Self-adjusting ring modules (SARMs) for flexible gait pattern generation

Using the principle of homeostasis, we derive a learning rule for a specific recurrent neural network structure, the so-called Self-Adjusting Ring Module (SARM). Several of these Ring Modules can be plugged together to drive segmented artificial organisms, for example centipede-like robots. Controlling robots of variable morphologies by SARMs has major advantages over using Central Pattern Generators (CPGs). SARMs are able to immediately reconfigure themselves after reassembly of the robot's morphology. In addition, there is no need to decide on a singular place for the robot's control processor, since SARMs represent inherently distributed control structures.