Experimental BrainResearch

Exp Brain Res (1987) 67:241-252

9 Springer-Verlag 1987

Trajectory control in targeted force impulses II. Pulse height control J. Gordon and C. Ghez Center for Neurobiology and Behavior, New York State Psychiatric Institute, College of Physicians and Surgeons, Columbia University, 722 W. 168th Street, New York, NY 10032, USA

Summary. The present report examines the control strategy adopted by subjects to modulate the amplitude of transient force responses aimed to a target. Previous studies (Freund and Budingen 1978; Ghez and Vicario 1978) suggest that subjects modulate the rate of rise of force while maintaining force rise time at a near-constant value, independent of peak force. Such studies, however, have examined only the most rapid responses where force rise time could have been at a physiological limit. We now examine whether this control policy is dependent on an instruction to produce the fastest possible trajectories, or whether it is freely selected by subjects to maximize accuracy when rise time is unconstrained. We compared responses made by six subjects, under two task conditions: 1. Fast, "make the force impulse as brief as possible"; and 2. Accurate, "be as accurate as possible without regard to rise time". Subjects were trained to produce monotonic flexion force impulses at the elbow to match the amplitudes of visually presented target shifts. Targets of three different were presented in randomized order. Responses made under the Accurate condition were less variable at each target amplitude than those under the Fast condition. Under both conditions, the initial peaks of the first and second time derivatives of force, early measures of trajectory dynamics, were strongly predictive of the peak force achieved and were correlated with the required force (target amplitude). Therefore, response trajectories must have been largely preprogrammed, and, further, the degree to which the initial peak d2F/dt 2 predicts the peak force achieved represents a measure of the contribution of a preplanned motor program to trajectory formation. Subjects showed two systematic differences in trajectories between conditions. First, in all subjects force rise time was greater in the Offprint requests to: C. Ghez (address see above)

Accurate condition than in the Fast condition. Second, while in the Fast condition there was a modest dependence of force rise time on peak force, in the Accurate condition this dependence disappeared. Thus, when subjects were attempting to be as accurate as possible, they more consistently regulated force rise time around a constant value. This pulse height control policy allows responses of different amplitudes to be produced by proportional scaling of a stereotyped waveform. We conclude that a pulse height control policy with regulation of force rise time is a strategy adopted by subjects to simplify accurate control of response amplitude.

Key words: Human subjects - Isometric - Trajectory control - Motor programs - Accuracy - Agonistantagonist EMG pattern

Introduction On the basis of experiments performed almost a century ago, Woodworth (1899) postulated two phases in the control of targeted limb movements. The first phase was an "initial adjustment" which governed most of the limb's trajectory to the target, while the second phase of "current control" brought the limb to the target through a series of corrections. Woodworth believed that the initial adjustment was the expression of a plan that specified "...the innervation of different muscles one after the other... [and]...also a command to stop after a certain distance" (p. 55). He considered the corrections occurring during the phase of current control to be reactions to visual or kinesthetic stimuli arising from the movement itself. A similar view of the control of targeted movements is still widely held today (see Brooks 1979 for review). Simple limb movements to

