6.1 Abstract
This chapter examines how the development of grip force prediction for a single object being m anipulated is dependent on previous m anipulative ex perience. The virtual object paradigm was used to create objects that were either linked together so they acted as a single object, or unlinked as if two separate objects are being m anipulated between the hand. The extent of grip force m odulation seen in one hand, when the other pulled on the ob ject was found to depend in a systematic way on the object's properties experienced over at least the previous three trials.
6.2 Introduction
In this chapter the effect of experiencing the objects as a single object or as two separate objects on previous trials was used to examine the current grip force prediction. In Chapter 5 it was shown that the occurrence of a linked trial prior to an unlinked trial significantly increased the m odulation that occurred during the trial. Here subjects experienced "Linked" trials, where the motion of the left hand caused equal and opposite forces on the object held in the right hand, and therefore the system behaved as though there were a real physical object held between the hands; a situation where
Effect o f previous experience 130
predictive grip force modulation w ould normally occur in the right hand. In ''U nlinked" trials, motion of the left hand was decoupled from the right hand object, and therefore there was no movement of the right hand object.
Events in the past that effect the future response m ade are ubiquitous feature of m any tasks, including serial-reaction time (Bertelson, 1965; Williams, 1966; Remington, 1971). The influence of previous trials has been show n to occur in the predictive m odulation of grip force. The scaling of grip force to the properties of an object has been described in a process of 'anticipa tory param eter control' (Johansson and Cole, 1992). W hen a subject was given an object of a high density to lift that was identical in appearance to a previously m anipulated object, the grip force on the first lift was scaled appropriate for the lighter load force (Gordon et al., 1993).
Analysis of the effect of previous experience in reaction time tasks is of ten confined to the examination of the effect of the imm ediately preceding stimulus, that is the first-order effect, and more specifically to the repetition effect (Bertelson, 1963; Bertelson, 1965; Remington, 1969; Peeke and Stone, 1972; Kirby, 1972). Such studies have examined the effects of previous ex perience in serial reaction time tasks. Bertelson (1963,1965) show ed that reaction time in a serial choice reaction time task, was strongly determ ined by the sequence of preceding stimuli. Differing effects of previous experi ence have been found that were task dependent. Tasks thought to rely on automatic, unconscious processes showed a repetition effect; where the ef fect of repeated experience is to reduce the response time to that stimuli. Conversely, in tasks that have a greater cognitive component, repetition of a stimulus has the effect of subjective expectancy, w here subjects tend to
expect more alternations than repetitions in a series (Cleeremans and Mc Clelland, 1991).
However, effects of previous experience are not isolated to the last stim uli presented and analysis of the pattern of higher-order effects is necessary to fully understand the mechanisms underlying sequential effects (Maljkovic and Nakayama, 1994; Maljkovic and Nakayama, 1996).
As m ost sequential effects research has been related to the first-order repetition effect, most explanations of the effect have been consistent with the idea of facilitatory memory traces or the strengthening of links between the processing nodes.
To examine the effect of previous m anipulative experience in grip force m odulation a random series of linked and unlinked trials was presented.
Three situations were compared, firstly where 30% of trials were linked, secondly w hen 50% of trials were linked and finally, w hen 70% of trials were linked. Linked trials are expected to result in predictive grip force modulation, and unlinked trials m ay suppress predictive m odulation. To assess the effect of previous experience on the grip force response, the m ag nitude and timing of grip force m odulation on both linked and unlinked trials can be examined as a function of the preceding history of linkage. The grip force response seen in the right hand during unlinked trials, where the object behaved as two separate objects reflects a purely predictive response, as in these trials no load force is generated on the fingertips of the right hand. Therefore, the grip force response on unlinked trials will be used to assess the influence of the previous manipulative experience on the current predictive response, separate from any reactive component.
Effect of previous experience 132
6.3 M ethods
Eight right-handed subjects (5 male, 3 female) aged 19-30 w ho were naive to the research aims gave informed consent and participated in the study. None of the subjects reported any sensory or m otor deficits. Each subject participated in all three conditions in a random ized order (30%, 50% and 70% Linked). Subjects perform ed 250 trials in each condition.
The task was a bimanual task, using the virtual object paradigm , as de scribed in Chapter 2, with two objects attached to separate torque motors and encoders under computer control. Before each trial, subjects positioned the objects in a starting position.
The start of a trial was signaled by a tone approximately every 3 s. On hearing this tone, subjects m ade a short upw ard pulse with their left hand. The position of the object held in the subject's left hand was displayed as a scrolling trace on a computer monitor. The required am plitude of 6 m m was displayed as a constant horizontal line on the scrolling trace. Either linked or unlinked trials were experienced, with the trials linked either 30%, 50% or 70% of the trials depending on the condition. The linked and unlinked trials occurred in a pseudorandom series. In linked and unlinked trials the m otion of the left hand acted against a simulated stiff spring of 1 N /m m attached to the left hand object's initial position. In a linked trial the objects behaved as if a single object was held between the hands and a load force equal and opposite to the left-hand load force, was applied to the right hand object. On unlinked trials, no load force was generated on the right hand object.
linked, based on cues from accidental small m ovem ents of the left hand, the force on the right hand was zero until the tone in all trials. To prevent fatigue, short rest periods were given every 40 trials in all conditions.
