A Hormone-Based Controller for Evolutionary Multi-Modular Robotics: From Single Modules to Gait Learning

Heiko Hamann, J├╝rgen Stradner, Thomas Schmickl, Karl Crailsheim
IEEE Congress on Evolutionary Computation (CEC'10) (2010)


  For any embodied, mobile, autonomous agent it is essential to
  control its actuators appropriately for the faced task. This holds
  for natural organisms as well as for robots. If several such agents
  have to cooperate, the coordination of actions becomes important. We
  present an artificial homeostatic hormone system which is a
  bio-inspired control paradigm. It allows to control both, a single
  robot as well a set of cooperating modules in multi-modular
  reconfigurable robotics. Our approach is inspired by chemical
  signal-processing and hormone control in animals. Evolutionary
  computation is used to adapt controllers for two distinct
  morphological robot configurations (uni- and multi-modular),
  different environmental conditions, and tasks. This approach is
  compared to artificial neural networks. Our results indicate, that
  the proposed control paradigm is well adaptable to different robot
  morphologies and to different environmental situations. It is able
  to generate behaviors for several robotic tasks and outperforms
  neural networks in terms of evolvability in the tested multi-modular
  robotic setting tested.

Download PDF: