1025
IH.!' TRANSACTI ONS ON BIOMEDICAL ENGINEERING. VOL. 37. 1\0. II. I\'OVEMBER 1990
Functional Assessment of Control Systems for Cybernetic Elbow Prostheses-Part
I:
Description of the Technique CARY
Abstracl-This
paper (Part
I
of
II)
J.
ABUL-HAJ
presents a novel control scheme
with which an amputee commands an elbow prosthesis using myoelec tric activity. By mimicking some important characteristics of the intact neuromuscular system. the proposed controller attempts to make the prosthesis respond as the natural elbow to both voluntary commands from the amputee and applied moments from the environment. Also presented is the description of a novel experiment for functionally as sessing elbow prothesis controllers. The experimental design calls for an amputee to perform a constrained motion task while operating a prosthesis capable of implementing a wide variety of controllers. Due to the natnre of the constraint, the task emphasizes the prosthesis re sponse to both inputs: voluntary commands and external moments. Application of the experiment to assessment of the proposed control scheme and the control scheme used in a state-of-the-art prosthesis is presented in Part
II.
l. iNTRODUCTION H E state of the practice in artificial limbs reflects a long-term deficiency of handicapped research. Today, the most common upper-extremity p rosthesis in use is the conventional cable-operated arm [41], [67], which was designed by aeronautical engineers after the Second World War. Since then the design of the device has changed very little. This prosthesis still utilizes power from the body I rather than external power from an actuator, and hence. does not represent the many recent advancements in elec tronics and actuator technology. Power and control of the cable-operated arm are derived from shoulder movement. In the case of the above-elbow prosthesis, rounding of the shoulders gives both flexion of the elbow and prehension of the terminal device (artificial hand) with a s ingle cable harnessed across the shoulders. Normally, the shoulder movement flexes the artifi c ial e lbow. To operate the ter minal device, the amputee first must lock the elbow via a mechanism that is engaged by shrugging the shoulder.
T
Manuscript received February 13, 1989; revised J anua ry 22,
1990.
AND
NEVILLE HOGAN
Many problems with body-powered artificial arms greatly limit restoration of function to the amputee user. F irst of all, since relevant groups within the remaining musculature are not utilized, control is awkward and re quires much concentration. The resulting movements are nonanthropomorp hic and noncosmetic. Secondly, each independent degree of freedom of the prosthesis requires an intact degree of freedom for control. Therefore, not only is the number of available control sites limited , but also. each control site used reduces function of the rest of the body. Thirdly, voluntary actuation of the device is restricted to one direction because the cable can only carry a tensile load. Typicall y, gravity or an elastic device gives the required restoring force. Development of cybernetic prostheses [48] has brought solutions to these control problems. The term cybernetic implies a natural synergy at the man-machine interface or , more specificall y. between the central nervous system (CNS) and the prosthesis. Motor commands sent by the C N S to the upper extremity provide the appropriate chan nel for commands to the artificial l imb. The electro myogram (EMG), or more correctly myoelectric activity (MEA), sensed with electrodes over the muscle [7), pro vides an excellent means of detecting the se motor com mands. Remaining musculature whose function is rele vant to control of the upper extremity is the most app rop riate source of M E A. Control of the prosthetic joints requires less learning, and does not interfere with control of intact body segments. Furthermore, exploiting antagonist activity e nables bidirectional powered move ment. Exploiting the synergistic action of multiple muscle groups enables multiple-axis prosthesis control [35], [36), [46], [59], [65].
Although other signals such as neuroelectric activity
This
[15] may be used to control a cybernetic prosthesis, only
work was perfonned in the Eric P. and Evelyn E. Newman Laboratory for
myoelectric control is considered here, Consequently, in the remainder of thi s report the terms cybernetic pros thesis and myoelectric prosthesis are used interchange ably. Two cybernetic upper-extremity prostheses represen tative of the state of the art are the Boston E lbow [3], [37] and the Utah Arm [34]. Each device has an electrically powered elbow. Attachment of an externally powered ter minal device is optional. Motion of the elbow joint is con-
Biomechanics and Human Rehabilitation at M.LT. and was supported
in
part by the National Science Foundation under Grants PFR-7917348 and ECS-8307461 and by the Fairchild Foundation.
C. J. Abul-Haj is with the Survivability Office, the Aerospace Corpo
ration, EI Segundo. CA
90245.
