서지주요정보
Robust multisensor fusion for autonomous spacecraft navigation during final approach for mars entry, descent, and landing = 화성 진입, 하강 및 착륙 과정 중 최종 접근 단계에서의 자율 우주선 항법을 위한 강건한 다중센서 융합
서명 / 저자 Robust multisensor fusion for autonomous spacecraft navigation during final approach for mars entry, descent, and landing = 화성 진입, 하강 및 착륙 과정 중 최종 접근 단계에서의 자율 우주선 항법을 위한 강건한 다중센서 융합 / Natnael Shewangizaw Zewge.
발행사항 [대전 : 한국과학기술원, 2024].
Online Access 원문보기 원문인쇄

소장정보

등록번호

8043056

소장위치/청구기호

학술문화관(도서관)2층 학위논문

DAE 24017

휴대폰 전송

도서상태

이용가능(대출불가)

사유안내

반납예정일

리뷰정보

초록정보

An integrated navigation scheme for autonomous spacecraft operation during the final approach phase of a Mars entry, descent, and landing (EDL) mission is considered. Optical camera-based observations of natural celestial bodies are aided by X-ray pulsar signal time of arrival (TOA) measurements to provide accurate three-dimensional position and velocity solutions. Once the proposed navigation system is initialized, it operates with no need for assistance from man-made assets on the ground or in orbit, providing complete autonomy. As a reference for the research community, a catalog of 41 X-ray pulsars that are deemed most promising for navigation purposes is presented. The list builds on top of the work of previous researchers and contains the expected range measurement error margins from each source together with the relevant parameters used to compute these error margins. This catalog is utilized in an optimization-based technique for selecting the best $j$ pulsars for navigation from among a candidate list of $k$ pulsars wherein, geometry and measurement accuracy are jointly considered. Afterward, measurements from the optical sensors and data from the chosen pulsars are fused in a distributionally robust nonlinear optimization-based filtering framework. The proposed robust navigation filter provides significant resilience under severe measurement outliers which cause non-Gaussian noise behavior that subsequently results in the degradation of navigation performance. The results of the pulsar selection algorithm as well as the robust sensor fusion scheme are illustrated on a simulated final approach trajectory of the Mars Science Laboratory (MSL). The pulsar selection algorithm is further deployed on an embedded platform to demonstrate potential onboard implementation. A new robust approach to image processing for planet/moon centroid and apparent diameter extraction is also provided with actual mission images used to validate the concept. Although this research takes Mars landing as a case study, the methods developed herewith can be used for other interplanetary missions, asteroid landings, and as an aid to spacecraft rendezvous and docking applications.

본 논문에서는 화성 진입, 하강 및 착륙(EDL) 임무 중 최종 접근 단계에서의 자율 우주선 운항을 위한 통합 항법 방안을 제안한다. 광학 카메라 기반의 천체 관측과 엑스선 펄서 신호의 도달 시간 측정을 통해 우주선의 정확한 3차원 위치와 속도 결과를 제공한다. 또한, 제안된 항법 시스템을 사용한다면 지상이나 궤도로부터의 인공적인 도움 없이 완전한 자율성을 확보한다. 참고 자료에 따르면, 41개의 엑스선 펄서 목록이 항법 목적으로 유망한 것으로 간주된다. 목록은 이전 연구자의 연구를 기반으로 만들어졌으며 각각에 대한 예상 측정오차 마진과 관련 매개변수를 포함한다. 목록에서는 지형과 측정 정확도를 고려한 최적화 기법을 통해 후보 펄서 중 항법을 위한 최선의 펄서를 찾아낸다. 이후, 광학 센서의 측정값과 선택된 펄서의 데이터를 결합하여 분산적으로 강건한 비선형 최적화 기반 필터를 구성한다. 이렇게 구성된 강건한 항법 필터는 가우시안이 아닌 노이즈를 동반하는 심각한 측정 이상에 대해 강력한 복원력을 제공한다. 본 논문에서는 펄서를 선택하는 알고리즘과 강건한 센서 융합 과정을 화성 과학 실험실(MSL)의 화성 최종 접근 경로에 적용하였다. 또한, 제안된 강건한 필터링 기술을 화성 EDL임무의 다른 단계로 확장하여 수행하였다. 행성/달 중심 및 겉보기 직경 추출 이미지 처리에 대한 새로운 강건한 접근 방식도 미션 이미지와 함께 제공하였다. 본 연구에서는 화성 착륙만을 고려하였지만 행성간 임무, 소행성 착륙, 우주선의 랑데부 및 도킹에도 사용할 수 있을 것으로 기대된다.

