A Nonlinearly Compliant Transmission Element for Force Sensing and Control by Andrew W. Curtis B.S., Mechanical Engineering Rice University, 1992 Submitted to the Department of Mechanical Engineering in Partial Fulfillment of the Requirements for the Degree of Master of Science in Mechanical Engineering at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY January 2000 @2000 Massachusetts Institute of Technology All rights reserved NSTITUTE MASSACHUSETTS INSTITUTE OF TECHNOL OGY SE P 2 0 LIBRARI ES Signature of Author...... Depa rtment of Mechanical Engineering January 31, 2000 Certified by ........ Pt. Kenneth Salisbury, Jr. P iciple Research Scientist Thesis Supervisor Accepted by ................ Professor Ain A. Sonin Chairman Committee on Graduate Students A Nonlinearly Compliant Transmission Element for Force Sensing and Control by Andrew W. Curtis Submitted to the Department of Mechanical Engineering on January 31, 2000 in partial fulfillment of the requirements for the Degree of Master of Science in Mechanical Engineering Abstract Recently, other researchers have demonstrated that force sensing and control capabilities can be implemented on a robot manipulator without the use of an explicit force/torque sensor. Instead, a measured flexible, or compliant, mechanism is added to an otherwise rigid joint transmission assembly and the joint forces are determined by measuring the This thesis compliant mechanism displacement [Williamson, 1995; Shah, 1997]. design mechanism transmission joint compliant nonlinearly an alternative investigates manipulators. that may be used to improve the dexterity of future planetary exploration The proposed nonlinear compliant element is based on the concept of a wrapping spring that undergoes a continuous decrease in spring free length and a resulting increase in stiffness for greater displacements. Multiple generations of prototypes were constructed to explore and evaluate mechanism design issues regarding the stiffening profile, the robustness, and the manufacturing, assembly, and integration methods. A four degree-offreedom robot arm was designed and constructed in conjunction with [Katz, 1999] to demonstrate integration of the proposed elements into a manipulator and to conduct evaluations of system level trajectory and force control techniques. The results indicate that the desired stiffening behavior can be achieved with this design and that it can be relatively easily integrated into a manipulator design, especially one with a cable transmission. However, the research also indicates that the mechanism performance can be significantly altered by small changes in part or assembly dimensions, thus indicating a potential lack of robustness as currently implemented. Thesis supervisor: Dr. J. Kenneth Salisbury, Jr. Title: Principal Research Scientist 5 Acknowledgements I would like to thank and acknowledge the following people for their time, encouragement, and assistance that contributed to my ability to complete this work: Dr. Ken Salisbury for inviting me to join his research group, where I have been able to explore several robotics topics while pursuing my specific interest in space robotics; The members of my research group, Arrin, Brian, Jesse, Ela, and Mark, for providing valuable technical advice and moral support on the numerous occasions when I encountered obstacles during this research; My friends and colleagues at Lockheed Martin Space Operations for providing me an opportunity to stay involved in ongoing Shuttle RMS activities as I work to improve my space robotics skills and knowledge base; My friends at Stewart Automotive Research for continuing to inspire me with their ingenuity, persistence, and intelligence; My parents and grandparents for teaching me at a young age that I can do anything if I put my mind to it; and My brothers, Adam, Byron, and Matthew. 7 Contents A B STRA C T ........................................................................................................................ 3 A CKN O WLED G EM EN TS .......................................................................................... 5 C ON TEN T S........................................................................................................................7 FIG U R E S............................................................................................................................9 TA BLE S..............................................................................................................................9 IN TRO D U C TION ............................................................................................... 1 1.1 1.2 1.3 11 O VERVIEW AND SCOPE ....................................................................................... 11 BRIEF SURVEY OF PRIOR TRANSMISSION COMPLIANCE WORK .......................... O UTLINE OF THESIS ................................................................................................ 13 THEO RY .................................................................................................................. 2 2.1 11 15 THE U TILITY OF EXPONENTIAL JOINT COM PLIANCE ............................................... 15 2.1.1 Exponential vs. Linear Compliance............................................................ 16 17 18 2.2 INTERPRETING JOINT TORQUES............................................................................... 2.3 CONTROL ................................................................................................................ 19 19 2.3.1 Position (Trajectory) Control ..................................................................... 20 2.3.2 Force Control............................................................................................. 20 2.3.3 Hybrid Position/ForceControl................................................................... 23 2.3.4 Controller Options...................................................................................... 2.1.2 3 Physicalvs. Virtual Compliance................................................................ H A RD W A R E D E SIG N ........................................................................................ 25 REQUIREM ENTS AND G OALS............................................................................... 25 3.1.1 NonlinearStiffness...................................................................................... 26 3.1.2 3.1.3 3.1.4 3.1.5 Compact..................................................................................................... Modular..........................................................................................................26 Robust (Durable) ....................................................................................... Scaleable..................................................................................................... 26 3.2.1 3.2.2 Description................................................................................................. Satisfaction of Requirements .................................................................... 27 28 3.1 26 27 3.2 W RAPPING SPRING CONCEPT ................................................................................27 3.3 M ODIFICATIONS & IMPLEMENTATION ISSUES.........................................................30 3.3.1 3.3.2 3.3.3 3.3.4 3.4 Addition of a Cam to Improve the Stiffening Characteristic..................... Spring Term inations ................................................................................. Manufacturingtolerances......................................................................... Measuringthe D isplacement .................................................................... D ESIGN VARIABLES ............................................................................................ 3.4.1 3.4.2 Maxim um Torque........................................................................................ Torque Dynamic Range ............................................................................. 30 30 30 31 33 33 33 8 3.5 CONFIGURATIONS ............................................................................................... 3.5.1 3.5.2 3.6 4 Capstan Configuration............................................................................. Coupling Configuration.............................................................................. W RAPPING SPRING A NALYTICAL M ODEL ............................................................... DESIGN EVOLUTION AND RESULTS..............................................................37 4.1 PROOF OF CONCEPT ............................................................................................. 4.1.1 Sim ulation of the Wrapping Spring Concept............................................. 4.1.2 Initial Elem ent Prototype........................................................................... 4.2 ONE D EGREE-OF-FREEDOM TEST STAND ........................................................... 4.2.1 Description................................................................................................. 4.2.2 Experim ental Procedures........................................................................... 4.3 V ERSION 1 COM PLIANT ELEMENT PROTOTYPE................................................... 4.3.1 Objective ........................................................................................................ 4.3.2 Description................................................................................................. 4.3 .3 R e s u lts ............................................................................................................ 4.4 V ERSION 2 COM PLIANT ELEMENT PROTOTYPE................................................... 4.4.1 Objective ........................................................................................................ 4.4.2 Description................................................................................................. 4 .4 .3 R e su lts ............................................................................................................ 4.5 COMPLIANT ELEMENT FINAL DESIGN ................................................................. 4.5.1 4.5.2 5 Objective ........................................................................................................ Description................................................................................................. 37 37 38 39 39 39 40 40 41 42 44 44 44 45 49 49 49 COMPLIANT ARM DESIGN FOR DIGGING (CADD).................51 5.1 5.2 6 33 34 34 34 O BJECTIVE ........................................................................................................... D ESCRIPTION ....................................................................................................... C O N C LU SIO N S .................................................................................................. 6.1 6.2 REVIEW OF THESIS W ORK ...................................................................................... FUTURE W ORK........................................................................................................53 BIBLIO G R A PH Y ............................................................................................................ 51 51 53 53 55 MATLAB CODE OF ANALYTICAL WRAPPING SPRING APPENDIX A. 57 M O D EL ............................................................................................................................. APPENDIX B. COMPLIANT ELEMENT PART DRAWINGS..............61 9 Figures Figure 1. Series Elastic Actuator Spring Designed by Williamson.............................12 13 Figure 2. Compliant Capstan Designed by Shah ......................................................... 13 Figure 3. NSCA Element Designed by Katz ............................................................... Figure 4. Phase Portrait and Equations for a Mass-Linear Spring System...................17 Figure 5. Phase Portrait and Equations for a Mass-Exponential Spring System...........17 Figure 6. Hybrid Position/Force Controller for a Compliant Joint Manipulator..........21 28 Figure 7. W rapping Spring Concept............................................................................. Figure 8. Inner Shaft with Original (Concentric) Shaft Adapter and Cam...................30 Figure 9. Original Potentiometer Signal Conditioning Circuit....................................32 32 Figure 10. Improved Potentiometer Signal Conditioning Circuit ............................... 34 Figure 11. Model Predictions and Sampled Data ......................................................... 37 Figure 12. Working Model 3D Simulation Images ...................................................... 38 Figure 13. Working Model 3D Simulation Results .................................................... 39 Figure 14. 1-D O F Test Stand........................................................................................ 41 Figure 15. Version 1 Compliant Element Prototype .................................................... 42 Figure 16. Shaft Pin D etail............................................................................................ 