Research On Goal-Oriented Motion Control Of Joint Structure

Joint is a widespread kind of structure in natural world and artificial machines. The modeling and motion control of joint are researched in computer animation and CNC machine tools. Forward kinematics and inverse kinematics are two types of usually used motion control methods. Inverse kinematics is goal-oriented. The solutions to inverse kinematics are grouped into two categories, one is based on nonlinear equations such as Newton-like iterative method, pseudo-inverse matrix, SR-inverse matrix, damped leased squares method; the other is based on optimizations such as cyclic coordinate descent, time-space constraint, dynamic constraint. Furthermore, there are inverse kinematic methods that using redundant degrees of freedom to satisfy kinematic chain constraints or achieve secondary task. These methods established solid foundation for further modeling and solution.Curve inverse kinematics for joint structure is proposed in this thesis in the case of end effector move along curve. Curve inverse kinematics establishes direct contact of kinematic chain motion curve in joint angle space and end effector trajectory in3-D space. The recursive equation of joint angle is derived, and two kind of error control mechanism are proposed to construct a complete algorithm. Weighted least norm method are used together with curve inverse kinematics to handle singularity.The offset coursed by cutter shape are researched for CNC machine tool. Instead of traditional cutter contact offset, a machine-cutter kinematic chain model and corresponding motion control method are proposed. The machine-cutter kinematic chain has unified form, and introduces two more passive degrees of freedom to exert the most flexibility of CNC machine.Another inverse kinematics algorithm based on spatial clustering is proposed. By using a mathematical modeling method, the forward kinematics can be seen as the mapping from the point in the space of joint angles to the position of the end effector in the3D space, while the inverse kinematics seen as the mapping from the end effector position in3D space to the points in the joint angle space. The method uses a double layer Hierarchical Inherit Model, one layer is the position of the end effector, denoted by P layer, and the other layer is the joint angle layer, referred to Q layer. The set of points in both P and Q layers are clustered successively, and the establishment of the correspondence relationship between the clusters could be extracted as motion model. The solution of the inverse kinematic is to find a suitable cluster in Q layer by given a cluster in P layer.Motion synthesis system based on motion capture data and Bayesian network is proposed. The system can synthesize human walking animation according to given footprints sequence. The system extracts keyframes and gait parameters from motion capture data and for samples for Bayesian network learning. Footprint sequences are input to the trained Bayesian network, and a complete virtual human walking motion is generated automatically. This approach combines the control of extremities by gait parameters and non-deterministic reasoning ability of the Bayesian network. Experiments show that synthetic motion fully meet time and space constraints and maintain good realism.