242 a target are generally thought to result from the concatenation of two mechanisms. The limb is propelled toward the target by an initial impulse which is controlled by a prior plan, or motor program. Then, if there is enough time to assess the results of the original impulse, one or more corrections are made to acquire the target. In the most rapid movements, sometimes referred to as "ballistic" (Stetson 1905), it is often considered that the entire trajectory depends on a motor program and that no corrections are possible (Kornhuber 1971; Hallett et al. 1975; Desmedt and Godaux 1978). The distinction between preprogrammed responses and corrective adjustments is dearly crucial to an understanding of the neural control of goaldirected movement. However, despite the venerable history of these concepts of movement control, many of the issues raised by Woodworth's original findings remain unresolved. In particular, questions concerning the mechanisms responsible for motor programming and for error correction and their relative contributions to trajectory formation have been difficult to resolve. Because operational criteria or objective measures of these hypothetical mechanisms have been lacking, it has proven difficult to determine whether a movement is guided solely by a program or whether it has been modified by corrective adjustments. The assumption that trajectory parameters of rapid movements are planned in advance usually rests on the notion that such movements are completed in less time than the latency to respond to a stimulus. This idea is extended to suggest that rapid movements cannot contain corrective adjustments and must of necessity be performed open-loop (Stetson and McDill 1923; Taylor and Birmingham 1948; Desmedt and Godaux 1978). However, a reliance on movement duration alone for defining the functional role of preplanned motor programs encounters difficulties (see also Brooks 1979). First, experiments have demonstrated corrective adjustments at a latency too brief after movement onset for them to be based on feedback from the results of the movement itself (Higgins and Angel 1970; Georgopoulos et al. 1981; Vicario and Ghez 1984). Second, this definition ignores the potential role of internal feedback pathways (Evarts 1971; Oscarsson 1973) which, in principle, could modify trajectories early in their course (e.g. Higgins and Angel 1970). Finally, attempts at defining the contribution of motor programs from comparisons with reaction times fail completely to address the issue for responses which have a longer time course. In order t o study the contribution of motor programs to movement, we have selected a model system in which motor programs are likely to play a

major role and in which their contributions to trajectory formation can be clearly defined. The responses we studied, rapid force impulses produced at the elbow to match visual targets, are sufficiently brief (rise times < 120 ms) that they must be largely dependent upon preplanning. In addition, we tried to exclude overt corrections by instructing subjects not to correct their responses once they were initiated. Previous studies suggested that both human subjects (Freund and Biidingen 1978) and cats (Ghez and Vicario 1978) adopt a characteristic "control policy" or "general rule'! to adjust the peak force of such targeted isometric responses. In both species, subjects vary the amplitude of the force responses by modulating the rate of force rise (measured as the peak of the first time derivative of force, dF/dt) while the force rise time is kept relatively invariant. This control policy has been referred to as "pulse height control" (Ghez 1979; Ghez et al. 1983). To the degree that force rise time is invariant, trajectories are stereotypic and response amplitude can be matched to different targets by the proportional scaling of a common waveform. Therefore, regulation of rise time around a preset value simplifies accurate control of response amplitude by reducing the number of variables that must be controlled (Bernstein 1967; Ghez et al. 1983; Enoka 1983). Until now, it has not been clear whether and to what extent rise time invariance is an important and general feature of targeted responses. Most of the supportive findings have been derived from the study of very rapid motor responses. In Freund and Btidingen's study, for instance, subjects were explicitly instructed to produce forces rising as rapidly as possible. Subjects may have interpreted this instruction to require a constant rise time, or neural or mechanical constraints alone may have limited the rise time. In order to examine the generality of pulse height control and its functional significance, we have now compared responses under two conditions. In one condition, the instructions required subjects to make the force rise as rapidly as possible to its peak (Fast condition). This instruction constrained rise time. In the other condition, subjects were instructed to produce the most accurate force impulses possible (Accurate condition). This instruction left rise time unconstrained. We reasoned that, if rise time invariance is present when subjects are instructed to maximize accuracy without regard to rise time, it would establish that such invariance does in fact represent a preferred Strategy and would also support the suggestion that rise time regulation is used to enhance accuracy. Establishing the presence of invariant relations between trajectory variables allows us to infer the

