| In the nuclear environment,as the premise of nuclear robot movement between regions,door opening plays an important role in the maintenance of nuclear facilities.At the same time,in the case of nuclear emergency,once the door cannot be opened and the source of the accident cannot be handled quickly,the loss caused by nuclear leakage is immeasurable,so the door opening method is very important for the daily maintenance of nuclear power plant and the emergency rescue under nuclear emergency.This subject is a part of the national 13 th five year plan nuclear energy development project,which has important research significance.Most of the nuclear robots performing tasks in the nuclear environment are teleoperated.Although the teleoperation method is more practical for some tasks with lower accuracy requirements,it can not complete the contact tasks with higher accuracy requirements such as screwing the valve and opening the door.Due to the influence of light,radiation and other interference conditions,the vision guided autonomous nuclear robot can not work stably in the nuclear emergency environment.In view of the above problems,this paper constructs a "teleoperation + task level autonomy" open door operation system framework.The framework combines the advantages of teleoperation and robot task level autonomous operation to complete the door opening task.First of all,this project deeply analyzes the characteristics of the nuclear robot’s door opening task,and designs a nuclear robot door opening system framework composed of three parts:the master-slave heterogeneous teleoperation method,the mathematical model of the manipulator,and the method of self-adaptive turning the door handle based on force sensing guidance.The workspace of master-slave manipulator is solved,and the mapping method of master-slave heterogeneous teleoperation is carefully studied and analyzed.According to the existing equipment,the teleoperation system is constructed.At the same time,the mathematical model of manipulator in the system is established as the theoretical support of the system framework.Then,an adaptive door handle rotation method based on force sense guidance is proposed,that is,the door handle rotation method based on l-ddpg.This method uses the dual network structure of ddpg algorithm and experience playback pool.In addition,LSTM network is added to the actor network of ddpg algorithm to prevent the algorithm from forgetting the previous experience knowledge in subsequent decision-making.At the same time,in the experience playback pool,LSTM network is added to the actor network of ddpg algorithm,In order to improve the convergence speed of the algorithm,the weight of sampling according to the size of reward is added.According to the force information,the state space and action space of the algorithm are designed.According to the geometric relationship between the end of the manipulator and the door handle,the reward reward function is designed.Finally,in the simulation and physical experiments,the proposed method is used to make the manipulator complete the door opening task,and the performance of the proposed method is evaluated.The experimental results show that the force on the end of the manipulator is small and can complete the task well.Compared with the traditional algorithm,the experimental results show that the algorithm in this paper plans the path in real time according to the change of the force on the end of the manipulator,so as to make the force on the end of the manipulator as small as possible.In addition,physical experiments also verify the migration ability and generalization of the algorithm. |
