SPATIAL AWARENESS DEMO 20 juni 2008 detection of self motion sensing body orientation in space...

SPATIAL AWARENESS DEMO

20 juni 2008

detection of self motion sensing body orientation in space visual perception in earth-centric coordinates

VESTIBULAR SENSORS

canals

otoliths

CANALS DETECT ROTATION

high-pass filter

insensitive to constant velocity rotation

nerve fibers code head velocity

CONSTANT ROTATION IN DARKNESS

• rotation percept decays

• after stop, percept of rotation in opposite direction

• reflects cupular mechanics

OTOLITHS

sensitive to tilt and translation

OTOLITH SIGNAL IS AMBIGUOUS

hair cells cannot distinguish tilt and translation

SPATIAL ORIENTATION ILLUSION

pilot is upright, but feels tilted

AMBIGUITY PROBLEM

otolith signal may have various causes:

• translation (a)• force of gravity due to tilt (g)• combination of a and g

How can the brain resolve this ambiguity ?

inverse problem

CANAL- OTOLITH INTERACTION MODEL

• canals detect rotation during tilt changes

• their signal helps to decompose otolith signal

Angelaki et al. (1999)

CANAL–OTOLITH INTERACTION MODEL

basic principle:- tilt stimulates otoliths AND canals- translation stimulates only otoliths

Merfeld and Zupan (2002) J. Neurophysiology

tilt angle

linear acceleration

angular velocity

OVAR

Vingerhoets et al. (2006) J. Neurophysiol.

Vingerhoets et al. (2007) J. Neurophysiol.

TESTING THE MODEL

THE ACTUAL MOTION

- rotation about tilted axis

- in darkness

- constant velocity

MODEL PREDICTIONS

rotation signal decays gradually

wrong interpretation otolith signal: illusory translation percept

SCHEMATIC SUMMARY OF RESULTS

confirms prediction

rotation percept

translation percept

Actual motion:

Percept:

TRANSLATION AND ROTATION PERCEPT DATA

rotation percept

translation percept

SPATIAL PERCEPTION IN STATIC TILT

SENSING THE DIRECTION OF GRAVITY

Two different tasks:

1. Set line to vertical (SVV)

2. Estimate your body tilt (SBT)

Van Beuzekom & Van Gisbergen (2000) J. Neurophysiol.

Van Beuzekom et al. (2001) Vision Res.

Kaptein & Van Gisbergen (2004, 2005) J. Neurophysiol.

De Vrijer et al. (2008) J. Neurophysiol.

experiments in darkness

ACCURACY vs PRECISION

Accuracy:

How close is the response to the true value?

Precision:

How reproducible is the response?

darts analogy:

ACCURACY AND PRECISION IN LINE TASK (SVV)

accuracy

precision

De Vrijer et al. (2008) J. Neurophysiol.

De Vrijer et al. (2008) in progress

ACCURACY IN LINE TASK

due to underestimation of body tilt?

NO UNDERESTIMATION OF BODY TILT

SVV SBT

• Subjects know quite well how they are tilted (SBT)

• Yet, their line settings undercompensate for tilt (SVV)

Van Beuzekom et al. (2001) Vision Res.

Kaptein and Van Gisbergen (2004) J. Neurophysiol.

PRECISION IN LINE TASK

is scatter in SVV simply reflection of noise in body tilt signal?

De Vrijer et al. (2008) J. Neurophysiol.

De Vrijer et al. (2008) in progress

SVV LESS NOISY THAN SBT

De Vrijer et al. in progress

psychometric experiments at 0o and 90o tilt:

SVV LESS NOISY THAN SBT

SUMMARY SBT AND SVV DATA

Two paradoxical findings:

1. subject knows tilt angle, yet makes biased line settings

2. more certain about line setting than about body tilt

estimate body tilt (SBT) adjust line to vertical (SVV)

SBT DATA SHOW:

• An unbiased head tilt signal is available

• Noise increases with tilt angle

SIGNALS REQUIRED FOR SPATIAL VISION

retinal signal

to compute line in space (Ls), brain must combine info about line orientation on retina (LR) and head tilt (HS)

head-tilt signal

SIMPLY USING RAW TILT SIGNAL …

would not explain SVV bias !!spatial vision would be accurate, but noisy

raw tilt signal

A BAYESIAN PERSPECTIVE

IDEAL OBSERVER MODEL

IDEAL OBSERVER STRATEGY

1) Use sensory data: noisy tilt signal suggests range of possible tilt angles (likelihood)

2) Use prior knowledge: we know that large tilt angles are very uncommon (prior)

3) Most likely tilt angle (posterior) is product of likelihood and prior

Eggert (1998) PhD Thesis, Munich

MacNeilage et al. (2007) Exp. Brain Res.

De Vrijer et al. (2008) J. Neurophysiol.

IDEAL OBSERVER STRATEGY

Tilt prior has 2 effects on SVV:

• Less noise

• Bias at large tilt

WHY WOULD THIS MAKE SENSE?

1) Less noise in spatial vision

2) Downside: bias at large tilts

3) Average performance improves (large tilts are rare)

DEMO

BIAS EFFECT INCREASES WITH TILT

no bias

De Vrijer et al. (2008) J. Neurophysiology

no bias

small bias

small bias

large bias

large bias

MODEL PARAMETERS

1) head tilt noise level in upright

2) increase of head tilt noise with tilt

3) prior width

4) eye torsion amplitude

MODEL FITS: SVV ACCURACY

MODEL FITS: SVV ACCURACY

< 0

MODEL EXPLANATION OF NOISE LEVELS:SVV vs SBT PRECISION

• SVV is less noisy than the SBT (remarkable, but explained by model)

• SBT becomes more noisy at larger tilt (supports model assumption)

• SBT noise levels compatible with head-tilt fit results

CONCLUSION

Accuracy-precision trade-off in spatial vision:

• Bayesian strategy reduces noise at small tilts

• causes systematic errors at large tilts

ACTIVE-TILT RESULTS ARE SIMILAR

body tilt estimates are quite accurate

but large errors in line task

line task body tilt estimate

Van Beuzekom et al. (2001) Vision Res.

psychometrische curvevan 0o tilt percept

proefpersoon wordt vaak in allerlei standen rond 0o gekanteld

beoordeelt elke stand als links of rechts

geen scherpe drempel door ruis in tiltsignaal

Ruis tiltsignaal bij 00

psychometrische kromme

ruis in tiltsignaal

Ruis tiltsignaal bij 0 en 90o

meer ruis bij 90o

Ruis tiltsignaal bij 0 en 90o

resultaten 5 proefpersonen

meer ruis bij 90o