloy [43] E E E+08 In the linear system, two different loading analyses ensure that the system can withstand the worst case loading. The first analysis configuration and results can be seen in Table 5 below. Table 5: Finger flexion force simulation constraints and results Parameters: Description Bottom of each mounting Fixed Support bracket 50 N Tension at end of Force center shaft Acceleration Gravity acting on system Moment.5 N-m at gear Nodes Elements Solution Number Part Max center Deformation m shaft Max Strain rack Max Stress e7 Pa rack Safety Factor rack As seen in Table 5, the first simulation run emulated the max loading the system will see during finger flexion exercise. The configuration was run as if the hand was exerting the max rated load of 50 N on the handle shaft, motor was employing a.5 N*m torque to negate the loading of the 49

50 hand. The bottom of each mounting bracket was confined to a fixed direction, and the acceleration of gravity was defined in the y direction to simulate the weight of the system. The analysis uses the default meshing pattern of Ansys workbench, yielding 87,877 elements in the system. To simplify the analysis, the roller bearings, thrust bearings and shaft collars were inserted as solid steel parts of the same dimensions. This is because these components are rated to loading greater than will be seen in the device and the fine geometry of the ball bearings and thrust bearings were greatly slowing down the meshing and solving processes of the computer. This was done during all simulations. The analysis finds the max deformation will occur in the center shaft of m, as seen in Figure 43 below. Figure 43: Deformation of finger flexion force simulation Because the loading is exerted at the end of the long, thin rail, it makes sense that the max deformation will take place at the end of the rail. However, it was found that the minimum safety factor of the center shaft is greater than 10, which means that shaft deflection is of little concern in the design of the system. The analysis found that the max stress experienced in the system occurs on the teeth of the rail, which is the expected result, as seen in Figure 44 below. 50

51 Figure 44 44:: Stress analysis of system at rack and pinion interface When more ore closely analyzed, it was found that the max stress experienced was 5.75e7 Pa, at the center of the tooth making contact with the gear, as shown in Figure 45 below. However, when considering the yield strength of the rack (made of structural steel), it was found that the safety factor of the system was 4.34, as seen in Figure 46,, meaning that the system could withstand quadruple the loading without yielding. Figure 45 45:: Stress analysis at tooth interface, point of max stress Figure 46:: Point of minimum safety factor (tooth/rack interface) 51

52 The next analysis run for the linear assembly simulates the max loading experienced during finger extension exercises. During these exercises, the fingers exert a max load of 50 N compressive force on the steel shaft, with the motors creating an offsetting torque of.5 N*m. The parameters and results of the analysis can be seen in Table 6 below. Table 6: Finger extension force simulation constraints and results Parameters: Fixed Support Force Acceleration moment Bottom of each mounting 50 N compressive at end of axis Gravity acting on system -.5 N-m at gear Nodes Elements Solution Number Part Max Deformation center shaft Max Strain nylon sleeve Max Stress 5.52E+07 Rack Safety Factor Rack As in the previous simulation, the bottom of each mounting bracket is fixed to simulate the device being bolted to the base plate. A 50 N compressive force is exerted on the handle s end of the shaft, and a moment of 0.5 N-m was applied to the outer face of the gear. With a gear pitch radius of 1 cm, the moment of the gear offsets the force of the shaft. Also, gravitational acceleration was applied to the system to simulate the weight of the system. As seen in below, the maximum deformation of the shaft is 2.82e-4 m, which occurs at the edge of the shaft. This makes sense given that the load is exerted on the end of the long, thin shaft. However, the stress analysis shown below found that the minimum safety factor of the center shaft was 10, making the shaft a non-critical factor in the design. 52

53 Figure 47: Deformation of finger extension force simulation Next, the stress analysis of this configuration found that the maximum stress experienced by the system is 55.2 MPa, as shown in the Figure 48 below. The max stress is found to be at the teeth of the rack, which would be expected given the loading conditions. Figure 48: Finger extension exercise stress analysis of system at rack and pinion interface As seen in Figure 49 and Figure 50 below, the maximum stress occurs at the center of the gear resulting in a minimum safety factor of With a minimum safety factor of 4.53, the system will be safe under any of the linear loading conditions. 53

