Unmanned system embodiment cognition evaluation and collaborative reconstruction method and system

Abstract

The application provides a kind of unmanned system somatic cognitive evaluation and collaborative reconstruction method and system, it is related to unmanned system intelligent collaborative control field.The method comprises: obtaining the multidimensional state data of unmanned system node and carrying out space-time registration and feature standardization processing, obtain somatic state data;Based on somatic state data, the ability of hardware state, software running state and energy state is measured, hardware variability, software adaptability and energy affordability are obtained, and nonlinear coupling dynamic solution is carried out, to obtain the boundary of ontology ability;With ontology ability boundary as task cognitive constraint, task cognitive dimension reconstruction deduction is carried out using mask autoencoder, to obtain task cognitive result;Fusion task cognitive result and carry out distributed consistency solution, according to global consensus result, collaborative reconstruction instruction is generated.The application improves the adaptability and reliability of unmanned system collaborative execution in complex environment by constructing the closed-loop mechanism of ability evaluation, task cognition and collaborative reconstruction.

Description

Technical Field

[0001] This application relates to the field of intelligent collaborative control technology for unmanned systems, and in particular to a method and system for embodied cognitive assessment and collaborative reconstruction of unmanned systems. Background Technology

[0002] With the development of automatic control, artificial intelligence, sensor, and communication technologies, unmanned systems such as drones, unmanned vehicles, and unmanned ships are widely used in scenarios such as inspection and monitoring, emergency rescue, target search, material transportation, and regional operations. Compared to a single unmanned device, an unmanned system cluster composed of multiple unmanned devices can expand the operational scope, improve task execution efficiency, and adapt to more complex operational requirements through task division and information exchange. During the execution of tasks by unmanned systems, task planning, path control, and collaborative scheduling are typically required based on task objectives, environmental information, and the operating status of the unmanned devices to ensure the continuous progress of the task.

[0003] Existing unmanned systems typically assess the operating status and task execution conditions of unmanned equipment based on pre-defined equipment performance parameters, task rules, or partial operational data collected by sensors, and generate corresponding control or coordination schemes accordingly. However, in actual operation, the operating status of unmanned equipment is continuously affected by various factors such as equipment wear and tear, environmental changes, resource consumption, communication quality, and task load, causing its actual task execution capability to change over time. Current technologies for assessing these changes are often fragmented or rely on preset conditions, making it difficult to reflect the actual task execution capability of unmanned equipment in a timely and comprehensive manner. This can lead to a mismatch between the generated task scheme and the actual capabilities of the unmanned equipment. Furthermore, when some unmanned equipment in the cluster experiences abnormal status, communication limitations, or changes in task conditions, existing coordination methods often struggle to promptly integrate the status and task requirements of each unmanned equipment to adjust the original task allocation and coordination relationships, thus affecting the task continuity and overall adaptability of the unmanned system in complex environments.

[0004] Therefore, in the process of unmanned system mission execution in complex and dynamic environments, it is difficult to accurately assess the actual capability status of unmanned equipment, to judge the executability of the mission in a timely manner, and to dynamically adjust the division of labor and coordination among unmanned system clusters according to changes in status. These are problems that urgently need to be solved. Summary of the Invention

[0005] This application provides a method and system for embodied cognitive assessment and collaborative reconstruction of unmanned systems, aiming to solve the technical problems that existing unmanned systems are difficult to accurately assess the actual capability status of unmanned system nodes when performing tasks in complex and dynamic environments, difficult to judge the executability of tasks in a timely manner, and difficult to dynamically adjust the task division and collaborative relationship of unmanned system clusters according to state changes.

