Effects of target disappearance during pursuit maintenance
EFFECT OF TARGET BLINKS
The “normal-maintenance” and “blinked-maintenance” conditions presented target motion at a constant 15°/s. For the blinked-maintenance condition, a 200-ms target blink occurred after eye velocity had reached target velocity. shows a response of monkey P to a single trial of the blinked-maintenance condition. The onset of target motion caused brisk eye acceleration after a latency of ~90 ms. A catch-up saccade occurred just as eye velocity reached target velocity. Eye velocity stayed close to target velocity until after the time of the target blink. About 120 ms after the onset of the target blink, eye velocity declined, reaching a minimum near 5°/s before starting to accelerate again ~70 ms after the target reappeared. Eye velocity again reached and remained near target velocity, and a saccade brought the eye back to the target position.
FIG. 2 A representative response to the blinked-maintenance condition in which a target blink occurred after accurate pursuit was underway at 15°/s. Thin black lines show eye position and velocity. The sharp vertical deflections in the eye velocity trace (more ...)
The response shown in was typical: target blinks during pursuit maintenance always produced a moderate to large “dip” in eye velocity, beginning ~120 ms after the offset of the target. , shows, for one monkey, average eye velocity during responses to the normal-and blinked-maintenance conditions. , shows (solid lines) average responses to the blinked-maintenance condition for each of the five monkeys tested. Target blinks, indicated by the solid horizontal bars below the traces, produced large dips in eye velocity for every monkey. Each solid trace is actually a pair of closely overlapping traces, showing two averages of the same data, one made using the saccade-excluded method (thin black traces) and one made using the saccade-interpolated method (bold gray traces). The saccade-excluded method makes averages of eye velocity that treat time points during saccades as missing data. The saccade-interpolated method makes averages of eye velocity after estimating the underlying pursuit velocity during each saccade via linear interpolation (see METHODS). In each example, the averages of eye velocity are nearly identical for these two methods.
FIG. 3 Average responses from all monkeys tested for the 3 maintenance conditions. Responses to the blinked- and occluded-maintenance conditions are shown, respectively, by the solid and short dashed traces. For these 2 conditions, the closely overlapping thick (more ...)
We quantified the magnitude of the dip in eye velocity by taking the difference between eye velocity measured at a time before the dip and the minimum eye velocity during the dip. The former was measured 80 ms after the onset of the blink, which was always before eye velocity began to decline. Measurements were made from the saccade-interpolated averages but were virtually identical if made from the saccade-excluded averages. For the blinked-maintenance condition, (□) shows the size of the dip in eye velocity for each monkey. Target blinks produced dips in eye velocity from 7 to 12°/s with a mean of 10.6°/s. For comparison, eye velocity changed little over the same interval for the normal-initiation condition (
FIG. 4 Quantitative summary of the changes in eye velocity induced by target blinks and occlusion during pursuit maintenance. □,
, and , measurements made from the blinked-, normal-, and occluded-maintenance conditions. Decreases in eye velocity (more ...)
Each monkey performed numerous trials that contained target blinks and had ample opportunity to understand that the target always reappeared after the blink. Trials ended only after the target had reached the far end of the display and, for most experiments, only after it had stopped and remained stationary for 500–700 ms. The presence of a blink in a particular trial was entirely predictable from the starting location of the target, and the precise time and place of the blink was also entirely predictable from the location of the nearby occluder. Yet monkeys improved little in their ability to maintain normal pursuit during the target blink. The most improvement was shown by monkey P, who performed the blinked-maintenance condition 133 times during the course of his first experimental session. For the first 20 trials, the target blink produced a dip in mean eye velocity of 13.5°/s. During the last 20 trials the dip was 10.6°/s, an improvement of 21%. gives results of this analysis for the five monkeys tested. Although some monkeys showed small improvements, none were able to ignore the target blinks despite their predictable nature. , and show the behavior of monkey P on his first and second experimental days. The size of the dip in eye velocity was similar in the two instances. Even after 2 days and hundreds of repetitions of the exact same target motion, he was unable to ignore a target blink that lasted only a fifth of a second.
