Gaze during visually-guided locomotion in...
Transcript of Gaze during visually-guided locomotion in...
Research report
Gaze during visually-guided locomotion in cats
Garth A. Fowler *, Helen Sherk
Department of Biological Structure, University of Washington, Seattle, WA 98195, USA
Received 2 January 2002; received in revised form 25 March 2002; accepted 25 March 2002
Abstract
Visual guidance is often critical during locomotion. To understand how the visual system performs this function it is necessary to
know what pattern of retinal image motion neurons experience. If a locomoting observer maintains an angle of gaze that is constant
relative to his body, retinal image motion will resemble Gibson’s (The Perception of the Visual World (1950)) well-known optic flow
field. However, if a moving observer fixates and tracks a stationary feature of the environment, or shifts his gaze, retinal motion will
be quite different. We have investigated gaze in cats during visually-guided locomotion. Because cats generally maintain their eyes
centered in the orbits, their gaze can be monitored with reasonable accuracy by monitoring head position. Using a digital
videocamera, we recorded head position in cats as they walked down a cluttered alley. For much of the time, cats maintained a
downward angle of gaze that was constant relative to their body coordinates; these episodes averaged 240 ms in duration and
occupied 48�/71% of the total trial time. Constant gaze episodes were separated by gaze shifts, which often coincided with blinks.
Only rarely did we observe instances when cats appeared to fixate and track stationary features of the alley. We hypothesize that
walking cats acquire visual information primarily during episodes of constant gaze, when retinal image motion resembles Gibson’s
conventional optic flow field.
# 2002 Elsevier Science B.V. All rights reserved.
Keywords: Cat; Vision; Gaze; Locomotion; Visually-guided; Optic flow; Retinal image motion
1. Introduction
How does the visual system use information for
guidance during locomotion? To answer this question,
one needs to describe the pattern of retinal image
motion that occurs during locomotion. Helmholtz [13]
and subsequent investigators (e.g. Gibson, [9]; Koender-
ink, [15]) have described the image motion that is
generated by an observer’s motion through the environ-
ment: elements move approximately radially outward
from the heading point, expanding and accelerating as
they go (see Fig. 1A, the cross indicates the heading
point). However, the pattern of retinal image motion
may be quite different because of various gaze strategies
employed by the observer [23,26]. Fig. 1B shows the
motion of images seen by a walking observer who fixates
and tracks a stationary feature of the environment to the
left of his heading point. If the same observer instead
shifts his gaze to the left while walking during the gaze
shift he will experience a pattern of retinal image motion
like that in Fig. 1C.
In order to understand visual processing during
locomotion, we need to know what pattern of motion
neurons actually see. This is a challenging proposition
because it requires gaze to be monitored in a locomoting
subject. In stationary subjects, Robinson’s [25] magnetic
coil method for measuring eye position has been highly
successful, and has been refined and extended to
assessment of head position by Collewijn et al. [2,3].
But this method does not work on subjects who
locomote from one point to another, because they
quickly move out of the magnetic field. Solomon and
Cohen [27] have applied the magnetic coil method to
tethered monkeys that are walking or running in a circle,
but this paradigm cannot tell us about gaze during
visually-guided locomotion.
We have taken another approach to the problem of
gaze measurement in locomoting cats. Although it seems
natural to monitor gaze by measuring eye position, in
* Corresponding author. Tel.: �/1-206-543-1861; fax: �/1-206-543-
1524.
E-mail address: [email protected] (G.A. Fowler).
Behavioural Brain Research 139 (2003) 83�/96
www.elsevier.com/locate/bbr
0166-4328/02/$ - see front matter # 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 6 - 4 3 2 8 ( 0 2 ) 0 0 0 9 6 - 7
the cat one can assess gaze with moderate accuracy by
looking at head position only. Cats can deviate their
eyes up to �/258 in the orbits [31,32], but such
deviations are brief. Guitton et al. [10] found that,
except during gaze shifts, cats maintain their eyes within
28 of the center of the orbits. They also noted that even
small gaze shifts include movement of the head as well
as the eyes. With the advent of digital VCs, it has
become possible to record head position in a locomoting
cat without attaching any sensors to the animal, and this
is the approach that we have taken. In this paper, we will
use the term gaze when the cat’s gaze was inferred from
static head position. We will avoid this usage when the
cat’s head was moving, and the linkage between gaze
and head position is less certain.
During locomotion, the importance of vision may
vary depending upon the environment. Walking across
an empty room or across an open field presents almost
no visual challenge, but picking one’s way across a
cluttered floor or up a rocky mountain trail requires
much more continuous visual attention. It is the latter
situation that we wished to simulate, and so we have
monitored gaze in cats that were walking down an alley
that was densely cluttered with small objects (Fig. 2).
