A Preliminary Study on BN-Robots' Dynamics

Abstract

Boolean network robotics concerns the use of Boolean networks as robot programs. In this brief contribution, we outline preliminary results on the analysis of the dynamics of BN-robots trained to accomplish a composite task. We find that the successful performing robots, which show the capability of robustly attaining the learned behaviours while adapting to new tasks to perform, are characterised by both number of fixed points and complexity higher than those of unsuccessful ones. These results are in accordance with the conjecture that systems able to balance robustness and evolvability work at the border between order and chaos and may suggest improvements to the automatic design of robot programs.