- Current Catalog Online First Archive
-
CHEN Qian, WANG Xiaobing, WANG Zhicheng, WANG Tianqi, LIAO Yi
2026,43(1):1-12 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.001
Abstract:
This paper focuses on the intelligent combat requirements in complex electromagnetic environments，and systematically explores the transformations and challenges of missile detection technology driven by artificial intelligence (AI).First，the ‘hyper-saturated’ state of modern battlefield electromagnetic environments is analyzed，and the multi-dimensional threats posed by ‘electromagnetic fog’ constructed by intelligent jamming equipment to missile detection systems are elucidated.Second，typical application cases of countries such as the U.S.and Russia in the field of missile intelligent detection are summarized，revealing the tactical advantages achieved through target recognition and autonomous decision-making algorithms.Third，the application challenges of AI technology in missile detection are analyzed，and four targeted breakthrough directions are proposed，i.e.，generative sample intelligent enhancement technology，edge missile-borne intelligent computing power enhancement technology，multi-mode and multi-missile cross-domain collaborative detection technology，and human-machine fusion bidirectional enhancement technology.Finally，development suggestions are presented from the aspects of multi-modal model ecosystem construction and algorithm evaluation framework，emphasizing the need to establish causal inference perception framework，lightweight model technology，and dynamic adversarial testing environment.The research demonstrates that the deep integration of AI and missile detection will significantly enhance situational awareness，anti-jamming capabilities，and collaborative combat capabilities of missiles in complex electromagnetic environments，providing key technical support for future intelligent warfare.
-
XU Feng, LUOMEI Yixiang, WEI Jiangtao, XU Jingwei, QIU Xiaolan, WU Junjie, WAN Xianrong, JIN Yaqiu
2026,43(1):13-30 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.002
Abstract:
To address future demands such as detection and perception for autonomous intelligent unmanned systems，this paper elaborates on the concept of embodied radar—a platform-radar integrated autonomous sensing system that deeply integrates radar perception with platform mobility and intelligent decision-making.The core idea lies in breaking through the limitations of traditional radar systems characterized by “fixed-mode，unidirectional processing，and passive perception，” and advancing toward a closed-loop “perception-decision-action” paradigm.This enables the radar to actively select detection modes，maneuver paths，and interaction strategies，thereby enhancing performance in dynamic target tracking，partially observable environments，and highly contested electromagnetic scenarios.Traditional radars mostly follow a mission-specific design approach，featuring fixed detection modes，non-adjustable parameters，and predefined trajectories.Their signal processing chains are primarily open-loop with unidirectional data flow，lacking the capability for autonomous optimization based on environmental and target awareness，which makes it difficult to meet the real-time modeling and decision-making requirements of unmanned systems in complex environments.Embodied radar couples platform mobility，detection perception，and agent planning strategies，constructing an electromagnetic world model to characterize the dynamic relationships among “electromagnetic fields，targets，environment，platform，and radar.” Through an interactive information processing framework that enables real-time closed-loop feedback，it achieves joint optimization of detection and maneuver strategies.By leveraging unmanned systems to advance the paradigm of embodied intelligent perception，embodied radar is expected to significantly improve detection effectiveness and autonomous operation capabilities in complex scenarios，which holds great importance for reshaping social production modes and future unmanned combat systems.
-
JIA Wei, LIN Daoshen, ZHU Jianpeng
2026,43(1):31-41 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.003
Abstract:
Artificial intelligence (AI) technology is reshaping the research and development paradigm in the aerospace industry，providing new paths to address core challenges such as extreme environmental adaptability，system complexity，and high-cost constraints.This article focuses on key technologies such as machine learning，reinforcement learning (RL)，and computer vision，exploring their theoretical breakthroughs and potential applications in spacecraft lifecycle management.Research has shown that，deep learning significantly improves the accuracy of remote sensing image analysis and fault detection through high-dimensional feature modeling，RL optimizes the autonomous control ability of spacecraft by combining dynamic decision frameworks，and evolutionary algorithms break through the efficiency boundaries of traditional methods in multi-objective optimization tasks.Despite the phased achievements in technological applications，the particularity of aerospace scenarios poses multiple constraints on AI implementation:data scarcity limits the generalization ability of models，extreme environmental disturbances pose risks to algorithm robustness，and the reliability controversy of black box models in safety critical scenarios urgently needs to be resolved.Future development trends present three evolutionary directions.First，build a hybrid intelligent model integrating physical mechanisms and data-driven integration to enhance the interpretability and environmental adaptability of the decision-making process.Second，develop a lightweight edge computing architecture to solve the real-time autonomous decision-making problem under the constraints of computing power of on-board equipment.Third，establish an intelligent enhancement system for human-machine collaboration to balance the algorithm efficiency with the value of human experience in complex tasks.Through the integration of interdisciplinary technologies and the improvement of engineering verification systems，AI is expected to promote the evolution of aerospace systems from preset logic driven to autonomous cognition，providing sustainable technical support for major tasks such as deep space exploration and constellation networking.
