Based on the research of the lunar exploration mission plans of the world’s major space-faring countries，four major trends of lunar exploration were summarized. With the aim of achieving long-term and sustainable development of lunar exploration in the future，suggestions were put forward for conducting key technology research and development in aspects such as enhancing the ability to reach the moon, achieving stable and all-day supply of energy at lunar surface，establishing a high-speed information network，enhancing the operational capabilities of intelligent equipment，and realizing large-scale exploitation and utilization of lunar resources. These suggestions can provide references for the subsequent planning of lunar exploration missions.

Drawing on the practical experience of celestial body detection projects at home and abroad，the urgent need for autonomous navigation technology in the tasks of extraterrestrial surface exploration and development was analyzed in detail. Based on this requirement，the relevant engineering practices of lunar，Mars，and small celestial body detectors in the landing and roaming missions were reviewed. And in view of the environmental challenges such as extreme illumination interference，dust occlusion effect，and complex terrain features in these tasks，technical difficulties of landing and roaming autonomous navigation were deeply discussed. On this basis，key technologies involved in autonomous navigation，including obstacle detection，autonomous localization，and trajectory planning，were systematically summarized and concluded. Moreover，the development trends of landing and roaming navigation technologies for extraterrestrial surface detectors were prospected.

A navigation system state estimation accuracy evaluation index was proposed based on the geometric relationship between the landmark and the probe for asteroid exploration attachment section with high precision requirements. Under the condition of limited resources on board and with a large number of available navigation landmarks for the probe，the index was derived by analyzing the geometric relationship between the navigation landmarks and the asteroid probe. Combined with the Fisher information matrix，a scalar calculation method for the lower bound of the system state estimation variance was designed. The method avoided the complex matrix calculations in the traditional accuracy evaluation process and adopted the “simulated annealing-enumeration method” to optimize the selection of navigation landmarks，which ensured the system state estimation accuracy and improved the navigation landmark optimization selection efficiency. The method was applied to the optical navigation scene of the Eros asteroid probe detachment section. Simulation results show that the method can effectively improve the efficiency of navigation landmark optimization selection and system state estimation accuracy.

To address optical parameters are affected by the complex environment in space will produce a certain bias in spacecraft on-orbit operation，it will seriously affect the accuracy of spacecraft attitude estimation. The traditional on-orbit calibration method generally base on star angular distance invariance，but it has high computational complexity，the computational and storage resources of deep space probe are very limited，it is difficult to be realized. This paper introduces a calibration method based on singular value decomposition invariance，the size of the singular value is used to measure the observability of the calibration system，on this basis based on the observability to optimize the selection of star distribution and combination，combined with extended Kalman filtering to calibrate the optical parameters of the star tracker. The simulation results show that compared with the traditional star optimal selected model，the optimal selected model proposed in this paper has an advantage in calibrated accuracy and can better suppress the star observational error.

Due to Jupiter’s long distance from us，the positional error of Jupiter can reach hundreds of kilometers，posing significant challenges to autonomous navigation for spacecraft. To address this issue，a star angle/StarNAV integrated navigation method was proposed. By analyzing the impact of ephemeris error of Jupiter on StarNAV，it was observed that velocity error of Jupiter had a greater influence on StarNAV，while positional error had a relatively smaller effect. Considering that Jupiter's positional error is relatively large while its velocity error was small，the integration of star angle navigation and StarNAV was applied to Jupiter exploration. In this method，Star Angle navigation provides the spacecraft's position relative to Jupiter，while StarNAV primarily supplies the velocity information of the spacecraft's relative to Jupiter. Simulation results demonstrate that the proposed integrated navigation method is nearly unaffected by Jupiter's ephemeris error and achieves significantly higher accuracy than Star Angle navigation or StarNAV alone，providing high-precision autonomous navigation information for Jupiter exploration.

A multi-source fusion adaptive filtering method based on the quantitative characterization of observability was proposed to address the problem of severe resource constraints of deep space probes and the difficulty of realizing autonomous navigation with multi-source heterogeneous data fusion. By constructing a variable-channel adaptive fusion structure，evaluating the observability degrees of the filter channel subsystems of each sensitizer on-line based on the quantitative characterization of system observability analysis，and flexibly configuring and dynamically adjusting the number of channels and weights of the source channels，adaptive fusion of multi-source heterogeneous information for autonomous operation of deep space probes was realized. Compared with the traditional pre-fusion structure，the dynamic adjustment of the filter structure with the change of the observability degrees of the measurement data not only solves the computational burden of the spatial-temporal alignment of heterogeneous information，but also avoids the influence of the degradation of the performance of a single sensitizer or subsystem on the fusion accuracy，and minimizes the complexity and redundancy of the filter structure. Through mathematical simulations，it has been verified that in the navigation processes of the long-distance approaching phase and the close-range landing phase，this method has basically the same navigation accuracy as traditional fusion methods. However，due to its ability to adaptively optimize the selection of measurement data and the filtering structure，the computational load is significantly reduced. This method can provide theoretical and technical support for autonomous navigation in deep space exploration.

