The rapid expansion of satellite constellations is leading to increased congestion in Earth's exosphere, driving the need for efficient space surveillance network (SSN) to monitor the growing number of resident space objects (RSOs). This paper addresses the scheduling of space surveillance resources for emergency tasks, emphasizing the importance of swift task allocation within strict timeframes, typically the current orbital period. To meet this need, We propose a two-stage SSN emergency task scheduling (ETS) method based on a heuristic algorithm. The approach incorporates a multi-level fuzzy comprehensive evaluation model to dynamically prioritize heterogeneous resources, including phased-array radar, mechanical tracking radar, and optical devices, while minimizing disruption to preexisting schedules. Resource preference rules are applied to streamline the solution space and enhance computational efficiency. Experimental results validate the superiority of the proposed framework. ETS outperforms conventional scheduling methods, achieving a 1.8 times improvement in tracklet valuation, a 48.79% reduction in orbit determination errors, and consistently higher task completion rates across varying scales of emergency events. The developed emergency scheduling framework provides a scalable, robust, and practical solution for real-time SSN operations. It ensures timely and high-quality observation of high-value RSOs, effectively addressing the urgent demands posed by the increasingly congested space environment.

Space Surveillance Network (SSN) plays a crucial role in maintaining the catalog of Resident Space Objectives (RSOs). With the rapid development of large constellations, the resource scheduling problem of the SSN is increasingly prominent. To enhance the flexibility of SSN and improve scheduling efficiency, a Data-driven Three-phase Fast Scheduling Method (DTFSM) is proposed, which includes task matching, optimization solving, and conflict resolution. In Phase I, tasks are allocated to appropriate observation resources through a well-trained neural network. In Phase II, an improved heuristic algorithm is employed to search high-quality solutions within each observation resource. Phase III utilizes improved contract net protocol and dynamic neighborhood structure to resolve time conflicts between tasks. Simulation results show that the completion time of DTFSM is reduced by 2.77-75.96 %, with completion rate 0.18-34.27 % higher compared to baselines. The DTFSM demonstrates superior performance, particularly when dealing with a large number of RSOs. (c) 2025 COSPAR. Published by Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

Owing to the chaotic and non-integrable nature of three-body dynamics, the conventional Keplerian elements are rendered inadequate for cataloging cislunar space objects. Currently, there has been a conspicuous absence of universally recognized parameters for the characterization and cataloging of such objects, thereby posing an urgent challenge to cislunar space situational awareness. This paper proposes a novel approach to parameterize the orbits of Earth-Moon collinear libration points by leveraging the theoretical frameworks of canonical transformations. First, under the Hamiltonian-form dynamical equations of the libration point, symplectic transformations are employed to extract 3 modes of motion from locally linearized part. A subsequent canonical transformation then decouples the hyperbolic invariant manifold from the center manifold within the nonlinear remainder. Finally, 6 characteristic parameters obtained via action-angle variables are established in a bijective correspondence with the state variables, where two parameters characterize the motion of the invariant manifold and four parameters characterize the motion of the central manifold. Furthermore, a distribution map of the Earth-Moon libration point orbits is drawn utilizing Poincaré sections, which can be used to describe the distribution of libration point object. Simulation results demonstrate that the proposed parameters are not only applicable to orbit identification and object cataloging but also exhibit remarkable consistency and robustness against variations in observation arc length and observational errors.