N. Hogan is with the Department of Mechanical Engineering, Massa chusetts In"itute of Technology. Cambridge, MA 02139. IEEE Log Number 9038608. l In t h i s context "external" refers to "outside the human bodv." De
d
vices that are actuated by the amputee a re called "body powere ." This jargon is widely accepted in the field of prosthetics.
0018-9294/90/1100-1025$01.00 © 1990 IEEE
1026
IEEE
TRANSACTIONS ON BIOMEDICAL ENGINEERING. VOL.
17. NO.
1!' NOVEMBER 1990
that is, the a lpha motoneurons-are still intact with the residual muscles, many of the afferent pathways from sen sory organs such as the joint receptors, muscle spindles, and golgi tendon organs are not intact. Therefore, al though the amputee is able to command the prosthesis via MEA, his ability to monitor the response is very l imited. Typically, cutaneous stimulation was used for artificial proprioception [6], [49], [58], [62]. However, such stim ulation generate s afferent activity quite unlike that due to the natural o rgans, and the success of these methods was l imited. Recent neurophys iological studies directed at under standing control of movement w ith the intact limb have shed insight into the problem of improving prosthesis control. First of all, control of state-of-the-art elbow prostheses is quite unlike control of the natural limb. W ith [59], [65]. Despite the many benefits of myoelectric protheses, the state-of-the-art elbow prostheses the amputee's motor these devices have been poorly accepted by amputees. commands, which are sensed as MEA, are used to p ro Some investigations during the early 1970's demonstrated portionally control the angular velocityZ of the joint [31, that o f the amputees who operated a myoelectric p ros [34], [37J. In contrast, with the intact limb, motor pro thesis most preferred the conventional body-powered de grams generated by the CNS do not command the a ngular vice [42], [49]. Moreover, many amputees preferred to velocity of the joints [60]. Another important finding is wear no prosthesis at all [43]. Even today relatively few that many movements can be performed in the complete myoelectric prostheses have been fitted to above-elbow absence of sensory feedback [8], [12], [29], [63], L64] amputees. even in the presence of external mechanical disturbances The work presented here is part of an ongoing project [II], LI2], [141. [40], [55], [64]. Thus, the amputee's directed at improving cybernetic prostheses. The under need for direct feedback from an artificial limb may be lying hypothesis is that improving the prosthesis control more due to poor controller architecture than to loss of ler architecture can enhance the amputee's functional ca proprioception. Consistent with this hypothesis is the observation that pability thereby inspiring him to use the prosthesis. Only control of a single-axis artificial e lbow has been studied; performance of the cable-operated p rosthesis is greatly however, the results may be generalized to control of a l imited despite the wealth of sensory feedback due to ex broader class of devices [4]. T wo elbow-prosthesis con tended physiological proprioception [59]. As is the case trollers have been examined: the conventional control with the myoelectric pro sthesis, control of the cable-op scheme implemented in the Boston E lbow [3], L37] and a erated prosthesis is quite unlike control of the natural proposed control scheme [5] that mimics some important l imb. Hogan postulated that the amputee's dependence on di characteristics of the intact neuromuscular system. Part I of this paper includes: 1) a detailed description of the two rect feedback with a myoelectric prosthesis may be de prosthesi s controllers and 2) the description of a technique creased given correct interpretation of muscle activity [25J, [27], [29], [30]. The object would be to have the for functionally assessing an amputee's perfo rmance with each controller. Part II reports on the experimental appl i p rosthesis controller determine from available M E A what cation of the technique, which demonstrated superior per the operator's motor intent is and then deduce what the natural limb would have done in response. Subsequently, formance of the proposed controller. the appropriate command signal would be sent to the pros II. IMPROVING CYBERNETIC PROSTHESES thesis actuator such that the pro sthesis responded as the One of the more l ikely factors leading to the poor suc natural limb. Two aspects of natural limb response are of interest cess of myoelectrically controlled prostheses is the am putee's strong dependence on direct feedback, particu here: 1) the response to voluntary motor commands ( i.e., "internal inputs"); and 2) the response to environmental larly via vision [39], [49], [51], L57], [661. If a prosthesis required visual feedback during operation, then it would loads (i.e., "external inputs"). Other investigators have draw the amputee's attention a way from the task at hand developed prosthesis controllers that attempt to restore natural motion of a n artificial arm in response to the am diminishing his functional capability. To reduce the need for visual feedback, some investi putee's commands [35], L361, [46], [59], [61], [65]. gators began examining alternat ive information transfer However. no prior attempt has been made to restore the pathways [6], L46], [49], [58], [62]. The approach taken 'Strictly speaking. the angular velocity is a vector quantity. which has was to provide an artificial substitute for some of the af magnitude and direction. Of interest in the conte>t of a prosthesis control· ferent neural communication e liminated or impaired by ler is not only the rale of movement (i.e . the magnitude or speed) but also the ablation of the limb. Although the efferent pathwaysthe direction (i.e flexion-extension). trolled by the amputee with a single antagonist muscle pair such as the biceps and triceps. Sensing MEA from these previously dysfunctional muscles allows control of the artificial joint with relevant musculature and does not interfere with other body movements. Using a dilferent control site for the terminal device gives fully indepen dent control of the elbow. A more elaborate scheme de veloped by Jacobsen et at. [35], [36] uses muscle signals from the shoulder girdle to control three prosthetic de grees of freedom: elbow flexio n , humeral rotation, and wrist rotation. The underlying idea is that the shoulder serves as a foundation for the upper extremity, and a ny movement of the upper extremity generates reactive loads that must be supported by the shoulder muscles. Other multiple-axis control schemes have been examined [46],
.
