Research On Autonomous Vision Navigation Methods For Spacecraft Landing On Mars

Spacecraft autonomous landing capability, a critical prerequisite for Mars safelanding and sample return, has been considered as the important design goal for theMars exploration missions. Due to the large distance between the Mars and earth,the short landing time, and the complex and changeable Mars environment, newchallenges for spacecraft autonomous real-time landing navigation are proposed.After an in-depth investigation of the Mars probe descent and landing phase, thebasic theory and key technology involved in spacecraft pose estimation are studiedin this dissertation in order to develop the new generation autonomoushigh-precision landing navigation method with strong adaptive capacity to thechange of environment. The major research contents of this dissertation are given asfollows:The attitude determination methods based on line-of-sight vector observationsare studied. According to the choice of nonlinear vector measurement models, thenonlinearity of two measurement models were compared using the differentialgeometry curvature nonlinearity measure theory. Considering the determinationproblem using the landmark features measurement information, the attitudequaternion estimation algorithms are developed based on line-of-sight vectormeasurements. By applying the constrained extended Kalman filter algorithm to thepart constrained multiplicative attitude estimation, a multiplicative quaternionconstrained filter algorithm is first designed from the view of information fusion.Then a generalized optimization object function under quadratic constraints isdeveloped through borrowing the traditional QUEST batch algorithm framework,and an extended QUEST optimal attitude estimation algorithm is presented bysolving the eigenvalue-eigenvector problem.The absolute navigation method based on optical and inertial measuremeninformation fusion is studied by combination of high resolution3D feature pointslocation information from Mars orbiter. For the information fusion of sensors withdifferent measurement frequency, a state vector augmentation algorithm is presentedto deal with the measurement delay and the six degrees-of-freedom sensorscalibration parameters are also estimated in real-time. Then the observability of theintegrated navigation system is investigated theoretically by the observability matrixrand condition based on Lie derivatives, and the landing navigation algorithm’sperformance under various operational conditions are validated by simulation.The landing navigation method in unknown environment and limitedobservation condition is studied. By fusing relative motion measurement and inertial navigation, a landing navigation algorithm based on inter-frame homographytransformation is proposed. The homography matrix estimation is directly fed intothe filter as vision measurement to solve the poorly conditioned problem inhomography decomposition. Then the measurement model explicitly related to thestate of the lander is provided. The image mosaicking processes are used to enlargethe images overlap regions and therefore increase the available feature measurementinformation. In view of the strongly nonlinear and multiplicative noise problem oflanding navigation system, the correlation between random variables is eliminatedby decoupling the sample space.The landing navigation algorithms based on multi-frame feature tracking arestudied from the perspectives of batch processing and filtering. First of all, thenavigation algorithm based on simultaneous localization and mapping (SLAM) isimproved. The dimensions of the state vectors are adaptively regulated by thefixed-lag smoothing principle, and a landing navigation algorithm based onmulti-frame sliding windows optimization is presented. In order to realize thereal-time feedback control during the landing process, the navigation system andmap construction processes are decoupled by adding a delayed initializationestimation of the features position, and a landing navigation filter algorithm basedon multi-view geometry constraint is designed. A measurement model relating themulti-frame geometric constraint is proposed and the computational complexity ofthe navigation algorithm is further reduced with the null space projection method.