Visual Reorientation Illusions and Sensory Weightings in Self-Orientation
QUESTION
Write up a short summary of these 2 examples of my research works. Be sure to highlight
similarities and differences and indicate two findings that you think are important.
Abstract. A tilted furnished room can induce strong visual reorientation illusions in stationary subjects. Supine subjects may perceive themselves upright when the room is tilted 90V so that the visual polarity axis is kept aligned with the subject. This `upright illusion’ was used to induce roll tilt in a truly horizontal, but perceptually vertical, plane. A semistatic tilt profile was applied, in which the tilt angle gradually changed from 0V to 90V, and vice versa. This method produced larger illusory self-tilt than usually found with static tilt of a visual scene. Ten subjects indicated self-tilt by setting a tactile rod to perceived vertical. Six of them experienced the upright illusion and indicated illusory self-tilt with an average gain of about 0.5. This value is smaller than with true self-tilt (0.8), but comparable to the gain of visually induced self-tilt in erect subjects. Apparently, the contribution of nonvisual cues to gravity was independent of the subject’s orientation to gravity itself. It therefore seems that the gain of visually induced self-tilt is smaller because of lacking, rather than conflicting, nonvisual cues. A vector analysis is used to discuss the results in terms of relative sensory weightings.
1 Introduction
Tilt of a visual scene about an earth horizontal axis is known to elicit sensations of self-tilt in the opposite direction in erect observers (Asch and Witkin 1948a). The magnitude of visually induced self-tilt depends on field size, but primarily on the contents of the scene (Howard and Childerson 1994; Allison et al 1999). A scene containing familiar objects with recognisable `tops’ and `bottoms’ has a polarity axis that identifies perceptual `up’ and `down’. Such a scene induces more illusory self-tilt than does a visual frame that merely contains cues to horizontal and vertical. However, subjects in a furnished, richly polarised room that is tilted 120V experience self-tilt of only about 15V (Howard and Childerson 1994; Groen 1997). The limited effect is attributed to the conflicting information from the otolith organs in the inner ear, pressure sense organ, and proprioceptors in the body. These `graviceptors’ indicate that the body remains upright while the visual scene indicates that the body is tilted. Judgments of self-orientation involve a compromise between judgments in a visual frame of reference and judgments in a frame of reference defined by information from graviceptors. Apparently, erect observers assign less weight to the visual frame of reference than to graviceptors.
One way to explore the contribution of graviceptors to self-orientation is to compare visually induced self-motion (vection) produced by rotation of a visual scene about a horizontal axis, which affects perceived self-orientation relative to gravity, with vection produced by scene rotation about a vertical axis, which keeps perceived self-orientation to gravity constant. For example, vection has been shown to be stronger for visual scene rotation about a vertical axis than for rotation about a horizontal axis (Howard et al 1987). This observation complies with the finding that vection is more pronounced in weightless conditions (Young and Shelhamer 1990; Mueller et al 1994), and indicates that visual information has a greater effect when gravitational cues are absent or irrelevant. Constant rotation of a visual scene about a vertical axis generally produces
1478 E L Groen, H L Jenkin, I P Howard
unrestrained vection, whereas visual motion about a horizontal axis typically produces a paradoxical sensation of continuous vection combined with limited self-tilt (Held et al 1975).
Another approach is to use the `rod-and-frame’ effect in which a vertical rod enclosed by a tilted frame appears to tilt in the opposite direction to the frame (Wertheimer 1912; Kleint 1936; Asch and Witkin 1948). The rod-and-frame effect is stronger in supine observers than in erect observers (Lichtenstein and Saucer 1974; Goodenough et al 1981), which confirms that visual information becomes more dominant when orthogonal to gravity. The influence of body position on the rod-and-frame effect also demonstrates that judgments of the orientation of a visual object (rod) with respect to a frame are related to judgments of self-orientation with respect to the frame. However, it was not directly investigated whether perceived body orientation itself was affected by the tilted frame. Some evidence for such an effect came from Templeton (1973), who found that subjects made larger errors in setting a line parallel to their longitudinal body axis when they were recumbent than when they were erect. Moreover, in the recumbent posture the error further increased in the presence of a tilted frame. Although this study showed an interaction between the frame effect and body posture, it still did not provide a direct estimate of perceived body orientation with respect to vertical. We could find only one attempt in the literature to investigate visually induced self-tilt about horizontal and vertical axes. For self-`tilt’ of a supine observer about a vertical axis there is no gravitational axis to relate to. To approach this problem, Goodenough et al (1982) asked subjects to indicate their perceived body orientation in a frontal plane with respect to the unseen laboratory walls. They found that a frame subtending a visual angle of 40 deg and tilted 28V in the frontal plane, induced a mean change in apparent body orientation in the opposite direction of about 5V in erect observers and 12V in supine observers. The supine observers were aware that they were supine. Therefore, information from the graviceptors was irrelevant to the task of judging the orientation of the body to the room.
A third way to study the interaction between visual and nonvisual cues to self- orientation is to measure visually induced self-tilt with observers tilted in a vertical plane to various degrees (eg Asch and Witkin 1948b). It has been found that visual orientation cues are less contradicted by gravitational cues when the body is not upright, that is when it is not in the orientation where the otoliths and other graviceptors are most sensitive (Dichgans et al 1974; Young et al 1975; Bischof 1978).
Objectively, self-tilt to gravity has meaning only for rotation about body axes that are not aligned with gravity. However, there could be illusory self-tilt with respect to a perceived direction of gravity. Recently, we found a visual-reorientation illusion that offers a unique opportunity to study visually induced self-tilt in supine subjects who perceive the gravity axis to be in the horizontal plane (Howard et al 1997). This reorienta- tion illusion is created by placing a supine subject inside a furnished room tilted 90V to gravity so that the normally vertical wall is horizontal and above the subject. Many subjects (generally more than 60%) then feel that they are upright in an upright room. In the present study we enquired whether it is possible to induce illusory self-tilt in an objectively horizontal plane in these `visually upright’ subjects by `tilting’ their whole body relative to the visual polarity axis of the 90V-tilted room (that is, tilt about the subject’s roll axis). We hypothesised that in this situation the perceived direction of gravity (the `subjective vertical’, SV) would exactly follow the visually defined vertical, since this is the cue for verticality that the subject is using. Nonvisual cues to verti- cal operate about an orthogonal axis. In contrast, erect subjects exposed to 90V tilt of the room about a horizontal axis were expected to experience only limited self-tilt since it is contradicted by graviceptor inputs that operate about the same axis as the visual cues. We compared visually induced self-tilt in erect observers inside a furnished
Perception of self-tilt in a true and illusory vertical plane 1479
room tilted up to 90V about a horizontal roll axis, with self-tilt about an illusory horizontal axis in supine observers. To obtain a complete set of relative sensory weight- ings, we also studied perceived self-tilt in three additional conditions: actual self-tilt about the horizontal roll axis in a lighted and in a darkened furnished room, and actual self-tilt while the furnished room remained parallel with the longitudinal body axis.
2 Method
2.1 Room
The furnished room consisted of a cubic frame with 8 foot sides lined with plywood on the floor and foam plastic sheets on the walls and ceiling. One side-wall had an interior house door that allowed access to the room. The walls were papered and decorated with various pictures and hanging objects. On the right-hand wall closest to the door, a bookcase held books, a teddy bear, and three cups hanging from cup hooks. In front of the bookcase there was a small kitchen table set with cutlery, plates, cups and saucers, a breadbasket, and a box of tissues. On either side of the table were two wicker chairs. The floor was carpeted and there were baseboards against the walls. Additional details consisted of a full-sized human manikin affixed to one of the wicker chairs in the right-hand corner closest to the door. Another manikin stood in the left-hand corner closest to the door. All objects were firmly held in place by hidden bolts, glue, or Velcro. In the centre of the ceiling was a simple 60 W fluorescent light fixture with a plastic cover. Thus the room provided a clear polarised visual frame and a rich variety of information about the direction of `down’.