. .
1027
ABUL-HAJ AND HOGAt': CONTROL SYSTEMS FOR ELBOW PROSTH�S�S-l
natural response of the arm to externally app lied mo ments. Since many activities of daily living involve ma nipulation of objects, both constrained amI unconstrained, the response of the limb to applied loads is quite impor tant. In the present study the approach taken to develop an improved cybernetic prosthcsis has been to: I) design a prosthesis controller that, to a crude degree, mimics the response of the natural limb to voluntary inputs and en vironmental loading [5] and 2) implement the controller in a prosthesis. From here on, the proposed cybernetic control scheme is called "natural control." The approach taken to evaluate the natural controller has been to compare its performance with the perfor mance of the conventional velocity scheme implemented in the Boston Elbow [371. Contrasting natural contro l with conventional control , the following questions are ad dressed: I) can the amputee successfully operate the "natural prosthesis"? 2) Does natural control offer the amputee improved functional capability? III. PROSTHESIS-CONTROLLER EYlULATION Both the velocity and the natural controllers were im plemented and tested with a system that can emulate a broad range of prosthesis characteristics [4], [5]. This system consists of a high performance artificial elbow that is worn by an amputee subject and interfaced to a digital computer. Desired characteristics are programmed via the computer thereby eliminating the need to construct new hardware for each control scheme tested. The emulator allows easy programming of controller architectures that give the isolated prosthesis equivalent dynamics of the form
lot)
Bo( wyir - w) + Ko(fJyir
e)
(I) where e and ware, respectively, the elbow angle and an gular velocity; 10, Bo and Ko are, respectively, the net in ertia, damping, and stiffness about the elbow joint: eyir and Wvir are , respectively, the commanded angle and an gular velocity: and Text is the resultant moment on the foreaml due to the environment. The stiffness and damp ing both are variable; however, the inertia is fixed at about 0.05 N . m . 52. Both eyir and Wvir are internal inputs to the prosthesis that is , they are functions of the amputee's myoelectric commands. In contrast, Text is an exlernal input. As is ap parent from (1), the controller forces the states e and W toward the correspond ing command inputs. However, due to other loads present, the states may not equal the inputs; for example, the actual position differs from the com manded position because of viscous, inertial, and external loads. Furthermore, values of an input may be specified outside the achievable range of the corresponding slate. As a consequence of the disparity between a state and the commanded state, the latter quantity is called a virtual state [23], [26] , [28]. Note that in general, =
dem 1= dt
-
+
Text
That is, unl ike the true position and velocity, the virtual position and virtual velocity may be i ndependent.
To maintain well controlled experimental conditions, the designer must make a controller appear the same to all amputee subjects each time it is used. Unfortunately, the range of a myoelectric signal varies significantly not only from one subject to the next but also for a given subject each time the electrodes are applied. Conse quently, before being used to command the prosthesis, the MEA must be normalized. Raw MEA from the sensing electrodes is first processed with a circuit that rectifies and low-pass filters as described in Part II of this paper. The resulting signal is unipolar. Next, the level of the pro cessed MEA duri ng maximum voluntary contraction (MVC) is estimated for each muscle as discussed in Part II of th is paper. Thereafter, the processed M E A is divided by the estimated level at MVC producing a s ignal that varies roughly from zero to one. Denoting the processed MEA for the ith muscle as M, and the level of the pro cessed MEA at MVC as Mi yields the following for the normalized MEA ex ,
=
Mi/Mi
where i is b for biceps or agonist and t for triceps or an tagonist. The MEA level at MVC is estimated for the sub ject's muscles each time he dons the electrodes. The seemingly simple dynamic form represented by (1) allows emulation of a wide variety of controller types i n cluding velocity control and natural control. Implemen tation of these two controllers with the emulator is briefly presented here. A.
