Deep Learning Based Robotic Grasp Technology

Object pose estimation and robot manipulator path planning can make the robot flexibly complete the work of object grasping,parts assembly and parts processing,so it plays an important role in the field of robot grasping.Scenes such as cluttered objects,mutual occlusion of objects and complex background pose new challenges to existing pose estimation and path planning algorithms of robotic arms.The main problems solved in this paper are as follows.With the development of in-depth learning technology,neural network technology is more and more used in object pose estimation.However,in order to obtain high prediction ability of neural network,a large number of high-quality labeling data need to be used for training.In the field of robot grasping,there are few databases about object pose labeling.When a new object does not exist in the database,it needs to be labeled by human.This wastes a lot of time and the quality of labeling varies from person to person,which makes the robot unable to predict the new object's pose in a short time.On the basis of summarizing the above shortcomings,this paper proposes a network architecture of object pose estimation based on deep learning,which uses semi-supervised learning and domain adaptive method to train the network.It can not only improve the accuracy of pose estimation,but also cope with the problem of new objects without labeled dataAfter obtaining the pose of the object,we need to plan a route through the planning algorithm,and make the manipulator move along this route to make the gripper approach the object and grasp the object.This path can not collide with objects and the path is as smooth as possible.It also requires real-time and high success rate of the planning algorithm.This paper summarizes and compares the latest path planners.On this basis,a path planning algorithm based on deep learning is proposed.This algorithm can not only understand dynamic scenes in real time,but also determine the generation of motion paths according to experience.