(1993) human extenders

8/2/2019 (1993) Human Extenders

1/10

H. KazerooniMechanical Engineering Department,University of California at Berkeley,Berkeley, CA 94720Jenhwa Guo

Institute of Naval Architectureand Ocean Engineering,National Taiwan University,Taipei, Taiwan

Human ExtendersA human's ability to perform physical tasks is limited by physical strength, not byintelligence. W e coined the word "extenders" as a class of robot manipulators wornby humans to augment human mechanical strength, while the wearer's intellectremains the central control system for manipulating the extender. O ur researchobjective is to determine the ground rules for the control of robotic systems wornby humans through the design, construction, and control of several prototype experimental direct-drive/non-direct-drive multi-degree-of-freedom hydraulic/electricextenders. The design of extenders is different from the design of conventionalrobots because the extender interfaces with the human on a physical level. The workdiscussed in this article involves the dynamics and control of a prototype hydraulicsix-degree-of-freedom extender. This extender's architecture is a direct drive systemwith all revalue joints. Its linkage consists of two identical subsystems, the arm andthe hand, each having three degrees of freedom. Two sets of force sensors measurethe forces imposed on the ex tender by the human and by the environm ent (i.e., theload). The extender's compliances in response to such contact forces were designedby selecting appropriate force compensators. The stability of the system of human,extender, and object being manipulated was analyzed. A mathematical expressionfor the extender performance was determined to quantify the force augmentation.Experimental studies on the control and performance of the experimental extenderwere conducted to ve rify the theoretical predictions.

1 IntroductionRobot manipulators perform tasks which would otherwisebe performed by humans. However, in even the simplest oftasks, robot manipulators fail to achieve performance comparable to the human performance which is possible with thehuman intellect. For example, humans excel at avoiding obstacles, assembling complex parts, and handling fragile objects.No manipulator can approach the speed and accuracy withwhich humans execute these tasks. But manipulators can exceed human ability in one area : strength. The ability of a human

to lift heavy objects depends upon muscular strength. Theability of a robot manipulator to lift heavy objects dependsupon the available actuator torques: a relatively small hydraulicactuator can supply a large torque. In contrast, the muscularstrength of the average human is quite limited.To benefit from the strength advantage of robot manipulators and the intellectual advantage of humans, a new classof manipulators called "extenders" was studied [14, 15]. Theextender is defined as an active manipulator worn by a hum anto augment the human's strength. The human provides anContributed by the Dynamic Systems and Control Division for publication

in the JOURNAL OF DYNAMIC SYSTEMS, MEASUREM ENT, AND CONTROL. Manuscriptreceived by the Dynamic Systems and Co ntrol Division Janu ary 11, 1993.

intelligent control system for the extender, while the extender'sactuators provide most of the strength necessary for performing the task. Figure 1 shows the experimental hydraulic extender designed and built at the University of California atBerkeley.The key to this new concept is the exchange of both information signals and physical power. T raditionally, the exchangeof information signals only has characterized human-machineinteraction in active systems [14, 26]. But the extender is distinguished from conventional master-slave systems in an important way: the extender is worn on the human body for thepurpose of direct transfer of power. So the human body exchanges both information signals and physical power with theextender [15]. There is actual physical contact between theextender and the hu man body. Because of this unique interface,the human becomes an integral part of the extender and "fe els"the load that the extender is carrying. In contrast, in a conventional master-slave system (i.e., when there is no forcereflection), the human operator may be either at a remotelocation or close to the slave manipulator, but the human isnot in direct physical contact with the slave. The human canexchange information signals with the slave, but not mechanical power. So the input signal to the slave is derived from a

Journal of Dynamic Systems, Measurem ent, and Control JUNE 1993, Vol. 11 5/ 28 1Copyright 1993 by ASME

Downloaded 14 Jul 2008 to 128.95.205.12. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm


2/10

Fig. 1 Experimental sixdegreeoffreedom hydraulic extendernaturally in the extender. Without a separate set of actuatorsthe human arm feels the actual forces on the extender, bothdirection of motion and a scaled-down version of mass. Forexample, if an extender manipulates a 200 lbf object, the human may feel 10 lbf while the extender supports the rest ofthe load. The 10 lbf contact forces are used not only formanipulation of the object, but also for generating the appropriate signals to the extender controller. The contact forcesbetween the human and the extender, and the load and theextender are measured, appropriately modified via control theory to satisfy performance and stability criteria, and used asan input to the extender controller.The objective of this research effort is to determine the rules

a transfer function matrix representing thestability controllerthe degrees of freedom of the extendera vector; the actual position of the extender endpointa vector; the desired position of the extender endpointa matrix; the performance indexthe Laplace operatora matrix; the sensitivity transfer functionmapping load force fe to extender position,pa matrix; the sensitivity transfer functionmapping human force fh to extender position, pa vector; the human muscle force which initiates a maneuvera transfer function matrix representing thecontroller operating on the human force, fha vector, operator representing the nonlinear dynamics in the human and the loadfrequency range of operationscalarscalar constants

npPde,

K(s)

a(s)

R(s)sSe(s)