[0006] Firstly, this application provides a method for embodied cognitive assessment and collaborative reconstruction of unmanned systems, the method comprising: Acquire multi-dimensional state data of each unmanned system node in the unmanned system. The multi-dimensional state data includes environmental perception data, system operation data, computing resource data, communication status data, and energy status data. For any unmanned system node, spatiotemporal registration and feature standardization are performed on the multidimensional state data of the unmanned system node to obtain the specific state data of the unmanned system node. Based on the embodied state data, the hardware state, software operating state, and energy state of the unmanned system node are measured to obtain hardware variability, software adaptability, and energy affordability. The nonlinear coupling relationship between hardware variability, software adaptability, and energy affordability is dynamically solved to obtain the ontological capability boundary; wherein, the ontological capability boundary includes a dynamic dynamic envelope surface, a computing power redundancy boundary, and an energy attenuation boundary. Using the aforementioned ontological capability boundary as a task cognition constraint, a mask autoencoder is used to perform mask reconstruction deduction on multiple task cognition dimensions of the task to be executed, thereby obtaining the mask reconstruction deduction results for each task cognition dimension. Based on the mask reconstruction results of each task cognition dimension, the task cognition results of each task cognition dimension are obtained; The cognitive results of each task are fused to obtain candidate collaborative information of the unmanned system nodes; Distributed consensus solution is performed on the candidate collaboration information generated by each of the unmanned system nodes to obtain a global consensus result; When the global consensus result meets the preset reconstruction triggering conditions, a collaborative reconstruction instruction is generated based on the global consensus result. The collaborative reconstruction instruction is used to adjust the role, function, or topology of each of the unmanned system nodes.

[0007] In one possible design, the ontological capability boundary is used as a task cognition constraint, and a mask autoencoder is used to perform mask reconstruction deduction on multiple task cognition dimensions of the task to be executed, to obtain the mask reconstruction deduction results for each task cognition dimension, including: The multiple task cognition dimensions are defined, including basic task capabilities, spatial accessibility, time constraints, access security, information and communication connectivity, action coordination, and organizational restructuring and fault tolerance. Obtain task feature data corresponding to each of the aforementioned task cognitive dimensions; Under the constraints of the ontology capability boundary, some features in each task feature data are masked, and the masked features are reconstructed using a mask autoencoder corresponding to each task cognitive dimension to obtain the reconstructed features of each task cognitive dimension, and the latent semantic features of each task cognitive dimension are extracted. Based on the reconstruction features and latent semantic features of each task cognitive dimension, a mask reconstruction deduction result is formed for each task cognitive dimension.

[0008] In one possible design, when masking a portion of the features in each of the task feature data under the constraints of the ontology capability boundary, and reconstructing the masked features using a mask autoencoder corresponding to the cognitive dimension of each task: For the spatial reachability, based on the dynamic dynamic envelope, energy decay boundary and available time constraint of the task to be executed, some spatiotemporal nodes in the candidate motion trajectory are masked, and the candidate motion trajectory that satisfies the dynamic dynamic envelope, energy decay boundary and available time constraint is reconstructed using the mask autoencoder corresponding to the spatial reachability, so as to obtain the reconstruction feature of the spatial reachability. Regarding the organizational restructuring and fault tolerance, the nodes and their connections in the network topology formed by the unmanned system nodes are masked alternately. Then, using the mask autoencoder corresponding to the organizational restructuring and fault tolerance, the backup network topology or functional compensation path is reconstructed based on the unmasked node capability information and the relationship information between nodes, thus obtaining the reconstruction features of the organizational restructuring and fault tolerance.

[0009] In one possible design, the task cognition results for each task cognition dimension are obtained based on the mask reconstruction deduction results for each of the task cognition dimensions, including: Based on the task feature data and reconstruction features corresponding to each of the task cognition dimensions, the reconstruction error of each of the task cognition dimensions is determined; Based on a preset noise tolerance baseline and adjustment parameters corresponding to the urgency of the task to be performed, each of the reconstruction errors is mapped to the corresponding task feasibility confidence level. Based on the feasibility confidence of each task and the corresponding latent semantic features, the task cognition results of each task cognition dimension are obtained.

[0010] In one possible design, the fusion of the cognitive results of each task to obtain candidate collaborative information for the unmanned system nodes includes: Based on the feasibility confidence of each task, the corresponding latent semantic features are weighted to obtain the task perception features of each task cognitive dimension. The task-aware features are mapped to a shared feature space, and context features are determined based on the task intent of the task to be executed and the running state of the unmanned system nodes. Attention fusion is performed on each of the task-aware features in the shared feature space using the context features to obtain a fused feature with task priority attributes; Candidate collaborative information for the unmanned system nodes is generated based on the fusion features.

