USF Libraries
USF Digital Collections

Design, construction and testing of a wheelchair-mounted robotic arm


Material Information

Design, construction and testing of a wheelchair-mounted robotic arm
Physical Description:
Edwards, Kevin D
University of South Florida
Place of Publication:
Tampa, Fla.
Publication Date:


Subjects / Keywords:
Machine design
Dissertations, Academic -- Mechanical Engineering -- Masters -- USF   ( lcsh )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )


ABSTRACT: A wheelchair-mounted robotic arm (WMRA) was designed and built to meet the needs of mobility-impaired persons, and to exceed the capabilities of current devices of this type. The mechanical design incorporates DC servo drive, with all actuator hardware at each individual joint, allowing reconfigurable link lengths. It has seven principal degrees of freedom and uses a side mount on a power wheelchair. A simple, scalable control system allows coordinated Cartesian control, and offers expandability for future research, such as coordinated motion with the wheelchair itself. Design payload including gripper is 6 kg, and the total arm mass with controller is 14 kg. These and other design attributes were confirmed through testing on the completed prototype.
Thesis (M.S.M.E.)--University of South Florida, 2005.
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
Statement of Responsibility:
by Kevin D. Edwards.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 79 pages.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001670401
oclc - 62364773
usfldc doi - E14-SFE0001276
usfldc handle - e14.1276
System ID:

This item is only available as the following downloads:

Full Text
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001670401
003 fts
005 20051216093354.0
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 051129s2005 flu sbm s000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0001276
TJ145 (Online)
1 100
Edwards, Kevin D.
0 245
Design, construction and testing of a wheelchair-mounted robotic arm
h [electronic resource] /
by Kevin D. Edwards.
[Tampa, Fla.] :
b University of South Florida,
Thesis (M.S.M.E.)--University of South Florida, 2005.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
Title from PDF of title page.
Document formatted into pages; contains 79 pages.
3 520
ABSTRACT: A wheelchair-mounted robotic arm (WMRA) was designed and built to meet the needs of mobility-impaired persons, and to exceed the capabilities of current devices of this type. The mechanical design incorporates DC servo drive, with all actuator hardware at each individual joint, allowing reconfigurable link lengths. It has seven principal degrees of freedom and uses a side mount on a power wheelchair. A simple, scalable control system allows coordinated Cartesian control, and offers expandability for future research, such as coordinated motion with the wheelchair itself. Design payload including gripper is 6 kg, and the total arm mass with controller is 14 kg. These and other design attributes were confirmed through testing on the completed prototype.
Adviser: Dr. Rajiv Dubey.
Machine design.
Dissertations, Academic
x Mechanical Engineering
t USF Electronic Theses and Dissertations.
4 856


Design, Construction and Testing of a Wheelchair-Mounted Robotic Arm By Kevin D. Edwards A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science of Mechanical Engineering Department of Mechanical Engineering College of Engineering University of South Florida Major Professor: Rajiv Dubey, Ph.D. Glen Besterfield, Ph.D. Daniel Hess, Ph. D. Date of Approval: July 14, 2005 Keywords: wmra, rehabilitation, reconfigurable, machine design, adl, assistive Copyright 2005, Kevin D. Edwards


i Table of Contents List of Tables iv List of Figures v Abstract vi Chapter One Introduction 1 1.1 Motivation 1 1.2 Objectives 2 Chapter Two Background 4 2.1 History of Rehabilitation Robotics 4 2.2 Workstation-Based Systems 5 2.3 Mobile Systems 10 2.4 Integrated Robotic Systems 12 2.5 Research WMRAs 14 2.6 Commercially Available WMRAs 16 2.6.1 The Manus 16 2.6.2 The Raptor 17 2.7 WMRA Mounting Locations 18 2.7.1 Rear Mount 19 2.7.2 Side Mount 19 2.7.3 Front Mount 20 2.8 Control System Types 22 2.8.1 Closed Loop Control 22 2.8.2 Open Loop Control 23 Chapter Three Design 25 3.1 General Goals for the Design 25 3.1.1 Mechanical Constraints 25 Weight 25 Mounting Location 25 Stiffness 26 Payload 26 Reconfigurability 27 Power Supply and Consumption 27 3.1.2 Cost Constraints 28 3.1.3 User Requirements 29


ii 3.2 Types of Systems Considered 30 3.3 Final Design 32 3.3.1 Kinematic Arrangement 32 3.3.2 Component Selection 34 Gearhead Selection 34 Motor Selection 37 Encoders 38 Controllers 39 3.3.3 Material Selection 41 3.3.4 Joint Design 41 3.3.5 Wrist Design 42 3.3.6 Control System 44 3.3.7 Final Design Overview 45 Chapter Four Construction 47 4.1 Considerations for Manufacture 47 4.2 Completed WMRA 48 Chapter Five Testing 49 5.1 Safety Tests 49 5.2 Stiffness Testing 51 5.3 Strength Testing 53 5.4 Joint Speed Measurements 54 5.5 Energy Consumption Testing 55 5.6 Further Testing 56 Chapter Six Summary and Future Work 58 6.1 Design Summary 58 6.2 Design Insights 59 6.2.1 Degrees of Freedom 59 6.2.2 Reconfigurability 59 6.2.3 Side Mounting 59 6.2.4 Cartesian Control 59 6.3 If I Had To Do It Again 60 6.4 Future Work 61 6.4.1 Develop High-Level Controller 61 6.4.2 Orientation Locking 61 6.4.3 Gripper Development 62 6.4.4 Trials 62 6.4.5 Integration with Power Chair 62


iii 6.4.6 Machine Vision Assist Functions 63 6.5 Conclusion 63 References 64 Appendices 66 Appendix A Cost Estimate 67 Appendix B Kinematic Transformation Matrix 68 Appendix C Joint Torque Calculations 70


iv List of Tables Table 3.1: DH Parameters for the USF WMRA 34 Table 3.2: Table 3.3: HD Systems Gearhead Selections for Each Joint 36 Table 3.3: Motors Used in USF WMRA 38 Table 5.1: Arm Deflections vs. Applied Load 51 Table 5.2: Maximum Joint Loads 54 Table 5.3: Joint Speed Measurements 54 Table 5.4: Power Usage 55 Table 6.1: USF WMRA Specifications 58 Table Appendix A.1: Cost Estimate 67 Table Appendix C.1: Available Harmonic Drive Gearhead Specifications 70 Table Appendix C.2: Joint Torque Design Spreadsheet 70


v List of Figures Figure 2.1: Puma 250 Arm 6 Figure 2.2: Handy-1 7 Figure 2.3: RAID Workstation 8 Figure 2.4: Robot Assistive Device 9 Figure 2.5: ProVAR System 10 Figure 2.6: MoVAR 11 Figure 2.7: MoVAID 12 Figure 2.8: FRIEND Robotic System 13 Figure 2.9: TAURO Robotic System 14 Figure 2.10: Weston Arm 15 Figure 2.11: Asimov Arm 16 Figure 2.12: Manus Arm 17 Figure 2.13: Raptor Arm 18 Figure 2.14: Manus Joystick Controller 23 Figure 2.15: Manus Keyboard Controller 23 Figure 3.1: Complete SolidWorks Model of the USF WMRA 32 Figure 3.2: Kinematic Diagram, with Link Frame Assignments 33 Figure 3.3: HD Systems Harmonic Drive Gearhead 35 Figure 3.4: Pittman PMDC Brush Motor with Gearbox and Encoder 38 Figure 3.5: J.R. Kerr PIC-SERVO Controller Board 39 Figure 3.6: Typical Joint Bracket Design, Showing Motor, Gearhead and Bracket 42 Figure 3.7: Nonorthogonal Wrist Concept 43 Figure 3.8: 3-DOF Orthogonal Wrist Concept 43 Figure 3.9: Complete SolidWorks Model of the USF WMRA 46 Figure 4.1: Andrew Bridges Milling a Joint Bracket 48 Figure 4.3: Completed Arm on Power Chair 48 Figure 5.1: Automatic Shutdown when Force Limit is Exceeded 50 Figure 5.2: Arm Stiffness Measurement 52 Figure 5.3: Strength Testing of Joint 1 53 Figure 5.4: WMRA in a Feeding Pose 56 Figure 5.5: Low-Right Reach 57 Figure 5.6: Mid-Right Reach 57 Figure 5.7: Left Side Doorknob 57 Figure 5.8: Right Side Doorknob 57 Figure 6.1: SolidWorks model 58 Figure 6.2: USF WMRA as Built 58


vi Design, Construction and Testing of a Wheelchair Mounted Robotic Arm Kevin D. Edwards ABSTRACT A wheelchair-mounted robotic arm (WMRA) was designed and built to meet the needs of mobility-impaired persons, and to exceed the capabilities of current devices of this type. The mechanical design incorporates DC servo drive, with all actuator hardware at each individual joint, allowing reconfigur able link lengths. It has seven principal degrees of freedom and uses a side mount on a power wheelchair. A simple, scalable control system allows coordinated Cartesian control, and offers expandability for future research, such as coordinated motion with the wheelchair itself. Design payload including gripper is 6 kg, and the total arm mass with controller is 14 kg. These and other design attributes were confirmed through testing on the completed prototype. This paper discusses the current state of the art in WMRAs; describes the design goals and user requirements for this device; explains the component selection process; discusses details of the mechanical design, electrical system and low-level controller; covers manufacturing concerns; and describe s the testing of the completed arm. Suggestions for further development are also included.


