特集 センサフュージョン ニューラルネットによる感覚運動融合の階層型モデル

Abstract

Human beings perceive the physical world by integrating various sensory inputs, information about own motor system and their knowledge. To process sensory information, the human brain has a hierarchical parallel distributed processing mechanism. Sensor fusion technology focuses on simulating the brain's sensory information processing mechanism and is intended for advanced sensing systems which cannot be constructed with unimodal sensory information processing.Our study of sensor fusion aims to develop a hierarchical sensory-motor fusion mechanism for achieving intentional sensing, which is a concept that sensing has a goal of perception (or intention of sensing) and sensing behaviors must be oriented to achieve the goal.In this paper, we propose a hierarchical sensory-motor fusion model with neural networks for intentional sensing, and also propose an iterative inversion method which takes advantage of multi-layer neural networks as a solution to the ill-posed inverse problem. We applied the hierarchical sensory-motor fusion model to a three-dimensional object recognition system and a vision-based robot arm control system, and demonstrated the effectiveness of the proposed model by computer simulations. We confirmed that the model accepts and propagates intentions, enables to tightly couple recognition and action, and can perform various tasks without rebuilding or relearning the sensing system.