Abstract – Second Version:
Today, using robots has grown dramatically in various applications. Robots are used in many different ways, such as manufacturing, logistics, medicine, home, education, defense, factory, etc. Thus, studying about robots will be very useful. One of their most important uses is using in factories to assemble some parts by screw. Robot arms usually are used to do it. Thus it will be important to recognize the screw holes and reach them automatically. Because sometimes there are a lot of screw holes what have a different position in a part. Random target reaching is a popular topic in robot arms among researchers. Usually, they use reinforcement learning (RL) algorithms by the help of vision methods to solve it; Such as Q-learning (QL), deep Q network (DQN), Deep Deterministic Policy Gradient (DDPG), etc. Although, these algorithms are rapid to solve the problem but they are so complex and applying of them in a factory may be so difficult. On the other hand these algorithm are not comfort. Actually, there is a trade-off between being rapid and being user friendly. Thus it may be better to use simple algorithms.
In this paper, we used a camera and image processing algorithms to find the position of the screw hole and we used the found position as an input of SARSA algorithm to find an appropriate angle of robot arm joints. SARASA found is one of the simple RL algorithms. The results show that the SARSA algorithm work as well as other algorithms mentioned above. So we can use this method in factories that need to assemble some parts by screw, Such as automobile maker companies, electrical devices companies, home appliances maker industries, etc. This simple algorithm help engineers to use it easily in factories.
Ali's Researches 