42 Figure 17. "Top Hat" Retaining Ring Clasp Detail .................................................... Cylinder without Secure Figure 18. Version 1 Static Test Results without Cam and 43 T erm in atio n ................................................................................................................ Figure 19. Version 1 Static Test Results with Cam.....................................................43 44 Figure 20. Version 2 Compliant Element Prototype .................................................... 45 Figure 21. Tw o V iew s of the T-block........................................................................... Figure 22. Version 2 Static Displacement vs. Torque Test Results with Cam.............46 Figure 23. Typical Torque Sensitivity Test Data.........................................................47 Figure 24. Version 2 Torque Sensitivity vs. Applied Torque......................................47 48 Figure 25. Version 2 Motor Driven Step Response Test Data ..................................... Simulations............49 Test Data and Response Offset Step Figure 26. Version 2 Initial 50 Figure 27. Final Element Design - Exploded View ..................................................... Figure 28. Compliant Arm Design for Digging CAD Model......................................52 Figure 29. Compliant Arm Design for Digging...........................................................52 Figure 30. Compliant Element Part Overview..............................................................61 Tables Table 1. Linear and Exponential Spring Displacements for the Same Force...............17 Table 2. Summary of Encoder and Potentiometer Attributes......................................31 Table 3. Index of Compliant Element Part Drawings...................................................61 11 1 Introduction 1.1 Overview and Scope The commonly accepted robot design methodology for serial chain manipulators is to make the mechanism as stiff as possible so that the endpoint position can be calculated with reasonable accuracy given measurements of the joint angles. To perform force controlled tasks with these manipulators, some type of compliance is usually incorporated in to the system. This compliance may be implemented using hardware or software methods and can be active (i.e. adapting to sensor data) or passive (e.g. compliant coverings). Besides enabling force control, hardware based compliance techniques can reduce peak shock loads on the motor/transmission assembly and have the potential to store energy, thus making the manipulator safer and more efficient. The consequence of adding compliance to the manipulator is that the trajectory control bandwidth and possibly the positioning accuracy will be reduced. For certain natural tasks, such as the process of extraterrestrial excavation that drives this research, this trade can be justified. This thesis presents the design of a nonlinearly compliant transmission element developed to improve robot force sensing and control capabilities. It was designed to meet requirements regarding nonlinear compliance, compactness, modularity, robustness, and scalability. The choice of a nonlinear, stiffening compliance characteristic was made to increase the force dynamic range of the mechanism without introducing unduly large displacements. The other requirements are intended to drive the development of a mechanism that may be integrated into both new and existing robot manipulators with minimal redesign effort. The proposed mechanism is based on the concept of a wrapping spring that undergoes a continuous decrease in spring free length and a resulting increase in stiffness as the displacement is increased. The mechanism was evaluated using a one degree-of-freedom test stand and a four degree-of-freedom robot arm was designed to demonstrate integration of the element into a manipulator and to conduct evaluations of manipulator control techniques. 1.2 Brief Survey of Prior Transmission Compliance Work Most research in the area of transmission compliance and joint flexibility has focused on developing control methods to compensate for this compliance with the objective of improving the trajectory tracking performance [Book, 1991; Hung, 1991; Readman, 1994; Spong, 1987]. The general assumptions associated with these research efforts are that the joint flexibility can be modeled by a linear spring, that the magnitude of the displacement caused by the flexibility is small, and that the natural frequency of the compliance is high compared to the bandwidth of the total arm motion. While these research efforts can provide useful guidance for modeling, analysis, and control techniques, the results are not directly applicable to the current research, which is interested in force control and which violates all of these assumptions to some degree. 12 Many other researchers have investigated active control of joint compliance using additional motors or other devices. [Sugano, et. al., 1992] designed a three actuator mechanism to simultaneously control the position, compliance, and damping characteristics of a robotic finger joint. [Morrell, 1996] investigated the use of separate but coupled position and force control motors for each joint. Recently, there have been some efforts to use passive joint compliance devices to provide improved force control capabilities. By measuring the deflection of a properly designed passive element, much of the benefit of the actively controlled methods can be achieved without a significant increase in the complexity of the original control system. In an effort to improve the capabilities of the humanoid robot Cog at the MIT Artificial Intelligence Lab, Williamson developed a passive compliant transmission mechanism called a series elastic actuator [Williamson, 1995] and integrated it in to Cog's arms to facilitate "natural" movements. These series elastic actuators are essentially linear torsion springs with a cross shaped cross section (see Figure 1) that are used as a torque coupling between the drive trains and the respective arm links. The deflection of the spring is measured using strain gauges. In practice, the series elastic elements in Cog's arms have demonstrated a successful implementation of a passive joint compliance technique. Figure 1. Series Elastic Actuator Spring Designed by Williamson The series elastic element concept was carried forward by the MIT Leg Lab during their development of several walking robots. [Matteo, 1997] provides a comprehensive overview of their efforts that resulted in a self contained linear actuator that uses compression springs to produce compliance. During research to develop a more capable and modular robotic finger, [Shah, 1997] designed a compliant capstan that provides a nearly exponential torque vs. angular displacement characteristic. This performance was achieved through the use of 6 Buna-N rubber balls in pie shaped slots formed by extrusions from opposite sides of the element (see Figure 2). The balls are compressed in the slots when the two sides of the element are rotated relative to each other. This relative rotation is measured with a potentiometer. While this is an elegant and compact mechanism that exhibits the desired stiffening behavior, there are a few drawbacks. First, in the high torque region, the two sides of the element are separated by the axial forces exerted by the compressed balls, resulting in inconsistent behavior. Second, the use of rubber may be appropriate for terrestrial 13 applications, but the potential pressure and temperature extremes of extraterrestrial applications would greatly degrade the performance of the rubber. Third, under repeated high load conditions, the rubber is likely to undergo some permanent deformation that would adversely effect the performance. Extrusions Figure 2. Compliant Capstan Designed by Shah In concurrent work to this research, [Katz, 1999] designed a nonlinear series compliance This actuator (NSCA) with a stiffening force vs. displacement characteristic. transmission drive cable a in pairs in used be to is intended concept spring compression to achieve joint compliance. The element uses a conical spring to achieve the desired nonlinear stiffening behavior and the displacement is measured using a linear potentiometer. II Figure 3. NSCA Element Designed by Katz Outline of Thesis 1.3 Section 2 discusses the theoretical issues associated with the use of exponentially stiffening compliant elements in a robot. Following a description of the value of exponentially stiffening springs relative to linear springs in Section 2.1, Section 2.2 provides an overview of the issues associated with the interpretation of joint forces to 14 calculate applied loads. Section 2.3 provides an overview of control techniques that are applicable to flexible joint manipulators. Section 3 provides a description of the hardware design process of the nonlinearly compliant transmission element. Section 3.1 defines the requirements used to develop the hardware and Section 3.2 describes the wrapping spring concept that is the basis for The following sections elaborate on some of the the new mechanism design. implementation and integration issues associated with incorporating the element in to manipulator hardware. Section 3.6 presents an analytical model of the wrapping spring mechanism that provides a reasonable estimate of actual performance. Section 4 describes the evolution of the design from the initial simulations and the evaluation of prototypes through to the final design of the element that will be integrated into the Compliant Arm Designed for Digging described in Section 5. Finally, Section 6 presents some conclusions of this investigation along recommendations for follow up activities. with 15 2 Theory 2.1 The Utility of Exponential Joint Compliance In general, the benefits of adding compliance to a serial manipulator include the potential to improve force control, to improve safety, to reduced wear, and to increase energy efficiency. The potential improvements in force control capabilities are derived from the dynamic stability characteristics of the combined manipulator and environment system. When both the robot and the contacted environment are rigid, small motions can generate large contact forces. Joint compliance alleviates these loads. However, the original rigid manipulator controller must be appropriately modified to prevent introducing new unstable behaviors through excitation of the added flexibility. Safety and reduced wear of the gearhead and motor are a direct result of the lower impact loads experienced by a serial manipulator that incorporates compliance. Furthermore, since the compliant components can store energy, the energy efficiency of the manipulator may be improved if an appropriate control strategy is employed. The two primary methods of adding compliance to robot manipulators, other than through joint compliance, are to use flexible links or to add a compliant covering to the parts of the manipulator that contact the environment. The major benefit of joint compliance compared to these other methods is that a relatively simple measurement of the joint compliance displacement is all that is required to calculate the endpoint position with the same accuracy as a rigid robot. With flexible links, a sophisticated combination of strain gauges and dynamic models are usually required to calculate accurate endpoint positions. The use of compliant coverings tends to insulate the manipulator from the environment and can introduce uncertainty about the exact point of contact even if contact sensing capabilities are incorporated into the covering. The key attribute of an exponentially stiffening compliant spring is that the incremental displacement, and thus the measurable incremental force, is a constant percentage of the applied load [Salisbury, in Mason and Salisbury, 1985]. To put it another way, an exponentially compliant element can provide a constant force resolution over the entire force range of the joint. This relationship can be derived by examining the following torque versus displacement equation for an exponential spring: e Equation 1. O where 0 is measured from the neutral, or zero displacement position. An incremental change in the torque is represented by: Equation 2. 6t = AeA06 , and thus the torque resolution is given by the ratio: Sz, = AeA196 eAO Se A = A80. Equation 3. 16 Equation 3 expresses the torque resolution as a linear function of displacement and illustrates how the compliant element displacement angle measurements will provide torque data with a constant resolution over the entire torque range. 