Research Of Compliance Control Methods Oriented To Uncertainty

Robot is a leading subject which develops rapidly, its core is the control system. In recent decades, control problem of robot has being researched by many studies in the world, especially robot compliance control which attracted the attention of many scholars and experts. Many scholars constantly using the new control theory and methods, from different angles, trying to have progress in theory and practical application.However, robot is highly nonlinear, and have many strong coupling, and contains many uncertainty factors. Besides, when the end of the robot actuator contact with the external environment, the difference of environment’s location and stiffness also have great influence on control performance. Therefore, application range of robots and is greatly restricted.Based on such problems, the thesis studies on robot compliance control, focuses on solving uncertainties of robot model as well as the environment.The thesis firstly introduced the basic strategies and there characteristics of compliance control, then choose impedance control through analysis. Secondly, it introduced the kinematics and dynamics model and the simulation model of robot contact with the environment is established. Again, under the idea of impedance control it derived a impedance algorithm based on torque and initial adjustment principles of impedance parameters are obtained through the simulation and analysis. Then, to cope with the uncertainties of constraint environment, it used a trajectory correction module to adjust the original reference trajectory based on real time force error. Besides, to cope with the uncertainty of robot model, it used a estimation of inertia matrix and delay module which improves the system performance greatly. However, the algorithm above requires very small sampling time. Finally, based on the algorithm, in view of the robot model uncertainty, it designed a RBF neural network to compensate uncertain items which can achieve good tracking results when sampling time not so small.