Changes in eye velocity during target blinks
EFFECT OF TARGET OCCLUSION
The occluded-maintenance condition was identical to the blinked-maintenance condition except that the target traveled at a vertical location 3° lower and a piece of tape occluded the blink, creating the impression that the target went behind the tape. Comparison of the middle and bottom traces in reveals that the behavioral effect of target absence can be altered dramatically by the presence of an occluder. The dip in eye velocity after the offset of the target was much smaller in the occluded-maintenance condition than in the blinked-maintenance condition. The bold dashed traces in , show that a reduction in the blink-induced dip was obtained for all monkeys tested. The
in show the amplitude of the dip in eye velocity during the occluded-maintenance condition. The error bars show the SE. Thus the change in eye velocity during target occlusion was statistically very different from the change in eye velocity during target blinks (□). The decline in eye velocity was reduced by 45–83% by the presence of the occluder (alternately, the dip was 83–477% larger in the absence of the occluder). Nevertheless, eye velocity always declined somewhat more during occlusion than during normal uninterrupted pursuit (thinner, long-dashed traces in ,
The presence of the occluder also had some small effects on pursuit even before the blink began. For three monkeys (P, Q, and N), maintained pursuit was slightly slower when the target was approaching the occluder as opposed to when the target was 3° higher for the blinked-maintenance condition or 6° higher for the normal-maintenance condition. For one monkey (O), maintained pursuit was slightly faster when approaching the occluder, and for another one (M), maintained pursuit was slightly higher for the normal-maintenance condition. The differences in maintained pursuit are probably due to the leftward retinal motion of the occluder, the expectation of the monkey regarding the upcoming blink/occlusion, or the vertical difference in target location. Regardless, small differences in the gain of pursuit maintenance have little impact on our measurement of the dip in eye velocity, which for each condition was made relative to eye velocity measured before the dip. For example, in eye velocity is considerably lower during occlusion than during the normal-maintenance condition. However, this was true even before the target disappeared. The change in eye velocity during the relevant interval differed by <2°/s between the two conditions (, P2).
The effect of the occluder was not primarily due to learning or experience with this task. Consider monkey P, who had not pursued targets with blinks before the experiment shown in . For the first 10 trials of the occluded-maintenance condition, the average dip in eye velocity was only 5.4°/s, much smaller than in the blinked-maintenance condition (13.5°/s for the first 20 trials, and 11.4°/s for all trials). Similar results were obtained for the other monkeys: even for the first 10 trials, the occluder reduced the size of the dip in eye velocity by 35–80%.
Might the occluder indirectly enhance pursuit by inducing saccades (Lisberger 1998
)? It seems likely that more saccades might be evoked during occlusion than during blinks. This was in fact true. During the interval from the target offset to the bottom of the dip in eye velocity, the percentage of trials that contained a saccade ranged from 35 to 96% for the occluded-maintenance condition, from 11 to 34% for the blinked-maintenance condition and from 13 to 46% for the normal-maintenance condition. We addressed this issue by making averages of eye velocity that excluded trials if a saccade occurred within a window that started 150 ms prior to the onset of the dip in eye velocity and ended at the time eye velocity began to recover. For the occluded-maintenance condition, the dip in eye velocity was still small. Across monkeys, the dip was a mere 10% larger for trials in which saccades were not present, compared with 83–477% larger for trials in which the occluder was not present. Thus the effect of the occluder cannot be mediated by saccades.
Effects of target disappearance during pursuit initiation
As described in the preceding text, we found that target blinks during pursuit maintenance cause a decline in eye velocity, while occlusion of the target allows the continuation of near normal pursuit. What maintains eye velocity during target occlusion? Two possibilities are that pursuit is driven by a memory of target velocity (in space, presumably, rather than on the retina) or that it is driven by a motor memory of eye velocity itself. There is theoretical and behavioral precedent for both possibilities. Various pursuit models have proposed that pursuit is supported by internal feedback that creates either a target-velocity signal or a motor “eye-velocity memory” (Churchland and Lisberger 2001a
; Dicke and Thier 1999
; Huebner et al. 1990
; Krauzlis and Lisberger 1989
; Newsome et al. 1988
; Pack et al. 2001
; Pola and Wyatt 1997
; Robinson et al. 1986
; Stone et al. 2000
; Young et al. 1968
). Behaviorally, image stabilization experiments have been cited as evidence that pursuit possesses an eye-velocity memory mechanism (Morris and Lisberger 1987
). On the other hand, image stabilization experiments are also consistent with the use of an internal target-velocity signal. The existence of an oculomotor memory for target-velocity is demonstrated by the ability of monkeys to saccade to the virtual position of a moving target during a long target blink (Barborica and Ferrera 2003
). To discriminate between these two possibilities, we tested the effect of target occlusion during pursuit initiation before eye velocity had reached target velocity. If pursuit during target occlusion is supported by eye-velocity memory, then eye acceleration should halt, and eye velocity should be maintained until after the target reappears. If pursuit during occlusion is supported by a target-velocity memory, then eye acceleration should continue until target velocity is reached as the memory of target velocity should be close to the actual target velocity.
