Solving POMDPs with Automatic Discovery of Subgoals
Le Tien Dung, Takashi Komeda, Motoki Takagi · BiblioBoard Library Catalog (Open Research Library) · 2009
In this chapter, we have proposed Reinforcement Learning using Automatic Discovery of Subgoals to accelerate learning of RL with RNN by profiting useful skills. Hidden units and their connections of RNNs, which are used by generated skills, are integrated into the RNN of the main policy. Experiment results of the E maze problem and the virtual office problem show the potential of this method.