Abstract
This paper presents a Machine Learning Method suitable for actual environments and treats Collision Avoidance Problem for multiple autonomous mobile robots as a case study.It is necessary that all autonomous mobile robots equip collision avoidance system as a function of movinig task commonly.Therefore, the ability of collision avoidance independently with moving objects is indispensable and should not be depending on devices constructing robots.The purpose is to emerge traffic rules that heterogeneous robots can avoid one onother with communicating only simple information.Using Reinforcement Learning suitable for uncertainty environment, the experiment was conducted in both distinctive and simple settings and gave some good results that collision avoidance actions emerged.In addition to, this paper shows stability of presented method by discribing the deference of performance in respect to the number of state partition which is one of the important parameters.