2.1.1 Exponential vs. Linear Compliance The force dynamic range of a mechanism is defined as the ratio of the maximum controllable force divided by the minimum controllable force. It serves as an indication of the sensitivity and the range of the force sensing capability, with larger values being better. For reference, few multi-degree-of-freedom manipulators have a dynamic range better than 100 while the dynamic range of a human finger is on the order of 10000. When attempting to achieve a large dynamic range using a compliant force detection mechanism, linear springs are less practical than stiffening springs since they require a significantly larger displacement to measure high forces, assuming the sensitivities of the two are the same at low forces. The following equations demonstrate this property for the ideal case of an exponentially stiffening spring. Without loss of generality, the linear and exponential springs can be assumed to start at a neutral position where zero displacement corresponds to zero force. Equation 4 and Equation 5 represent the force equations for a linear spring and an exponential spring, respectively. Equation 6 and Equation 7 show the results of solving the first two equations for the displacement as a function of force. Equation 4. f= kx g x k(e = - 1) Equation 6. f/k y=IlnC1 A Assuming that k Equation 5. +1) Equation 7. k = 1 and solving for a force of f= g = 1, the resulting displacements are: x = 1/1 =1 Equation 8. y = ln(1/1 + 1)/A = ln(2)/A = .6931/A Equation 9. In Equation 9, the parameter A can be chosen to be .6931 so that the applied force of I causes a displacement of 1 in the exponential case to match the displacement in the linear case. Using these values for k and A, Equation 6 and Equation 7 can be used to demonstrate the drastically greater displacements necessary to measure or apply larger forces. Some results are shown in Table 1. By the time the dynamic range reaches 1000, the displacement of the linear spring is approximately 100 times greater than the displacement of the exponential spring. 17 Table 1. Linear and Exponential Spring Displacements for the Same Force Force X (Linear Y (Exponential 1 10 100 1000 Spring) 1 10 100 1000 Spring) 1 3.4597 6.6587 9.9679 Some insight in to the relative stability of an exponentially compliant mechanism can be obtained by examining phase portraits. As with a mass-linear spring system, the phase portrait of a mass-exponential spring system is of the form of a center point, indicating marginal stability (stable in the sense of Lyapunov) and the potential for limit cycle behavior (see Figure 4 and Figure 5). However, the difference in the case of the exponential spring is that the curves are not circular (or elliptical in the general case), but instead follow a more rectangular path. Of course, near the origin, the phase portrait of the exponential spring approaches a circular appearance, as would be expected since a linear approximation is applicable over this region. X m3 + kx = 0 x2 +x2 =C Figure 4. Phase Portrait and Equations for a Mass-Linear Spring System m+k(eAx -1)=0 + 2(A-LeAx - x)= C Figure 5. Phase Portrait and Equations for a Mass-Exponential Spring System The addition of damping, primarily due to friction, to either of these cases will tend to change the phase portraits into stable foci (asymptotically stable). 2.1.2 Physical vs. Virtual Compliance While the use of virtual compliance in the form of a control algorithm based on force/torque sensor data can be used to artificially create a wide range of compliance behaviors in a manipulator, the addition of physical compliance has a number of advantages in the context of a planetary exploration robot. Most of these benefits are 18 derived from the fact that the use of physical compliant mechanisms changes the fundamental open loop dynamics of the manipulator. Virtual compliance relies on closed loop routines and control algorithms that may be limited by sampling and control frequencies, noise, and actuator saturation. First, virtual compliance algorithms require the use of force/torque data to gather the data needed to issue force control commands. The use of an explicit force torque sensor becomes redundant when the proposed compliant elements are used since each joint torque can be measured directly. Second, physical compliance enables safer, more robust interaction with the environment. While software implemented compliance is dependent on the continued correct operation of all sensor and motor control hardware and circuitry, a physically compliant robot will maintain its compliant behavior when these components fail and even when power is removed from the control system entirely. The trade is an introduction of some physical component failure modes, which are likely to be more manageable on long duration space flights than failure modes for electronic components. Third, the physical compliant elements have the capacity to store energy, which can not be done using virtual compliance methods. This stored energy may be useful during digging or shoving tasks and may even augment striking tasks if a back swing motion of the appropriate amplitude and frequency is used. 2.2 InterpretingJoint Torques Given the kinematic configuration of a particular robot, the measured joint torques represent a set of wrenches located at the joint positions and aligned with the joint axes. The resultant force represented by the summation of these wrenches can be transformed in to the end effector coordinates, the base coordinates, or any other applicable coordinate system using kinematic equations, joint angle measurements, and transformation matrices [Bicchi, et. al., 1990, Murray, et. al., 1994]. The calculation of an applied load using joint angle torque measurements is dependent on the kinematic configuration, knowledge of the unloaded dynamic characteristics of the manipulator, and the number of applied loads. First, by their nature, joint torque measurements are in the same directions as the joint motion. Therefore, to acquire a complete six degree-of-freedom characterization of an endpoint force/torque couple requires that the manipulator have at least six joints and that it is not in a singular configuration. Using the manipulator Jacobian (Equation 10) and the duality principle, the joint torques are seen to be related to the endpoint forces by the transpose of the Jacobian matrix (Equation 11). For loads applied at locations other than the endpoint, the same process applies with an appropriately modified Jacobian matrix. = Equation 10. A ,r = j F Equation 11. 19 For many tasks, such as the basic trenching tasks expected for future planetary missions, the three components of the endpoint force are of primary interest and the endpoint torques are usually not particularly useful. In these cases, a minimum of three torque sensing compliant joints could provide complete force data. Second, the measured joint loads are the sum of the internal loads due to gravity, inertia, and motion effects (Coriolis and centrifugal forces) and of the externally applied loads. One way to separate these components to identify the contribution of the external load is to use an analytical dynamic model of the manipulator, such as Equation 12, to predict the internal loads. Alternatively, empirical data may be gathered by maneuvering an unloaded manipulator throughout its workspace at different speeds and in different directions to characterize the system before using it with applied loads. H(O6+ C(6, 0)0 + G Equation 12. Third, to accurately determine the magnitude of external loads on the manipulator from joint torque measurements, the number and location of the external loads must be determined or assumed. For multiple external loads on a single link of the manipulator, the joint torques can only be used to calculate the magnitude of an equivalent resultant load applied at some point on that link. For the cases in which there are at least one joint between two applied loads, the in-between joint data will reflect the effects of one load while the remaining joints will measure the resultant of the two loads. If the kinematic alignment is favorable and if there are enough joints between the two loads, both may be calculated. The control system could be designed to suspect multiple forces if the measurements from the last one or more joints indicate a sufficiently different force than the other joint measurements. 2.3 Control This section provides a brief description of the fundamental control methods used with flexible joint manipulators. The method chosen for a particular manipulator will depend on the task description. 2.3.1 Position (Trajectory) Control Most techniques for controlling joint flexibility during trajectory commanding boil down to a separation of the controller in to a fast inner force control loop and a slower outer position control loop [Book, 1991; Hung, 1991; Readman, 1994; Spong, 1987]. The goal of the inner loop is to counteract the joint flexibility and thus virtually stiffen the joint. This control technique is based on the assumption that the joint dynamics are relatively fast and of small amplitude when compared to the dynamics of the entire arm. This is not a good assumption for the designed compliant element, which introduces enough compliance to dominate most other sources of manipulator flexibility. Furthermore, since virtually stiffening the joints is counter productive to the objective of providing compliance to facilitate safe interactions with the environment, a different strategy must be employed. 20 Another difference between the proposed compliant element and the standard treatment of flexible joints is that the compliance is measured. This provides a direct measurement of the joint angle for use in the control algorithm rather than relying on a model based estimation. Thus, the dual nature of the measured compliant element potentially enables better force control as well as better positioning accuracy than previous flexible joint concepts. 2.3.2 Force Control When the manipulator is constrained by its environment in one or more directions, it becomes desirable to control the contact forces in those directions using a force controller. The successful use of another compliant element design to perform force control was demonstrated by [Shah, 1997]. As with the current compliant element, the measured feedback consisted of the motor angle and the compliant element displacement angle for each joint. The controller consisted of an outer, slower, PID torque control loop and an inner, faster, PD motor position control loop. 2.3.3 Hybrid Position/Force Control The fundamental concept of hybrid control schemes is to combine the utility of both position and force controllers in to a single controller that will automatically issue trajectory commands in the unconstrained directions and force commands in the constrained directions. Figure 6 illustrates one such control scheme. The inputs to this system are the vectors defining the commanded position of the manipulator endpoint in the fixed (base) reference frame and the desired endpoint force vector, also expressed in the reference frame. The measured outputs are the motor angles (from the motor encoder) and the compliant element displacement angles (from the potentiometer). The controlled state variables are the joint angles and the compliant element displacement angles. Use of these state variables allows decoupled control of the endpoint position and force through the kinematic and compliant element stiffness relationships respectively. 21 0 Xd Inverse Kinematics AO Od A'AO Trajectory Control (3p + T cmd = cto Ir 1m 'Tm N tT gemTra Motor Equation oo V. Robot + faT J Td -- > -Ce -1 Ce Limiter Tf Oced+,~ e'~'A, A0~ KlA Gf - Force Fre Control - Environment t 0~0 Ke ce Forward Dynamics + 'Txt Q4 x Tn 'y0O 2 rY '=HO+GS 0 A Figure 6. Hybrid Position/Force Controller for a Compliant Joint Manipulator The following is a description of the components of this hybrid controller concept. Inverse Kinematics - The inverse kinematic relationship from manipulator endpoint position to joint angles. Manipulator Jacobian, J 1 - This is the standard Jacobian that relates the joint rates to the endpoint rates. = J1 6 Equation 13. Transmission Jacobian, Jt - This matrix is the transfer function from the output side of the compliant elements to the joint angles. When the compliant elements are collocated with the joints, the transmission Jacobian reduces to the identity matrix. 0 = J,0t Equation 14. Compliant Element Stiffness Function, Kce - The compliant element stiffness has been deliberately designed to be nonlinear. In the ideal case, it can be represented by an exponential function of the displacement angle, T = sign(O,,) * A(e BIO,,l _ 1) Equation 15. 22 where A and B are constants. The specific function used will be determined by the characteristics of the actual stiffening compliant element used. Inverse Compliant Element Stiffness Function, Kee-' - This function is necessary to transform the commanded forces into pseudo joint angle commands for combination with the position control inputs to generate the composite motor control commands. In the ideal case of an exponentially stiffening compliant element, the inverse stiffness relationship is a logarithmic function of the torque applied to the compliant element. Oce = sign(r) * [1-%ln B (A) I Equation 16. Gear Ratio, Ng - This matrix represents the total gear ratio between the motor output and the input to the compliant element, including any intermediate transmission coupling effects. Motor Equation - The motor equation is derived from the motor specifications to transform the commanded torques into appropriate voltage signals to drive the motors. For the simplified case when the motor dynamics can be neglected, the motor voltage is a linear function of the commanded torque derived from the motor constant, the rotor inertia, and the terminal resistance. For the full dynamics case, the motor equation will also include rotor velocity and acceleration dependent terms. Forward Dynamics - A forward dynamics calculation of the joint torques from the measured joint angles and their derivatives provide a torque vector that is associated with the actual velocity of the manipulator endpoint. As shown in Figure 6, the traditional serial chain dynamics equation has been simplified to eliminate the Coriolis and centrifugal terms based on the assumption that the joint velocities will be relatively slow. Selection Matrices, a'f and a'p This hybrid control scheme employs a slightly different implementation of the selection matrices than a typical hybrid controller. Rather than being in Cartesian space, these selection matrices are in joint space and represent the relative weights on the torque and position controller commands passed to each joint motor. The torque calculated by the forward dynamics equation is subtracted from the measured joint torques to determine the component of the measured torque that is generated by external forces on the robot. This component is assumed to be in the direction that must be force controlled. Thus, the 'f selection matrix is calculated as the normalized diagonalization of the external torques calculated for each joint. The u's selection matrix is then calculated by subtracting O'f from the Identity matrix. [ Te'= I ext 0 0 0 *. 