-
DONG Runyang, LIU Haifeng, XUE Guangtao, 陈潜, YANG Lanqing, MA Rong
2026,43(1):42-53 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.004
Abstract:
Addressing the limitation of insufficient spatial resolution in space-based ocean remote sensing satellites and their heavy reliance on sparse and high-cost in-situ ground truth data for calibration，this paper proposes a smartphone-enabled salinity detection technique based on acoustic channel state information (CSI).The method leverages the propagation characteristics of acoustic waves in liquids.With proper designedorthogonal frequency division multiplexing (OFDM) signals，both the amplitude and phase features of the acoustic CSI are extracted and analyzed，enabling non-contact salinity measurement.The feasibility of constructing a high-density and low-cost nearshore salinity ground truth network is analyzed，and experimental validation under various salinity levels and environmental conditions is conducted.The analysis based on laboratory scenarios demonstrates that the method achieves excellent separation for eight distinct salinity levels with intervals of 5‰.By leveraging the widespread prevalence of smartphones，this salinity detection approach can potentially establish a‘capillary-level’ ground observation network through a crowdsourcing model.This network could provide massivereal-time calibration and validation support for ocean salinity satellites，offering a potential technical pathway to alleviate the current bottleneck of ‘sufficient space-borne capacity but inadequate ground-based data’ and meet the closed-loop requirements of a ‘space-ground collaborative’ ocean salinity remote sensing system.
-
YANG Jiuwen, WU Haichao, YU Zhijian
2026,43(1):54-62 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.005
Abstract:
In aerospace launch missions， the accuracy of tracking and identifying sub-stage debris is directly related to mission safety. Traditional radar cross section(RCS) features such as mean and variance often lead to misidentification of targets like sub-stage debris and fairings， owing to the neglect of sequence temporal characteristics. To address this issue， in this paper， a joint feature recognition method for spectrum and autocorrelation RCS is proposed， in which two novel RCS features， i.e.， cumulative spectrum mean and cumulative autocorrelation mean， are introduced. The separability of different features is evaluated， and an optimized combination of three features is used to train and test datasets from six aerospace launch missions. The experimental results demonstrate that the proposed method effectively enhances the clustering performance of similar targets and achieves favorable classification outcomes. The proposed approach can be applied to classification and recognition scenarios for multi-stage rocket separation targets， demonstrating practical engineering application value.
-
LI Xinsheng, ZHANG Haichao, WU Chuze, FENG Shuyi, HAO Yuzhe, LI Yuanxiang
2026,43(1):63-73 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.006
Abstract:
With the rapid development of satellite remote sensing technology，a single data source no longer meets the needs of ship target tracking.The fusion of multi-source satellite observation data can provide comprehensive and accurate Earth observation information，overcome the limitations of a single data source，improve the target tracking performance，and thus support accurate analysis and decision-making.In this paper，the observation data from space-based microwave radar，electronic reconnaissance satellites，and synthetic aperture radar (SAR) satellites are adopted to study how to effectively fuse data from multiple satellite payloads to achieve accurate tracking of ship targets.First，a data fusion method based on the convolutional neural network (CNN) and attention mechanism is proposed，which can effectively integrate data from different modalities to enhance the performance of the model in complex tasks.Then，a data association algorithm based on graph neural networks (GNNs) is proposed，which ensures the consistency and continuity of each target during the tracking process.Simulation validation is carried out with the simulated dataset generated by the ship automatic identification system.The results show that the method obtains good fusion accuracy and tracking stability in three ship distribution density scenarios of 5 km×5 km，10 km×10 km，and 20 km×20 km，and has high value for engineering applications.
