Perception-Action Coupling in the Development of Visual

Journal of Experimental Psychology: Copyright 1997 by the American Psychological Association, Inc. Human Perception and Performance 0096-1523/97/$3.00 1997, Vol. 23, No. 6, 1631-1643

Perception-Action Coupling in the Development of Visual Control of Posture

Bennett I. Bertenthal, James L. Rose, and Dina L. Bai University of Virginia

In this investigation of developmental changes in the coordination of perceived optical flow and postural responses, 4 age groups of infants (5-, 7-, 9-, and 13-month-olds) were tested while seated on a force plate in a "moving room." During each trial the walls oscillated in an anteroposterior direction for 12 s, and the postural sway of the infant was measured. The results revealed that infants perceived the frequency and amplitude of the optical flow and scaled their postural responses to the visual information. This scaling was present even before infants could sit without support, but it showed considerable improvement during the period when infants learn to sit. Taken together, these results suggest that the visuomotor coordination necessary for controlling sitting is functional prior to the onset of independent sitting but becomes more finely tuned with experience.

An emerging theme in the developmental literature is that perception and action are dynamically coupled in the regulation of coordinated movements (Bertenthal, 1996; Thelen, 1990; von Hofsten, 1993). This coupling is necessary to ensure stable action patterns that are appropriate to the specific environmental demands of the situation. Con- sider the case of locomotion. It would be nearly impossible to move from one location to another if the motor response synergies necessary for this action were not modulated by perceptual information specifying the layout of surfaces and the spatial orientation of the organism.

One of the most useful paradigms for studying the early development of this coupling involves the visual control of posture in a "moving room" (see Figure 1). In this paradigm, the infant sits or stands on a stationary floor while the surrounding walls and ceiling move forward and backward. This movement produces visual information congru- ent with movement of the head in the opposite direction. If the optical flow is perceived as specifying self-motion (as opposed to object motion), 1 then a compensatory response that varies with age and motor development will be produced.

In the initial study with this paradigm, Lee and Aronson (1974) reported that infants who were just beginning to stand on their own would sway or stagger when the walls moved. This finding was subsequently replicated and extended to show that optical flow restricted to the peripheral portions of the visual field was sufficient to induce postural

Bennett I. Bertenthal, James L. Rose, and Dina L. Bai, Depart- ment of Psychology, University of Virginia.

This research was supported by National Institutes of Health (NIH) Grants HD16196 and HD23144 and by a predoctoral fellowship from NIH Grant HD07323. We thank Dawn Chandler, Sharon Lambert, Leslie Ferree, and Stephen Garrison for their assistance in testing infants and scoring data.

Correspondence concerning this article should be addressed to Bennett I. Bertenthal, Department of Psychology, Gilmer Hall, University of Virginia, Charlottesville, Virginia 22903. Electronic mail may be sent via Intemet to [email protected].

compensations (Bertenthal & Bai, 1989; Stoffregen, Schmuckler, & Gibson, 1987). Additional evidence for the coupling between vision and posture was reported by Butterworth and Hicks (1977) and Bertenthal and Bai (1989), who showed that infants who could sit independently would produce postural compensations with their trunks when tested in the moving room. Some of the most recent evidence (Jouen, 1990; Jouen, Lepecq, Gapenne, & Bertenthal, in press) suggests that even newborn infants show postural compensations with their heads when stimulated by an optical flow pattern.

This last finding is especially provocative because it suggests that the rate-limiting factor in the visual control of posture is not the perception of optical flow. Even neonates detect optical flow information as specifying self-motion. Nevertheless, the detection of this visual information does not guarantee that the infant will produce the appropriate postural responses in all situations. Systematic postural compensations while sitting or standing occur later than compensations with the head because the necessary muscle strength and coordination develop at different rates for each posture (Woollacott & Sveistrup, 1994). Moreover, perceptual sensitivity to visual information improves rapidly during the first few months after birth. In particular, thresholds for spatial and temporal frequencies improve, as do thresholds for detecting motion (Banks & Dannemiller, 1987; Bertenthal & Bradbury, 1992). As these thresholds improve, the minimal information for inducing a postural compensation will change.

Thus, it is the development of both sensory and motor skills that are necessary for improvements in postural

1 It is important to emphasize that the perception of self-motion from optical flow does not imply any awareness of the self. The perception of self-motion is based on infants' showing a directionally appropriate postural response to the optical flow information. It would be misleading and unnecessary to conclude any more about the perception of the self from this response. See Bertenthal and Rose (1995) for further discussion of this point.

1631

1632 BERTENTHAL, ROSE, AND BAI

Figure 1. Schematic drawing of moving room. Depicted inside the room is a child falling backward as a function of the room's moving toward the child. This compensatory response would occur if the child perceived the optical flow produced by the room movement as specifying a forward sway rather than a movement of the walls.

control, but it would be misleading to suggest that these developmental changes are sufficient. In order to ensure a stable posture, it is necessary for the infant not only to detect any departure from a state of equilibrium but also to produce the appropriate muscle activity to restore balance. The success of this compensatory response depends specifically on the processing of the sensory information into motor com- mands that produce the appropriate muscle activity (Hirschfeld & Forssberg, 1994; Pick, 1989). For this reason, it is the coordination, or coupling, of the sensory and motor processes that is central to the development of postural control.

Previous research on the development of visual control of posture has been restricted primarily to fairly general questions of perception-action coupling, such as whether infants sway in the appropriate direction to compensate for their perceived displacement. More specific questions concerning how infants use visual information to modulate their postural responses and when they begin to scale their compensatory responses to the perceived displacement have rarely been addressed. This limitation in the literature is at least in part attributable to the fact that most previous studies used only observational measures and induced postural compensations only with very simple forms of optical flow. The purpose of the current study was to begin to investigate the development of the visual control of posture more quantitatively. To do so, we focused on the development of infants' ability to maintain postural equilibrium while sitting in a moving room.

Effects o f Frequency and Ampli tude of Optical Flow on Postural Sway

In the current research, we manipulated two dimensions that are useful for defining the visual stimulus in optical flow (Andersen & Dyre, 1989; van Asten, Gielen, & Denier van

der Gon, 1988). The first of these dimensions concerns the temporal frequency (f) or the period (I/f) of the visual stimulus. In order to operationalize this dimension in the moving room paradigm, one must oscillate the room at a specified frequency, such as once every 2 s. The second dimension concerns the amplitude of the visual stimulus. In the moving room, this dimension corresponds to the distance that the walls move before reversing direction. It is useful to note that the covariation between these two dimensions specifies the velocity of the visual stimulus.

These dimensions of frequency and amplitude can be used for parameterizing the postural response as well. More specifically, the infant can be described as swaying at a given frequency and at a given amplitude. By measuring the entrainment (or the covariation) between the driving stimulus and the postural sway, it is possible to quantitatively assess the coupling between the visual information and the appropriate motor responses. As postural responses become better modulated by the visual information, the entrainment between the visual information and the sway response will increase.