1 Chapter One: Introduction 1.1 Motivation A wheelchair mounted robotic arm can enhance the manipulation capabilities of mobility-impaired persons, and reduce dependence on human aides. Unfortunately, the present state of the art for this applicati on has not met with much commercial success, which may be due to poor usability and low pa yload. It is difficult, cumbersome, and sometimes impossible to accomplish everyday tasks with the WMRAs currently on the market. This project attempts to surpass previous devices in terms of performance, while maintaining cost competitiveness. Data from the US Census Bureau Sta tistical Brief of 1993 showed that over 34 million Americans had difficulty performing functional activities. Of this number, over 24 million were considered to have severe di sabilities. Every year more and more people become disabled in a way that minimizes their use of upper extremities. These can be motor dysfunctions due to accidents, disease, aging, or genetic predispositions. The field of Rehabilitation Robotics has been created in an attempt to increase the quality of life and to assist in activities of daily living. Rehabilitation Robotics addresses assistive technologies as well as the traditiona l definition of rehabilitation: increasing or expanding the individuals mental, physical, or sensory capabilities. The primary focus of


2 Rehabilitation Engineering and robotics is to increase the quality of life through increasing functional independence, and decreasing the costs associated with the assistance required by the individual. We are interested in people who have limited or no upper extremity mobility. Robotic aids used in these applications vary from advanced limb orthosis to robotic arms. Persons that benefit from the devices are t hose with severe physical disabilities, which limit the ability to manipulate objects. These de vices increase self-sufficiency, and reduce dependence on caregivers. The following are ex amples of those who could benefit from a robotic arm. In the case of spinal injury or dysfunction, robotic aids are most appropriate for individuals with spinal deficiencies ra nging from cervical spine vertebra 3 through cervical spine vertebra 5. Below the cervical spine vertebra 5, individuals often can be served with simpler, more traditional assistive technology. Persons with these injuries can generally make use of their upper limbs, and robotic arms are not necessary, nor significantly improve quality of life. Simila rly, persons with spinal fractures above cervical spine vertebra 3 are also not serv ed well by robotic assistance. Injuries are usually so debilitating with this type of inju ry that a respirator and daily attendants are required, thereby reducing the benefit of assistive devices. Other individuals that could benefit from a robotic arm are persons with neuromuscular deficiencies such as multiple sclerosis. Since persons with the above disabilities re quire mobility assist devices, such a power wheel chair, this power wheelchair is the natural platform for adding further mobility assistance. There have been several attempts in the past to create commercially


3 viable wheelchair mounted robotic arms. The power wheelchair not only provides an excellent structure with which to mount th e device, but also provides a power supply. Currently there are only two commercial WMRAs available, the Manus (Exact Dynamics, Inc., Netherlands) and the Raptor (Applied Resources, Inc, NJ USA). 1.2 Objectives The main objective of this thesis was to design and build a prototype WMRA. This manipulator had to be lightweight, able to carry a 4 kg payload at full reach, and be capable of Cartesian control. In additi on, it had to be cost competitive with other WMRAs and with traditional human assistants. This paper discusses the many decisions that led to a product meeting these and other specific requirements.


4 Chapter Two: Background 2.1 History of Rehabilitation Robotics There have been various attempts over th e years to create robotic assistants for individuals with various levels of disabilities. For over 30 years research has progressed in the field with only partial commercial success. An early attempt at telemanipulators was done at the Case Institute of Technology during the early 1960s. The Case system was a floor mounted, powered exoskeleton. The operator controlled the device by wearing a head-mounted light source that triggered light sensors in the environment. By looking at specific points in the room, the operator could trigger the light sensors, and initiate one of several preprogrammed gestures that were stored on magnetic tape. A later devel opment allowed for Cartesian movement and direct control of individual joints, along with myoelectric signals for velocity control. One of the first attempts at rehab ilitation robotics included the Rancho Golden arm, designed in 1969 at Rancho Los Amigos Hospital in Downy, California (Reswick 1990). The arm was an electrically driven 6 DOF robotic arm mounted to a powered wheelchair, and was controlled at the joint level by an array of tongue-operated switches. Further discussions on the topic of the c ontrollability of the arm commented on both successes and failures the design and those successe s with the project can be attributed to


5 the important role that proprioceptive feedback plays in the control of extremities (Allen et al., 1972). These were pioneering research projects that often provided a framework for future development. This early work expanded the field of a ssistive robotics to the wide variety of devices found today. Different de sign goals and approaches to the problem have yielded many typs of robotic devices. For clarity, assistive robotics can be divided into several categories: 1. Workstation robots, which are for stationary, well-structured environments. 2. Mobile assistive robots, which travel about the room and have a manipulator arm. 3. Wheelchair mounted robotic arms (WMRAs) that mount a manipulator arm onto the individuals wheelchair to provide assistance throughout the day. 2.2 Workstation Based Systems The very first rehabilitation robotics app lications focused on using commercially available industrial manipulators and modifying them for rehabilitation applications. A factor limiting the use of industrial r obotic arms in rehabilitative robotics roles is the basic differences in operational require ments. Industrial arms are designed to work at high speed and accuracy in an envi ronment where there are no humans. For applications in a human intensive workspace, assistive robotic arms must be mechanically limited to low velocities and accelerations. Further, high stiffness and accuracy found in industrial robots is unnecessary in rehabilitation.


One workstation robot example is The Robotic Aid Project. This was an attempt to create a system for users with quadrip legia, by adapting an already commercially available industrial robotic arm. This was the integration of a PUMA 250 arm, microprocessor, multi-line monochrome disp lay and speech synthesis and recognition systems. The PUMA 250 is shown in Figure 2.1. Limitations with the speech recognition systems of the day and computati onal deficiencies limited the success of the program. The computational power of computer s of the day did not allow for real time inverse kinematics of the arms, which lim ited the arm to replaying preprogrammed actions. Individual joints of the arm could be manipulated but coordinated real time multi-joint maneuvers were impossible. Figure 2.1: Puma 250 Arm As more application-specific robotic arms and computers with increased computational power became available, arms with controllers could now be mounted onto mobile platforms. At first these systems were simple rolling bases, and later they increased in complexity and degrees of freedom to include powered mobile robots. 6


Handy-1 is a robotic arm mounted to a nonpowered wheeled base to assist in very specific activities of daily living (ADL). Handy-1 was developed in 1988 to provide persons with severe disabilities assistance at mealtimes. The unit is capable of providing assistance in personal hygiene, eating, and the application of make-up. During user trials women specifically asked if the unit would be capable of putting on cosmetic products. Shortly after the trials, the design was upgrad ed with a new tray and gripper accessory. Each task has a specific tray to accomplish its goal. Handy-1 is shown in Fig. 2.2 and is based on a 5-DOF, lightly modified industrial manipulator (Topping 1999). Figure 2.2: Handy-1 In the feeding mode, the operator controls the robot through an interface that uses lights that move across the available food tray s, and a button to select the item desired. Once the button is pressed the robot scoops up the food and brings it to a predetermined place near the operators mouth. Once the user has consumed the food, he presses another button, and the robot returns to the food selection mode. This assistive device does not eliminate the need for a personal assistant but allows for individuals to have an increased leve l of self-sufficiency. In user trials, almost invariably the users believed the device significantly increased their quality of life. 7


8 The Wessex robot (Bath Institute of Medical Engineering) is a trolley-mounted mobile robot with modified SCARA geometry. A SCARA arm has two revolute joints in the horizontal plane, allowing it to reach any point within a horizontal planar workspace defined by two concentric circles. In modifi ed SCARA configuration most of the joints operate in the horizontal plane. All vertical movement is achieved through the use of a single vertical actuator. The Wessex suffered from several shortcomings; one example was its limited reach making it unable to pick up items off the ground. The arm also had limited reach beyond the tray at the top of th e trolley. The trolley was not powered and was pushed into location by the daily assistant. In user trials the operator felt limited by its programmability and fact that the trolley was not powered. The user felt that if the trolley were able to be remote controlled it could be used to retrieve or manipulate objects within the same room. For example, th e operator could adjust the thermostat or retrieve a drink from an attached kitchen (Hillman and Gammie 1994). The RAID workstation shown in Fig. 2.3 was designed to be a workstation assistive robot system. It is comprised of a 6 DOF robotic arm mounted onto a linear track in a well-controlled environment (Da llaway 1992). In the figure the manipulator can be seen near the top of the shelf in the center of the cabinet.


Figure 2.3: RAID Workstation The RAID system benefits greatly from the formal structure provided by the workstation environment. This organization allows the manipulator arm to repeatedly move and acquire items needed by the operator using preprogrammed functions and routines. Another robotic arm under development is The Robotic Assistive Device is by the Neil Squire Foundation in Vancouver, Ca nada. The RAD is a 6 DOF workspace mountable manipulator that uses a serial port computer interface. The manipulator is controlled through a graphical user interf ace (GUI), utilizing icons to symbolize predefined tasks. The arm can be mounted on various surfaces and has good repeatability at 3mm, and relatively large payload cap acity of 4.3 kg. Most rehabilitation specific manipulators have maximum payloads of 2 kg or less (Squire 2004). 9


Figure 2.4: Robot Assistive Device A similar system is The ProVAR (Stanf ord, CA), which is based on a Puma 260 robotic arm, and is designed to operate in a vocational environment. The ProVAR manipulator shown in Fig. 2.5 is the next generation of the DeVAR system and expands upon the previous research by reducing opera ting costs and increasing overall usefulness (Katevas 2000). Figure 2.5: ProVAR System 10


The ProVAR uses a web-based virtual e nvironment to model the functionality of the manipulator. In this way the operator can examine potential arm movements for a given task and if the simulation is successful the action can be performed. The primary goals for ProVAR are more functionality per dollar, easier operator control, and higher system reliability compared with the previous generation of vocational assistive robots. 2.3 Mobile Systems The Mobile Vocational Assistant Robot (MoVAR), shown in Figure 2.6, utilized an omnidirectional mobile platform mounti ng a PUMA-250 robotic arm, remote viewing camera, force sensors and gripper proximity sensors. Figure 2.6: MoVAR MoVAID is an advanced version of th e MoVAR system design specifically for home use. MoVAID increases the effectivene ss of the previous model by applying the lessons learned in the laboratory to assist in common tasks around the home such as 11


cleaning and food preparation. MoVAID incor porates a variety of sensing devices both mounted to the manipulator and the base. In Figure 2.7, MoVAID can be seen along with various sensors located on the arm. Sensors m ounted to the first link of the arm include a pair of cameras used for stereo vision, and a laser localization system used in task execution. MoVAID also uses homing beacons placed around a room, to navigate. In addition to position detection, the unit also has ultrasonic detectors and an active bumper that disable the device should an impact occur. Figure 2.7: MoVAID The robotic arm is an 8 DOF arm, and the gripper has three fingers with two degrees of freedom. The grippe r was specifically designed as a prosthetic device that provides the manipulator with excellent de xterity. The increased dexterity provided by the gripper over more traditional end effectors allows MoVAID to be more effective in the unstructured environment of a home. 12


2.4 Integrated Robotic Systems Further integration of robotic arms and ot her sensors has led to some increasingly capable designs. Although still in development, they offer even greater potential as assistive devices. For example, the FRIEND system is a Manus arm integrated with stereovision and dedicated computer contro l and specialized software. Beside the standard programming methods the FRIEND sy stem, shown in Fig. 2.8, is capable of being programmed via a haptic interface glove. The haptic glove allows the operator / programmer to feel what the robot feels thr ough feedback to the user. A Haptic glove is donned, and the action, such as pouring a glass, is completed and then stored into the computer for future use. The action can then be replayed as a user function. The user may also control the arm verbally using an inte grated voice recognition system (Borgerding et al.). Figure 2.8: FRIEND Robotic System 13