Emulation of the Boston-Elbow Controller
The Boston-Elbow controller uses the difference of two processed myoelectric s ignals to proportionally command the angular velocity about the joint. Typically, an antag onist pair such as the biceps and triceps is used for the control sites. A positive difference of the muscle signals corresponds to flexion and a negative difference to exten sion. Maintaining zero velocity while the muscles are re laxed is achieved with a "reverse-locking" clutch, which is a pass ive mechanical device in the drive train. The clutch allows the motor to actuate the elbow but locks the elbow if external loads exceed the motor torque. Th is controller is well modelled as a velocity servo (see Fig. I) with very high gain. The forearm is treated as an inertia driven by: 1) the actuator torque, which is propor tional to the velocity error: 2) the clutch torque Tclu1ch; and 3) externally applied loads Text· The corresponding equa tion of motion is lotj
=
BO(Wvir
-
w)
+
Tdul,h
+
Text
(2)
where Wvir is the i nput command to the servo. The Boston Elbow is primarily i ntended for use in a vertical plane. Under this condition, gravity assists the elbow during extension, and as a consequence, the de signers of the device selected a lower velocity-loop gain for extension-about one half the value for flexion.
1028
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING.
Text
+
VOL .17. NO. I L NOVEMBER 1990
Gclutch
actuator torque I
: IcY'
Fig. L Velocity servo used to model the Boston-Elbow controller.
�
Therefore. the damping factor for the emulator Bo is halved during extension. Computation of "',ir requires the difference of muscle activities �a
=
ab - at.
Before computing the velocity command, some simple nonlinear processing of �a is performed. First of all, a dead zone nonlinearity is applied to �a to prevent the low level of activity due to "muscle tone" from causing ex traneous motion of the prosthesis. The half width of the dead zone is denoted �amin' Also, a saturation nonline arity is applied to �a so that the amputee need not flex maximally for maximum output of the prosthesis. The saturation level is denoted �amax' Given that �amin and �amax are both positive constants, the net nonlinear char acteristic is mathematically represented by the following if I �a I
o �a
-
sign
if sign
(a)
::; �amm;
�a . �amin
�amin
�ama,;
and is graphically represented by Fig. 2(a). Proportional control of velocity is now achieved by setting
�amax "'max
where "'max is a positive constant representing the maxi mum speed. Simulation of the reverse-locking clutch is accom plished by switching on the position term shown in (1) and setting the stiffness high: locked; unlocked; where the position command Ovir is maintained equal to the value of the elbow angle 0 at the time the clutch is locked. Locking of the clutch occurs when the external load on the forearm begins to stall the motor. Therefore, (2) gives the following the criterion for locking: Bo"'m _
+
Text
-+ O.
Unlocking occurs when the subject commands an elbow torque that overcomes the external torque. Implementa-
Bo
=
7.0 N· m' s/rad (flexion)
= 0.5
Bo
=
3.5 N' m' s/rad (extension)
= 2.0 rad/s
Ku
=
150 N· m/rad
B. Emulation of the Natural Controller Previous studies have shown that the "spring-like" property of muscle plays an important role in controlling posture [9J, [ 11], [12], [18], [26], [53], [56] and move ment [8]-[10], [26], [28], [44], [55]. The agonist and an tagonist muscles spanning a joint determine the net joint stiffness [2 1], [22], [54], [56], which in tum determines the equilibrium position of the joint [12], [14], [ 1 8], [40], [55], [56]. Modulation of the muscle activities changes the net joint stiffness and, hence, may be the means by which the CNS controls posture. Control of movement may then be achieved by continuously varying the muscle activities [9], [ 1 0] thereby specifying a trajectory for the limb. Furthermore, since the net stiffness of the neuro muscular system determines the response of the limb to perturbations from the environment [22], [52], [55], the CNS may adaptively tune the net stiffness when com manding posture and movement in the presence of exter nal mechanical disturbances.
