Development of a Neuromorphic Control System for a Biped Robot

Abstract

Humanoid robots are developed around the world with the purpose to assist humans in their domestic and public activities and operate in unstructured and hazardous environments. To accomplish this effectively, intelligent humanoids should be autonomous, to accomplish high-level human tasks without help, and adaptable, to be able to react to dynamic changes and external disturbances in operating environments. The primary objective of this thesis is to investigate how biologically plausible methods such as reservoir computing and rewardmodulated learning can be used for generating robust sensory-motor outputs and achieving adaptability of the biped system. Recurrent neural networks architecture is studied on two robot systems: first is Asimo humanoid and second is biped developed at Nazarbayev University.