For each trial the position of both hands, and the grip force, and load force on the object in the right hand were recorded at 200 Hz. To quantify the m agnitude and timing of anticipatory grip force, the am plitude and timing of the peak grip force modulation was found for each trial. Grip force m od ulation was taken as the difference between the peak grip force (maximum grip force w ithin a 400 ms window on either m axim um left hand discursion) and the baseline grip (average value of the grip force in the first 100 ms of each trial). This measure of modulation of grip force, rather than actual grip force was determ ined as it is increased m odulation that is the characteristic feature of predictive grip responses (Johansson and Cole, 1992). The grip force m odulation lag was calculated as the difference betw een the time of the peak grip force and the time of peak left hand discursion (with negative values indicating grip force precedes discursion). To examine the effects of linkage and condition on these measures a MANOVA was perform ed on grip m odulation and lag as a function of condition (3 levels) and linkage (2 level).
To display the influence of the history of linkage for the predictive re sponse in unlinked trials, the grip force m odulation was displayed as a binary tree structure. In such a tree, the first layer is the grip force av eraged over all unlinked trials. The second layer is represented by two nodes, one for the m odulation on unlinked trials which followed a linked trial and the other for unlinked trials which followed an unlinked trial. The third layer divides each of these nodes into tw o representing the na
Effect of previous experience 134
ture of the trial two time steps ago, and so on. To quantify the influence of previous trials, the measures for the unlinked trials { X j , the measure
of interest on unlinked trial j ) were regressed against the history of link
age. This was achieved by regressing these measures against indicator vari ables, I j , over the last 5 trials, where I j = 1 if trial j was a linked trial
and I j = 0 is trial j was unlinked. The regression therefore fitted X j = clq -f- CLiIj—i 4- CL2l j—2 T (X z lj—z 4" cL^Ij—A 4" CL^Ij—^. Separate regressions were
perform ed for each of the eight subjects and three conditions and a t-test was used to examine the significance of the regression param eters for each condition (5 levels a i - a ^ ) .
This behaviour was simulated in a simple model of subjects' predictive behavior p . The m odel assumes that subjects alter their future response
according to their prediction error on the current response. The response change is assum ed to be proportional to the prediction error. Prediction ranges continuously from no prediction (p = 0) to full prediction of linkage (p = 1). If the trial experienced was linked, then the corresponding predic tion error e = 1 — p, (if the trial was unlinked, then e = p). The prediction on the following trial will be updated to reduce e, by an am ount propor tional to e: p —> p 4- p • e, where g is the proportionality gain. To reflect that
prediction m ay be slower (or faster) to decay than to build up, the gain af ter experiencing a linked trial gi may differ from that after experiencing an
unlinked trial g u -
To construct a linkage tree from this model, the average prediction for an unlinked trial is given by p = where Pl is the overall probabil
given that the previous trial was linked is derived a s p L = p - f (1 — p ) 9l-
Analogously, average prediction given that the previous trial was unlinked
i s p u = p { l - Q u )- Further tree linkages can be calculated recursively. To
fit the peak grip force data with this model, we took grip force as G m i n +
p { G m a x - G m i n ) , where G m i n and G m ax are the m inim um and m axim um pre
dictive grip force recorded during each condition. The two free param eters
Qu and q l were fit to the data using a least squares minimisation.
6.4 Results
Three typical runs of 20 trials from the three conditions for one subjects are show n in Figure 6.1. The yellow lines show the maxim um grip force m od ulation which is in general higher on linked trials (red) than on unlinked trials (blue). Flowever, the linked trials represent a combination of a pre dictive and a reactive response whereas the unlinked trials reflect purely predictive responses. Although the data show variability, the influence of the previous trials on a given unlinked trial can be seen. For example, com pare the trial immediately before a linked trial (red) w ith one after. It can be seen that the one after is usually higher, and the height depends on the num ber of preceding linked trials. The MANOVA show ed that the condi tion had no effect on grip force m odulation, baseline or lag. However, there was significant m ain effect of linkage on both lag and m odulation. For the linked trials the peak m odulation was 1.3 N greater (p<0.0001) and 30 ms delayed (p<0.00001) compared to the unlinked trial. This is consistent w ith our previous findings that the reactive component of grip force m odulation both amplifies and delays the grip force response.
Effect of previous experience 136 a) b) T r ia l 100 T r ia l T r ia l
Figure 6.1 — Maximum grip force modulation (yellow o) in unlinked trials (blue) and linked trials (red) of a typical subject on trials 100-120 for a) 30% linked con dition b) 50% linked condition and c) 70% linked condition. Note that some trials were off the scale.
