Target Attitude Estimation And Tracking Control In Non-cooperative Rendezvous And Docking Tasks

With the development of space technology,spacecraft rendezvous and docking technology have been widely used in on-orbit monitoring and services such as space station berth,on-orbit replenishment,and space capture.Compared with cooperative situations,the rendezvous and docking technology of non-cooperative targets has broader application prospects.Still,there are also more significant challenges due to the inability to provide information communication actively.The critical link to realize the rendezvous and docking technology of non-cooperative targets is attitude estimation and tracking control.Among them,most of the attitude estimation methods currently used are based on artificially designed a priori features,which have the disadvantages of high task cost,low adaptability,and poor robustness,and the control methods generally used in the process of performing tracking tasks do not consider execution practical physical limitations of the device and issues of timeliness.Therefore,this paper focuses on the target pose estimation and tracking control methods for non-cooperative target rendezvous and docking tasks.Aiming at the problem of low detection accuracy of small and blurred targets in complex backgrounds,a non-cooperative target detection method HRF-Faster RCNN is proposed that combines STAGAN-L and HRF-Faster R-CNN.First,a sample expansion method based on generative adversarial networks is designed to provide sufficient data sets for target detection networks to supplement training.Then,after improving the target detection network by using the design idea of the hierarchical residual network structure and combining the feature pyramid network structure,the target detection work can be completed,and then the target that needs to be estimated for the pose is determined.The simulation results show that the target detection method designed in this paper improves the accuracy rate of space target detection to 96.8%,among which the detection accuracy rate of small targets and fuzzy targets increases to 92.3%.Aiming at the lack of adaptability and robustness of pose estimation methods,a pose estimation method UrsoNet-Improve based on the UrsoNet algorithm,is designed.UrsoNet-Improve obtains the surrounding information of the target through the adaptive dual-channel feature extraction module to improve the adaptability to the environment.The convolutional attention mechanism module enables the network to learn features more accurately in the case of complex backgrounds,thereby obtaining precise pose information.Finally,the corresponding simulation experiments are carried out,and the results show that UrsoNet-Improve improves the performance of UrsoNet by 18.1% compared with UrsoNet in the public data set.In order to realize the fast-tracking of non-cooperative target attitude estimation results,the controller is designed using backstepping technology,adaptive control technology,and finite-time control theory.On the one hand,a robust adaptive control scheme is proposed based on finite-time control theory,and an appropriate adaptive update rate is derived from dealing with the spacecraft attitude system’s unknown parameters so that the parameter uncertainty synchronization error system converges in a finite time at zero.On the other hand,in the presence of actuator constraints,control laws are designed that take into account input saturation.The controller’s performance is verified by numerical simulation of the motion equations of the nonlinear control system.The results show that the designed controller can achieve the finite time convergence of the closed-loop system within the 20 s and avoid external disturbances when the actuator’s output is limited.