A(p)Wp

EAoo A"", {3oc,{3"", {3in

difference in the control variables (i.e., position and/or ve-locity) between the master and the slave, but not from any setof contact forces.In the extender system, the input signal to the extender isderived from the set of contact forces between the extenderand the human. These contact forces are part of the physicalforce needed to move objects, and additional ly are used togenerate information signals for controlling the extender. Ina typical master-slave system, such natural force reflection doesnot occur because the human and the slave manipulator arenot in direct physical contact. Instead, a separate set of ac-tuators are required on the master to reflect forces felt by theslave back to the human opera tor. Force reflection occurs

Nomenclature - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -All matrices and vectors are n x nand n x I, where n isthe number of degrees of freedom for the extender. The Laplace arguments of the transfer functions and the arguments

of nonlinear operators are shown in the Nomenclature but aredropped in the text.E(p) a vector, nonlinear operator; representingthe load dynamicsEo(s) a transfer function matrix; representing thelinear portion of the load dynamicsfe a vector; the force imposed on the extenderby the load

fex! a vector; any force imposed on the load byexternal objects other than the extender orthe human armfh a vector; the force imposed on the extenderby a human armf: a vector; the force imposed on the extenderby an arm at no-load condition

G(s) a matrix; the closed-loop transfer functionmapping desired position, Pde, to extenderendpoint position, PH(p) a vector, nonlinear operator representinghuman arm dynamicsHo(s) a transfer function matrix representing thelinear portion of the human arm dynamics

282/ Vol. 115, JUNE 1993 Transactions of the ASMEDownloaded 14 Jul 2008 to 128.95.205.12. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm


3/10

for the control of robotic systems worn by humans throughthe design, construction, and control of a prototype experimental extender. Section 2 reviews the h istory of devices similarto the extender. Section 3 gives an overall view of the systemsdynamic behavior. Sections 4 and 5 discuss the control systemarchitecture. Section 6 presents experimental results pertainingto the stability and performa nce of this experimental extender.Section 7 offers conclusions drawn from this effort.

2 HistoryThe concept of a device which could increase the strengthof a human operator using a master-slave system has existedsince the early 1960s. The concept was originally given thename "man-amplifier." The man-amplifier was defined as atype of manipulator which greatly increases the strength of ahuman operator, while maintaining the human's supervisorycontrol of the manipulator.In the early 1960s, the Department of Defense was interestedin developing a powered "su it of arm or' ' to augment the liftingand carrying capabilities of soldiers. The original intent wasto develop a system which would allow the human operatorto walk and manipulate very heavy objects. Research was donefor the Air Force in 1962 at the Cornell Aeronautical Laboratory to determine the feasibility of developing a master-slavesystem to accomplish this task. This study found that duplicating all human motions would not be practical, and thatfurther experimentation would be required to determine whichmotions were necessary. It was also determined tha t the mostdifficult problems in designing the man-amplifier were in theareas of servo, sensor and mechanical design [4]. Further workat the Cornell Aeronautical Laboratory determined that anexoskeleton having far fewer degrees of freedom than the human operator would be sufficient to carry out most desiredtasks. In 1964, Cornell developed a preliminary arm and shoulder design. This design demonstrated that the amplifying ability of the manipulator was limited by the physical size of thehydraulic actuators required to amplify the human operator'sstrength [22, 23, 24].

From 1966 to 1971, General Electric completed work on theman-amplifier concept through prototype development andtesting. This man-amplifier, known as the Hardiman, was asystem of two overlapping exoskeletons worn by the humanoperator and was designed as amaster-slave system. The masterwas the inner exoskeleton, which moved due to the motionsof the oper ator. The slave was the hydraulically-actuated outerexoskeleton, which followed the motions of the m aster. Thus ,the slave exoskeleton also followed the motions of the humanoperators [9, 10, 11, 20].

It is important to note th at previous systems (described above)operated based on the master-slave concept, rather than onthe direct physical contact between human and manipulatorinherent in the extender concept. Unlike the Hardiman andother man-amplifiers, the extender is not a master-slave system(i.e., it does not consist of two overlapping exoskeletons.)There is no joystick or other device for information transfer.Instead, the human operator's commands to the extender aretaken directly from the interaction force between the humanand the extender. This interaction force also helps the extendermanipulate objects physically. In other words, informationsignals and power transfer simultaneously between the humanand the extender. The load forces naturally oppose the extendermotion. The controller developed for the extender translatesthis interaction force signal into a motion command for theextender. This allows the human to initiate tracking comm andsto the extender in a very direct way. The concept of transferof power and information signals is also valid for the load andextender. The load forces are measured directly from the interface between the load and the extender and processed bythe controller to develop electronic compliancy [13, 16, 17,

! \i *y' 3U3 H1 J ..

