Cognitive Neural Network Driving DoF-Scalable Limbs in Time-Evolving Situations

Abstract

Object handling and manipulation are vital skills for humans and autonomous humanoid robots. The fundamental bases of how our brain solves such tasks remain largely unknown. Here we develop a novel approach that addresses the problem of limb movements in time-evolving situations at an abstract cognitive level. We exploit the concept of generalized cognitive maps constructed in the so-called handspace by a neural network simulating a wave simultaneously exploring different subject actions independently on the number of objects in the workspace. We show that the approach is scalable to limbs with minimalistic and redundant numbers of degrees of freedom (DoF). It also allows biasing the effort of reaching a target among different DoF.