Another example of an integrated system is the TAURO. This system uses offthe-shelf components such as a power wh eelchair, ultrasonic sensors, camera and computers. TAURO is a mobile service robot being developed for inspection, stocktaking and documentation tasks in indoor environm ents. The TAURO system integrates the movement of the wheelchair and the operation of the manipulator. In this way if the goal is out of reach, of the manipulator the wh eelchair will move on a path toward the goal until the manipulator is within reach. This coor dinated control is a significant advance in the use of WMRA. Although not specifically designed for rehabilitation robotics, it would be readily adaptable to the task. The TAURO system can be seen in Figure 2.9. Figure 2.9: TAURO Robotic System 2.5 Research WMRAs Wheelchair mounted robotic arms (WMRAs) combine the idea of a workstation and a mobile robot. WMRAs mount a manipulat or arm onto a power wheelchair. In the past, manipulators have been so large and heavy as to hinder the operators ability to maneuver the chair. 14


15 Currently there are two production wh eelchair mounted robotic arms: The Manus, manufactured by Exact Dynamics; and the Raptor, manufactured by Applied Resources. Other WMRAs under developmen t are the Helping Hand System, Weston Arm from the United Kingdom, and the Asimov from Sweden. Under development is The Helping Hand system (Kinetic Rehabilitation Instruments, Hanover Massachusetts), which is a 5 DOF robotic arm. Its design is modular in nature and can be mounted to the side of a power wheelchair. It is controlled at the joint level via switches controlling each individual motor (Sheredos et al. 1995). Another arm under development is The We ston robotic arm (Bath Institute of Medical Engineering), shown in Figure 2.10. This is the continuation of the trolley mounted Wessex robot. It uses a vertical actuator mounted to a wheelchair with the main rotary joints (shoulder, elbow, and wrist), constrained to move in the horizontal plane. This arm approaches the design rather differently than others. The first joint of the arm (the shoulder) is prismatic, which actua tes in a sliding motion along a track. This necessarily makes The Weston arm larger th an both the Manus and the Raptor designs. The other joints of the arm utilize a modifi ed SCARA design as described in the Wessex manipulator.


Figure 2.10: Weston Arm Another arm currently under development is the Asimov (Bolmsjo et al.). The Asimov is a modular design, with its cont rol system and motors are distributed throughout the arm. A computer rendering of the Asimov is shown in Figure 2.11. The modularity of the design allows for multiple mounting locations and various workspace geometries. This approach of modularity shows great promise in creating one robotic system that can be used in both mobile or workstation environments. Asimov models have been shown with all three possible m ounting positions: front, side and rear. Without physical models to test the efficacy of the design it is unknown how well the design would integrate into real world applications. 16


Figure 2.11: Asimov Arm 2.6 Commercially Available WMRAs 2.6.1 The Manus The Manus manipulator arm is fully de terministic manipulator: It can be programmed in a manner comparable to industrial robotic manipulators. The Manus has been under development since the mid 1980s and entered into production in the early 1990s. A picture of Manus mounted onto a Permobil Max90 wheelchair is shown in Figure 2.12. It is a 6 DOF arm, with sevomotors all housed in a cylindrical base. 17


Figure 2.12: Manus Arm 2.6.2 Raptor Another production WMRA is the Raptor [Applied Resources, Inc.], which mounts the robotic arm to the right side of the wheelchair. This manipulator has four degrees of freedom plus a planar gripper and can be seen mounted to a power wheelchair in Figure 2.13. The user directly controls the arm with either a joystick or 10-button controller. The Raptor uses an 8 position joys tick-like input device that is mounted to the armrest of the wheelchair. Typically, the joys tick that controls the manipulator arm is located on the armrest opposite to the input devi ce that controls the steering of the power wheelchair. Because the Raptor does not have encoders, the manipulator cannot be preprogrammed in the fashion of industrial robot s. This compromise was done to minimize overall system cost and make the product more readily available to the public. The simplicity the Raptor arm allows it to cost ha lf that of the MANUS arm given the current exchange rate. 18


Figure 2.13: Raptor Arm The Raptor uses an 8 position joystick-lik e input device that is mounted to the armrest of the wheelchair. Typically, the joys tick that controls the manipulator arm is located on the armrest opposite to the input devi ce that controls the steering of the power wheelchair. 2.7 WMRA Moutning Postitions 19 All power wheelchairs have different st ructural designs. There are several possible mounting locations for a WMRA (War ner and Prior 1994). The mount may be in the front, side or rear of the wheelchair. T hus, there are several possible ways to mount an assistive robotic arm. In order to mount a robotic arm to a power wheelchair, several design considerations must be met. Foremost is the safety of the operator. The mount must be sturdy and rigid and not compromise the structural integrity or the functionality of the chair in any way. Next, the robotic arm must be mounted in such a way that it does not excessively increase the width of the wh eelchair. Often, powere wheel chairs are near the maximum width that allows access through standard doors, etc. For some


20 mobility devices that have the frame hi dden under a fairing, mounting may be more difficult. Some mounting positions may not be possible with all commercially available power wheelchairs and custom brackets may need to be fabricated to facilitate mounting. All of these factors must be considered. 2.7.1 Rear Mount One of the potential benef its of a rear-mounted arm is that it will not increase the width of the wheelchair when not in use. Assu ming that the arm can be able to be stowed behind the wheelchair, the arm would not creat e a distraction for individuals interacting with the person. Additionally a rear-mounted arm would not be a physical obstruction during transfer into and out of the wheelchair. Rear mount units suffer from placement issues. Due to excessive link lengths required to design a robotic arm with a dor sal (rear) mounting there are higher torques and loads on the bearings that further increase weight and size. Manipulation in front of the chair is also reduced. One possibility is to have a support provided for the arm, on the side near the front of the wheelchair. The arm would be swung around from the stowed position, and then locked into a rigid support. This would combine the convenience of a side mount with the stowage capability of a rear mount. At this time there are no commercially av ailable WMRAs that are mounted to the rear of the wheelchair. It should be noted that there is an optional rear-mounting bracket available for the Raptor but this eliminates most of the ability of the arm to reach directly in front of the chair.


21 2.7.2 Side Mount Like front mounted manipulators, side mount units also have deficiencies such as increasing the width of the power wheelchair. With the side mount located lower than the armrest (under the wheelchair) the arm will always add at least the width of the first link to the width of the wheelchair. This makes it even more difficult to for the operator to maneuver through doorways and tight hallway s. This exacerbates mobility problems already encountered with power wheelchair us ers. The side mount also requires longer link lengths to allow for manipulation of object s in front of the power wheelchair. These increased link lengths require larger and more powerful motors and gear-heads to move and stabilize the links actuation. These factors often increase the weight and cost of designing arms for this application. The Raptor is a side-mounted arm. The pr imary joint motor of the robotic arm is an exposed gear motor, and it must be m ounted onto the frame of the wheelchair under the seat. The motor is slightly in front of th e operators lap and the first rotational axis is horizontal, oriented laterally to the wheelchair (i.e. parallel with the drive wheel axles). The side mount is mostly hidden underneath th e chair and when the arm is not it use and when stowed, the Raptor arm can be relativ ely innocuous. However, when the arm is retracted and not in use, the Raptor is be low the operators waist level and is fairly unobtrusive. 2.7.3 Front Mount The front mount offers greater access to the operators immediate working environment. The lap, tray top on arm rests, and the mouth location can all be considered the immediate environment of the operator. Manipulating objects in these areas is


22 optimized with this mounting location. Due to the high mounting point the front mount near the knee allows for good access to high obj ects such as items on shelves or operating doors on high cabinets. Objects in front of the chair are also readily manipulated. The Manus utilizes a front mounting locati on to the left of the operators left knee. The first joint of the arm rotates about the z-axis (floor to ceiling) and is located approximately 5 cm above the level of the ar mrest of the power wheelchair. This location allows for ready manipulation of objects that are above the plane of the wheelchair seat, and most importantly the operators face and lap. However, the front mount style also ha s limitations. The first is the visual distraction of having a large piece of technol ogy between the operator and those they are interacting with. This was noted as a hindr ance in long-term Manus trials (Eftring and Boschian 1999). The mounting location also limite d the ability of the operator to put their legs under desks, tables, and sinks in clinical evaluations. Also, front mount limits access to tables and other furniture that requires driving the legs of the individual under the object. Because there are many standards that ha ve been set forth to allow individuals in mobility assist devices to maneuver close to desks and sinks this is a significant limitation. Another complaint from surveyed us ers was that, even when fully retracted, a front-mount arm inhibits the user from being able to move close to a table or a sink. Finally, users have also commented that th e front mounting makes the manipulator arm obtrusive and can create uncomfortable social tensions with people unfamiliar with robotic technology.