The rapid development of large-scale constellations has led to a dramatic increase in the number of Resident Space Objectives (RSOs), significantly intensifying the complexity of Space Situational Awareness (SSA). Furthermore, the maneuver behaviors of non-cooperative RSOs pose potential threats to space safety, making the real-time monitoring of their post-maneuver orbital becomes more critical. In particular, the maneuvering characteristics of large-scale constellation satellites impose more stringent demands on the timeliness and adaptability of existing scheduling algorithms for observation resources. To address the emergency scheduling demands of heterogeneous ground-based observation resources, this paper proposes an Emergency Task Three-phase Scheduling Framework (ETTSF) based on Automated Machine Learning (AutoML) and auction algorithm. The framework collaboratively optimizes resource allocation through three phases: resource matching, task scheduling, and rescheduling. First, AutoML combined with an auction algorithm predicts and assigns emergency tasks to the most appropriate resources, reducing solution space complexity, simultaneously, the auction algorithm's corrected results are fed back to AutoML for model fine-tuning. Second, a heuristic algorithm with dynamic neighborhood structures efficiently inserts emergency tasks into the routine observation plan. Finally, affected routine tasks are rescheduled to minimize the operational impact. Simulation results demonstrate that compared to baselines-Real-Time Dynamic Scheduling (RTDS) and Improved Adaptive Large Neighborhood Search (IALNS)-ETTSF achieves a 2.82% improvement in Completion Rate of Emergency RSOs (CRER) and a 27.83% reduction in Impact Rate (IR) on routine tasks. Ablation experiments further validate the effectiveness of the resource matching and rescheduling phases.

The issue of constructing the closed-loop equilibrium for close-range orbital pursuit-evasion games was addressed and a computation method that integrates Bellman's principle of optimality,the finite difference method,and interpolation techniques was proposed.A dimension-reduction dynamics of the game system in the line-of-sight coordinate frame was derived,establishing a close-range orbital pursuit-evasion game model and reducing the dimensionality of the system's state space.Based on Bellman's principle of optimality,the original problem was reformulated as a Hamilton-Jacobi-Isaacs partial differential equation terminal value problem,enabling the simultaneous handling of multiple game scenarios through reverse-time analysis.The state space was discretized using Cartesian grids,and the finite difference method was employed to calculate the dynamic evolution process of the equilibrium driven by the dynamics,and analyze the game situation.Utilizing the relationship between control and the spatial gradient of the equilibrium,numerical interpolation was applied to construct the closed-loop control function.The effectiveness of the proposed method was demonstrated through numerical simulations.

Accurately and rapidly identifying the motion intention of space non-cooperative targets is crucial for reducing collision risk and ensuring space mission success. However, existing methods suffer from incomplete intention coverage, insufficient timeliness, and lack of dynamic adaptability. To address these issues, this paper proposes a deep learning method called MSA-LNN for space non-cooperative target intention recognition, which integrates Multi-Head Self-Attention (MSA) and Liquid Neural Network (LNN) frameworks. The research first decouples the relative motion trajectory into six physically interpretable basic modes based on modal decomposition, achieving a unified representation of various motion intentions in three-dimensional space. Subsequently, MSA is employed to focus on key spatiotemporal features, capturing long-range dependencies and temporal evolution patterns; then LNN enables dynamic response, outputting motion intention labels with high timeliness. Simulation results demonstrate that the MSA-LNN model achieves an average recognition accuracy of 98.32%; even under strong noise conditions and two-body nonlinear perturbations, the intention recognition accuracy remains stably above 80%, exhibiting excellent robustness and good engineering applicability.

Space Surveillance Network (SSN) faces inherent challenges in coordinating space and ground-based sensors tasks to maintain the catalog of Resident Space Objects (RSOs). To address operational silos between autonomous scheduling centers, this work introduces a holonic organizational framework that systematically decouples SSN scheduling into two hierarchical process: RSO allocation to scheduling centers and intra-center task scheduling. Based on this framework, an SSN Heterogeneous Resource Joint Scheduling (SSN-HRJS) model is established, including the decision variables, objective function and constraints. Given the complexity of SSN-HRJS, a Data-driven Three-phase Joint Scheduling (DTJS) methodology is proposed. First, A Graph Convolutional Network (GCN) is employed to generate RSO-to-center allocations scheme and probability matrices by learning orbital and resource characteristics. Next, a Hierarchical Combination Optimization (HCO) produces sensor task scheduling plans using the allocation scheme. Subsequently, an Adaptive Simulated Annealing with Priori Search (ASAPS) algorithm leverages the probability matrix and initial scheduling plan to achieve global optimization. Extensive experiments and comparative studies are conducted to verify the efficiency of SSN-HRJS model and DTJS algorithm.

