Intelligent Control And Recognition Of Space Robot Capturing Non-cooperative Targets

Space robots play an irreplaceable role in space-on-orbit services.The increasingly complex space missions and the increasingly severe space environment.Human space teleoperation can no longer meet the requirements of future space missions.More importantly,space robots need their own intelligence to make decisions and actions independently in the face of changes in the external space environment.Therefore,unmanned intelligence is an important development direction of space robots in the future.In this paper,the space robot captures the non-cooperative target as the research object.According to the non-cooperative characteristics of the captured target,the artificial intelligence method is used as the technical background,and the capture control problem and target recognition in the process of capturing the non-cooperative target by the space robot are studied.The main research contents of this paper include the following aspects:(1)Taking the free-flying space robot as the research object,the multi-body dynamics is used to simplify its modeling,the Lagrange equation is used to obtain the dynamic equation of the space robot,and the collision effect in the process of catching the target spacecraft is studied.Finally,the dynamics of space robot system is simulated and analyzed.In the process of space robot motion,the manipulator and the base platform constitute a time-varying non-linear coupling dynamic system.(2)In the process of capturing non-cooperative targets by space robot,a dual-loop control method consisting of enhanced learning control and PD control is proposed to control the orbital attitude and manipulator motion of the space robot base platform.In the inner loop,the reinforcement learning is combined with the fuzzy theory to design the controller to control the end motion of the manipulator.In the outer loop,PD control is used to stably control the attitude orbit state of the base platform.Finally,the proposed control method is used for numerical simulation and compared with the traditional PD control method to verify the effectiveness of the control method.The results show that the motion of the manipulator under reinforcement learning control is stable and the control accuracy is high.The manipulator has a certain selflearning ability,and is more suitable for the non-cooperative characteristics of the capture target.(3)Using the deep learning method to image recognition of the non-cooperative target.A deep learning method is proposed for image recognition of non-cooperative targets.The non-cooperative characteristics of the target spacecraft make the manual identification difficult.In this paper,the YoloV3 network structure in deep learning is used to train the neural network of the non-cooperative target spacecraft mission,and the experimental verification of the target recognition is carried out.The experimental results show that the method of in-depth learning can effectively recognize the target spacecraft and is not restricted by the non-cooperative characteristics of the target spacecraft.It is insensitive to external information such as sunlight and has important research significance.