[0011] In one possible design, the distributed consensus solution for the candidate collaborative information generated by each of the unmanned system nodes to obtain a global consensus result includes: Based on the candidate collaborative information generated by each of the unmanned system nodes, the local decision variables of each of the unmanned system nodes are determined, and a local objective function corresponding to each of the local decision variables is constructed. Based on the consistency constraints among the local decision variables and the local objective functions, a distributed consensus optimization problem is constructed. The alternating direction multiplier method is used to iteratively update each of the local decision variables, global consensus variables and dual variables, and the original residuals are determined based on the updated local decision variables and global consensus variables, and the dual residuals are determined based on the global consensus variables of adjacent iterations. The iteration stops when the original residual and the dual residual satisfy the corresponding preset convergence conditions, and the global consensus result is obtained based on the global consensus variables at the time of stopping the iteration.

[0012] In one possible design, the dynamic calculation of the nonlinear coupling relationship between hardware variability, software adaptability, and energy affordability to obtain the ontological capability boundary includes: Based on the aforementioned hardware variability, software adaptability, and energy affordability, an ontology capability state tensor is constructed that includes the aforementioned hardware variability, software adaptability, and energy affordability. Based on the ontological capability state tensor, an ontological capability nonlinear evolution model is established. The ontological capability nonlinear evolution model is used to characterize the cross-domain coupling relationship between hardware variability, software adaptability and energy affordability. Based on the aforementioned nonlinear evolution model of the ontological capability, the coupled change trends of hardware variability, software adaptability, and energy affordability are predicted to obtain the ontological capability evolution trajectory. A stability analysis is performed on the evolution trajectory of the ontological capability to determine the capability dissipation safety boundary corresponding to the evolution trajectory of the ontological capability. Based on the evolution trajectory of the ontological capability and the capability dissipation safety boundary, the dynamic dynamic envelope, the computing power redundancy boundary, and the energy decay boundary are determined to obtain the ontological capability boundary.

[0013] In one possible design, based on the embodied state data, the capability measurement of the hardware state, software operating state, and energy state of the unmanned system node is performed to obtain hardware variability, software adaptability, and energy affordability, including: Based on the execution instructions and corresponding execution feedback of the unmanned system node, the operational health status of the actuator is determined, and based on the operational health status and the dynamic constraints of the unmanned system node, the range of available motion capabilities of the unmanned system node is determined, thus obtaining the hardware variability; Based on the task queue status, computing resource occupancy status, and resource requirements and timeliness requirements of the unmanned system node to be executed, the real-time execution capability and computing resource reserve of the unmanned system node for the task to be executed are determined, and the software adaptability is obtained based on the real-time execution capability and computing resource reserve. Based on the real-time power consumption, battery temperature, and remaining state of charge of the unmanned system node, the available range of the unmanned system node is determined, and based on the available range, the energy affordability is obtained.

[0014] In one possible design, the spatiotemporal registration and feature normalization processing of the multidimensional state data of the unmanned system node to obtain the specific state data of the unmanned system node includes: Based on the acquisition time corresponding to each of the multidimensional state data, time synchronization is performed on each of the multidimensional state data to obtain time synchronization data. Based on the unified spatial reference frame of the unmanned system node, the data with spatial attributes in the time synchronization data are spatially aligned, and according to the corresponding acquisition time, the data without spatial attributes are state-associated with the spatially aligned data to obtain spatiotemporal registration data. The spatiotemporal registration data is subjected to noise suppression, feature extraction and normalization, and the processed features are fused to obtain the embodied state data of the unmanned system node.