A recent study by van Asten et al. (1988) presents an excellent example of how the relation between these stimulus dimensions and postural sway is investigated in adults. In that study, adult observers were shown a rotating windmill pattern while they were instructed to stand upright. Their postural responses were time sampled and transformed to the frequency domain so that the effect of the driving frequency on postural sway could be evaluated. The results revealed a clear correlation between the frequency of the visual stimulus and the frequency of the postural sway for driving frequencies of 0.3 Hz or below. On the other hand, sway frequency was essentially independent of the amplitude of the visual stimulation.

PERCEPTION-ACTION COUPLING 1633

One caveat about the preceding findings should be noted. Postural sway was induced by a two-dimensional rotating stimulus. It is not clear that the same sensorimotor relations would be observed if the visual stimulus was produced by a moving room. Optical flow can be decomposed locally into four basic components: translation, rotation, expansion and contraction, and a pure shear (Koenderink, 1986). The stimulus used by van Asten et al. (1988) contained only one of the components available in the flow field (i.e., rotation), whereas the stimulation produced by a moving room contains all four of the components in the flow field. It is thus possible that the findings reported by van Asten et al. (1988) would not necessarily generalize to a situation involving the moving room.

For this reason, the results from a preliminary study (Delorme, Frigon, & Lagace, 1989) that tested the sway frequency of young children's responses in a moving room are especially noteworthy. In this study, 7-, 9-, and 13-month- old infants, as well as 2- and 4-year-old toddlers, were tested while standing inside of an oscillating moving room. The younger infants were not able to stand without support; all children were therefore requested to hold onto a T-shaped support bar that was attached to a force platform. In this way, postural changes while standing were indexed indirectly by the changes in the force applied to the T-bar.

For our purposes the most relevant finding is that 50% of the 7-month-olds, 100% of the 9-month-olds, and 60% of the 13-month-olds showed a peak sway frequency that was identical to the driving frequency of the moving room (0.54 Hz) on at least one of two trials. Regrettably, infants were tested with only one sway frequency, and thus it is possible that the driving frequency chosen for the moving room corresponded to the natural sway frequency used by children whenever they are induced to sway. Testing with at least two different frequencies is necessary to establish that this potential confound was not responsible for the results.

Visual Control of Sitting

In the current investigation, we focused on the development of the visual control of sitting. One reason for studying sitting rather than standing is that we were interested in collecting as many trials of data as possible from each infant, and we anticipated that we could better ensure compliance during sitting. (Our experience from previous studies is that standing infants will not remain in the moving room for any extended period of time.) A second reason for studying sitting was that we were interested in testing infants who straddled an age at which independent control of a posture was first observed. By choosing to test infants while they were sitting, we could provide passive support so that sitting independently was not a criterion for inclusion in the study.

On average, infants begin sitting independent of support at 6.6 months of age (range = 5 to 9 months; Bayley, 1969). The emergence of this new posture represents a major motoric milestone, because the center of mass is now suspended above the surface of support and must remain balanced within its stability limits. Relatively rapid postural adjustments in hip, trunk, and neck muscles are required to

resist or compensate for sudden perturbations. Recent evidence suggests that the muscle synergies necessary for maintaining equilibrium during sitting are adultlike but are more variable by the onset of independent sitting (Hadders- Algra, Brogren, & Forrsberg, 1996a; Hirschfeld & Forss- berg, 1994; Woollacott & Sveistrup, 1994).

Our primary goal in this study was to investigate when and how infants begin to use visual information to modulate their postural responses while sitting. Accordingly, we chose to test infants at 5, 7, and 9 months of age, ages that span the range within which sitting develops. In addition, we thought it would be useful to investigate whether additional changes in the coupling between vision and posture would be observed following more experience with sitting. For this reason, we tested an additional group of infants at 13 months of age.

All infants were tested in the moving room while seated on a force plate. We manipulated the timing of the visual stimulus by oscillating the walls of the moving room at two frequencies (0.3 Hz or 0.6 Hz). The amplitude of the visual stimulus was manipulated by moving the walls a distance of 9 cm or 18 cm. Unlike the tasks in most previous moving room studies, this task demanded that infants not only show directionally appropriate compensations but also respond quickly and accurately to changes in the perceived direction of postural sway. By measuring how the magnitude and frequency of the postural response covary with the visual stimulus, we sought to gain new insights into how the coupling between visual information and posture develops as a function of age and sitting experience.

Method

Participants

The final sample included 40 infants divided evenly between four ages: 5 months (20-22 weeks), 7 months (28-30 weeks), 9 months (36-38 weeks), and 13 months (56-58 weeks). Twenty-six additional infants were excluded from the study (21 because of fussiness, 4 because of equipment failures, and 1 who showed a significant delay in motor development). Infants were recruited from birth records in the local newspaper, and the majority were from middle-class families. Additional information about socioeco- nomic status and ethnic background was not obtained in this study.

Apparatus

Infants were tested inside a moving room similar to the one used by Bertenthal and B ai (1989). The room is shown schematically in Figure 1. It consist~ of a boxlike enclosure with three walls, a ceiling, and a floor (1.2 × 1.2 x 2.1 m). The walls are mounted on wheels that roll on tracks located on the floor. This arrangement allows the walls and ceiling to be moved back and forth by hand along the room's long axis while the floor remains stationary. The inside of the room is lined with a vertically striped green and white material (stripes = 3.3 cm); the floor is padded and covered with white muslin. Two fluorescent lights located at the top of the two side walls illuminate the room.


A rectangular window (21.5 × 18 cm) is positioned behind the middle of the front wall at a height of 83 cm. Located directly behind the window is a wooden box that contains a toy dog. The box is illuminated and the dog is electronically activated during each trial in order to control infants' direction of gaze. A low-light video camera is mounted on the back side of the front wall for videotaping infants' responses through a small circular opening located below the window.

Procedure and Design

Infants were seated in the moving room and faced the front wall. After allowing them a brief period of time to become familiar with the room, we began the first trial. Each trial followed the same sequence of events. First the toy was activated to direct the infant's attention to the front of the room, and then the walls began to oscillate forward and backward for approximately 12 s. Following each trial, infants were given a brief opportunity to rest before the next wall movement began, Infants were tested for a total of 12 trials, which lasted approximately 15 to 20 min. After testing in the moving room was completed, information about the infant's motor development (e.g., onset of sitting, crawling, and standing) was obtained from the parent.

