Robot Grasping And Assembly Research Based On Demonstration Teaching And Autonomous Learning

Industrial robots have the disadvantages of low programming efficiency,low intelligence and poor human-robot interaction when performing grasping and assembly operations.If deep learning and reinforcement learning methods can be integrated with industrial robots,robots can learn operational tasks from the demonstration teaching and learn relevant skills autonomously,which can significantly improve robot programming efficiency and intelligence.To this end,this paper investigates the demonstration and teaching,autonomous learning techniques for robot grasping and assembly,as follows:(1)A vision-based robot grasping and assembly demonstration of the teaching system is developed.The system consists of a target recognition and center point positioning module,a demonstration action recognition module and a robot action generation module.In order to accurately identify and locate target objects,a robot grasping method with target object centroid localization is proposed in the designed target recognition and center point positioning module,which uses an instance segmentation algorithm to identify object classes and calculates object 3D pose information by mask homogenization processing and coordinate transformation.In order to accurately identify the behavior information appearing in the action video,an action classification recognition model based on deep learning algorithm is established in the demonstration action recognition module,which has the input of assembled action video frames and the output of action classification labels.The abnormal action classification labels are cleaned and sorted by the proposed data cleaning filtering algorithm to obtain the desired action classification labels.The robot action generation module plans robot gripping and assembly action based on the information of object category,object 3D pose and action classification labels,and controls the robot to perform the gripping and assembly task.An experimental study was carried out on the example of shaft-hole assembly to verify the effectiveness of the above method.The system has a reference value for robot demonstration and teaching research.(2)A robot autonomous grasping and assembly skill learning method based on deep reinforcement learning is proposed,which includes a grasping and assembly prior knowledge information extraction module,a robot grasping deep reinforcement learning network module,and a robot assembly deep reinforcement learning network module.Grasping and assembly a priori knowledge information extraction module extracts the center point of the part model to be grasped,the centroid and rotation angle prior knowledge information of the part model to be assembled.In a robot grasping deep reinforcement learning network module,in order to obtain the optimal robot action in the grasping direction,the action strategy is optimized by adding the centroid of the part model to be grasped and the probability information of the part to be grasped itself.In order to increase the success rate of grasping,the module adds the grasping judgment threshold and the assembly judgment threshold for the prediction of the grasping situation.In a robot assembly deep reinforcement learning network module,the centroid and rotation angle of the part to be assembled as a priori knowledge information are input to the PPO network.The aim is to reduce the training time and interaction data required by the policy learning algorithm,while improving the assembly success rate and generalization performance.In order to verify the effectiveness of the above method,experiments were conducted on the assembly of a square shaft and a square hole,and the experimental results showed that the method effectively improved the success rate and accuracy of autonomous robot grasping and assembly.In this paper,we build a vision-based robot grasping and assembly demonstration teaching system based on demonstration teaching and autonomous learning of robot grasping and assembly process,and study the robot autonomous grasping and assembly skills method based on deep reinforcement learning.For the realization of robot demonstration teaching and independent learning,which is meaningful to promote intelligent manufacturing.