Research On Robot Screwing Skill Method Based On Demonstration Learning

With the advancement of manufacturing industry,assembly automation can greatly improve industrial production efficiency and workpiece quality,reduce the labor intensity of workers and improve the intelligent level of factory operation.Almost all device assembly will involve screw assembly.At present,robots are mainly pre-programmed to complete repetitive and labor-intensive tasks.When a task is changed,it results in a low utilization rate of previous operating experience and a high cost of relocation time.Learning from demonstration(LFD)can promote the learning and generalization of robot screwing skills,extract features and model the robot screwing task,and flexibly apply to more scenes and screwing objects.Therefore,the combination of skill learning and robot screwing is the key to unleash the potential of robot screwing.The characteristics of extensive screwing assembly tasks,complex process and various models will lead to the difficulty of robot screwing operation and poor environmental adaptability.On the basis of analyzing the characteristics of robot screwing and assembly,the method of robot screwing skills based on teaching and learning has been mainly studied.Combined with the experience of human operation,the learning and generalization model of robot screwing skills has been established,and a physical experiment platform has been built to verify the algorithm.The main research work is as follows:(1)Analysis and modeling of robot screw assembly process.The screwing process and characteristics were analyzed,and the contact model between bolt and threaded hole during robot screw assembly was established,so as to obtain the maximum allowable attitude deviation in the contact process.On this basis,the force information in the screwing process of multiple types of bolts was collected,and the force information in the screwing process of the same bolt before and after screwing and in the screwing process of different bolts of the same model were compared,so as to lay a foundation for the modeling and characterization of the screwing skills of the robot.(2)A robot hole-finding strategy based on dynamic movement primitives was established.In view of the problems of task point switching and obstacles in the operating environment and poor adaptability of robot operation in the process of screw hole finding,the Dynamic Movement Primitive(DMP)learning model was studied,the potential field function of obstacles was added,and the robot hole finding strategy with minimum error was established.When encountering obstacles,the obstacle avoidance curve adapting to obstacles can be automatically planned without additional calculation.The results of experiment show that the strategy is effective in complex environment.(3)A screwing skill learning and generalization method based on Gaussian Mixture Model-Gaussian Mixture Regression(GMM-GMR)was proposed.The demonstration information in the robot screwing process was collected for many times,and the Dynamic Time Warping(DTW)was used for data alignment.Then the screwing features were extracted by GMM method,and GMR regression fitting was used to screen out the smooth screwing feature curve.The generalization of screwing skills can be realized by changing the starting point and end point.The results of experiment show that this method can be effectively used in robot screwing task and generalization.(4)Robot screwing platform construction and skill generalization verification.The screwing platform based on KUKA iiwa 7 R800 robot was built.Through the design of robot bolt tendency and screwing,plastic bottle cap tendency and obstacle avoidance tendency to screw faucet experiments,the effectiveness of the robot screwing method based on teaching learning has been verified and generalized to different screwing tasks in a variety of scenes.Finally,the work of this paper and the results and experience were summarized.The screwing skill method based on teaching learning studied in this paper has important research significance for improving the flexibility and generalization of robot performing assembly tasks.