Spatial coding of the Predicted Impact

Location of a Looming* Object

M. Neppi-Mòdona

D. Auclair

A. Sirigu

J.-R. Duhamel

Approaching Objects

• When will it arrive or pass – time to contact (TTC)

• Will it hit me?– How far to one side will it pass?– Where will it strike me?

• In interception of an object where and when are not independent problems– In this study where is

constrained to the plane of the eyes

Main Questions• How accurate is performance

when predicting impact location on the face– For targets originating straight

ahead– For targets with eccentric origin

• What reference frame are approaching objects represented in?– Retinal, visuotopic, intermediate– Manipulated alignment of the

observers retinal and visuotopic frames

LED stimuli (5mm diameter) originated from 40 cm in a dark and frameless environment

Travelled at 20 cm/s for 1 s, then occluded for final 1 s of approach trajectory

3 start points, 7 end points, one of which is the cyclopean eye (or midline of the head)

Simplest case is fixation directly ahead, and midline aligned with central Fixation LED, causing retinotopic and visuotopic frames to lie on top of one another

Task is to judge left / right of midline (forced choice)

Fixating central with central LED origin and central impact location produces same retinal image as fixation to left with left LED origin and central impact

When the start point is 17 degrees to the left of the fixation and midline there is an ipsilateral bias in the judgment, i.e. a bias to predict an impact location on the same side as the start point.

For this observer, for left origins 100% of impacts on the nose, or 15 mm to the right of the nose were reported as impacts on the left of the face. 75% of impacts 30 mm to the right of the nose were also reported as left.

This pattern is symmetrical, and there is no increase in JND for the peripheral presentation, and there is still a ceiling and floor problem for JND estimation.

The effect of eccentric origin on prediction

For straight ahead origin perception of impact location is unbiased, with the 50% point judgment point occurring for the central impact location of the 7 tested

JND (difference between 50% and 75% “right”) 2.2mm

The measurement was inadequate to fit functions and estimate JND. 3 of the 7 impact locations give a ceiling effect, and 3 give a floor effect.

The impact locations were spaced 15 mm apart, so how can JND in the range of 2 mm be measured?

Why ipsilateral bias?

• The bias occurs because the judgment is made under conditions of uncertainty, forcing the use of a heuristic strategy– The level of uncertainty is greater than

in normal interception conditions because the “where” judgment must be made at TTC 1000 msec, whereas TTC 500 msec would be a more natural point to initiate an action. (See handout)

– At TTC 1000 all stimuli starting on the left are still on the left (with 2 degrees separating each of them), all from the right are still on the right, and those from the centre are split 3/1/3

– At TTC 500 the separation is more like 5 degrees between each stimulus, and one of them has crossed the midline

– Between TTC 1000 and 500 whether angle alpha is growing, shrinking, or constant is above threshold, and this gives unambiguous information about the destination of the stimulus. (See Table)

– Authors do not discuss this optical variable, which is also available, but may be below threshold in the portion of the trajectory they do show.

• Allowing spatial vision of background would help bring it above threshold

– Given the impoverished information at TTC 1000, a simple heuristic is to respond left if it is on the left, and right if it is on the right. For the 18 stimuli that don’t hit the nose this strategy performs at 66% (100% for straight origin), and produces the observed biases.

• A more general point is that natural selection would favour a mechanism to intercept impacts, not predict where they will occur, which is has less adaptive value.

• Those objects observers judged incorrectly as “left” would not have crossed the midline until very late in their trajectory (see diagram), so actual interception would occur on the left.

Why ipsilateral bias?

Visuotopic or retinotopic space? (Misaligning the angle of gaze and the midline)

• Is the approach angle of an object relative to the observer correctly perceived, independently of retinal position?

• Or, is there an influence of retinal position, producing biases in perception of approach angle?

• By pointing the midline towards the central target location as before, but fixating one of the eccentric positions the foveal origin of the retinal coordinate frame is no longer aligned with the midline origin of the visuotopic frame

• Observed biases might indicate an influence of the retinotopic frame, but a purely retinotopic observer could in principle perform accurately using the direction of change of alpha

Midline and fovea aligned with target origin (no bias)

Both misaligned with target origin (large bias)

Midline aligned, fovea misaligned (? bias)

Midline misaligned, fovea aligned (? bias)

Right

bias

Group data (individuals differ qualitatively from each other)

Impact prediction takes place in an intermediate reference frame?

Individual Differences

Partial-correlation between prediction error and position in each of the reference frames

Presume based on trials where frames misaligned

Red line is bias for central origin trajectories during right fixation

Retinal frame dominant

Visuotopic frame dominant

Intermediate frame dominant

Subject 4 is not consistent with use of direction of change of alpha strategy at TTC 1000 because basic ipsilateral bias for central fixation is still present

Conclusions• I predict that all the observed biases,

whether they are ipsilateral with straight ahead fixation or related to the misalignment of retinal frame from visuotopic frame would disappear if the stimulus was allowed to develop to a more realistic value of TTC 500 msec– This is because direction of change of

alpha would be well above threshold. Note that this would allow observers to behave independently of eye position as if they had a visuotopic representation, without actually having one at all.

• I predict that allowing spatial vision of the background would bring change of alpha above threshold for earlier TTC values– Deletion and accretion of texture as alpha

changes would be a powerful cue, and would certainly make it obvious when alpha was unchanging, specifying a central impact ( in the presented data central impacts are sometimes mistaken for lateral ones)

Conclusions

• Authors agree with me that one explanation for the ipsilateral bias is covert interception

• I think the question of reference frames for this task is interesting, but I don’t believe the conditions tested establish that a visuotopic frame is needed for this task under more natural viewing conditions

• To convince me, authors need to show effects under more natural viewing, and rule out the alpha explanation