Research On Intelligent Control Schemes For Multi-fingered Robot Hands

As an end-effector of a robot manipulator, the hand is a key part of a robot to accomplish tasks effectively, and also is a concrete expressing device of robot"brain"activities. In order to meet the needs of more and more complicated and flexible tasks, the hand has become one of the robotic research areas, which has been paid more attention and is being greatly developed. Based on the synthetic analysis of the state and development of the multi-fingered robot hand, this thesis mainly deals with the real-time intelligent control schemes for the multi-fingered robot hand and shows simulation results of a two-fingered robot hand with the control schemes mentioned later. The main contents of the thesis are as follows:Firstly, considering the constraint relationship between the multi-fingered hand and the grasped object, the kinematics and dynamics of the hand are built. To implement coordinated manipulating of an object, a position/force hybrid control scheme, which allows position/force feedback loops to split into independent control with respect to position and force, is presented, and simulation of the multi-fingered robot hand with two fingers is performed.Secondly, to deal with adverse effects caused by unknown robot finger masses and unknown object mass, in the thesis presented are two intelligent control schemes, the control scheme based on reinforcement learning and adaptive fuzzy sliding mode control scheme. The former one is a new approach combining reinforcement learning with feedback control. Through reinforcement learning, the trajectory tracking error caused by these uncertainties can be compensated and high performance will be achieved. The later one combines sliding mode control with adaptive fuzzy nerve network which is utilized to approach the nonlinear control part of sliding mode control. The parameter of the adaptive fuzzy control is adjusted on-line to deal with these uncertainties, thus the robustness and high performance of the system are obtained.Finally, considering the feature of the multi-fingered robot hand system, in the thesis presented is a control schemes for the hand which, based on multi-Agent theory, has been implemented on the JADE platform.