Truss Assembly Behavior Decision Of Space Cellular Robots Based On Reinforcement Learning

With the development of aerospace technology,some space systems tend to be large,and their scale often exceeds the launch capability of launch vehicles.The current expandable structure is also affected by the restraint of the launch vehicle and the load applied by the launching environment.Limiting the quality and size of the spacecraft system,system construction by means of on-orbit assembly is a good solution to this problem.Traditional space robots and robotic arms can perform specific on-orbit assembly tasks,but it is not possible to launch new assembly tools for each new task.Space Cell Robot is a multi-robot system consisting of multiple small robots with docking interfaces and relatively simple.It has many characteristics such as self-reconfiguration and intelligent self-adaptation.It has great flexibility and not only has the space machinery.The full functionality of the arm,and because it is self-reconfigurable,can accommodate a wide variety of space system in-orbit assembly tasks,and each cell replaceable can provide higher reliability for on-orbit assembly.The traditional assembly sequence planning mainly considers the order in which the components of the top layer are installed,and ignores the actions of the assembly robot in the middle layer.In the future,the assembly robot is becoming more and more intelligent,and its self-construction assembly sequence planning after considering its own behavior will become an important direction to complete the unknown assembly task.In this paper,the task of on-orbit assembly for space cell robots is to decompose the large truss assembly into the assembly process of the simplest structural unit and the transportation process of the handling robot.The two work in parallel,and then the two can be independently planned.The expression system of the whole assembly system is modeled,and the state parameters of the assembly object and the assembly robot are defined.The behavior of the assembly robot,ie the space cell robot,and the transfer relationship between state and behavior are defined.Finally,the behavioral decisionmaking research based on reinforcement learning is carried out.Through reinforcement learning,the cell robot can carry out the object planning and selection of the next assembly and the behavioral action planning choice for the next assembly state in the assembly task,with triangular prism truss and quadrangular pyramid truss.For example,the simulation verification of the simplest structural unit assembly planning is carried out,and the optimal assembly assembly sequence and the behavior sequence of the cellular robot are obtained.This research provides support for future intelligent on-orbit assembly tasks and has important theoretical and engineering significance.