Firstly, the repetition effect was examined. Figure 6.2 illustrates the de pendence of the grip force response profile on the previous history. This shows grip force profiles averaged over the subjects in unlinked trials in the 50% condition. These show a systematic relationship betw een the profiles and the previous linkage history. The thick black line shows the grip profile w hen all unlinked trials are considered (***U, where the stars represent all possibilities over the previous three trials). The other lines represent sub sets of this data depending on the history. Lines above the thick line are for conditions in which either the last (**LU), previous two (*LLU) or previous three trials (LLLU) were linked (representing successively smaller subsets of the data). The grip force m odulation is thereby systematically increased by the experience of linked trial. Similarly the profiles below the thick line are for trials in which either the last (**UU), previous two (*UUU) or previous three trials (UUUU) were unlinked showing that the m odulation decreases with experience of unlinked trials.
These differences in the magnitude of the grip force response show a dependence of the current grip force response on the preceding trials that were experienced. To examine the specific influence of previous m anipula tive experience on the m agnitude of the predictive grip force response, the m agnitude of grip force m odulation on each possible combination of pre ceding linkage was compared for unlinked trials over the previous three trials. Figure 6.3a.
In the unlinked trials of the 30% condition, previous experience of a linked or unlinked trial affects the m agnitude of the grip force response in the current trial (Figure 6.3). The MANOVA showed that the last trial, un-
Effect of previous experience 138 50% Linked Trials LLLU 8.5 LLU LU ^ 7.5 uu uuu UUUU 6.5 5.5 -300 -200 -100 100 200 300 Time (ms)
Figure 6.2 — Grip force response for the unlinked trials when 50% linked. Average of the subject's average grip force profiles, aligned to the maximum left hand po sition (vertical line). The grip profile is the average of each subject's average grip force response on linked trials where the previous three trials are unlinked; below the thick line; to averages of grip force responses when the previous three trials are linked; above the thick line .
linked or linked has the largest effect on the m agnitude of grip force m od ulation in the following unlinked trial (P<0.001). However, the influence of linkage history is not restricted to the last unlinked trial, w ith a signifi cant influence (P<0.01) of the penultimate trial, and a significant influence (P<0.05) on grip force m agnitude dependent on the linkage of the trial three prior to the current trial. In the 50% condition, the effect of previous linkage on the grip force response was significant over the last three trials. During the 70% condition, there was a significant effect of previous linkage on the grip response during unlinked trials over the last two trials. There was no effect of lag as function of linkage history.
Statistics were then perform ed on the regression analysis; t-tests over the regression param eters showed a significant effect of the history of linkage in all three conditions. In the 30% condition, there was a significant influ ence of the last trial (P<0.00001); penultimate trial (P<0.0001); third last trial (P<0.005) and fifth trial (P<0.05), in this condition (Figure 6.4). In the 50 % condition, the last (P<0.01) and third from last trial (P<0.05) significantly influenced the m agnitude of the grip force response. In the 70 % condition, the last trial was significant, (P<0.001). None of the regression param eters were significant for the lags.
The data is captured by a simple model in which the prediction builds up at a different than it decays (Figure 6.3 b; see Analysis for details). Using only two free param eters for the build-up and the decay, this m odel cap tures the essential features of the data. First, it reproduces the shape of the linkage tree (Figure 6.3 a). Secondly, the model matches the incremental decay of prediction after a series of unlinked trials.
Effect of previous experience 140
a)
i) 30% Linked Trials n) 50% Linked Trials Hi) 70% Linked Trials
2.5
8
1.5 3 3 < 2.5 / 2 ' 0 2 1.5 1.5 --- ,--- '--- ' 1 --- --- '--- . 1 0 1 2 3 0 1 2 3 0 1 2 3History of Linkage History of Linkage History of Linkage
b)
8
Gu GI 0.85 0.55 3 2 1 3 0 1 2 History of Linkage Gu Gi 0.60 0.51 0 2 3 History of Linkage Gu Gi 0.74 0.55 0 1 2 3 History of Linkage Figure 6.3 — a)Average of each subject's average maximum grip force modulation in unlinked trials plotted as a function of the history of linkage of those unlinked trials. Grip force modulation with trials with previous unlinked trials (diamonds) can be compared with those trials with previous linked trials (squares) for i) 30% linked trials ii) 50% linked trials and iii) 70 % linked trials, b) Simulation of the data in the same format as a). GI and Gu are the two free parameters fit to the data.a) 30% b) 50% 0 70% œ 0.4 CL 0.4 CL 0.4
iiiè
P 0.2 2 0.2 9? 0.2i i
a1 a2 aS a4 a5 History of Linkage a1 a2 aS a4 a5 History of Linkage a1 a2 aS a4 a5 History of LinkageFigure 6.4 — Average of the regression parameters a i - a ^ with standard error bars
for a)30% condition b)50% condition c) 70%condition
6.5 D iscussion
In this chapter, the history of experience affects grip force m odulation dur ing the bim anual m anipulation of a virtual object has been examined. The forces on each object were controlled so that the two objects acted as though a single object was being m anipulated between the hands, or unlinked, so that they acted as two independent objects. Linked and unlinked trials oc curred in random series of 30, 50 or 70% linkage dependent on the condi tion. The unlinked trials in the 30%, 50% and 70% conditions show ed an effect of previous experience of object linkage on the m agnitude of the grip force response. It was found that the linkage experienced three trials prior