0 0 Equation 17. 23 2.3.4 Controller Options In the above discussions of position, force, and hybrid control, the specific control laws were not identified. While PID control laws are common and can generate adequate control commands under many circumstances, some model based control laws may be able to improve performance. It is not the intent of this section to review all possible methods for controlling manipulators with compliant joints, but to point out a few of the options. The four degree-of-freedom arm described in Section 5 has been designed and constructed to further explore control techniques for digging, trenching, scooping, and shoving tasks that might be expected of a planetary exploration manipulator. PID Controllers The use of PID controllers with the proposed nonlinear compliant element requires the application of some method to perform gain scheduling based on the current displacement angle (the current torque on the joint). One approach, demonstrated by [Shah, 1997], is to select gains for the current cycle that will produce a (constant) desired control bandwidth. This gain selection process relies on a locally linear approximation of the compliant element performance in its current configuration. Alternatively, one could adapt this process to choose optimized gains that exhibit increased bandwidth when the actuator is subjected to higher loads and the compliant element is relatively stiff. Model Based Controllers Control systems that incorporate a dynamic model of the manipulator, such as computed torque control, sliding control, and adaptive control, can often provide better trajectory tracking performance than PID controllers, but at the cost of increased computational workload. However, for a control system like the hybrid controller described above that already incorporates a dynamic model of the manipulator (to differentiate internal from external forces), the additional workload is minimal. 25 3 Hardware Design 3.1 Requirements and Goals This research effort was undertaken to develop a nonlinearly compliant transmission element that is compact, modular, robust, and scaleable. The target application for this mechanism is to improve the capabilities of interplanetary exploration robots. The fundamental goals are threefold: * Improve force control over a wide dynamic range, * Simplify actuator design and instrumentation, and * Increase overall manipulator durability. Improvements in force control are necessary to facilitate dexterous interaction of manipulators with the environment. To date, operational space manipulators such as the Shuttle Remote Manipulator System and the experiment arm on the Mars Sojourner rover have been position controlled devices without any force feedback capabilities. In recognition of the utility of force control to enable more dexterous operations, the partners in the International Space Station program plan to incorporate force/torque sensors in to the smaller, precision task robots such as the Special Purpose Dexterous Manipulator (SPDM) and the Japanese Small Fine Arm (SFA) [Brimley, et. al., 1994]. Additionally, improved manipulator dexterity through design as well as through software algorithms is one goal of NASA's Planetary Dexterous Manipulator program at the Jet Propulsion Laboratory (JPL) to enable more productive remote geology operations on Mars and possibly on other celestial bodies [Das, 1999]. With this recognition of the value of force control comes the engineering challenge of designing improvements. The power, mass, and volume limitations as well as the harsh operating environment of space and on foreign planets, etc., lead to the adoption of the goal for a simple design. The compliance should be implemented in a minimal package (mass and volume), that uses a minimal amount of power (for sensors and computation), and that has built in redundancy or graceful degradation. For space manipulators to become a more useful tool for exploration and investigation, they need to become more robust to external disturbances and capable of sustaining extensive intentional and unintentional contact with their surroundings. The current generation of arms is often operated to maximize clearance between the manipulator and the surrounding structure except for occasions when an end effector interaction is required. Operations would be facilitated by not being concerned about incidental contacts because the manipulator has enough compliance to absorb impacts with minimal physical damage to itself or to the environment. The specific requirements for this research effort are documented in the following sections. 26 3.1.1 Nonlinear Stiffness The case for exponential stiffness is made in Section 2.1 above. While this characteristic is a reasonable goal, it is not necessary to match an exponential curve exactly to realize the benefits of nonlinear compliance. Thus, the requirements are: " The compliant element stiffness shall increase nonlinearly such that at high loads, an incremental displacement measurement corresponds to a greater incremental force than the same displacement measurement at a lower applied load. The ideal characteristic for the nonlinear characteristic is exponential. * The compliant element stiffness characteristics shall be known (able to be modeled), symmetric, and constant for the lifetime of the element. 3.1.2 Compact Since the target application of this project is interplanetary robotics, it is important to develop a component package that is sufficiently compact and power efficient to be integrated into small, lightweight manipulators. Specifically: " The compliant element shall have a size and mass comparable to a conventional coupling or capstan used in current (space) manipulator designs. " The compliant element shall not add significant power, data, or computation requirements to the manipulator. 3.1.3 Modular As part of the effort to meet the simplification goal, it should be possible to retrofit compliant elements into existing hardware. Thus: * The compliant element shall be designed to be similar enough to existing capstans or couplings to allow for retrofitting into existing robotic devices. Additionally, satisfaction of this requirement should lead to a design that can be readily replaced without significant disturbance to the rest of the manipulator, thus facilitating repairs and upgrades or change outs to update the specific nonlinear compliance characteristics for each joint. 3.1.4 Robust (Durable) In order to survive the harsh environment of space for long periods of time, either in operation or in transit to its destination, the compliant elements must be designed using appropriate materials and techniques for minimizing performance degradation. " The compliant element shall use appropriate materials for space based applications. * The compliant element shall be designed to have minimal performance variation due to ambient thermal and pressure conditions. 27 * The compliant element shall be designed to withstand anticipated operational and non-operational (i.e. launch and landing) dynamic and shock loads. 3.1.5 Scaleable The two fundamental characteristics of the compliant element that must be scaleable are the overall package size and the force range. The range of displacement for a rotary element will be determined by the dynamic range and the maximum force requirements. * The compliant element shall be designed to be scaleable such that it can be incorporated into robots from rover scale to space station scale. 3.2 Wrapping Spring Concept Upon examination, none of the previously developed compliant element mechanisms fully satisfied the design requirements just presented. The [Williamson, 1995] element has a linear torque vs. displacement characteristic and is not particularly compact. The [Shah, 1997] element has an appropriate nonlinear characteristic and is very compact, but it uses inappropriate materials for space applications. Thus, a new mechanism concept was sought that could satisfy all of the requirements, leading ultimately to the wrapping spring concept that is employed by the compliant elements evaluated in this research. 3.2.1 Description Figure 7 illustrates the fundamental components of the wrapping spring concept, including the central shaft (blue), a concentric cylinder (red), and two opposing springs (yellow and green). The springs are semicircular wire forms with 90' hook terminations on both ends. The bottom terminator must be securely fastened to the center shaft while the top terminator is fastened to the inside of the cylinder. As the cylinder rotates concentrically with respect to the shaft, one spring will begin to wrap around the shaft as the other is pressed against the inside wall of the cylinder. Both of these results have the effect of shortening the free length of the springs. Thus, while the springs themselves have a linear stiffness characteristic, the progressively shorter free length causes a nonlinear increase in torque per unit displacement in the mechanism. The maximum relative rotation of the cylinder relative to the shaft is limited by the spring being compressed against the inner wall of the cylinder. 28 Figure 7. Wrapping Spring Concept 3.2.2 Satisfaction of Requirements Nonlinear Stiffness While the modeling and initial simulation efforts predicted a desirable spring behavior, the actual performance of the Version 1 prototype did not exhibit as much of a stiffening profile as desired (see Section 4.3). This motivated the examination of the addition of a non-concentric cam element to the inner shaft (see Section 3.3.1). This design modification succeeded in creating a more desirable stiffening characteristic. The stiffness behavior of the mechanism does not fully satisfy the requirement for symmetry or constant behavior. As discussed further below, the test results indicate a definite asymmetry in the torque vs. displacement performance of the element that is likely due to a combination of part dimension uncertainty and/or assembly misalignment. Also, the mechanism has been observed to exhibit hysteresis -- the displacement measurements can be affected by the previous state of the apparatus and whether it is being loaded or unloaded. Compact The final design of the nonlinearly compliant element is 1.5" in diameter by 1.25" long, including clearance for the potentiometer wiper element. At this size, it is no larger than a standard capstan element that would have otherwise been used to implement the cable transmission that was chosen for the four degree-of-freedom manipulator described in Section 5. The power requirements for the designed element are minimal. Only a 5 Volt power supply using less than 10 gA is required to drive the potentiometer signal. The analog output voltage signal from the potentiometer does require some signal conditioning (described in Section 3.3.4) before it can be read by a typical A/D converter. 29 Modular The most significant aspect of the final compliant element design that reduces its modularity is the need to modify the inner shaft with holes for the springs. For manipulator designs with readily removable shafts, such as the shoulder yaw joint of the four degree-of-freedom manipulator described in Section 5, the designed element is entirely modular. When used in the capstan configuration, the element is of an appropriate size to be For use in a shaft coupling directly interchangeable with standard capstans. configuration, the design requires a modification to the end cap to incorporate a shaft clamp, pin, or set screw feature. At 1.5" in diameter and approximately 1.75" long (to accommodate the end cap modification), this compliant element is about 1.5 to 3 times larger than a standard shaft coupling for a 0.25" diameter shaft. While different sets of springs (different spring wire diameters) will require modifications to the termination hardware (the T-block, shaft, and shaft adapter parts), all other parts are reusable. Robust (Durable) This design uses metal springs to meet the materials requirement for robots designed for operation in space. While [Shah, 1997] demonstrated promising results with compliant elements using rubber balls as the compliance mechanism, rubber may not be a wise choice for use in space where the temperature and pressure extremes can make the rubber brittle. While metal springs are less affected by thermal and pressure conditions, the prototypes have revealed that the performance of the mechanism is very sensitive to the exact relative geometries of the parts. Therefore, further thermal analysis would be prudent to evaluate whether the use of metals with different thermal coefficients (steel and aluminum) is acceptable, or if a single material (steel) must be used. Scalable The physical package size of the compliant element is inherently scalable to fit virtually any shaft size. The force range of the element is scaleable by choosing different spring wire diameters, by using multiple spring pairs, or by using flat springs of various widths. Several design details are worth noting. First, the specific means for terminating the ends of the springs must be carefully planned for each spring wire cross section and size. Second, an appropriate rotary encoder that accommodates the selected shaft size must be identified. Third, the minimum outer diameter of the compliant element will be determined either by the chosen spring geometry (force dynamic range and maximum displacement) or by the rotary potentiometer diameter. Fourth, the minimum length of the element will be determined by the spring wire diameter (or width), the number of spring pairs, the shaft bearing width, and the potentiometer width. 