Kinematic And Dynamic Control Of A Humanoid Pingpong Manipulator

Humanoid pingpong manipulator, imitating the mechanical structure of human arm, is built to play pingpong, with 6 or more degrees of freedom. It can move with dexterity like a human arm. The humanoid pingpong manipulator is a kind of traditional serial link manipulator with rotational joints. It is more compact in structure, lighter in weight, faster and more dexterous in moving than the normal industrial robots. Besides the pingpong playing task, the grasping task could be fulfilled if the pingpong pat at the end of the manipulator is replaced with gripper or dexterous robotic fingers. The control methods for pingpong manipulator could be adopted to welding, spaying or educational robots. The study on the control law of pingpong manipulator offers technical help to the widespread use of humanoid manipulators in the service robot field.The three editions of humanoid pingpong manipulators we built could run at position mode, velocity mode and current mode. The control mode is divided into three based on the input of the servo amplifier which drives the motor of the joint. The input of amplifier is the joint angles when the manipulator is running at position mode. Velocity mode means that the input is joint velocities and current mode corresponds to motor currents. The current mode is called torque mode, since the current of DC servo is in direct proportion to the motor torque. The major research works and contributions of this thesis include:1) In order to handle the uncertainty of DH parameters which is lead by imprecisely manufac-ture and assembly, a kinematic calibration method is presented. The Leica Geosystems Absolute Tracker is used to get the exact position of the manipulator end-effector. The kinematic errors are determined by least-norm calculation of numbers of measuring data. The position accuracy of the end-effector is approved with the compensation kinematic errors. And, a dynamic parameter identification method is proposed, which is based on quadratic programming. A new method to make the inertia matrix positive definite is developed via doing quadratic programming twice. The dynamic parameters estimation based on the Solidworks model is also given.2) A lower dimensional task function method for point-to-point tasks of non-redundant manip- ulator is developed in this thesis. A series of mapping functions which map the six-dimensional point-to-point task to a lower dimensional task space are given. In the lower dimensional task space the manipulator is changed to redundant one and redundant control laws could be adopted for the subtasks. And the conditions that the mapping function should satisfy are also given.3) A generalize-weighted least-norm method for the control of redundant manipulator with multi subtasks is developed. This method maintains the advantage of weighted least norm method which has fantastic ability to avoid joint limits. With a new concept of virtual joints, the subtasks performance criterions are changed to the virtual joints limits problem and solved by weighted (?)t norm method. With this generalized method, the number of subtasks to be deal with could be larger than the number of joints of the manipulator.4) An adaptive dynamic friction compensation method is developed in this thesis. With the dy-namic LuGre friction compensation, obtaining the internal bristle-deflection state is the most important in the control design. A simple nonlinear observer, which is a simplification of De Wit's nonlinear observer, is developed in this paper. The stability of the adaptive control is an-alyzed using the second Lyapunov theory. A low-pass filter is added to remove the influence of the high frequency noise of joint velocities measurements. The experiments on single joint or multiple joints verify the advantage of this method on the dynamic compensation of friction.5) For the collision safety problem of manipulator with inertia parameter uncertainties, a new collision detection algorithm is presented. A disturbance observer which is deduced from preser-vation of momentum is designed to evaluate the influence of the collision force. The variation of the observed results is added into the collision detection thresholds. For the damage limiting of the robot or the safety of the human body, two different safe reaction methods for velocity mode and torque mode control are also presented. The experiments demonstrate that the collision could be detected and the safety of the manipulator and human are guaranteed with this method.