[0015] Secondly, this application provides an embodied cognitive assessment and collaborative reconstruction system for unmanned systems, the system comprising: The data acquisition module is used to acquire multi-dimensional status data of each unmanned system node in the unmanned system. The multi-dimensional status data includes environmental perception data, body operation data, computing resource data, communication status data, and energy status data. The state data processing module is used to perform spatiotemporal registration and feature standardization on the multidimensional state data of any unmanned system node to obtain the specific state data of the unmanned system node. The on-body capability measurement module is used to measure the hardware status, software operation status and energy status of the unmanned system node based on the on-body status data, and to obtain hardware variability, software adaptability and energy affordability. The capability boundary calculation module is used to dynamically calculate the nonlinear coupling relationship between hardware variability, software adaptability and energy affordability to obtain the ontological capability boundary; wherein, the ontological capability boundary includes a dynamic dynamic envelope surface, a computing power redundancy boundary and an energy attenuation boundary. The task cognition deduction module is used to use the ontology capability boundary as the task cognition constraint, and to use the mask autoencoder to perform mask reconstruction deduction on multiple task cognition dimensions of the task to be executed, to obtain the mask reconstruction deduction result of each task cognition dimension, and to obtain the task cognition result of each task cognition dimension based on the mask reconstruction deduction result of each task cognition dimension. The cognitive result fusion module is used to fuse the cognitive results of each task to obtain candidate collaborative information of the unmanned system nodes; The distributed consensus solving module is used to perform distributed consensus solving on the candidate collaborative information generated by each of the unmanned system nodes to obtain a global consensus result. The collaborative reconstruction module is used to generate a collaborative reconstruction instruction based on the global consensus result when the global consensus result meets the preset reconstruction triggering condition. The collaborative reconstruction instruction is used to adjust the role, function or topology relationship of each of the unmanned system nodes.

[0016] Thirdly, this application provides an electronic device, including: a memory and at least one processor; The memory stores computer-executed instructions; The at least one processor executes computer execution instructions stored in the memory, causing the at least one processor to perform the method described in the first aspect or various possible designs of the first aspect.

[0017] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed, implement the method described in the first aspect or various possible designs of the first aspect.

[0018] This application provides a method and system for embodied cognitive assessment and collaborative reconstruction of unmanned systems. The method acquires multi-dimensional state data from each unmanned system node, including environmental perception, system operation, computing resources, communication, and energy. This data is then spatiotemporally registered and feature standardized to form embodied state data, enabling comprehensive analysis of previously dispersed node state information under a unified state representation. Based on this, capability measurements are performed on hardware state, software operation state, and energy state. The nonlinear coupling relationship between hardware variability, software adaptability, and energy affordability is dynamically solved, thereby obtaining an ontological capability boundary that reflects the current actual capability limit of the node, thus improving the accuracy of the assessment of the actual capability state of unmanned system nodes. Further, this method… The multi-dimensional task cognition deduction of the task to be executed, constrained by the body's capability boundary, matches the obtained task cognition results with the current actual dynamics, computing power, and energy conditions of the nodes, thereby enabling timely judgment of task executability. Finally, by fusing the cognition results of each task to form candidate collaborative information, and performing distributed consistency solution on the candidate collaborative information of each node, collaborative reconfiguration instructions for adjusting node roles, functions, or topological relationships are generated when the reconfiguration trigger conditions are met. This enables the unmanned system cluster to adjust the division of labor and collaborative relationships in a timely manner according to the dynamic changes in node capabilities and task conditions, solving the problems of difficulty in accurately assessing the actual capabilities of nodes, difficulty in timely judging task executability, and difficulty in dynamically adjusting cluster collaborative relationships in complex dynamic environments. Attached Figure Description

[0019] Figure 1 A flowchart illustrating an embodied cognition assessment and collaborative reconstruction method for an unmanned system provided in this application embodiment; Figure 2 A schematic diagram of the overall processing architecture for embodied cognitive assessment and collaborative reconstruction of an unmanned system provided in this application embodiment; Figure 3 A schematic diagram of the ontology capability state tensor and capability boundary calculation provided in this application embodiment; Figure 4 A schematic diagram of a task feasibility assessment process based on a mask autoencoder is provided for an embodiment of this application; Figure 5 This is a schematic diagram of a task cognition result fusion framework provided in an embodiment of this application; Figure 6 This is a schematic diagram of a collaborative reconstruction process based on distributed consensus solving, provided as an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims and drawings of this application are intended to cover non-exclusive inclusion.

[0022] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of the phrase "embodiment" in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0023] In this article, the term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists, A and B can exist simultaneously, and B exists. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0024] Furthermore, the terms "first," "second," etc., in the specification and claims of this application or in the aforementioned drawings are used to distinguish different objects rather than to describe a specific order, and may explicitly or implicitly include one or more of the features.

[0025] In the description of this application, unless otherwise stated, "multiple" and "at least two" mean two or more (including two), and similarly, "multiple groups" and "at least two groups" mean two or more (including two groups).

