Influence of the space segmentation and its adaptive automation for reinforcement learning
Akira Notsu, Yuki Komori, Katsuhiro Honda, Hidetomo Ichihashi · 2011
We performed a single pendulum simulation and observed the influence of the situation space segmentation pattern in reinforcement learning processes in order to propose a new adaptive automation for situation space segmentation. Usually, in real-world reinforcement learning processes, infinite states and actions and the uncertainty of the optimum solution make the learning process more difficult than the finite Markov decision process. In a numerical experiment, a single pendulum simulation is performed in order to demonstrate the influence and adaptability of the proposed method.