Research On Capture Control Strategy Of Space Manipulator Based On Reinforcement Learning

With the development of society,the number of spacecraft in orbit is increasing because of the demand for spacecraft.On-orbit service technology will fully demonstrate its military value and significant economic benefits.Capturing space target is a key technology of space manipulator to perform various on-orbit service tasks.However,the dynamic characteristics of space manipulator system are complex,and it is difficult to establish an accurate dynamic model,which makes it difficult for the control system to achieve better control effect.At present,with the continuous development of artificial intelligence technology,data-efficient intelligent control algorithm has been proposed one after another,which don’t need mathematical modeling of the system.Therefore,in this paper,the reinforcement learning method is applied to the acquisition process control of space manipulator,and the deep reinforcement learning acquisition control method and the model-based reinforcement learning control method are proposed respectively.The main work and conclusions are as follows:Firstly,the virtual simulation environment of space manipulator is established.Taking the acquisition of floating target by space manipulator as the research task,the simulation environment of space manipulator is established in V-rep simulation platform for reinforcement learning training and dynamic simulation analysis,and a traditional acquisition control method is designed and simulated for comparison with reinforcement learning based acquisition control method.Then,a deep reinforcement learning acquisition control method is proposed.The acquisition control method is designed based on the idea of "pre-training".The acquisition task of space target is achieved by training in virtual simulation environment.Compared with the acquisition control method without "pre-training",it is verified that "pre-training" can greatly improve the learning efficiency of training samples.Finally,a model-based reinforcement learning acquisition control method is proposed.Aiming at the problem of low learning efficiency of deep reinforcement learning samples,a acquisition control method based on PILCO algorithm is proposed.The capture control strategy can be obtained through a few rounds of training.The comparison between the acquisition control method based on reinforcement learning and the traditional acquisition control method is made,and the feasibility of the application of reinforcement learning in the acquisition process of space manipulator is demonstrated.The results show that the acquisition control method based on reinforcement learning can achieve the task of capturing space targets in virtual environment,which takes less time than the traditional acquisition control method.However,in the acquisition process,the smoothness of control moment and the accuracy of acquisition position can not reach the level of the traditional acquisition method.The training efficiency of model-based reinforcement learning acquisition control method is obviously better than that of deep reinforcement learning.The learning task can be completed in a few rounds.The effect of task completion is comparable to that of deep reinforcement learning acquisition control method.Therefore,model-based reinforcement learning acquisition control method has great application potential in the acquisition process of space manipulator.