To address the challenge of noise uncertainty affecting filtering performance in Jupiter’s complex environment，an optical autonomous navigation scheme was established based on the relative line-of-sight information from multiple Jovian moons，using a simplified QLEKF（Q-Learning Extended Kalman Filter）algorithm with a single filter to estimate the position and velocity of the probe. The QLEKF-single（Single Filter Q-learning Extended Kalman Filter）designs a reward function based on the innovation of a single EKF filter. The Q-learning algorithm adaptively selected the values of the noise covariance matrix，while the SoftMax strategy was employed for action selection，ultimately achieving iterative system state estimation by EKF filter. Through simulation by randomly generating initial state estimates and measurement noise，the simplified model of Jupiter's real orbital dynamics was verified. It demonstrated that in scenarios with noise uncertainty，the QLEKF-single algorithm effectively improved navigation accuracy compared to traditional filtering methods. Moreover，compared to the QLEKF algorithm，the run time was reduced by more than 10% with little change in accuracy.

Based on traditional image-based obstacle detection methods can only locate obstacles in 2D image plane，requiring additional measurement methods such as stereo matching to obtain depth information and then determine the 3D positions of obstacles. However，stereo matching faces challenges of high computational cost and decreased accuracy when dealing with complex environments. Therefore，we propose an implicit 3D representation learning method for extraterrestrial obstacle detection was proposed. It encodes the potential three-dimensional coordinates of each point into image features，and the generated features can effectively establish an implicit conversion from 2D images to 3D space，thereby enabling direct prediction of the 3D positions of obstacles. Experiments conducted on Mars surface images collected by the Spirit rover demonstrate that the proposed method can effectively identify locations and sizes of obstacles，achieving 85.5% average precision. The proposed method in this study presents an innovative framework for planetary surface obstacle detection，with substantial potential to advance autonomous navigation capabilities in lunar/Martian exploration rovers.

Due to the demand for safe obstacle avoidance in the autonomous navigation of Mars rover in complex terrain and the double constraints of computational resources and energy supply of the onboard platform，a lightweight detection model, YOLOv8-LMD, was constructed，aiming at realizing the requirements of high precision and lightweight characteristics of the rock detection algorithm on the surface of Mars. First，the lightweight backbone network was reconstructed based on the HGNetv2 architecture to realize the preliminary compression of model parameters. Secondly，a multi-scale feature fusion network structure was designed，and the neck network was reconstructed by integrating slim-neck and ASF-YOLO to effectively improve the feature characterization of rock targets at different scales. In addition，a lightweight detection head was designed by using the convolutional sharing strategy，which reduced the computational complexity and enhanced the classification and localization accuracy at the same time. Finally，a pruning algorithm was used to prune the model parameter redundancy to further compress the model, and the knowledge distillation technique was used to achieve the compensation and optimization of the accuracy. Through experiments， it is found that compared with YOLOv8n，YOLOv8-LMD accuracy was improved by 1.7%，the computational amount was reduced by 68%，the parameter amount was reduced by 77%，and the model size was reduced by 75%. Therefore，it can be concluded that the model proposed in this paper is more suitable for the task of rock detection on the surface of Mars.

Lunar impact crater detection is crucial for lunar surface studies and spacecraft landing missions, yet deep learning still struggles with accurately detecting small craters, especially when relying on incomplete catalogs. In this work, we integrate Digital Elevation Model (DEM) data to construct a high-quality dataset enriched with slope information, enabling a detailed analysis of crater features and effectively improving detection performance in complex terrains and low-contrast areas. Based on this foundation, we propose a novel two-stage detection network, MSFNet, which leverages multi-scale adaptive feature fusion and multi-size ROI pooling to enhance the recognition of craters across various scales. Experimental results demonstrate that MSFNet achieves an F₁ score of 74.8% on Test Region1 and a recall rate of 87% for craters with diameters larger than 2 km. Moreover, it shows exceptional performance in detecting sub-kilometer craters by successfully identifying a large number of high-confidence, previously unlabeled targets with a low false detection rate confirmed through manual review. This approach offers an efficient and reliable deep learning solution for lunar impact crater detection.