서지기타정보

서지기타정보
청구기호 {DAE 24017
형태사항 x, 120 p. : 삽도 ; 30 cm
언어 영어
일반주기 저자명의 한글표기 : 저우게 나다니엘
지도교수의 영문표기 : Hyochoong Bang
지도교수의 한글표기 : 방효충
수록잡지명 : "A Distributionally Robust Fusion Framework for Autonomous Multisensor Spacecraft Navigation during Entry Phase of Mars Entry, Descent, and Landing". Remote Sensing, v.15.no.4, pp. 1139(2023)
Including appendix
학위논문 학위논문(박사) - 한국과학기술원 : 항공우주공학과,
서지주기 References : p. 110-116
주제 Final approach navigation towards Mars
Entry descent
Entry descent, and landing (EDL)
Optical navigation (OPNAV)
X-ray pulsar navigation (XNAV)
Integrated navigation
Pulsar selection
Robust filtering
화성 최종 접근 항법
진입
하강 및 착륙(EDL)
광학 항법(OPNAV)
엑스선 펄서 항법(XNAV)
통합 항법
볼록 최적화
펄서 선택
강건한 필터링
QR CODE

책소개

전체보기

목차

전체보기

이 주제의 인기대출도서

SPIRE inertial navigation system (Image credit: DRAPER)

Concept design for a navigation fix for a Mars probe mission through Earth to Venus, to Mars, and back to Earth[5]

Chronology of Mars lander missions

Startship spacecraft and launch vehicle from SpaceX (Image credit: SpaceX

Entry, descent, and landing sequence for the Mars 2020 mission (Image credit: NASA/JPL Caltech)

Uncertaintly ellipses for the most notable Mars missions (Image credit: NASA/JPL

The Crab Nebula and Crab pulsar (in X-ray spectrum) observed by Chandra X-ray observatory. (Image credit: NASA/CXC/SAO)

Magnetic environment ofa neutron star (lighthouse model). (Image credit: (ESA medialab / XMM-Newton / ASTRON-LOFAR))

LN-200, used in a variety of Mars missions (Image credit: Northrop Grumman).

Inertial navigation system computation. The steps shown are for inertial frame mecha- nization. bis the body frame, i is the inertial frame. f and w represent specific force and angular rate respectively. C8 is the transformation matrix from frame a to 3. Y, r, and U are gravitation model. position, and velocity respectively.

The 3D point with coordinates (in the camera frame) of (X,Y,Z) gets projected onto the image place with corresponding coordinates (x,71,)

Illustration of the epoch folding procedure.

Relevant parameters for Mars final approach dynamics model.

Extraction of planet horizon and centroid from images of the MarCO-B Cubesat as it approached Mars just before landing of InSight spacecraft.

Extraction of planet horizon and centroid from images of the MarCO-B Cubesat as it approached Mars just after landing of Insight spacecraft.

Configuration for apparent diameter measurement.

Configuration for relative bearing measurements of multiple stars from planet center (exaggerated scales).

Angle measuring equipment and readout technique for Apollo navigation

Illustrations of the navigation functions of the Apollo lunar mission.

X-ray pulsar navigation system measurement configuration.

Signal delay caused by space-time curvature (Shapiro delay) [64]

Distribution ofX-ray pulsars in galactic coordinate system.

Distribution ofX-ray pulsars in equatorial coordinate system.

Right ascension and declination (with ID labels from catalog in 2.4

The cones over which SDP and SOCP problems are defined

Contours ofthe Gaussian densityinR2. Thesquare root ofthe eigenvalues VX1 and VXi,of Ce are thelength ofthesemi-major axes oftheellipses measured alongthe axis U1 correspondingtok =1 and k=2. Vx and V길 correspond to the axis u2. The ui are obtained via the eigendecomposition 1 CH1=Cr- x, ujuT.

Scalar metrics for optimal experiment design.

Absolute estimation error with no outlier contamination (6 = 0)

Absolute estimation errors with 5% outlier contamination (E = 0.05)

Absolute estimation error with 20% outlier contamination (E = 0.20).

Absolute estimation error with 40% outlier contamination (6 = 0.40).

The space of distributions considered for nonparametric, parametric, and robust statistics approaches.

Huber's robust loss function and its derivatives.

(2-based and M-estimation-based penalties and their applications to a simple data fitting problem.

Heavy-tailed distributions in contrast to the Normal distribution (we note the higher value of the density at the tails for Laplace, Cauchy, and E-contaminated Gaussian distributions).

Mars Science Laboratory (MSL cruise configuration: Image Credit: NASA/JPL).

Raspberry Pi 4 Model-B for testing SOCP-based pulsar selection.

Specifications of embedded platform for running SOCP-based pulsar selection.

Further lower power alteratives for testing SOCP selection.