EFFECT OF TARGET BLINKS
Target blinks during pursuit initiation provide control data against which the responses to target occlusion can be compared. In the blinked-initiation condition, the target was blinked so as to interrupt pursuit while the eye was still accelerating. The effect of target blinks during pursuit initiation was similar to that during pursuit maintenance. In the example shown in , initial eye acceleration was on course to bring eye velocity swiftly to target velocity but was interrupted ~60 ms after the onset of the target blink (, 1st oblique arrow). Eye velocity remained constant near 23°/s for ~70 ms (i.e., until 130 ms after the target disappeared) and then declined sharply (, 2nd oblique arrow). Approximately 60 ms after the target reappeared, the eye began again to accelerate, and a subsequent saccade brought the eye back to the target. Eye velocity then fluctuated at speeds slightly slower than target velocity and fell to near zero after the target stopped moving. The solid traces in , show averages of eye velocity in response to the blinked-initiation condition for three of the four monkeys tested. Eye acceleration halted or was abruptly reduced after the start of the target blink, and eye velocity declined during the blink for most monkeys (monkeys P
, and also monkey M
, whose average response is not shown for this condition). One monkey (O
, ) showed no dip in eye velocity during the target blink, although eye velocity was much reduced compared with the normal-initiation condition (long-dashed traces). The □ in , left
, summarize these findings, and plot the size of the blink-induced dip in eye velocity. In all but one instance, eye velocity declined as a result of the target blink. Even for the exception, eye acceleration was greatly reduced relative to the normal-initiation condition (
FIG. 5 A representative response during the blinked-initiation condition, in which the target was blinked shortly after the onset of target motion. Black traces show eye position and velocity. The sharp vertical deflection in the eye velocity trace corresponds (more ...)
FIG. 6 Average eye velocity responses for conditions in which the target was blinked or occluded during the initiation phase of pursuit (A–C) or during the accelerating pursuit response to a change in target velocity (D). Saccade-interpolated and -excluded (more ...)
FIG. 7 Quantitative summary of the changes in eye velocity induced by target blinks and occlusion in the initiation, velocity-change, and delayed-initiation conditions. Increases in eye velocity are graphed upward, and decreases are graphed downward. Each triplet (more ...) EFFECT OF TARGET OCCLUSION
The occluded-initiation condition was identical to the blinked-initiation condition except that a piece of tape occluded the blink, creating the impression that the target went behind the tape. The short-dashed traces in , illustrate responses to the occluded-initiation condition. As with the blinked-initiation condition, eye acceleration was reduced or abolished starting ~60 ms after the target disappeared. However, the dip in eye velocity that was observed during target blinks was largely absent during occlusion. Eye velocity either remained roughly stable () or accelerated moderately (). The
in , left
, show the change in eye velocity during occlusion. There was always less of a decline in eye velocity than was seen for the blinked conditions (□) and never as much increase as was seen for the normal condition (
). For the blinked-, occluded-, and normal-initiation conditions, eye velocity during the analysis interval changed on average −8.4, +0.1, and +15.4°/s, respectively. Thus while occlusion largely or entirely abolishes the drop in eye velocity produced by a target blink, it does not restore normal pursuit initiation. On average, eye velocity changes little during the interval affected by occlusion, consistent with the hypothesis that pursuit during occlusion is sustained by an eye-velocity memory. Note, however, that there was often some
eye acceleration during the relevant interval, a finding we discuss in a later section. Thus sustained eye-velocity memory cannot account for all aspects of pursuit during occlusion.