23 2.8 Control System Types WMRAs, like any robotic system, have two types of control system options, closed loop or open loop. Clos ed loop control is more capable, but open loop control is less expensive. For robotic devices intended to be affordable in the consumer market, the decision to use an open or closed loop scheme is non-trivial. 2.8.1 Closed Loop Control Closed loop systems are in common use in industrial robotics applications. These closed loop systems permit accurate repeated motions of robotic manipulators. Most specifically these systems are most effective in the structured environments such as rehabilitation workstations. These rehabilitation workstations mimic the habitat originally designed for the industrial robot the manufact uring cell. These are highly structured environments, which permit high productivity due to eliminating positioning variances. These systems are very useful in reha bilitation robotics applications by allowing preprogrammed actions and gestures. Preprogr ammed gestures can be as simple or as complicated as required such as reaching for a light switch or eating and drinking. Closed loop control also allows furthe r integration of the arm into more complicated and intelligent systems that can a ssist the operator. These assist functions may include stereo vision, object recognition, target distance determination, etc. The MANUS system is a version of a closed l oop system. A joystick and a keypad control the manipulator. The joystick used to manually ope rate the manipulator is shown in Figure 2.14. Manus can also carry out coordinated control of multiple joints with preprogrammed gestures using the 16-butt on keypad shown in Figure 2.15. Gestures can be taught to the Manus and stored for future use via the keypad. With the use of the two


input devices the operator can run preprogr ammed routines or directly control the manipulator in real time. The controller conve rts the inputs from a haptic interface into a signal that directly controls the robotic ar m. There may be a direct or indirect link between the input device and the output signal. This may be a simple proportional control or more complex method where input position is converted into arm velocity output. The downside to closed loop systems is th eir higher initial cost. The drives for the links must have encoders or some other form of feedback to the controller. Often the increased productivity, programmability, and system interoperability can compensate for this increased cost by offering more bang for the buck. Figure 2.14: Manus Joystick Controller Figure 2.15: Manus Keyboard Controller 2.8.2 Open Loop Control An open loop controller places all error correction responsibility on the human operator. The operator continuously directs th e arm into its final position. This type of system is inherently tolerant of positioning e rrors from a variety of causes. These errors may be inherent in the robotic device such as play in the motors, gears, bearings or 24


25 compliance within the links due to loading or environmental conditions such as thermal effects, wind, and movement of the base with respect to the reference frame. The ability of the open loop controller to to lerate and correct for various types of error is because the operator continuously upda tes its position correcting any errors that may occur during the manipulation. The operato r actually considers the sum of all the errors and moves the arm according to the act ual perceived position of the end effector and not what the arms internal sensors are telling the operator. Because computer-controlled, coordinate d motion is not possible, motion is limited to one joint at a time. Open l oop control thus requires higher levels of concentration and eye hand coordination than other forms of control, which may be programmed or assisted. This is more taxing for the operator and this fatigue can limit the use of the assistive robotic device. Because of the human in the operational loop an open loop system is unable to make precisely reproducible motions. These cost-saving measures may not be justified in light of the reduced performance of the end product.


26 Chapter Three: Design 3.1 Design Goals An entirely new WMRA has been developed at the University of South Florida. The goal was to produce an arm that has bette r manipulability, greater payload, and easier control than current designs. The arm is al so reconfigurable, which increases the number of applications, and returns more benefit from the engineering investment. 3.1.1 Mechanical Constraints Weight In a mobile application, minimal weight is of primary importance. Power wheelchairs have a rated payload, and a hea vy arm reduces the payload available for the user. Our goal was to have a total system mass under 14 kg, including the arm, controller, and all peripherals. Mounting Location As found in our previous research xviii side mounting is preferable overall because it provides the best balance between manipul ability and unobtrusiveness. However, care must be taken to prevent widening of the pow er chair. The new arm is mounted as far


27 forward and upward as possible while still in a side mount configuration, and only increases chair width by 7.5cm. This mounti ng location allows the arm to be stowed by folding it back and then wrapping the forearm behind the seat. It virtually disappears when not in use, especially when the arm is painted to match the chair. This is good because most users want the assistance, without the stigma that these devices often bring. The arm must be slightly longer than with a forward mount, requiring greater shoulder joint torque and heavier gearboxes. This is compensated by the inherent efficiency of harmonic gearheads used in th e drivetrain, allowing greater payload at less weight than the MANUS. Stiffness This is one of the greatest differen ces between our WMRA and a typical industrial manipulator. As we anticipate te leoperation will be the most common use for the robot, great precision is not required. With a human participating at all times, inaccuracy due to a compliant structure is easily and transparently corrected. Recognizing this allowed the structure to be made much lighter than an industrial manipulator with the same payload. However, the low stiffness and large backlash of other WMRAs is an impediment to accurate control. With this design, we attempted to find an optimal balance, and arrived at a st ructure stiffer than other WMRAs, but less stiff than an industrial manipulator. Payload This manipulator is intended for use in Activities of Daily Living, and for job tasks typical of an office environment. As such, it is important that the arm be strong enough to move objects that are common in these environments. A gallon jug of milk is


28 a good upper limit for a typical around-the-house obj ect that must be manipulated. As this is an approximately 4 kg mass, this was set as the baseline payload for the arm at full horizontal reach. Then, an extra margin of 2 kg was added to allow for a choice of end effector that would also be capab le of this load. After all, what use is a strong arm with a weak hand? The 4 kg useful payload is signi ficantly larger than the 1 kg payload of the Raptor. Reconfigurability Even though a side mount was chosen for this prototype, it is important to note that the basic design can be adapted to a front or rear wheelchair mount, or a fixed workstation mount. The arm can be speciali zed for these workspaces by adjusting link lengths. Longer lengths can be specified for a rear mount on a power chair, but this will necessarily reduce payload and reduce ma nipulability in front of the chair. Reconfigurability places a strong constraint on the drivetrain type, to be discussed in section 3.2. Power Supply and Consumption In the power wheelchair industry, a 24-volt lead-acid battery pack is standard, and is the natural choice for the power supply of a WMRA. All motors, controllers, input devices, sensors and so on must be able to work with 24vdc, or through a voltage regulator at under 24vdc. Typically, two Group 24 gel cell lead-acid batteries are used, providing roughly 73 amp-hours of capacity. Energy consumption is important as well. A power chair is expected to run all day on a single charge, and users would reject an arm that worked well but left them stranded!


29 Therefore, efficient power electronics, motors and drivetrain were chosen to keep power consumption low. Merely powering the control electronics takes 0.35A, but a motor may use up to 4A during heavy use. While holding position, consumption ranges from 0.5A with no payload, to 1.7A with a 6kg load and the arm fully outstretched. This is because no brakes are used, and current must be app lied to hold position. Average use will depend on application, but for typical household and o ffice work this will be roughly 2A. This draw on a 73Ah battery would allow 37.5 hour s of continuous operation, assuming only the arm was used. A typical day with 6 hours of arm use would consume 12Ah, leaving 61Ah, or 84% of battery capacity for the trac tion motors. This would reduce a typical 30 km driving range to 25 km negligible for most users. Intensive arm use is likely to be in one location, and charging while the arm is in use is an option to extend battery life. 3.1.2 Cost Constraints Of course, cost and ease of manufactur e have been considered from the beginning, and the new WMRA has to exceed the performance of current WMRAs without increasing cost. We feel that cost has not been the major hurdle to widespread adoption of these devices, but rather poor utility and difficulty of use. The target was to come between the Raptor and Manus system s in terms of cost, while exceeding the performance of both. In hard numbers, we e xpect that this system can be produced and sold profitably at 30,000 USD retail. Deta ils may be found in Appendix A: Cost Estimate.


30 3.1.3 User Requirements People want a useful payload, and a simple intuitive control. A major drawback of the Raptor system is the single-joint, non cartesian controller. Raptor lacks encoders and therefore control is manual, one joint at a time. Quadrature encoders are a costeffective way to provide closed-loop control. The controllers of the new WMRA have PWM voltage regulation, and have builtin support for acceleration limits. The controllers communicate with the host PC over a RS-485 serial link, which is daisy chain connected to each one. The system easily scales to control grippers or even the base wheelchair, all through one standard control system. Extra degrees of freedom are a sure wa y to improve manipulability. This is evidenced by the considerable increase going from Raptors 4 DOF to the 6 DOF of MANUS. Our new design incorporates 7 join ts, allowing full 6 DOF pose control even in difficult regions of the workspace, such as reaching around the wheelchair, or up to a high shelf. Reconfigurable arm lengths allow greater leverage on the engineering input, as a single basic design may be adapted to numerous applications. This is only practical with electric drive and actuator placement directly at each joint. The MANUS, for example, houses all drive motors in the base and us es a complex drivetrain involving gears and synchronous belts to drive the joints and grippe r. Reconfiguration in this context means a complete redesign. The new USF WMRA de sign requires only a few hundred dollars in parts and an hour of a technicians time to reconfigure it according to the users needs and the desired uses of the arm.


31 3.2 Types of Systems Considered Some ideas were more seriously considered than others, but before beginning design we spent quite a while researching po ssible ways to actuate and read the position of our arm. Actuation: 1. Stepper motors with gearboxes at each joint 2. Steppers with screw jacks 3. DC servos, gearboxes, directly acting on joints 4. DC servos with screw jacks 5. Servo or stepper motors at base, driving gearboxes or screws using flexible drive shaft 6. Hydraulic pump and electric valves in base, cylinders on arm 7. Same but pneumatic would require electric brakes 8. Master/slave hydraulic system, driven by electric motors 9. Muscle wire Most actuation alternatives were restricted due to our requirement for reconfigurability. Imagine changing the length of an arm that is driven through linkages or flex cables from motors in the base. So many parts would have to change, it would be a whole new design. Muscle wire was rejected because it is weak, slow and inefficient; pneumatics was thrown out due to positioning difficulty and compressor noise. We decided to drive the joints electrically through harmonic gearheads, with the entire actuator positioned at each joint.


32 The only serious choice was whether to use stepper or servo motors. Due to recent improvements in servo controllers, the co st of this option is not much higher than for stepper motors. Brush DC servomotors allow closed-loop control, and are much quieter, lighter and more efficient than ste ppers. For these reasons, DC Servo drive was selected. Position sensing: 1. Limit switches to prevent damage 2. Steppers with limit switches for initialization 3. Potentiometers at joints 4. Relative optical encoders at motors, limit switches 5. Absolute optical encoders at joints 6. LVDT on inboard or outboard hydraulic cylinder or on screw jacks Options 1 and 2 do not allow servo cont rol and were rejected. Option 3 was considered, but potentiometers are electrically noisy and have a poor life span. Absolute encoders are very attractive because they do not require an initialization routine. However, for the required resolution they are rather expensive, adding $2000 or more to the overall robot cost. Mounting at the joint is required, and is more difficult than at the motor. LVDTs and resolvers were considered as well, but they are analog devices not supported by our controllers. They are also more expensive than quadrature encoders. Quadrature encoders, mounted on the motors, were selected for their ease of integration, accuracy, simplicity and low cost. Optical limit switches ease initialization upon power-up. These encoders are also directly supported by the controller hardware


we selected, unlike many other sensor type s. These are the most common feedback devices in servomotor robots. It is no coin cidence that Pittman manufactures motors with quadrature encoders built-in. 3.3 Final Design 3.3.1 Kinematic Arrangement The arm is a 7-DOF design, using 7 revolute joints. Revolute joints were chosen over prismatic and other types because of their better packaging and mechanical simplicity. The basic layout is anthropomorphic, with joints 1, 2 and 3 acting as a shoulder, joint 4 as an elbow, and joints 5, 6 and 7 as a wrist. Th e 3 DOF shoulder allows the elbow to be positioned anywhere along a s pherical surface, whereas with the Raptor arm, elbow movement is limited to a fore-aft circle. Figure 3.1: Complete SolidWorks Model of the USF WMRA 33