ABUL HAl AND HOGAN: CO]\;TROL SYSTE\
IH.!' TRANSACTI ONS ON BIOMEDICAL ENGINEERING. VOL. 37. 1\0. II. I\'OVEMBER 1990
Functional Assessment of Control Systems for Cybernetic Elbow Prostheses-Part
I:
Description of the Technique CARY
Abstracl-This
paper (Part
I
of
II)
J.
ABUL-HAJ
presents a novel control scheme
with which an amputee commands an elbow prosthesis using myoelec tric activity. By mimicking some important characteristics of the intact neuromuscular system. the proposed controller attempts to make the prosthesis respond as the natural elbow to both voluntary commands from the amputee and applied moments from the environment. Also presented is the description of a novel experiment for functionally as sessing elbow prothesis controllers. The experimental design calls for an amputee to perform a constrained motion task while operating a prosthesis capable of implementing a wide variety of controllers. Due to the natnre of the constraint, the task emphasizes the prosthesis re sponse to both inputs: voluntary commands and external moments. Application of the experiment to assessment of the proposed control scheme and the control scheme used in a state-of-the-art prosthesis is presented in Part
II.
l. iNTRODUCTION H E state of the practice in artificial limbs reflects a long-term deficiency of handicapped research. Today, the most common upper-extremity p rosthesis in use is the conventional cable-operated arm [41], [67], which was designed by aeronautical engineers after the Second World War. Since then the design of the device has changed very little. This prosthesis still utilizes power from the body I rather than external power from an actuator, and hence. does not represent the many recent advancements in elec tronics and actuator technology. Power and control of the cable-operated arm are derived from shoulder movement. In the case of the above-elbow prosthesis, rounding of the shoulders gives both flexion of the elbow and prehension of the terminal device (artificial hand) with a s ingle cable harnessed across the shoulders. Normally, the shoulder movement flexes the artifi c ial e lbow. To operate the ter minal device, the amputee first must lock the elbow via a mechanism that is engaged by shrugging the shoulder.
T
Manuscript received February 13, 1989; revised J anua ry 22,
1990.
AND
NEVILLE HOGAN
Many problems with body-powered artificial arms greatly limit restoration of function to the amputee user. F irst of all, since relevant groups within the remaining musculature are not utilized, control is awkward and re quires much concentration. The resulting movements are nonanthropomorp hic and noncosmetic. Secondly, each independent degree of freedom of the prosthesis requires an intact degree of freedom for control. Therefore, not only is the number of available control sites limited , but also. each control site used reduces function of the rest of the body. Thirdly, voluntary actuation of the device is restricted to one direction because the cable can only carry a tensile load. Typicall y, gravity or an elastic device gives the required restoring force. Development of cybernetic prostheses [48] has brought solutions to these control problems. The term cybernetic implies a natural synergy at the man-machine interface or , more specificall y. between the central nervous system (CNS) and the prosthesis. Motor commands sent by the C N S to the upper extremity provide the appropriate chan nel for commands to the artificial l imb. The electro myogram (EMG), or more correctly myoelectric activity (MEA), sensed with electrodes over the muscle [7), pro vides an excellent means of detecting the se motor com mands. Remaining musculature whose function is rele vant to control of the upper extremity is the most app rop riate source of M E A. Control of the prosthetic joints requires less learning, and does not interfere with control of intact body segments. Furthermore, exploiting antagonist activity e nables bidirectional powered move ment. Exploiting the synergistic action of multiple muscle groups enables multiple-axis prosthesis control [35], [36), [46], [59], [65].
Although other signals such as neuroelectric activity
This
[15] may be used to control a cybernetic prosthesis, only
work was perfonned in the Eric P. and Evelyn E. Newman Laboratory for
myoelectric control is considered here, Consequently, in the remainder of thi s report the terms cybernetic pros thesis and myoelectric prosthesis are used interchange ably. Two cybernetic upper-extremity prostheses represen tative of the state of the art are the Boston E lbow [3], [37] and the Utah Arm [34]. Each device has an electrically powered elbow. Attachment of an externally powered ter minal device is optional. Motion of the elbow joint is con-
Biomechanics and Human Rehabilitation at M.LT. and was supported
in
part by the National Science Foundation under Grants PFR-7917348 and ECS-8307461 and by the Fairchild Foundation.
C. J. Abul-Haj is with the Survivability Office, the Aerospace Corpo
ration, EI Segundo. CA
90245.
