Basic research of reinforcement learning with Self Organizing Map

The conventional Reinforcement Learning(RL) uses pre-defined state space for choosing actions from environmental information. To apply RL in actual continuous space, the division of the space is required. For this purpose, we proposed RL with Self Organizing Map(SOM). The continuous space is mapped to the lattices by SOM, and the actions for each lattice are learned by RL. In this paper, the basic characteristics of this algorithm is studied.