30 3.3 Modifications & Implementation Issues 3.3.1 Addition of a Cam to Improve the Stiffening Characteristic The initial tests of the compliant element prototypes (see Section 4.3.3) indicated that the torque versus displacement characteristic of the element was not exhibiting as much stiffening as desired. To increase the stiffening behavior, a cam part was added to the assembly. This cam is a cylindrical part with a 5/8" diameter (the original shaft adapter is 1/2" diameter) that is mounted on the inner shaft adapter in a non-concentric fashion as shown in Figure 8. Thus, as the inner shaft rotates with respect to the cylinder, the radius of the cam part in contact with the spring increases and causes a more rapid increase in the stiffening behavior than the original design. Test results demonstrate the success of this modification (Sections 4.3.3 and 4.4.3). Figure 8. Inner Shaft with Original (Concentric) Shaft Adapter and Cam 3.3.2 Spring Terminations The springs in this mechanism must have rigid terminations that will accommodate the maximum anticipated torque loads. Secure terminations are vital to maintaining the required component geometry that is responsible for generating the nonlinear performance. Depending on the spring wire diameter and the relative dimensions of the shaft, the cam, and the wire curvature and cylinder diameters; it may be necessary to use different means to adequately terminate the springs. Several methods were explored using the prototypes, including a pin restraint, a retaining ring restraint, and a setscrew restraint. Descriptions of these methods and of how well they worked are included in Sections 4.3.3 and 4.4.3. 3.3.3 Manufacturing tolerances Since the stiffening characteristic of the element is a function of the geometry of the springs and of the parts they contact, the manufacturing tolerances of all these elements must be considered. While the spring wire diameter is extremely uniform (+/- 0.001" or less for wire diameters of 0.1" or less), the wire forming process to create the spring curvature and the termination segments can not be controlled with such precision. It was necessary to select closely matched pairs of springs following the bending process, 31 especially for the Version 2 prototype, which used a larger spring wire diameter and thus had smaller clearances between the springs and the other parts. 3.3.4 Measuring the Displacement Choice of Sensor The usefulness of the compliant element as a force sensor is dependent on the ability to accurately measure the displacement angle. Both a through-shaft encoder and a rotary potentiometer were considered for this purposes. Both were implemented on the one degree-of-freedom test stand to evaluate their relative performance characteristics. A summary of the attributes of both pieces of hardware is provided in Table 2. Table 2. Summary of Encoder and Potentiometer Attributes Attribute Package Size * Package Mass * Encoder (HEDS 6505) 2.6" X 2.2" X .81" 30 g 4.63 in 3 Power Requirement 5 Volts, 5 mA max -> 25 mW Resolution 1024 CPR Absolute Zero Output Signal Additional Equipment No Digital Mounting Hardware Potentiometer (Novotechnik Model P45a502) 1.5" Diam. X .32" = 2.26 in 3 3.3 g 5 Volts, 10 tA -> 50 jiW Limited only by A/D conversion process Yes Analog (Voltage) Mounting Hardware Signal Conditioning Circuitry *Without mounting hardware. PotentiometerIntegration With instrument resolution and size as the most important attributes, the potentiometer was chosen for use in the final design. The technical challenge associated with the integration of a potentiometer in to the compliant element mechanism is to amplify the output signal so that the compliant element range of motion (+/- Degrees) corresponds to the A/D converter input range (+/- Volts) without introducing excessive noise. Figure 9 provides an illustration of the original potentiometer signal conditioning circuit used with the Version 1 prototype. The potentiometer acts as a voltage divider with R 1, which was chosen to be large enough to keep the current below the 10 gA rating of the potentiometer. The first operational amplifier circuit amplifies the potentiometer output by -R1/R 2 and the second one inverts the signal so that the output is positive. Since R 2 is the same resistance as the maximum potentiometer resistance, the output will be in the range of 0 to 5 volts. In practice, this circuit exhibited a significant amount of noise during the static tests of the prototype. Figure 19 shows the wide variation of 40 consecutive potentiometer data 32 samples taken at each load condition. without much success. R,, Various filtering techniques were attempted R = 505 kQ R2= 5 kQ 24 kQ Vout = 0 to 5 V R3 rR2 +R3 R2 :Pot, Gain = RI/R2 Figure 9. Original Potentiometer Signal Conditioning Circuit A second potentiometer circuit, modeled after the circuit used by [Shah, 1997] and shown in Figure 10, was used with the Version 2 prototype. The first operational amplifier divides and inverts the supply voltage to provide a voltage difference of -5 to 0 volts across the potentiometer. The second operational amplifier provides a unity gain and prevents any current from flowing through the output lead of the potentiometer. The last operational amplifier provides a gain of -R4/R 3. The variable resistor R 2 is used to set the zero bias of the output voltage as shown in Equation 18. Equation 18. R R R3 ' 2 The performance of this improved circuit proved to be quite good during the static tests of the Version 2 prototype. Figure 22 shows the dramatic reduction in the variation of 40 consecutive potentiometer samples at each load condition compared to those in Figure 19. However, a significant amount of noise was observed in the potentiometer signal Since this amplifier and circuit board when the motor amplifier was activated. configuration are not the same as the one that will be used in the four degree-of-freedom manipulator described in Section 5, further trouble shooting will be conducted with the real hardware. R= 138 kQ R 2 = Oto 100 R1/3 R2 Ri R4 kg R3= 4 kQ R4 = 37 kQ Vout = -I to I V +r - -R-1 Gain = R4/R3 15V' -ot Zero bias set by adjusting R 2 Figure 10. Improved Potentiometer Signal Conditioning Circuit 33 Calibration Calibration of the potentiometer output (using the circuit in Figure 10) consists of setting the zero bias and establishing the conversion between output volts and the displacement angle. The compliant element can be set to the zero displacement position by either removing all applied loads or by applying a symmetric torque to the element. At this position, the variable resistor R 2 is adjusted to establish the zero bias. It should be emphasized that this zero output voltage is dependent on the supply voltage and on the potentiometer wiper position. If either of these change, the zero bias must be reestablished. Resistors R 3 and R4 will determine the conversion factor between the amplified output voltage and the displacement angle. The gain (R4/R 3) should be selected so that the maximum range of element displacement fills the A/D signal detection range. This will maximize the resolution of the displacement measurements. A test set that includes an encoder, such as the one degree-of-freedom test stand described in Section 4.2, can be used to determine the calibration ratio of the selected resistors. 3.4 Design Variables The two primary variables used to define a compliant element design are the maximum torque and the torque dynamic range, as described in the following sections. 3.4.1 Maximum Torque Since the proposed compliant element is intended to be installed at the joints, it must be properly designed to accommodate the maximum dynamic loading conditions expected at each joint. This is a significant factor in selecting the spring wire diameter or, in the case of multiple or non-circular springs, the total spring cross sectional area. The maximum load, in combination with the spring cross section, will define a minimum inner shaft diameter for the spring to wind on. 3.4.2 Torque Dynamic Range Selection of the desired torque dynamic range following the identification of the maximum torque will drive many of the remaining design parameters. Most notably, the dynamic range will drive the relative dimensions of the inner shaft diameter, the cylinder diameter, and the spring radius of curvature. Consequently, these parameters will define the outer dimensions of the compliant element as well as the details of the T-Block and inner shaft termination designs. 3.5 Configurations The compliant element proposed in this thesis can be used either in a capstan or in a shaft coupling configuration as described in the following sections. 34 3.5.1 Capstan Configuration The capstan configuration is intended for use in a cable transmission robotic system such as the four degree-of-freedom manipulator described in Section 5. In this configuration, the motor or gearhead output is rigidly coupled to the inner shaft and the cable transmission uses the outer cylinder as a capstan. 3.5.2 Coupling Configuration The coupling configuration is intended to be used as a replacement for a rigid shaft coupling. Instead of extending the inner shaft through both end caps as in the capstan configuration, the inner shaft terminates at the bearing on one side and the end cap is modified to provide a rigid coupling between the cylinder and a second shaft. Wrapping Spring Analytical Model 3.6 An analytical model was developed using geometric considerations and beam bending theory to predict the performance of a mechanism based on the wrapping spring concept. The model contains calculations for the contributions of both the winding and the unwinding spring segments. For both cases, the model assumes ideal spring terminations and that the entire induced torque is due to the bending moment of the remaining free length of the spring at the current displacement angle. The model predictions and sampled data for the .045" diameter springs of the Version 1 prototype and for the .092" diameter springs of the Version 2 prototype are shown in Figure 11. Displacement Torque: Version 1 CE with Cam, Model Prediction and Sampled Data 30 Displacement vs. Torque Version 2 CE with Cam, Model Prediction and Sampled Data 20 15 20 10 10 1D - 0 - c"-5In * -10 -10 -20 -15 -500 -d00 -300 -200 0 100 -100 Static Torque (mNm) 200 300 400 500 -2000 -1500 -1000 0 500 -500 Static Torque (mNm) 1000 1500 2000 Figure 11. Model Predictions and Sampled Data While the correlation between the model predictions and the sampled data is reasonably good, there are two discrepancies that merit discussion. First, the measured data indicates a nearly constant linear behavior of the mechanism near zero displacement that is not reflected in the model prediction. This behavior is likely caused by some bending of the spring termination segments and by preload stresses in the springs caused by part and assembly variations. Second, there is a noticeable asymmetry of the measured data about the zero displacement configuration. This deviation is also attributed to part variation, 35 most likely of the cam (not ideally aligned or circular), since the same asymmetry was observed with two different sets of springs. Other part variations that could contribute to this type of discrepancy include the spring radius of curvature and the precise alignment of the T-block to terminate the springs symmetrically on the cylinder. For reference, a Matlab script file of the wrapping spring analytical model is included as Appendix A. 37 4 Design Evolution and Results 4.1 Proof of Concept 4.1.1 Simulation of the Wrapping Spring Concept The initial proof of concept was accomplished using the Working Model 3D software package from MSC.Software Corporation to conduct a quasi-static torque vs. displacement simulation of the wrapping spring concept. The model was designed to evaluate the winding half of the concept about the center shaft. The spring was modeled as eight cylindrical solid bodies joined by linear torsion springs with neutral positions chosen to create the semicircular spring shape (see Figure 12A). The top of the spring was fixed while the bottom of the spring was constrained to move with the center shaft as it was rotated. The surfaces of the center shaft and of the spring elements were defined as solid contact surfaces. A. Theta 0 B. Theta 18 Figure 12. Working Model 3D Simulation Images The results of slowly rotating the center shaft and measuring the resulting axial torque generated by the springs on the fixed support (corresponding to the outer cylinder) are shown in Figure 13. Co-plotted with the measured data is the closest fit exponential curve. While the results are not exactly exponential, they do trend in the desired direction. One apparent artifact of the non-continuous spring model used in this simulation is the appearance of noticeable increases in the stiffness as subsequent spring elements make initial contact with the inner cylinder (at approximately 6, 13, and 18 degrees.) This behavior validates the wrapping spring concept as a means to generate a nonlinearly stiffening mechanism. The other artifacts that appear as spikes between 8 and 10 seconds were caused by the contact model dynamics of the Working Model 3D simulation that were occasionally excited as the shaft rotated. 