抄録
Reinforcement learning approaches attract attention as a technique to learn mapping functions between sensorsmotors of an autonomous mobile robot through trial-and-error. However, traditional reinforcement learning algorithms need to prepare state space and action space that are appropriately divided by a designer beforehand. This paper proposes a new reinforcement learning algorithm that can learn mapping functions between continuous state space and grid action space, and the mapping functions are described by neural networks. The proposed method has two features. One is that the number of middle neurons and the initial value of weight parameters of neural networks are set through the dimensional number of state spaces automatically. The other is that an action is selected by using Griddy-Gibbs sampler. The proposed method is demonstrated through a navigation problem of an autonomous mobile robot, and comparing with when a designer sets the parameters of neural networks heuristically.