The whole structure was mounted on external bearings so that it could be rotated about a horizontal axis passing through the centre of the room. Room rotation was controlled by a computer-controlled DC servomotor. The subject was firmly strapped onto a flat vertical bed, with the feet supported by a small foot-rest about 10 cm above the floor. The chest and knees were secured between cushioned wooden plates, and the head was held in a head-support. In one condition, the bed was supported on a boom protruding through the wall facing the bookcase. The boom was aligned with the axis of rotation of the room so that the bed was orthogonal to the axis of room rotation. In a second condition, the bed was supported on an axis in the middle of the wall facing the door, so that it was parallel to the axis of room rotation. In both positions the bed could be rotated by a computer-controlled motor about the roll axis passing through the centre of the subject’s body. Although the subject faced the wall with the bookcase in one condition, and the wall with the door in the other condition, we assumed that the fully furnished room was equally powerful for each position of the bed.
2.2 Conditions
There were five conditions of actual or apparent whole-body tilt about the subject’s roll axis. In all conditions, the tilt angle was varied systematically between 0V and 90V. Figure 1 shows a cartoon of each condition. In conditions Self light and Self dark, the room remained upright and the subject was tilted about the horizontal roll axis with eyes open or eyes closed. In the Self Room condition, both subject (with eyes open) and room were tilted so that the body remained aligned with the polarity axis of the room. In a purely visual condition (Room), the subject was erect and the room was tilted about the horizontal roll axis. In all these conditions, gravity was in the plane of tilt. In the Supine condition, the bed was rotated about an axis orthogonal to the axis of room rotation and the subject was tilted about a vertical roll axis. Thus gravity was orthogonal to the plane of tilt. The subject was first strapped onto the vertical bed. The bed and the room were then slowly tilted 90V backwards at the same speed so that the same wall remained in the frontal plane of the subject. For those subjects who perceived themselves and the room as upright in the supine position,
1480 E L Groen, H L Jenkin, I P Howard
Self light Self dark Room Self Room
i gggg
ii
Figure 1. Upper part: Cartoon of the five conditions of tilt of the subject, the room, or both. Bottom part: vector diagrams showing the corresponding combinations of the gravitational (g), visual (v), and body (i) frames of reference (for explanation, see section 4).
rotation of the bed about a vertical axis produced apparent tilt of the self with respect to the illusory direction of gravity.
As shown in the upper part of figure 1, these five conditions combined visual and gravitational cues in different ways, providing the opportunity to estimate their (relative) weighting in the determination of the SV. The two cues were aligned and congruent (Self light), oppositely directed and incongruent (Room, Room Self), or one of the two was absent or irrelevant (Self dark, Supine).
2.3 Tilt profile
Each condition started with the subject and the polarity axis of the room aligned (relative orientation of 0V). Then the subject or the room was tilted slowly with max- imum velocity of 4V s1 up to 90V, either clockwise or anticlockwise. At 90V of tilt there was a 5 s pause before changing direction and tilting in one sweep to 90V on the other side. After another 5 s stop in that orientation, the subject or room finally returned to the 0V orientation. Throughout these changes, subjects indicated their SV. Figure 2 depicts the velocity and position trace of the stimulus, together with an example of the time history of simultaneous registration of the SV of one subject (see section 2.4 for a description of the task). The orientation of the subject, the room, and the response will be referenced with respect to the subject’s roll axis, with right-ear-down-tilt signed positive.
To minimise stimulation of the semicircular canals, angular acceleration at the beginning and the end of tilt motion was very low (0.7V s2 ). A constant velocity of 4V s1 prevented the build-up of vection. A previous study on vection showed that 5V s1 of angular motion of a wide visual display did not induce circular vection as long as there were no stationary objects in view that provided relative-motion cues (Howard and Howard 1994). To remove relative-motion cues from the subjects’ body, subjects were asked to look straight ahead. In preliminary trials, three subjects, includ- ing two of the authors, experienced no vection. The percept was described as a gradual change of position, without the feeling of any motion.
2.4 Procedure
Twelve subjects, five women and seven men (average age 37 years) participated in all five conditions. Because reconfiguring the bed between the two axes took about one day, all conditions with tilt in the vertical plane (Self light, Self dark, Room, Self Room)
v
Supine
v
i vvi
Perception of self-tilt in a true and illusory vertical plane
1481
angular velocity
indicated self-tilt
subjective vertical (SV)
25 s
Figure 2. Time histories of angular velocity (upper trace), angle of actual self-tilt (middle), and indicated SV (bottom) in a Self dark trial. Note that the reversal in SV setting at the 90V positions is related to a subtask where subjects briefly indicated the perceived direction of their body axis (see section 2.4).
were presented in the one session, and the Supine condition was presented in a second session on a separate day. Within the first session, the order of conditions was counter- balanced over subjects. Likewise, the starting direction of tilt was counterbalanced over subjects and alternated between conditions within subjects. All subjects experienced each condition once. During each trial, subjects continuously indicated the direction of their SV (pointing to perceived `up’) by setting a 10 cm long rod to the apparent vertical. The rod was fixed to the bed on the right-hand side of the subject at the height of the hips. The rod was not visible, so that settings were done by touch. The rod could rotate in the roll-plane through 360V with just enough friction to keep it in position when released. The orientation of the rod was read by a potentiometer with a resolution of 0.5V. Since the rod was fixed in the pitch-plane, the amount of perceived pitch had to be reported verbally. Subjects were instructed to point the rod towards their feet when they felt confused about their orientation, so that these episodes could be recognised in the offline analysis.
In the vertical-plane conditions, sensors monitored the tilt angle and the velocity of the stimulus (bed or room). Together with the potentiometer output of the tactile rod, these signals were written to a data logging system with a sample frequency of 16 Hz.
30V 30V 4V s1
1482 E L Groen, H L Jenkin, I P Howard
In the Supine condition, stimulus angle was derived from the angular velocity. Since the velocity profile in Supine conditions closely matched that in the other conditions, we felt confident in taking one of the sensor recordings from a Self dark condition as a substitute for tilt angle. Rather than integrating the velocity signal, this was used to synchronise the subjects’ response with the tilt angle.
During the 5 s pause in each extreme tilt position (90V and 90V) subjects were asked to briefly set the rod to the perceived longitudinal body axis. Consequently, the SV recordings show a brief interruption, as in the bottom trace of figure 2. This measure- ment was intended to indicate whether the perceived orientation of the body axis relative to the rod was also affected by the visual frame of reference. However, subjects often were confused about this subtask, and the data will not be further discussed. The indication of the body axis sometimes introduced a considerable bias in the SV recording, since the tactile rod was never returned exactly to the orientation it had just before the interruption. To facilitate the interpretation of tactile-rod settings, subjects were also asked to verbally report on their perceived self-orientation and give an estimate of the maximum angle of perceived tilt in degrees.