We found that cats performing this task spent most of
their time with their gaze fixed relative to their bodies
(or heading points), so that the pattern of retinal image
motion resembled the relatively simple optic flow field
of Fig. 1A. But these episodes of constant gaze were
usually rather short, with cats making frequent gaze
Fig. 1. Patterns of retinal optic flow generated by locomotion along a
straight path. The simulated cat had an eye height of 18 cm and moved
at 80 cm/s, heading towards the cross. Vectors show the motion of
objects lying on a ground plane that extends 904 cm in front of the cat,
with each black spot indicating an object’s position after 67 ms of
motion. (A) Optic flow seen when the cat maintained a constant angle
of gaze relative to its heading. (B) Optic flow seen when the cat fixated
and tracked a stationary object 28 cm to the left of its path, and 70 cm
ahead (square). (C) Optic flow seen when the cat shifted its gaze to the
left at 508/s.
Fig. 2. Test alley, viewed from outside the exit end. Black objects can
be seen scattered across the alley floor. The cat’s speed was monitored
with photosensors, P1, P2, and P3, and gaze was monitored with a
digital VC directed down the first leg of the alley.
G.A. Fowler, H. Sherk / Behavioural Brain Research 139 (2003) 83�/9684
shifts. Their directions of gaze clustered near their mid-
sagittal planes and were angled downward. Only rarely
did walking cats fixate and track stationary features.
Some of these data have been published in abstractform [8].
2. Methods
2.1. General procedure
Six adult cats, five female and one male, were used in
this study. All methods were approved by the AnimalCare Committee of the University of Washington, and
conformed to NIH guidelines.
Cats were trained to walk down an L-shaped alley
whose floor was cluttered with small objects, typically
there were 44 objects in the first leg (Fig. 2). These were
irregularly shaped pieces of cardboard, about 3�/7 cm2,
on which were glued slices of PCV tubing, 0.8 cm in
height. With practice, cats learned to avoid stepping onthese objects. Because the locations of objects varied
randomly from trial to trial, cats relied on vision for
accurate foot placement [29]. Cats’ accuracy was
assessed by looking at the impressions left by their feet
in the sand that covered the alley floor. When the cat
stepped on an object, it indented the sand, or sometimes
was shifted out of position. In our analysis of gaze
during locomotion, we considered only trials in whichthe cat made 0 or 1 error in the first leg of the alley (the
great majority of trials analyzed had no errors). Since by
chance we would expect 4�/5 errors in this leg of the alley
[29], an error-free run indicated that the cat’s attention
was focused on the task. Visual distractions inside the
alley were minimal. The alley had blank white walls,
interrupted by fixed features that included vertical
struts, a fluorescent light, photosensors and bicyclereflectors. Practiced cats paid not evident attention to
these features.
Cats were initially taught to traverse the alley by
feeding them with canned tuna fish. One animal was
rewarded in this fashion on all subsequent trials, but the
other cats were thereafter rewarded only with petting.
No cat was food-deprived. Cats were free to move at
their own pace, and were rewarded regardless of thenumber of errors made. However, all cats became
increasingly skilled at avoiding stepping on objects,
suggesting that this is a natural behavior [29]. The
amount of practice required to reach a stable level of
performance varied from cat to cat, ranging from 30 to
190 days.
To assess gaze, we videotaped the cat as it walked
down the first leg of the alley, directly toward thevideocamera (VC in Fig. 2). For the first two cats
studied (Brie and Cheddar), we used a digital VC with a
frame rate of 30 Hz and a resolution of 720�/480 pixels,
interlaced (Canon ZR). For the other four cats, a similar
VC (Canon Elura) was used; it was capable of non-
interlaced image capture, thus doubling the effective
vertical resolution. The image quality obtained in thesevideotapes is illustrated in Fig. 3 and Fig. 6.
Out of a large number of trials (472 total), we chose
for analysis 10�/18/cat, based primarily on image
quality. The VC’s automatic zoom did not always
maintain the image in sharp focus, and we selected
only videotapes that showed excellent focus throughout.
However, the chief limitation on the size of the data set
was the amount of time required for the analysis.Videotaped images were downloaded to a computer
using PhotoDV (Digital Origin) and analyzed frame by
frame using Adobe Photoshop. In addition, movies were
reconstructed from captured frames using QuickTime
(Apple Computer, Inc.), and these movies were viewed
in slow motion or frame by frame to look for rapid eye
movements. We could not accurately measure eye
position but we could detect saccades. In the gazecalibration procedure described in the next section, in
which cats made gaze changes of known size, we found
that we could readily detect saccades as small as 3.58.Although we could not measure the size of saccades in
locomoting cats, we could use them to determine the
beginning or end of some episodes of constant gaze.
