Đang chuẩn bị liên kết để tải về tài liệu:
Biomimetics Learning from nature Part 4

Tham khảo tài liệu 'biomimetics learning from nature part 4', kỹ thuật - công nghệ, cơ khí - chế tạo máy phục vụ nhu cầu học tập, nghiên cứu và làm việc hiệu quả | Neurobiologically inspired distributed and hierarchical system for control and learning 83 inverse the feedforward controller commands the body more dominantly. Fig. 4 illustrates the FEL scheme proposed by Gomi and Kawato Kawato Gomi 1992 .The feedback controller can be linear for example as . . . . Tfb K10b-0 K 20b-0 K 30b-0 1 To acquire the inverse model different learning schemes could be used. In general a TT TT TT learning scheme T f ộ ỡđ 0 0đ 0 0đ 0 W can be expressed where W represents the adaptive parameter vector 0d the desired position vector and 0 the actual position vector. The adaptive update rule for the FEL is as follows. T ext 2 where Text is the external torque and the learning ratio which is small. Fig. 4. The FEL model. Adapted from Kawato and Gomi 1992 . The convergence property of the FEL scheme was shown Gomi Kawato 1993 Nakanishi Schaal 2004 . The FEL model has been developed in detail as a specific neural circuit model for three different regions of the cerebellum and the learning of the corresponding representative movements 1 the flocculus and adaptive modification of the vestibuloocular reflex and optokinetic eye movement responses 2 the vermis and adaptive posture control and 3 the intermediate zones of the hemisphere and adaptive control of locomotion. The existence of inverse internal model in the cerebellum is argued based on studies Wolpert Kawato 1998 Wolpert et al. 1998 Schweighofer et al. 1998 that the Purkinje cell activities can be approximated by kinematic signals. There have been many other models of the cerebellum Barto et al. 1998 Miall et al. 1993 Schweighofer et al. 1998 . In those models the cerebellum is also either feedforward or feedback control system. Yet uniform descriptions for various models would be necessary to support one model over the other as there are multiple ways to describe one model. Interestingly a probabilistic modelling approach has been applied to explain the inverse 84 Biomimetics Learning from .