作者: Rodolphe J. Gentili , Hyuk Oh , Javier Molina , José L. Contreras-Vidal
DOI: 10.1007/978-1-4419-1452-1_6
关键词:
摘要: One fundamental problem for the developing brain as well any artificial system aiming to control a complex kinematic mechanism, such redundant anthropomorphic limb or finger, is learn internal models of sensorimotor transformations reaching and grasping. This since mapping between sensory motor spaces generally highly nonlinear depends constraints imposed by changing physical attributes hand changes in brain. Previous computational suggested that development visuomotor behavior requires certain amount simultaneous exposure patterned proprioceptive visual stimulation under conditions self-produced movement—referred ‘motor babbling.’ However, geometrical specific human arm finger have not been incorporated these performance 3D. Here we propose large scale neural network model composed two modular components. The first module learns multiple inverse features an fingers having seven four degree freedom, respectively. Once 3D kinematics limb/finger are learned, second simplified strategy whole shaping during grasping tasks provides realistic coordination among fingers. These bio-inspired functionally mimic cortical able reproduce movements. high modularity this makes it suited high-level neuro-controller planning grasp motions actual robotic system.