| With the rapid iterations of offshore oil extraction technology,underwater ROV robots are developing towards the direction of intelligence.However,the current ROV robot grasping operation is still at the stage of remote control operation,which is prone to collision damage and other problems.Although there has been work on robot arm grasping using binocular vision deep learning methods,not much research has been done to optimize the motion of the robot,so it is difficult to solve the effects of robot shaking and extreme environments during grasping.To address the current domestic and international problems regarding underwater robot grasping,this paper investigates a binocular vision-based underwater ROV robot identification and grasping method,and improves the non-stop maintenance simulation system.Firstly,for the problems of wobbling and deep sea environment influence encountered when the robot works underwater,this paper studies the high order sliding mode control thruster assignment method based on the invertible method.The power distribution matrix of eight propellers is solved by using the pseudo-inverse method,and the propeller operation parameters are optimized according to the matrix,which reduces the energy consumption when the propellers work and improves the travel speed of the ROV robot at the same time.When there is a harsh environment,such as the influence of ocean currents,the balance of the ROV robot is adjusted using the method of high-order sliding mode control to reduce the influence of the environment on the ROV robot’s travel route.At the same time,when one or more propellers are damaged,the method of calculating the distribution matrix is used to ensure that the ROV can complete the established tasks normally,to enhance its working toughness in harsh environments,and to provide the necessary conditions for the planar grasping position of the binocular vision robot arm.Second,for the problems faced by manual grasping such as high difficulty,high cost and low accuracy,this paper proposes a binocular vision grasping strategy based on deep learning.The URDF model of the new robotic arm is constructed by soildworks,and the Stereo SGBM binocular vision stereo matching network in Open CV technology is used to generate depth maps and depth sequences,and then the generated depth maps are used for the training work of robotic arm grasping.At the same time,the seven-dimensional grasping mode of the traditional robot arm is optimized to planar grasping by combining the underwater dynamics allocation method,which reduces the amount of judgment in two dimensions.Compared with the traditional method,this optimization method improves the speed of grasping training analysis by about 30%.The experimental results show that the binocular vision robotic arm based on deep learning can improve the accuracy of underwater grasping and reduce the labor cost.Finally,a complete ROV robot grasping system is designed by combining binocular vision and propeller power distribution methods.And the system is successfully applied to the simulation case of non-stop maintenance in underwater oil extraction,which provides a new idea of intelligent grasping in the field of underwater oil extraction. |