N. Hogan is with the Department of Mechanical Engineering, Massa chusetts In"itute of Technology. Cambridge, MA 02139. IEEE Log Number 9038608. l In t h i s context "external" refers to "outside the human bodv." De
d
vices that are actuated by the amputee a re called "body powere ." This jargon is widely accepted in the field of prosthetics.
0018-9294/90/1100-1025$01.00 © 1990 IEEE
1026
IEEE
TRANSACTIONS ON BIOMEDICAL ENGINEERING. VOL.
17. NO.
1!' NOVEMBER 1990
that is, the a lpha motoneurons-are still intact with the residual muscles, many of the afferent pathways from sen sory organs such as the joint receptors, muscle spindles, and golgi tendon organs are not intact. Therefore, al though the amputee is able to command the prosthesis via MEA, his ability to monitor the response is very l imited. Typically, cutaneous stimulation was used for artificial proprioception [6], [49], [58], [62]. However, such stim ulation generate s afferent activity quite unlike that due to the natural o rgans, and the success of these methods was l imited. Recent neurophys iological studies directed at under standing control of movement w ith the intact limb have shed insight into the problem of improving prosthesis control. First of all, control of state-of-the-art elbow prostheses is quite unlike control of the natural limb. W ith [59], [65]. Despite the many benefits of myoelectric protheses, the state-of-the-art elbow prostheses the amputee's motor these devices have been poorly accepted by amputees. commands, which are sensed as MEA, are used to p ro Some investigations during the early 1970's demonstrated portionally control the angular velocityZ of the joint [31, that o f the amputees who operated a myoelectric p ros [34], [37J. In contrast, with the intact limb, motor pro thesis most preferred the conventional body-powered de grams generated by the CNS do not command the a ngular vice [42], [49]. Moreover, many amputees preferred to velocity of the joints [60]. Another important finding is wear no prosthesis at all [43]. Even today relatively few that many movements can be performed in the complete myoelectric prostheses have been fitted to above-elbow absence of sensory feedback [8], [12], [29], [63], L64] amputees. even in the presence of external mechanical disturbances The work presented here is part of an ongoing project [II], LI2], [141. [40], [55], [64]. Thus, the amputee's directed at improving cybernetic prostheses. The under need for direct feedback from an artificial limb may be lying hypothesis is that improving the prosthesis control more due to poor controller architecture than to loss of ler architecture can enhance the amputee's functional ca proprioception. Consistent with this hypothesis is the observation that pability thereby inspiring him to use the prosthesis. Only control of a single-axis artificial e lbow has been studied; performance of the cable-operated p rosthesis is greatly however, the results may be generalized to control of a l imited despite the wealth of sensory feedback due to ex broader class of devices [4]. T wo elbow-prosthesis con tended physiological proprioception [59]. As is the case trollers have been examined: the conventional control with the myoelectric pro sthesis, control of the cable-op scheme implemented in the Boston E lbow [3], L37] and a erated prosthesis is quite unlike control of the natural proposed control scheme [5] that mimics some important l imb. Hogan postulated that the amputee's dependence on di characteristics of the intact neuromuscular system. Part I of this paper includes: 1) a detailed description of the two rect feedback with a myoelectric prosthesis may be de prosthesi s controllers and 2) the description of a technique creased given correct interpretation of muscle activity [25J, [27], [29], [30]. The object would be to have the for functionally assessing an amputee's perfo rmance with each controller. Part II reports on the experimental appl i p rosthesis controller determine from available M E A what cation of the technique, which demonstrated superior per the operator's motor intent is and then deduce what the natural limb would have done in response. Subsequently, formance of the proposed controller. the appropriate command signal would be sent to the pros II. IMPROVING CYBERNETIC PROSTHESES thesis actuator such that the pro sthesis responded as the One of the more l ikely factors leading to the poor suc natural limb. Two aspects of natural limb response are of interest cess of myoelectrically controlled prostheses is the am putee's strong dependence on direct feedback, particu here: 1) the response to voluntary motor commands ( i.e., "internal inputs"); and 2) the response to environmental larly via vision [39], [49], [51], L57], [661. If a prosthesis required visual feedback during operation, then it would loads (i.e., "external inputs"). Other investigators have draw the amputee's attention a way from the task at hand developed prosthesis controllers that attempt to restore natural motion of a n artificial arm in response to the am diminishing his functional capability. To reduce the need for visual feedback, some investi putee's commands [35], L361, [46], [59], [61], [65]. gators began examining alternat ive information transfer However. no prior attempt has been made to restore the pathways [6], L46], [49], [58], [62]. The approach taken 'Strictly speaking. the angular velocity is a vector quantity. which has was to provide an artificial substitute for some of the af magnitude and direction. Of interest in the conte>t of a prosthesis control· ferent neural communication e liminated or impaired by ler is not only the rale of movement (i.e . the magnitude or speed) but also the ablation of the limb. Although the efferent pathwaysthe direction (i.e flexion-extension). trolled by the amputee with a single antagonist muscle pair such as the biceps and triceps. Sensing MEA from these previously dysfunctional muscles allows control of the artificial joint with relevant musculature and does not interfere with other body movements. Using a dilferent control site for the terminal device gives fully indepen dent control of the elbow. A more elaborate scheme de veloped by Jacobsen et at. [35], [36] uses muscle signals from the shoulder girdle to control three prosthetic de grees of freedom: elbow flexio n , humeral rotation, and wrist rotation. The underlying idea is that the shoulder serves as a foundation for the upper extremity, and a ny movement of the upper extremity generates reactive loads that must be supported by the shoulder muscles. Other multiple-axis control schemes have been examined [46],
.