54 Figure 49: Finger extension exercise stress analysis at tooth interface, point of max stress Figure 50: Finger extension exercise point of minimum safety factor (tooth/rack interface) As seen from the results of the analysis, the system is mechanically safe, and will be able to withstand the maximum loading that it will face. 3.4 Rotational Actuation The rotational dive system consists of a geared DC motor using a belt drive system to rotate the handle. A belt drive actuation system was chosen because it adds no additional backlash to the system. Figure 51 below displays the configuration of the rotational actuation system. 54

55 Figure 51: Rotational actuator layout Motor Selection To select the proper motor for the rotational actuation system, a motor is needed that can satisfy the max torque and rotational velocity from the specifications. To maximize system backdrivability, a 1:1 torque ratio was selected for the belt drive. Therefore, the direct output of the motor needs to match the specifications in Section 3.1. The motor needs to have outputs of at least 1.6 N*m and 100 rpm. A Maxon RE30 brushed motor with a gear ratio of 23, and a rated torque output of 2.5 N*m and 300 rpm is used Sensor Integration For the rotational actuation system to be capable of impedance control, the system needs both position and force sensing capabilities. Because the rotational drive is not series elastic, a torque sensor is necessary for implementing impedance control. Because the system is nearly rigid, the sensors were place towards the proximal end of the system, near the motor, to reduce tabletop area. A Futek TFF 325 reaction torque sensor is used. The torque sensor uses a strain gauge to measure the torque difference between the torque input from the user, and the torque input from the motor. The necessity of this measurement will be discussed further in Section To sense the position of the system, a HEDS encoder was inserted between the torque sensor and the 55

56 belt drive. The encoder has a resolution of.352. The configuration of the sensors can be seen in Figure Finite Element Analysis Similarly to the handle and linear actuation system, a finite element analysis was conducted on the rotational actuation system to ensure saftey. Ansys Workbench software was used. Overall, the same materials used in the linear actuation system were used in the rotational actuation system, which can be seen in Table 4. To analyze the rotational subsystem, the actuation system was analyzed to simulate the max torque of 1.6 N*m being exerted on the system. To create a worst case scenario, the system was analyzed such that 1.6 N*m was exerted on the belt drive, with an offsetting torque of 1.6 N*m exerted by the motor in the opposite direction. As with the linear system, the bottom of each ABS mounting bracket was fixed to simulate the brackets being mounted to the base plate. The acceleration of gravity was applied in the y direction to simulate the weight of the system. To simulate the torque exerted on the pulley, torque is considered by simulating the tension forces of the pulley. According to SDP-SI, a major vendor of gear and pulley systems, the minimum tension for an XL pulley system needs to be 5.1 lbf (22.7 N) per inch of separation between pulleys [44]. With a center distance of 2.5 in (6.35 cm) between each pulley, the tension of the pulley needs to be lb. (56.72 N), which will be exerted at both contacts for the pulley. When the pulley experiences a torque, one side of the pulley will feel the tension, while the other side of the pulley will feel both the tension and a torque-exerting force, shown in Figure 52 below. 56

57 Figure 52: Free body diagram of pulley With these concepts, the torque analysis was conducted on the pulley. On one side of the pulley, the tension force of lb. (56.72 N) was exerted. One the opposite tooth of the pulley, both the tension and the torque-exerting force were applied of 35 lb. (155.7 N) was applied. With a pitch diameter of in ( cm), the final output torque of these 2 forces is 1.6 N-m. The maximum deformation was found to be 9.86e-5 m on the gear tooth, as seen in Figure 53 below. Considering the tooth is the point that directly experiences the tension force, it is logical that the tooth will experience the greatest deformation. Figure 53: Deformation of rotational subassembly In the stress analysis of the rotational system, it was found that the maximum stress of the system was 65.6 MPa, experienced in the roller bearing which supports the mounting shaft for the pulley and encoder, shown in Figure 54, below. Although this was not the expected result, it makes 57