On each trial, one of six stimulus conditions was presented; each condition was presented twice. The first condition was a baseline in which the walls did not move. This condition (henceforth referred to as the no-movement condition) was always presented as the first and last trials of the series. The next four conditions involved crossing two stimulus frequencies (0.3 and 0.6 Hz) and two stimulus amplitudes (9 and 18 cm) to produce wall movements of 0.3 Hz that moved a distance of 9 and 18 cm and wall movements of 0.6 Hz that moved 9 and 18 cm. Mean velocities in these four conditions (collectively referred to as the constant frequency condition) ranged between 5.4 and 21.6 cm/s (see Table 1). A final condition involved moving the walls at a variable frequency and amplitude that changed in pseudorandom fashion from one cycle to the next. In this condition (henceforth referred to as the variable frequency condition), frequencies ranged between 0.1 and 0.9 Hz, and amplitudes ranged between 4 and 16 cm. These last five conditions produced a total of 10 trials that were presented in random order.

Data Collection and Reduction

Postural responses were measured while infants sat in a standard infant bicycle seat with a rigid back inclined at 70 ° from the horizontal plane. The seat was mounted on a metal plate attached to four pressure transducers that measured the vertical ground reaction forces in response to the forces produced by the infant sitting in the chair. The center of pressure of these vertical forces is a standard metric for measuring changes in posture over time (Winter, 1979). If all four forces are equal, the center of pressure is at the exact center of the force plate. A change in the distribution of

Table 1 Mean Velocity (in Centimeters/Second) as a Function of Frequency and Amplitude of Wall Movement

Amplitude (half-cycle)

Frequency (in hertz) 9 cm 18 cm

0.3 5.4 10.8 0.6 10.8 21.6

d I (F 1 + F 2 ) " (F 3 + F 4 ) ] x : --/-[ Fz

Y = d 2 (F 1 + F 4 ) ° (F 2 + F 3 ) ] 2 [ F z

Figure 2. Schematic representation of a force plate and vector components of ground reaction forces. F = force (in newtons); M = moment (in newton centimeters); d = distance (in centimeters); X and Y correspond to the center of pressure along the two orthogonal axes of the plate. The force transducers are labeled F~ through F4.

forces will shift the center of pressure along the two orthogonal axes of the plate. Figure 2 presents a schematic representation of a force plate and the formulas used for computing center of pressure along both anteroposterior and mediolateral directions. Note that the center of pressure is normalized relative to the total vertical force or weight of the infant.

In this study, center of pressure was time sampled every 20 ms (50 samples/s) and stored in a computer for subsequent analysis. A potentiometer attached to one wall of the moving room output a voltage that changed linearly with the position of the wall. This voltage was also sampled every 20 ms and was subsequently transformed to a measure of distance and direction so that postural responses by infants could be compared directly with the displacement of the walls.

Postural sway was operationalized as the change in the center of pressure along the anteroposterior axis from one sample to the next. Wall movement was operationalized as the change in the anteroposterior position of the room from one sample to the next. For each trial, we calculated the time series for walt movement (see Figure 3A) and postural sway (see Figure 3B) by extracting the first 512 samples (10.2 s) of the data collected. The frequency of wall movements and of postural sway was calculated by transforming the time series data to the frequency domain with a fast Fourier transform. Prior to this transformation, we subtracted the mean of the time series from each sample to eliminate any direct current bias, and the we passed the series through a Hamming window to reduce the contribution of spurious spectral values created by the


(A)

10

2 _ s

o ~ ~ ~ ~ 1'o TIME (sec)

(e)

°i g

- 3 i i i i i i

(c)

1'2

70-

uJ 60- U

50-

40- >

30- Z

Z0- n , ,

~ 1 0 -

0:6

(o)

l'!z 1:8 z:4 FREQUENCY (Hz)

30-

uJ 25- tJ Z < 20- m.

> 1 5 - I---

,,z, ~o. r e '

~ 5"

0 0 2 4 6 8 10 1'2 0~6 1 ~2 1 ~8 2~4 :~

TIME (sec) FREQUENCY (Hz)

Figure 3. An example of wall movement and postural sway plotted in the time and frequency domains. A and B depict two time series corresponding to the time-sampled wall movement and sway data, respectively. C and D depict the power spectral density functions for the wall movement and sway data, respectively.

presence of noninteger frequencies within the time series window (Jenkins & Watts, 1968; Ramirez, 1985). 2 We calculated a power spectrum from the amplitude portion of the Fourier transform to provide an estimate of the variance accounted for by each 0.1-Hz increment in frequency (limited by the Nyquist sampling theorem to 25.0 Hz in the current analysis); the resulting spectral density function was normalized so that the total variance was equal to 1.0 (Jenkins & Watts, 1968).

After calculating the spectral density function, we determined the magnitudes (i.e., percentages of variance) of postural sway at those frequencies corresponding to the stimulus frequencies. Figure 3 shows spectral density functions for the wall movement (Figure 3C) and postural sway (Figure 3D). As can be observed, the peak frequency of the wall movement is 0.6 Hz, and the peak frequency of the postural sway is also 0.6 Hz. Note that the peak frequency of the wall movement is expected to correspond to the specified stimulus frequency for that trial, but the peak frequency of the postural sway is free to vary. A value in the sway frequency spectrum is considered significant if its magnitude (height) exceeds the 95% confidence limits of the spectral estimates of white noise (Jenkins & Watts, 1968). (See Figure 4.)

Resu l t s

The results are divided into four sections. First, we present descriptive data summarizing infants ' sitting experience at each age. Second, we present analyses examining the covariation between the frequency of postural sway and the frequency of the wall movements. Third, we assess the

timing between postural sway and wall movements when the walls move at constant and variable frequencies. Last, we examine how the changing stability in infants ' sitting contributes to the variabil i ty of their postural sway. Unless stated otherwise, each dependent measure is based on the mean of the two replicates at each condition.

Onset of Sitting Without Support

The mean age for sitting without support was 25.6 weeks (SD = 2.9 weeks). Table 2 lists the number of infants sitting without support at each age and the mean number of weeks of sitting experience for each age group. As can be observed, only one 5-month-old infant sat without support, but the majority of 7-month-old infants were capable of sitting. By 9 months of age, all infants were sitting without support.

2 It was unnecessary to detrend the data because the linear component of the time series contributed only 1-2% to the variance. Moreover, we were not interested in the entire spectrum but only in the results from those frequency bins corresponding to the stimulus frequencies of the wall movements, that is, 0.3 and 0.6 Hz. A preliminary comparison between subsets of trials that were and were not detrended revealed that the results from the relevant frequency bins differed by less than 0.2% of the variance.


100" ILl

~ 80"

60"

o. 40"

f

J

• f

• ~ s S

2O" • • • I s s •

0 I

o 5

. - f s t* f ! "

• s s f f s,, ~*

I I

10 15 FREQUENCY (Hz)

I i

20 25

301 25

15

O" 0.6

95% Confidence Limit

! v I I

1.2 1.8 2.4 3 FREQUENCY(Hz)

Figure 4. Spectral density functions for postural sway and white noise. The top panel depicts the cumulative spectral density distribution for postural sway (solid line) and white noise (large dashed line). The small dashed lines depict the 95% confidence limits for the white noise function. This plot shows that almost all the spectral energy in postural sway is concentrated below 2.0 Hz. The bottom panel depicts a spectral density function showing that one peak centered at 0.6 Hz is significant because it exceeds the 95% confidence interval for a white noise function.