Effects of target disappearance after a change in target velocity
We also tested the effect of blinks and occlusion that interrupted the response to a change in target velocity. The three conditions used were the normal-velocity-change condition, the blinked-velocity-change condition, and the occluded-velocity-change condition. Each of these presented a target trajectory similar to that in the analogous initiation condition, but stable pursuit was achieved at 5°/s before the target accelerated to 35°/s. Not surprisingly, results for the velocity-change conditions were similar to those for the initiation conditions as illustrated by the data for monkey M in . The bars in , middle, summarize the responses to these conditions. Target blinks caused declines in eye velocity, while target occlusion led to eye velocities that were relatively unchanged over the analysis interval. Across the three monkeys, the average changes in eye velocity for the normal, blinked, and occluded conditions were 10.7, −0.3, and −8.7°/s.
For both the initiation and velocity-change conditions, saccades were a common, sometimes universal feature of pursuit for the time period under study. The frequency of saccades typically differed between the blinked, occluded, and normal conditions. We calculated the percentage of trials with at least one saccade during a time interval defined relative to the blinked condition response: from target offset to the point at which eye velocity began to recover after the target reappearance. Across monkeys, the mean percentages were 48% (range of 15–89%) for the blinked conditions, 80% (60–100%) for the occluded conditions, and 84% (69–100%) for the normal conditions. Given the nearly universal presence of saccades for the occluded and normal conditions, it was not practical to construct averages of the saccade free trials as we had for the maintenance conditions. There were often no such trials, and when there were, pursuit tended to be quite poor overall, consistent with the idea that the entire oculomotor system was making little effort to track the target. Nevertheless, an attempt to factor out the contribution of saccades can be made. For two monkeys, saccadic frequency was fortuitously similar across the three conditions. For monkey P, saccadic frequencies for the blinked-, occluded-, and normal-velocity-change conditions were, respectively, 70, 70, and 69%. For monkey M, saccadic frequencies for those three conditions were 89, 98, and 99%. Both these monkeys showed the typical pattern of results: eye velocity declined during the blink, stayed constant during occlusion, and rose during normal pursuit (, middle). From these comparisons, and from the previous analysis of the maintenance conditions, it is very unlikely that the facilitation provided by the occluder is effected primarily via saccades.
Effects of target disappearance on the initiation of pursuit to a previously moving target
We motivated the occluded-initiation condition by asking whether pursuit during occlusion is sustained by a memory of eye velocity or by a memory of target velocity. We stated that in the latter case, the eye ought to continue to accelerate toward target velocity during the occluded interval. This claim presumes that the internal representation of target velocity nears actual target velocity soon after the sensory information reaches cortex. What if, instead, the internal representation of target velocity takes considerable time to “charge”? To test this possibility, monkeys were trained to withhold pursuit responses until the fixation point was extinguished (the delayed-initiation task), allowing time for an internal representation of target velocity to “charge.” The target moved at 35°/s for 400 ms before the fixation point was extinguished, and for another 100 ms before reaching the occluder, or undergoing a blink. , right
, shows that the effect of blinks and occlusion were much the same in this paradigm as in the standard initiation and velocity-change paradigms: eye velocity declined during blinks (□) and remained relatively constant during occlusion (
Anticipatory eye acceleration during target occlusion
As seen in , target occlusion causes an abrupt reduction or halt in eye acceleration during pursuit initiation. In most cases, eye velocity then remained relatively constant until the visually driven response to the target reappearance. Nevertheless, there was typically some eye acceleration prior to the visually driven response. In , monkey Q shows eye acceleration just prior to the visually driven response to the reappearance of the target (i.e., in the short-dashed trace, before the time indicated by the last thin vertical line). Monkey O () showed particularly robust eye acceleration during occlusion: the eye accelerated for the duration of the relevant interval (short dashed trace). Monkey O even showed some eye acceleration when pursuit initiation was interrupted by a blink (solid trace).