-
WANG Peng, ZHOU Kaili, ZHU Hao, WANG Xingyun, DU Jun
2026,43(1):74-81 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.007
Abstract:
Remote sensing image captioning (RSIC) is a task that combines computer vision and natural language processing，aiming to convert remote sensing images into natural language descriptions.In this paper，an image captioning method based on dual-branch attention and Mamba is proposed.In the dual-branch attention Mamba network，a bidirectional scanning Mamba module is designed.The latest Mamba architecture is adopted to encode global image features，and a bidirectional scanning mechanism is used to enhance the model’s spatial perception and understanding of the image space.In the dual-branch attention module，a lightweight attention mechanism is used to effectively focus on and optimize local image features，thereby improving the overall model performance.Tests on image captioning based on the UCM-Captions dataset and Sydney-Captions dataset show that the method proposed in this paper performs better than existing methods.
-
KONG Xianglei, AN Hongyang, ZHANG Chi, YANG Haiguang, RAN Ruilin, LI Zhongyu, WU Junjie, YANG Jianyu
2026,43(1):82-90 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.008
Abstract:
Synthetic aperture radar (SAR) plays a vital role in surface imaging of terrestrial and maritime environments.However，with the increasing complexity of the electromagnetic environment，SAR systems are vulnerable to various forms of active jamming，which severely degrade the imaging performance of SAR.To enhance the anti-jamming capability of SAR，effective scheduling of transmission resources is essential.To address the anti-jamming problem under complex and diverse jamming scenarios，in this paper，a proximal policy optimization (PPO)-based anti-jamming strategy generation method for radar is proposed.An anti-jamming model for SAR is established，and a policy gradient-based optimization framework is developed.By flattening the state and action spaces and carefully designing the reward function，the proposed method effectively mitigates the challenges of slow policy generation and convergence to local optima in high-dimensional radar decision spaces.The simulation results demonstrate that，compared with the dueling double deep Q-network (D3QN)，the proposed approach significantly accelerates the policy generation under combined jamming conditions，particularly in high-dimensional transmission parameter decision spaces，with the optimal number of pulses increased by 2.86 times.
-
A Self-calibration-based Enhanced Difference-guided Method for Remote Sensing Image Change Detection
LI Shuying, WANG Yu, ZHANG San, NIU Saisai
2026,43(1):91-101 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.009
Abstract:
The high-precision change detection of remote sensing images is of great value in fields such as geographic analysis，urban monitoring，and land use assessment.In recent years，change detection networks based on convolutional neural networks and vision transformers have made significant progress，and have demonstrated outstanding performance in fusing dual-temporal image features.However，existing networks have deficiencies in geometric modeling and edge representation，which often results in incomplete boundary details and thus inaccurate positioning of change regions.To address these limitations，in this paper，an enhanced difference-guided change detection network based on self-calibration (SEDGNet) is proposed.First，an adaptive square calibration module (ASCM) is constructed.The global context along the horizontal and vertical axes is modeled to explicitly capture the structural patterns in change regions.While enhancing geometric awareness，it combines a multi-scale fusion module to effectively integrate the differential information from dual-temporal images.Second，a differential fusion guidance module (DFGM) is designed，which integrates encoder features，decoder outputs，and high-frequency differential features to strengthen the edge representation in change areas.Finally，tests are conducted on three public datasets to validate the proposed network.The results show that the proposed network outperformed existing advanced networks across multiple evaluation metrics，verifying its effectiveness and superiority in high-precision change detection tasks.
-
ZHANG Zhaoxiang, ZHANG Jianqiao, ZHOU Shuopeng, HAN Aojia, XU Yuelei
2026,43(1):102-113 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.010
Abstract:
Non-cooperative target pose estimation is the key of on-orbit servicing missions including space capture，debris removal，and spacecraft maintenance.In this paper，a deep-learning-based pose estimation and generalized tracking method for non-cooperative space targets is proposed to address challenges such as unknown three-dimensional (3D) models，complex illumination，and tracking drift.First，an improved EfficientPose network is adopted to determine the initial pose quickly，while dilated convolution modules are introduced to enhance the ability to capture the detailed features of spatial targets.Second，an enhanced SuperPoint model is utilized to extract sub-pixel keypoints from RGBD images，and a multi-channel matching algorithm with triplet loss is designed for high-precision keypoint correspondence.Finally，a non-iterative outlier removal algorithm is proposed to reduce tracking errors，while the batch normalization layer adaptation technology is used to enhance the generalization for unseen targets.Six distinct types of non-cooperative spatial target datasets are constructed，and tests under various lighting and resolution conditions are carried out.The results demonstrate that the improved initial pose estimation network achieves an average deviation distance (ADD) of 91.11% on the Hubble target，outperforming existing state-of-the-art methods in accuracy and robustness.