Frequency Entrainment Between Postural Sway and Wall Movements

We conducted three analyses to test whether sway frequency was systematically related to the wall movements.

Table 2 Mean Age of Sitting Onset and Weeks of Experience for Each Age Group

Sitting onset Sitting experience (in weeks) (in weeks)

Age group n a M SD M SD

5-month-olds 1 19.0 - - 2.0 - - 7-month-olds 9 23.8 2.4 4.9 3.1 9-month-olds 10 26.8 3.7 11.6 4.1 13-month-olds 10 26.1 2.7 26.6 2.9

aNumber of infants sitting without support at each age.

First, we tested whether sway frequencies at 0.3 Hz or 0.6 Hz were significantly greater than predicted by chance. Second, we tested whether the magnitudes of the sway frequencies matching the wall movement frequencies were different during the constant frequency trials and the no- movement trials. Finally, we tested whether sway frequency varied as a function of the frequency or amplitude of the moving walls.

The first analysis tested whether sway frequencies at 0.3 Hz or 0.6 Hz showed a magnitude that was significantly greater than chance (i.e., white noise). The results from this analysis are grouped as a function of age, wall frequency, and wall amplitude (see Table 3). At each age, tests were conducted on 80 trials divided evenly between four conditions. Even 5-month-old infants showed significant sway frequencies on some of the trials, but it was not until 7 months of age that infants showed a significant sway frequency on a substantial percentage (i.e., 44%) of trials. By 9 months of age, infants showed a significant sway frequency on 65% of the trials, and this percentage was approximately the same (i.e., 62.5%) at 13 months of age. Chi-square analyses revealed that the number of trials showing a significant sway frequency increased with age, ×2 (3, N = 40) = 44.01, p < .001, and with frequency, ×2 (1, N = 40) = 8.47, p < .005, but not with amplitude, ×2 (1, N = 40) = 2.45, ns.

The next analysis evaluated whether the driving frequency of the moving walls entrained infants to sway at that same frequency. In order to answer this question we compared the magnitude (percentage of variance) of sway at the driving frequency (constant frequency condition) with the magnitude of sway at that same frequency calculated during the baseline trials (no-movement condition). For example, the magnitude of the spectral density function calculated at 0.3 Hz for the 0.3-Hz wall movement condition was compared to the magnitude of the spectral density function at 0.3 Hz for the no-movement condition. This comparison between no-movement and constant frequency conditions was evaluated as a function of age. As can be seen in Figure 5, the magnitude of sway in the constant frequency condition showed a substantial increase as a function of age, whereas the magnitude of sway in the no-movement condi-

Table 3 Number of Trials Showing Significant Sway Frequencies as a Function of Age, Frequency, and Amplitude

Amplitude

Age group Frequency (in hertz) 9 cm 18 cm

5-month-olds 0.3 2 5 0.6 3 5

7-month-olds 0.3 6 11 0.6 7 11

9-month-olds 0.3 8 13 0.6 14 17

13-month-olds 0.3 9 15 0.6 14 12


tion showed a much more modest increase with age. An 1 analysis of variance revealed a significant effect for condi- Z0 tion, F(1, 36) = 136.54, p < .001, and a significant effect for tu age, F(3, 36) = 11.67, p < .001. Also, the interaction between these two variables was significant because the <_ 15

t~ difference between them increased with age, F(3, 36) = ,( 9.50, p < .001. >

F- In spite of the preceding age effect, the above results z

kl.I confirm that postural compensations were systematically coupled with wall movements, even at 5 months of age, F(1, tu r t

9) = 14.56,p < .005. It is especially noteworthy that this relation was significant at an age when only 1 infant was sitting without support. Apparently, postural adjustments are modulated by visual information even before sitting is sufficiently coordinated and functional to maintain postural equilibrium.

The next analysis assessed whether the magnitude (or percentage of variance) of sway at the driving frequency of the walls varied as a function of age (5, 7, 9, and 13 months), wall frequency (0.3 and 0.6 Hz), or wall amplitude (9 and 18 1 cm). As can be seen in Figure 6, the magnitude of the sway 20 frequency increased with both the amplitude and frequency of the wall movements. Moreover, this measure of entrain- z ment between posture and wall movements increased with ~< 15 age. An analysis of variance revealed that all three of these ,~

> effects were statistically reliable: age, F(3, 36) = 11.19, p <

I . . .

.001; wall frequency, F(1, 36) = 16.08, p < .001; wall 10

amplitude, F(1, 36) = 10.09, p < .01. None of the interactions between these variables was significant. Neu- ~- 5 man-Keuls post hoc comparisons (p < .05) confirmed that the percentage of entrainment shown by 7-month-old infants was significantly greater than that shown by 5-month-old 0 infants and, likewise, that the percentage of entrainment shown by 9-month-old infants was significantly greater than that shown by 7-month-old infants. The percentage of entrainment increaSed only slightly between 9 and 13 months of age, and this difference was not significant.

As a follow-up to this finding, we tested whether the increase in the percentage of entrainment was related to sitting experience. Correlation coefficients were computed

• .3 Hz [ ] .6~lz I ¢ J ~ , ¢ /

~ ' / / / / • i ~" # ¢ ' f / / , ~ / / / J

~// / , ///,4/

z 9 3 AGE (months)

¢ , / ' / / - ~

g i l i A 11 f1~ 111 f~ v111~

I f f f ~

5 7 9 13 AGE (months)

Figure 6. Mean magnitude of spectral density function (+SE) calculated at the wall movement frequency as a function of stimulus frequency, stimulus amplitude, and age. Top panel summarizes the results of the two wall movement frequencies; bottom panel summarizes the results of the two wall movement amplitudes.

20-

815- Z <c

X IO. F- Z , , i

,,, S- a-

O-

[ ] Constant Freq. ] • No Movement I


Figure 5. Mean magnitude of spectral density function (+SE) calculated at the wall movement frequency (Freq.) as a function of condition and age.

between percentage of entrainment and sitting experience at 7 and 9 months of age. The results revealed a relatively low correlation at 7 months, r(18) = .13, but the correlation at 9 months was significantly higher, r(18) = .72. Recall that 9-month-old infants were reported to have had twice as much sitting experience (11.5 weeks) as 7-month-old infants (5.4 weeks). This difference suggests that 5 weeks of sitting experience contributes only modestly to the visual control of sitting but that twice as much experience makes a significant difference. Of course, the direction of this relation is ambiguous, and thus it is just as likely that improvements in the visual control of posture contribute to a more coordinated and stable sitting posture. In either case, the relation between sitting experience and visual control of posture is noteworthy.