TE

loa r

i + J 1s!f *oext

p 1

f hextender

Sh(s); r i i ! n.cSc(s)

G(s) a*i'' !

f

performance''filter a ( s )1 1 W +

K(s)stabilityontrolle

aaa ( ^

r1 +

Fig. 2 The overall block diagram for the extender. The extender dynamics, which are linearized by the primary stabilizing controller, arerepresented by G, Sh an d Se. The human and the load dynamics arerepresented by two nonlinear operators, H and Two linear controllers(a and K) modulate the forces fh an d fe.

and 18] in response to load forces. In other w ords, informa tionsignals and power transfer simultaneously between the loadand the extender [14, 15].3 Modeling

This section models the dynamic behaviors of the extender,the human arm and the load being maneuvered; these modelsare combined in Fig. 2. It is assumed that the extender primarilyhas a closed-loop position controller, which is called the primary stabilizing controller. The resulting closed-loop systemis called the primary closed-loop system. The design of theprimary stabilizing controller must consider the following threeissues.(1) Exact dynamic models for the extender are difficult toproduce because of uncertainties in the dynamics of theextender actuators, transmissions and structure. Theseuncertainties become a major barrier to the achievementof the desired extender performance, especially whenhuman dynamics are coupled with the extender dynamicsin actual machine maneuvers. The extender's primarystabilizing controller minimizes the uncertainties in theextender dynamics and creates a more definite and lineardynamic model for the extender. Therefore, it is assumedthat the dynamics of the extender are linearized by theprimary stabilizing controller over a range of operation.This linear model may then be used to design othercontrollers that operate on forces//, (i.e., the humanforce) and/e (i.e., the load force). For the experimentalextender employed in this research effort, the computed-torque m ethod along with a PD controller [1, 2] is usedas the primary stabilizing controller to create a moredefinite and linear dynamics for the extender.

(2) Extender stability must be guaranteed when the humanis not maneuvering the extender. This is a very imp ortantsafety feature: when the human separates his/her handfrom the extender in emergency situations, the primarystabilizing c ontroller must hold the extender stationaryat the configuration at which the human arm separatedfrom the extender.(3) The design of the primary stabilizing controller must letthe designer deal with the effect of the extender uncertainties without concern for the dynamics of the humanopera tor. The human arm dynam ics, unlike the extenderdynamics, change significantly with each human and alsowithin one person over time [12, 25]. Considering thecontrol difficulties arising from the human and loadnonlinear dynam ics, it is a practical matter to make everyJ ou rna l o f D yn a m i c Sy s t em s , M eas u rem e n t , and C o n t ro l J UN E 19 9 3 , Vo l . 115 / 2 83



4/10

effort in developing a linear dynamic behavior for theextender.The selection of the primary stabilizing controller is notdiscussed here; a variety of controllers may be used to stabilizethe extender in the presence of unc ertainties and nonlinearities.These controllers result in uncoupled and linearized closed-loop behavior for the extender within a certain frequency range[28].Regardless of the type of primary stabilizing controller, theextender position, p, results from two inputs: (1) the desiredposition command, />des, and (2) the forces imposed on theextender. The transfer function matrix G represents the primary closed-loop position system which maps pde s to the extender position, p. Two forces are imposed on the extender:/;, is imposed by the human, and fe is imposed by the load.Si,, an extender sensitivity transfer function, maps the humanforce, //,, onto the extender position, p. Similarly, Se, an extender sensitivity transfer function, maps the load force, fe,onto the extender position, p. If the primary stabilizing controller is designed so that Si , and Se are small, the extender hasonly a small response to the imposed forces fh and fe. A high-gain controller in the primary stabilizing controller results ina small Sh and Se and consequently a small extender responseto /A and/ e . Using G, S e and Sh, Eq. (1) represents the dynamicbehavior of the extender.

p = Gpde s + Shfh + Sefe (1)The middle part of the block diagram in Fig. 2 representsthe extender model (i.e., Eq. (1)) interacting with the humanand the load. The upper left part of the block diagram represents the human dynamics. The human arm's force on theextender, //,, is a function of both the human muscle forces,uh, and the position of the extender, p. Thus, the extender'smotion may be considered to be a position disturbance occurring on the force-controlled human arm. If the extender isstationary (i.e.,p = 0), then the force imposed on the extenderis solely a function of the hum an muscle force command produced by the central nervous system. Conversely, if the extender is in motion and uh = 0, then the force imposed onthe extender is solely a function of the human arm impedan ce,H(p). H is a nonlinear operator representing the human armimpedance as a function of the human arm configuration; His determined primarily by the physical properties of the humanarm [5 ,19,27]. Based on the above, Eq. 2 represents a dynamicmodel of the human arm.

fh = uh-H(p) (2)The specific form of Ui, is not known other than that it resultsfrom human muscle force on the extender. A simple study ofhow the central nervous system generates the desired forcecommand uh is given in [6, 8]. The experimental procedure tomeasure H from various subjects is given in Section 6.It is assumed that the extender is maneuvering a load. Theload force impedes the extender motion. The extender controller translates the two measured interaction forces ( i.e., thehuman forces and load forces) into a motion command forthe extender to create a desired relationship between the huma nforces and the load forces. E is a nonlinear operator representing the load dynamics. / ex t is the equivalent of all theexternal forces imposed on the load which do not depend onp and other system variables. Equation (3) provides a generalexpression for the force imposed on the extender, fe, as afunction of p.