Throughout the arm, adjacent joint axes are oriented at 90 degrees. This helps to meet two goals: Mechanical design simplicity and kinematic simplicity. Machining parts on a conventional milling machine is easier with right angles. And the coordinate transform equations simplify greatly, with si nes and cosines of these angles becoming ones and zeroes (especially the zeroes are appreci ated!). All adjacent joint axes intersect, also simplifying the kinematics. There was a choice to be made in the wrist kinematics. While 3 degrees of freedom are certainly required here for maximum manipulability, there were two primary ways to arrange this. One is with each successive joint oriented at 90 degrees. The other is to place the middle joint at 45 degrees to the others. The advantage of this nonorthogonal layout is that it can help redu ce difficulty due to singularities in the equations. However, the packaging of this layout was quite unattractive, and a much more aesthetically pleasing layout was developed, helped by a 90-degree gearbox. This elegantly places the joint 6 motor inside the forearm tube, rather than protruding out the side of the forearm. Section 3.3.5, Wrist Design, describes the wrist in more detail. 34 Figure 3.2: Kinematic Diagram, with Link Frame Assignments


35 Table 3.1 DH Parameters for the USF WMRA i i-1 (degrees) a i (mm) d i (mm) i 1 0 0 0 1 2 90 0 146 2 3 -90 0 549 3 4 90 0 130 4 5 -90 0 241 5 6 90 0 0 6 7 -90 0 179 7 3.3.2 Component Selection Emphasis was placed on using off-the-shelf parts wherever possible. The basic arrangement for each joint is a high-reduction gearhead, a motor with encoder and spurgear reduction, and a bracket that holds th ese two parts and attaches to the two neighboring links. Gearhead Selection Next the question was which gearboxes to use. For joint 1, in the shoulder, the required torque output is roughly 100 Nm. As our servomotors on joints 1 through 4 have only 1.2 Nm output after their built-in gearboxes, reduction of nearly 100:1 is required. Planetary gears were considered, but the desired torque and reduction required a large, 180mm long gearbox. This would pose a significant packaging problem. Harmonic drive gearheads were chosen because they can achieve 100:1 reduction in a single stage, with only 64mm axial length. In addition, they have bearings suitable for supporting overhung loads, enabling the next arm segment to be bolted directly to the


output flange of the gearhead. This greatly simplifies the design, reducing weight and cost through lower part count. Figure 3.3: HD Systems Harmonic Drive Gearhead Gearheads were chosen based on require d overhung loads and torques, with the size of gearhead gradually reducing in each more distal joint. This is not a closed-form problem, because the weight of one gearhead affects the torque required of the more proximal joints. Once the basic type of gearhead was selected, information on the available sizes was collected, namely the mass and recommended maximum torque. Maximum recommended torque here was taken to be the lesser of two specifications from the manufacturer: Maximum output to rque and maximum overhung torque. A simple spreadsheet model of a horizontally outstretched arm was made, which accounted 36


37 for link lengths and self-weight of the structure. The target payload (taken here to mean the end effector and object grasped totaling 6 kg) was also applied to the end of the arm. For each joint, the torque due to gravity acting on the more distal joints was applied. For instance, joint 4 was subjected to the torque from the weight of joints 6 and 7, plus the payload. The weight of a joint was taken as the sum of the gearhead, motor with encoder, an aluminum bracket at 500g, a nd the link tube attached to it. The link lengths were specified, and the spreadsheet gave the required gearhead size that would meet the torque applied to it. An exampl e spreadheet, showing torque estimates, is included in Appendix C: Joint Torque Calculations. The goal was to find an optimal selection of gearheads that would meet payload and reach requirements, with minimum tota l arm weight. This model allowed many design iterations to be quickly evaluated, on ce the spreadsheet was set up. Because the maximum torque increases stepwise with one size larger gearhead, it was found that some combinations were much more efficient in terms of payload/structure mass ratios. Eventually a combination was found that me t all requirements and had evenly stressed components. The selected gearheads are shown in Table 3.2. Table 3.2: HD Systems Gearhead Selections for Each Joint Joint Model Selected Torque (N m) OD (mm) Mass (kg) 1 CSF-25 140 107 1.50 2 CSF-25 140 107 1.50 3 CSF-20 70 93 0.98 4 CSF-17 46 79 0.68 5 CSF-17 46 79 0.68 6 CSF-14 19.5 73 0.52 7 CSF-11 6.6 58 0.15


38 Motor Selection Brush DC motors were chosen because they are the least expensive way to achieve servo control. While brushless moto rs are a future possibility, performance gains are dubious, and would increase the cost of the robot by roughly $1000. The marginal increase in efficiency is relatively unimporta nt, and gear train noise is already greater than commutator noise. The main benefits for brushless motors are increased service life before maintenance, and possibly better packaging. We maintain that Brush DC servo drive is the best overall compromise for a WMRA. Once maximum joint torques were known, a nd targets were set for joint speeds, and the gearhead ratios were selected, motor selection could begin. The goal here was to minimize weight and bulk, while meeting performance specifications, and without incurring undue cost. Pittman motors were se lected that are off-the-shelf, meet all performance criteria, and have integrated g earboxes and encoders. Joints 1 through 4 use Pittman model GM9234C212-R3. While the elbow (joint 4) has a lower torque demand than joints 1 through 3, the same motor was used to reduce part inventory required. As much less torque is required at the wrist, smaller gear motors are used to reduce weight. Pittman model GM8724S009 motors actuate join ts 5 and 7, and a similar motor, model 8322S003, drives joint 6. For good packaging, th e gearhead on joint 6 is driven through a precision right-angle gearbox, allowing the moto r to be hidden inside the link tube. Since the right angle gearbox has a reducti on of 5:1, the motor does not have an integrated gearbox. All 7 motors are designed to operate on 0 24 VDC. The larger motors stall at about 4 amps, which is the limit of our controllers but still safe (the controllers


automatically limit current to prevent damage ). The duty cycle at full current is only 25%, but tests have shown this to be accepta ble even during extended use the motors barely rise above room temperature. This is because full rated power is only rarely required in normal use. Figure 3.4: Pittman PMDC Brush Motor with Gearbox and Encoder Table 3.3: Motors Used in USF WMRA Motor Type Applied to Weight (g) Speed (RPM) Continuous Torque (Nm) Stall Torque (Nm) GM9234C212-R3 Joints 1-4 505 900 0.431 2.147 8322S003 Joint 6 218 7850 0.011 0.052 GM8724S009 Joints 5 & 7 316 1400 0.102 0.297 Encoders Quadrature encoders on the motors provide relative motion information. The arm is initialized using optical limit switches mount ed at the output side of each gearhead. Pittman produces motors with integrated enc oders, and these were used to reduce cost and design complexity. One note: with 500 count/revolution encoders and 600:1 39


reduction between the motor and gearhead output, 300000 counts per output flange revolution are recorded. This is excessive for the application, but does not cause any ill effects, and is useful for accurate veloc ity and acceleration measurement. The PICSERVO boards read the encoders directly and only report position back to the main controller when queried, so serial bus traffic is unaffected by the high encoder resolution. Controllers If there was any doubt that DC servo actuation was the right choice, the PICSERVO controller removed it. At 5cm x 7.5cm, this unit has a small microprocessor that drives the built-in amplifier with a PWM signal, handles PID position and velocity control, communicates over a simple RS-485 se rial link, and can be daisy-chained up to 32 units. It can also read quadrature encode rs, limit switches, an 8 bit analog input, and supports coordinated motion control. It is a bargain at just $150 per controller. Figure 3.5: J.R. Kerr PIC-SERVO Controller Board 40


41 Here are the basic specifications for this motor controller: 1. PIC-SERVO SC Motion Control Board 2. Part Number: KAE-T0V10-BDV1 3. Motor Type: DC Servo Motor (brush-type) 4. Driver Ratings: 3A cont./6A peak, 12-48vdc 5. 32-bit position, velocity and acceleration control 6. Trapezoidal and velocity profiling permit on-the-fly parameter changes 7. 16 bit PID servo gains can be changed on-the-fly 8. Multiaxis coordinated motion control support 9. 2 or 3 channel encoder input, limit switch inputs, hall sensor inputs 10. Optional Step and Direction inputs 11. Amplifier includes overcurrent, overvoltage, undervoltage and thermal overload protection 12. May also be used with external amplifiers 13. 4-wire RS485 communications interface can be connect to additional controllers (up to 32 total) 14. Nominal size: 5cm x 7.5cm These controllers handle all of the necessary low-level tasks, freeing up resources on the main computer and also preventing a bo ttleneck in the serial interface. Software development was eased by the carefully doc umented example code included with the controllers.


42 3.3.3 Material Selection 6061 Aluminum was chosen for the join t brackets because of machinability, weldability, relatively low cost, good strength-to weight ratio, and availability. This material was also chosen for the link tubes, for the same reasons. Steel was considered but rejected due to its high density. In many places, the thickness of a bracket is not determined by strength or stiffness, but by si mple packaging constraints. Steel would unacceptably increase mass in these areas. Composites were considered for the li nk tubes. Especially carbon fiber/epoxy was investigated, due to the increase in s tiffness and reduction in weight possible. Aluminum was ultimately chosen, although payl oad could be increased by 0.5 kg or more using carbon/epoxy. Perhaps this could be an upgrade option in a production arm, as the link tubes are easily changed out. Carbon fibe r becomes especially attractive for a longreach option, and may make a rear-mount arm mo re feasible. This is an area we will explore in our future development of this arm. 3.3.4 Joint Design Once all components were selected, design of each joint was rather straightforward. The typical arrangement for a joint is to have a gearhead and motor held together by an angle bracket. This bracket mounts to the previous joint or link. The output flange of the gearhead attaches to the next joint bracket or link.