[0026] In the description of this application, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, "connected" or "linked" can refer not only to a physical connection, but also to an electrical connection or a signal connection. For instance, it can be a direct connection, i.e., a physical connection, or an indirect connection through at least one intermediate component, as long as the circuit is connected. It can also refer to the internal connection between two components. A signal connection can refer not only to a signal connection through a circuit, but also to a signal connection through a medium, such as radio waves. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0027] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, different technical features in this application can be combined with each other.

[0028] When unmanned systems perform collaborative tasks in complex environments, it is typically necessary to assess the task execution capabilities of each node based on its equipment operating status, computing resource status, communication status, and energy status. Existing capability assessment methods often employ preset performance parameters, fixed rules, or thresholds to judge mechanical performance, computing resources, and energy status separately, usually assuming that the capabilities of unmanned system nodes remain relatively stable over a certain period. However, because factors such as equipment wear and tear, localized failures, increased computing load, and continuous energy consumption can occur simultaneously and interact, assessing different states separately fails to reflect the comprehensive impact of various state changes on the actual execution capabilities of nodes, easily leading to inconsistencies between the capability assessment results and the actual task execution capabilities of the unmanned system nodes.

[0029] In terms of task feasibility assessment, existing methods typically focus on whether the target can be identified, whether a planned path exists, or whether individual performance indicators meet requirements. They lack comprehensive analysis of factors such as task timeliness, regional permissions, communication connectivity, inter-node coordination, and task succession capabilities in case of failure. Therefore, it is difficult to accurately determine whether unmanned system nodes are suitable to continue executing the corresponding tasks under their current operating state. Furthermore, when some nodes malfunction, communication conditions deteriorate, or task requirements change, existing unmanned systems usually need to transmit the status information of each node to the central node, which then re-plans the task and coordinates control. In cases of limited communication bandwidth, significant transmission delays, or central node failure, decision-making lags or coordination interruptions can easily occur, making it difficult for the unmanned system cluster to adjust node roles, task functions, and inter-node coordination relationships in a timely manner.

[0030] To address the aforementioned issues, this application provides a method and system for embodied cognitive assessment and collaborative reconstruction of unmanned systems. First, it acquires multi-dimensional state data of each unmanned system node, including environmental perception, ontological operation, computing resources, communication, and energy. State data from different sources are processed into embodied state data that uniformly represents the current operating state of the nodes. Then, the hardware variability, software adaptability, and energy affordability of the nodes are measured, and the nonlinear coupling relationship between these three is dynamically solved to determine the ontological capability boundary of the node in its current state. Based on this, the ontological capability boundary is used as a task cognitive constraint, and a masked reconstruction deduction is performed on multiple task cognitive dimensions of the task to be executed, ensuring that the task executability assessment is based on the actual dynamic capabilities, computing capabilities, and energy capabilities of the nodes. Furthermore, the results of each task cognitive step are integrated to form candidate collaborative information for each node, and a global consensus result is obtained through distributed consensus solving. When the global consensus result meets the preset reconstruction triggering conditions, the roles, functions, or topological relationships of the unmanned system nodes are adjusted, thus forming a closed-loop processing process from node state perception, ontological capability assessment, task executability judgment to cluster collaborative reconstruction.

[0031] Figure 1 This is a flowchart illustrating an embodied cognitive assessment and collaborative reconstruction method for an unmanned system, provided as an embodiment of this application. Figure 1 As shown, the method includes S101 to S109, and S101 to S109 are described in detail below.

[0032] It should be noted that the method provided in this application embodiment can be executed by a processor deployed in an unmanned system node, or by a processor in an edge computing device, collaborative control device, server, or other electronic device that is communicatively connected to each unmanned system node, or by the processors of each unmanned system node jointly executing the method in a distributed collaborative manner. The unmanned system can be an unmanned system cluster composed of multiple homogeneous or heterogeneous unmanned system nodes. Unmanned system nodes can be, for example, drones, unmanned vehicles, unmanned boats, underwater unmanned devices, or other devices capable of autonomous or semi-autonomous task execution. Each unmanned system node can be equipped with environmental perception components, motion execution components, computing components, communication components, energy components, a processor, and a memory. The memory stores a computer program that can be executed by the processor, and when the processor executes the computer program, it implements the method provided in this application embodiment.