243

general rules by which motor programs operate (Bouisset and Lestienne 1974). These rules must be major elements in the process by which the central nervous system transforms target information into response parameters (Ghez 1979). An understanding of these rules may also provide a basis for distinguishing the influences of motor programming from those of corrective adjustments. In this study, we have examined force impulses made in the Fast and Accurate conditions to determine whether such rules exist. We have also explored how such rules affect accuracy arid whether they provide evidence concerning the operation of motor programs. Finally, we have examined the EMG activity of the biceps and triceps to determine how the rules are implemented by contractions of the agonist and antagonist muscles. In the following paper (Gordon and Ghez 1987), we will extend our analysis to determine whether corrective adjustments are also present in these responses. These results formed part of a doctoral thesis (Gordon 1985) and some of the data have appeared as an abstract (Gordon and Ghez 1985). Methods Six neurologically normal adult volunteers (S1-$6) between the ages of 25 and 45 participated in the experiment, including 5 males and 1 female. Five of the subjects were right-handed; one subject reported that he was ambidextrous, preferring to use his right hand for precision tasks (e.g. writing) and his left hand for power tasks (e.g. lifting and throwing). All subjects were given at least three sessions of training prior to testing, in order to allow them to become accustomed to the apparatus, task, and experimental procedures. All subjects had participated in one or more of the experiments reported in the previous paper in this series (Ghez and Gordon 1987). Apparatus, training, and data collection procedures have previously been described in detail (Ghez and Gordon 1987). Subjects were tested sitting with right arm placed in a rigidly fixed manipulandum. A force transducer was coupled to the manipulandum in such a way as to measure isometric torque produced at the elbow joint. Subjects viewed a video monitor showing two horizontal lines, representing the target (or required force) and the actual force output. At the beginning of a trial, subjects were to align their force output with the target. After a variable period (1-2 s), the target shifted upwards and the subjects then had to produce an impulse of flexion force at the elbow so that the oscilloscope trace just reached the target. They were then to allow a passive return of force to baseline. Subjects were instructed to produce a single smooth impulse of force and were further told not to attempt to amend their responses once initiated. Responses with multiple peaks in the first derivative of force were considered to show evidence of overt corrections (Brooks 1974) and were excluded from data analysis. These were always less than 5% of a subject's responses. The overall task of producing targeted force impulses was modified by one of two sets of instructions. In one, termed here the Fast condition, subjects were urged to minimize the force rise time, that is, to "make the force rise as fast as possible to its peak".

A 150 Z V

o n/ O L,_

i00

v CE L.d

./ /.'

5(3

/// o

/..../

FAST

+ ACCURATE

. .y"/' .... I

I

5o

I

. . . . . .

16o

I

B Z

.3-

(D (I ~X C~ >

.2

b. 0

t-Z W

F

,i

r h. Ld O (.3

'

'

5'0

!

!

16o

TARGET AMPLITUDE

1go

'

(N)

Fig. 1. A Mean peak force as a function of target amplitude for subject $2. Circles indicate means of responses made in Fast condition, crosses indicate means in Accurate condition. Dotted line indicates perfect scaling of average peak force to required force. B Coefficient of variation of peak force (ratio of standard deviation of peak force to mean peak force) as a function of target amplitude for subject $2. Circles indicate Fast condition, crosses indicate Accurate condition

In the other, termed the Accurate condition, subjects were to minimize deviations of the peak force from the target, that is, to "make responses as accurate as possible"; in this condition, there were no constraints on force rise time. In both conditions, no constraints were imposed on reaction time; subject were specifically instructed to respond "when ready". As in the previous study, subjects were able to review their responses on a storage oscilloscope placed just below the primary monitor. This allowed subjects to assess the accuracy of their responses, to determine when trajectories were not smooth, and to optimally conform their responses to the instructions. Testing sessions consisted of 200-300 trials of 1.8 s duration. Inter-trial interval was varied randomly between 4 and 6 s. Trials

244

0.24

0.20 Z 0

0.16

h. 0

0.12

Z

hJ hl 0 0

0.08

0.04

0.00

$1

$2

S3

[771

FAST

$4

$5

$6

MEAN

TASK CONDITIONS ~ ACCURATE

Fig. 2. Mean coefficient of variation of peak force for each subject (S1-S6) by condition. The height of each bar represents a subject's mean coefficient of variation (across target amplitudes) for either Fast (lightly hatched) or Accurate (densely hatched) conditions. The two bars on the extreme right show the mean values for each condition across all six subjects. The difference between Fast and Accurate conditions is statistically significant (Mann-Whitney U test, p < 0.05)

