An information theory analysis of critical Boolean networks as control software for robots

This work is an analysis of critical random Boolean networks used as control software for robots. The main goal is to find if there are relations between information theory measures on robot's sensors and actuators and the capability of the robot to achieve a particular task. Secondary goals are to verify if just the number of nodes of the networks is significant to obtain better populations of controllers for a given task and if a Boolean network can perform well in more than one single task. Results show that for certain tasks there is a strongly positively correlation between some information theory measures and the objective function of the task. Moreover Boolean networks with an higher number of nodes tend to perform better. These results can be useful in the automatic design process of control software for robots. Finally some Boolean networks from a random generated population exhibit phenotypic plasticity, which is the ability to manifest more phenotypes from the same genotype in different environments. In this scenario it is the capability of the same Boolean network (same functions and connections) to successfully achieve different tasks.