Research On The Application Of Deep Reinforcement Learning In The Assembly Of Low-voltage Electrical Appliances

This paper focus on the difficulties in the automated assembly process of small electrical appliances represented by low-voltage electrical apparatus,such as complex parts structure,fixed initial position of parts,fixed production lines,and poor flexibility,etc.We apply Deep Reinforcement Learning to the assembly research of low voltage electrical appliances.By using the good perception and decision-making ability of Deep Reinforcement Learning,we complete the automatic assembly of low-voltage electrical appliances represented by small circuit breakers,then verify it in simulation environment and real environment.and finally achieve good results.After investigating and summarizing the status of low voltage electrical apparatus assembly and visual assembly,the main works of this paper are as follows:1.This paper sets up the assembly system of the small circuit breaker.Then we establish the kinematics model of the iiwa robotic arm with seven degrees of freedom of the execution module and analyze the forward and inverse kinematics of the robotic arm;After that,we design the assembly process of the small circuit breaker and formulate the Visual Servo mode as the assembly Strategy.Finally,the motion process of the robotic arm is analyzed in the simulation software.2.This paper designs an assembly process evaluation system based on template matching and build a vision module based on Kinect v2.The acquired images are preprocessed before the assembly work.The template matching algorithm is used to determine whether the assembly is successful and the assembly evaluation system is designed according to the template matching.3.In this paper,we study the Deep Reinforcement Learning algorithm based on value function.Then we analyze the expression of state space and action space in assembly process,design the reward function according to the assembly evaluation system,and construct the Markov model of the assembly process of the upper and lower caps of small circuit breaker.Finally,we formulate the assembly action strategy and design the pseudo code of assembly algorithm based on Deep Q Network(DQN).4.The various modules of the assembly system are built under the Gazebo simulation platform,and the overall design and requirements of the simulation experiment were analyzed.The algorithm flow is designed in the simulation environment and implemented under the framework of Tensorflow algorithm.Through the training and testing of simulation and real experiments,it is shown that the proposed assembly algorithm can effectively solve the problem of automatic assembly of small circuit breakers.Finally,we summarized the work done and the research results in this paper,and analyzed the shortcomings in this paper and the problems that need need further solution.