Research On Visual Grasping System For Humanoid Arm By Artificial Muscles

With the transformation of social production patterns and the development of science and technology,essential safety and human-machine coexistence have become the developing trend of the new generation of manipulator.Essential safety means lightweight,slow and compliant manipulator ontology.Rigid joint driver based on the traditional motor and hydraulic has disadvantages of low power density,heavy weight,poor compliance and safety,etc.But compliant joint driver based on artificial muscle has advantages of high power density,light weight,good compliance and safety,and effective simulating the contraction movement of human skeletal muscles,etc.Human-machine coexistence and cooperation mean that manipulator performs tasks autonomously such as grabbing and handling in the same workspace with people,and ensures people’s safety.In order to develop a new generation of manipulator,this thesis develops a low-cost,compliant,safe and light hybrid drive humanoid arm based on compliant artificial muscle actuators by pneumatic artificial muscle(PAM)and shape memory alloy(SMA),and builds a verification platform for visual grasping system based on the camera and humanoid arm.This thesis focuses on the execution,planning,perception,and security of the system.Execution layer includes humanoid arm system design.Firstly,motion data of human joints which are obtained through the Vicon motion capture system,combined with the parameters of the human arm are used to design humanoid arm mechanical system.Secondly,execution layer designs motor-driven shoulder joint control system,PAM-driven elbow joint control system,SMA-driven wrist joint and bionic hand control system for the humanoid arm platform.Planning layer includes humanoid arm kinematics and trajectory planning.Firstly,the kinematic model of humanoid arm is established by MD-H parameter method.Secondly,planning layer analyzes the robot motion planning algorithm in dynamic unstructured environment,and implements the commonly used joint space and cartesian space trajectory generation algorithm for the current space environment of humanoid arm.Perception layer includes recognition and positioning of target object.Firstly,perception layer analyzes the camera imaging model,implements camera calibration by comparing related algorithms,and gets the conversion matrix between the camera and the humanoid arm by simplifing the hand-eye calibration algorithm.Secondly,in order to identify the grab,perception layer implements image denoising,edge detection and feature extraction on the target scene image.To summarize the above three aspects of work,this thesis designs comprehensive test experiments for visual grasping system for humanoid arm by artificial muscles.Firstly,humanoid arm grabs general objects using planning-executing control methods,and ball based on visual guidance using perception-plan-execute control methods.Secondly,in order to improve system security,this thesis designs the admittance controller based on the PID position control,and performs the compliant control experiment on the humanoid arm.