2.5 Data analysis
The tactile-rod settings were analysed as a function of stimulus tilt angle. For this purpose, samples were pooled into stimulus intervals (bins) of 1V. Figure 3 shows a trace of rod settings plotted against stimulus angle. The interruptions at the 90V positions, due to the indication of the body axis, were removed. The loop in the figure resulted from the bias induced by this body-alignment task.
| IV
|
|
| III
|
I
II
|
90 60 30
0 30 60
90
90 60 30 0 30
Self-tiltaV
Figure 3. Example of raw setting of the tactile rod as a function of stimulus angle. In this case the stimulus consisted of actual self-tilt, starting in a positive direction (right- ward) in segment I, then changing direction in segment II, tilting entirely to the other side (segment III), and finally returning to upright (segment IV). The negative slope of the signal is due to the fact that the sub- ject compensated for perceived body tilt as to indicate the SV. The irregularities at the end-points are related to the body-alignment task.
60 90
Next, the whole SV trace for each condition was divided into four segments corre- sponding to (for a subject starting with rightward tilt):
I. the first tilt sweep from 0V to 90V;
II. the first half of the sweep to the other side: from 90V to 0V; III. the second half of the sweep to the other side: from 0V to 90V; IV. the final segment from 90V to 0V.
In each segment the response gain was estimated by linear regression analysis on the response. This yielded a slope and intercept. The intercept was ignored, since it was considered an artifact from the body-alignment task in the endmost positions. There- fore, the results will be discussed in terms of the estimated slope or `gain’ of perceived self-tilt, where a value of 1 means that indicated self-tilt changed in an equal fashion
Indicated SVaV
Perception of self-tilt in a true and illusory vertical plane 1483
with the stimulus angle. The gain was made positive when the response was in the expected direction. For example, rod settings to vertical were expected to be in the opposite direction as actual self-tilt, but in the same direction as room tilt, since room tilt induces apparent self-tilt in the opposite direction.
A within-subjects analysis of variance (ANOVA) was used to test differences between conditions. Effects will be considered significant with p S 0X05.
3 Results
The data of one subject were not registered completely. Another subject was very confused in the visual conditions Room, Self Room, and Supine, and produced inconsistent responses. The data of both subjects were discarded, so that data of ten subjects were finally included in the analysis.
Four subjects did not experience the `upright illusion’ in the Supine condition, and were thus unable to indicate their SV as a function of illusory roll tilt when they were supine. For all tilt angles in the Supine condition, these subjects reported veridical tilt of both themselves and the room. The other six subjects experienced an `upright illusion’, which means that they continuously perceived themselves and the room as being upright or within 20V to upright [in accordance with the criterion used in a study of Howard and Hu (2001)] during simultaneous tilt of self and room. Obviously, these subjects strongly relied on the visual frame of reference, and will be referred to as `visually dependent’, as opposed to the four subjects who were `visually independent’. One subject experienced the upright illusion for angles of roll tilt of the body up to 60V but, with larger angles of roll tilt, he reported a combination of roll tilt of 80V and backwards pitch of 45V. He reported to be confused and pointed the tactile rod towards his feet. The slope of his SV settings was estimated only over the first 60V.
The differences in rod settings between the Supine condition and the other four conditions could be examined only in the visually dependent subgroup. The bar plot in figure 4 shows the gain values per condition averaged for both subgroups. Table 1 also gives the corresponding standard deviations, together with the descriptive statistics of the verbal estimates of self-tilt. The verbal gain was calculated by dividing the maximum reported angle by the maximum stimulus angle of 90V (except for the subject who experienced the upright illusion only up to 60V: his verbal gain was calculated at 60V). In figure 5, the individual gain values derived from the rod settings are plotted against the corresponding verbal estimates of perceived self-tilt. This shows that the verbal estimates tend to cluster at 15V intervals, which is especially evident at 90V. Since the rod settings showed a higher resolution than the verbal estimates, our analysis was concentrated on the rod settings.
Visually dependent group Visually independent group
1.0
0.8
0.6
0.4
0.2
0.0
Figure 4. Mean slope of the rod settings, ie the `gain’, of six visually dependent subjects who experi- enced the upright illusion, and of four visually independent subjects who did not experience the illusion in the Supine condition.
Self light
Self dark Room Condition
SelfRoom
Supine
Slope of SV (gain)
1484 E L Groen, H L Jenkin, I P Howard
Table 1. Mean and standard deviation of the estimated gain derived from the rod settings (SV gain), and the corresponding gain values derived from the verbal judgments of maximum self-tilt (verbal gain), separated for the visually dependent and visually independent groups.
Condition
Self light Self dark Room
Self Room Supine
Visually dependent group
Visually independent group
SV gain 0X79 0X25
0X73 0X33 0X44 0X34 0X13 0X20 0X49 0X17
verbal gain 1X06 0X14
1X02 0X11 0X69 0X34 0X32 0X27 1X04 0X14
SV gain 0X90 0X39
0X91 0X33 0X33 0X32 0X77 0X46 n/a
verbal gain 1X18 0X15
1X19 0X17 0X54 0X37 0X81 0X42 n/a
1.2
1.0
0.8
0.6
0.4
0.2
0.0
Figure 5. Gain values of the rod setting plotted against the verbal estimates of maximum perceived self-tilt of all subjects over all con- ditions. The dotted line is the linear regression line.
0 30 60 90 120 Verbal estimate of tilt angleaV
Since there was no effect of stimulus direction in the first tilt segment (from 0V to 90V) of the Self dark condition, we pooled the gain data for initial stimulus direction in all conditions. To test differences between the Supine condition and the other condi- tions for the visually dependent group we used the design Condition (5)TSegment (4), and tested differences between the four vertical plane conditions for all ten subjects using the design Visual subgroup (2)TCondition (4)TSegment (4).
On the basis of the analysis of the four vertical-plane conditions, there was a main effect of condition and a significant two-way interaction between Condition and Visual subgroup. There was no main effect of Segment, nor a significant three-way interaction including Segment as factor. According to a posteriori Tukey test, signif- icant differences existed between conditions Self light and Self dark on the one hand, and Room on the other hand. Subjects were quite accurate in indicating actual self- tilt (gain close to unity), whether in a visible room or with their eyes closed. Apparent self-tilt in a tilting room, however, followed the visual cues with a gain of only about 0.4. The verbal reports on self-tilt show a similar pattern (table 1). Furthermore, in the Self Room condition the responses of the visually dependent group were significantly different from those of the visually independent group. The high gain found in the visually independent group, indicates that these subjects almost fully compensated for the angle of self- (and room) tilt, and the tactile rod was set close to the earth-vertical. In contrast, the low gain in the visually dependent group indicates that these subjects set their SV close to the visual vertical defined by the room. This is consistent with the behaviour of both groups in the Supine condition. For the visually dependent group, the mean of rod settings in the Supine condition did not differ from that in the Room condition. This does not conform to our hypothesis that visually induced self-tilt would be restricted by gravitational cues in the Room, but not in the Supine, condition. However, the standard deviation of rod setting in the Supine condition was smaller than in the Room condition. The filled circles in figure 6 show a scatter plot
Gain of rod setting
Perception of self-tilt in a true and illusory vertical plane 1485
3
62
2
1
45 1
4 6
3
5
1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0
Verbal estimate Rod setting
Figure 6. Correspondence between the rod- setting gain (filled symbols) of all six visually dependent subjects between the Room con- dition and the Supine condition. The open circles show the correspondence between the verbal estimates of self-tilt in both con- ditions. Numbers identifying subjects are shown next to the data points.
0.0 0.2 0.4
Room condition
0.6 0.8 1.0
of the rod-setting gain between the two conditions for the visually dependent subjects. The open circles in figure 6 show the corresponding gain of verbal responses. Clearly, verbal estimates were more scattered in the Room condition than in the Supine con- dition, where the data seemed to cluster around a gain of 1 (indicating a purely visual response).