Figure 3.6: Typical Joint Design, Showing Motor, Gearhead and Bracket Joints 1 4 were designed this way, and produced from single blocks of 6061 aluminum. Billet construction was chosen fo r its high strength-to-weight ratio and high dimensional accuracy. 3.3.5 Wrist Design As noted before, there were two basic choices for a 3-DOF wrist: Orthogonal and nonorthogonal. The following renderings show each type of wrist. For clarity, obscuring brackets have been omitted. 43


Figure 3.7: Nonorthogonal Wrist Concept Figure 3.8: 3-DOF Orthogonal Wrist Concept 44


45 The conventional orthogonal arrangement wa s selected due to better packaging. All three axes are mutually orthogonal and all axes converge at a single point. This is common in industrial manipulators, such as the Puma 560. It is done to simplify the kinematic equations and guarantee a closed-f orm inverse kinematic solution. As this manipulator is intended for use on a wheel chair, processor power is limited and a numerical inverse-kinematics routine would be unacceptable. The brackets for the wrist (joints 5, 6 and 7) were designed to be fabricated from machined plates, which reduces production time a nd cost. Joint 5 is much like the rest of the arm, with an angle bracket holding the motor and gearhead at a right angle to the output flange of Joint 4. Joint 6 has a design unlike the others in this manipulator. A right angle gearbox between the motor and gearhead greatly improves packaging, but does increase complexity of design. A single bracket wa s designed to hold all 3 parts in proper alignment, and to carry the load to the link tube and joint 5. Joint 7 is coaxial with the last link, so that no matter the pose of the arm, rotation about this axis is assured. The gearhead mount s to a flange welded to the end of the link tube, and the motor is hidden inside this tube. Again, this was done to improve appearance of the arm. 3.3.6 Control System While the details of the high-level contro l system is outside the scope of this design project, it is appropriate to discuss here the provisions made for such a system. This is a fully deterministic manipulator arm. Each joint controller is individually addressable, and can be controlled in positi on, velocity, or current (torque) mode. In


46 position mode, velocity and acceleration limits may be specified for smooth operation. These controllers automatically track position and velocity data; the central computer need only query each controller when necessary. This greatly reduces bandwidth required. Data for the entire arm is interfaced to the main computer using a single serial link. The PIC-Servo controllers use RS-485, a nd a hardware converter interfaces this with the RS-232 port on our host PC. The hos t PC right now is an older IBM laptop, running Windows 2000. However, the communications protocol is simple and open, and could be adapted to virtually any hardware/s oftware platform with an RS-232 port. We now also have an Rs-485/USB 1 adapter, allo wing this arm to be used on any PC with a USB port. Some of my ideas for future development of the control system are included in the Future Work section of Chapter Six. 3.3.7 Final Design Overview Figure 3.7 shows the complete assembly mode l of this arm. It is shown in a pose typical of a right-hand side mount, although either side mount is possible without mechanical modification. Motor covers have b een left off to show more design details. And of course, the gripper shown is only repres entative; any of a wide range of grippers can be used. We have a BarrettHand BH-8 th at can be mounted to various manipulators in our lab, including this one.


Figure 3.9: Complete SolidWorks Model of the USF WMRA 47


48 Chapter Four: Construction 4.1 Considerations for Manufacture This WMRA was entirely built by graduate and undergraduate students working for the Rehabilitation Engineering and Technol ogy Program at USF. Due to this, it was designed to be made with the equipment av ailable in our robotics machine shop. All machining was done with a conventional milling machine (with a rotary table) and with a conventional lathe. All welding was done on our Hobart TIG welder. While some parts could be made to look a little fancier with CNC equipment, we felt that having all production done in-house was much more valuable Especially for a prototype such as this, having a close-knit design/build team speeds production. Inevitable problems are quickly recognized and corrected, whereas in a typical over-the-wall engineering production environment such errors can cost days. In addition, simple manufacturing techniques will help to reduce production cost in the future.


Figure 4.1: Undergraduate Research Assistan t Andrew Bridges Milling a Joint Bracket 4.2 Completed WMRA 49 Figure 4.2: Completed Arm on Power Chair


50 Chapter Five: Testing 5.1 Safety Tests Of course, safety is a primary concern w ith any product, but this is especially the case for a WMRA, as we may assume that the user is unable to move out of the way of the manipulator. A balancing act is necessar y, because the arm must be slow and weak for safety, but not so much that users rej ect it. Fortunately, WMRAs do not have to operate at the high speeds and accelerations of industrial manipulators. Here we outline some simple safety testing done on our prototype. One feature of the PIC-Servo controllers is a software-selectable current limit. As current is proportional to motor torque, this is a simple and effective way to limit the force that may be accidentally applied to the us er. Some simple tests were done to see if the controllers responded quickly enough to avoid harm to the user. The arm was intentionally run at full speed, with the curre nt limit set to maximum, directly into the body of a volunteer (the head researcher on this project).


Figure 5.1: Automatic Shutdown when Force Limit is Exceeded The tests did not cause any damage to th e user or to the arm. However, some discomfort was experienced. One suggestion that can be implemented in the control system is a virtual safety bubble around the user, inside of which the maximum speed and force of the arm are limited. Maximum jo int torque is only required when reaching straight out, far away from the user. This safety improvement would therefore cause no noticeable decrease in performance. Another control possibility is to have joint torque limits set lower than maximum all the time. When a larger force is required, the GUI would prompt the user with an Are You Sure? message. This would help prevent unintentional use of the full force of the manipulator. It should be noted that these safety test s are not meant to certify this robot for any purpose other than research. The intent here is to merely get some estimate of the risk involved in development and use of the robot. Much more rigorous testing will be done later. 51


52 5.2 Stiffness Testing The stiffness required of this manipulator is much less than for an industrial manipulator. This is because teleoperation is the normal control mode, and the working environment is unstructured anyway. The us er easily corrects any compliance errors in the arm. However, too much compliance woul d annoy the user. Good stiffness leads to a feeling of quality construction. Stiffness was tested by extending the arm straight out in front of the wheelchair. A dial indicator was set to measure deflecti on in the vertical direction, and then a known mass was applied to wrist plate at the end of the arm. Deflections were measured at the wrist plate (100.3 cm from axis 1), joint 4 (50.8 cm from joint 1) and directly on the joint 1 gearhead. These deflections are shown in Table 5.1: Table 5.1: Arm Deflections vs. Applied Load Load (kg) Wrist Deflection (mm) Elbow Deflection (mm) Joint 1 Deflection (mm) 2 4.4 1.8 0.2 4 8.7 3.7 0.4 6 13.3 5.5 0.7 Deflection is essentially linear with app lied load. While not noted in the table, these deflections are recovered upon removal of the load, to within 0.1mm.


Figure 5.2: Arm Stiffness Measurement Backlash is another matter. More so than excessive compliance, backlash can make a device feel shoddy. Fortunately, Harm onic Drive gearheads have virtually zero backlash, as demonstrated during testing. For the application, stiffness and backlash values are excellent. Compare this to the Raptor arm, which at the end effector has +/50mm of play in all directions. 53


5.3 Strength Testing Each joint was individually tested for th e maximum load it could lift. This was done by placing the arm in a pose most adverse for the joint in question. For example, the arm was placed fully outstretched, poi nting forward parallel with the ground. Weights were progressively added, and the join t was given full power to try to raise the weights. Figure 5.3: Strength Testing of Joint 1 54


55 Table 5.2: Maximum Joint Loads Joint Max Load (kg) Note 1 6 2 6 3 6 2/3 power used 4 6 Only uses 1/2 power to lift 6kg 5 6 2/3 power used 6 6 All joints were tested up to the design lo ad. However, some joints met this load with less than full power. Testing shows that joints three and four are overpowered, and smaller motors could be substituted here. Joint 7 was tested differently as it does not have a moment arm already attached to it. A mechanics torque wrench was attached, and the maximum torque of this joint was found to be 25 N-m, sufficient for all anticipated tasks. 5.4 Joint Speed Measurements The maximum, unloaded speeds of each joint were measured using a known arc (90, 180, or 360 degrees as geometry permitted). Time to traverse this arc was measured with a stopwatch and joint angular velocities in RPM were calculated. From this, and the distance from the joint axis to the wrist plat e, a maximum wrist plate linear velocity was calculated. Table 5.3: Joint Speed Measurements Joint RPM Wrist Distance (mm) Wrist Speed (m/s) 1 5.8 889 0.54 2 5.8 889 0.54 3 7.1 508 0.38 4 10.0 508 0.54 5 11.0 254 0.29 6 6.5 254 0.18 7 16.0 0 0.00


56 In practice, maximum speeds will be limite d by the controllers to less than these values, especially when the end effector is near the user. 5.5 Energy Consumption Testing With any battery-operated device, energy use is very important. In this case it is especially so because if the arm were to discharge the wheelchairs battery, the user may be stranded. A digital multi-meter was set to current sensing mode and connected inline with the power feed from the wheelchair batte ry. Then, various operations were tested and power consumption recorded. The results are shown in Table 5.4. Table 5.4: Power Usage Condition Current (A) Idle all motors off, controller only 0.36 Holding self-weight outstretched 0.58 Holding 6kg fully outstretched 1.70 Lifting 6kg with joint 1 3.30 While more testing will be instructive, a reasonable estimate is that typical household and office tasks will lead to an aver age current of 2 Amperes. Six continuous hours of arm use would therefore consume 12 Ah. This would leave a 73 Ah battery (group 24 gel cell) with 61 Ah for propulsion, or 84% of capacity. Thus, driving range would be reduced, from perhaps 30 km to 25 km. This should be acceptable for most users. If not, most manipulation occurs with th e platform stationary, such as at an office desk. The arm is capable of plugging the wh eelchairs charger into the socket without any assistance, allowing a recharge during the workday.


5.6 Further Testing While the inverse-kinematic controller software is not yet complete, some testing was done to demonstrate the large, usable workspace of the manipulator. The following figures show the workspace envelope extr emes, the ease of reaching doorknobs on both the left and right, and how the arm may be unobtrusively parked behind the chair. Figure 5.4 WMRA in a Feeding Pose 57


Figure 5.5 Low-Right Reach Figure 5.6 Mid-Right Reach Figure 5.7 Left Side Doorknob Figure 5.8 Right Side Doorknob 58


Chapter Six: Summary and Future Work 6.1 Design Summary Table 6.1 USF WMRA Specifications Arm Mass 12.5 kg Max reachable height above floor 1.37 m Chair width increase with side mount 7.5 cm Average Current Draw 2 A Design Payload (including gripper) 6 kg Deflection at design payload 13.3 mm Degrees of Freedom 7 Actuator Type Brush DC Servo Transmission Harmonic Drive Controller Type Pic-Servo SC Figure 6.1 SolidWorks model Figure 6.2 USF WMRA as Built 59


60 6.2 Design Insights 6.2.1 Degrees of Freedom Invest in degrees of freedom. Increasing the joint count from just 4 up to 6 or 7 does increase cost somewhat, but makes the arm much more versatile. 6.2.2 Reconfigurability Reconfigurable arm lengths allow greater leverage on the engineering input. This is only practical with electric actuators placed at the joints. The MANUS, for example, houses all drive motors in the base and uses a complex drivetrain to drive the joints and gripper. Reconfiguration in this context means a complete redesign. Our design requires only a few hundred dollars in parts and an hour of a technicians time. 6.2.3 Side Mounting Side mounting is preferable overall. Ho wever, care must be taken to prevent widening of the power chair. Our arm is m ounted as far forward and upward as possible while still in a side mount configuration, and does not significantly increase chair width. This allows the arm to be parked by foldi ng it back, then wrapping the forearm behind the seat. The arm must be slightly longer th an with a forward mount, requiring greater shoulder joint torque and heavier gearboxes. This is compensated by the inherent efficiency of harmonic gearheads used in our drivetrain, allowing greater payload at less weight than the competing MANUS. 6.2.4 Cartesian Control Cartesian control is necessary. Raptor lack s encoders and therefore control is just single joint at a time, with a human doing all the work. Quadrature encoders are a costeffective way to provide closed-loop control.


