Research On Intelligent Grasping Technology Based On Deep Reinforcement Learning

With the development of robotics,more and more researchers are engaged in the research on the application of robots.As a result,robots are being more and more widely used in the home.Domestic robot can not only take care of and accompany the lonely elderly and leftbehind children,but also effectively help the disabled.For some of the elderly whose memories may fade,and they may forget necessary behaviors such as drinking water and taking medicine.Domestic robot,however,can help them by handing over cups of water and medicine boxes for them at set times,while increasing their joy in life and help regulate their lives.The application and research background of this paper based on the preset scene,which robots delivering objects to users at home,how a robot can successfully grab the desired target object in a complex home environment.First,it is necessary for the robot to detect multiple objects from a complex background and select the target object to obtain relevant information;then the robot arm need to be controlled to reach at the target location,but it' not easy to control the robot arm with different degrees of freedom.Continuous guidance and training are needed before it can grab objects along the appropriate trajectory and avoid collision with things outside the target object.The main work we did in this paper is to verify the effect of deep reinforcement learning algorithm in the control of the robotic arm,so that the robot can easily grasp the required objects like a human.The final experimental results show that the processing methods and improved methods proposed in this paper are effective in accomplishing tasks,are robust to the extra complexity of actual robot operations,the task of grabbing target object can be done with high accuracy and take shorter time.