38 8 Element Winding Spring Model -- Linear Torsion Spring/Damper Joints -- Diameters: 6,12,1.5,9 Wnrking Madel 3D Y k n.04865*exp(0.2601*theta) 0.5 0.4 0.3 0.2 0.1 - 0 2 4 6 12 8 10 Relative Twist Theta (Degrees) 14 16 18 20 Figure 13. Working Model 3D Simulation Results 4.1.2 Initial Element Prototype An initial, uninstrumented prototype element was constructed to help identify the assembly and manufacturing issues for the compliant element. It was constructed using a 0.5" diameter aluminum shaft, an acrylic, semi-transparent cylinder with an inner diameter of 1.0", two .045" diameter wire springs with a radius of curvature of 0.375", and two bearings. The ends of the springs were only passively secured in the shaft and in the cylinder by the geometry of the spring termination segments. The initial element prototype provided some useful insights: " Secure spring termination techniques are required for proper performance; * Performance is dependent on the spring mounting angle -- springs that are not mounted with the plane of curvature perpendicular to the inner shaft have less contact with the inner wall of the cylinder and thus will have a greater range of motion; " The method for assembling the springs with the inner shaft relies on the flexibility of the springs to expand around the shaft during assembly - modifications to the inner shaft will be required to accommodate less flexible springs; " The approximate range of motion (± 300) for the geometric configuration used (0.1:1:1.5:2 spring wire, shaft, spring curvature, cylinder diameter ratio) was verified. 39 4.2 One Degree-of-Freedom Test Stand 4.2.1 Description Figure 14 shows the one degree-of-freedom test stand. Starting from the left in the figure are the vertical support, the compliant element, an encoder, the motor support, a 72.38:1 gearhead, the motor, and a motor encoder. The potentiometer can bee seen on the left side of the compliant element. The encoder attached to the compliant element is a Hewlett Packard HEDS 6505 model through-shaft encoder with 1024 counts per revolution and was used to calibrate the compliant element potentiometer circuit to provide the best possible resolution for the displacement range of the compliant element. Figure 14. 1-DOF Test Stand 4.2.2 Experimental Procedures Static Displacementvs. Torque Test The one degree-of-freedom test stand was used to perform static tests to determine the displacement vs. torque characteristics of the compliant element prototypes. This was accomplished by immobilizing the inner shaft (with the clamps visible in Figure 14) and then applying a torque load to the element using hanging weights (not visible in Figure 14). With the static load applied to the element, one encoder measurement and forty The mean, maximum, consecutive potentiometer measurements were sampled. data were potentiometer of samples forty the of minimum, and standard deviation of this results The signal. calculated to evaluate the amount of noise in the potentiometer an and 11) test were compared to the analytical model predictions (Section 3.6, Figure exponential curve fit of the data was used to perform some simulations (Section 4.4.3). 40 Torque Dynamic Range Test With an exponential curve fit of the data generated by the static displacement vs. torque test and with an estimate of the noise level in the potentiometer signal, predictions can be made about the expected torque resolution of the compliant element. This prediction can be verified by observing the potentiometer signal when the incremental torque is applied. The procedure for this test is to apply the desired initial load, begin data recording, and add the incremental load. If the load is not detectable, then a slightly greater load is applied until the signal difference is clearly discernable from the noise. Motor Driven Step Response Test This test is intended to assist in the evaluation of the dynamic characteristics of the compliant element prototypes. In an attempt to generate a step input with the motor, a maximum motor drive command is issued to the test set for a user-specified length of time when the test is initiated. When the time limit is reached, a PD controller that uses the motor encoder is activated to hold the current motor position for the remainder of a four-second window. Then, the same process is executed in the opposite direction. The results of this test provide a dynamic characterization of the combined motor, gearhead, and compliant element system. Initial Offset Step Response Test This is another test to evaluate the dynamic characteristics of the compliant element prototype. In this test, the center shaft is immobilized and the cylinder is displaced by applying an external torque. After data recording is started, the external load is released and the dynamic response of the prototype returning its neutral position (zero displacement) is recorded. Unlike the motor driven step response test, the compliant element prototype is isolated from the motor and gearhead dynamics, permitting dynamic characterization of just the prototype. 4.3 Version 1 Compliant Element Prototype 4.3.1 Objective Version 1 of the 1-DOF element prototype was constructed to accomplish the following objectives: * Evaluate the performance of the selected potentiometer hardware; * Evaluate methods for securing the spring ends to the center shaft and to the cylinder; * Characterize the performance of one or more spring pairs with different spring wire diameters for comparison with predicted behavior; * Evaluate the performance of the mechanism with the addition of a non-concentric cam added to the inner shaft to increase the nonlinear behavior. 41 4.3.2 Description The Version 1 compliant element prototype is shown in Figure 15. The inner shaft consists of a .25" aluminum shaft and a .50" aluminum shaft that are joined using a rigid shaft coupling. The outer cylinder was machined out of semi-transparent acrylic to permit some visibility of the internal components. A Hewlett Packard HEDS 6505, 1024 CPR encoder is installed on one side of the element to provide redundant displacement data for calibration of the potentiometer data. The rotary potentiometer is installed on the opposite side of the element (on the left side in Figure 15). The Version 1 prototype was assembled using .045" diameter springs mounted at a 60' angle to the perpendicular plane. This configuration minimizes contact between the spring and the inside of the cylinder wall, resulting in a nearly pure wrapping behavior. Following the disappointing initial results (see below), this prototype was modified to include a non-concentric cam on the inner shaft to amplify the nonlinear stiffening behavior. Figure 15. Version 1 Compliant Element Prototype Two spring attachment techniques were investigated with this prototype. The spring was attached to the inner cylinder using a keyway technique. Notches were cut in to the terminator section of the spring and a pin was used to lock them in place (see Figure 16). The spring was attached to the cylinder using a ring and groove technique. The ends of the spring were inserted through the "brim" of a top hat shaped retention clasp and were held in place by a retaining ring that fit into groves cut in to the spring terminator section (see Figure 17). 42 Figure 16. Shaft Pin Detail Figure 17. "Top Hat" Retaining Ring Clasp Detail 4.3.3 Results First, the displacement vs. torque performance of this prototype indicates that the .045" diameter springs do not provide a stiff enough mechanism for use in the four degree-offreedom robotic digging arm under construction as a follow-on activity (see Section 5). Second, the retaining ring technique for securing the ends of the spring was only marginally successful and may not be adequately strong for larger diameter springs and correspondingly larger loads. Third, the keyway method for attaching the spring to the inner shaft was dependent on the geometry of this specific case (.045" diameter springs attached at a 600 angle) and may be more difficult to implement with other geometries. Fourth, some difficulties were experienced during assembly that led to the cylinder design modifications made in the Version 2 prototype. Finally, the potentiometer signal conditioning circuitry was observed to be unacceptably noisy. Figure 18 shows the results of the Version 1 compliant element static displacement vs. torque test without the cam part -- the winding surface was a concentric .50" diameter concentric shaft. A slight asymmetry can be observed, but no stiffening behavior. The lack of stiffening behavior is speculated to be due to part and assembly variations. First, the shape of the spring wire forms -- both the radius of curvature and the correct location of the termination segments along the circumference are vital to the formation of the correct geometry to induce the stiffening behavior. The wire forming process leaves some variation among the springs and there may have been additional deformation caused during assembly, when the springs were integrated with the shaft. Second, any gaps between the springs and the shaft or the cylinder at the termination locations would tend to delay the appearance of the stiffening behavior. Third, any slippage of the spring terminations would prevent the spring from properly wrapping around the shaft to create the nonlinear profile. Slippage of this nature was occasionally observed with the cylinder terminations during these tests. 43 Displacement vs. Torque: Version 1 CE without Cam and without Secure Cylinder Termination 40 - 30 2010CX x 0 E a -10 - x CX X -20 -30 -40 -300 -200 -100 0 Static Torque (mNm) 100 300 200 Figure 18. Version 1 Static Test Results without Cam and without Secure Cylinder Termination Figure 19 shows the results of the Version 1 compliant element static displacement vs. torque test with the cam part. A similar asymmetry of the data still exists, but definite stiffening behavior can be observed. This plot also shows the excessively noisy range of potentiometer values measured using the original signal conditioning circuitry described in Section 3.3.4. Displacement vs. Torque: Version 1 CE with Cam, 40 Sample Potentiometer Data (Low, High, and Mean) 30 20 10 (Di ~-10 -IPI o)I -20- -30 -40 -500 -400 -300 -200 100 0 Static Torque (mNm) -100 200 300 400 Figure 19. Version 1 Static Test Results with Cam 500 44 4.4 Version 2 Compliant Element Prototype 4.4.1 Objective The Version 2 compliant element prototype was designed to the same external and interface dimensions as the Version 1 element to facilitate direct substitution in to the one degree-of-freedom test stand. It was constructed to improve on the performance of the Version 1 element in the following ways: " Increase the element stiffness by an order of magnitude and reduce the maximum displacement angle; " Improve the method for terminating the springs in the cylinder; " Investigate alternative methods for terminating the springs in the shaft; " Make modifications to facilitate easier assembly; " Reduce the level of noise in the potentiometer signals. 4.4.2 Description The Version 2 compliant element prototype shown in Figure 20 was designed to use .092" diameter springs to increase the stiffness and to decrease the maximum displacement angle. A quick calculation using the beam bending equation (Equation 19) and the cross section inertia equation for a cylinder (Equation 20) indicates that this spring should be about 17 times stiffer than the .045" springs. Equation 19. where: M -- bending moment EI E -- elastic modulus M =R I-- cross section moment of inertia R -- bending radius of curvature I =7 Equation 20. 4 4 The cylinder of the Version 2 prototype was machined out of two pieces of aluminum that are fastened together with countersunk screws. This design change was implemented to facilitate easier assembly of the final element within a joint assembly and to permit the use of a pin retention technique to secure the spring to the cylinder. Figure 20. Version 2 Compliant Element Prototype 45 A T-block retaining part (see Figure 21) was designed to retain the springs in the cylinder. This part is similar in concept to the top hat retaining device in that the spring ends penetrate the T-block, but in this case, they are retained by a pin, rather than by a retaining ring. The use of a pin is facilitated by the redesign of the cylinder to be a two piece assembly that is assembled around the shaft after the springs have been assembled with both the shaft and with the T-block. Figure 21. Two Views of the T-block The Version 2 prototype was also used to investigate the use of set screws to secure the springs to the inner shaft. Circular grooves on the spring termination segment were found to work better than flats and retained the springs as well as the pin technique. Finally, the potentiometer signal processing circuitry was modified (see Section 3.3.4) to reduce the level of noise in the output. 