4 Discussion
In this study we have shown that it is possible to induce a percept of lateral body tilt in supine observers who experience the `upright illusion’ in a 90V tilted visual scene. Earlier reports, in which a simple visual frame was used, showed that supine subjects perceive a change in body orientation with respect to the unseen surroundings when viewing a simple frame tilted about a vertical axis. But these subjects veridically perceived themselves as supine and therefore judged self-orientation in a correctly perceived hori- zontal plane. Our supine subjects perceived themselves as being upright. They therefore judged self-orientation in a truly horizontal but perceptually vertical plane. Similarly, in weightlessness, where the gravitational frame of reference is absent, astronauts nevertheless experience a given direction as `down’. One can speak of perceived self-`tilt’ to gravity only when subjects perceive themselves as upright. When subjects perceive themselves as supine, self-orientation can be defined with reference to a visual frame of reference or an imagined frame of reference.
4.1 Condition Room versus Supine
Our main hypothesis was that the gain of visually induced self-tilt would be smaller in upright observers than in supine observers who perceive themselves as upright, because gravitational cues contradict self-tilt in the former situation but are irrelevant in the latter. However, for visually dependent subjects, tactile judgments of the SV in the Room and Supine conditions were indistinguishable, with an average gain of about 0.5. One would expect that in the Supine condition subjects would follow the visual cues in an `all-or-nothing’ manner (gain of 0 or 1), since visual cues provided the only frame of reference for orientation about the roll axis. The verbal responses showed evidence of this type of behaviour (figure 6), since all estimates were close to 90V in the Supine condition but not in the Room condition. Apparently, in the Supine condition, the visual cues affected verbal estimates more strongly than they affected rod settings. This may be because verbal estimates fell within discrete categories. The similarity of rod settings in the Supine and Room conditions suggests that visually induced self-tilt
Supine condition
1486 E L Groen, H L Jenkin, I P Howard
in the erect observer is not limited by counteracting inputs from graviceptors, but by the lack of appropriate tactile and proprioceptive inputs. This fits in with the comment of subjects that, in the Room and Supine conditions, they experienced a convincing though unusual sensation of self-tilt without feeling pressure against a side of the bed. The idea of missing proprioceptive cues may offer an alternative explanation for the observation that circular vection about a horizontal roll or pitch axis is strongest when the subject is not erect (Asch and Witkin 1984b; Young et al 1975; Bischof 1978). It is usually assumed that gravitational cues are strongest when head and body are aligned with gravity. However, the asymmetrical pressure cues when the subjects are tilted in a vertical plane may facilitate sensations of vection, since they suggest self- rotation in the same plane as the visual stimulus. In contrast, pressure cues in erect and supine subjects, as in our Room and Supine conditions, are symmetrical and suggest that the body is stationary.
4.2 Semistatic tilt stimulus
Our subjects indicated actual self-tilt with a gain close to unity, which is close to ideal performance. It is slightly surprising that performance was not different between Self light and Self dark conditions. Apparently, nonvisual cues were sufficient in the absence of visual cues. Therefore, it seems that congruent visual cues do not improve gain. They may reduce response variability, but our method did not provide information on this point.
The effect of semistatic room tilt in the present study was considerably higher than that usually found with static tilt of a visual stimulus. For example, in previous studies with static tilt of the same room by 120V, induced illusory self-tilt of erect subjects was at most 15V (Howard and Childerson 1994; Groen 1997). In the present study, erect subjects indicated visually induced self-tilt up to about 60V. We measured the SV `on the fly’ during gradually increasing tilt at a barely perceptible angular velocity. We did this to avoid suddenly repositioning the subjects. Repositioning stimulates the vestibular system (especially the semicircular canals) and may have unwanted after- effects. Although visual motion itself can induce vection, we chose the angular velocity under the threshold of 5V s1 that has been reported for yaw vection (Howard and Howard 1994). Indeed, our subjects did not report any vection. Rather than visual motion, we think that the continuous presentation of the visual scene itself enhanced its effectiveness. A gradual change of visual orientation cues is more likely to be associated with a change of self orientation than a `snapshot’ of a scene that is usually presented in the traditional static method.
4.3 Relative sensory weightings
So far, we have considered the importance of two objective frames of reference: one determined by the direction of gravity, and the other determined by the polarity axis of the room. However, it is known that the main body axes may determine an addi- tional frame of reference for judgments of self-orientation. Here we will make an attempt to quantify the relative weighting between these three frames of reference. We will assume that in a given situation the SV is determined by the sum of three vectors: a gravitational vector g defined by gravity, a visual vector v defined by the visual frame of reference, and an `idiotropic’ vector i defined by the longitudinal body axis. The concept of the idiotropic vector originated with Mittelstaedt (1983). The tendency to use the body axis as a reference for `down’ becomes especially evident in astronauts who, after a few days in weightlessness, consider the surface below their feet as the floor.
Depending on the experimental condition, the three vectors can be aligned in different ways. They can all be congruent, or the visual and/or gravitational frame of reference can be out of alignment with the body (figure 1). For instance, during actual self-tilt in a lighted room (Self light), the idiotropic vector tilts with respect to
Perception of self-tilt in a true and illusory vertical plane 1487
the combined gravitational and visual vector. Conversely, in condition Self Room, the combined ideotropic and visual vector tilts relative to the gravitational vector. Thus, one can generalise that, in each condition, there is a `stimulus’ vector (s) that is coupled to the stimulus and tilts relative to a `fixed’ vector ( f ). Both vectors s and f may consist of g, v, or i, or a combination of two of these. The left diagram in figure 7 shows how the vectors sum up to a resultant vector that represents the combined effect of the three different frames of reference. Theoretically, this resultant vector represents the SV, and c represents the angle that subjects indicated with the tactile rod.(1) This angle is determined by stimulus angle f, and also by the length of the composing vectors. The lengths of g, v, i reflect the extent to which the SV conforms to the respective frames of reference. We can obtain an estimate of these sensory weightings from the relation between response angle c and stimulus angle f. Since we are inter- ested in the relative weighting between the three frames of reference, we can normalise the fixed vector ( f 1). Then the relation between c and f can be described as
follows: ws sin f carctan ws cosf1
where ws is the relative length of the stimulus vector. The right-hand side of figure 7 shows how function c(f) varies for different ws . To estimate ws in each condition, we took the average gain of the visually dependent subjects (ie the slope of the regression line fitted to the rod settings), and calculated the value of ws that would yield this gain. Thus we approximated the function c(f) with a linear regression line, similar to the analysis of the rod settings.
yx
s
1.0 0.8
0.6 0.4
0.2 0
c
f
vector SV
s
45
30
15
0
Figure 7. Vector diagram showing how, in each condition, a stimulus vector s tilts away over angle f from a fixed vector f. The resultant vector reflects the SV, which makes an angle c with vector f. The right-hand plot shows how response angle c varies as a function of stimulus angle f for various strengths of s.
Table 2 shows the calculated values of ws for each condition. The table gives five equations with three unknown variables (g, v, and i). Because we normalised vector f, we finally obtain the lengths of v and i relative to that of g. Condition Self dark yields a ratio between i and g of 0.4. When substituted in the equation of Room condition, this
(1) In fact, since the SV was measured by means of a tactile rod that was fixed to the bed, the angle indicated by the subjects was either c (in conditions Room and Supine) or 1 c (in conditions Self light, Self dark, and Self Room).
