Evolutionary robot simulations with competitive-cooperative neural network and adaptive synaptic couplings
Tatsuro Shinchi, Masayoshi Tabuse, A. Todaka, Y. Tokushige, Tetsuro Kitazoe · 2002
This paper describes a new approach to control systems for an autonomous mobile robot by using two different kinds of neural networks: 1) a neural network for recognizing sensor information with a mechanism of competition and cooperation, where synaptic couplings are fixed; and 2) a neural network with adaptive synaptic couplings corresponding genotype in animals and is used for self-learning of wheel controls. In the computer simulation model with both the neural networks, we successfully developed a typical robot with good performance when moving along a curved wall. The first part of the networks plays the role of decision making among sensor signals under noisy environment, while the second part is effective to adjust the synaptic couplings through genetic operations so that it may transfer the outputs from the first stage with the competition-cooperation neural network (CCNN) to the rotation of the robot wheel. The test performed shows the superiority of the CCNN.