fe=~E(p)+fta (3)In the example of accelerating a point mass m along a horizontal line, the load force, fe, can be characterized by fe= m^p. In this case E = ms? and/ex t = 0 where p is the massposition and 5 is the Laplace op erator. If the load is large and

cannot be represented by a point mass, then E can be calculatedusing Lagrangian formulation.The diagram of Fig. 2 includes two linear controllers, a(s)and K(s), which modulate the forces//, and /,, a and K (whichare implemented on a computer) must be designed to producea desired performance in the extender system; this is describedin the next section. As the Fig. 2 block diagram shows, theperformance filter a lets designers choose the approp riate performance for the extender, and the stability filter K (whichoperates on both fh and /, ) guarantees the system stabilitywhen the extender is used by people with various arm impedances (strengths).4 Control

To understand the role of controllers a and K, assume fora moment that neither controller is included in the system. Ifthe commanded position, /? des , the human muscle forces, uh,and the external forces,/ex t, all equal zero, then the extenderposition, p, equals zero, and no motion is transmitted to theload. T his is the case when the h uman is holding the extenderwithout intending to maneuver it. If the human decides toinitiate a maneuver, then uh takes on a nonzero value, and anextender motion develops from /,. The resulting motion issmall if Sh is small. In other words, the human may not haveenough strength to overcome the extender's primary closed-loop controller.To increase the human's effective strength, the extender'seffective sensitivity to //, must be increased by measuring thehuman force, //,, and passing it through the controllers a andK. Figure 2 shows that GKa, parallel with Sh, increases theeffective sensitivity of the extender to// , . To retain a sense ofthe load in the extender operating, the load force, /


5/10

dynamic cha racteristics of the free ma ss, yet the load essentially"feels" l ighter.With these heuristic ideas of system performance, the extender perform ance is captured in Eq. (4) where / / ,* is thehuman force applied to maneuver the extender when no loadis present. R is the performance matrix, and [0, wp] is thefrequency range of the human arm motion.ifh-fh*)=-Rfe for all we[0,Wp ] (4)

Equality 4 guarantees that ( / / , - / / ,* ) , the por t ion of the huma nforce that is actually applied to maneuver the load, is proportional to the load force, fe. The performance matr ix R isan n x n l inear transfer function matrix. Suppose R is chosenas a diagonal matrix with all members having magnitudessmaller than unity over some bounded frequency range, [0,u> p\. Then the human force is smaller than the load force bya factor of /? . Suppose R is chosen as a diagonal matrix withall members having magnitudes greater than unity. Then thehuman force is larger than the load force by a factor of R. Ina more complex example, the transfer function matrix R m aybe selected to represent l inear passive dynamic systems (i .e.,combinations of dampers, springs and masses). The frequencyrange [0, wp], implies the desired frequency range in which thedesigners wish to operate the extender (i .e., human motion).Specifying oi p allows the designers to achieve equality (4) onlyfor a bounded frequency range; there is no need to achieveEq. (4) for an infinite bandwidth. Establishing the set of performance specifications described by Eq. (4) gives designers achance to express what they wish to have happen during amaneuver. Note that Eq. (4) does not imply any choice ofcontrol technique for the extender. We have not even said howone might achieve the above performance specification. Equation (4) only allows designers to translate their objectives intoa form that is meaningful from the standpoint of control theory.

By inspecting Fig. 2, the extender position is written as afunction of//, an d/, , .p=(GKa + Sh)fh+(GK+Se)fe (5)

Now suppose that the human m aneuvers the ex tender th roughthe same trajectory indicated by p in Eq. (5), except withoutany load . The no- load hu man force, / , * , i s then obtained byinspection of Fig, 2 where E = 0 a n d / e x , = 0:

p=(GKa + Sh)fh* (6)Equating the trajectories from Eqs. (5) and (6) results in Eq.(7).

( / * - / * ) = - {GKa + Shy\GK+Se)fe (7)Comparing Eqs. (4) and (7) shows that to guarantee the performance represented by R in Eq. (4), inequality (8) must besatisfied.oma x[{GKct + Shr[(G K + Se)-R] for all we[0,iop] (9)

Inequality (9) suggests that, since e is a small number, thedesigner must choose K to be a transfer function matrix withlarge magnitude to satisfy inequality (9) for frequencies co[0,up] and for a given e, R, Sh and Se. The smaller e is chosen tobe, the larger K must be to achieve the desired performance.K may not be arbitrarily very large: the choice of K must alsoguarantee the closed-loop stabili ty of the system shown in Fig.2, as discussed in the next section.5 Stability

It has been shown that, to achieve the system performanceindicated by R, a must be equal to R~ ' and K must be a largetransfer function matrix satisfying inequality (9). However,the designers must realize that the closed-loop system of Fig.2 must remain stable for these choices of a an d K. The selectionof K is particularly important since H and E generally containnonl inear dynamic compo nents . For example, the human armimpedance H changes from person to person and also withinone person over t ime. The load dynamics E is also a nonlinearelement, as discussed earlier. Compared to the human armand load dynamics, the extender dynamics (G, Se, Sh) ar egenerally well-defined due to the primary stabilizing co ntroller.Consequently, this analysis focuses on designing a stabilizingcontroller K in the presence of all bounded variations of thenonlinear operators H and E with G, Se and Sh being knownand linear dynamics. The following questions i l luminate ourapproach to the design of K.