. .
1027
ABUL-HAJ AND HOGAt': CONTROL SYSTEMS FOR ELBOW PROSTH�S�S-l
natural response of the arm to externally app lied mo ments. Since many activities of daily living involve ma nipulation of objects, both constrained amI unconstrained, the response of the limb to applied loads is quite impor tant. In the present study the approach taken to develop an improved cybernetic prosthcsis has been to: I) design a prosthesis controller that, to a crude degree, mimics the response of the natural limb to voluntary inputs and en vironmental loading [5] and 2) implement the controller in a prosthesis. From here on, the proposed cybernetic control scheme is called "natural control." The approach taken to evaluate the natural controller has been to compare its performance with the perfor mance of the conventional velocity scheme implemented in the Boston Elbow [371. Contrasting natural contro l with conventional control , the following questions are ad dressed: I) can the amputee successfully operate the "natural prosthesis"? 2) Does natural control offer the amputee improved functional capability? III. PROSTHESIS-CONTROLLER EYlULATION Both the velocity and the natural controllers were im plemented and tested with a system that can emulate a broad range of prosthesis characteristics [4], [5]. This system consists of a high performance artificial elbow that is worn by an amputee subject and interfaced to a digital computer. Desired characteristics are programmed via the computer thereby eliminating the need to construct new hardware for each control scheme tested. The emulator allows easy programming of controller architectures that give the isolated prosthesis equivalent dynamics of the form
lot)
Bo( wyir - w) + Ko(fJyir
e)
(I) where e and ware, respectively, the elbow angle and an gular velocity; 10, Bo and Ko are, respectively, the net in ertia, damping, and stiffness about the elbow joint: eyir and Wvir are , respectively, the commanded angle and an gular velocity: and Text is the resultant moment on the foreaml due to the environment. The stiffness and damp ing both are variable; however, the inertia is fixed at about 0.05 N . m . 52. Both eyir and Wvir are internal inputs to the prosthesis that is , they are functions of the amputee's myoelectric commands. In contrast, Text is an exlernal input. As is ap parent from (1), the controller forces the states e and W toward the correspond ing command inputs. However, due to other loads present, the states may not equal the inputs; for example, the actual position differs from the com manded position because of viscous, inertial, and external loads. Furthermore, values of an input may be specified outside the achievable range of the corresponding slate. As a consequence of the disparity between a state and the commanded state, the latter quantity is called a virtual state [23], [26] , [28]. Note that in general, =
dem 1= dt
-
+
Text
That is, unl ike the true position and velocity, the virtual position and virtual velocity may be i ndependent.
To maintain well controlled experimental conditions, the designer must make a controller appear the same to all amputee subjects each time it is used. Unfortunately, the range of a myoelectric signal varies significantly not only from one subject to the next but also for a given subject each time the electrodes are applied. Conse quently, before being used to command the prosthesis, the MEA must be normalized. Raw MEA from the sensing electrodes is first processed with a circuit that rectifies and low-pass filters as described in Part II of this paper. The resulting signal is unipolar. Next, the level of the pro cessed MEA duri ng maximum voluntary contraction (MVC) is estimated for each muscle as discussed in Part II of th is paper. Thereafter, the processed M E A is divided by the estimated level at MVC producing a s ignal that varies roughly from zero to one. Denoting the processed MEA for the ith muscle as M, and the level of the pro cessed MEA at MVC as Mi yields the following for the normalized MEA ex ,
=
Mi/Mi
where i is b for biceps or agonist and t for triceps or an tagonist. The MEA level at MVC is estimated for the sub ject's muscles each time he dons the electrodes. The seemingly simple dynamic form represented by (1) allows emulation of a wide variety of controller types i n cluding velocity control and natural control. Implemen tation of these two controllers with the emulator is briefly presented here. A.
