DECENTRALIZED NEURAL NETWORK BASED POSITION CONTROL OF MULTI-FINGERED DEXTEROUS HAND

A decentralized neural network based position control method is presented for multi-fingered dexterous hand with nonlinear tendon-drive system. Multiple decentralized neural network is acquired via learning nonlinear relationship of tendon-driving system using the simple input output data that obtained by applying different input to the complex drive system. Then the inverse model of the tendon-drive system can be established using the decentralized neural networks according to the characteristics of the tendon-drive system. In the meantime, an online neural network is used to keep well running state for a long time. Experiment illustrates the proposed method can improve greatly the tracking accuracy.