In designing K, is it possible to work with a hu manand a load with linear dynamics (represented by H 0an d EQ ) instead of the nonlinear dynam ics representedby operators H and E? If the answer is yes, then whatproperties should H 0 and E0 have so that the designedK both satisfies inequality (9) and stabilizes the Fig.2 system in the presence of a family of nonlinearoperators H and E?The block diagram of Fig. 2 is transformed into the blockdiagram of Fig. 3 in order to group the nonlinear oper ators,H and E, into one block on the diagram.

Sh and Se represent the sensitivities of the extender positionto the human and load forces. G, Sh and Se depend on thenature of the extender 's primary stabilizing controller. If aprimary stabilizing controller with a large position loop-gainor integral control is chosen to insure sm all steady-state errors,then S/, and Se are extremely small compared to G, approachingzero at steady-state. Prototype extender designs have producedsensitivities on the order of 103 to 106 t imes smaller than G[15]. If the extender actuators are non-backdrivable, as insystems with geared large transmission ratios, then Sh and Seare zero, regardless of how carefully the extender 's primarystabilizing controller is chosen. Since Sh an d Se are typicallymuch smaller than G, their effect on the overall system dynamics is negligible. Therefore Sh and se are disregarded in thefollowing stability analysis. Figure 4 presents the resulting sim-

- f ' ' ' ' 'DO! E 3

( y* 1Fig. 3 Sim plified block d iagram of Fig. 2. / is a unity m atrix.Journal of Dynamic Systems, Measurement, and Control JUNE 1993, Vol. 115/285



6/10

ocu, + f ext^v K -J n PFig. 4 In systems with very small sensitivity transfer functions, Se andSh are much smaller than G and their effect on the overall system dynamics is negligible

" U h + f e x l ,

H+ E 0

* ( " ) - * Gc = [ I + G K (a H 0+ E0)]"1 GK

Fig. 5 A is the stable nonlinear operator representing the nonlinearhuman and load dynamics

plified stabili ty diagram. Note that the stacked nature of theinput signal, the h u m an and load dynamics in Fig. 4 are nowreduced to single entities where (aH + E) is a nonl inear operator , and K and G are transfer function matrices.If H0 and E0 are assumed to represent a particular set ofl inear human and load dynamics , Eq. 10 represents the generalform of H and E where A is the stable nonlinear part of thedynamics. In cases where the nonlinear dynamics may not bedirectly separated from the l inear dynamics, anestimate of H0and E0 must be formed from experiments such as those inSection 6.

aH(p)+E(p) = [aH0 + E0]p + A(p)Note that [aH0 + E0] is a transfer function matrix operatingon p. The block diagram of Fig. 4 can be t ransformed in tothe block diagram of Fig. 5 which separates the nonlinear termA from the l inear terms [aH0 + E0]. With respect to Fig. 5,the design of K is approached as follows. A stabilizing controller K must be designed for a set of linear H0 and E0 so thatthe closed-loop system of Fig. 5 remains stable and inequality(9) (indicating the performance) is satisifed 2. This controllermust guarantee enough stabili ty robustness for all boundedvalues of H and E. Therefore, the goal is to find a particularclass of H0 and 0 and a stabilizing controller K that together

yield the largest stabili ty robustness for a given largest A,Equat ion 11 represents the forward loop in Fig. 5(b).Gc=[I+GK(aH0 + E0)]- lGK (11)

The stabili ty of each element in Fig. 5(b) is described byinequalities (12), (13), and (14):l l a +/ e x = /3 in


7/10

Xdirectionforce sensor

Fig. 6 The extender arm has three degrees of freedom. The upper armisdriven by actuator A 2, while the forearm is driven by actuator A3 . Bothactuators 2 and 3 are located on a bracket which rotates in response tothe rotation of actuator A1.

human arm

load \ glovegripper to grasp the loadFig. 7 The extender hand has three degrees of freedom. Actuator A4twists the whole arm while actuators AS drives the axis of wrist flexionand extension. Actuator A6 opens and closes the gripper.

piezoelectric force sensor that measures the first three components of human force, fh.The extender hand mechanism is shown in Fig. 7; it hasthree degrees of freedom. The need to minimize operator fatigue requires that the extender hand be easily integrated withhuman hand postures. Actuator A5 drives the axis of wristflexion and extension. For tasks that require secure grasping,humans apply opposing forces around the object. Opposingforces in the extender hand are created by motion of the thumbrelative to the wrist via actuator 6. A5 and A6 are both locatedon a bracket connected to actuator A4 which twists the wholehand mechanism. The human hand is located in a glove whichhas force sensors mounted at the interfaces between the handand the glove for measuring the human force,/A, in six directions. Figure 8 shows the exact points at which the human isin contact with the extender glove for the measurement of thesecond three components of//,. A thin elastic material is usedat these contact points for the comfort of the human hand.The human operator can adjust the tightness of this glovesimply by moving his/her elbow position horizontally. Severalforce sensors, not shown in the figure, are also mounted atthe gripper to measure the load force, fe, in six directions. Thematerials and the dimensions of the extender components arechosen to preserve the structural-dynamic integrity of the extender. Each link is machined as one solid piece rather thanas an assembly of smaller parts. The links are made of highstrength 7075 aluminum alloy to reduce the weight of theextender.5

