Research On Operation Path Planning Of Flange Flexible Assembly Robot Based On Vision

With the continuous development and improvement of industrial robots,they have been widely used in industrial production.Assembly robots have been widely used in industrial manufacturing,mainly for the assembly tasks of electrical appliances,automobiles,electronics,other products and their components.Assembly robots have also changed from the early program-controlled mode to teaching mode,and then to the intelligent mode.The precision and intelligence of assembly robots were being higher and higher.Target recognition and path planning were also two of the important capabilities of assembly robots,which determined their degree of flexibility and intelligence.Deep learning had achieved great success in the field of machine vision.Using deep learning method instead of traditional control method in the target feature recognition and path planning of assembly robots can significantly improve their flexibility and intelligence.Therefore,For specific assembly tasks,the integrated application of deep learning can improve the generalization ability of target feature recognition and path planning,and also compensated the defects of traditional methods.This paper established the assembly system framework for the flange assembly task and built the simulation environment.This paper studied the principle of camera calibration and the principle of robot hand-eye calibration.Based on Coppelia Sim,the camera,robot,and system structure were simulated and modeled.A virtual environment for assembly robot simulation was built,and the control method for controlling Coppelia Sim simulation with external applications was studied.A feature recognition method based on SSD algorithm was applied to the flange hole positioning recognition,and the method of contour recognition of the flange cover based on the depth information map was completed.A flange assembly feature recognition network based on SSD target recognition algorithm was established.A series of fixed-size bounding boxes were generated,and the categories of parts contained in the bounding boxes were scored.Finally,the final prediction was generated by the non-maximum suppression link,and its results were compared with results of the traditional flange recognition method.Flange cover contour recognition based on depth map was realized and verified in the simulation environment.A robot operation path planning method based on deep reinforcement learning was designed,and the operation path planning and assembly process were simulated and verified in the simulation environment.A network model of robot operation path planning based on deep reinforcement learning was established.The thing that robot reaches the target position was the task of the Markov decision process.State space,action space and reward function were designed reasonably.The multi-layer perceptron network model was used to approximate the action-value function in Deep Q learning,and the best moving action was selected by maximizing the Q value.Then the simulation verification of the flange assembly robot system was carried out in Coppelia Sim.