Abstract:In view of the shortcomings of current mechanical arm,such as fixed grabing and laying methods, single instruction, difficult to deal with complicated unknown situation,a mechanical arm grasp control method based on deep reinforcement learning and RRT was proposed.In this method, the objects grabing and laying problem was taken as a Markov process, autonomous grasp for unknown object was realized through objects field elements description and improved deep reinforcement learning algorithm (Dueling Network), objects were placed at target location accurately through the key points and RRT algorithm based on task requirements. The experimental results show that this method is simple and effective, and the robotic arm is flexible in grasping and placing independently, which can further improve the autonomous control ability of the robotic arm against unknown objects and meet the requirements of different objects grasping and laying tasks.