Research On Space Intelligent Collaborative Capture Strategy Of Dual Robotic Arm Based On Reinforcement Learning

With the development of aerospace technology and the deepening of aerospace applications,spacecraft operating in orbit has experienced explosive growth,and there is a strong demand for space services.Invalid spacecraft need to be cleaned by on-orbit services.The failed spacecraft requires on-orbit service for maintenance.Normal spacecraft can be extended and upgraded by on-orbit service.In the initial stage,on-orbit service by astronauts has problems such as high risks and high costs,and it is difficult to adapt to the diversification and huge demand of on-orbit services.As a typical space robot,the space manipulator demonstrates powerful application capabilities and broad application space.Therefore,this paper takes the space manipulator as the object,and studies the cooperative motion planning of the dual robotic arm for the space target acquisition problem,focusing on the Rapidly Exploring Random Tree(RRT)and its improved algorithm.A two-arm cooperative motion planning algorithm based on Deep Deterministic Policy Gradient(DDPG)is proposed.The DDPG and RRT algorithms are compared and analyzed.The main work completed by the thesis and the conclusions obtained are as follows:Firstly,the kinematics model of the manipulator is established by D-H coordinate method.Each arm has a three-degree-of-freedom structure.The joint coordinate system between the links is established by D-H method.On the basis of this,the forward kinematics and inverse kinematics of the manipulator are analyzed respectively,and the kinematics modeling is completed.Secondly,a mathematical model of the cooperative motion task of the dual robotic arm is established.The mathematical model and evaluation criteria of the motion planning target are given.Based on the constraint analysis of the robotic arm motion,the axial Axis-aligned bounding box method is used to establish the obstacle model and the capsule body model of the robotic arm.An algorithm for detecting collisions between mechanical arms and obstacles and between arms using a hierarchical bounding box tree algorithm is given.Thirdly,the dual robotic arm motion planning algorithm based on RRT and its improved algorithm is analyzed.The RRT algorithm based on the fast random extension path of the sampling space is used to study the motion planning.Considering that the algorithm has a certain blindness and low efficiency,on the basis of this,two improved algorithms,the bidirectional RRT algorit hm and the RRTstar algorithm,are studied.The simulation and performance comparison of these three algorithms in the V-REP simulation environment.Finally,a dual robotic arm coordinated motion planning algorithm based on DDPG is proposed.The DDPG algorithm model of the dual robotic arm coordinated motion planning is given,and the network structure and parameter design are carried out.The V-REP simulation model is used to train the model to obtain the planning strategy of the task.The trained model is used to carry out the dual robotic arm motion planning experiment,and carry out a comparative analysis with the RRT algorithm.The results show that various algorithms can effectively complete the robotic arm motion planning task.Among them,the RRT algorithm has different planning results and low efficiency due to the expansion of randomness.The RRT-connect algorithm improves the greedy guided exploration and improves the planning efficiency,but still cannot guarantee the quality of the planning result s.The RRTstar algorithm adds the path optimization comparison step in the exploration process to make the planning result have a progressive optimality.However,the comparison of each expansion step increases the computational burden and reduces the plan ning efficiency.The DDPG algorithm gradually optimizes the planning strategy by continuously testing and arranging the model.The trained model can directly use the current strategy for motion planning,which can not only obtain an optimized path but also ensure the efficiency of the planning.