2.2. Measurement of head azimuth
We measured head azimuth and elevation, usually in
alternate frames, but sometimes in every frame when the
head was turning rapidly. Azimuth was found by taking
the ratio of the distances illustrated in Fig. 3A, that is,
from the temporal corner of the left eye to the left edge
of the head, and from the corner of the right eye to the
right edge of the head. To determine head azimuth fromthis ratio, we used a look-up table that was compiled
from measurements made in videotaped frames of
stationary cats that were fixating at known locations.
In this procedure, the cat sat in one investigator’s lap
with its head directly in line with the VC. In front of the
cat was a horizontal board with small holes at positions
corresponding to particular locations in the cat’s body
coordinate system (for example, 108 to the left and 108inferior). The cat fixated on the tip of a pipecleaner that
emerged, wiggling, from one hole. We found that cats
fixated repeatedly and reliably in this situation. Fifteen
different locations, 58 apart, were tested. Two cats were
used (Jack and Kraft), and we videotaped 256 fixations.
Repeated fixations at a given location yielded very
similar images (e.g. Fig. 3B), and values found for the
two cats were also quite similar. Close to the mid-sagittal plane, gaze was measured to within 9/18 of the
same value on repeated trials, while at 208 of azimuth,
repeatability was 9/28.
G.A. Fowler, H. Sherk / Behavioural Brain Research 139 (2003) 83�/96 85
2.3. Measurement of head elevation
Head elevation was measured by taking a ratio of two
distances: muzzle length (M ) (from the lower edge of the
eye to the point of the nose), and chin length (C ) (from
the point of the nose to the bottom of the chin) (Fig.
3C). For every cat we found the ratio C /(C�/M) that
corresponded to a horizontal gaze by videotaping the
cat from the side while it fixated an object that was at
eye height (Fig. 3D). This image was imported into
Adobe Illustrator and rotated in 58 increments, and M
and C were measured at each position. A linear
equation was found using these values that related
head elevation to the ratio C /(C�/M ). Equations varied
somewhat from one cat to another because of individual
variation in the lengths of muzzle and chin.
2.4. Predicted retinal image motion
When the walking cat’s gaze was constant relative to
its body, as occurred for much of the time during each
trial, we could determine the speed and trajectory of
images that corresponded to objects on the alley floor as
they passed through the center of the cat’s visual field.
The variables required for this calculation are (1) thewalking speed of the cat (measured using the photo-
sensors shown in Fig. 2); (2) the height of the cat’s eyes
above the alley floor (measured from the videotaped
Fig. 3. Measurement of head azimuth and elevation. (A) Azimuth was found by taking a ratio of the horizontal distances from the outer margin of
each eye to the edge of the head. (B) Four different fixations by one cat at a location with azimuth�/158 and elevation�/�/158 in the cat’s body
coordinates. (C) Front view showing the distances M and C as they were measured for the calculation of elevation. (D) Side view of cat fixating on a
point at eye level, also showing M and C .
G.A. Fowler, H. Sherk / Behavioural Brain Research 139 (2003) 83�/9686
image after taking into account the cat’s distance from
the camera); (3) the angle of gaze relative to the cat’s
heading direction (head azimuth and elevation). These
data were entered into a program that computed thespeed, azimuth, and elevation, in the cat’s visual field, of
an image passing through the center of gaze. These
variables were of interest for determining the expected
trajectory of the cat’s gaze if, instead of maintaining a
constant gaze angle, it tracked the image.
3. Results
All cats showed broadly similar patterns of gaze asthey walked down the alley. The examples in Fig. 4
illustrate the most common features of gaze that we
observed. (1) During every trial, there were several
intervals during which head azimuth was constant
(defined as a variation of 38 or less throughout the
interval). In Fig. 4, azimuth during these constant gaze
episodes (CGEs) is plotted using open circles. (2) CGEs
were separated from each other by either a gaze shift, a
blink (vertical lines indicate frames in which the cat’s
eyes were closed), or both. (3) Some gaze shifts
culminated in a brief glance to one side or the other,
followed by an immediate shift back towards the midline
(see arrow in Fig. 4D). (4) Head elevation was often
more variable than azimuth, and commonly oscillated in
time with the cat’s footfalls. At such low frequencies (�/
2�/3 Hz), the cat’s VOR is capable of fully compensating
for vertical head oscillation [5,6], and thus it seems likely
that counter-rotation of the eyes compensated for these
movements.
CGEs occupied the greatest amount of time, on
average 48�/71% of each trial (Fig. 5). Cats spent the
remainder of the time either shifting their gaze or closing
Fig. 4. Head azimuth and elevation, in degrees, during 4 trials by 4 different cats. When gaze was constant (that is, with azimuth maintained within
1.58 of the mean for the CGE), data points are marked with open circles. During gaze shifts, data points are marked with filled circles. In the azimuth
plots, points above 0 indicate head rotation to the left, and points below 0 indicate head rotation to the right. Cats turned their heads to the right at
the end of each record because they anticipated the right turn into the second leg of the alley. Vertical lines mark videotape frames in which the cat’s
eyes were closed (blinks).