The extender's primary closed-loop controller measures angular position using encoders. The extender linear model hasbeen resulted because of use of a computed torque method forcancellation of the extender gravity and coriolis forces and aPD controller for stabilization and robustness. (For brevity,the selection of the primary stabilizing compensator (i.e., computer torque method and PD controller) is not discussed here.See Reference [2] for a detailed description of such a control

'This article is dedicated to the dynamics and control of the extender as itinteracts with the human and the load. (The details kinematic, dynamic, andstructural analysis of the extender is not being described here in this article. )

finger attachment pointforearm attachment pointthumb attachment point

Fig. 8 The human hand is constrained by the glove. Any intent tomaneuver by the human arm results in measurement of the human forcesby the force sensors in the glove.

Fig. 9 The experimental extender used to verify the analysis. Bothforces fh and f, have six components.

101frequency (rad/s)

Fig. 10 Theoretical and experimental frequency response of the primary closed-loop system from input command to extender position

method.) Employing this primary controller, the extenderclosed-loop dynamics in the Cartesian coordinate frame (Fig.9) in all directions can be described by Eq. (17). This equationhas been verified experimentally and its magnitude is shownin Fig. 10 within a bounded frequency range. The 30 rad/sbandwidth in the plot implies that the extender does not respond to frequencies after 30 rad/s (about 5 Hertz).

G(s)={ g2 1.565 ft/ft (17)\3158 18.28-+1.565

Journal of Dynamic Systems, Measurement, and Control J U N E 1993, Vol. 115/287Downloaded 14 Jul 2008 to 128.95.205.12. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
http://0.0.12.86/http://0.0.12.86/http://0.0.12.86/


8/10

Note that we have two concerns in achieving the bandwidthof the primary stabilizing controller: (1) hum an arm frequencyrange of operation, and (2) extender unmodeled dynamics frequency range. Note that the stabili ty robustness to high frequency unmodeled dynamics in the extender requires smallgain for the primary stabilizing controller at high frequencieswhile tracking the human arm motion requires large gain forthe primary stabilizing controller within the frequency rangeof human arm maneuvers. We will learn in the next sectionthat the normal human arm maneuvers contain frequency components of about up to 3.Hertz. Therefore the primary stabilizing controller loop has been shaped to cross over atfrequencies less than the frequency range of extender unmodeled dynam ics. This resulted in 5 He rtz bandw idth for theextender which is sufficient for various norma l mane uvers. T oachieve a wide bandwidth for the extender, designers shouldhave a good model of the extender at high frequencies andconsequently, a weak set of stabili ty robustness specificationsat high frequencies. It will be shown later that the 5 Hertzbandwidth will become the major impediment in achieving thedesired force amplification at high frequency maneuvers.The primary stabilizing controller also develops very smallsensitivity for the extender. Equations (18) and (19) show theextender sensitivity in response to a human force and the loadforce in actuator Al only. The small DC gains given by Eqs.(18) and (19) imply strong no n-backd rivabili ty; this has resulted

from the large supply pressure and large PD gains for theprimary stabilizing controller. Since the sensitivity of all actuators is within the range given by Eqs. (18) and (19), the restof the experimental analysis assumes Sh = 0 and Se = 0. SeeReference [21] for dynamics of the servovalves and hydraulicro tary actuators .

frequency (irad/sec)Fig. 11 The experimental and theoretical plot of H; the human operatorholds the extender loosely

101frequency (rad/sec)Fig. 12 The experime ntal and theoretical plot of H; the human operatoris holding the extender tightly

S(s) = 8 5 . 0 x 1 0 "

S e( s ) = 1 5 1 x 1 0 -

18.' ; + ls2 s+ TT- r r+1 .5 6 53158 18.28

18.6 + 1f s+ 7 TT ^ + L5 6 53158 18.28

rad/lbf (18)

rad/lbf (19)

Human Arm Dynamics. The human arm dynamics do notaffect performance (defined by inequality (9)) but do play amajor role in extender stabili ty. Several experiments were conducted to measure the human arm impedance represented byH. Then the largest of these impedances was chosen to be H 0in the stabili ty analysis. In the exp eriments, the subject graspeda handle on the extender. The extender was commanded tooscillate via sinusoidal functions. (See Reference [3] for moredetailed information about the nature of such experiments.)At each oscillation frequency, the human operator tried tomove his/her hand to follow the extender so that zero contactforce would be maintained between the hand and the extender.The human arm, when trying to maintain zero contact forces,cannot keep up with the high-frequency motio n of the extender.Thus, large contact forces and consequently a large H ar eexpected at high frequencies. Since this force is equal to theproduct of the extender acceleration and human arm inertia(New ton's Second Law ), at least a second-orde r transfer function is expected for H at high frequencies. At low frequencies(in particular at DC), the operator can follow the extendermotion comfortably, and can establish almost constant contactforces between the hand and the extender. Thus, small contactforces at all extender positions and consequently a constanttransfer function for H are expected at low frequencies.Figures 11 and 12 show the experim ental values and thefit ted transfer functions for two different experiments. In the

first set of experiments (Fig. 11), the subject holds the extenderhan dle loosely. In the seco nd set of experim ents (Fig. 12), thesubject holds the extender very tightly, at low frequencies, thehuman arm impedance is smaller when the subject holds thehandle loosely than when the subject holds the handle verytightly. Also note that both plots show a crossover at about20 rad/s; this is in the neighborhood of frequencies at whichthe central nervous system can no longer keep up with theextender motio n. The largest imped ance, th at of Fig. 12, ischosen for use in the stability analysis.

