Visual Reinforcement Learning Algorithm using Self Organizing Maps and Its Simulation in OpenGL Environment

摘要

Recently, the camera systems becomes more available for mobile robots. But scene analysis for generating control signals is still difficult and consumes large computational power. For this problem, the control method which generates the control signals directly from the raw camera images will be effective. In this paper, we use the reinforcement learning using the camera image as input data. For the division of the states represented with camera images, self organizing map is introduced. The division of the states and learning of the control signal using reinforcement learning are executed simultaneously on the map. For examining the performance of this algorithm, we made the simulation system with Graphical User Interface using OpenGL.