0 30 60 90
faV
f
caV
1488 E L Groen, H L Jenkin, I P Howard
Table 2. Calculated value of weighting factor ws of the tilt stimulus in each condition that would produce the measured slope of the SV settings. This ws gives the length of stimulus vector s relative to the fixed vector f. Depending on the conditions, s and f are composed of different combinations of g, v, and i, yielding five different equations for the determination of the relative strengths of g, v, and i.
Condition
Self light Self dark Room
Self Room Supine
gives a ratio between v and g
of Self light, we obtain another estimate for the ratio between i and g of about 0.5. This is reasonably consistent with the first estimate of 0.4. However, when we sub- stitute the ratio of 0.4 in the equation of Self Room we get another estimate for the ratio between v and g of 5.6, which is more than three times the first estimate for this ratio. This nonlinearity suggests that the combined effect of the aligned visual and idiotropic frames of reference is stronger than the effect of each frame of reference by itself. This finding is in agreement with the study of Howard and Hu (2001), who found that visual reorientation illusions occur most frequently when these two frames remain aligned.
4.4 Practical implications
The upright illusion is interesting in relation to vehicle simulation. It is evident that moving base simulators can only make limited displacements. For this reason, low- frequency linear accelerations producing relatively large displacements are usually simulated by tilting the simulator (Rolfe and Staples 1988). By this procedure, gravity is oriented with respect to the simulator in a similar way as the gravito-inertial force would be oriented with respect to real cockpit. When the angular velocity of simulator tilt is kept below a certain limit, the simulator pilot perceives a linear acceleration rather than the simulator tilt itself (Groen and Bles 1999). Thus one can say that the pilot experiences a type of upright illusion. The visual display of a moving base simulator provides a stimulus in a coordinate system that is fixed to the pilot, analogous to condition Self Room. Our results in that condition showed that the combined effect of the aligned visual and idiotropic frames of reference is stronger than the effect of a tilted visual frame of reference alone. This may explain why a simulator’s visual stimulus effectively suppresses the perception of self-tilt in favour of the perception of linear self-motion in a horizontal plane. Obviously, the visual stimulus of a flight simulator differs in many aspects from our visual stimulus. One important difference is that of simulated vehicle motion (optic flow). In our study, the upright illusion was produced by a clear and stable visual frame of reference in combination with a clear polarity axis (up and down). It is questionable whether optic flow alone (eg when flying through clouds) is sufficient to suppress the perception of simulator tilt. It seems remarkable that simulator motion is not adjusted for different types of weather. Still, to our knowledge, simulator pilots do not complain about disturbing simulator motion when entering a cloud. Not unlikely, their expectations about the motion and their proficient interpretation of flight instruments may act as a substitute for the lacking visual frame of reference.
s f Equation
i gv iwsgv
i g iwsg
v gi vwsgi vi g viwsg
v i vwsi
Slope SV ws
0.79 0.2 0.73 0.4 0.44 1.2 0.13 6.0 0.49 2.0
of 1.7. When this ratio is substituted in the equation
Perception of self-tilt in a true and illusory vertical plane 1489
5 Conclusions
(i) Six out of ten supine subjects experienced visually induced self-tilt in a perceptually vertical plane when immersed in a tilted furnished room of which the visual polarity was in a horizontal plane.
(ii) Visually induced self-tilt was similar for erect subjects and supine subjects who experienced the `upright illusion’, with a gain of about 0.5. This does not confirm our hypothesis that gravitational cues counteract visually induced tilt in erect observers. Presumably, the lack of proprioceptive cues is a more important factor.
(iii) Semistatic tilt of the visual scene, where the observer can keep track of visual contours, produced larger illusory self-tilt than previously reported static tilt of a visual scene.
(iv) The combined weighting of the aligned visual and idiotropic frames of reference during simultaneous tilt of subject and room is larger than the weighting of each frame alone.
Asymmetrical representation of body orientation
Michael Barnett-Cowan Heather L. Jenkin
†
Michael R. Jenkin
Laurence R. Harris
The perceived orientation of objects, gravity, and the body are biased to the left. Whether this leftward bias is attributable to biases in sensing or processing vestibular, visual, and body sense cues has never been assessed directly. The orientation in which characters are most easily recognized—the perceived upright (PU)—can be well predicted from a weighted vector sum of these sensory cues. A simple form of this model assumes that the directions of the contributing inputs are coded accurately and as a consequence participants tilted left- or right-side-down relative to gravity should exhibit mirror symmetric patterns of responses. If a left/right asymmetry were present then varying these sensory cues could be used to assess in which sensory modality or modalities a PU bias may have arisen. Participants completed the Oriented Character Recognition Test (OCHART) while manipulating body posture and visual orientation cues relative to gravity. The response patterns showed systematic differences depending on which side they were tilted. An asymmetry of the PU was found to be best modeled by adding a leftward bias of 5.68 to the perceived orientation of the body relative to its actual orientation relative to the head. The asymmetry in the effect of body orientation is reminiscent of the body-defined left-leaning asymmetry in the perceived direction of light coming from above
$
Richard T. Dyde
Centre for Vision Research, Department of Psychology, York University, Toronto, Ontario, Canada
Centre for Vision Research, Department of Psychology, York University, Toronto, Ontario, Canada
Centre for Vision Research, York University, Toronto, Ontario, Canada
Centre for Vision Research, Department of Computer Science and Engineering, York University, Toronto, Ontario, Canada
Centre for Vision Research, Department of Psychology, York University, Toronto, Ontario, Canada
#
#
#
#
and reports that people tend to adopt a right-leaning posture.
Introduction
A reference orientation direction is fundamental to perception and action. Knowing one’s orientation and the orientation of surrounding objects in relation to gravity affects the ability to maintain postural stability (Kluzik, Horak, & Peterka, 2005; Wade & Jones, 1997) as well as the ability to identify (Dyde, Jenkin, & Harris, 2006), predict the behavior of (Barnett-Cowan, Flem- ing,Singh,&Bu ̈lthoff,2011)andinteractwith, surrounding objects (McIntyre, Zago, Berthoz, & Lacquaniti, 2001). Gravity is ideally suited as a reference direction for orientation because it is univer- sally available. However, the senses provide different types of information about the direction of gravity. Thus optimal performance requires integrating orien- tation cues from multiple senses. Several studies in the perceptual literature suggest that the orientation which best enables the brain to reconstruct the three-dimen- sional structure of objects, rather than being aligned with gravity, is biased towards the observer’s left
$
$
$
Citation: Barnett-Cowan, M., Jenkin, H. L., Dyde, R. T., Jenkin, M. R., & Harris, L. R. (2013). Asymmetrical representation of body orientation. Journal of Vision, 13(2):3, 1-9, http://www.journalofvision.org/content/13/2/3, doi:10.1167/13.2.3.
doi: 10.1167/13.2.3 Received August 13, 2012; published February 1, 2013 ISSN 1534-7362 Ó 2013 ARVO Downloaded from jov.arvojournals.org on 07/29/2023
Journal of Vision (2013) 13(2):3, 1-9 Barnett-Cowan et al. 2
(Jenkin, Dyde, Jenkin, Harris, & Howard, 2003; Jenkin, Jenkin, Dyde, & Harris, 2004; Mamassian & Goutcher, 2001; McManus, Buckman, & Woolley, 2004; Metzger, 1975; Sun & Perona, 1998). Howard, Bergstro ̈m, & Ohmi (1990) found that obtaining shape from shading is predominantly performed in a headcentric frame of reference, and while trends of a leftward bias relative to the head were found regardless of head tilt they did not reach significance. Results from Jenkin et al. (2003) and Jenkin et al. (2004) also show a small trend for preferred lighting to be to the left relative to the head when oriented upright (Jenkin et al., 2003: see figure 4a; Jenkin et al., 2004: 5.78 to the left, figure 3) and even when upside down (Jenkin et al., 2003, see figure 4b). This latter observation is crucial because it indicates that the bias is in a constant direction relative to the head.