With the rapid increase in the number of spacecraft in low Earth orbit, Space Situational Awareness (SSA) capabilities are facing increasingly severe challenges. Traditional ground-based detection resources are limited by the Earth's curvature, detection range, and fixed geographical locations, leading to observation blind spots and discontinuous temporal coverage. In response, major spacefaring nations are actively developing space-based space object surveillance systems, with optical surveillance satellites being a key representative. Against this backdrop, how to plan trajectories for in-orbit spacecraft clusters that can effectively evade continuous surveillance by space-based optical sensors has become a critical technical challenge in the field of space security. To address this issue, this paper proposes a closed-loop trajectory planning and control method that integrates real-time situational updates and autonomous decision-making. First, a continuous-time optimal control model is formulated, incorporating cluster dynamics, collision avoidance, and sensor line-of-sight constraints. The model is then discretized and approximated using Bezier curves, transforming it into a tractable nonlinear programming problem. To enhance robustness and real-time performance, the solution is embedded within a receding horizon control framework equipped with a state error correction mechanism, forming a complete closed-loop control system. Simulation results demonstrate that the proposed method can successfully plan trajectories for spacecraft clusters that avoid the sensor fields of view of optical surveillance satellites, strictly satisfy inter-spacecraft collision avoidance constraints, and maintain high computational efficiency. These findings validate the reliability and effectiveness of the method as a solution for low-observability cooperative maneuvers of spacecraft clusters. This research provides key technical support for the intelligent planning of cooperative stealth maneuvers for spacecraft clusters in complex space situational awareness scenarios.

Research on the safe operation of on-orbit spacecraft is increasingly shifting from single-spacecraft scenarios to those involving multiple spacecraft. Based on the Multi-Agent Deep Deterministic Policy Gradient algorithm, this paper addresses the problem of cooperative pursuit of a target spacecraft by multiple interacting spacecraft with communication-based coordination. The orbital pursuit-evasion games is formulated as a partially observable Markov games, with carefully designed observation and action spaces. To leverage cooperative advantages among pursuers, a communication target allocation mechanism is designed, enabling the formation of a dynamic ring topology during action execution. A hybrid reward structure combining global and local rewards is proposed. Local rewards incorporate dynamic characteristics to differentiate contributions of individual spacecraft, while global rewards are designed based on an ideal encirclement configuration with its associated constraints. For network architecture, an Encoder-Head structure is adopted: the Encoder, implemented with Long Short-Term Memory networks, effectively captures temporal patterns in the observation space, while the Head, composed of Multi-Layer Perceptron networks, maps encoded latent features into actions. Simulations first investigate a scenario with six pursuers under observational noise, and comparisons against three communication mechanisms and two distinct evader strategies demonstrate that the proposed approach effectively enhances pursuit performance and exhibits generalization capability.

This paper investigates the problem of orbital pursuit-evasion games under bounded distance constraints. The objective function is designed based on the relative distance and continuously processed using an exponential function to facilitate solution via two-player zero-sum differential game theory. Further, the necessary conditions for saddle-point strategies are derived and transformed into a nonlinear optimization problem solved through using the shooting method. Simulation results demonstrate that the designed optimized objective function can effectively maintain the relative distance within the bounded region while avoiding excessively close proximity between the two spacecraft. Additionally, extra simulation cases numerically verify the bilateral optimality of the computed strategies.

This study addresses multi-sensor task scheduling for Geostationary Orbit (GEO) spacecraft observation, developing a model incorporating visibility, task, and resource constraints. To enhance computational efficiency, an Improved Contract Net Protocol Algorithm with Up-Bottom mechanism (UB-ICNPA) is proposed, employing hierarchical computation distribution to reduce solution space complexity. Experimental comparisons with Rapid-Time Dynamic Scheduling (RTDS) demonstrate UB-ICNPA’s superior performance: 70-80% higher Profit Completion Rate (PCR) with reduced computation time and enhanced stability. The algorithm’s layered architecture effectively balances workload across nodes while maintaining scheduling precision, offering significant advancements for space-based sensor networks in managing increasing geostationary orbit surveillance demands.

