Sensory-motor fusion based on dynamics matching: learning of strategies in target tracking task

摘要

The authors propose a novel concept named "dynamics matching " for designing sensory-motor fusion system. This means a strategy to adjust adaptively the contents of the information processing and the control to maximize overall performance of the system under constraints such as restricted computational resources, physical properties of the sensor and motor systems and requirements by given tasks. The authors construct an algorithm utilizing reinforcement learning where the system performance is regarded as the "reward". The effect of the proposed concept is illustrated by a numerical experiment of a target-tracking task.