Research On Robotic Grasping Based On Visual Servo Contro

Robots have gradually become commodities.It is no longer an illusion to enter thousands of households.In the future,robots will provide services for human beings in various industries and jointly create a better future.In order to achieve this goal,a large number of researchers are working hard in this field and looking forward to a greater breakthrough.The random grab of the manipulator is one of the starting points.If you want a robot to be able to freely grasp any object you want to grasp like a human being,you can’t simply set up a fixed program to let the robot arm execute according to instructions.Only by sensing and analyzing the surrounding environment and making scientific decisions can you achieve intelligent grasping.At present,one of the most possible ways to realize the random grab task of the robot arm is to first plan a motion path with low cost and high safety factor for it,and then control it to accurately reach the grab position and pose to achieve the target grab.When there are obstacles in the environment,it must be ensured that the planned path can reasonably avoid the obstacles.This paper studies this task,aiming at solving the problem of how to control the mechanical manipulator to grasp the target object quickly and accurately with environmental obstacles.It mainly focuses on the planning of motion path and image-based visual servo control.The main research tasks of this paper are as follows:(1)The first part of grasping is the path planning of manipulator.An improved RRT * FN manipulator path planning algorithm is proposed to solve the real-time and path optimization problems of path planning in multiple complex scenes.Aiming at the problem that the existing RRT * FN planning takes a long time,a heuristic sampling strategy and a dichotomy based greedy expansion and quadratic expansion method are proposed to ensure that the algorithm can plan a lower cost collision free path while meeting the real-time performance;Considering that it is difficult to find a path in a narrow environment,this paper proposes a path planning method based on local environment sampling boundary expansion for complex narrow channel environment;In ROS,a general 6-DOF manipulator is used as the research object to carry out the simulation experiment,which proves the performance and effectiveness of the algorithm.(2)The second part of grasping is visual servo control manipulator.Aiming at the low servo accuracy,poor stability and lack of visibility constraints of visual servo control system,it is proposed that a multi-strategy fusion visual servo control method based on depth reinforcement learning adaptive gain.Firstly,a controller is designed to integrate the sliding mode control(SMC)and the classical proportional control image-based visual servo(C-IBVS)into the SMCC-IBVS controller;Then,in view of the limited field of view of the camera and the feature loss,and the current heuristic or empirical manual selection of servo gain is easy to cause poor robustness and slow convergence of the system,a method of adaptive adjustment of servo gain based on depth deterministic strategy gradient(DDPG)is proposed,and the DDPG adaptive servo gain algorithm is designed using servo simulation and physical data;Finally,the simulation and scene experiments on the manipulator show that the improved servo control method has the characteristics of no loss,higher accuracy and better stability under large displacement.(3)Finally,the grasping experiment is carried out in the real robotic arm scene.The path planning algorithm proposed in this paper and the visual servo control are combined to carry out grasping experiments in the obstacle environment.Firstly,the path planning algorithm proposed in this paper is called to guide the robotic arm to reach the location near the target object,and then the visual servo control method is called to control the robotic arm to accurately reach the grasping position,and then the grabbed object is placed in the specified position.Results of experiment reveal that the proposed method in this paper can be applied to the working scene of real robotic arm.