For the problem of semantic segmentation for space objects in on-orbit service tasks, a deep learning-based real-time and highly generalized semantic segmentation model named Space-SAM is proposed. Firstly, to overcome the difficulties of the traditional segmentation model, which is difficult to run in real-time due to the arithmetic power of the satellite hardware, it is proposed that the original Segment Anything Model is compressed based on knowledge distillation, building the student network with lower computational complexity to achieve lightweight deployment. Furthermore, to adapt to the space object imaging environment, a low-rank adaptive (LoRA) fine-tuning method is used, which adds only very few external parameters to adjust the space object data distribution. Finally, due to signal transmission noise disturbance in on-orbit images, the image denoising module is developed and joined with the semantic segmentation network to eliminate random noise based on powerful denoising algorithms. Experiments on the UESD benchmark show that Space-SAM achieves 91.23% mIOU, outperforming state-of-the-art (SOTA), a series of ablation studies have also demonstrated the effectiveness of the individual operations proposed in this paper. These advances provide a reliable vision method for on-orbit services tasks in resource-constrained and noisy environments. (c) 2025 COSPAR. Published by Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

The exploration of ice giant systems represents one of the priority areas in deep space exploration for the coming decade. Owing to the vast orbital distances of these planets and the need for extensive in-system transfers during missions, trajectory design and optimization constitute a critical enabling technology for the exploration of ice giant systems. While numerous mission concepts and orbital design methodologies have been proposed to date, a comprehensive review of these methodologies is currently lacking, which hinders the further development of novel design techniques and the formulation of new mission proposals. This survey systematically synthesizes both established and state-of-the-art methods across four primary mission phases: interplanetary transfer, ice giant capture, planetary satellite tours, and other scientific observations targeting planets and comets. For each phase, different design strategies are introduced, with their advantages and capabilities described and analyzed to reveal technological progress. Finally, perspectives on future developments are provided, aiming to establish a reference framework for further research in trajectory design and optimization for ice giant system exploration.

This article presents a rapid trajectory replanning method based on deep neural networks (DNNs) for Earth-centered low-thrust rendezvous missions, considering thrust degradation due to partial thruster failure (PTF) events. The three-stage many-revolution rendezvous trajectory computed by a recent staged strategy is adopted as the nominal trajectory. Based on the effects of PTF events with different occurring times and degraded thrust magnitudes on orbital-element adjustments, PTF events are classified into six types. A series of staged trajectory models are formulated for different PTF events. Using the orbital averaging technique, the initial control parameters are analytically derived. The relevant Jacobian matrices are also derived, enabling the parameters to be efficiently corrected to reduce the terminal errors of replanned trajectories. In particular, for the time-consuming parts in the numerical method due to multiple iterations and dynamic integrations, a DNN technique is used to generate high-quality predictions of the control parameters, significantly improving the efficiency of trajectory replanning. The promising results in an Earth-centered low-thrust rendezvous mission verify that the proposed method achieves rapid replanning of many-revolution trajectories with high terminal accuracy.

In gravitational wave detection missions, the test mass capture phase is a critical operation for drag-free satellites, which demands rapid stabilization to prevent collisions with the cavity boundaries. However, existing control schemes typically rely on frequent control-command updates, which may increase the burden on resource-constrained onboard computing and communication systems. To address the tradeoff between control performance and update demand, this paper proposes a dynamic event-triggered adaptive fixed-time control strategy for the test mass capture problem subject to asymmetric output constraints. For the controller design, a backstepping framework is utilized, with a fixed-time command filter to efficiently handle the differentiation of virtual control signals. Crucially, to ensure the physical limitations, a unified barrier Lyapunov function is employed to establish asymmetric output-constraint satisfaction in the continuous-time closed-loop analysis, while an improved auxiliary system is designed to address input saturation. Furthermore, a dynamic event-triggered mechanism (DETM) is constructed to regulate control-command transmission. Compared with the static event-triggered mechanism, the proposed DETM introduces an internal dynamic variable and reduces redundant command updates under the same sampled implementation. Theoretical analysis proves practical fixed-time stability and excludes Zeno behavior. Simulation results demonstrate that the proposed method achieves high-precision practical fixed-time capture with reduced command-update demand.

Great challenges in real-time robust control have been put forward in asteroid exploration in recent years. To deal with the complicated environment full of uncertainties and to satisfy the position as well as attitude requirements of the mission, a convex approach is proposed for stochastic six-degree-of-freedom (six-DOF) trajectories near asteroids. A stochastic six-DOF dynamic model based on modified Rodrigues parameters is built with covariance control and feedback control. Integrated path constraints and field-of-view (FOV) constraints are imposed on account of collision avoidance and sensor orientation, respectively. The constraints with uncertainties are converted to chance constraints and are convexified losslessly using auxiliary variables and virtual control. The stochastic convex problem is applied to hopping transfers on the asteroid 1996 HW1 and solved by successive solutions. The method shows great performance and applicability in optimizing trajectories under uncertainties, as proved by Monte Carlo simulation with numerous samples.