-
ZHAO Wei, YAN Huaicheng, GAO Sheng, Lü Yunkai
2026,43(1):114-124 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.011
Abstract:
To address the issue of degraded positioning accuracy，particularly along the vertical axis，in traditional fast LiDAR-inertial odometry systems operating in GPS-denied environments—a problem stemming from the default initialization of height at the global coordinate origin and insufficient observational constraints in the Z-axis，which limits further improvement of overall localization performance—this paper proposes a fusion scheme that integrates a low-cost 2D LiDAR with the FAST-LIO framework.Methodologically，the approach begins by converting polar coordinate data from the 2D LiDAR into a 3D point cloud，followed by Random Sample Consensus line fitting，multi-stage validation filtering，and coordinate transformation to obtain a centimeter-level initial height estimate.Subsequently，a tightly-coupled system is constructed by combining the 2D Light detection and ranging (LiDAR) with the inherent inertial measurement unit (IMU) and 3D LiDAR of FAST-LIO.Observations from the 2D LiDAR are incorporated into the observation matrix of the iterated error state kalman filter (IESKF)，thereby enhancing constraints in the Z-axis.The proposed method is low-cost，easy to integrate，and effectively improves the positioning and pose estimation accuracy of unmanned aerial vehicles (UAVs)，supporting reliable autonomous navigation in GPS-denied scenarios.Future work will explore the use of 3D plane fitting to further optimize adaptability.
-
LIU Dong, CHEN Junli, WANG Kaizhi, SHAN Zhengliang, LIN Xin, QIN Gengze
2026,43(1):125-136 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.012
Abstract:
The electronic reconnaissance equipment and systems of various world military powers represented by the United States represent the highest level today.This article systematically introduces the development status of typical electronic reconnaissance equipment systems abroad，with a particular focus on the Pegasus electronic reconnaissance system and the ALR-69A radar early warning receiver in the United States，and briefly describes the relevant technological progress in Russia，the United Kingdom，and France.On this basis，the development of typical electronic reconnaissance technologies，e.g.，digital reception technology，microwave photon broadband reception technology and signal processing，time lens and information optics technology，is discussed.Finally，the future development trends of electronic reconnaissance equipment are prospected.With the technology advancementand the increasing demand of the country for multi-source information reconnaissance and perception，electronic reconnaissance equipment will serve as the core perception node，accelerating the deep integration development with radar，communication reconnaissance，and various interference and anti-interference equipment.By utilizing the close coupling relationship between active reconnaissance and passive reconnaissance，multi-source information fusion and multifunctional integrated processing will ultimately be achieved to meet the needs of different battlefield application scenarios.
-
JI Jiaxin, GUO Mengyao, DI Xinpeng, XU Jing
2026,43(1):137-148 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.013
Abstract:
In cooperative encirclementmissions on non-cooperative space targets，it is hard for traditional control methods to satisfy the requirements ofreal-time response and stable encirclement under highly adversarial and dynamic conditions，primarily due to the decoupling ofgame decision-making and formation execution.To this end，a hierarchical architecture based on the differential game theory and prescribed-time control is proposed.The upper layer of this architecture constructs a target-attacker-defender (TAD) game model，generating the optimal encirclement trajectory for theswarm’s master satelliteby solving for the Nash equilibrium online.The lower layer，in turn，designs a distributed prescribed-time formation controller，ensuring that the swarm can autonomously achieve formation establishment，switching，and maintenance within a prescribed time.The simulation results demonstrate that the proposed method achieves tight coordination between the upper-layer game decision-making and lower-layer formation control，enabling the satellite swarm to complete a rapid approach and stable encirclement of the target within the prescribed time.
