Using Genetic Network Programming to Get Comprehensible Control Rules for Real Robots
Tadahiko Murata, D. Okada · 2006
We have already proposed several derivatives of genetic network programming (GNP), and examined its performance by computational experiments. GNP is a kind of genetic algorithms designed for controlling some agent in a certain virtual environment. In this paper, we firstly apply a rule structure developed by a GNP to control a real robot. In order to develop a rule structure for controlling a real robot, we should simplify a state-action space for a rule structure in our simulator. We show how to simplify it and results of computational and real experiments. That is, we apply our genetic network programming not only in computational experiments but also to a real robot, AIBO ERS-7M2, an entertainment robot produced by Sony. Our experiments show that we can obtain comprehensible control rules by genetic network programming.