Missed thrust event is an important factor to consider in the design of low-thrust trajectories. Different from interplanetary transfer trajectories, the Jovian moon gravity assist trajectories used for Jupiter system exploration possess shorter orbital period, shorter intervals between moon flybys and require consideration of the impact on subsequent flybys. Therefore, this paper selects typical scenarios for Jupiter moon gravity assist trajectories and proposes different trajectory recovery strategies to address missed thrust events in each scenario based on the characteristics and requirements of each mission phase. The proposed strategies for flight time extension are based on the orbital periods of the Jovian moon and the spacecraft. This approach permits direct calculation of the required time extension, obviating complex optimization, and simultaneously mitigates the impact on subsequent flybys. A comprehensive sampling of trajectories was conducted, and two metrics were proposed to evaluate the performance of different strategies. The results show that the strategies with extended flight time enable addressing a longer duration of missed thrust event while consuming less fuel than a strategy of directly returning to the nominal trajectory in certain cases. These strategies can be selected based on a trade-off between mission duration and fuel costs. The strategies and analysis results presented in this paper can serve as a reference for trajectory design and for responding to missed thrust events.

In the context of gravitational wave detection, the capture control of the test mass (TM) following its release phase is a critical prerequisite for the transition to scientific operation mode. The primary objective is to rapidly stabilize the TMs at the center of the drag-free satellite cavity while avoiding collisions with the cavity boundaries. Notably, the capture control relies on the extremely weak electrostatic forces furnished by electrode plates, thereby presenting substantial challenges for the design of the controller. This article commences by implementing specific state transformations on the original system and introducing a position constraint mechanism that restricts the TM within a predefined time-varying boundary. To tackle the input saturation issue arising from the weak electrostatic forces, an enhanced dynamic auxiliary system is developed. Building on these improvements, a nonsingular terminal sliding mode control algorithm is proposed, which integrates a radial basis function neural network and the improved auxiliary system. Under input saturation condition, the integrated framework ensures that the relative trajectory tracking errors of TMs converge to a small neighborhood around zero within a fixed time. Finally, the simulation results demonstrate the feasibility and efficacy of the proposed method when compared with existing approaches.

This paper presents a reinforcement-learning (RL)-based multiphase robust trajectory design method for low-thrust exploration with gravity assist (GA) under uncertainties. To alleviate the complexity due to the interior-point constraints caused by intermediate GA, a trajectory segmentation strategy is used. The low-thrust trajectory legs before and after GA are segmented into interplanetary transfer phases (ITPs) and approaching phases (APs), respectively, and Markov decision processes with stochastic dynamics are modeled in each phase. Moreover, for the issue of failing to reach the target state of the AP due to the low terminal accuracy of the preceding ITP, a tradeoff factor is defined to modify the nominal initial state of the AP to better guide the exploration of robust policies. In particular, a reachability constraint of the target state of the AP is modeled analytically and incorporated into the reward of the ITP, which can significantly improve the reachability from the end states of RL-based trajectories in the preceding ITP to the target of the AP. Besides, low-thrust robust guidance laws in the AP are trained to deal with the uncertainties over the AP. The promising results in an Earth-Earth-Jupiter mission show that the proposed method can not only effectively deal with various uncertainties but also achieve the desired accuracy for intermediate GA and terminal rendezvous.

A multi-maneuver approach to transition from Lunar Frozen Orbit(LFO)to a cislunar L2 Near Rectilinear Halo Orbit(NRHO)was developed in this research.LFO was utilized to provide the navigation or communication platform for lunar exploration.They are long-term stable orbits with constant orbital elements on average.Subsets of Halo orbit families,known as NRHO,are orbits that are nearly stable.Due to their important locations in cislunar space,the L2 NRHO are considered as possible launching platforms for deep space or lunar exploration mission.For upcoming cislunar space exploration missions,the low fuel consumption transfer approach between these orbits is very valuable.Utilizing the maximum stretching direction to determine the insertion maneuver,the spacecraft may rapidly approach NRHO.In order to optimize the transfer trajectories,a nonlinear programming problem was developed.The optimization results with different transfer windows in the high-fidelity model were given for the transfer from LFO to NRHO,demonstrating the reliability of the proposed strategy.