4.4.3 Results Static Displacement vs. Torque Test Results Figure 22 shows the static displacement vs. torque test results for the Version 2 compliant element prototype with the cam part. The data indicates that this mechanism is about 10 times stiffer than the Version 1 prototype. The maximum range of motion has also been reduced to about 2/3 of its previous value. The T-block and set screw retention techniques were found to work well and the new potentiometer circuit successfully reduced the noise levels, as can be seen in the much narrower spread of displacement values of Figure 22 compared to Figure 19. However, there still exists an asymmetry in the displacement vs. torque data. Since the asymmetry is similar to that observed for the Version 1 prototype, the most likely cause is the alignment or circularity of the cam part, which was used in both prototypes. It is desirable to correct this asymmetry so that the performance of the mechanism is the same in both directions and so that a constant calibration can be used to convert displacement measurements to torque estimates. Assuming an exponentially stiffening profile as discussed in Section 2.1, a curve fit of this displacement vs. torque data results in Equation 21. 46 ,= 270(e0. 130 - where: t is measured in mNm e is measured in degrees I) Equation 21. Equation 21 is the basis for a model of the compliant element that is used to help analyze the additional test results presented in the following sections. Displacement vs. Torque: Version 2 CE with Cam, 40 Sample Potentiometer Data (Low, High, and Mean) 20 15 - 10 - S5 0 E 'a -5 -10 -15 -201 -2500 -2000 -1500 -1000 500 0 -500 Static Torque (mNm) 1000 1500 2000 2500 Figure 22. Version 2 Static Displacement vs. Torque Test Results with Cam Torque Dynamic Range Test Results Following the methodology described in Section 2.1, the derivative of Equation 21 is examined to estimate the torque sensitivity of the Version 2 compliant element (Equation 22). To evaluate Equation 22, the minimum detectable change in displacement angle (60) is estimated to be 0.14", corresponding to 5 potentiometer counts. This choice is based on the noise levels observed (-3 counts) and on the calibration setting (36.7 counts per degree) of the potentiometer measurements. 5, = (270X0.13Xe 0)130 Equation 22. Figure 23 shows a typical sample of the data gathered to determine the torque sensitivity of the prototype under different initial load conditions. In this case, an additional incremental load of 3.7 mNm was added (in the negative measured direction) to the unloaded prototype at approximately 0.75 seconds, causing a momentary spike in the load response. The measured load is observed to stabilize at a new value (-2 mNm) that is distinct from the original value, even though it is not quite the same as the actual load added. 47 Multiple tests of the type shown in Figure 23 were used to find the minimum detectable incremental torque under different initial load conditions and are plotted in Figure 24. Also plotted in Figure 24, as the solid line, is the expected relationship between these values based on the model expressed by Equation 21 and Equation 22. This line has a constant slope of .049, indicating that at any given applied load, the torque resolution is 4.9%. Figure 24 demonstrates that the actual torque resolution was observed to be slightly better than predicted for most cases. Typica I Torque Sens itivity Test D ata Torque Sensitivty vs. Applied Torque - Predicted and Measured 650 z 40 z UX -2 30 -4 0 20- -6 E 10 -8 -10 0 0.5 1 1.5 Time (ueconds) Figure 23. Typical Torque Sensitivity Test Data 0 500 1500 1000 Applied Torque (mNm) 2000 2500 Figure 24. Version 2 Torque Sensitivity vs. Applied Torque The force dynamic range of the element, taken as the ratio of the maximum controllable force to the minimum force, can be taken from Figure 24 to be 2500:5, or about 500. While this range is quite large compared to most other robot manipulators, it is clear that it could be further increased if the noise in the potentiometer signal can be reduced further. Motor Driven Step Response Test Results Figure 25 shows the results of a typical motor driven step response test. The gearhead output angle response is the trajectory of the inner shaft of the prototype compliant element while the joint angle output is the trajectory of the cylinder and reflects the contribution of the compliant element. This test revealed that the combination of motor and gearhead inertias with the limited current capability of the test set motor power amplifier dominates the dynamic response characteristics. While it is difficult to quantify the dynamic behavior of the compliant element given that the driving function has been significantly altered from the intended step, the results do provide some insight in to its general dynamic behavior. 48 Version 2 CE Motor Driven Step Response Test Data 160 - - - - - 140 -- ointAngle Command Gearhead Output Angle Joint 0Outp ut A ng le / 120 100 - 80- S 60 0 40 20 - 0 0 / 0.1 0.2 0.3 Time (Seconds) 0.4 0.5 Figure 25. Version 2 Motor Driven Step Response Test Data Initial Offset Step Response Test Results Figure 26A shows a typical result from an initial offset step response test. Based on the settling time, a value for the damping coefficient was calculated and was used in a simulation with the exponential spring model (Equation 21) and the value of the applied load to generate plot B of Figure 26. The measured response in plot A exhibits a higher frequency oscillation than can be accounted for by this original model. Based on an observation that the hanging masses of the test set did not appear to move during the final oscillation, a modified model was developed that employs variable damping and variable inertia schemes. The results of the modified model are shown in plot C. Plot D shows the variable damping values used in the model as a function of displacement angle. Damping was set at the nominal value for displacements greater than 5' and was scaled down by a factor of 50 for displacements less than 50, except for a notch of +/- 0.05' around zero that was only scaled down by a factor of 20. The greater energy loss of this center notch can be reasonably interpreted as being caused by a "dead zone" around the zero displacement position when the springs change from wrapping to unwrapping. Plot E shows how the load inertia was assumed to vary as a function of displacement angle in the modified model. For small displacement angles, the hanging masses of the test set did not appear to move and the compliant element was observed to oscillate at a frequency consistent with an unloaded condition. Physically, this may be accounted for by some flexibility in the cable used to hang the weights and/or by the tendency for the weights to swing a little bit, rather than move straight up and down. As shown in plot E, the modeled inertia is decreased linearly from the nominal value above 12' displacement angles to the unloaded value (very small compared to the load, but not zero) at zero displacement. 49 B. Simulation -- Constant Damping, Constant Inertia A. Version 2 CE Initial Offset Step Test Data 20 15 15 2 10 10 5 c 5 -a CO0 0 -5 -5 -10 0 0.2 0.6 0.4 Time (Seconds) 0.8 016 0 '4 0.2 0 . . . 0 1 0.8 Time (Seconds) D Variable Dam png Coefficient 0 07 0 06 C. Simulation -- Variable Damping, Variable Inertia 20. E0 -5 15 2D 10 -20 2D .15 -10 5 .5 5 E. Variable X 10- 20 2 -5 0 15 35 0 .101 10 Inertia 0.2 0.4 0.6 Time (Seconds) 0.8 - 1 0 -20 -15 -10 -5 0 10 15 20 Figure 26. Version 2 Initial Offset Step Response Test Data and Simulations 4.5 Compliant Element Final Design 4.5.1 Objective The final design of the compliant element is as a component in the four degree-offreedom manipulator described in Section 5 that will be used to investigate robotic excavation techniques. This element will be integrated in to the shoulder yaw and pitch joints of the robot while a second compliant element concept developed concurrently by [Katz, 1999] will be used to add compliance to the elbow pitch and roll joints. 4.5.2 Description Figure 27 provides an exploded view of the final compliant element design. With only eight different machined parts (10 parts total), it can be regarded as a relatively simple mechanism. However, as indicated by the test results above, the performance of the mechanism requires a high degree of precision in the manufacturing and assembly of these parts. 50 Potentiometer End Cap with Bearing - T-Block Cam Shaft - Cylinder (Half) Spring Shaft Adapter End Cap with Bearing Figure 27. Final Element Design - Exploded View The final design is modeled after the Version 2 prototype with the following modifications: * The final design uses a single .25" diameter shaft and a shaft adapter rather than using a rigid coupling to join a .25" shaft to a .50" shaft; * The length of the compliant element was minimized to facilitate integration in to the CADD manipulator; * The hardware for mounting an encoder was eliminated (on the prototypes, the encoder provided a redundant, lower resolution measurement of the compliant element displacement angle to validate and calibrate the potentiometer). 51 5 Compliant Arm Design for Digging (CADD) Objective 5.1 The Compliant Arm Design for Digging (CADD) is a four degree-of-freedom robot arm designed and constructed to meet the following objectives: * Evaluate the performance of two compliant transmission strategies in a manipulator as a follow up to the one degree-of-freedom evaluations; " Investigate excavation control strategies for a manipulator that employs a nonlinearly compliant joint transmission; " Design a manipulator that is approximately interplanetary rover scale that uses readily available components when possible. Description 5.2 The kinematic configuration of the manipulator was chosen to be Yaw-Pitch-Pitch-Roll. The first three joints match the configuration of a typical industrial backhoe. The elbow roll joint was implemented to facilitate the use of oscillatory motions that have been observed to aid in excavation tasks by reducing stiction between the tool and the soil [Hong, 1999]. Figure 28 shows a CAD representation of the four degree-of-freedom manipulator and Figure 29 shows the manipulator itself. The yaw guide track on the base segment is used with the yaw compliant element to produce an additional 3:1 output speed reduction with an output range of motion of 1800. At the top of the base segment is a tapered roller bearing pack to support the rest of the robot. All four motor and gearhead assemblies are mounted on the first link that moves with the yaw joint. While it was recognized that this motor placement adds a significant amount of additional inertia that must be moved by the shoulder yaw joint, the choice is justified by the resulting simplification of the transmission and the planned use of the robot to conduct relatively slow maneuvers. A short cable transmission is used to connect the shoulder pitch gearhead output to the shoulder pitch compliant element, which is located along the axis of the shoulder pitch joint. Longer cable transmissions are used to drive the elbow pitch and roll joints through a differential assembly located at the elbow joint. The long runs of cable along the first link are necessary to accommodate the in-line compliant elements contributed by [Katz, 1999]. Not shown in Figure 28, but visible in Figure 29 are a one degree-of-freedom (pitch) end effector mechanism that will be used with a scoop to interact with the soil and a six degree-of-freedom force/torque sensor that will measure loads at the interface between the end effector and the arm link. Additionally, some preliminary designs have been examined to add counterbalance masses to the elbow and shoulder pitch joints so the compliant elements will operate around their zero displacement states when the manipulator is unloaded. 52 Figure 28. Compliant Arm Design for Digging CAD Model Figure 29. Compliant Arm Design for Digging 53 6 Conclusions 6.1 Review of Thesis Work The focus of this research was to develop a nonlinear, compact, modular, robust, and scalable compliant transmission element that can be used to add compliance to both new and existing robot manipulators with minimal redesign effort. A nonlinear, stiffening compliance characteristic was sought to increase the force dynamic range of the mechanism without unduly large displacements. To preserve positioning accuracy and to measure the applied forces, the element must have a known compliance characteristic and must provide a means to measure the displacement. The final design of the compliant element presented in this thesis exhibits the desired nonlinear stiffening behavior and a force dynamic range of about 500. However, the design appears to be overly sensitive to part and assembly tolerances. An analytical model is provided that may be used to predict the torque vs. displacement profile of While the use of a alternative spring, shaft (or cam), and cylinder diameters. potentiometer was successfully demonstrated during static tests, some concerns about noise were raised during the dynamic tests. The compliant element design has been integrated in to the design of a four degree-of-freedom manipulator that will be used to further investigate excavation control strategies using a robot with a compliant transmission. 