G.A. Fowler, H. Sherk / Behavioural Brain Research 139 (2003) 83�/96 87
their eyes. Although we had thought that they might
sometimes fixate and track stationary objects in the
alley, we rarely found evidence of such behavior (seebelow).
3.1. Constant gaze episodes
Every trial included several CGEs, typically about 5/
trial in the first leg of the alley, giving an averagefrequency of 2.5/s (Table 1). Fig. 6 illustrates the
constancy of gaze during 3 such episodes; in each case,
head azimuths were within 18 of each other in the first
and last frames of the CGE. CGE duration varied
considerably, from 67 ms (the briefest CGE that we
could identify) to 1000 ms, and averaged 247 ms (Fig. 7).
There was some cat-to-cat variability: Brie and Jack, for
example, tended to have prolonged CGEs.It is possible that we under-estimated the duration of
some CGEs because we over-estimated the duration of
the preceding gaze shift. If the cat combined a head
movement with a saccade, the saccade might bring gaze
to its final position before the head had finished moving,
resulting in an interval when the head continued to
move but the gaze was fixed because the eyes counter-
rotated [10,11]. However, this potential problem doesnot arise for a CGE that immediately follows a blink,
since the CGE cannot start until the cat opens its eyes.
We thus assumed that the durations of CGEs following
blinks were accurately measured, and compared them to
the durations of other CGEs. There was no difference in
the durations of CGEs that followed blinks compared to
those that did not follow blinks, suggesting that we had
not substantially under-estimated CGE duration.
During CGEs, the cat’s gaze was almost always
directed downward, and tended to remain close to its
midline. Fig. 8 shows gaze during CGEs in a coordinate
system that was centered on the cat’s heading point.
Open circles show episodes in which the center of gaze
would have fallen on the alley floor, and filled circles
show episodes in which gaze would have fallen on the
alley walls. All cats looked at the floor most of the time,
with a gaze angle on average 198 below the horizon
(Table 1). We speculated that CGEs might be longer
when gaze was close to the cat’s anticipated path than
when gaze was eccentric, but found only weak support
for this hypothesis. For one cat (Kraft) there was a slight
but significant tendency for more eccentric CGEs to be
briefer, but for the other cats duration and azimuth were
uncorrelated.
One might ask where walking cats typically looked in
physical space*/how far ahead, and how close to their
intended path. These locations are shown in Fig. 9 for
all CGEs in which gaze fell on the alley floor. Cats never
looked directly downward, but instead looked at points
on average 58 cm ahead (see ‘distance ahead’ in Table
1). Given a stride length of �/19 cm, cats were thus
Fig. 5. Time spent by each cat in different gaze behaviors while walking down the alley. CGEs dominated the behavior of all cats. The number of
trials analyzed for each cat is given as n .
Table 1
Cat CGEs/s Average CGE azimuth (8) Average CGE elevation (8) Average distance ahead (cm)
Brie 2.16 3.5 right �17.4 69.7
Cheddar 3.08 0.8 right �22.3 50.0
Gouda 2.60 1.8 left �16.9 69.0
Jack 2.28 3.5 right �20.6 51.0
Kraft 2.57 1.9 right �17.2 62.0
Leaf 2.25 3.7 left �22.0 47.0
Average 2.49 0.8 right �19.4 58.12
G.A. Fowler, H. Sherk / Behavioural Brain Research 139 (2003) 83�/9688
looking at points about 3 steps beyond their current foot
placement. Gaze strategies varied somewhat between
cats. Jack and Leaf focused their gaze over a rather
narrow range of distances ahead, so that their standard
deviations were only 9 cm, while Gouda and Kraft had
standard deviations that were twice as great.
If gaze is constant relative to the body, as during a
CGE, images move through the visual field in the optic
flow pattern shown in Fig. 1A. It was thus possible to
calculate the speed of images passing through the area
centralis during CGEs, and these speeds are shown for
each cat in Fig. 10. The most critical determinant of
image speed was the cat’s angle of gaze*/cats that
tended to look steeply downward experienced relatively
rapid image motion, and cats that maintained a less
steep angle of gaze experienced slower motion. Image
speeds were generally modest, averaging 158/s and rarely
exceeding 258/s for most cats.
3.2. Gaze changes
Between CGEs cats usually shifted their gaze. Gaze
shifts were moderate in size, with the vast majority being
less than 108 (Fig. 11A). Cats sometimes made a brief
but large gaze shift, usually to the left or right, which we
refer to as a glance. Glances were generally substantially
larger than gaze shifts between CGEs (compare histo-
gram in Fig. 11B with those in Fig. 11A). Fig. 4D shows
a 208 glance to the left. Cats varied in the frequency of
glances, from 0.9/trial (Leaf) to 0/trial (Cheddar) on
average.