Timing Between Postural Sway and Wall Movements

One question unanswered by the preceding analyses is whether the improved covariation between the visual infor-


mation and the postural response is related to improvements in the temporal coupling of this perception-action system. In order to answer this question, we assessed coupling between postural sway and wall movements by correlating these two variables at all possible temporal lags between 0.0 and 1.67 s (i.e., the period of the faster wall movement condition) to produce a cross-correlation function for each trial (see Figure 7). 3 From this function, we calculated the temporal lag at the peak correlation as an index of the temporal coupling (i.e., latency to respond) of the perception-action system. For example, the peak correlation in the cross- correlation function depicted in Figure 7 is .29, and this correlation occurs at a temporal lag of 0.17 s.

The first set of analyses calculated the peak correlations and temporal lags for the constant frequency condition. Both measures showed improvements as a function of age. As can be seen in Figure 8, the correlations showed a gradual improvement between 5 and 9 months of age, whereas the temporal lags showed a more abrupt improvement between 5 and 7 months of age. (Note that lags decrease as performance improves.) An analysis of variance revealed that the peak correlations between wall movement and sway increased not only as a function of age, F(3, 36) = 6.34, p < .001, but also as a function of wall frequency, F(1, 36) = 16.40, p < .001, and wall amplitude, F(1, 36) = 37.26, p < .001. As such, these results are completely consistent with those based on the spectral density functions.

The temporal lags between postural sway and wall movements were also analyzed as a function of age, wall frequency, and wall amplitude. These analyses revealed a significant effect for age, F(3, 36) = 7.29, p < .001, and showed a strong trend as a function of wall amplitude, F(1, 36) = 3.83, p < .06. It is interesting that the latency to respond to the visual stimulus did not vary as a function of wall frequency, F(1, 36) = 2.44, ns.

Thus far, our analyses have been restricted to conditions involving a constant frequency of oscillation. In most natural situations, the visual information available to the

• 5"1 , PEAK .4/ i CORRELATION

~ ,,

"'21 -.4-

-1 . -112 -J8 -)4 :4 18 112 126

TIME (sec)

Figure 7. An example of a cross-correlation function showing the change in correlations between postural sway and 0.3-Hz wall movements from a temporal lag of - 1.67 to + 1.67.

z ~ kU g

LU o o

z 0

uJ

o 0

[ ] Constant Freq. ]


.8 "G

.6 (.9 5 LU .4 ~E B

.2 ̧

S 7 9 13 AGE (months)

Figure 8. Covariation between postural sway and wall movements for constant and variable frequency (Freq.) conditions. Top panel depicts the mean peak correlation (+SE) for wall movements and postural sway as a function of condition and age. Bottom panel depicts the mean temporal lag (+SE) between wall movements and postural sway as a function of condition and age.

observer does not conform to a periodic stimulus oscillating at a single frequency. At the very least, the visual information is more variable as the trunk and head change their period and amplitude of oscillation in response to local perturbations. In this next set of analyses, we tested the coupling between sway and wall movements in a condition that more closely approximates natural situations, that is, the variable frequency condition.

3 Typically, a cross-correlation function extends from negative to positive lags. In our analysis, temporal lags between 0.0 and 1.67 s corresponded to room movements leading postural sway; temporal lags between 0.0 and -1.67 s corresponded to postural sway leading room movements. This study was concerned with the control of postural sway by the visual stimulus and thus was restricted to temporal lags between 0.0 and 1.67 s.

PERCEt~ION-ACTION COUPLING 1639

As can be seen in Figure 8, the peak correlations and temporal lags both showed improvements with age. Similar to the preceding findings, the peak correlations showed a gradual improvement between 5 and 9 months of age, whereas the temporal lags showed a more abrupt improvement. Analyses of variance confirmed that the peak correlations showed a linear trend with age, F(1, 36) = 5.82, p < .03, and that the temporal lags between postural sway and wall movements decreased as a function of age, F(3, 36) = 5.75, p < .005.

The final set of analyses compared the peak correlations and temporal lags for the constant and variable frequency conditions. Peak correlations did not differ as a function of condition, F(1, 36) = 0.22, ns, nor was there an interaction with age, F(3, 36) = 0.89, ns. By contrast, temporal lags differed as a function of condition, F(1, 36) = 8.45, p < .01, and this difference interacted with age, F(3, 36) = 4.24, p < .02. As can be observed in Figure 8, these differences are attributable primarily to 7-month-old infants' showing shorter temporal lags in the constant frequency condition than in the variable frequency condition. By 9 months of age, the temporal lag is very similar in the two conditions and averages no more than 0.25 s. This tight coupling between perception and action is quite impressive given the inertia of the limbs and the temporal lags imposed by neural transmis- sion of signals (Bertenthal, 1996; von Hofsten, 1993).

Postural Stability and Variability o f Performance

One interpretation for the improvement in visuomotor coupling with age is that younger infants are less posturally stable and thus their responses are more variable. As a consequence of this variability, posture will show less of a systematic relation to the movement of the walls. In order to evaluate this interpretation, we assessed the variability or the total magnitude of infants' postural response by calculating the root-mean-square (RMS) of their changing center of pressure on each trial. The first analysis was restricted to wall movement trials in the constant frequency condition and compared RMS as a function of age, stimulus frequency, and stimulus amplitude. This measure of postural sway revealed a significant effect of age, F(3, 36) = 15.62, p < .001, but no effect of stimulus frequency, F(1, 36) = 1.44, ns, or stimulus amplitude, F(1, 36) = 2.99, ns. As can be observed in Figure 9, the shape of the developmental function was surprisingly nonlinear. Trend analyses revealed a significant linear effect with age, F(4, 33) = 2.78, p < .05, a significant quadratic effect with age, F(4, 33) = 3.88, p < .01, and also a significant cubic effect with age, F(4, 33) = 5.12, p < .005. The reason for the significant quadratic and cubic effects is that 5-month-old infants showed more sway than 7-month-old infants, but 9-month-old infants also showed more sway than 7-month-old infants. This U-shaped function is quite familiar in the motor learning literature. During development of new motor skills, it is not uncom- mon for individuals to temporarily reduce degrees of freedom by increasing muscle stiffness in order to reduce the chances of producing highly variable movements that dis- rupt postural stability or the outcome of the action (Bern-

Z.O~

1.5"

E u (~ 1.0" ~E

.5-

O-

Constant Freq. i Variable Freq. No Movement


Figure 9. Mean root-mean-square (RMS) of postural sway (+SE) as a function of condition and age. Freq. = frequency.

stein, 1967; Bertenthal & Clifton, in press; Newell & Corcos, 1993).