58 sense given that the tension of the pulley will exert and upward force on the pulley, thus creating an upward load for the pulley, shaft, and roller bearings. However, the minimum safety factor 3.81 of the subsystem also occurs in the roller bearing, as shown in Figure 55. Considering the high safety factor, the system is more than capable of withstanding the rated loads. Figure 54: Stress analysis of rotational subassembly Figure 55: Safety factor of rotational assembly Overall, with a minimum safety factor of 3.81, it was found that the rotational actuation system is mechanically safe for the loading that will be exerted on the system. 58

59 4 System Control 4.1 Introduction to Impedance Control Originally developed in 1985, Neville Hogan presented impedance control as a means of control for robotic manipulation tasks. Prior to his work, the main industrial application of robotics were in position control, especially in the areas of painting and welding. However, manipulation, in which the robot interacts with its environment, was a problem not sufficiently solved for implementation before Hogan presented impedance control. According to Hogan, in a haptic system with force feedback (i.e. any type of end effector with a force feedback for the user), the feel of the system is an extremely important measure. The feel of the system is directly related to the interaction dynamics between the actuation force and the input force from the user. Therefore, for controlling robots with human interaction, the best method is to control the interaction dynamics of the robot, rather than purely the actuation forces or input forces from the person. 4.2 Derivation of Impedance Control for Device According to Hogan, control of position or force alone is inadequate for manipulation, but rather controlling the dynamic behavior of the system while considering the interaction between user and actuator. To do this, the system must be controller by considering the impedance of the robot. At the most basic level, impedance can be defined as an object's resistance to motion. In mathematical terms, it can be defined as: Where Z() is the impedance, F() is the force as a function of the system's angular frequency, and V() is the velocity as a function of the system's angular frequency. Considering this controller in the Laplace domain, the parameters for control become both the force and position 59

60 of the system. Figure 56 below shows a general block diagram for the impedance control of this device. Figure 56: General impedance control loop In the Navigator device, there are two independent impedance control systems: one to control the linear DOF (finger flexion and extension), and the other to control the rotational DOF (wrist pronation and supination). Each control loop begins with the reference position and measured position, which using a PD controller yields a force command. This force command is used as the reference force, and with force measurement from the load cell, a PI controller is run to yield a command for the actuator Effect of Force Feedback Loop In impedance control, the main function of the force feedback loop is to compensate for the dynamics of the system, making the force reading much more accurate. In the case of the Navigator system, the force control loop can be simplified to the control of a mass which the actuator and user both act on, as seen in Figure 57 below: Figure 57: Dynamic interaction of forces on end effector 60

61 Overall, the basic dynamic equation of the system is: Where F act is the force of the actuaotr, and F ext is the force exerted by the user on the system. When a close loop control is applied to the system, Fact is the output of the force loop. Therefore, When the equations are combine, the following can be derived: 1 For the sake of simplification, assume that the reference force is null, which allows for the equation: 1 Meaning that as the user exerts force on the system, the apparent mass of the system will go down by a factor of G+1. However, because the force control is a PI controller, it is important to keep in mind: Much of this derivation of force gains is a continuation of previous work presented in the doctoral dissertation of Ozer Unluhisarcikli [45]. 4.3 Implementation of Impedance Controller The impedance control calculation for the Navigator system consists of a single microcontroller, an Arduino Mega equipped with an ATmega1280 microcontroller. The Arduino Mega has both digital and analog I/O capabilities as well as 16 MHz clock speed for its processor, allowing for both sensor input, and all controller computation on the single microcontroller. Using a microcontroller running only the impedance control loop, the programmer can avoid timing issues that can arise from performing the task on a computer running a general OS, such as Windows. These timing issues are especially crucial when running time-sensitive computations such as PD and PI controllers. 61