Fig. 6. The first and last frames of typical CGEs for 3 different cats.
The time shown is that elapsed between the two frames. In each case,
the difference in head azimuth between the two frames was no greater
than 18.
Fig. 7. Durations of CGEs for each cat. Arrowheads indicate average
values.
G.A. Fowler, H. Sherk / Behavioural Brain Research 139 (2003) 83�/96 89
3.3. Fixation and tracking
Do locomoting cats fixate and track static features of
the environment? If a cat fixated a feature while walking
down the alley, its gaze would rotate smoothly relative
to its own body (a behavior that we shall refer to as
‘tracking’). Tracking might have occurred during either
presumptive CGEs or during gaze shifts. During brief
CGEs we could not rule out this possibility since the
magnitude of gaze rotation that would occur during
tracking would be small. During longer CGEs, however,
the head rotation predicted during tracking would be
large enough to observe. We therefore identified all
CGEs that were long enough that gaze would shift by 58or more if the cat started tracking at the beginning of the
CGE, and for which tracking speed would be at least
108/sec, which is beyond the limit for smooth eye pursuit
by the cat (see Section 4). For each of 63 such CGEs we
calculated the head trajectory that would be predicted if
the cat were tracking rather than maintaining a constant
gaze (see Section 2). The match between predicted and
actual gaze trajectories is moderately good in Fig. 12A
and B, and somewhat less so in Fig. 12C. The remaining
plots (Fig. 12D�/L) are typical of the other 60 CGEs,
none of which showed a match. Thus fixation and
tracking during presumptive CGEs appeared to be rare.
Tracking might be more likely to occur during gaze
shifts than during CGEs. Although we observed a large
number of gaze shifts, blinks coincided with many (see
below), effectively precluding tracking. There were 223
gaze changes that were free of blinks and long enough to
compare to trajectories predicted by a tracking hypoth-
esis. In 179 cases the cat’s head moved in a direction
opposite to that predicted. In the remaining instances,
the head moved in the predicted direction, but not at the
predicted rate (usually it moved too fast). Thus none of
the gaze changes that we observed could be interpreted
as a tracking episode.
Fig. 8. Location of gaze during CGEs for each cat, plotted in a body-centered coordinate system. Azimuth and elevation are given in degrees. When
gaze would have fallen on the alley’s wall rather than on the floor, the corresponding circle is shaded.
G.A. Fowler, H. Sherk / Behavioural Brain Research 139 (2003) 83�/9690
3.4. Blinks
All cats blinked from time to time while walking down
the alley. Blinks could be brief, visible in only a single
videotaped frame, but more commonly lasted 100 ms or
more (Fig. 13). The frequency of blinks, and the total
amount of time spent with the eyes closed, varied
considerably among cats. Kraft and Leaf each spent
17% of the time with their eyes closed, while Jack, at the
other extreme, spent only 2% of the time blinking (Fig.
5).Blinks occurred most commonly during gaze shifts. In
the sample as a whole, 85% of blinks coincided with gaze
shifts, a pattern that is clear in the examples of Fig. 4.
Less commonly, a blink occurred in the middle of what
would otherwise be a single CGE (see, for example, the
blinks in Fig. 4C).
The high incidence of blinks during many trials
surprised us, and we wondered whether stationary cats
that were actively fixating visual targets blinked equally
frequently. We counted blinks made during the gaze
calibration task, in which the cat sat still and looked at a
wiggling pipecleaner (see Section 2). Both cats tested in
this fashion blinked very rarely when stationary. Jack
exhibited 0.018 blinks/s when stationary, compared to
0.16 blinks/s when walking. Kraft exhibited 0.035
blinks/s when stationary, compared to 1.31 blinks/s
when walking.
4. Discussion
The primary goal of this study was to determine what
pattern of retinal image motion locomoting cats experi-
ence. We found that, like stationary observers, cats
displayed many short episodes of constant gaze that
were separated by gaze shifts. During these CGEs,
images would have moved through the visual field in a
pattern approximating a Gibsonian optic flow field (Fig.
1A). Because cats almost always looked downward,
their fixation points did not coincide with the focus of
expansion, and thus even at the center of the area
centralis they saw image motion.
Fig. 9. Location of gaze during CGEs, plotted in physical space.
Ordinate plots distance in cm to the left or right of the cat’s mid-
sagittal plane. Only episodes in which gaze fell on the floor of the alley
are shown.
Fig. 10. Retinal image speeds during CGEs in which gaze fell on the
alley floor. Arrowheads indicate average values.
G.A. Fowler, H. Sherk / Behavioural Brain Research 139 (2003) 83�/96 91
4.1. Methodological considerations
Because conventional methods for monitoring eye
and head movements are not feasible for locomoting
observers, we have used a novel method to assess gaze.