The total magnitude of postural sway showed a similar developmental pattern for the variable frequency and no- movement conditions (see Figure 9). An analysis of variance comparing RMS as a function of age and condition (constant frequency, variable frequency, and no movement) revealed a significant effect of age, F(3, 36) = 14.82, p < .001. Trend analyses revealed a significant linear effect with age, F(3, 34) = 4.53, p < .01, a significant quadratic effect with age, F(3, 34) = 3.73, p < .02, and a significant cubic effect with age, F(3, 34) = 8.22; p < .001. There was also a significant effect of condition, F(2, 72) = 12.51, p < .001, but the interaction between condition and age was not significant, F(6, 72) = 1.45, ns. Neuman-Keuls post hoc comparisons (p < .05) revealed that RMS was significantly higher in the no-movement condition than in the other two conditions. One interpretation for this finding is that infants move less when they detect that they are unstable and losing balance.

In sum, these results show that the RMS of postural sway does not decrease linearly with age. Instead, the variability of postural sway declines around the age when infants first begin sitting without support, but then it begins to increase as infants gain more experience with sitting. Apparently, the improved coupling between visual information and posture as a function of age is not attributable to a decrease in the variability of the postural response. In fact, it appears from this most recent set of analyses that it is just as plausible to suggest that increased variability in postural sway is associated with improved coupling between posture and walt movements. We return to this finding in the Discussion section.

Discussion

A clear and consistent developmental pattern emerges from the results of this study. Most infants began to sit alone by 6 to 7 months of age, but they began to visually modulate


their postural response even before they were able to maintain postural equilibrium. Even 5-month-old infants showed some entrainment to the driving frequency of the walls, and this response became more consistent with age and experience. After 9 months of age, the response leveled off, which suggests that it improves primarily during the period of time when infants are learning to sit without support. Infants also showed developmental changes in the total magnitude of their postural responses. Unlike the measure of frequency entrainment, this measure did not show a linear trajectory with age. Infants showed a significant decrease in the RMS of their center of pressure at 7 months of age, but this decrease was transient and RMS returned to higher levels at older ages. Taken together, these findings suggest that the developmental changes related to visual control of posture are somewhat complex and vary with immediate context or stimulus information, infants' sitting experience, as well as the variables used to measure postural control. Each of these issues is discussed in turn.

Perception of Stimulus Information

The results from both the spectral analyses and the cross-correlations revealed that infants at each age showed some entrainment of their postural responses to the driving frequency of the wall movements, and this entrainment improved significantly between 5 and 9 months of age. In addition to showing improvements with age, infants also showed improvements as a function of increases in stimulus amplitude and frequency. One likely reason that frequency entrainment continued to increase through 0.6 Hz for infants is that the natural sway frequency for sitting infants is closer to 0.6 Hz than to 0.3 Hz. Preliminary findings from a new study in our lab support this conjecture.

The improvement in performance as a function of stimulus amplitude was somewhat unexpected. Recall that the magnitude of entrainment by adult observers in the van Asten et al. (1988) study was unrelated to stimulus amplitude. One hypothesis for this discrepancy is that adults respond to the stimulus information as consistently for slow velocities as for fast velocities (which covary with stimulus amplitude) but that infants show an advantage for the faster velocities. Another hypothesis for this discrepancy is that the stimulus information in the study by van Asten et al. (1988) was restricted to one component of optical flow (i.e., rotation) that drove posture along the mediolateral axis, whereas the stimulus information in the current study included all four components of optical flow driving posture along the anteroposterior axis. It is possible that either stimulus or sway axis differences or both contributed to the discrepancy between the two studies. Apparently, the process by which infants respond to optical flow information requires further investigation.

In spite of these unanswered questions, it is clear that infants are quite successful in modulating their posture. This form of postural control is impressive, because to keep pace with the back-and-forth movement of the walls it is necessary not only to perceive the changing direction of the trunk and head but also to scale the restorative forces to the

magnitude of the displacement. The findings from this investigation thus provide evidence that infants perceive not only the direction of their postural sway but also the amplitude of their sway. If both dimensions were not perceived veridically, then it would not be possible for infants to keep pace with the driving frequency of the wall movements. It is important to emphasize that for successful performance in the moving room task it is not necessary for the direction or the amplitude of the postural sway to be perceived explicitly. Rather, the key is that both dimensions are used implicitly by the visuomotor system to produce the appropriate compensatory actions (Bertenthal & Rose, 1995; Lee, 1993). It is necessary to detect both the orientation and the inertia of the trunk to restore it to a more stable posture. This information will vary from one moment to the next during the trial and will require different muscle activation patterns as a function of the currently perceived posture. The high level of entrainment shown by infants reveals their success in adjudicating this complex process.

Some additional insight into this process is provided by the similar performance displayed in the constant and variable frequency conditions. If the frequency of sway was controlled by some higher level representation of the oscillation rate, then it would seem reasonable to expect that the repetitive presentation of the same stimulus frequency would increase the entrainment shown by the infant because the representation would improve over time. If, however, the visual control of posture is more similar to a real-time control system, then repetition of the same stimulus would offer little advantage. On the basis of the findings from this investigation, it does not appear that stimulus repetition was necessary for optimal performance. Recall that the correlation coefficients between room movements and postural sway showed little difference as a function of condition. It is interesting that Schmuckler (1997) presents similar findings based on a related moving room study with 3--6-year-old children. It thus appears that visual control of posture operates in real time with little contribution from higher level processes.

One possible exception to this conclusion is the finding that the temporal coupling between postural sway and wall movements improved between 5 and 7 months of age in the constant frequency condition but did not show a comparable improvement in the variable condition until 7-9 months of age. Although this finding could be interpreted as a processing advantage for the repetitive stimulus condition, a more parsimonious interpretation is that the variable frequency condition included higher frequencies and smaller amplitudes that were physically more challenging to the younger infants (i.e., they lacked the muscle strength necessary to maintain a closer temporal relation between wall movements and their postural responses). Additional evidence supporting the similarity in performance in the two conditions is that infants showed a substantial decrease in their latency to respond to the variable wall movements between 5 and 9 months of age, and this decrease was comparable to that shown by infants to the constant wall movements. Currently, it is not clear whether the decrease in the lag time is specifically a function of improvements in the speed of


detection or the speed of responding to the perceived displacement, but it is intriguing that this change coincides with the development of independent sitting. This correspondence suggests that the decrease in lag time is related to the development of muscle strength and coordination that accompanies sitting.

Effects of Age and Sitting Experience

Our sample showed good consistency with the developmental norms on the onset of independent sitting. The mean age of onset was 6.0 months. At 5 months of age, only 1 infant was able to sit unassisted, whereas by 9 months of age all infants were sitting independently. It is during this time period that infants show marked improvements in the coupling between vision and posture.

