A BAYESIAN PERSPECTIVE ON SPATIAL PERCEPTION Maaike de Vrijer Jan van Gisbergen February 20, 2008
Jan 15, 2016
A BAYESIAN PERSPECTIVE ON SPATIAL PERCEPTION
Maaike de Vrijer
Jan van Gisbergen
February 20, 2008
VISUAL STABILITY
Brain must combine visual and vestibular information to preserve a stable percept of the world when we make head movements
SEMICIRCULAR CANALS
Three set of mutually perpendicular canals measure rotations in three dimensions
High-pass properties
FUNCTION OF THE OTOLITHS
Sensitive to tilt and translation
Nerve fibers code deflection of the hair cell cilia (gravito-inertial force)
EXPERIMENTS
PERCEPTION OF VERTICALITY
How well can a tilted subject adjust a line to the direction of gravity?
Van Beuzekom & Van Gisbergen (2000) J. Neurophysiol.
Kaptein & Van Gisbergen (2004) J. Neurophysiol.
PASSIVE TILT EXPERIMENTS
RESULTS LINE ADJUSTMENT
”adjust the line to the direction of gravity”
De Vrijer et al. (2008)
SYSTEMATIC ERRORS LINE TASK
is tilt angle underestimated?
ACTIVE TILT RESULTS REVEAL PARADOX
tilt estimates are quite accurate
but large errors in line task
line task body tilt estimate
SPATIAL PERCEPTION TESTS
Paradox: subject knows tilt angle but has biased line settings
estimate body tilt adjust line to direction of gravity
BAYESIAN STRATEGY
BAYESIAN MODEL
OPTIMAL OBSERVER THEORY
De Vrijer et al. (2008)
• Noisy otolith signal would spoil precise visual signal
• Most of the time, the head is near upright (prior knowledge)
• Brain takes prior knowledge into account when using the sensory tilt signal
BAYES MODEL OPTIMAL OBSERVER
1) The incoming (noisy) otolith signal may have been caused by a range of possible tilt angles (likelihood function)
2) On a priori grounds, not all tilt angles are equally probable: mostly the head is near upright (prior)
3) The most likely tilt angle is the product of likelihood en prior (posterior)
BAYESIAN MODEL
Noisy tilt signal causes uncertainty about the actual tilt angle (likelihood)
Brain applies correction, based on experience of which tilt angles are most common (prior)
Result is a weighted compromise (posterior)
BAYESIAN MODEL FITS
De Vrijer et al. (2008)
tilt compensation
error
MODEL ASSUMPTIONS
1. Prior has a fixed width, independent of tilt angle
2. Noise in tilt signal increases with tilt angle: ………true?
MEASURING UNCERTAINTY:
PSYCHOPHYSICAL EXPERIMENTS
De Vrijer & Van Wamel (2007)
PSYCHOMETRIC CURVE OF 0o BODY TILT PERCEPT
Subject is tilted at various angles around 0o
Makes right or left judgment (forced-choice)
No sharply defined threshold, due to noise in tilt signal
NOISE IN TILT SIGNAL AT 00
Psychometric curve
Noise in tilt signal
NOISE BODY TILT SIGNAL AT 0o AND 90o
more noise at 90o
NOISE AT 0o AND 90o
results 5 subjects
more noise at 90o
VISUAL VERTICAL AND BODY TILT SIGNAL AT 90o
less accuracy but better precision in visual vertical
VISUAL VERTICAL AND BODY TILT SIGNAL
visual vertical
body tilt
Rechtop is de perceptie van lijnoriëntatie bijna foutloos
gekantelde proefpersoon ziet de visuele wereld geroteerd
klokschaal methode
Effect van sensor ruis
prior heeft des te meer effect op het percept naarmate de sensorische meting ruiziger is
Bayes model visuele verticaal
de Vrijer et al. (2007)
Oculomotor verticaal
Van Beuzekom & Van Gisbergen (2000)
fouten in visuele en oculomotor verticaal zijn vergelijkbaar
RENS VINGERHOETS
AMBIGUITY PROBLEM OTOLITHS
Detect gravito-inertial force (GIF)
No distinction between tilt and translation
tilt
translation
models vestibular signal processing
SOLVING THE AMBIGUITY PROBLEM: TWO MODELS
Two putative strategies for decomposition of the otolith signal:
• frequency-filtering model
• canal-otolith interaction model
inverse problem
CANAL-OTOLITH INTERACTION
• canals respond to rotation during tilt
• brain can use their signal to decompose otolith signal
SPATIAL ORIENTATION ILLUSIONS
PILOT AIRCRAFT CARRIER
pilot is upright during launch in the dark, but feels tilted backward
ROTATION IN DARKNESS
rotation percept decays slowly; after stop subject feels rotation in opposite direction
reflection of cupula mechanics
EXPERIMENTS
Percepts during rotation about a tilted axis
Vingerhoets et al. (2006) J. Neurophysiol.
Vingerhoets et al. (2007) J. Neurophysiol.
THE ACTUAL MOTION
- rotation about tilted axis
- in darkness
- constant velocity
PERCEIVED MOTION
Results support canal-otolith interaction model
rotation percept
translation percept
TRANSLATION AND ROTATION PERCEPT
rotation percept
translation percept
MODEL SIMULATION
otolith signal, actually caused by tilt changes, is partially ascribed to translation when the rotation signal dies out
SPATIAL PERCEPTION TESTS
Paradox: subject knows tilt angle but has biased line settings
estimate body tilt adjust line to direction of gravity