under a specific task condition were organized as blocks of trials with 1-2 min rest periods allowed between blocks. Subjects were presented with three different targets whose amplitudes were in a ratio of 1 : 2 : 3; the largest target amplitude corresponded to a force that ranged from 37% to 53% of subjects' maximum voluntary flexion force. Maximum target amplitude for each subject was chosen during practice sessions to be the largest peak force a subject could consistently achieve without observable assistive movements of the trunk and without causing fatigue. In testing sessions, the experimental trials were presented in blocks of 20-30 trials. Within a block, the three different target amplitudes were presented in randomized order. The two task conditions, Fast and Accurate, were alternated within a session. Two to three blocks were performed under one condition, then an equal number of blocks under the other condition. This order was then repeated once. Three of the subjects began with the Fast condition; the other three began with the Accurate condition. At the end of a session, maximal voluntary flexion and extension forces produced by the elbow were recorded, along with the associated EMG. The force target was displayed on the monitor as a horizontal line prior to the shift from baseline to the required peak force. After the step change in target level, the line on the monitor appeared as a solid bar with a width that was constant for all target amplitudes and equal to about 10% of the largest target amplitude. This was achieved by summing the target voltage with a 10 kHz square wave signal of appropriate amplitude. A running tally of hits and misses was kept, and the subject was informed of the score after each block of trials. The purpose of this procedure was to maintain a high level of motivation throughout the testing session by presenting the subject with a challenging task. The score of hits and misses was not used in subsequent data analysis.

Results

Effect of accuracy instruction on task performance All subjects were capable of producing force impulses whose amplitudes closely approximated the required target values under both Fast and Accurate conditions. This is illustrated for one subject ($2) in Fig. 1A, in which the mean peak force is plotted as a function of target force for each condition. The mean amplitudes of the force responses produced in Fast and Accurate conditions were very similar although, as shown in the example illustrated here, Fast responses were typically slightly overshot while Accurate responses were slightly undershot. There was no systematic trend for the mean errors to be larger in either the Fast or the Accurate condition. In contrast, task condition did influence the variability or dispersion of the peak forces of responses. In the Accurate condition, peak force variability at each target amplitude was significantly reduced compared to the Fast condition for all target sizes. In order to allow comparisons of response variability across target amplitudes and across subjects, the coefficient of variation of peak force at each target amplitude (i.e. the ratio of the standard

245

A

ACCURATE CONDITION

FAST CONDITION

TARGETS

i

FORCE dF/dt

NORMALIZED

.-

FORCE dF/dt

..

j&

-...... =

"

-

,,

o

F onset

~

....

I

I

-dod

-

z0o

40o~

-z'00

....

0

e00

" 400ms

F onset

Fig. 3. A Average force, dF/dt, and d2F/dt 2 trajectories for responses to 3 target amplitudes by one subject ($3) in Fast condition (left) and Accurate condition (right). Each average consists of 27-30 trials; trial data were aligned on force onset prior to averaging. Responses to large target shown as solid lines, responses to medium target as widely spaced dots, responses to small target as closely spaced dots. Calibration: 100 N, 2920 N/s, 116,800 N/s2. B Same average trajectories as in A are normalized by peak force

deviation of peak force to the mean peak force) was used as a measure of relative dispersion. Figure 1B plots coefficient of variation as a function of target amplitude in Fast and Accurate conditions for the same subject as in Fig. 1A. The relative dispersion of peak forces around the mean peak force for each target amplitude was less in the Accurate condition, though this effect was more pronounced for responses to the smaller target amplitudes. The slope of the relationship between coefficient of variation and target amplitude was flatter in the Accurate condition than the Fast condition, indicating that the relative variability of peak force was more consistent across target amplitudes in the Accurate condition. The same results were observed in all subjects: the principal effect of the instruction to be as accurate as possible was that subjects reduced the variability of

their responses. Figure 2 shows the mean coefficient of variation across target amplitudes for each subject in the Fast and Accurate conditions, as well as the mean across subjects. In all subjects, the mean coefficient of variation was significantly reduced in the Accurate condition (Mann-Whitney U test, p < 0.05). Similarly, the dependence of the coefficient of variation on target amplitude, noted earlier in the Fast condition, was either reduced or absent in the Accurate condition for all other subjects.

General features of isometric force .trajectories We next analyzed relations between trajectory parameters in the Fast and Accurate conditions to determine whether subjects used similar control policies in

246

A o

~q \ Z v v .i--,

o

o

oo

,o

,e

o

.i,

et:, o