Abstract
In this paper, we propose a memory discrimination method of an autonomous mobile robot such that when it acts, while storing some environment and its behavior, the robot distinguishes current environment and as a result, it acts using the memory that is appropriate for its environment. In this system, if the robot can not distinguish between correct storage and other storage appropriately, it does not work well. For this reason we propose two new methods to distinguish between the storage environments. One is the method using Fourier Transform. The other is the method using the mixed normal distribution function. Here, as the learning system, Q-learning method, a kind of reinforcement learning, is used and as the memory system, plural chaotic neural networks with bidirectional associative memory are used. Finally, the effectiveness of the proposed method constructed by three parts, learning, memorizing, and the proposed distinguishing function of the environment is verified through the simulation applying it to the optimal path searching problem in mazes.