The Earth–Moon libration points no longer exhibit the dynamical characteristics of “equilibrium points” due to perturbation effects when applying the ephemeris model. By decoupling the forced motions within the ephemeris model and computing the dynamical substitute trajectories, we can reconstruct a dynamical system that recovers the “equilibrium points” feature. Diverging from the conventional analytical approach rooted in the framework of Newtonian mechanics, this paper presents a novel method for calculating dynamical substitute based on the Hamiltonian mechanics framework. First, the Hamiltonian equations for the ephemeris model are formulated. Subsequently, the problem of decoupling forced motions is reformulated as solving a nonautonomous differential equation through canonical transformations. Then, an iterative method based on frequency analysis is employed for the computation. Eventually, approximate analytical solutions for five libration points over a 360 yr period are provided. Simulation results demonstrate that the computed approximate analytical solutions are in excellent agreement with the numerical integration results derived from the ephemeris model, thereby validating the efficacy of the proposed method. The Hamiltonian dynamical system derived herein enables the analysis of nonlinear central manifold motions via canonical transformations, facilitating the construction of higher-order analytical solutions for libration point orbits. This framework also provides a robust foundation for exploring characterization parameters of libration point orbits within the real Earth–Moon system.

The advancement of space technology has led to a continuous increase in the number of Resident Space Objectives (RSOs), while the observation resources remain relatively limited, posing higher demands on the resource allocation of Space Surveillance Network. Currently, research has primarily focused on the observation of Unmanned Aerial Vehicles or maintenance of RSO catalog, with insufficient study on the criteria for objective allocation prioritization or the cooperative observation of high-precision tracking RSOs. This paper introduces a resource allocation method based on task-priority’s time-density, aiming to optimize the efficiency of observation task allocation and the accuracy of RSO tracking. Through simulation experiments, the traditional “start-time first” and “priority first” algorithms are compared, and the results show that the task priority’s time density algorithm performs better in terms of computational time, task completion rate, and facility idle rate.

This study proposed an efficient approach to address the boundary constraints in the context of rendezvous with the noncooperative target within the vicinity of elliptical orbits. The problem was transformed into a two- point boundary value problem (TPBVP) constituted by series of boundary constraints and differential equation constraints by deriving the necessary conditions for the saddle-point strategy. The switching functions embedded with the boundary constraints equations were derived through the Theory of Functional Connections (TFC) to deal with the boundary constraints. Subsequently, the nested function structure with two levels was applied to be the free functions, which was involved in the constrainted expressions together with switching functions to treate the differential equation constraints. Simulation outcomes confirmed the superior computational efficiency of this method when compared to previous studies. Furthermore, a comprehensive analysis was undertaken to explore the impact of orbital eccentricity and true anomaly on the game's results, offering critical insights for enhancing spacecraft safety during on-orbit operations.

Spacecraft relative motion control with inter-craft magnetic force/torque has many advantages, such as no propellant consumption and plume contamination, while providing high-precision, continuous, reversible, synchronous, and non-contacting control capability. This technology has been exploited in many spacecraft operation fields, such as docking and separation, swarm flight, noncontact assembly and reconfiguration, etc. In October 2021, two papers on dexterous magnetic manipulation of conductive non-magnetic object were published in Nature, representing one of the novel developments in magnetic control and highlighting a research hotspot of orbit debris removal with rotating magnets. Aiming to reveal the physical mechanism, enabling technology and corresponding application, from introductions to dynamic modeling and dynamics analysis, magnetic device optimization design, control methods, ground and on-orbit experiments, this paper presents a comprehensive review on magnetic control technology of spacecraft relative motion. In addition, this paper also puts forward two key bottleneck problems of spacecraft relative motion magnetic control that need to be solved in the future: feasible magnetic control of far-distance objects and dexterous magnetic control of non-magnetic objects. (c) 2025 COSPAR. Published by Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

Rapidly increasing space debris poses a serious threat to space activities. Space nets are promising in removing space debris. The impact response of space nets and space debris is rarely investigated despite its importance for a successful net capture. In the present paper, a new rhombic-mesh net-based impact dynamic model is established with the appending constraint method. A ground net impact test is conducted to validate the developed model, after which the dynamic characteristics of space net are explored based on numerical simulations. Then, space net capture parameters are firstly analyzed considering impact dynamics. It is found that the impact force during the impact process contributes to net wrapping and net closing, and the tether in the impact area is likely to reach a maximum force value. Moreover, the energy of a space net is transferred drastically, which can well mitigate the impact response during the impact process. As the impact kinetic energy of the space net increases, the observed impact response becomes drastic. Furthermore, the influence of the impact position is examined, showing that an eccentric impact is negatively correlated with net wrapping and closing.