From a developmental standpoint, one of the most significant findings emerging from this investigation is that infants showed visual modulation of their sitting posture even before they were able to sit without support. This result implies that the integration between visual information and motor response synergies does not depend on infants developing first the necessary muscle strength and coordination to maintain an independent sitting posture. The processing of visual information into motor responses that are scaled to the direction and displacement of the trunk and head is already functional a few weeks prior to the onset of independent sitting. This finding is reminiscent of the report by Jouen et al. (in press) that neonates show head movements that are scaled to the velocity of the visual stimulus before they are capable of supporting their heads. Likewise, the study by Delorme et al. (1989) reports that infants show visual modulation of their upright posture before they are able to stand without support. It thus appears that visual information is coupled with postural responses even before they function independent of external support.

Still, it is clear from the results of this study that visual control of posture continues to improve with age. What factors contribute to these changes in visuomotor coordination? One likely candidate is sitting experience, which shows a high correlation with postural entrainment by 9 months of age. This finding is consistent with a proposal by Hirschfeld and Forssberg (1994) that postural adjustments necessary for sitting are organized at two different levels. Initially, the muscles involved in postural adjustments of the trunk show an incomplete spatial pattern of muscle activation when stimulated by a loss of balance. As infants gain experience with sitting, the precise order of the spatial- temporal muscle activation pattern becomes configured and the magnitude of the activity is scaled to the mass and displacement of the trunk. Presumably, this improvement in the muscle activation pattern is a function of the perceptual modulation that develops with sitting experience.

Although Hirschfeld and Forssberg (1994) did not test this hypothesis directly, two recent studies offer evidence consistent with their proposal. In the first study (Sveistrup & Woollacott, in press), standing infants were given intensive experience with postural perturbations for 3 days, and then muscle activation patterns were assessed and compared with

the pattern displayed by a control group of infants. The results revealed that infants who had experienced the additional postural perturbations showed more consistent and shorter latency muscle activation patterns than did control infants. Similar findings were reported by Hadders- Algra, Brogren, and Forssberg (1996b), who showed that daily practice with less stable sitting postures facilitated the development of a more consistent muscle activation pattern. It thus appears that practice and experience are associated with improvements in the perceptual modulation of posture.

Comparison Between Different Response Measures

Two different responses were assessed in this study--total sway magnitude and frequency entrainment. As previously discussed, frequency entrainment showed significant improvements with age. By contrast, sway magnitude showed a significant decrease at 7 months of age but increased again at 9 months of age. Sway magnitude was measured by calculating the RMS of the infant's center of pressure. In most studies with adults (Ashmead & McCarty, 1991; Paulus, Straube, & Bran&, 1987; Turano, Herdman, & Dagnelie, 1993), RMS and other measures of sway magnitude are reported to decrease as stability increases. If the same relation applied to the current findings, it would seem to imply that older infants are less posturally stable than younger infants. This conclusion is clearly wrong and logically implausible. A more likely interpretation for this discrepancy between adult and infant data is that the adult results are based on studies of quiet stance, whereas the current findings are based on a dynamic task involving a more active response to a continuously changing visual scene. The process of dynamic equilibrium is clearly different from the process of static equilibrium (Alexander, 1992; Freedland & Bertenthal, 1994; Raibert, 1986), and maintaining a balanced posture in a dynamic task requires different strategies than does maintaining posture in a static task (Bertenthal, Boker, & Rose, 1995; Pozzo, Levik, & Berthoz, 1995). It is precisely for this reason that sway magnitude increased rather than decreased at 9 months of age.

In some ways, this finding is similar to the classic study by Arutyunyan, Gurfinkel, and Mirsky (1969) showing that expert pistol shooters produce more movement between their wrists and shoulders when aiming their guns than do novice pistol shooters. This movement allows the experts greater flexibility, which translates into greater accuracy. Similarly, the increased sway movements of the 9- and 13-month-old infants provide them with greater flexibility to maintain a balanced posture.

It is unlikely that this same explanation applies to the performance of the 5-month-old infants because they were truly unstable (i.e., they could not yet sit without support). Thus, the interpretation for the magnitude of RMS differs as a function of whether infants are capable of controlling their own sitting posture. It is interesting that RMS was lowest at 7 months of age, which corresponds to the time when infants are just beginning to sit unsupported. Learning to sit is similar to the development of other motor skills and involves an extended period of practice, during which time degrees of


freedom relating to the behavior are decreased (Bernstein, 1967). This process is accomplished by increasing muscle stiffness and locking some of the joints. As a consequence, movement is significantly reduced during this period of time.

Concluding Remarks

Two generalizations emerge from this research. The first is that the visual control of sitting is not rate limited by the perception of the stimulus information. Our findings confirm that infants are sensitive to the visually specified direction of self-motion well before they are able to sit unsupported. As such, this finding is consistent with those from many previous studies suggesting that perception of sensory information specifying self-motion is available at very young ages (e.g., Butterworth, 1992; Jouen, 1990; Woolla- cott & Sveistrup, 1994). For this reason, the developmental task for the infant is geared primarily toward learning to scale or map the sensory information to the motor response synergies.

The second generalization is that the coordination of visual and motor responses appears to show significant improvements with sitting experience. Initially, infants show considerable variability in their postural responses to visual information. This variability is manifested by less entrainment and longer temporal lags between posture and wall movements at younger ages. As infants gain greater experience with controlling their sitting posture, they tend to increase the speed and precision of the coupling between their perceptions and their actions. Thus, a plausible conclusion is that infants are tutored by their own experiences to improve their coordination between vision and posture.

As a final comment, it is instructive to note that the coordination observed between vision and posture in this study is not likely to show much evidence of generalization to other postures, such as upright stance. When infants begin standing without support, they again show difficulties in maintaining postural stability and scaling their responses to the perceived displacement (Bertenthal & Bal, 1990; Wool- lacott & Sveistrup, 1994). It thus appears that little savings accrues from the development of visuomotor coordination of sitting to the development of visual control of upright stance. This finding is consistent with recent neuroanatomi- cal speculations that visual control of actions involves a direct mapping to specific motor response loci, and thus the mapping between vision and any one action, such as sitting, does not generalize to other mappings, such as visual control of upright stance (Goodale, 1988; Milner & Goodale, 1995). As a consequence of this neural organization, visuomotor learning is highly specific and somewhat redundant, but once learned, visuomotor responses are extremely rapid and sensitive to contextual variables. The development of the visual control of sitting offers compelling evidence for this conclusion.

References

Alexander, R. M. (1992). Exploring biomechanics: Animals in motion. New York: W. H. Freeman.

Andersen, G. J., & Dyre, B. E (1989). Spatial orientation from optic flow in the central visual field. Perception & Psychophys- ics, 45, 453-458.

Arutyunyan, G., Gurfinkel, V., & Mirsky, M. (1969). Investigation of aiming at a target. Biophysics, 13, 536-538.

Ashmead, D. H., & McCarty, M. E. (1991). Postural sway of human infants while standing in light and dark. Child Develop- ment, 62, 1276-1287.

Banks, M. S., & Dannemiller, J. L. (1987). Infant visual psychophysics. In P. Salapatek & L. Cohen (Eds.), Handbook of infant perception. Vol. 1. From sensation to perception (pp. 115-184). San Diego, CA: Academic Press.