62 The microcontroller can still communicate with a desktop computer via serial communication, sending data to the computer for data collection, or game compatibility, which will be discussed in Chapter 5. This configuration also allows the device to be compatible with any generation of Windows without worrying about unexpected timing issues that can arise from older OS versions or older computer hardware Arduino One of the primary design goals of the Navigator system was to create a device with similar functionality to the Herri, but at a much lower cost. Reducing the cost of the system included finding lower cost electronics and software. Therefore, it was necessary to change the primary source of computation from a very expensive Labview Real-Time machine and OS to a much lower cost source of computation. With this in mind, the primary choices were either to compile the impedance controller directly on the computer, using the motor amplifiers to interact directly with a computer, or move to a low-cost microcontroller. Because of possible timing issues experienced on general-purpose operating systems discussed earlier, use of a low-cost microcontroller was chosen as the method of control. Therefore, the Arduino Mega microcontroller was used, shown in Figure 58 below. Figure 58: Arduino Mega microcontroller board [46] The key performance capabilities of the Arduino Mega are shown in Table 7 below. The Arduino Mega typically costs approximately $60, making it an extremely low-cost solution for the 62

63 computation of an impedance controller [47]. Therefore, one of the main research goals of the project is to characterize how well an Arduino Mega can implement an impedance controller. Table 7: Arduino Mega performance characteristics [46] Controller ATmega1280 Clock Speed 16 MHz Digital I/O 54 Pins Analog Input 16 Pins Analog Resolution 10 bits (1024 values,.0049 V) Analog Input Range 0-5V Digital Output PWM In implementing the impedance controller, the Arduino receives all sensor inputs, computes the impedance controller, and using pulse width modulation (PWM) sends an output signal to the motor amplifiers. PWM is a method of emulating an analog output voltage when only digital output is available. Rather than using the amplitude of the voltage as an output signal, PWM uses the duty cycle of digital signal. This means that the digital signal is HIGH/LOW, but the percentage of time the signal is HIGH is used to create a linear relationship between the PWM output of the Arduino and the current output of the DC motor amplifier [48]. Figure 59 shows PWM working at multiple different duty cycles. Figure 59: Example of Pulse Width Modulation at a)10% duty cycle, b) 50% duty cycle, and c) 90% duty cycle [48]. 63

64 4.3.2 Motor amplifiers Due to availability, 2 different motor amplifiers were used for the Navigator system: the Junus Digital Servoamplifier from Copley Controls, and the Escon 50/5 Servo Controller from Maxon Motor Company. However both work very similarly. Figure 60: Junus and Escon 50/5 Motor Amplifiers [49] [50] Using a PI control, the amplifiers allow for extremely accurate control of current to the motor using a PWM input signal. The main performance characteristics of both amplifiers can be seen in Table 8 below, which shows that both the PWM frequency and PI controller sampling rate more than match the frequency needed for a human interface device. Table 8: Specifications of Motor Amplifiers [49] [50] Specification: Escon 50/5 Junus Nominal Operating Voltage 10-50V 20-90V Output Current (Continuous) 5 A 10 A PWM Frequency 53.6 khz 40 khz PI Current Control Sampling Rate 53.6kHz 20 khz Because of the following approximation of the relationship between current and torque in a DC motor: With simple algebra it can be found that the PWM output of the Arduino has a linear relationship with the torque exerted by the motor. Because PWM signal ranges from 0-100%