We refer to such eye acceleration as “anticipatory” because it anticipates the reappearance of the target and because of its resemblance to previous examples of eye acceleration prior to a predictable event (e.g., Kao and Morrow 1994
; Wells and Barnes 1999
). To quantify anticipatory eye acceleration, we calculated average eye acceleration during two 40-ms intervals placed 1
) just after the halt in visually driven eye acceleration, 80–120 ms after the target disappearance and 2
) just before the visually driven response to the target reappearance, 20–60 ms after that reappearance. All measurements were made from responses to the occluded-initiation condition. Eye acceleration during the first interval ranged from −78°/s2
, day 1) to 50°/s2
), with an average of −23°/s2
. Eye acceleration during the second interval ranged from −9 (monkey P
, day 1) to 50°/s2
), with an average of 29°/s2
. By comparison, during the interval from 80 to 120 ms after the target reappearance, visually driven eye acceleration averaged 161°/s2
, or 550% greater than the average anticipatory acceleration measured prior to target reappearance.
For each experimental session, we measured average eye acceleration in the second measurement interval described in the preceding text for both the first 10 and last 10 trials of the occluded initiation condition. Eye acceleration during the first 10 trials ranged from −68 (monkey P, day 1) to 12°/s2 (monkey Q), with an average of −22°/s2. Eye acceleration during the last 10 trials ranged from 28 (monkey P1) to 51°/s2 (monkey M), with an average of 40°/s2. In summary, anticipatory eye acceleration was greatest immediately prior to the reappearance of the target, and grew with experience.
Latencies of the effects produced by target blinks
There were consistent differences in the response latencies to the onset of target motion, the disappearance of the target at the start of the blink, and the reappearance of the target at the end of the blink. We first consider the blinked-maintenance condition. The thin vertical lines in mark the expected onset of the response to these three events, given the expected latency of 65 ms. This expectation is based on the typical duration of the open loop interval for our monkeys, defined as the response latency for a change in target velocity during pursuit (e.g., Priebe et al. 2001
). Consider first the responses of monkey M
in . A 65-ms latency accounts nicely for the onset of pursuit (1st vertical line) and for the start of eye acceleration after the reappearance of the target (3rd vertical line). However, the decline in eye velocity produced by the target blink began later than expected. For quantitative comparison, shows latencies measured by hand from the averages of eye velocity. The latency of pursuit initiation was always slightly longer than the latency of the response to the target reappearance, consistent with our prior finding that the latency of pursuit initiation is usually slightly longer than the true open loop interval (Priebe et al. 2001
). More surprisingly, response latencies for the disappearance of the target were roughly twice as long as the response latencies for the reappearance of the target. Findings were similar if we measured latency using an automatic algorithm. The average latency across monkeys is shown at the bottom of for both the manual and automated measurements.
Latencies from changes in the target: maintenance conditions
Response latencies in the blinked-initiation and -velocity-change conditions show a pattern similar to that found for the blinked-maintenance condition. The thin vertical lines in mark the expected start of the response, given a 65-ms latency to the onset of target motion, the disappearance of the target at the start of the blink, and the reappearance of the target at the end of the blink. A 65-ms latency accounts reasonably well for the cessation of eye acceleration after the disappearance of the target, and for the eye acceleration that begins after the reappearance of the target. However, for the blinked-initiation and -velocity-change conditions, the decline in eye velocity lagged the disappearance of the target by considerably greater than 65 ms. This feature is most clearly illustrated in (solid lines), where eye velocity was maintained with little acceleration or deceleration for a 50- to 70-ms period after the cessation of eye acceleration. Only at the end of this period did behavior diverge strongly for the occluded and blinked conditions. In the next section, we will model the existence of two latencies by assuming two mechanisms in pursuit: one that provides visuo-motor drive with a short latency and one that disengages eye velocity memory with a longer latency.
summarizes the results of our latency analysis for the blinked-initiation and -velocity-change conditions. The average latency for the eye to begin to accelerate after the reappearance of the target was 59 ms. The average latency for the eye to cease accelerating, after the disappearance of the target, was similar: 60 ms. In contrast, the average latency for the eye to begin decelerating following the disappearance of the target was twice as long: 125 ms. The average latency of pursuit initiation was 82 ms.
