Abstract
In this paper, we propose a move-action learning model for animal-like robots. A robot using the model learns to move to satisfy a desire. In our previous study, the model used a Recurrent Neural Network (RNN). The RNN learns movement by Back Propagation Through Time (BPTT) in our previous study. However, there is a problem that it is difficult to establish BPTT learning periods of learning for this model. Instead, we use a Genetic Algorithm (GA) to optimize connection weights for the RNN. We performed a simulation about a move-action learning model because we want to verify whether the robot can move from any location to satisfy its desire.
