Sensory estimates are limited in resolution and corrupted by noise. When multiple cues to the same property are available, such as visual and haptic cues to shape, it is possible to improve the accuracy (reduce the variance) of the final estimate by integrating these individual estimates [1]. Recent research shows human adults to be proficient at integrating multisensory cues in a manner that is optimal for reducing variance. This entails averaging the cues while taking their individual reliabilities into account. We will present two studies examining the developmental basis for this ability. Even young infants know the correspondences between multisensory stimuli, and respond more rapidly to multimodal than unimodal events. We asked whether these early-emerging multisensory capacities provide a basis for optimal integration based on cue reliability, or whether this kind of integration depends on a different, still unknown developmental process.

We first tested 4-8 year olds’ and adults’ accuracy in homing to a location in a darkened room using (i) visual landmarks and (ii) the internal sense of direction updated with self-motion (e.g. vestibular) cues, when these were presented either separately, together, or under a small conflict [2]. While adults integrated landmark and self-motion information near-optimally to reduce the variance of their spatial estimates, children as old as 8 years did not integrate the cues but followed one or the other. This suggests a surprisingly late development for integration of visual and vestibular spatial cues. We followed this up with a study posing the same information-processing problem in a different way: a psychophysical two-alternative-forced-choice (2AFC) task with minimal motor and memory demands, testing how accurately participants could judge the slants of surfaces based on disparity and texture– two cues within the visual modality. When both disparity and texture cues were available, adults integrated these to reduce the variance of their estimates of surface slant. By contrast, children as old as 10 years did not show a significant improvement given both cues together rather than one alone.

Taken together, these results suggest that young children either (1) cannot compute weighted averages based on multiple cues, or (2) use a perceptual decision rule (criterion for how much sensory evidence to correct before responding) that is optimised for speed rather than accuracy. I will discuss some experimental and modelling work on the time course of perceptual decisions based on multiple cues that addresses this second hypothesis.


[1] Ernst, M. O. & Banks, M. S. (2002). Humans Integrate Visual and Haptic Information in a Statistically Optimal Fashion. Nature, 415, 429-433.

[2] Nardini, M., Jones, P., Bedford, R., & Braddick, O. (2008). Development of Cue Integration in Human Navigation. Current Biology, 18, 689-693.