H * = 1 2 ' l { < L + s k + 1 ) M/ft (20)Performance. The expe rimental extender is capable of l ifting of objects up to 500 lb when the supply pressure is set at3000 psi. Since the high frequency maneuvers of a 500 lb loadis rather unsafe, the experimental analysis on the extenderdynamic behavior was carried out at low level of force amplification. In order to observe the system dynam ics within theextender bandwidth, in particular the extender instabili ty, thesupply pressure was decreased to 800 psi and low force amplification ratios were chosen for analysis. This allows us to

maneuver the extender within 2 Hz. Matrix R in Eq. (21) ischosen as the performance matrix in the Cartesian coordinateframe.i ? " ' = a = 5 00 7 for all co [0,5 Hertz] (21)

The above performance specification has force amplificationsof seven times in the jy-direction and five times in the x-direc-tion. Figure (13) depicts the history of the extender positionalong the x-direction as a function of t ime in an experimentwhere the human operator maneuvers the extender irregularly(i .e., random ly). Figure 14 shows the experimental values ofthe human force, / / , , and the load force, fe, in the x-direction.

2 8 8 / V o l . 115, J U N E 1 9 9 3 Transactions of the ASMEDownloaded 14 Jul 2008 to 128.95.205.12. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm


9/10

5 6 7Time (Seconds)Fig. 13 Extender position, x, shown in Fig. 9

5 6 7Time (Seconds)

Fig. 17 Extender position, y, measured from horizontal

5 6 7Time (seconds)Fig. 14 Experimental hum an and load forces along the x direction.Force amplification ratio is five.

5 6 7 8 9 10Time (seconds)

Fig. 18 Experimental human and load forces along the y direction.Force amplification ratio is seven.

30

20

i f ih*" 0o' -1 0ii

-2 0 experimenttheory-3010-1 101 10!Frequency (rad/s)

Fig. 15 Theoretical and experimen tal force amplification ratio alongthe x direction

.-a

101 102 103frequency (rad/s)

Fig. 19 Theoretical and experime ntal force amplification ratio alongthe y direction

fh-fhdbf)Fig. 16 Load force versus human force along the x direction. Slope isapproximately five.

Figure 15 shows the FFT offe/(fh - fh*) along thex-directionwhere the load force fe is more than (//, - f* ) by a factorof five. It can be seen that this ratio is preserved only withinthe extender ban dwidth. Figure 16 shows fe versus (fh - fh")

along the ^-direction w here the slope of +5 represents theforce amplification by a factor of five. Figures 17 through 20are similar to Figs. 13 through 16, except that they have aforce amplification of seven in the ^-direction. Figure 18 showstwo graphs versus time, (f h - fh*) an d /e along the ^-directionwhere the load force, fe, is more than (f h - //,*) by a factorof seven. Figure 19 shows the FFT of /,, /( //, - fh*) along the.y-direction where the amplification of seven is preserved withinthe extender band width. Figures 21 and 22 show the extenderposition, human force and extender force in the /-directionwhen a = - 19 which violates the stability criteria in inequality(15).7 Summary and Conclusion

This article describes the dynamics of human machine interaction in robotic systems worn by humans. These robotsare referred to as extenders and amplify the strength of thehuman operator, while utilizing the intelligence of the operatorto spontaneously generate the command signal to the system.Extenders augment hum an physical strength. System perform-Journal of Dynamic Systems, Measurement, and Control J UN E 1 9 9 3 , Vo l . 115 / 2 89



10/10

0 2 Afh-fhdbf)

Fig. 20 Load force versus human force along the y direction. Slope isapproximately seven.

0.1 0.2 0.3 0.4 0.5 0.6 0.7Time (Seconds)Fig. 21 Extender position, y, measured from horizontal, unstable maneuver

0.3 0.4 0.5Time (seconds)

Fig. 22 Experimental human and load forces. Force amp lification ratiois nineteen.

ance is defined as a linear relationship between the humanforce and the load force. A condition for stability of the totalsystem (Extender, hum an and the load) is derived, and, throughexperimentation, the performance is demonstrated. A six-de-gree-of-freedom extender has been built for experimental verification of the analysis.

References1 An, C . H. , Atkeson, C . G. , and Holle rbach, J. M., "Model-Based Controlof Robot Manipula tor ," MIT Press , 1988 .2 Asada , H ., Slotine, J. E., "Robot Analysis and Control , " Wiley, NewYork, 1986.3 Berthoz, A. , Metra l , S. , "Behavior of Muscular Group Subjected to aSinusoidal and Trapezoidal Variation of F or c e , " J. of Applied Physiol., Vol.