61 6.3 If I Had To Do It Again There are several areas where improvements can be made, primarily in better packaging. This is very important for improving aesthetics and increasing user acceptance. The following are some of my thoughts on how to make an even better arm. Joints 3 and 4 could be improved with so me rearrangement. The motor for joint 3 hangs out a bit. If the motor-gearhead assembly was turned around 180 degrees, the motor could be neatly placed inside the main arm link tube. Likewise, the motor for Joint 4 could be placed inside the main link t ube, by using a right angle gearbox. These modifications would not change the performa nce or kinematics of the robot, but would certainly help to improve the appearance. While the wrist of this robot is f unctional and reasonably compact, I think improvement is possible. I have two ideas that may be investigated in future design projects: A differential drive in the wrist and a nonorthogonal joint arrangement. Because of the actuator-at-the-joint ser vo arrangement in this robot, there is a necessary offset of roughly 15 cm between the intersecting axes of the wrist and the final output flange to which the gripper mounts. In tight areas, this offset can restrict the range of possible orientations, and so it is desirable to reduce it. This is accomplished in other manipulators by use of a differential gear tr ain, which allows the three motors to be housed in the forearm. This can reduce the o ffset from the wrist axes intersection to the output flange, from 15 cm to perhaps 5 cm. The drawback is increased complexity and some backlash. A nonorthogonal wrist, also known as a 3-roll wrist, is another possible arrangement. While all 3 joint axes would inte rsect as before, the middle joint (joint 6 in


62 this case) would be placed at an odd angle relative to joint 5, perhaps 45 The advantage here is that this can help avoid singularities in the inverse kinematic solution, leading to more satisfying operation. I think this would make an interesting future project. 6.4 Future Work This was a project to design and bu ild a WMRA, up through the PC interface layer. Of course, a robot is useless without a good control system, and my work finishes with simple single-joint control. The next step for our group is to develop a Cartesian control scheme based on this hardware. As th is arm is fully programmable, I expect this process to be readily doable. This and other extensions of my work are listed here. 6.4.1 Develop High-Level Controller Features of the control system are al ready under consideration. The finished system will incorporate multiple input devi ces, to accommodate various user abilities. The main control mode will likely be velocity control, but there will also be provision for user-programmable positions. This will greatly speed repetitive tasks. 6.4.2 Orientation Locking One simplification for the user is an opti on to lock the end effector orientation. This can come in two varieties. First is a 3-DOF lock. This is useful for such a task as sliding open a drawer, where any orientati on change is undesired. Second is a 2-DOF lock. This allows the gripper to rotate a bout the world z-axis, thus keeping a glass of water level. Controlling the three position va riables is plenty of work for a human, without the added complication of constantly leveling an object.


63 6.4.3 Gripper Development The gripper itself is another considerati on. We have a mount for the BarrettHand BH-8, but its power supply is too bulky to be truly portable. This is a smaller project, but developing a small power supply for the BH8, that runs off the 24VDC wheelchair battery, will make the system truly portable again. 6.4.4 Trials Once the arm is fully operational, trials w ill begin. One aspect of these trials will be testing various link lengths against a set of tasks. Some tasks require a longer reach, such as reaching a high kitchen shelf or into a freezer. But longer lengths will reduce manipulability close to the mount, reduce payl oad somewhat, and also appear bulkier when stowed. Only real-world testing can determine what the best general-purpose dimensions are. Testing wont end with normal-ability researchers playing with hardware. Disabled volunteers will be enlisted to try out the device in our model apartment. Their comments will be noted and used to further develop the arm, especially the controls, GUI, and input devices. At a later stage, the arm may be lent to a disabled person to try out in a true real-world test. 6.4.5 Integration with Power Chair Yet another upcoming project deals with integration between the WMRA and the power wheelchair itself. As the chair po ssesses two degrees of freedom, with PMDC motors, retrofitting it to be a true Servo syst em is not difficult. We plan to mount encoders to each gear motor, and replace th e stock control system with two more PIC-


64 Servo controllers (and suitable power amplifiers ). The controllers will then be installed with the arms daisy chain of controllers, providing seamless integration from the hardware perspective. Once the platform is operational, work will begin on coordinated motion of the total arm-wheelchair system. Th is will lead to interesting capabilities, such as opening and holding doors while driving through. 6.4.6 Machine Vision Assist Functions Another area we have been developing sepa rately is machine vision. This system uses a camera on the end effector, coupled with advanced software that recognizes userselected objects and provides an assist function to home in on an object. This eases what can be a tedious process for the user. 6.5 Conclusion A wheelchair-mounted robotic arm (WMRA) was designed to meet the needs of mobility-impaired persons, and to exceed the capabilities of current devices of this type. The mechanical design incorporates DC servo drive with actuators at each joint, allowing reconfigurable link lengths and thus greater adaptability to a range of workspaces. Seven principal degrees of freedom allow full pose control, even while operating in the constricted workspace afforded by a side mount on a power wheelchair. A simple, scalable control system allows coordinated Cartesian control, and offers expandability for future research, such as coordinated motion with the wheelchair itself. We feel that this design will surpass previous attempts at building wheelchair mounted robotic arms that are truly useful and convenient. Subsequent testing, and ultimately the market, will determine if we are right.


65 References Reswick J.B., The Moon over Dubrovnik -A Tale of Wo rldwide Impact on Persons with Disabilities ," Advances in External Contro l of Human Extremities, 1990. J.R.Allen, A. Karchak, Jr., E.L.Bontrager, Design and Fabrication of a Pair of Rancho Anthropomomorphic Arms, Technical Report, The Attending Staff Association of the Rancho Los Amigos Hospital, Inc, 1972 T.Rahman, S.Stroud, R.Ramanathan, M.Alexander, R.Alexander, R. Seliktar, W. Harwin. Consumer Criteria for an Arm Orthosis, Applied Science and Engineering Laboritories. nathan/ publications/t_and_d.pdf M.J. Topping. The Development of Handy-1, A Robotic System to Assist the Severely Disabled, Proc ICORR 244-249 Hillman and Gammie. The Bath Institute of Medi cal Engineering Assistive Robot. ICORR 211-212 Dallaway JL, Jackson RD (1992) RAID a Vocational Robotic Workstation. Proceedings of the third International Conference on Rehabilitation Robotics. Neil Squire Foundation, Robotic Assistive Device. N. Katevas (Ed), Mobile Robotics in Health Care Services. IOS Press Amsterdam, 2000, pp. 227-251 H.F.M. Van der Loos, VA/Stanford Rehabilitation Robotics Research and Development Program: Lessons Learned in the Application of Robotics T echnology to the Field of Rehabilitation. IEEE Trans. Rehabilitation Engineering Vol. 3, No. 1, March 1995, pp. 46-55. Bernhard Borgerding, Oleg Ivlev, Christian Martens, Nils Ruchel: FRIEND Functional Robot Arm with User Friendly Interface for Disabled People. Institute of Automation Technology (IAT) rojekte/HTML_e/FRIEND.htm S. Sheredos, B. Taylor, C. Cobb, E. Dann. The Helping Hand Electro-Mechanical Arm. Proc. RESNA 493-495, Vancover, Canada (1995) G. Bolmsj, M. Olsson, P. Hedenborn, U. Lorentzon, F. Charnier, H. Nasri: Modular Robotics Design System Integration of a Robot for Disabled People Warner, P.R and Prior, S.D. "Investigations In to the Design of a Wheel chair-Mounted Rehabilitation Robotic Manipulator." Proceed ings of the 3rd Cambridge Wo rkshop on Rehabilitation Robotics, Cambridge University, England, April 8 1994. Holly A. Yanco. "Integrating Robotic Research: a Survey of Robotic Wheelchair Development." AAAI Spring Symposium on Integrating Robotic Research Stanford, California, March 1998.


66 H.Eftring, K.Boschian, Technical Results from Manus User Trials. Proc. ICORR 136-141 Edward McCaffrey, Kinematic Analysis and Evaluation of Wheelchair Mounted Robotic Arms. University of South Florida, Novermber 13, 2003.




68 Appendix A Cost Estimate Adaptive technologies must not only work well but must be affordable. The following is a brief estimate of the production cost of the USF WMRA. Table Appendix A.1: Cost Estimate Item Cost 7 controllers $897.33 7 motors $1,064.00 USB adapter $80.00 Wiring, connectors $200.00 PC controller $1,125.00 Human Interface hardware $2,000.00 2 CSF-25 Gearheads $1,615.00 1 CSF-20 $680.00 2 CSF-17 $1,275.00 1 CSF-14 $552.50 1 CSF-11 $467.50 Plastic covers $750.00 Aluminum stock $500.00 Fasteners and hardware $300.00 Gripper (undecided) $3,000.00 Machine Time (75 hours @ $60/hr) $4,500.00 Assembly (10 hours @ $25/hr) $250.00 Overhead ($250k/yr, 50 units) $5,000.00 Total $24,256.33 Prices for some parts, such as the motors, controllers, and gearheads, are based on price breaks given for large-quantity purchases. While rudimentary, this production cost estimate shows that we have met our goal of producing a capable arm that can be sold at retail for approximately $30000. The main variability in the estimate comes from the gripper design not yet being finalized, plus the Overhead catch-all category. The human interface hardware also may vary in cost, depending on individual needs.