Leftward biases that are constant with change in head orientation can be found in the spatial perception literature as well. Barnett-Cowan and Harris (2008) reported that the perceived direction of the body is significantly biased leftward by 48-68 when setting a tactile rod or moving the eyes relative to these reference directions and that this bias persists with the body tilted to the left or right by 608. Similar results were also reported by Schuler, Bockisch, Straumann, & Tarnutzer (2010) who found that the leftward bias when setting a tactile rod to vertical was stronger when upside down relative to gravity. The perceptual upright (PU)—the orientation in which characters appear most upright relative to the observer—has also been shown to be tilted around 48 to the left in upright observers (Dyde et al., 2006) and in microgravity environments (Dyde, Jenkin, Jenkin, Zacher, Harris, 2009).
It seems rather odd that the brain should exhibit such a bias given that there is no reason to suspect that space itself and the average environments in which we find ourselves have such biases. Indeed even if space were biased itself then the bias from the perspective of the observer should reverse by simply turning round. Any such bias must arise in the sensory systems or internal representations involved. If, for example, the eyebrows stuck out further to one side, the vestibular organs were not symmetrically arranged on both sides of the head, or a person habitually tilted their head or body to one side, then a leftward bias could be attributed to how the sensory information about orientation was normally received. Alternatively, a bias may arise during central processing. A general attentional bias towards the right side of space has previously been reported (Spence, Shore, & Klein, 2001) and there is considerable evidence that neurologically intact individuals have a tendency to over-represent the left portion of space (see Bowers & Heilman, 1980; Jewell & McCourt, 2000, for reviews). This phenomenon, known as pseudoneglect, is demon-
strated where individuals bisect a line slightly to the left of the line’s true midpoint. Recently, Cattaneo and colleagues (2011) found that in a haptic line bisection paradigm, both sighted and early blind individuals also exhibit a leftward bias, suggesting that the leftward bias is not intrinsic to visual space. Note that a relationship between the leftward bias in pseudoneglect and in spatial tasks is merely speculative as these refer to bias of the left side of space and left tilt respectively. To our knowledge there is no direct evidence suggesting a casual relation.
Here, we set out to determine in which sensory modalities a leftward bias in orientation may arise by measuring the PU in different body orientations while viewing a visual scene rolled by various amounts relative to gravity. This work builds on preliminary results originally presented in poster format (Jenkin et al., 2008). First, we compared the PU with participants oriented left-side-down (LSD), upright and right-side- down (RSD) relative to gravity without visual orien- tation cues to reveal any body based asymmetry. Next, we compared the effect of rolling the visual back- ground to the left or right with the body upright to reveal any possible bias of visual origin. Finally, as the perceived direction of gravity has been shown to change from an over- to an under-estimation through increasing body tilt angles (Mittelstaedt, 1983; Van Beuzekom & Van Gisbergen, 2000), we also measured the PU in multiple body orientations and show that the PU can be well modeled by the linear vector sum model augmented by the addition of a constant leftward bias of the body.
Methods
Participants
Seventeen participants between the ages of 21 and 43 took part in the experiment (10 males, 14 right- handed). All participants were tested with the body upright and when lying left- and right-side-down. Ten of these participants were also tested at oblique body orientations (see below). All participants had normal or corrected-to-normal vision and reported no history of vestibular dysfunction. All participants gave their informed written consent. Experiments conformed to the Ethics Guidelines of York University and the 1964 Declaration of Helsinki.
Convention
The orientation of all stimuli is defined with respect to the body midline of the participant where 08 refers
Downloaded from jov.arvojournals.org on 07/29/2023
Journal of Vision (2013) 13(2):3, 1-9 Barnett-Cowan et al. 3
Figure 1. Apparatus. The visual background (upper left) and the body postures used in this experiment. Participants were tested while lying on their side on a foam mattress for all orientations other than upright with their head supported by foam blocks to ensure that their head was aligned with the long axis of the body. A participant is shown wearing the climbing harness for testing at 1208. Participants viewed the display through a shroud to obscure all peripheral vision (central insert). Viewed through the shroud, the screen subtended a 358 diameter circle at a distance of 25 cm.
to the orientation of the longitudinal axis of the body. Positive orientations are clockwise (‘rightwards’) from this reference point, negative orientations are counter- clockwise (‘leftwards’). Thus the gravity defined ”up” is at þ908 when the participant is lying on their left side. The p symbol used in the OCHART test (see below) is described as being 08 when the vertical shaft of the symbol is aligned with the body axis with the letter bowl to the right (i.e., the symbol appears as an upright p).
Positioning of the body
Participants stood upright or lay on a foam mattress on their right or left side with their head supported by foam blocks to ensure that their head was comfortably aligned with the body. The bed was oriented at the following tilt angles (in roll relative to gravity): 08 (participant upright), 308, 458, 608, 908, and 1208 right- side-down (RSD), and 908 left-side-down (LSD) (Figure 1). For the 1208 body orientation, participants wore a climbing harness while lying on their side, were secured to the top of the bed frame using climbing rope and were then suspended by tilting the bed and frame into the desired orientation (see Figure 1). Participants remained in this partially upside down position for no longer than 35 minutes and were permitted to take breaks upon request.
Stimulus presentation
Participants observed images presented in their fronto-parallel plane on an Apple iBook laptop computer (Apple, Inc., Cupertino, CA) with a reso- lution of 48 pixels/cm (21 pixels/8). The screen was masked to a circle subtending 358 when viewed at 25 cm by a black circular shrouding tube that obscured all peripheral vision (Figure 1). The laptop was mounted in an aluminum frame and fixed to the bed to maintain the screen at a fixed orientation relative to the observer (the top of the screen corresponded to the top of the participant’s head) and to hold the shroud. Participants responded using two buttons on a game-pad (Gravis Game Pad Pro, Kensington/Gravis, Redwood Shores, CA).
Test for perceptual upright (OCHART probe)
The Oriented CHAracter Recognition Test (OCHART) (Dyde et al., 2006) is an indirect measure of the PU. The OCHART has the observer discriminate between the letter p and the letter d. As the letter p when rotated 1808 becomes the letter d, the transitions from p-to-d and d-to-p when averaged define the PU. The beauty of this technique is that observers are not asked to make judgments with respect to any frame of reference, but rather only identify the character. Using the method of constant stimuli, the probe character, a
Downloaded from jov.arvojournals.org on 07/29/2023
Journal of Vision (2013) 13(2):3, 1-9 Barnett-Cowan et al. 4
letter p (3.18 · 1.98), was presented rotated around its geometric centre in one of 24 static orientations 08-3458 in 158 increments. The probe character was presented for 500 ms either on a background picture rich in polarized cues that could be presented in one of 18 orientations 08-337.58 in 22.58 increments (Figure 1), or against a neutral grey background. After 500 ms, stimuli were replaced with a screen of the same mean luminance as the visually polarized image with a central fixation point (0.458 diameter). Thus for each body orientation there were 456 (24 · 19) character/back- ground combinations which were each presented eight times in a random order resulting in a total of 3,648 (456 · 8) presentations for each body orientation. Testing was divided into two sessions per body orientation (completed in two blocks of 1,824 trials each) which each took approximately 35 minutes to complete. The presentation of stimuli was randomized within each block. No feedback as to participant performance was given.
