Self scaling reinforcement learning for fuzzy logic controller

摘要

In this paper, we propose a new reinforcement learning algorithm for generating a fuzzy controller. The algorithm generates a range of continuous real-valued actions, and reinforcement signal is self-scaled. This prevents the weights from overshooting when the system gets a very large reinforcement value. The proposed method is applied to the problem of controlling the brachiation robot, which moves dynamically from branch to branch like a gibbon swinging its body in a pendulum fashion.