65 Where i max and i min are the maximum allowed current, dictated by the torque specifications. Combining these equations, the torque is found to be linearly related to PWM:.01 These equations allow for the assumption that both force in the linear DOF and torque in the rotational DOF can be controlled directly from the PWM output of the Arduino, which is proven from the automated controller testing discussed later in this chapter Sensors To execute the impedance controller in both DOF's, 4 main sensors are needed: a potentionmeter for linear position, a load cell to measure force, an encoder for angle, and a torque sensor. Each of these sensors are discussed in detail below. In terms of the analog sensors, one somewhat unintended benefit of the Arduino Mega is the 10 bit resolution of the analog input on the arduino. Because the 10 bit resolution measures the analog signals in 4.9 mv increments, the microcontroller itself acted as a sort of digital filter, eliminating noise that occurred in amplitudes below this threshold. Some analog filtering was still needed, and is discussed below Position Sensors For the linear degree of freedom, a linear potentiometer was used to measure position. The potentiometer, with a travel length of 100 mm, was found to have a 1.4% full scale error, with the largest error occurring near the end of the run path, which is quite common in potentiometers.. The results of the test can be seen in Figure 61 below. 65

66 Figure 61: Potentiometer accuracy test The other position sensor used in the system was a rotational encoder, an Avago HEDS-5605, seen in below. This encoder has 1024 counts, meaning that it outputs 1024 pulses per revolution, giving it a resolution of.352, which is quite accurate for the Navigator device's application. Like the potentiometer, the Arduino powered the encoder. The high setting of the output pulse is the 5V supply coming from the Arduino, and the low setting is the Arduino's ground. One benefit of an encoder in comparison to a potentiometer is that the output signal is digital, and therefore there is no signal noise associated with the position of the encoder. However, the main weakness of the encoder use is that it is an incremental encoder, rather than an absolute encoder. This means that the sensor does not have an absolute position reading, the position of the system when the Arduino is turned on is considered to be 0. Therefore, the handle needs to be at approximately 0. When the system is turned on. This will be solved in future iterations of the system. Figure 62: HEDS-5605 encoder [51] 66

67 Load Cells and Amplifiers Load cells are a crucial component of the impedance control loop, as they allow for the force feedback loop within the larger impedance control loop. This allows for a programmatically lowered apparent intertia, allow the user to interact with the system. The Navigator system uses two load cells: a load cell and a torque sensor, each seen in Figure 63 below. The load cells each contain strain gauges that measure the deflection of the metal within the load cell. This deflection causes a change in output voltage within the strain gauge. The output voltage is then amplified with an Industrologic SGAU Universal Strain Gauge Amplifier, seen in Figure 63 below. The SGAU has an output voltage of 5V, meaning that the strain gauge's output voltage is amplified and normalized to an output voltage of 0-5 V. Although amplifiers with higher output voltages, and therefore, higher resolution are often used, there were 2 main reasons in choosing the SGAU amplifier. First, main strain gauge amplifiers cost upwards of $300, and the SGAU costs $69. Second, the 5V output voltage makes the amplifier fully compatible with the Aruino's input capabilities. Figure 63: SGAU Universal Strain Gauge Amplifier [52] In running the closed loop control of the rotational DOF, an unacceptable amount of noise was found from the raw signal of the amplifier. Two actions were taken to reduce the signal noise. First, it was found that the electromagnetic field from the motor was causing a great amount of noise in the sensor, and creating instability in the controller. Therefore, a faraday cage was place 67

68 around both the torque sensor and the rotational DOF's motor, which can be seen on the torque sensor in Figure 64 below. Figure 64: Grounded faraday cage around torque sensor While this improved the torque signal, both DOF's still required analog filtering to create a smoother signal. Using a lowpass RC filter, each analog signal was filtered at approximately 100 Hz. The 100 Hz frequency was chosen for filtering because it was found that the impedance control loop on the Arduino runs at approximately 200 Hz, and therefore, all noise running at a higher frequency than the control loop needed to be eliminated. 4.4 Implementation of Controller With all electronic components implemented with proper signal filtering, the impedance loop can be executed. All position and load sensors are connected to the Arduino, allowing for the inputs to be used. For tuning and initial testing, the impedance controller was modeled in both degrees of freedom as a fixed spring and damper system. This means that the position reference is a fixed value, and the user can interact with the controller with resistance mode exercise only. Although this does not implement the assistive exercise feature of the Navigator device, it is a necessary step in tuning the system. Tuning was primarily done manually. Although position-only control it is generally standard practice to model the system beforehand using tools such as root-locus or bode plots, force control tuning is not as well characterized, and often done heuristically. The 68

