Research On Intelligent Motion Planning Method Of Space Robot To Pursuit-capture Non-cooperative Target

Along with the increasing number of international and national space projects,the number of non-cooperative space objects is increasing rapidly.This not only leads to a great occupation of orbital resources,but also threatens the safety of spacecraft in service.To reduce the risk of noncooperative target,as so to ensure sustainable development and safety of the space,it is of great urgency to improve the capacity of the space object removal and space operations.A space robot,which is composed of a spacecraft platform and a multi-freedom manipulator,exhibits several advantages such as high dexterity,strong adaptability,and excellent operation capability.It can implement a series of complex tasks,including the approach,capture,and manipulation of space targets.Aiming at noncooperative targets,the difficulty is greatly increased for the safe and autonomous movement of the space robot,due to the non-cooperative characteristics of the target,as well as the high degree of freedom and strong coupling feature of the robot itself.To enhance the safety,flexibility and synergy of the space robot,it is of great practical significance and engineering importance to research the intelligent motion planning methods of the space robot for the pursuing and capturing of noncooperative targets.Relying on the Innovation Fund Project of China Academy of Space Technology "Research on motion planning of space cluster system for noncooperative target pursuit task",this paper studies the intelligent motion planning method of space robot to pursuit and capture non-cooperative target,and the details are as follows:Firstly,for non-cooperative space targets,the maneuver recognition method and motion prediction are studied.By analyzing the mapping rule between the changes of orbital elements and different types of maneuvers,the featuring functions for maneuvering recognition are constructed.Then,a classification model is built for the detection and classification of the maneuver of non-cooperative target.To improve the motion prediction accuracy of the target,the error between the prediction based on the basic dynamic model and the measurements is analyzed,followed by a probability distribution prediction law.By compensating the error,the prediction accuracy can be improved,and the potential maneuver behavior of the target can be warned,which can facilitate the robot motion planning.Next,aiming at the demand of autonomous motion planning of space robots in the short range,the target-robot relative motion planning strategy fusing prediction data is studied.Considering the influence of nonlinear term in the relative kinematics model and the motion performance limitation of the robot,a real-time motion planner is designed for robot to steadily approach the target.Considering the maneuvering behavior of the target and the nonnegligible time delay in the data processing process,a correction strategy is designed combing with the prediction data analysis.Simulations show that the motion planner can make the position and velocity of the space robot quickly converge to be consistent with the target,such that the position error tends to zero,and the relative velocity error to be within 10-3 m/s.Combined with the prediction data,a good compensation can be achieved to cancel the error caused by the potential maneuvers of the target and the time delay.Then,combining the structure and motion characteristics of the space robot and the mounted manipulator,the optimal capture region is determined according to the installation location,the accessible workspace and the dexterity.The safety of ultra-close motion is analyzed by analyzing the collision probability and the safety motion margin.Furthermore,considering the robot’s characteristics and the Deep Deterministic Policy Gradient method,the robot is divided into three agents,for which independent motion planners are designed.For the multi-agent collaborative decision model training with large dimensional space and time complexity,a training framework of "separate training-centralized training-distributed execution" is designed to accelerate the training.The simulation results show that the end of the manipulator can quickly approach the desired capture position,by applying the proposed multiagent collaborative decision model.The position error is within 3 mm,while the attitude error within 1°.Moreover,this method can be applied and easily adjusted in real-time.Finally,the co-simulation of pursuing and capturing non-cooperative target is carried out based on intelligent motion planning method for a space robot.The co-simulation platform is built,and an appropriate experimental scheme is designed.Experiments on non-cooperative target both without and with maneuverability are carried out respectively,the results prove not only the effectiveness of the proposed theoretical methods,but also their abilities to complement the whole pursuit-capture mission.