Latencies from changes in the target: initiation conditions
Simulations of an “image-motion” model
The experiments described above demonstrate that the pursuit response to a target blink is profoundly affected by the presence of an occluder. To test if our results could be explained by descending modulation of the basic pursuit kernel, we have added suitable modulation to a version of the image-motion pursuit model (Churchland and Lisberger 2001a
; Krauzlis and Lisberger 1994
). The term image-motion is meant to indicate that the model has no access to the actual velocity of the target in space but only to an estimate of image velocity provided by the visual system. A prior version of this model is available with a simple graphical user interface at: http://keck.ucsf.edu/~sgl/top_pursuitmodel.htm
shows a minor adaptation of our model. The model includes two basic processes. In the first, visual motion signals create a “visuo-motor drive” that is a command for eye acceleration (Ë′). In the second process, a motor “eye-velocity memory” mathematically integrates eye acceleration commands, and sustains pursuit once target velocity has been reached.
FIG. 8 Architecture and behavior of an image motion pursuit model. A: schematic diagram of the model. The arrows show the flow of signals and the elements in boxes represent various transformations. , Ë, and Ý represent target, eye, (more ...) Visuo-motor drive
is provided by two parallel pathways shown on the left side of . The top pathway contributes an image velocity signal and includes a linear gain (8.9°/s2
of eye acceleration per °/s of image velocity) and a filter (exponential time constant of 30 ms). The bottom pathway contributes a command related to image acceleration and consists of a nonlinear gain element, a differentiator, and a filter (exponential time constant of 10 ms). The nonlinear gain had a slope that was 0.92 near zero and declined with increasing image velocity. The nonlinearity was included because it captures an important nonlinearity in pursuit (Churchland and Lisberger 2001a
) and helps produce realistic responses for the normal-maintenance and normal-initiation conditions. Visuo-motor drive was gated by a gain, g1
, whose purely pragmatic purpose was to abolish visuo-motor drive in the absence of a target.
is created by a positive feedback loop with a gain labeled g2
(Krauzlis and Lisberger 1994
). The feedback loop acts as a perfect integrator when g2
is set to one and as a leaky integrator with an exponential time constant when g2
is less than one.
Visual and motor delays are lumped together in a single 60-ms delay element (Δt
). A filter (filter3
) with an exponential time constant of 20 ms represents the compensated dynamics of the plant (Robinson et al. 1986
) and intervenes between the eye velocity command (Ë
′) and the “actual” eye velocity (Ë
). The values of the delay, the gain of the image velocity pathway, and the “plant” filter time constant were set based on prior modeling or experimentally measured values. The other parameter values were chosen to produce realistic pursuit trajectories.
We use the image motion model diagrammed in to test the hypothesis that the responses to target blinks and occlusion can be accounted for if we assume that visuo-motor drive is lost during both blinks and occlusion, due to the absence of image motion, and eye-velocity memory is maintained during occlusion but not during blinks. For simulations of normal pursuit, the gain of visuo-motor drive (g1 in ) had a value of one. For the blinked and occluded conditions, g1 fell to zero 60 ms after the target disappeared and returned to normal 60 ms after the target reappeared. For simulations of normal pursuit, the gain of eye-velocity memory (g2 in ) began at one and declined to 0.999 linearly over the course of the trial. This decline was included to account for the observed decline in maintained eye velocity over the course of a trial. For the occluded conditions, g2 retained this normal sequence of values. For the blinked conditions, g2 dropped to a value of 0.99 120 ms after the start of the target blink, rendering eye-velocity memory a leaky integrator with an exponential time constant of 100 ms. The value of g2 returned to normal 60 ms after target reappearance.
The simulations in , account for most features of pursuit during blinks and occlusion. Simulated eye velocity provides a reasonable emulation of normal pursuit (long dashed traces), falls appropriately after target blinks (solid traces), and is maintained at the preexisting eye velocity during target occlusions (short dashed traces). The latencies of the different effects are not emergent properties of the model but were determined by the times chosen to change the values of g1 and g2. For example, in the blinked-initiation condition (, solid trace), the model shows a 60-ms interval of maintained eye velocity between the cessation of eye acceleration and the start of eye deceleration because g1 was set to zero starting 60 ms after the blink onset, while g2 was reduced another 60 ms later. It is an assumption of the model that the latency to disengage eye-velocity memory is longer than the latency of visuo-motor drive.
Note that the model as currently configured does not account for the anticipatory eye acceleration produced during occlusion prior to the visually driven response to the target reappearance. This is expected if one accepts that anticipatory eye acceleration is produced by predictive mechanisms that are separate from the basic pursuit kernel that the image motion model is intended to capture.