69 impedance control loop implemented is represented by the block diagram, seen below in Figure 65. Figure 65: Impedance control loop implemented on Arduino While the above loop represents the overall impedance control implemented, some nuances were added to the program increase both the stability and safety of the device. In tuning the controller, the biggest challenge is to set a constant K i in the force feedback loop which will sufficiently increase the accuracy and responsiveness of the controller, while maintaining stability. In order to ensure that the controller maintained stability, the full output of the PI controller was set to reach only the max possible force output of the controller. Pseudo code of the function is given below: { } While this is not a fullproof measure of safety, it ensures that the output of the loop does not exceed the maximum allowable force for the user to feel, and thus will help prevent any injury that could result from using the device. Early on in the tuning process, it was found that the controller showed instability due to small (microsecond scale) fluctuations in runtime for loop iterations. To remedy this instability, a 1 millisecond delay was implemented between loop iterations. This delay maintains an acceptable sampling rate for the impedance loop, while making any issues that arise from timing inconsistency negligible. Refer to Appendix A for the full code of the impedance control implementation on the Arduino Mega. 69

70 4.5 Controller Results Tuning of the controller was primarily done manually. Although position control tuning is extremely well characterized, with many tools available, force control is generally performed manually in application. Because of this, the impedance controller for both degrees of freedom was tuned manually. Because the position control portion of the impedance loop outputs the force command, the control gains of the position PD controller actually represent a physical spring and damper constant, as seen below: From these equations, the F command is then used as the reference in the force feedback loop. Because of this output, Kp represents a physical spring constant (N/mm and N*m/degree) and Kd represents a physical damper (N*s/mm and N*m*s/degree). From the anthropometric data, the desired max force and torque were known to be 50 N and 1.6 N*m, respectively. Therefore, Kp in the linear direction was initially set at 2 N/mm. Kp of the rotational direction was initially set at.015 N*mm/degree. Kd, as well as the force gains, were tuned manually. The impedance controller was tuned using three separate tests. First, automated step input and sine input tests were conducted. After the automated tests, the controller was tuned using resistance mode grasp and release and pronation and supination exercises Linear Actuator Results The linear impedance controller was initially tested with a step input test. Using the impedance loop, a position step input of 25 mm was given to the system. After manually tuning the position and force controller, the best combination of gains found was the following: 70

71 Table 9: Controller Gains of Step Input Test Position Kp 3 Kd 0.3 Force Kp 1 Ki 0.5 Figure 66: Linear step input The settling time of the system is.224 s, and the sytem settles at mm, for a steady stata error of.21 mm. While ideally the settling time would be slightly faster, for a human interaction device this is acceptable. With the step input tuned to an acceptable result, a sine wave position input was implemented in the system. An amplitude of 15 mm was implemented, giving a total path of travel of 30 mm, which is approximately the distance travelled during a grasping exercise. A frequency of 1 Hz was used for testing. 1 Hz is approximately the speed a person performing a grasping exercise will use. Therefore, it was the frequency chosen for controller tuning. The controller gains implemented in the system can be found in Table 10 below, and the results of the test can be found in Figure 67. Table 10: Controller Gains of Sine Input Test Position: 71

72 Kp 2 Kd 0.35 Force Kp 1 Ki 0.5 From the sine input test seen above, it was found that the impedance controller is extremely responsive with low overshoot. The test shows that the average phase shift between the reference position and output was ms, with the max phase shift being 50 ms. This was measured by comparing the time between peak values in the reference and measured trajectories. More information can be seen in Table 11 below. Figure 67: Position sine input Table 11: Phase Shift Analysis Phase Shift (s) Average St. Dev Min Shift Max Shift The sine wave input had an acceptable overshoot as well. The max overshoot in the controller throughout the test was mm, which given the sensitivity of the persons hand is an acceptable amount of overshoot. More can be seen in Table 12 below. Table 12: Positional Analysis of Sine Input Test Position Error (mm): 72

