Research On Methods Of Autonomous Hazard Detection On Planetary Surface And Terrain Relative Navigation

Exploring activities on unfamiliar extraterrestrial objects can result in a huge scientific return,but it also challenges the navigation technology of deep space probes such as landers and planetary surface rovers.Widely existing rocks on Mars remain the major hazard for safe landing of Martian probes.Hence,precise rock-based hazard detection and avoidance(HDA)technique is the precondition for safe landing of the lander and follow-up exploration missions.Due to the limitation of communication delay,the aim of accurate soft landing requires that the spacecraft should have the ability to autonomously determine its position and attitude in real time.For the tasks such as planetary surface exploration,autonomously estimating the state of rover and sensing the dangerous terrain are the guarantee of improving task safety and efficiency,as well as reducing the requirement for ground support systems.Therefore,it is necessary to develop autonomous HDA techniques and autonomous navigation methods for spacecrafts and rovers in planetary surface applications.Combined with our China Mars exploration mission,this thesis presents the research on active and passive version-aided hazard detection and navigation techniques.The main contributions are as follows:Lidar-based hazard detection method is studied.In the 3D terrain data acquired by LIDAR,the key to detect topographic and undulating obstacles including rock is to eliminate the interference of undulating terrain.To handle the issue of fast and robust datum plane fitting,a morphological filter-based plane fitting method is proposed.In this new method,the point data are filtered with morphological operations first,which can eliminate most of the outliers including hazards.Secondly,a datum plane is generated by random sampling from the filtered point data.The procedure of morphological filtering sufficiently decreases the ratio of outliers,which can shorten the time cost of random sampling and improve the precision of plane fitting.Simulation results show that the new method works out efficiently.Image-based rock detection method is studied.Due to the fact that contrast of Mars surface is weak,a method based on multi-scale region contrast is proposed.New method deals with the detection in region level rather than pixel level,and enhances the contrast via calculating the feature difference of regions.Compared with edge-based detectors,the new method can significantly reduce the memory occupancy and improve the precision of rock detection simultaneously.The three-dimensional surface data of planet terrain obtained by laser radar or binocular vision can not only be used to detect and identify obstacles such as rock,but also determine the position and orientation of the detector by three-dimensional feature matching.Due to the objective conditions such as sensor technology and terrain occlusion,3D terrain data are usually sparse,which makes it difficult to extract and match terrain features.To solve this problem,a feature matching method based on the distance between features is proposed.Taking the distance between the features and the corresponding direction angle as the basic elements,the terrain feature description sets are constructed,and the feature similarity is used to determine whether the features match.Based on the feature matching method,an autonomous navigation method for terminal landing of the detector based on three-dimensional topography is established.There is a relatively large difference in scale and view between planetary surface images collected during the landing process and that in the navigation map database.This makes it difficult to match features of planetary surface images in the map database.In order to solve this problem,a visual navigation method based on assistant celestial 3D terrain is proposed.This method transforms the traditional two-dimensional navigation map carried by the spacecraft into the three-dimensional terrain in the landing area.During the landing process,the navigation system generates a navigation reference image online according to the predicted pose of the probe.That is,a navigation map,so as to reduce the scale difference between a planetary surface image and a navigation reference image.Through the template matching,the real position of the terrain feature in the collected planetary surface images during the landing is determined.The pixel coordinates of the terrain feature with known position are taken as the observations,and the navigation filter is designed to determine the state of spacecraft.Simultaneous Localization and Mapping(SLAM)in unknown environment is an effective way to solve this problem.But standard SLAM algrithnm is not effective enough to deal with accumulative error because there is not repetitive observation to navigation landmark.To solve this problem,an autonomous navigation algrithm with active repetitive observation in unknown environment is put forward.Landmark extracting and evaluation method based on local terrian siliency is designed.And trigger conditions for active repetitive observation is built.A method of selecting active repetitive observation path considering local silency,length of path and ratio of landmarks is given.Comparing to standard EKF-SLAM,active SLAM performs better in dealing with accumulative error.