[0033] like Figure 2 As shown, the overall processing architecture includes four interconnected processing parts: multi-dimensional state data acquisition and embodied state construction, ontology capability measurement and capability boundary calculation, task cognitive inference and task cognitive result generation, and task cognitive result fusion, distributed consistency solution and collaborative reconstruction.

[0034] In the multidimensional state data acquisition and embodied state construction section, environmental perception data, ontology operation data, computing resource data, communication state data, and energy state data are acquired, and embodied state data are formed through spatiotemporal registration and feature standardization processing, so that state data from different sources, different acquisition times, and different data scales can form a unified node state representation.

[0035] In the part of ontological capability measurement and capability boundary calculation, based on the embodied state data, the capabilities of the hardware state, software operation state and energy state of the unmanned system nodes are measured to obtain hardware variability, software adaptability and energy affordability; further, the nonlinear coupling relationship between the three is dynamically calculated to obtain the ontological capability boundary including the dynamic dynamic envelope surface, the computing power redundancy boundary and the energy decay boundary.

[0036] In the task cognition deduction and task cognition result generation section, the ontology capability boundary is used as the task cognition constraint. Mask reconstruction deduction is performed using a mask autoencoder corresponding to multiple task cognition dimensions. Based on the reconstruction features, latent semantic features and task feasibility confidence of each task cognition dimension, the task cognition result of each task cognition dimension is formed.

[0037] In the task cognition result fusion, distributed consensus solution and collaborative reconstruction section, the task cognition results are fused to generate candidate collaborative information for each unmanned system node; the candidate collaborative information generated by each unmanned system node is solved in a distributed consensus manner to obtain a global consensus result; when the global consensus result meets the preset reconstruction triggering conditions, a collaborative reconstruction instruction is generated to adjust the role, function or topology relationship of each unmanned system node.

[0038] S101. Obtain multi-dimensional state data of each unmanned system node in the unmanned system.

[0039] The multidimensional state data includes environmental perception data, ontological operation data, computing resource data, communication state data, and energy state data.

[0040] In some embodiments, multidimensional status data can be continuously or periodically acquired through sensors, actuator feedback interfaces, computing resource monitoring interfaces, communication status monitoring interfaces, and energy management interfaces deployed on each unmanned system node. Different unmanned system nodes can be configured with different data acquisition components according to equipment type, payload type, and task to be performed, and the acquired data can be associated with the corresponding node identifier, acquisition time, and data type.

[0041] Environmental perception data reflects the state of the environment and mission objects in which unmanned system nodes are located. It can include one or more of the following: image data, point cloud data, radar detection data, target location data, obstacle distribution data, terrain data, passable area data, environmental semantic data, and dynamic target status data. Environmental perception data can be collected by camera devices, laser detection devices, millimeter-wave detection devices, sonar devices, or other environmental sensors.

[0042] The system's operational data reflects the motion and operational status of its nodes, and may include one or more of the following: position, attitude, velocity, acceleration, angular velocity, executed commands, executed feedback, actuator rotation speed, body vibration status, load status, and equipment health status. This operational data can be collected by inertial measurement units, positioning devices, actuator sensors, and the equipment control bus.

[0043] Computational resource data reflects the current software operating status and available computing power of unmanned system nodes. This data may include one or more of the following: processor utilization, graphics processor utilization, memory usage, task queue status, thread running status, algorithm execution latency, and remaining computing resources. Computational resource data can be obtained by the operating system, task scheduler, or computational resource monitoring program within the unmanned system node.

[0044] Communication status data reflects the communication capabilities of the unmanned system node itself and between unmanned system nodes. It may include one or more of the following: communication link quality, signal strength, signal-to-noise ratio, transmission delay, available bandwidth, packet loss rate, communication connection relationships, and network topology status. Energy status data reflects the current energy supply capability of the unmanned system node. It may include one or more of the following: real-time power consumption, battery voltage, battery current, battery temperature, remaining state of charge, remaining energy, and predicted endurance.

[0045] Various types of multidimensional state data can have different sampling periods, data formats, coordinate reference systems, and numerical ranges. When acquiring multidimensional state data, timestamps, node identifiers, and data source identifiers can be added to various types of data to facilitate unified processing of state data from different sources, at different sampling times, and at different data scales.