When a cat fixates, head position indicates the direction
of gaze within about 28 because cats maintain their eyes
centered in the orbits, or nearly so [10]. During gaze
shifts, however, gaze does not necessarily coincide with
head position because cats may combine a saccade witha head rotation. We could detect fairly small saccades
(see Section 2) but we could not measure their size; thus
although we could describe the direction of gaze shifts,
we could not describe gaze trajectory during the shift.
As noted above, without knowing the size of such
saccades we might have over-estimated the duration of
associated gaze shifts, and consequently under-esti-
mated the duration of subsequent CGEs. However,when we compared potentially under-estimated CGEs
with ones whose starting times were unambiguous
because they immediately followed blinks, we found
no difference between the two groups.
The temporal resolution of our method was limited by
the VC’s frame rate, which was 30 Hz. Although modest
compared to the temporal resolution of magnetic coil
systems, this resolution was quite adequate for measur-ing the duration of gaze events (CGE’s, blinks, and gaze
shifts), since these events lasted at least 33 ms.
4.2. Constant gaze episodes
The most common gaze event that we observed was
the CGE. In duration and frequency, these resembled
the fixations exhibited by a stationary, unrestrained cat
in a study done by Collewijn [2]. Collewijn’s stationarycat made about 2.7 fixations/s, and our cats averaged 2.5
CGEs/s. The average fixation duration for the station-
ary cat was �/303 ms, and for our locomoting cats,
CGEs averaged 247 ms. In humans, the duration of
fixations depends on the task [12]: during reading
English text, for example, fixation durations generally
average 250 ms [28], and during free viewing of a
painting they averaged �/400 ms [35].It is generally assumed that stationary observers
acquire visual information primarily during fixations,
and it seems likely that our cats likewise obtained
information primarily during CGEs and not during
gaze shifts. In many cases this must have been true
because the cat closed its eyes during the gaze shift.
Similarly, Orchard and Stern [20] observed that humans
tend to concentrate their blinks during saccades ratherthan fixations when reading. Cats’ blinks during gaze
changes might serve the same function as saccadic
suppression, which decreases visual sensitivity during
rapid eye movements [34].
The conclusion that cats acquired information pri-
marily during CGEs implies that they ignored visual
cues during about 40% of each trial (the time occupied
by gaze shifts and blinks). Although 40% seems like asurprisingly large amount of time during which to
exclude visual information, this outcome is consistent
with percentages found by Patla et al. [21]. Their human
Fig. 11. Sizes of gaze changes (A), and of glances (B). Arrowheads
indicate average size of gaze change for each cat. Glance sizes were
pooled for all cats because they were too infrequent to show
individually.
G.A. Fowler, H. Sherk / Behavioural Brain Research 139 (2003) 83�/9692
subjects, who were fitted with liquid crystal goggles,
blocked out visual information for 60% or more of the
time while performing a visually-demanding walking
task (placing their feet on ‘stepping-stones’). This
parsimonious strategy was sufficient to produce a high
level of performance, just as our cats’ gaze strategy
sufficed for accurate foot placement in a difficult task.
During a CGE, the pattern of retinal image motion
presumably resembled the well-known optic flow field
illustrated in Fig. 1A. But locomoting cats did not keep
their gaze centered on their heading point, as has
sometimes been assumed (e.g. [1,24]). Instead, they
looked downward at the path ahead of them. The
average downward angle of gaze, 198 below the horizon,
Fig. 12. Examples of actual CGEs, and of gaze azimuth and elevation predicted if the cat had fixated and tracked the point on the alley floor where
its gaze fell at the beginning of the CGE. Actual values are indicated by filled circles, and predicted values are indicated by shaded circles. The best
matches are the first 3 examples (A)�/(C).
G.A. Fowler, H. Sherk / Behavioural Brain Research 139 (2003) 83�/96 93
placed gaze 58 cm ahead of the cat’s eyes. At a walking
speed of 43 cm/s (the average for cats in this study), the
cat would step on this location 1.3 s (or about 3 steps)
after its image crossed the center of gaze.
A strategy of maintaining a constant angle of gaze
during locomotion has both advantages and disadvan-
tages. A constant gaze yields a predictable, stereotyped
pattern of retinal image motion (the optic flow field of
Fig. 1A), but it also results in continuous image motion
across the center of gaze. Rapid image motion would
make precise visual analysis difficult. Typically, how-
ever, when cats looked at the alley floor, images movedat modest speeds through the area centralis, on average
158/s. The image speeds that we observed during CGEs
would strongly activate many cells in several areas of
visual cortex, including areas 17 [18], 18 [19], and the
lateral suprasylvian visual area ([30]; unpub. observ.).
Indeed, retinal image motion appears to be not only
consistent with visual analysis during locomotion, but
also necessary for accurate foot placement, since elim-ination of retinal image motion by the use of strobe light
has devastating effects on accuracy in this task [29].