Continuation methods are used in numerous applications that focus on generating families of periodic orbits with common topology and characteristics, rather than achieving single disjoint trajectories. However, they work by sequentially tracking parameters associated with the system as they vary, thus creating a solution path. This study proposes a nonsequential hybrid method to compute families of resonant periodic orbits traversing total bifurcations of types 1 and 2 in the planar circular restricted three-body problem, including members with arbitrary proximity to the secondary that preserve both the intrinsic resonance of the family and its topology. A thorough description of the development is given, which combines the theory of generating orbits with an analytical asymptotic approximation of resonant orbits. Unification is achieved numerically by incorporating a two-stage differential correction scheme that modulates the minimum distances to the secondary of the closest approaching orbits. Families are constructed orbit-by-orbit simultaneously. The operation of the hybrid method is illustrated with the exterior 5:9 and 6:7 resonant families in the Jupiter–Europa system. The mechanism of type 1 bifurcations is further explored to analyze the impact of family approximation geometries to the singularity on the nonlinear physical behavior in the nearest orbits.

This paper presents the results and design methods of team Nanjing University of Aeronautics and Astronautics in the 12th edition of the Global Trajectory Optimization Competition. To address the problem of sustainable asteroid mining, we focus on the following: analyzing the constraints and asteroids involved; selecting a candidate set of asteroids for which mining missions can be performed easily; establishing an algorithmic flow using phasing indicators, multiobjective beam search, and a genetic algorithm to determine the sequence of asteroid visits for mining ships; and optimizing low-thrust trajectories via an indirect method and global optimization. In addition, a central-node method is proposed to simplify the design process and reduce the computational cost of performing repetitive asteroid-rendezvous missions. The methods developed in the competition enable the mining of 161 asteroids via 20 mining ships, with a total collected mass of 11,513 kg.

Multiple gravity-assist trajectory design is an indispensable step in giant planetary system exploration missions. In view of China's future deep space exploration plan, this paper optimizes and designs the Jovian tour trajectory for the Callisto capture mission, and proposes a multi-stage tour trajectory design method to realize the low-consumption transfer of the spacecraft from the Jupiter capture to multi-moon flyby detection to the Callisto scientific orbit in the patch-conic model. First, Corresponding trajectory design methods are given according to different mission phases. Then, the whole mission is divided into three phases, and the Jupiter flyby sequence is given by the Tisserand graph analysis, and three methods of branch and bound, dynamic programming, and leveraging transfer are used to design and optimize, respectively. With the objective of minimizing the velocity increment, the trajectory design schemes are given and analytically compared in the ephemeris model, which can further save the fuel consumption required for the mission compared with the existing results. The proposed multi-stage design method can also be applied to other planetary systems, providing a reference for the trajectory design of future deep space missions.

The study of resonant flybys has been increasingly addressed by incorporating three-body effects in early stages of complex tour designs. In the past, a resonant transition mechanism revealed the existence of a close interplay between the invariant manifolds associated with unstable resonant orbits at certain energies, coined as interconnection, and has been used to benefit from large-scale natural transport across phase space. However, the state-of-the-art has focussed either on point resonances at few energies, or on promoting resonant transitions via heteroclinic connections that generally do not occur in the most direct way. This paper intends to continue the work of previous researchers and fill a gap left behind, by carrying out a systematic analysis to search for interconnections between fifteen resonant families at multiple energy levels, covering the spectrum of resonances between the secondary in the CR3BP Jupiter-Europa system and Ganymede. To this end, an unsupervised procedure is implemented using standard dynamical systems techniques, and including a simple geometric detection method to capture the clearest homoclinic connections, so that the explicit use of Poincare maps is not necessary. The results presented here can potentially be used to train deep neural networks in order to mitigate the computational burden.

This paper presents a fuel-saving Jovian capture approach utilizing solar gravity perturbation (SGP). A scheme of utilizing SGP with prograde arrivals is proposed, enabling the combination of SGP utilization with multiple-moon-aided Jovian capture. To efficiently utilize SGP, the effects of SGP on the change in the perijove are analyzed via parametric study in the Sun–Jupiter circular restricted 3-body problem. Thereafter, the mechanism of utilizing SGP with respect to the phase angle of the perijove, the perijove radius, and the eccentricity of the capture orbit is concluded. The best condition of SGP utilization and the characteristics of the required flight time are also revealed. Finally, moon-aided capture trajectories utilizing SGP are designed in both simplified and high-fidelity dynamical models. The simulation results indicate that the proposed method can reduce the velocity increment substantially.