This manuscript examines the spacecraft rendezvous pursuit-evasion scenario, employing the relative fuel consumption between two spacecraft as the optimization objective. Anchored in Pontryagin's Minimum Principle, the saddle point control strategy is deduced, revealing that the control laws converge to be identical at the saddle point. Considering the difficulty caused by discontinuities due to the potential glider coast phase of the aircraft, the time-optimal solution is leveraged as a preliminary solution to enhance the resolution of this issue, followed by the application of the L-BFGS-B algorithm to facilitate further optimization. The simulation outcomes substantiate the efficacy of this approach in addressing the challenge effectively. Additionally, the study elucidates that, when considering relative fuel consumption, one can optimally harness the gliding phase subsequent to the shutdown of propulsion system realizing fuel conservation alongside the attainment of mission goals.

Proximity observations of spacecraft swarms are crucial for ensuring the safe operation of geostationary earth orbit satellites and the smooth execution of on-orbit servicing missions. Current spacecraft swarm observation planning typically employs a two-layer framework, which has problems such as complex nested optimization, extensive computational requirements, and insufficient applicability. Therefore, a rapid task allocation method is proposed for close-proximity observation missions of spacecraft swarm based on deep neural networks. In the initial phase, deep neural networks is adopted in lieu of numerical optimization for the foundational prediction of fuel expenditure to augment the efficiency of online task allocation. Bayesian optimization is implemented for the automated optimization of hyperparameters. Secondly, an adaptive rapid genetic algorithm has been constructed to address the issues of high computational complexity and susceptibility to local optima in genetic algorithms, which can adjust the crossover and mutation probabilities based on the current population's fitness. Simulation results demonstrate that, compared to traditional approaches, the rapid task allocation method remarkably enhances the efficiency of task allocation while ensuring computational accuracy. Furthermore, the presented method has stronger robustness and can be adapted to different scenarios.

This paper studies many-to-one orbital pursuit-evasion games under impulsive propulsion. A multi-agent environment is designed using Clohessy-Wiltshire (CW) dynamics, where pursuers observe only partial states (self and evader), differing critically from globally observable Markov decision process. The evader follows the maneuvers against all pursuers inspired by on the optimal one-to-one game maneuvers, while pursuers employ maneuvers given by MADDPG with two enhancements: (1) dense rewards based on relative dynamics, and (2) imitation learning for policy initialization and convergence. Simulations show 85% interception success despite the evader’s maneuver advantage, proving the method’s efficacy.

Telemetry data is high-dimensional, non-stationary, and highly coupled. Modeling the process that generates telemetry data is crucial to ensuring the safe and stable operation of large, complex equipment. In this paper, to study the data generation mechanism of telemetry data, we first use multivariate transfer entropy to identify the information sources of telemetry parameters and construct a causal Bayesian network representing the relations among telemetry parameters. Next, we propose the structural causal model individual data generating process (IDGP) to describe the data generation process. Finally, we extend IDGP by learning the dynamic effects of interventions, resulting in the prediction model intervention IDGP (IIDGP). This model integrates causal inference and Kalman filters, allowing for accurate predictions under soft intervention and robust long-term predictions after soft intervention. We evaluate IIDGP on two representative telemetry datasets, focusing on short-term predictions during soft interventions and long-term predictions following soft interventions. These results are compared with several popular baseline methods. The results demonstrate that IIDGP more effectively captures the changing dynamics for the majority of parameters in both prediction scenarios. This effectiveness is demonstrated by significant reductions in mean squared error and mean absolute error, confirming the superiority of the proposed approach.