-
LI Zihan, LU Shan, HOU Yueyang, LIU Chunyang
2026,43(1):149-158 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.014
Abstract:
To address the multi-spacecraft orbital game problem in low Earth orbit (LEO)，an artificial potential field (APF)-based approach for games is proposed.First，both the game players adopt potential-field strategies:the evader implements a composite avoidance algorithm that integrates multi-source repulsive fields and a velocity-retention potential，while the pursuers implement a distributed encirclement strategy combining predictive potential fields with inter-pursuer repulsive forces.Second，while the evader’s strategy remains unchanged，an enhanced terminal strategy integrating proportional navigation guidance (PNG) is designed for the pursuers to improve the endgame accuracy.The simulation results indicate that the potential-field method enables the pursuers to achieve preliminary encirclement of the evader;however，due to a lack of terminal directivity，capture is not achieved，and the pursuers experience a flyby.In contrast，the enhanced pursuit strategy significantly improves the terminal approach performance，allowing the pursuers to successfully capture the evader at 1 236 s
-
GU Chengpeng, ZHANG Wenqi, SHOU Xing, WANG Weijun, SHI Feizhou
2026,43(1):159-168 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.015
Abstract:
To meet the requirements of autonomous driving tasks of lunar rover vehicles and address the issues of path tracking and stability control in the lunar surface environment with low gravity and low adhesion，a strategy for optimizing linear quadratic regulator control (LQRC) parameters based on reinforcement learning is proposed.First，an linear quadratic regulator (LQR) controller is designed based on the vehicle dynamics model to control the front and rear wheel steering angles and additional yaw moment，and the preview point error model is integrated to adapt to the dynamic response constraints of the steering mechanism of lunar rover vehicles.Second，a reinforcement learning framework based on the soft actor-critic (SAC) algorithm is developed，and a reward function for achieving the optimal tracking accuracy and the sideslip angle is constructed.Through training，an intelligent agent capable of optimizing the LQR weight coefficients and preview point distance is obtained.Finally，a full-vehicle simulation model and double lane change test conditions with different curvatures are built in the Simulink environment.The results show that，compared with fixed parameter control，the reinforcement learning method reduces the lateral position errors by 28.1% and 59.2% and the sideslip angles by 6.2% and 29.8%，respectively.This indicates that the reinforcement learning strategy proposed in this paper can significantly improve the path tracking accuracy and stability control of lunar rover vehicles，providing a solution for realizing autonomous driving in the complex lunar surface environment.
-
LI Yinkang, WANG Hao, YUAN Ronghao, WANG Yangyang, LIU Xiaokun, TAN Shengyong, LI Shuang
2026,43(1):169-179 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.016
Abstract:
Aiming at the bottleneck problems such as lengthy decision-making chains and high dependence on manual intervention in the “human-in-the-loop” mode of traditional on-orbit service mission planning，this paper proposes an autonomous mission planning method based on “large language model (LLM)+agent”.First，an agent decision-making architecture based on “LLM+agent” is constructed to realize the deep collaboration between the semantic understanding of LLM and the accurate calculation of algorithm toolkits.Second，a heterogeneous model interaction framework based on the model context protocol (MCP) is designed to achieve efficient data flow between the LLM and heterogeneous algorithm tools as well as flexible system expansion.Third，a standardized MCP service algorithm toolkit is established based on general on-orbit service mission planning algorithms，and a prompt template oriented to the semantics of space on-orbit service missions is designed to improve the planning reliability of the large model.Finally，through closed-loop tests from task instruction parsing to execution feedback，it is verified that the proposed technology can realize autonomous mission planning for on-orbit services and improve the efficiency of task decision-making.
