Research On Hand-eye Cooperative Grasping Control Of Explosion-proof Robot

Explosion-proof robots are robots that operate in hazardous environments.They can replace manual labor for dangerous operations in industries such as petroleum,chemicals,explosive ordnance disposal,and demining,improving production efficiency and ensuring personnel safety.Grasping is one of the main forms of operation for explosion-proof robots,but at the current stage of development of explosion-proof robots,there are issues such as low efficiency and difficulty in effectively controlling grasping force due to the low intelligence of explosion-proof robots in grasping and dependence on operator experience.This thesis focuses on the intelligent grasping of targets in dangerous environments,combining robot vision-guided grasping technology and mechanical hand grasping control technology.We have developed an intelligent hand-eye coordination grasping system.The main research work of this thesis includes:1)A blast-proof robot motion system used in this thesis was constructed,including the selection of hardware,software design,and information transmission between software and hardware.A hand-eye coordinated motion control system composed of a computer,a four-degree-of-freedom robotic arm,an Astra Pro depth camera,and a two-fingered robotic hand was determined.Kinematic modeling of the four-degree-of-freedom robot used in this thesis was conducted to clarify the principles of robot motion.Through camera calibration and hand eye calibration experiments,the conversion relationships of various coordinates in the robotic arm,camera,and robot system were obtained,providing a hardware platform for the implementation of attitude estimation and force control algorithms.2)In response to the low efficiency problem of intelligent grasping in explosion-proof robots,a grasp pose estimation algorithm based on deep learning was constructed by analyzing the grasping tasks of explosion-proof robots.To address the issue of simplicity and limited accuracy in the GGCNN2 model,the Inception feature extraction module multi-branch convolution structure to optimize the structure of GGCNN2 network.The input channels of the network were increased after optimization,allowing the network to be trained using RGB-D data.Through experiments,the average accuracy of the model was obtained.By conducting comparative experiments,the impact of the Inception feature extraction module and multiple input channels on the average accuracy of the network was discussed.The analysis of the results shows that the network trained through the RGB-D channels has an average accuracy improvement of 8.47% compared to the original network trained through the D channel.3)To address the problem of difficult control of gripping force during the grabbing process of explosion-proof robots,a gripping force control system model was constructed.The MATLAB system identification toolbox was used to obtain the transfer function of the constructed system.The particle swarm optimization fuzzy PID self-tuning control algorithm was introduced into the gripping force control field.This algorithm combines particle swarm optimization algorithm,fuzzy control theory,and PID control algorithm,optimized the cumbersome parameter tuning process of simply using PID algorithm,improving the robustness of the system.The system model was combined with the algorithm in Simulink for modeling,and disturbance-free simulation and disturbance simulation were performed in a simulation environment to verify the effectiveness of the gripping force control algorithm.4)An experimental platform was built to conduct functional tests on the hand-eye coordination grasping control system module of the explosion-proof robot.This includes experiments on grasping pose estimation,physical grasping experiments in a static environment,and dynamic environment grasping experiments using the upper computer interface developed in this thesis to communicate with the explosion-proof robot.Analysis of the experimental results revealed no significant deformation of the grasped object,and the gripping force could be adjusted in real-time during the gripping process,while eliminating the effects of joint vibration and inertia caused by acceleration,maintaining the stability of the grasp,and verifying the effectiveness of the posture estimation algorithm and the robot gripping force control algorithm in hand-eye coordination grasping.