Extended Homeostatic Adaptation: Improving the Link between Internal and Behavioural Stability

摘要: This study presents an extended model of homeostatic adaptation designed to exploit the internal dynamics a neural network in absence sensory input. In order avoid typical convergence asymptotic states under these conditions plastic changes are induced evolved neurocontrollers leading renewal that may favour sensorimotor adaptation. Other measures taken loss variability (as caused, for instance, by synaptic strength saturation). The method allows generation reliable morphological disruptions simple simulated vehicle using neurocontroller has been selected behave homeostatically while performing desired behaviour but non-homeostatically other circumstances. performance is compared with controllers have only positive link between and behavioural stability. networks perform much better more adaptive never experienced before agents.