Analysis
The percentage of presentations that participants identified the character as a p was plotted as a function of the character’s orientation for a given background. Two sigmoidal cumulative Gaussian functions (Equa- tion 1) were fitted to the participants’ response rate to determine each of the p-to-d and d-to-p transitions for each visual background in each body orientation.
Figure 2. The effect of body orientation on the perceptual upright (PU). The average PU is shown relative to the body measured against a gray background for participants oriented LSD, UP, and RSD relative to gravity. RSD data are shown as inverted (1[RSD]) to better illustrate the asymmetry of the perceptual upright with body tilt. The direction of gravity relative to the body for each body orientation is indicated by the thick horizontal black lines. *: p , 0.05, **: p , 0.01, ***: p , 0.001.
were significantly different from each other at the p , 0.01 level (see Figure 2). With the body upright, the PU was significantly biased to the left (one-sample t-test:
3.08 SE 1.08; t(16) 1⁄4 2.9, p 1⁄4 0.010). In order to test the bias between LSD and RSD, we compared LSD results against the negative of the RSD results for the grey background. A repeated-measures ANOVA shows a significant effect of body orientation, F(1,16) 1⁄4 23.88, p , 0.05. By comparing the LSD PU with the RSD PU, a significant leftward bias of the PU of 9.58 ([þ7.78 26.78]/2 1⁄4 9.58) is evident.
Visual scene
In order to assess whether there was a similar bias in the internal representation of the direction of visual cues to upright, we compared the effect of tilting the background left and right. To do so we compared the effect of a scene rotation of given angle (e.g., 22.58; to the right) minus the effect for a scene rotated in the opposite direction (e.g., 337.58; 22.58 to the left). A response bias from responses at 22.58, 458, 67.58, 908, 112.58, 1358, and 157.58 from upright was computed for each observer as their response to a scene rotated clockwise minus their response to the same display rotated counterclockwise. A repeated-measures AN- OVA was run on 17 observers on this display bias. Mauchly’s test for sphericity indicated that the assumption of sphericity was violated, v2(20) 1⁄4 204.2, p , 0.05. Therefore degrees of freedom were corrected
y1⁄4 100 % 1 þ exx0
r
ð1Þ
where: x0 corresponds to the 50% point and r is the standard deviation. The average of the orientations at which these two transitions occurred was taken as the PU.
Results
Body-based bias
A repeated-measures ANOVA was run on the gray background dataset. There were three body orienta- tions (upright, RSD, LSD) and 17 participants. Mauchly’s test for sphericity indicated that the assumption of sphericity was violated (v2(2) 1⁄4 11.79, p ,0.05). Therefore degrees of freedom were corrected using Greeenhouse-Geisser estimates of sphericity (E 1⁄4 0.647). The results show a significant effect of body orientation on the PU, F(1.3, 20.7) 1⁄4 37.78, p , 0.05. Post hoc tests revealed that all three body orientations
Downloaded from jov.arvojournals.org on 07/29/2023
Journal of Vision (2013) 13(2):3, 1-9 Barnett-Cowan et al. 5
Figure 3. Average and predicted PU as a function of visual background orientation for different body orientations. The data are plotted relative to the body (a) and relative to gravity (b). Negative values are to the left (counterclockwise). Solid curves are the vector model PU predictions (R2: 0.90). Error bars represent 61 SEM.
using Greenhouse-Geisser estimates of sphericity (E 1⁄4 0.203). No significant result was found, F(1.2, 19.5) 1⁄4 3.695, p 1⁄4 0.62. Importantly, when the scene was upright, no bias of the PU was observed (one-sample t- test: 0.98 SE 1.98; t(16) 1⁄4 0.48, p 1⁄4 0.635). These results indicate that a bias of the PU does not reside in the interpretation of visual orientation cues.
The effect of body posture on visual and body biases
In order to see if the leftward bias was present in multiple body tilts, we measured the PU for a range of body and visual combinations. A sample (N 1⁄4 10) of the original participants was tested in four additional body postures (308, 458, 608, and 1208). The full dataset for these participants was fit using a modified version of the model of Dyde et al. (2006) in which the PU is modeled as the simple linear sum of three vectors corresponding to the visual, body and gravity cues. We augmented the Dyde et al.’s model through the addition of a bias term to the visual and body vectors (see Equation 2).
~
PU1⁄4~vhv þbhb þ~g ð2Þ
where each vector is in the veridical direction relative to the body rotated by a bias term (hv, hb) in the roll plane
and the relative length of each vector indicates the
extent to which the PU is influenced by each factor. We
fitted the weighted vector model (Equation 2) to our
average PU data shown in Figure 3. The Marquardt-
Levenberg optimization algorithm technique was used
(Press, 1988) where ~v (vision) and b~(body) are vectors
of variable lengths in the orientations of the visual cues
and body respectively and ~g (gravity) is a unity vector
in the direction of gravity. The fit was also permitted to
~ add a rotational bias to both ~v and b.
The data were well predicted by the model (R2 1⁄4 0.90). The best fit weights for the magnitude of each
~
vector were ~v: 2.0 (SE 0.14), b: 4.5 (SE 3.1), relative to ~g
which was fixed to 1.0. A leftward bias of 5.68 which was significantly different from 08 (SE 0.7; p , 0.0001) had to be added to the body vector to obtain a best fit as predicted from the results in section 3.2. No significant bias was required to be added to vision (0.228 SE 2.0; p 1⁄4 0.85). Table 1 lists parameter estimates as magnitude ratios relative to gravity as well as percentages. The model fit to the data is drawn through the data plotted as a function of visual tilt for each body orientation with respect to the body (Figure 3a) and gravity (Figure 3b).
It has previously been reported that tilt of the body leads to increased influence of visual cues for estimates of the subjective visual vertical (SVV; Mittelstaedt, 1986) and also increased roll vection (Dichgans & Brandt, 1978). To assess whether body tilt in the
Downloaded from jov.arvojournals.org on 07/29/2023
Journal of Vision (2013) 13(2):3, 1-9
Barnett-Cowan et al. 6
Cue weight
2.0 4.5 1.0
Bias
0.28 5.68 R2 0.90
SE p
0.14 ,0.001 0.31 ,0.001
Relative cue weight
26.7% 60.0% 13.3%
Vision (~v) Body (~b) Gravity (~g)
SE
2.08
0.78 ,0.001
p
hv hb
0.895
Table 1. Output of the augmented vector sum model (with biases). Parameter estimates and R2 of the vector sum model for vision, body and gravity cues as well as bias terms for vision and the body.
present experiment increases the effect of visual cues on the PU, we calculated the difference between the minimum and maximum points for each participant’s PU estimates in each body orientation. We call this the ”visual effect.” We did not find a significant effect of body orientation on the visual effect (Greenhouse Geisser, F(2.002, 18.014) 1⁄4 0.8 , p 1⁄4 0.467) and the average visual effect was similar to the maximum visual effect predicted from the model (see Figure 4).
Discussion
We have identified a persistent leftward bias of the internal representation of the body’s orientation relative to the body’s true orientation. In order to accurately predict the perceptual upright from the orientations of vision, body and gravity at all body postures it is necessary to add a constant bias of approximately 5.68 to the left of the actual orientation of the body. It is possible that the representation of the body may also be tilted in the pitch plane relative to its true direction but we were unable to measure such a bias with the present protocol. The magnitude of the leftward bias is consistent with previous studies investigating the perceptual upright (Dyde et al., 2006), shape-from-shading (Jenkin et al., 2003; Jenkin et al., 2004; McManus et al., 2004), as well as the perceived directions of gravity and the body (Barnett-Cowan & Harris, 2008) that have all suggested a bias of the PU to the left (relative to the body). This leftward bias of the PU cannot be attributed to sensory information as the vision bias was not significant and the effect of information present in the visual scene on the PU relative to an upright observer was symmetrical (see also Haji-Khamneh & Harris, 2010).
Why should the brain have such a bias in its representation of the body? One possibility is that a leftward bias of the PU arises from asymmetries in brain function. It has been shown that the posterior
Figure 4. Average visual effect as a function of body orientation compared with model predictions. Error bars represent 61 SEM.
parietal and lateral frontal premotor regions are activated in the right hemisphere more than in the left hemisphere when estimating the midsaggital plane (Vallar et al., 1999). This may underlie the observation that visuo-spatial attention is preferentially directed toward the left (Corbetta, Shulman, Miezen, & Peterson, 1995; Niemeier, Stojanoski, Singh, & Chu, 2008; Posner & Rothbart, 2007). Further, unilateral damage to these brain regions—particularly in the right hemisphere—can result in spatial hemi-neglect (Vallar, 1998). Indeed, it has been shown that this hemispheric asymmetry manifests itself as pseudoneglect, where even normal participants tend to bisect a horizontal line with a leftward bias (see Jewell & McCourt, 2000 for a review). This bias persists when assessed in a supine orientation suggesting that the leftward bias is based predominantly in body coordinates (Nicholls, Smith, Mattingley, & Bradshaw, 2006). As these regions are implicated in contributing to the computation of an egocentric frame of reference (Anderson, Snyder, Bradley, & Xing, 1997), differential activity here between the hemispheres may very well be related to the leftward bias of the perceptual upright that we report.
An alternative hypothesis which could explain our results is that typically people tend to adopt a posture with their heads tilted around 108 to the right during much of everyday life (Greenberg, 1960; Previc, 1991, 1994; Putnam, Noonan, Bellia, & Previc, 1996). Indeed a quick sampling of your family photographs will likely confirm this typical head tilt. Thus, when forced to adopt an accurately upright posture in an experimental setting they perceive their head as tilted in the opposite direction from its ”natural” orientation (i.e., to the right). It has been suggested that this natural rightward head tilt results from asymmetries of the body such as leg length where the left leg tends to be slightly longer (Ludwig, 1970; Peters, 1988) as well as an asymmetry of the vestibular system such that people tend to lean towards the side of the weaker (typically right) otolith organ (Curthoys & Halmagyi, 1995; Gresty, Bronstein,
Downloaded from jov.arvojournals.org on 07/29/2023
Journal of Vision (2013) 13(2):3, 1-9 Barnett-Cowan et al. 7
Brandt, & Dieterich, 1992; Previc, 1991). Straightening of the head under laboratory conditions therefore would result in a leftward bias of perceived upright which otherwise would be veridical relative to gravity when adopting a natural, right-tilting head posture. We suggest that this explanation may also contribute to other leftward biases that have been reported in the shape-from-shading literature where it has alternatively been argued that a leftward bias for preferred lighting arises from preferentially arranging lighting sources to the left to facilitate right-handed writing (Metzger, 1975). Explanations based on brain asymmetries and on head tilt are not mutually exclusive as brain asymmetries may of course contribute to the natural right-tilting posture.
The PU was well predicted across multiple physical and visual tilts as arising from a weighted vector sum of the internal representations of the orientation of the body (60%) by vision (27%) and gravity (13%). This corresponds with the letter p appearing most like the letter p when it is oriented somewhere between the body and gravity. This is easily appreciated by noting the orientation in which you typically hold a book when reading it on your side. These sensory weightings are similar to those previously reported by Dyde and colleagues (2006; body: 54%, vision: 25%, gravity: 21%). Note that the weighted vector sum model of Dyde and colleagues (2006) is a simplification of the weighted vector model of Mittelstaedt (1983). Mittel- staedt’s model includes factors to describe adjustments that model torsional eye movements associated with rolled body and visual displays relative to gravity. While we did not model or measure ocular counter-roll, the bias term identified here is constantly to the left and torsional eye movements are leftward when right-tilted but rightward when left-tilted, the constant leftward bias term here can be applied to Mittelstaedt’s model as well and may help to reduce errors attributed to modeling ocular torsion.
This relative weighting of multisensory cues for the PU likely explains why other measures of perceived orientation such as the subjective visual vertical (SVV) do not yield similar leftward biases. Dyde et al. (2006) established that the PU and the SVV are distinct, where the SVV is primarily driven by gravity whereas the PU is primarily driven by body. Since the SVV is primarily driven by gravity, and not as much by body (where we show the bias to reside), one would not expect to see an asymmetry using the SVV measure since it is not as sensitive as the PU measure for the contribution of the body. The relative contribution of vision : body : gravityfortheSVVis0.1:0.2:1.0but1.2:2.6:1.0for thePU(Dydeetal.,2006)or2.0:4.5:1.0forthePUin the present study. It is thus not surprising that a leftward bias of approximately 5.68 is not typically reported using the SVV measure since the body
contributes only 15% of the SVV as opposed to 60% for the PU as shown here. The PU is therefore four times more sensitive to the body; by extension we would only expect a bias of approximately 1.48 for the SVV (5.6/4 1⁄4 1.4). Note that it is surprising that leftward biases have not been reported for subjective horizontal (Mittelstaedt, 1983) or vertical (Bisdorff, Wolsley, Anastasopoulos, Bronstein, & Gresty, 1996) body position, and therefore our results also confirm the marked dissociation between perceived posture and spatial perception tasks as previously noted by these authors.
Another distinction between the SVV and PU was also found when analyzing change in the visual effect as a function of body tilt. Mittelstaedt (1986) observed large changes in the influence of the visual field on the SVV when the body was tilted, which he resolved by implementing a gating mechanism to explain the interaction of visual and vestibular information. Lack of significant modulation of the PU visual effect in the present study points to further dissociation between the neural mechanisms, which govern the SVV and PU.
Conclusions
The OCHART has been used as a means of quantifying whether changes in the relative weightings of sensory cues occurs in microgravity (Dyde et al., 2009; Harris, Jenkin, Jenkin, Zacher, & Dyde, 2011; Jenkin, Dyde, Jenkin, Zacher, & Harris, 2011), between sexes (Barnett-Cowan, Dyde, Thompson, & Harris, 2010b) and in Parkinson’s patients with and without medication (Barnett-Cowan, Dyde, Fox, Hutchison, & Harris, 2010a). The ability of the OCHART to identify a robust leftward bias in the perceptual upright could be of practical importance in testing clinical popula- tions with asymmetrical symptoms such as those suffering from stroke, neglect, muscular dystrophy and pusher syndrome.
Keywords: body sense, gravity, leftward bias, percep- tual upright, multisensory integration
Acknowledgments
This work was supported by Natural Sciences and Engineering Research Council of Canada (NSERC) operating grants to LRH and MRJ and with the generous support of the Canadian Space Agency. MB-C was supported by a PGS-D3 NSERC Scholar- ship and a Canadian Institutes of Health Research Vision Health Science Training Grant. Our thanks to
Downloaded from jov.arvojournals.org on 07/29/2023
Journal of Vision (2013) 13(2):3, 1-9 Barnett-Cowan et al. 8
Jeff Sanderson, Elena Mazour, and Aisha Islam who helped conduct experiments.