-
NASHUN Buhe, CHEN Minghua, ZHAO Jiaqing, FU Zhekai, LIU Xin, ZHU Xinzhong
2026,43(1):180-188 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.017
Abstract:
The inter-satellite laser communication (ISLC) technology has been widely used in satellite constellation networks，owing to its high bandwidth and low latency characteristics.However，solar conjunction interference severely affects the stability of communication systems.Taking a typical constellation network as an example，this paper systematically analyzes the solar conjunction phenomenon in satellite constellations and its effects on laser communications.It is found out that，in large constellations，solar conjunctions occur periodically，approximately every 30 days，with up to 48 interference events during a single conjunction window，and the longest single interference duration can reach 100 s.To address these challenges，an on-board solar conjunction mitigation method based on autonomous computing is proposed.By calculating the angle between the laser optical head and the sun in real time，the method enables rapid laser link rerouting.Then，the rerouted link resources are used to dynamically adjust the communication paths.To resolve link interruptions caused by solar eclipses affecting laser links，an intelligent dynamic routing optimization system is constructed.First，the multi-dimensional link quality evaluation functions are designed.Then，combining routing pre-generation and caching strategies，the network can autonomously perceive topology changes，make intelligent decisions，and switch to the optimal communication path.The simulation results demonstrate that when multiple satellites in the same orbital plane experience simultaneous solar eclipses，this intelligent routing mechanism can effectively maintain the continuity and reliability of inter-satellite communication.The research findings provide theoretical support and implementation pathways for enhancing the communication robustness of constellation networks in complex space environments.
-
SONG Fei, GAO Chao, TIAN Yuan, BING Qi, BU Shichao, LIU Yanyang, WANG Tiancheng, CHEN Junli, SHAO Xiaowei
2026,43(1):189-197 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.018
Abstract:
To address the challenges of dynamic modeling and generalization in sequential decision-making environments with scarce interaction data，a latent dynamic prediction representation model (PRM) based on the joint embedding predictive architecture (JEPA) is proposed.A core innovation of the model is the "action-query" attention mechanism:using actions as queries and historical state sequences as keys and values.This allows for the direct and efficient learning and representation of action-state transition relationships in latent space，bypassing the computational burden and information redundancy of pixel-level reconstruction.The experimental evaluations in the Atari Learning Environment (ALE) demonstrate that the proposed model can accurately perform 15 steps of open-loop prediction in trained environments and achieve effective extrapolation for about 3 steps in unseen，unknown environments.The results confirm that this method can learn a world model with certain generalization capability under limited interaction data，providing effective support for general sequential decision-making.
-
WU Haoen, LIU Hui, ZHOU Yirui, CHEN Hao, WEI Jianyu, LIU Bo
2026,43(1):198-210 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.019
Abstract:
In the satellite mission planning based on reinforcement learning， the trial-and-error feedback mechanism is used to learn how to maximize mission benefits under resource， time， and orbital constraints while adapting to dynamic environments. To better simulate real-world scenarios involving multiple satellites observing multiple targets， multiple key factors are taken into account， including target imaging， storage， battery charge， and wheel speed. To this end， in this paper， a complex-environment multi-satellite collaborative mission planning (CE-MSCMP) framework is proposed， and the entire process， from the modeling of satellite mission scenarios with the Markov decision process (MDP) to solving collaborative planning policies， is systematically studied. The results show that the advantages of the CE-MSCMP framework lie mainly in three aspects. First， it builds a comprehensive dynamic environment model to improve the realism of simulation scenarios. Second， it introduces the heterogeneous agent proximal policy optimization (HAPPO) algorithm into the multi-satellite collaborative mission planning problem， providing a new paradigm. Third， it significantly enhances the generalization ability， real-time adaptability， and multi-objective flexibility of the planning policy， overcoming the limitations of traditional methods in scenario adaptability and scalability. The simulation results verify the rationality of the MDP modeling and the effectiveness of the HAPPO algorithm in satellite mission planning， demonstrating the superior performance of the CE-MSCMP framework.
-
SU Hao, JI Mingjiang, BAI Chengchao, WU Peng, MENG Ling, YAN Bin, CAO Lu, HUANG Hao
2026,43(1):211-220 ,DOI: 10.19328/j.cnki.2096-8655.2026.01.020
Abstract:
To address the cooperative task allocation problem in satellite swarm pursuit-evasion games，a team-consensus-based bundle auction method is proposed.First，the pursuit success probability is introduced as a decision variable to construct a cooperative task allocation model for satellite swarms incorporating team benefit.Subsequently，a team-consensus-based bundle auction algorithm is developed，which optimizes global task allocation benefit through cooperative bidding among satellites，and resolves target conflicts via iterative auction updates.Finally，simulation experiments are carried out to validate the effectiveness of the proposed method.The results demonstrate that the proposed method achieves task allocation benefit comparable to the global enumeration method while exhibiting higher efficiency.Moreover，it enables rapid task reallocation within limited iterations，and ensures system stability and mission continuity when individual satellites failure or new targets emerge.
