The purpose of our study is to establish a robust communication, while keeping security and privacy, between a handheld communicator and the surrounding information environment. From the viewpoint of low power consumption, we have been developing a reflectivity modulating communication module composed of a liquid crystal light modulator and a corner-reflecting mirror sheet. We installed a corner-reflecting sheet instead of light scattering sheet in a handheld videogame machine with a display screen with a reflection-type liquid crystal. Infrared (IR) LED illuminator attached next to the IR camera of a base station illuminates all the room, and the terminal send their data to the base station by switching ON and OFF of the reflected IR beam. Intensity of reflected light differs with the position and the direction of the terminal, and sometimes the intensity of OFF signal at a certain condition is brighter than that of ON signal at another condition. To improve the communication quality, use of machine learning technique is a possibility of the solution. In this paper, we compare various machine learning techniques for the purpose of free space optical communication, and propose a new algorithm that improves the robustness of the data link. Evaluation using an actual free-space communication system is also described.