Control And Planning Of Bionic Arm Based On Pneumatic Artificial Muscle

By mimicking the structure and characteristics of biology which evolved from millions of years,bionic robots have obtained a series of good performances such as adaptability,robustness and multi-flexibility.The development of bionic robots and related control and intelligent sensing technologies,has important implications for promoting robotic intelligence and lightweight.Among them,bionic fingers and arms driven by pneumatic artificial muscles have been widely studied and paid attention by scholars because of their light weight,safety and suppleness,and biological bone-muscle structure.This thesis studies the control,planning and vision technology of a pneumatic artificial muscle-driven bionic arm.The main contents are as follows:Firstly,the problem of positional control of antagonistic joints driven by pneumatic artificial muscles is studied.Accuratejoint position control is the basis of bionic arm movement ability.A model-free adaptive controller based on compact format dynamic linearization and full format dynamic linearization is designed.Based on the hysteresis of pneumatic artificial muscles,an improved model-free adaptive controller for lag nonlinear discrete-time systems is designed.The results of position tracking experiments show that the improved model-free adaptive controller has better tracking performance than the PID controller.Secondly,aiming at the five-degree-of-freedom bionic arm studied in this thesis,a bionic arm kinematics model is established and the trajectory planning method is studied.Using the DH parameter method,the link coordinate system of the bionic arm is established,and the analytical solution of the bionic arm kinematics and inverse kinematics model is solved.The polynomial,T-shaped and S-shaped trajectory planning methods are studied,and the trajectory planning of the bionic arm in the plane and Cartesian space is designed.The experimental results of the bionic arm end trajectory show that the three trajectory planning methods can give smooth,continuous and suitable for the trajectory of the bionic arm.Thirdly,in this thesis,the RGB-D camera is used as an external sensor to study the object pose estimation method in the grab task,and add visual perception ability to the bionic arm.The object pose estimation network model DenseFusion is studied.For the preprocessing and color feature extraction modules,an improved method of network structure merging and simplification is proposed to improve the running speed of the model.The experimental data shows that the improved DenseFusion model has an accuracy of 89.3%,and the speed is increased by 100%when the accuracy is only 4%lower than DenseFusion.Finally,based on the above three aspects of study,this thesis builds an algorithm verification platform for bionic arm system and RGB-D camera,and carries out visual servo grasp experiment.The experimental results verify the effectiveness and practicability of this thesis.