Walking cats sometimes glanced to one side or the
other (and occasionally downward). Glances might be
considered to be particularly brief instances of CGEs,
but they were generally much more eccentric than CGEs
(Fig. 10). In some cats they were also more stereotyped.
One cat habitually glanced far to the right as sheemerged from the start box, and another usually glanced
to the left just before turning the corner of the L-shaped
alley. Patterns of constant gaze, in contrast, varied
unpredictably from trial to trial.
4.3. Tracking during locomotion
Do cats sometimes fixate and track stationary fea-
tures in the alley while walking? One might at first think
such behavior unlikely since previous studies have
shown that cats with immobilized heads use smooth
pursuit eye movements to track only slowly moving
targets (2�/68/s at most; [4,7], Malpeli, pers. commun.see, however, [17]), much slower than the image motion
that our locomoting cats typically experienced (Fig. 10).
However, smooth pursuit gaze movements, in which a
cat is free to move its head, have not been investigated.
We have found that stationary, unrestrained cats use
head movements to track a moving object such as a
bouncing ball1 (Sherk and Fowler unpub. observ.), and
thus walking cats should be quite capable of trackingstationary features. Another line of evidence points to
the same conclusion: when a patterned drum rotates
around a head-fixed cat, the drum elicits an optokinetic
response in which the cat’s eyes track the pattern with
speeds up to �/308/s, though with a gain substantially
less than one at moderate or high speeds [7,33]. But even
though walking cats appear capable of tracking static
features, we found very little evidence that they did so inthe alley task. Like cats tracking bouncing balls, walking
cats would presumably employ head movements, but
head trajectories rarely matched those predicted by the
Fig. 13. Durations of blinks for each cat.
1 In unrestrained monkeys, the head also contributes most of the
motion during smooth pursuit [16].
G.A. Fowler, H. Sherk / Behavioural Brain Research 139 (2003) 83�/9694
cat’s direction of gaze and walking speed. Two caveats
should be mentioned. First, we only compared predicted
and actual head trajectories when the total predicted
head movement was at least 58, which excluded most
CGEs because of their brevity. However, there is no
reason to suppose that shorter CGEs differ from longer
ones in any way except duration, and thus the absence of
tracking during longer episodes suggests that cats did
not track during shorter ones either. Second, we
assumed that cats would track stationary objects with
a gain of one. If their gain was substantially lower, we
might well have missed tracking episodes. Cats follow-
ing such a strategy would see a pattern of image motion
intermediate between conventional optic flow (Fig. 1A),
and ‘tracking’ optic flow (Fig. 1B). Although possible,
this strategy would fail to provide the benefits of either
constant gaze, which yields a stereotyped and thus
predictable pattern of motion, or of fixation and
tracking, which minimizes image motion in the area
centralis.In two studies on visually-guided human locomotion,
investigators concluded that humans frequently fixate
and track stationary features [14,22]. This conclusion
must be considered tentative, since these studies mea-
sured eye movements (in one case, only horizontal eye
movements) but not head position. Assuming that
humans do fixate and track in these situations, does
this indicate a difference from the cat? Not necessarily,
since both of the tasks employed in the human studies
may have promoted a fixation and tracking strategy. In
one study subjects had to place their feet on small,
irregularly-spaced stepping-stones, which apparently
resulted in fixation of each stepping-stone until the
foot had touched it. As readers will know from their
own experience, this is not a normal visual guidance
strategy; one rarely looks at one’s foot as it makes
contact with the substrate. In the other study, subjects
had to step over a solitary barrier. Although not visually
complex, this latter situation may have prompted
fixation on the barrier because it was the only object
present. Interestingly, in this study subjects also spent
about 33% of their time in CGEs (events termed by the
authors ‘travel fixations’ [22]). Thus there are similarities
as well as differences between the gaze strategies
employed by humans and cats during locomotion.
Possibly the differences between the studies reflect
differences in the tasks used more than a fundamental
difference between species.
References
[1] Albright TD. Centrifugal directional bias in the middle temporal
visual area (MT) of the macaque. Vis Neurosci 1982;2:177�/88.
[2] Collewijn H. Gaze in freely moving subjects. In: Baker R, Berthoz
A, editors. Control of gaze by brain stem neurons Developments
in neuroscience, vol. 1, 1977:13�/22.
[3] Collewijn H, Martins AJ, Steinman RM. Natural retinal image
motion: origin and change. Ann NY Acad Sci 1981;374:312�/
329.
[4] Conway JL, Timberlake GT, Skavenski AA. Oculomotor changes
in cats reared without experiencing continuous retinal image
motion. Exp Brain Res 1981;43:229�/32.
[5] Donaghy M. The cat’s vestibulo-ocular reflex. J Physiol
1980;300:337�/51.
[6] Donaghy M. The contrast sensitivity, spatial resolution and
velocity of tuning of cat’s optokinetic relex. J Physiol
1980;300:353�/65.