Bayley, N. (1969). Manual for the Bayley Scales of Infant Development. New York: Psychological Corporation.

Bernstein, N. (1967). Coodination and regulation of movement. New York: Pergamon Press.

Bertenthal, B. I. (1996). Origins and early development of perception, action, and representation. Annual Review of Psychology, 47, 431-459.

Bertenthal, B. I., & Bai, D. L. (1989). Infants' sensitivity to optical flow for controlling posture. Developmental Psychology, 25, 936-945.

Bertenthal, B. I., & Bai, D. L. (1990). Application of biomechanical principles to the study of perception and action. In H. Bloch & B. I. Bertenthal (Eds.), Sensory-motor organization and development in infancy and early childhood (pp. 243-260). Dordrecht, The Netherlands: Kluwer.

Bertenthal, B. I., Boker, S. M., & Rose, J. L. (1995). Dynamical analyses of postural development. Journal of Sport & Exercise Psychology, 17, 8.

Bertenthal, B. I., & Bradbury, A. (1992). Infants' detection of shearing motion in random-dot displays. Developmental Psychol- ogy, 28, 1056-1066.

Bertenthal, B. I., & Clifton, R. K. (in press). Perception and action. In D. Kuhn & R. Siegler (Eds.), Handbook of child psychology, Vol. 2. Cognition, perception, and language.

Bertenthal, B. I., & Rose, J. L. (1995). Two modes of perceiving the self. In E Rochat (Ed.), The self in infancy: Theory and research (pp. 303-324). Amsterdam: Elsevier.

Butterworth, G. (1992). Origins of self-perception in infancy. Psychological Inquiry, 3, 103-I 11.

Butterworth, G., & Hicks, L. (1977). Visual proprioception and postural stability in infancy: A developmental study. Perception, 5, 255-263.

Delorme, A., Frigon, J., & Lagace, C. (1989). Infants' reactions to visual movement of the environment. Perception, 18, 667--673.

Freedland, R. L., & Bertenthal, B. I. (1994). Developmental changes in interlimb coordination: Transition to hands-and knees crawling. Psychological Science, 5, 26-32.

Goodale, M. A. (1988). Modularity in visuomotor control: From input to output. In Z. Pylyshyn (Ed.), Computational processes in human vision: An interdisciplinary perspective (pp. 262-285). Norwood, NJ: Ablex.

Hadders-Algra, M., Brogren, E., & Forssberg, H. (1996a). Ontog- eny of postural adjustments during sitting in infancy: Variation, selection and modulation. Journal of Physiology, 493, 273-288.

Hadders-Algra, M., Brogren, E., & Forssberg, H. (1996b). Training affects the development of postural adjustments in sitting infants. Journal of Physiology, 493, 289-298.

Hirschfeld, H., & Forssberg, H. (1994). Epigenetic development of postural responses for sitting during infancy. Experimental Brain Research, 97, 528-540.

Jenkins, G. M., & Watts, D. G. (1968). Spectral analysis and its applications. San Francisco: Holden-Day.

Jouen, E (1990). Early visual-vestibular interactions and postural


development. In H. Bloch & B. I. Bertenthal (Eds.), Sensory- motor organization and development in infancy and early childhood (pp. 199-215). Dordrecht, The Netherlands: Kluwer.

Jouen, E, Lepecq, J. C., Gapenne, O., & Bertenthal, B. I. (in press). Optical flow sensitivity in neonates. Child Development.

Koenderink, J. J. (1986). Optic flow. Vision Research, 26, 161-179. Lee, D. N. (1993). Body-environment coupling. In U. Neisser

(Ed.), Ecological and interpersonal knowledge of the self (pp. 68-88). Cambridge, England: Cambridge University Press.

Lee, D. N., & Aronson, E. (1974). Visual proprioceptive control of standing in human infants. Perception & Psychophysics, 15, 529-532.

Milner, A. D., & Goodale, M. A. (1995). The visual brain in action. New York: Oxford University Press.

Newell, K., & Corcos, D. M. (1993). Issues in variability and motor control. In K. M. Newell & D. M. Corcos (Eds.), Variability and motor control (pp. 1-12). Champaign, IL: Human Kinetics.

Paulus, W., Straube, A., & Brandt, T. H. (1987). Visual postural performance after loss of somatosensory and vestibular functions. Journal of Neurology, Neurosurgery, and Psychiatry, 50, 1542-1547.

Pick, H. L. (1989). Motor development: The control of action. Developmental Psychology, 25, 867-870.

Pozzo, T., Levik, Y., & Berthoz, A. (1995). Head and trunk movements in the frontal plane during complex dynamic equilibrium tasks in humans. Experimental Brain Research, 106, 327-338.

Raibert, M. H. (1986). Legged robots that balance. Cambridge, MA: MIT Press.

Ramirez, R. W. (1985). The FFT: Fundamentals and concepts. Englewood Cliffs, NJ: Prentice Hall.

Schmuckler, M. A. (1997). Children's postural sway in response to low- and high-frequency visual oscillation. Journal of Experimen- tal Psychology: Human Perception and Performance, 23, 528- 545.

Stoffregen, T. A., Schmuckler, M. A., & Gibson, E. J. (1987). Use of central and peripheral optical flow in stance and locomotion in young walkers. Perception, 16, 113-119.

Sveistrup, H., & Woollacott, M. J. (in press). Can practice modify the developing automatic postural response? Experimental Brain Research.

Thelen, E. (1990). Coupling perception and action in the development of skill: A dynamic approach. In H. Bloch & B. I. Bertenthal (Eds.), Sensory-motor organization and development in infancy and early childhood (pp. 39-56). Dordrecht, The Netherlands: Kluwer.

Turano, K., Herdman, S. J., & Dagnelie, G. (1993). Visual stabilization of posture in retinitis pigmentosa and in artificially restricted visual fields. Investigative Ophthalmology and Visual Science, 34, 3004-3010.

van Asten, W. N. J. C., Gielen, C. C. A. M., & Denier van der Gon, J. J. (1988). Postural movements induced by rotations of visual scenes. Journal of the Optical Society of America, A5, 1781- 1789.

von Hofsten, C. (1993). Prospective control: A basic aspect of action development. Human Development, 36, 253-270.

Winter, D. (1979). Biomechanics of human movement. New York: Wiley.

Woollacott, M. H., & Sveistrup, H. (1994). The development of sensorimotor integration underlying posture control in infants during the transition to independent stance. In S. P. Swinnen, J. Massion, & H. Heuer (Eds.), lnterlimb coordination: Neural, dynamical and cognitive constraints (pp. 371-389). San Diego, CA: Academic Press.

Received February 27, 1995 Revision received July 19, 1996

Accepted October 7, 1996 •

Documents

Perception-Action Coupling in the Development of Visual