Proposition of Memory Architecture for Real-Time Action Search of Autonomous Mobile Robot

Abstract

This paper proposes a new memory architecture which takes into consideration the computational resource constraint in the real-time search. In the proposed architecture, each data is hierarchically stored in the order of its frequency in the real-timesearch. Moreover, the number of stored data which are checked on-line varies according to the allocated time to find a solution. Therefore, the computational amount which is necessary to recall a stored data can flexibly change according to the current situation, and the efficient recall can be realized even if the allocated computational resource is not enough.
In this paper, the proposed method is applied to the action acquisition for an autonomous mobile robot in the moving obstacle avoidance problem. The usefulness is verified through some experimental results.