69 Appendix B Kinematic Transformation Matrix Notes: t1 is the angle of joint 1, 1 T is the 4x4 transformation matrix relating the wrist plate frame back to the base frame. Each element is separated into one paragraph. Each row is enclosed in brackets []. Units of length are millimeters. T = Row 1 [((((cos(t1)*cos(t2)*cos(t3)-sin(t1)*sin(t3))*cos(t4)-cos(t1)*sin(t2)*sin(t4))*cos(t5)(cos(t1)*cos(t2)*sin(t3)+sin(t1)*cos(t3))*sin(t5))*cos(t6)+(-(cos(t1)*cos(t2)*cos(t3)sin(t1)*sin(t3))*sin(t4)-cos(t1)*sin(t2)*cos(t4))*sin(t6))*cos(t7)(((cos(t1)*cos(t2)*cos(t3)-sin(t1)*sin(t3))*cos(t4)cos(t1)*sin(t2)*sin(t4))*sin(t5)+(cos(t1)*cos(t2)*sin(t3)+ sin(t1)*cos(t3))*cos(t5))*sin(t7), -((((cos(t1)*cos(t2)*cos(t3)-sin(t1)*sin(t3))*cos(t4)-cos(t1)*sin(t2)*sin(t4))*cos(t5)(cos(t1)*cos(t2)*sin(t3)+sin(t1)*cos(t3))*sin(t5))*cos(t6)+(-(cos(t1)*cos(t2)*cos(t3)sin(t1)*sin(t3))*sin(t4)-cos(t1)*sin(t2)*cos(t4))*sin(t6))*sin(t7)(((cos(t1)*cos(t2)*cos(t3)-sin(t1)*sin(t3))*cos(t4)cos(t1)*sin(t2)*sin(t4))*sin(t5)+(cos(t1)*cos(t2)*sin(t3)+ sin(t1)*cos(t3))*cos(t5))*cos(t7), -(((cos(t1)*cos(t2)*cos(t3)-sin(t1)*sin(t3))*cos(t4)-cos(t1)*sin(t2)*sin(t4))*cos(t5)(cos(t1)*cos(t2)*sin(t3)+sin(t1)*cos(t3))*sin(t5))*sin(t6)+(-(cos(t1)*cos(t2)*cos(t3)sin(t1)*sin(t3))*sin(t4)-cos(t1)*sin(t2)*cos(t4))*cos(t6), -179*(((cos(t1)*cos(t2)*cos(t3)-sin(t1)*sin(t3))*cos(t4)-cos(t1)*sin(t2)*sin(t4))*cos(t5)(cos(t1)*cos(t2)*sin(t3)+sin(t1)*cos(t3))*sin(t5))*sin(t6)+ 179*(-(cos(t1)*cos(t2)*cos(t3)-sin(t1)*sin(t3))*sin(t4)-cos(t1)*sin(t2)*cos(t4))*cos(t6)241*(cos(t1)*cos(t2)*cos(t3)-sin(t1)*sin(t3))*sin(t4)241*cos(t1)*sin(t2)*cos(t4)+130*cos(t1)*c os(t2)*sin(t3)+130*sin(t1)*cos(t3)549*cos(t1)*sin(t2)+146*sin(t1)] Row 2 [((((sin(t1)*cos(t2)*cos(t3)+cos(t1)*sin(t3 ))*cos(t4)-sin(t1)*sin(t2)*sin(t4))*cos(t5)(sin(t1)*cos(t2)*sin(t3)-cos(t1)*cos(t3))*sin(t5))*cos(t6)+ (-(sin(t1)*cos(t2)*cos(t3)+cos(t1)*sin(t3))*sin(t4)sin(t1)*sin(t2)*cos(t4))*sin(t6))*cos(t7)(((sin(t1)*cos(t2)*cos(t3)+cos(t1)*sin(t3))*cos(t4)sin(t1)*sin(t2)*sin(t4))*sin(t5)+(sin(t1)*cos(t2) *sin(t3)-cos(t1)*cos(t3))*cos(t5))*sin(t7),


70 Appendix B (Continued) -((((sin(t1)*cos(t2)*cos(t3)+cos(t1)*sin(t3))*cos(t4)-sin(t1)*sin(t2)*sin(t4))*cos(t5)(sin(t1)*cos(t2)*sin(t3)-cos(t1)*cos(t3))*sin(t5))*cos(t6)+ (-(sin(t1)*cos(t2)*cos(t3)+cos(t1)*sin(t3))*sin(t4)sin(t1)*sin(t2)*cos(t4))*sin(t6))*sin(t7)(((sin(t1)*cos(t2)*cos(t3)+cos(t1)*sin(t3))*cos(t4)sin(t1)*sin(t2)*sin(t4))*sin(t5)+(sin(t1)*cos(t2)*sin(t3)cos(t1)*cos(t3))*cos(t5))*cos(t7), -(((sin(t1)*cos(t2)*cos(t3)+cos(t1)*sin(t3))*cos(t4)-sin(t1)*sin(t2)*sin(t4))*cos(t5)(sin(t1)*cos(t2)*sin(t3)-cos(t1)*cos(t3))*sin(t5))*sin(t6)+ (-(sin(t1)*cos(t2)*cos(t3)+cos(t1)*sin(t3))*sin(t4)-sin(t1)*sin(t2)*cos(t4))*cos(t6), -179*(((sin(t1)*cos(t2)*cos(t3)+cos(t1)*sin(t3))*cos(t4)-sin(t1)*sin(t2)*sin(t4))*cos(t5)(sin(t1)*cos(t2)*sin(t3)-cos(t1)*cos(t3))*sin(t5))*sin(t6)+179* (-(sin(t1)*cos(t2)*cos(t3)+cos(t1)*sin(t3))*sin(t4)-sin(t1)*sin(t2)*cos(t4))*cos(t6)241*(sin(t1)*cos(t2)*cos(t3)+cos(t1)*sin(t3))*sin(t4)241*sin(t1)*sin(t2)*cos(t4)+130*sin(t1)*cos(t2)*sin(t3)-130*cos(t1)*cos(t3)549*sin(t1)*sin(t2)-146*cos(t1)] Row 3 [(((sin(t2)*cos(t3)*cos(t4)+cos(t2)*sin(t4))* cos(t5)-sin(t2)*sin(t3)*sin(t5))*cos(t6)+ (-sin(t2)*cos(t3)*sin(t4)+cos(t2)*cos(t4))*sin(t6))*cos(t7)((sin(t2)*cos(t3)*cos(t4)+cos(t2)*sin(t4))*sin(t5)+sin(t2)*sin(t3)*cos(t5))*sin(t7), -(((sin(t2)*cos(t3)*cos(t4)+cos(t2)*sin(t4))*cos(t5)-sin(t2)*sin(t3)*sin(t5))*cos(t6)+ (-sin(t2)*cos(t3)*sin(t4)+cos(t2)*cos(t4))*sin(t6))*sin(t7)((sin(t2)*cos(t3)*cos(t4)+cos(t2)*sin(t4))* sin(t5)+sin(t2)*sin(t3)*cos(t5))*cos(t7), -((sin(t2)*cos(t3)*cos(t4)+cos(t2)*sin(t4))*cos(t5)-sin(t2)*sin(t3)*sin(t5))*sin(t6)+ (-sin(t2)*cos(t3)*sin(t4)+cos(t2)*cos(t4))*cos(t6), -179*((sin(t2)*cos(t3)*cos(t4)+cos(t2)*sin(t4))*cos(t5)sin(t2)*sin(t3)*sin(t5))*sin(t6)+179*(-sin(t2)*cos(t3)*sin(t4)+cos(t2)*cos(t4))*cos(t6)241*sin(t2)*cos(t3)*sin(t4)+241*cos(t2)*c os(t4)+130*sin(t2)*sin(t3)+549*cos(t2)] Row 4 [0, 0, 0, 1]


71 Appendix C Joint Torque Calculations As this is a serial chain robot, torques at each joint may be modeled by summing the torque contributions of each more distall part. For example, Joint 7 sees only the torque due to mass at the gripper, the payload, shown here as 8.24 Nm. Joint 1 must resist torques from all six other joints plus the payload. These are listed in the bottom row of the spreadsheet, and when summed they equal 98.79 Nm. Inputs to the spreadsheet are the distances between each part, the type of gearhead used, and the payload. The distances shown represent the arm in a horizontal, full outstretched position. The gearheads shown are the ones selected in the final design. The payload is 6 kg, for a weight of 58.86 N. Table Appendix C.1: Available Harmonic Drive Gearhead Specifications Gearhead model Torque (N m) Weight (N) CSF 11 12 1.47 CSF 14 19.5 5.08 CSF 17 46 6.69 CSF 20 70 9.63 CSF 25 140 14.72 Table Appendix C.2: Joint Torque Design Spreadsheet Joint Distance from next (m) Gearhead type Weight at Joint (N) tgrip t7 t6 t5 t4 t3 t2 total torque Rated Torque ok? Grip 0.14 0 58.86 0.00 0 Yes 7 0.06 11 4.41 8.24 8.24 12 Yes 6 0.11 14 8.52 11.77 0.26 12.04 19.5 Yes 5 0.00 17 10.61 18.25 0.75 0.94 19.93 46 Yes 4 0.49 17 18.46 18.25 0.75 0.94 0.00 19.93 46 Yes 3 0.12 20 19.44 47.09 2.91 5.11 5.20 9.05 69.36 70 Yes 2 0.10 25 24.53 54.15 3.44 6.13 6.47 11.26 2.33 83.79 140 Yes 1 0.00 25 24.53 60.04 3.88 6.98 7. 53 13.11 4.28 2.45 98.28 140 Yes


72 Appendix C (Continued) For a selected gearhead type in column C, the spreadsheet automatically fills in the appropriate joint weight. This joint weight accounts for the gearhead weight plus the motor, bracket, and the previous link tube weights. The rated continuous torque is also automatically filled into column M. The individual torque components are calculated, and summed in column L. As a final check, the calculated torques are compared to the rated torques, shown in column N. This spreadsheet was used to quickly evaluate dozens of design possibilities. Since each part affects the more proximal joints, optimization starts at the distall end and works inward. Various link lengths were tried, and then the minumim gearhead sizes were found for each joint. Gearhead torque ratings do not scale linearly with gearhead masses. Smaller gearheads generally carry less torque per unit mass. As self-weight is very important, it was found that by increasing total arm mass just 20%, from 9 kg to 11 kg, available payload nearly doubled, from 3.5 kg to 6 kg. This was considered to be an improvement, for much greater performance was found without much increase in either cost or weight.