Space-based gravitational wave detection has emerged as a central focus of current gravitational wave missions, where the precise stability of spacecraft formations is essential to ensure accurate detection. A Model Predictive Control (MPC) strategy incorporating low-thrust magnitude is proposed in this paper to achieve high-precision configuration for space-based gravitational wave detection. On one hand, the MPC strategy takes the spacecraft's current state vector as the starting point and predicts the state evolution over multiple future time steps. The core idea is "Rolling Time Domain Optimization", where optimization is performed at each time step to determine the best thrust input. Specifically, the thrust magnitude constraints are incorporated into the optimization, making the results more aligned with practical engineering requirements. On the other hand, by relaxing the restrictions on arm length and strengthening the constraints on breathing angle, fuel consumption can be further reduced and mission life can be improved. Finally, the control strategy is applied to the scenario of the heliocentric space-based gravitational wave detection mission, and the simulation results demonstrate that the proposed method effectively supports the formation keeping process. (c) 2025 COSPAR. Published by Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

A benchmark test is presented for the computation of families of resonant periodic orbits in the planar circular restricted three-body problem using sequential and nonsequential methods. As sequential representatives, the natural parameter and the pseudo-arclength continuation techniques are employed. The nonsequential implementation is followed using a hybrid method, which is based on a numerical unification of the theory of generating orbits and an analytical asymptotic approximation of resonant orbits close to periodic second-species solutions. This paper explores the extent to which the hybrid method can benefit from parallel computing on multicore processors to build families orbit-by-orbit simultaneously. With eight cores, the results show better performance than the pseudo-arclength in family portions approaching the small primary and comparable to that of the natural parameter continuation in regular areas. The main advantages of the hybrid method are also discussed, and the existence of a lower threshold in the minimum distances to the secondary that periodic orbits can reach is investigated, which is a consequence of using the nonregularized equations of motion.

Earth-moon distant retrograde orbits (DROs) periodically circle around the moon with high stability, suitable for the deployment of a cislunar station. However, due to the navigation and control errors and uncertainty in dynamic parameter, the actual trajectory drifts from the nominal DRO. To further improve the on-orbit station-keeping (SK) accuracy, robust guidance methods for DRO SK are provided. For reducing the influences of navigation and control errors, a robust SK point design method based on Lyapunov exponents is developed. Trajectory convergence under initial position and velocity error is studied using an equivalent method. The robust DRO-SK point is designed to perform trajectory guidance periodically. To deal with the uncertainty in dynamic parameter, a model-free uncertainty fitting method is develop based on reinforcement learning (RL). The uncertainty is identified, fitted, and predicted using the nominal and actual trajectories to improve the accuracy of trajectory propagation. The optimal state and reward for RL-based uncertainty fitting are investigated. Furthermore, a double-loop guidance framework for on-orbit guidance is established based on a predictor-corrector guidance method. In the inner loop, the RL-based uncertainty fitting method is employed; in the outer loop, the robust SK point is taken as a periodic SK point for the cislunar station. The performance of the robust guidance method is analyzed through numerical simulations.

Space-based optical observation systems constitute critical technologies for enhancing space situational awareness. However, the design of large-scale optical observation constellations under complex operational constraints presents significant challenges, due to computational complexity and the high-dimensional design space. This paper proposes an efficient constellation design methodology which achieves an optimal balance between constellation size and observation performance for high-frequency revisit missions targeting Low-Earth-Orbit (LEO) objects within limited sensor fields of view. The proposed method framework implements a design strategy that combines pre-computed database construction with heuristic-based search algorithms. In particular, an accelerated database generation mechanism significantly reduces the computational complexity of parameter optimization, allowing efficient exploration of the solution space. To address the revisit observation requirements for the LEO objects, a heuristic constellation design rule proposed by relative motion analysis of the satellites is utilized to achieve a dimensionality reduction of the solution space thus allowing for the generation of high quality solutions. The results show that the performance index of the solution generated by this method exceeds the currently known best solution for the related competition problem. The developed method framework shows potential for extension to various complex constellation design problems in space systems engineering.

This paper proposes a method to rapidly estimate time-optimal transfers for solar-sail spacecraft using deep neural networks. First, we employ an indirect approach to derive the optimal control law for time minimization. Next, we introduce a backward-generation technique that rapidly expands the nominal dataset of optimal trajectories. This augmentation process quickly provides a wide range of training samples for our network. We train a deep neural network (DNN) to map orbital parameters to optimal flight time. Numerical results show that this data augmentation technique can rapidly generate a large number of training samples, thereby substantially enhancing the prediction accuracy. The method thus supports fast early-stage design for multi-asteroid exploration, accelerating mission analysis by providing near-instant transfer-time predictions once the DNN is trained. Copyright (c) 2025 The Authors. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/)