73 Max Position Min Position 33.9 Max Overshoot 1.88 The final tests conducted in tuning the linear actuator's controller was a resistance mode grasp exercise. For this test, position setpoint of the actuator was set to 60 mm, and a grasping exercise was performed at approximately 1 Hz. A diagram of this test can be found in Figure 68 below. Figure 68: Grasp exercise The goal of this test was to measure the force response of the controller. The force gains were kept the same as the sine wave input test, as changing the force gains risks instability. The gains of the position feedback were changed according to the inventor's comfort. The controller gains can be found in Table 13 below, and the results of the test can be seen in Figure 69. Table 13: Controller Gains of Grasp Exercise Position: Kp 1.6 Kd 0.2 Force: Kp 1 Ki

74 As seen above, the force feedback edback offset the position error in a very responsive manner, which is the desired outcome in a resistive mode exercise Rotational Actuator Results Figure 69: Grasp exercise force response The rotational controller was initially tuned using a step input test. Using the impedance control loop, a step input of 90 was input to the loop. As with the linear controller, the position gain Kp was driven by the max allowed torque of 1.6 N*m. Using the maximum allowed torque and max angle of 108, an initial Kp of.0148 N*m/degree was used. However, in tuning it was found that a stiffer constant was need for a responsive step input. Table 14 below shows the final parameters implemented in the controller, and Figure 70 displays the final results. Table 14: Controller Gains of Rotational Step Input Position Kp Kd Torque Kp 0.55 Ki

75 Figure 70: Rotational step input As seen above, the impedance controller yielded both a fast and accurate response. The full results can be found in Table 15 below. Table 15: Step Input Analysis Overshoot 0.78% Steady State Error 0.39% Rise Time (s) Settling Time (s) The next test implemented was a sine wave input. An amplitude of 90 was used at a frequency of 1 Hz. The gains used in the test can be seen in Table 16 below, and the results of the test can be seen in Figure 71. Table 16: Controller Gains of Rotational Sine Input Position Kp Kd Torque Kp 0.55 Ki

76 Figure 71: Rotational position sine input As seen above, the controller is very accurate and responsive. The average phase shift measured was 29 ms, measured by comparing the time between the peak values of the reference and measured positions. The maximum overshoot of the controller was Further analysis of the phase shift and overshoot can be seen in Table 17 Table 18 below. Table 17: Phase Shift Analysis Phase Shift (s) Average St. Dev Max Min Table 18: Position Analysis of Sine Input Test Position Error (degrees) Max 90 Min Max Overshoot 2.11 The final test to tune the rotational impedance controller was a resistive mode pronation and supination exercise. For this test, the reference position of the system was set to 0, and a pronation and supination exercise was conducted at a comfortable speed, which turned out to be approximately.5 Hz. The procedure of this test can be seen in Figure 72 below. 76

77 Figure 72: Pronation and supination exercise The final controller gains implemented can be seen in Table 19, and the controller's torque response can be seen in Figure 73 below. Table 19: Controller Gains of Pronation and Supination Exercise Position: Kp.035 Kd Force: Kp.55 Ki 0.2 Figure 73: Pronation and supination torque response As seen above, the controller's torque response is extremely responsive, but slightly noisy. It is noisiest when the wrist is fully pronated and supinated. This is because the torque is approaching the max torque that the system is able to output, and therefore the system reacts to rapidly lower 77

78 the torque. However, the force responsiveness is a much more important factor in the user's feel of the system, and as seen in the graph the torque response to the system is acceptable. 78