Specifications of embedded platform for running SOCP-based pulsar selection (for contrast, the IBM PowerPC 750FX onboard Orion runs at 900MHz).

Selected navigation stars

Assumed initial conditions for simulation.

Approach trajectory for the Mars Science Laboratory (MSL) from 12 hours before entry up to near entry interface.

View of approach trajectory looking into the positive y-axis

Selection results for direct SDP method (blue markers indicate the selections

Pulsar selection results using the proposed optimization-based approach (direct SDP).

Pulsar selection results using the proposed optimization-based approach (SDP with local refinement).

Selection results for SDP with refinement (blue markers indicate the selections)

Pulsar selection results using the proposed optimization-based approach (dire

Pulsar selection results using the proposed optimization-based approach (SOCP V efinement).

Pulsar selection results using the proposed optimization-based approach (SOCP on an embedded platform implemented in C).

Pulsar selection results using the proposed optimization-based approach (SOCP on an embedded platform).

Selection results for direct SOCP method (blue markers indicate the selections

Pulsar selection results using the proposed optimization-based approach (mixed-intA SOCP, exact).

Selection results for SOCP with refinement (blue markers indicate the selections)

Exact solutions using mixed integer approach.

State estimation performance comparison between optical navigation and XNAV-aided optical navigation.

Absolute value of navigation error comparison between optical navigation and XNAV-aided optical navigation.

Absolute value of navigation error comparison between EKF and LM-RIEKF (E = 0.05). Outlier affected regions are shown in light gray patches.

EKF estimation error with associated 30 bounds (c = 0.05). Outlier affected regions are shown in light gray patches.

LM-RIEKF estimation error with associated 3o bounds (e = 0.05). Outlier affected regions are shown in light gray patches.

Absolute value of navigation error comparison between EKF and LM-RIEKF (E= 0.40). Outlier affected regions are shown in light gray patches.

EKF estimation error with associated 3o bounds (e = 0.40). Outlier affected regions are shown in light gray patches.

LM-RIEKF estimation error with associated 3o bounds (6= 0.40). Outlier affected regions are shown in light gray patches.

Absolute error comparison for various candidate methods (E = 0.05).

Absolute error comparison for various candidate methods (E = 0.40).

Average RMSE for multisensor fusion navigation scheme for 1000 Monte Carlo runs represents the fraction of corrupted data)

Monte Carlo analysis (E = 0.05) for final position (projection on : - y axis

Monte Carlo analysis (E = 0.05) for final position (projection on x - 乙 axis)

Monte Carlo analysis (e = 0.05) for final position (projection on y - 2 axis)

Monte Carlo analysis (E = 0.40) for final position (projection on xc - y axis)

Monte Carlo analysis (e = 0.40) for final position (projection on x - 2 axis)

Monte Carlo analysis (E = 0.40) for final position (projection on y - 之 axis)

Results of absolute FPA error for 1000 Monte Carlo trials E : 0.05 and 6 : 0.40 outlier level.

RMSE comparison for multisensoI fusion navigation with and without Doppler inf integration (LM-RIEKF used)

Core elements of the navigation system for the Apollo program [57, 99]

Sensor and optics assembly for the optical navigation camera for the Orion spacecraft [101].

Perspective projection ofa sphere (frontal pinhole imaging model) onto an image plane

Extraction of planet apparent diameter measurement for navigation filter design

Steps followed in fitting an ellipse to an image of Rhea, a moon of Saturn (image courtesy: NASA/JPL).

The effect of outliers on ellipse fitting accuracy (uniform outliers at 40% are assumed). We note the significantly distorted fit in the case of outlier-corrupted edge points.

An ellipse in 2D with the five essential parameters for its identification

Synthetic image of Mars from Celestia software.

Edge extraction results with increasing levels of false edge detection starting from false detection-free case.

A qualitative view ofellipse fit from the various methods considered in this research. NLS green, DLS - yellow, RANSAC - purple, MSAC - brown, LmedS - Cyan, Proposed - red

Ellipse fitting performance for various ellipse fitting methods. We note the highly robust performance of the proposed method under all test cases.

Ellipse fitting performance for Mercury image data from MESSENGER mission (Image credit: NASA/JPL).

Ellipsefitting performance for an image ofTriton (the largest moon of Neptune) from the Voyager 2 mission (Image credit: NASA/JPL).

Comparison of centroid locations for methods under consideration against the baseline centroid location. The headings indicate the image of the target body and the corresponding mission the image is obtained from.

Ellipse fitting performance for Jupiter image data from Juno mission (Image credit: NASA/JPL).

Comparative view of various ellipse fitting methods on real mission images.

RMSE comparisons against baseline for candidate existing methods with the proposec iteratively reweighted robust nonlinear least squares approach.