Bayesian model for multimodal sensory information fusion in humanoid

In this paper, the Bayesian model for bimodal sensory information fusion is presented. It is a simple and biological plausible model used to model the sensory fusion in human’s brain. It is adopted into humanoid robot to fuse the spatial information gained from analyzing auditory and visual input, aiming to increase the accuracy of object localization. Bayesian fusion model requires prior knowledge on weights for sensory systems. These weights can be determined based on standard deviation (SD) of unimodal localization error obtained in experiments. The performance of auditory and visual localization was tested under two conditions: fixation and saccade. The experiment result shows that Bayesian model did improve the accuracy of object localization. However, the fused position of the object is not accurate when both of the sensory systems were bias towards the same direction.