Emulation of the Boston-Elbow Controller
The Boston-Elbow controller uses the difference of two processed myoelectric s ignals to proportionally command the angular velocity about the joint. Typically, an antag onist pair such as the biceps and triceps is used for the control sites. A positive difference of the muscle signals corresponds to flexion and a negative difference to exten sion. Maintaining zero velocity while the muscles are re laxed is achieved with a "reverse-locking" clutch, which is a pass ive mechanical device in the drive train. The clutch allows the motor to actuate the elbow but locks the elbow if external loads exceed the motor torque. Th is controller is well modelled as a velocity servo (see Fig. I) with very high gain. The forearm is treated as an inertia driven by: 1) the actuator torque, which is propor tional to the velocity error: 2) the clutch torque Tclu1ch; and 3) externally applied loads Text· The corresponding equa tion of motion is lotj
=
BO(Wvir
-
w)
+
Tdul,h
+
Text
(2)
where Wvir is the i nput command to the servo. The Boston Elbow is primarily i ntended for use in a vertical plane. Under this condition, gravity assists the elbow during extension, and as a consequence, the de signers of the device selected a lower velocity-loop gain for extension-about one half the value for flexion.
1028
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING.
Text
+
VOL .17. NO. I L NOVEMBER 1990
Gclutch
actuator torque I
: IcY'
Fig. L Velocity servo used to model the Boston-Elbow controller.
�
Therefore. the damping factor for the emulator Bo is halved during extension. Computation of "',ir requires the difference of muscle activities �a
=
ab - at.
Before computing the velocity command, some simple nonlinear processing of �a is performed. First of all, a dead zone nonlinearity is applied to �a to prevent the low level of activity due to "muscle tone" from causing ex traneous motion of the prosthesis. The half width of the dead zone is denoted �amin' Also, a saturation nonline arity is applied to �a so that the amputee need not flex maximally for maximum output of the prosthesis. The saturation level is denoted �amax' Given that �amin and �amax are both positive constants, the net nonlinear char acteristic is mathematically represented by the following if I �a I
o �a
-
sign
if sign
(a)
::; �amm;
�a . �amin
�amin
�ama,;
and is graphically represented by Fig. 2(a). Proportional control of velocity is now achieved by setting
�amax "'max
where "'max is a positive constant representing the maxi mum speed. Simulation of the reverse-locking clutch is accom plished by switching on the position term shown in (1) and setting the stiffness high: locked; unlocked; where the position command Ovir is maintained equal to the value of the elbow angle 0 at the time the clutch is locked. Locking of the clutch occurs when the external load on the forearm begins to stall the motor. Therefore, (2) gives the following the criterion for locking: Bo"'m _
+
Text
-+ O.
Unlocking occurs when the subject commands an elbow torque that overcomes the external torque. Implementa-
Bo
=
7.0 N· m' s/rad (flexion)
= 0.5
Bo
=
3.5 N' m' s/rad (extension)
= 2.0 rad/s
Ku
=
150 N· m/rad
B. Emulation of the Natural Controller Previous studies have shown that the "spring-like" property of muscle plays an important role in controlling posture [9J, [ 11], [12], [18], [26], [53], [56] and move ment [8]-[10], [26], [28], [44], [55]. The agonist and an tagonist muscles spanning a joint determine the net joint stiffness [2 1], [22], [54], [56], which in tum determines the equilibrium position of the joint [12], [14], [ 1 8], [40], [55], [56]. Modulation of the muscle activities changes the net joint stiffness and, hence, may be the means by which the CNS controls posture. Control of movement may then be achieved by continuously varying the muscle activities [9], [ 1 0] thereby specifying a trajectory for the limb. Furthermore, since the net stiffness of the neuro muscular system determines the response of the limb to perturbations from the environment [22], [52], [55], the CNS may adaptively tune the net stiffness when com manding posture and movement in the presence of exter nal mechanical disturbances.
ABUL HAl AND HOGAN: CO]\;TROL SYSTE\