79 5 Virtual Reality Interface 5.1 Virtual Reality in Rehabilitation In the past decade, virtual reality (VR) has been increasingly studied for physical therapy and neurorehabilitation. Research to this point has indicated that VR can be a useful tool in aiding rehabilitation, warranting further study into its efficacy and effectiveness [53]. Many VR systems originally intended for entertainment have been adapted for use in physical therapy and neurorehabilitation. The most common of these is the Nintendo Wii, which is a video game console using infrared sensors and accelerometers to sense the controller positions in 3 dimensions. Adapting existing entertainment VR systems for therapy has many benefits, including low cost and wide availability. These factors allow the system to be easily available for both the home and clinical setting. One of the most common problems therapists and patients experience during therapy is patient boredom [5]. Implementing a virtual environment can help keep the patient motivated, while also requiring less immediate attention from the therapist. As the population of patients needing neurorehabilitation grows, therapy methods that require less intensive attention from the therapist will become more valuable. 5.2 Game Concept Development In developing a virtual environment for the Navigator device, it is important to tailor the environment to the device's two degrees of freedom. Therefore, creating motions in the virtual environment that feel natural to the patient when performing grasp/release exercises, as well as pronation/supination is crucial to creating an effective environment for rehabilitation application. An intuitive feel is crucial for the patient being able to associate a specific task with the motions they are performing. 79

80 Aside from intuition of move movement, ment, another key consideration is the main user base of the virtual environment. The Navigator device's primary target is stroke victims, which means that the population using the device will generally be older and less familiar with using virtual environments. ments. Therefore, simplicity and intuitiveness must be kept as a high priority while maintaining interest and motivation from the user. 5.3 Virtual Environment Design A 3D virtual environment was created which allows the user to exercise both pronation and supination pination of the wrist, as well as flexion and extension of the fingers. The DOFs are represented in the virtual environment through the following actions: pronation of the wrist causes the character to rotate counter counter-clockwise clockwise (to the left) supination of the wrist causes the character to rotate clockwise (to the right) flexion of the fingers causes the character to move forward extension of the fingers causes the character to stop When the virtual environment is initiated, there are two non non-terrain terrain objects: the player character and the target sphere. The main goal of the game is to make contact with the target sphere. Upon reaching the target sphere, the user's score will increase by one point, and a new sphere will generate. The collision detection etection between the user and target sphere, as well as other game design components will be discussed in the following sections. Figure 74: Virtual environment for Navigator System 80

81 As seen above, there are multiple feedback mechanisms to make the game more interactive and useful to the user. The terrain obstacles designed into the game necessitate that the user use both flexion and extension to stop and go, while pronating and supinating to turn around the obstacles. Once the user reaches the target sphere, a new target will generate, and the user's score will increase by one point. As seen in the top right of Figure 74, a separate camera allows the user to see where on the map the player character and target sphere are. This allows the user to see where the player character and target sphere are on the map. 5.4 Game Design Components Collision Detection One of the primary components needed from the Unity engine was its collision detection capabilities. For a virtual environment to function properly, collision detection is needed for when 2 objects are in contact with each other. Unity has the capability of applying collision detection to both continuous shapes, or as a mesh in cases where the shape is quite complicated. The colliders used for collision detection in Unity are the capsule, sphere, and terrain colliders, shown below. Figure 75: Capsule collider 81

82 The capsule collider, shown in Figure 75 above, is the shape used for the first person camera of the player character. The shape utilizes a semi-sphere at the top and bottom of a cylinder. Therefore, the game engine uses the height of the cylinder, as well as the radius of the cylinder and semi-spheres to detect whether the player character has collided with another object. Figure 76: Sphere collider The sphere collider, shown in Figure 76 above, is the collider type used for the target spheres in the game. The sphere collider needs only the radius of the sphere is in contact with another object. This makes for relatively simple computation, which allows the virtual environment to run more efficiently. Figure 77: Top-down view of terrain The final type is the terrain collider. A top-down view of the terrain used can be seen in Figure 77. The terrain collider uses the heightmap to calculate if an object is colliding with the terrain. 82