[7] Evinger LC, Fuchs AF. Saccadic, smooth pursuit and optokinetic
movements of the trained cat. J Physiol (London) 1987;285:209�/
29.
[8] Fowler GA, Soden C, Brush S, Sherk H. Gaze during visual
locomotion. Soc Neurosci Abst 2000;26(2):2085.
[9] Gibson JJ. The perception of the visual world. Boston:
Houghton-Mifflin, 1950.
[10] Guitton D, Douglas RM, Volle M. Eye-head coordination in cats.
J Neurophysiol 1984;52:1030�/50.
[11] Guitton D, Munoz DP, Galiana HL. Gaze control in the cat:
studies and modeling of the coupling between orienting eye and
head movements in different behavioral tasks. J Neurophysiol
1990;64:509�/31.
[12] Harris CM, Hainline L, Abramov I, Lemerise E, Camenzuli C.
The distribution of fixation durations in infants and naı̈ve adults.
Vision Res 1988;28:419�/32.
[13] Helmholtz H von. Treatise on physiological optics, vol. III
(Translated by Southall, JPC). New York: Dover Publications,
1910.
[14] Hollands MA, Marple-Horvat DE, Henkes S, Rowan AK.
Human eye movements during visually guided stepping. J Motor
Behav 1995;27:155�/63.
[15] Koenderink JJ. Optic flow. Vision Res 1986;2:161�/79.
[16] Lanman J, Bizzi E. The coordination of eye and head movement
during smooth pursuit. Brain Res 1978;153:39�/53.
[17] Missal M, Lefevre P, Crommelinck M. Evidence for high velocity
smooth pursuit eye movements in the cat. Neurosci Abst
1994;20:858.
[18] Movshon JA. The velocity tuning of single units in cat striate
cortex. J Physiol (London) 1975;249:445�/68.
[19] Orban GA, Callens M. Influence of movement parameters on area
18 neurones in the cat. Exp Brain Res 1977;30(1):125�/40.
[20] Orchard LN, Stern JA. Blinks as an index of cognitive activity
during reading. Integr Physiol Behav Sci 1991;26:108�/16.
[21] Patla AE, Adkin A, Martin C, Holden R, Prentice S. Character-
istics of voluntary visual sampling of the environment for safe
locomotion over different terrains. Exp Brain Res 1996;112:513�/
22.
[22] Patla AE, Vickers JN. Where and when do we look as we
approach and step over an obstacle in the travel path? Neurore-
port 1997;8:3661�/5.
[23] Perrone JA, Stone LS. A model of self-motion estimation within
primate extrastriate visual cortex. Vis Res 1994;34:2917�/
2938.
[24] Rauschecker JP, von Grunau MW, Poulin C. Centrifugal
organization of direction preferences in the cat’s lateral supra-
sylvian visual cortex and its relation to flow field processing. J
Neurosci 1978;7:943�/58.
[25] Robinson DA. A method of measuring eye movement using a
scleral search coil in a magnetic field. IEEE Trans Biomed
Electron BME 1963;10:829�/37.
G.A. Fowler, H. Sherk / Behavioural Brain Research 139 (2003) 83�/96 95
[26] Royden CS, Crowell JA, Banks MS. Estimating heading during
eye movements. Vis Res 1994;34:3197�/214.
[27] Solomon D, Cohen B. Stabilization of gaze during circular
locomotion in light. I. Compensatory head and eye nystagmus
in the running monkey. J Neurophol 1992;67:1146�/57.
[28] Salthouse TA, Ellis CL. Determinants of eye-fixation duration.
Am J Psychol 1989;93:207�/34.
[29] Sherk H, Fowler GA. Visual analysis and image motion in
locomoting cats. Eur J Neurosci 2001;13:1239�/48.
[30] Spear PD, Baumann TP. Receptive-field characteristics of single
neurons in lateral suprasylvian visual area of the cat. J Neuro-
physiol 1975;38:1403�/20.
[31] Stein BE, Goldberg SJ, Clamann HP. The control of eye
movements by the superior colliculus in the alert cat. Brain Res
1976;118:469�/74.
[32] Straschill M, Rieger P. Eye movements evoked by focal stimula-
tion of the cat’s superior colliculus. Brain Res 1973;59:211�/22.
[33] Strong NP, Malach R, Lee P, Van Sluyters RC. Horizontal
optokinetic nystagmus in the cat: recovery from cortical lesions.
Dev Brain Res 1984;13:179�/92.
[34] Volkmann FC. Human visual suppression. Vis Res 1986;26:1401�/
16.
[35] Yarbus AL. In: Riggs LA, editor. Eye movements and vision
(Translated by Haigh, B). New York: Plenum Press, 1967.
G.A. Fowler, H. Sherk / Behavioural Brain Research 139 (2003) 83�/9696