Research On The Design Of Simulation Platform And Pose Estimation Algorithm In Asteroid Probe Landing Period

Asteroid detection is of great significance to the development of space resources,the exploration of solar system composition and the origin of life in the solar system..It has gradually become a research hotspot in various countries,which is also a significant component of China’s space strategy.In asteroid exploration missions,the probe is required to possess autonomous navigation capability due to the long distance,long delay of communication,and the complex and unknown asteroid environment.This means that the probe can complete necessary tasks such as attitude positioning and object tracking even when ground communication is completely interrupted.In this paper,the research of asteroid landing simulation platform and related experiment is carried out for the design of visual navigation scheme and autonomous pose estimation.The main content of the paper is as follows:Firstly,in order to simulate the landing process of probe,the simulator platform based on multi-rotor aircraft is proposed.The dynamics model of landing section of probe is established;the hardware and software design of the simulation platform were conceived.In the aspect of hardware,the system based on on-board computer is designed.In the aspect of software,a modular and distributed software system based on robot operating system(ROS)is built.Secondly,a visual-based navigation scheme is proposed aiming at the navigation difficulties in special environment during the landing process of simulation platform.In order to decrease the interference of horizontal displacement caused by asteroid rotation,two different artificial markers based on Ar Uco marker is proposed.Recognition algorithms of these two artificial markers were proposed and relative pose is estimated.Then the visual navigation scheme used in the autonomous landing simulation platform is selected by comparing and analyzing the performance of these two artificial markers in ranging experiment and occlusion experiment.Thirdly,a visual-inertial odometry(VIO)network model based on deep learning is proposed,which integrates visual information and IMU information,to solve the problem that traditional inertial navigation has strict requirements for camera calibration and synchronization in probe pose estimation.The public KITTI dataset is preprocessed and loss function is optimized for the training.Then the visual analysis of the model under different training epochs and comparison with the Deep VO method verifies feasibility of the proposed deep learning model.Finally,a prototype system of autonomous probe landing is built and experiments is carried out based on the research of multi-rotor UAV technology.In the probe autonomous landing system,a hardware platform is selected and built;the probe autonomous landing algorithm is designed,which is verified in the Gazebo simulation environment.Then the vision and IMU data collected in the Gazebo simulation environment is tested to verify the effectiveness of the VIO model.In the end,some experiments with static markers,linear motion markers and circular motion markers is carried out in a real environment to realize the autonomous landing of probe gradually,which verifies the stability of the platform built in this paper.