6.2 Future Work The next step in the evaluation of the compliant element presented in this thesis is to evaluate its performance as a component of the Compliant Arm Design for Digging. Through this apparatus, various position, force, and hybrid controllers can be evaluated. Furthermore, the accuracy of the joint torque calculations made from the compliant element displacement measurements can be compared to the results indicated by an independent force/torque sensor. While this research effort has largely demonstrated the potential usefulness of the proposed wrapping spring concept, there are several issues that merit further investigation. First, efforts could be made to better characterize the effects of part and assembly tolerances on element performance. Second, the design may benefit from a rigorous evaluation of the mechanism parts to identify ways to simplify the design and the manufacturing processes. Third, while the current analytical model provides a reasonable prediction of the expected behavior of a wrapping spring mechanism, it could benefit from the addition of a component that can account for the linear behavior of the mechanism near zero displacement. An extension of this model could also be used as a design tool to recommend geometric properties (relative shaft (or cam), spring, and cylinder diameters) to meet desired maximum torque and dynamic range requirements. Fourth, the potentiometer signal noise issue must be addressed for any future application. 55 Bibliography Asada, H. and Slotine, J.-J. E. 1986. Robot Analysis and Control. New York: John Wiley & Sons, Inc. Bicchi, Antonio J., Salisbury, J. K., and Brock, David L. 1990 "Contact Sensing from Force Measurements". Artificial Intelligence Laboratory Memo 1261, Massachusetts Institute of Technology. Book, W. J. and Paul, F. ed. 1991. Modelling and Control of Compliantand Rigid Motion Systems. New York: The American Society of Mechanical Engineers. Boresi, Arthur P. and Sidebottom, Omar M. 1985. Advanced Mechanics ofMaterials, 4th Edition. New York: John Wiley & Sons, Inc. Brimley, W., Brown, D., and Cox, B. 1994. "Overview of International Robot Design for Space Station Freedom." Teleoperation and Robotics in Space, AIAA Progress in Astronautics and Aeronautics, Volume 161: 411-441. Das, Hari. 1999. "Planetary Dexterous Manipulators Task." http://robotics.jpl.nasa.gov/tasks/pdm/homepage.html. Gere, James M. and Timoshenko, Stephen P. 1984. Mechanics of Materials,Second Edition. Boston: PWS Engineering. Hong, W.J. and Salisbury, J.K. 1999. "Obstacle negotiation in robotic excavation." Submitted to the IASTED Robotics and Applications 1999 Conference. Hung, John Y. 1991 "Control of Industrial Robots that Have Transmission Elasticity." IEEE Transactionson IndustrialElectronics (Vol. 38, No. 6): 421-427. Katz, Arrin. 1999. "The Design and Application of a Nonlinear Series Compliance Actuator for Use in Robotic Arms." MS Thesis, Massachusetts Institute of Technology. Mason, Matthew T. and Salisbury, J. K. 1985. Robot Hands and the Mechanics of Manipulation. Cambridge: MIT Press. Matteo, Benjamin C. 1997. "The Design of an Elastic Element for a Modular Series Elastic Actuator." Bachelor's Thesis, Massachusetts Institute of Technology. Morrell, John B. 1996. "Parallel Coupled Micro-Macro Actuators." PhD Thesis, Massachusetts Institute of Technology. Murray, Richard M., Li, Zexiang, and Sastry, S. Shankar. 1994. A Mathematical Introduction to Robotic Manipulation. Boca Raton: CRC Press, Inc. Pratt, G. A. and Williamson, M. W. 1995. "Series Elastic Actuators." Proceedingsof the IEEE/RSJ InternationalConference on Intelligent Robots and Systems (IROS-95 Volume 1): 399-406. Readman, Mark C. 1994. Flexible Joint Robots. Boca Raton: CRC Press, Inc. 56 Shah, Vinay K. 1997. "Design and Control of a Nonlinearly Compliant Robotic Finger." MS Thesis, Massachusetts Institute of Technology. Slotine, Jean-Jacques E. and Li, Weiping. 1991. Applied Nonlinear Control. Englewood Cliffs: Prentice Hall. Spong, M. W. 1987. "Modeling and Control of Elastic Joint Robots." Journalof Dynamic Systems, Measurement, and Control (Vol. 109, December): 310-319. Spong, Mark W. and Vidyasagar, M. 1989. Robot Dynamics and Control. New York: John Wiley & Sons, Inc. Sugano, S., Tsuto, S., and Kato, I. 1992. "Force Control of the Robot Finger Joint Equipped with Mechanical Compliance Adjuster." Proceedings of the 1992 IEEE/RSJ InternationalConference on Intelligent Robots and Systems: 2005-2013. Townsend, William T. 1988. "The effect of Transmission Design on Force-Controlled Manipulator Performance." PhD Thesis, Massachusetts Institute of Technology. Williamson, Matthew M. 1995. "Series Elastic Actuators." MS Thesis, Massachusetts Institute of Technology. 57 Appendix A. Matlab Code of Analytical Wrapping Spring Model 58 % % % Wrapping Spring Model Andrew W. Curtis January, 2000 % Constants E = 200e9; rsp = (.092/2)*2.54/100; I (pi/4)*rspA4; RO = (.375)*2.54/100; MO E*I/RO; = rc = (1.0/2)*2.54/100; rsh = (.5/2)*2.54/100; rcam = ((5/8)/2)*2.54/1 00; % cO = rcam - rsh; % 0 1s0 = pi*RO; % 0 oung's Modulus for steel pring wire radius ross section moment of inertia nitial spring radius of curvature nitial state of the wire form (Pa) (m) (m^4) (m) (Nm) ylinder radius haft radius am radius enter offset between cam and shaft nitial spring free length (m) (in) % Set up shaft rotation angle theta = (0:.001:.4)'; td = theta.*180/pi; % t heta converted to degrees (in) (in) (in) (deg) % Calculate the Moment generated by the wrapping side of the mechanism % Assumption: point of contact moves in opposite direction of the shaft rotation by same angle, theta %0 a = rc - rsp; b = rsh + rsp; c = rcam+rsp; % Calculate cam effect beta = abs(asin((c0/c)*sin(pi-2.*theta))); eta = 2.*theta - beta; b2 = sqrt(c0^2 + cA2 - 2*c0*c.*cos(eta)); % Calculate the chord length %lcf = sqrt(a^2 + bA2 - 2*a*b.*cos(pi-theta)); lcf = sqrt(aA2 + b2.A2 - 2*a.*b2.*cos(pi-theta)); without cam with cam % Calculate length of spring wrapped around the cylinder % without cam %lsw = b*2.*theta; % with cam lsw = c.*eta; % Calculate new free length of spring lsf = 1s0 - lsw; % Calculate new radius of curvature for the free length of spring for i=l:length(lcf), I = = solve('''num2str(lcf(i)) [R(i),psi(i)] eval([' 2*R*(sin(psi/2))'','''num2str(lsf(i)) ' = R*psi'');']); end for i=l:length(lcf), R2(i,1) = psi2(i,l) end str2num(char(R(i))); = str2num(char(psi(i))); 59 % Calculate the moment due to the free length of spring Mwf = MO - (E*I./R2); Calculate the component of Mwf that acts as a torque at the fixed point on the cylinder %without cam %phil = abs(asin((b./lcf).*sin(theta))); %with cam phil = abs(asin((b2./lcf).*sin(theta))); alpha = ((pi-psi2)./2)-phil; Mwc = (a.*Mwf./R2).*cos(alpha); % Calculate the Moment generated by the unwrapping side of the mechanism % Approximation: Free segment of spring maintains nearly constant curvature % Calculate change in spring endpoint position due to shaft rotation: dlsp = b.*theta; % Calculate angle required to accomodate this change at the radius of curvature when in contact with the cylinder % gamma = dlsp./(a-RO); % Calculate the new free length of the spring lsfu = lso - a.*gamma; % Calculate the chord length of the free length of spring lcfu = sqrt(a^2 + b^2 -2*a*b.*cos(pi-theta-gamma)); % Calculate the actual radius of curvature for the free length of spring for i=l:length(lcfu), = eval(['[Ru(i),psiu(i)1 = solve('''num2str(lcfu(i)) 2*Ru*(sin(psiu/2))'', ''num2str(lsfu(i)) ' = Ru*psiu'');']); end for i=l:length(lcfu), Ru2(i,l) = str2num(char(Ru(i))); psiu2(i,l) = str2num(char(psiu(i))); end % Compensate for possible sign inversion of psiu2 psiu3=abs(psiu2); Ru3 = lsfu./psiu3; % Calculate the moment due to the free length of spring Muf = (E*I./Ru3) - MO; % Calculate the component of Muf that acts as a moment at the fixed point on the shaft % Assumption: The radius vector Ru is nearly the same as the radius vector b % (to the fixed point on the shaft) Mus = b.*Muf./Ru3; % Sum the two torques Mt = Mwc + Mus; % End of Wrapping Spring Model 61 Appendix B. Compliant Element Part Drawings Table 3. Index of Compliant Element Part Drawings Part Cam Quantity 1 CSpring 092 Cylinder 92 (Half) End Cap End Cap Pot Inner Shaft Shaft Adapter T-Block 2 2 1 1 1 1 1 Potentiometer End Cap with Bearing T-Block Cam Cylinder (Half) GShaft Shaft Adapter End Cap with Bearing Figure 30. Compliant Element Part Overview H DWG. NO- REV. REVISIONS ZONE REV DESCRIPTION DATE APPROVED .0625 .625 500 1875 .062 5 .375 Co noad r 1/32' Typical .150 4* . 31 2 1875 045 .280 0TY REO FSCM NO NOMENCLATURE PART OR IDENTIFYING NO UNLESS OTHERWISE SPECIFIED DIMENSIONS ARE IN INCHES ARE: FRACTIONS DECIMALS ANGLES MATERIAL SPECIFICATION OR DESCRIPTION PARTS LIST CONTRACT NO. IOLERANCES YES- DO NOT TREATMENT SCAL E APPROVALS - DRAWI NG DATE ITLE DRAWN Cor CHECKED F INISH SIZE A ISSUED SIMILAR TO AC W1 OWG NO CALC WT SCALE I FSCM NO. 4 SHEET ITEM NO SAC. REV. SH NI REVISIONS REV ZONE .092 (Wire DATE DESCRIPTION APPROVED Diameter) .12 0165 .0625 _2 093 .375 4- 30 'T 0625 -. .204 .093 rCM D? PESO PART OR IDENTIFYIN No ITEM MATERIAL SPECI FICAT ION NOMENCLAIURE OF DE SCR IPT ION PARTS LISI G NO NO CONTRACT NO. UNLESS OTHERWISE SPECIFIES DIMENSIONS ARE IN INCHES TOLERANCES ARE: FRACTIONS DECIMALS ANCLES T-. DATE APPROVALS DO NOT SCALE DRAWING TI TI.E OSpring DRAWN TRPEAT MENT CHECK<ED F INISH SIZE FSCM 092 DWO NO, NO. ISSUEDA SIMILAR TO I ACT. PT CALC PT SHEET SCALE I __ _ _ __ _ _ I __ I _ _ _ I___ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 1 2 ING. NO. 4 IREV [S REVISIONS ZONE .250 D REV DESCRIPTION DATE APPRDVED .225 D .047 0-80 T op .117 - .0625 Radius Typical -- - .300 16D 550 .102 C .100 .1 1 .500 C 100 NJ 115 4--- 25 .150 1 .125 .0 25 .062 5 B B 1 .000 .116 Thru .1875 Countersink 4-40 Clear 4 .0635 Thru .125 Countersink 0-80 Clear 7 - .047 0-80 Tap 2 Places 2 PlIa Ce S Places N oty lwaiE0 &Ni~N tENlRACT No. IN IICHE NSN3M RANCES ARE, 0IM TOLE TRACI 10" D[CIMALS NO "MCL EI 2 :% - A DO N0l PAT SCLIUSE CP DE SN? IP1 IN PARTS L IST SCALE DRAWIrNG APPRFOVALS DATEr Cylinder 92 11 2 2 rALE NO. TO Ill, "I 1 A"W, S 3 r I ATlION L O~A F INISH SIMILAR PEVI SCA LE A (Half) oEe ET 1 N HEE T 4 4 - 3H DWG. NO. REV. REVISIONS ZONE REV DESCRIPTION DATE APPROVED 1.000 625 1 .125 .313 .063 .196 4-. 100 .250 .086 Thru 4-40 Top 4 P iaces FSCM OTY REDO NOMENCLAIURE MATERIAL ITEM OR DESCRIPTION SPECIF ICATION NO PART OR IDENT IFYING NC NO C UNLESS OTHERWISE SPECIFIED DIMENSIONS ARE IN INCHES TOLERANCES ARE: FRACTIONS DECIMALS ANGLES 555DO NOT SCALE DRAWING TREATMENT C CONTRACT PARTS L IST NO. APPROVALS DATE TITLE End Cap ORAWN CHECKED SIZE F IN ISWT ISSUEDA SIMILAR TO I I ACl 1 FSCM NO. DWG NO. CADET SCALE 4 SHEET SH GWG. NO. 079 1 88 .157 REV. REVISIONS ZONE REV DESCRIPTION DATE APPROVED 438 .196 .100 188 .750 1.000 1 .000 .625 1 .500 .313 1.12 5 4.567 .125 1 .339 .086 Thru 4-40 Top 4 Places Outside Inside 0-1Y REDO | PART OR IIENTIFYINC NO FSCM NO MATERIAL SPECIFICATION CONTRACT NO. UNLESS O-HERWISE SPECIFIED DIMENSIONS ARE IN INCHES TOLERAkNCES ARE; FRACTIONS DECIMALS ANGLES . NOMENCLATURE OR OESCRIPTION PARTS LIST I APPROVALS DO NOT SCALE DRAWING DATE End Cap DRAWN TREATMFENT Pot CHECKED F INIS SIZE ISSUEDA SIMILAR TO AC' W CALC FSCM No. OWG NO. WT SCALE 4 SHEET ITEM NO DWG0 ~H PLY. I NO. NO. DWG ISH REVISIONS ZONE REV DATE DESCRIPTION IFLv. I APPROVED .093 .092 4 1 .093 .092 .125- .250 4I DrY REO FSCM NO PART OR IENTIFYINC UNLESS OTHERWISE SPECIFIED DIMENSIONS ARE IN INCHES 1OLERANCES ARE: FRACTIONS DECIMALS ANGLES .15NOT SCALE DRAWING TREATMENT DO NOMENCLATURE OR DESCRIPTION PARTS LIST NO I MATERIAL ITEM SPECIFICAlION NO CONTRACT NO. APPROVALS DATE IITLE Inner DRAWN Shaft CHECKED F INISW CHSUED 5 FSCM NO. owG NO. ISSUED97: SIMILAR TO ACT W1 CALC WT SCALE SHEET ON DWG. SH NO. IS5H REV. IHV 00 REVISIONS 092 T hru 12u 2 Places NE REV DESCRIPTION DATE APPROVED .0 93 - Groove - to allow ing assembly ______spr .093 375 .106 6-32 lap 2 Places xtX~f2K 4- .0625 45 Deg Chamfer 2 Places -P---- .125 .5 .125 .25 FSCM ETY REO NOMENCLAIURE OR DESCRIPTION PART DR IDENTAIFYINC NO NO MATERIAL SPECIFICAT ION PARTS LIST UNLESS OTHERWISE SPECIFIED DIMENSIONS ARE IN INDIES TOLERANCES ARE: FRACTIONS DECIMALS ANGLES DO NOT TREATMENT CONTRAC-T .TTSCALE DRAWING ND. DATE APPROVALS Shaft DRAWN Adapter CHECKED SIZE E INISPl FSCM NO. DWG NO. ISSUEDA SIMILAR TO ACT T CALM 1T 5CAL E I SHEETi ITEM ND SH wOG. NO. .0635 Thru .1094 Counter si nk I REV. RE VIS IONS ZONE REVI DESCR IPT DATE ION APPROVED 0-80 C Ie ar 2 Places 0625 .092 .275 .0625 0930 - 180 -. 1 .50 .360 .080 115 4- 1 .00 313 .125 .0725 .125 0 15 REDO FSCM NO NOMENCLATURE OR DESCRIPTION PART OR IDENTIFYINC NO MATERIAL SPECIFICA~ ION PARTS LIST UNLESS ITHERWI SE SPECIF lET DIMENSIONS ARE IN INCHES CONTRACT NO. ITOLERA MCIES ARE: FRACTIONS DO TREATMENT NOT DECIMALS ANGLES APPROVALS SCALE DRAWING DATE CHECKED V INISW ISSUED ISSUEDA TITLE T - BIo ck DRAWN SZE AI FSCM NO. SCAL E T ISHEE WG NO. T ITEM NO