Reinforcement Learning Based on State Space Model using Growing Neural Gas for a Mobile Robot

摘要

In the application of Reinforcement Learning to real tasks, a state space construction is an important problem. In order to use in real world environment, we need to deal with the problem of continuous information. Therefore, we proposed a Growing Neural Gas method based on state space construction model. In our system, the agent constructs State Space Model from its own experience autonomously. Furthermore, it can reconstruct a suitable state space to adapt complication of the environment. Through the experiments, we showed that our method using state space performs as well as the conventional method by using a smaller number of states.