29, pp. 378-384, 1970.4 Cla rk, D. C , et a l . , "Explora tory Invest iga t ion of the Man-AmplifierConcept ," U.S . Air Force AMRL-TDR-62-89, AD-390070, Aug. 1962.5 Cooke , J. D. , "Dependence of Human Arm Movements in Limb Mechanica l Prop er t ies ," 'Brain Res., Vol. 165, pp. 366-369, 1979.6 Cooke , J. D. , "The Organiza t ion of Simple , Ski l led Movements ," Tu-torials in Motor Behavior, G. Stelmach and J. Requin, eds., Elsevier, Amsterdam, 1980.7 Desoer, C. A., Vidyasagar, M., "Feedbac k Systems: Input-o utput P roperties," Academic Press, 1975.8 Fe ldman, A. G. , "Func t iona l Tunning of the Nervous System With Control of Movement or M a in te na nc e , " J. of Biophysics, Vol. 11, pp. 565-578 .9 GE Company, "Exoske le ton Prototype Projec t , F ina l Repor t on PhaseI," Report S-67-1011, Schenectady, NY, 66.10 GE Company, " Hard iman I Prototype Project, Special Interim Study,"Report S-68-1060, Schenectady, NY, 1968.11 Groshaw, P . F . , "Hardiman I Arm Test , Hardiman I P r o to type , " R e po rtS-70-1019, GE Company, Schenectady, NY, 1969.12 Houk, J. C , "Neura l Control of Muscle Length and Tension," MotorC ontrol, V. B. Brooks, ed., American P hysiological Society Handb ook of Physiology, Bethesda , MD.13 Kazerooni, H., and Waibel, B. J. , "On the Stability of the Constra inedRobotic Maneuvers in the Presence of Modeling Uncer ta int ies ," IEEE Transaction on Robotics and Automation, Vol. 7, N o. 1, Feb. 1991.14 Kazerooni, H., "H um an-R ob ot Interaction via the Transfer of Power andInformation S igna ls ," IEEE Transactions on Systems and C ybernetics, Vol. 20,

No . 2, Mar. 1990.15 Kazerooni, H., and Mahoney, S. , " Dyna m ic s and Control of RoboticSystems Worn By H u m a n s , " A S M E JOURNAL OF DYNAMIC SYSTEMS, MEASUREMENT, AND CONTROL, Vol. 133, No. 3, Sept. 1991.16 Kazerooni, H., Sher idan, T. B., a nd Houp t , P. K., " F unda m e n ta l s ofRobust Compliant Motion for M a nipu la to r s , " IEEE Journal of Robotics andAutomation, Vol. 2, N o . 2, June 1986.17 Kazerooni, H., Waibel, B. J., and Kim, S . , "Theory and Exper iments onRobot Compliant Motion Control , " ASME JOURNAL OF DYNAMIC SYSTEMS,MEASUREMENT, AND CONTROL, Vol. 112, No. 3, Sept. 1990.18 Kazerooni, H., " O n the Contact Instability of the Robots When Constrained by Rigid Environments ," IEEE Transactions on Automatic C ontrol,Vol. 35, No. 6, June 1990.19 Marsden, C. D., Mer ton, P. A., and Mor ton, H. B., "Stretch Reflexesand Servo Actions in a Variety of Huma n M uscl es," Journal of Physiol, London ,Vol. 259, pp. 531-560.20 Makinson, B. J., "Research and Development Prototype for MachineAugmenta t ion of Human S trength and Endurance , Hardiman I Projec t , " Repor tS-71-1056, GE Company, Schenectady, NY, 1971.21 Merritt, H . E. , "Hydraul ic Control Systems," John Wiley, 1967.22 Mizen, N. J., "Pre l iminary Design for the Shoulders and Ar m s of aPowered, Exoske leta l S truc ture ," C orne l l Aeronaut ica l Labora tory, Repor t No.VO-1692-V-4, 1965.23 Mosher, R. S., "Force Ref lec t ing Elec trohydraul ic Servomanipula tor ,"Electro-Technology, pp. 138, Dec. 60.24 Mosher , R. S., " H a n d y m a n to Hardiman," SAE Repor t No. 670088 .25 Repperger, D. W., Morr is , A., "Discr iminant Ana lysis of Changes inHuman Muscle Function When Interacting with an Assistive Aid," IEEE Transaction on Biomedical Engineering, Vol. 35, No. 5, May 1988.26 Sheridan, T. B., and Ferrell, W. R., Man-Machine Systems: Information,C ontrol, and Decision, M odel of Human Performance, MIT Press, 1974.27 Stein, R. B., "What Muscles Variables Does the Nervous System Controlin Limb Movements? ," J. of the Behavioral and Brain Sciences, 1982, Vol. 5,pp. 535-577.28 Spong, M. W. , and Vidyasagar , M. , "Rob ust Nonl inear Control of RobotM a n ipu la to r s , " Proc. 24th IEEE C DC , Ft. Lauderdale, FL, Dec. 1985.

2 9 0 / V o l . 115, JUN E 1993 T rans ac t ions of t h e ASM E

(1993) human extenders

Documents