Research On Cross-Coupled Control Scheme For Dexterous Hand Based On Neural Network

As the most important end effector in robot system, the dexterous robot hand has been the hotspot in the field of robot research. Researches on the control of dexterous robot hand are very significant to improving the quality of the whole robot system. According to the project"Research on a new generation of five-finger anthropopathic dexterous robot hand and its cooperative control"(863 program) of National Science and Technology Ministry, this dissertation applies the cross-coupled control on finger base joint, and studies on the self-tuning of coupled control parameters based on neuron network at HIT/DLR II dexterous hand with it's real-time control platform. The main purpose of this research is to improve the tracing accuracy of fingertip position.The base joint of the finger is coupling-driven by two motors together. Due to the differences on electronic and mechanic processes, the synchronization of the two motors can not be guaranteed. As for common position control methods for series-robot such as PID control, their control target is the position error of independent driver without considering the synchronous error of coupled motion. So the position controlling methods mentioned above are not completely suitable for the base joint of the finger, and they may influence the tracking accuracy of finger position and the operation performance, thus influence the grasp performance of the dexterous hand. In this dissertation, after the introduction of the hardware and the controlling system of dexterous robot hand, the key of the research lies in the coupling error caused by non-synchronization between the two motors at finger's base joint. By designing the cross-coupling controller of the dexterous robot hand, coupling error is brought in the control item and compensated, which improves the tracking accuracy. Besides, the simulation system of the finger is constructed for validating the algorithm.There are four controlling parameters in cross-coupled controller, and it is difficult for time-varying and nonlinear system to achieve accurate adjustment. With respect to the problems above, the dissertation works on the influence of each coupled control parameter on the controlling performance, and then based on the strong nonlinear mapping and self-study ability of neuron network, tuning the coupling control parameters by backpropagation algorithm of neuron network. In order to diminish synchronization error quickly and ameliorate the performance index function, the quadrics which contains synchronization error is proposed. By constructing the parameter adjuster of neuron network, the self-tuning of coupling control parameters is achieved,and the adjustability of the dexterous hand to environment are improved as well. Furthermore, for the sake of improving the position-control accuracy more, the friction parameter of finger joints is identified and compensated. At last the validity of above-mentioned control algorithm are proved by experiments.