[0046] S102. For any unmanned system node, perform spatiotemporal registration and feature standardization on the multidimensional state data of the unmanned system node to obtain the specific state data of the unmanned system node.

[0047] Specifically, multidimensional state data can be time-synchronized based on the acquisition time of various types of multidimensional state data, and data with spatial attributes can be converted to a unified spatial reference frame corresponding to the unmanned system node. Data without spatial attributes can be associated with the spatial state within the same time period. On this basis, noise suppression, outlier processing, feature extraction, and normalization can be performed on various types of data. The processed features can then be associated or fused to form embodied state data that can uniformly represent the current environmental state, physical state, computing state, communication state, and energy state of the unmanned system node.

[0048] Embodied state data can be represented as eigenvectors, eigenmatrices, state tensors, or other data formats that can be processed by subsequent capability measurement processes, and it is dynamically updated as new multidimensional state data is continuously acquired. By forming embodied state data, the impact of differences in data sources, sampling times, and data scales on subsequent capability assessments can be reduced.

[0049] S103. Based on embodied state data, measure the capabilities of the hardware state, software operation state, and energy state of unmanned system nodes to obtain hardware variability, software adaptability, and energy affordability.

[0050] Specifically, hardware variability is used to characterize the motion, attitude adjustment, and task load control capabilities that unmanned system nodes can achieve under the current equipment health state and dynamic constraints; software adaptability is used to characterize the ability of unmanned system nodes to perform real-time calculations, algorithm execution, and resource adjustments for tasks to be executed under the current computing resource occupancy and task scheduling state; and energy affordability is used to characterize the ability of unmanned system nodes to continuously undertake tasks to be executed under the current power consumption, battery temperature, and remaining energy state.

[0051] Hardware variability, software adaptability, and energy affordability can be represented numerically, in intervals, as vectors, matrices, or other capability representations. Capability measurement is not based solely on the rated performance parameters of the equipment, but rather on combining the current specific state data of the unmanned system nodes to determine the actual hardware capabilities, software operational capabilities, and energy availability capabilities that each unmanned system node can provide at the current moment.

[0052] S104. Dynamically solve the nonlinear coupling relationship between hardware variability, software adaptability and energy affordability to obtain the ontological capability boundary.

[0053] The ontological capability boundary includes the dynamic dynamics envelope, the computing power redundancy boundary, and the energy decay boundary.

[0054] During the operation of unmanned system nodes, the hardware state, software state, and energy state are not independent of each other. For example, when the performance of an actuator degrades, the control program may need to increase its computational load to compensate for the change in hardware state, and the increased computational load may lead to a faster rate of energy consumption. Therefore, hardware variability, software adaptability, and energy affordability can be jointly characterized, and the evolution of the unmanned system node's capabilities can be dynamically calculated based on the interaction between these three factors over time.

[0055] The dynamic dynamics envelope is used to limit the range of motion or dynamic states that an unmanned system node can safely achieve under the current hardware conditions; the computing power redundancy boundary is used to limit the range of computing resources that can continue to be allocated while meeting the timeliness and real-time control requirements; the energy decay boundary is used to limit the power consumption, endurance, or task execution range that can be continuously supported under the current energy conditions. The ontological capability boundary can be updated with changes in hardware conditions, computing load, and energy conditions, thereby reflecting the current actual usable comprehensive capability limit of the unmanned system node.

[0056] S105. Using the ontological capability boundary as the task cognition constraint, a mask autoencoder is used to perform mask reconstruction deduction on multiple task cognition dimensions of the task to be executed, and the mask reconstruction deduction results of each task cognition dimension are obtained.

[0057] Specifically, task feature data corresponding to multiple task cognitive dimensions can be constructed based on the task content, task objectives, task environment, and task constraints of the task to be executed. The dynamic dynamic envelope, computing power redundancy boundary, and energy decay boundary are used as insurmountable capability constraints during task deduction. Some task features are masked, and the masked autoencoder is used to reconstruct the masked parts based on the unmasked task features and ontological capability boundaries.

[0058] Through the above processing, the mask reconstruction inference for each task cognitive dimension is no longer based solely on pre-set rated performance or ideal environmental conditions, but is constrained by the current real dynamic capabilities, computing power, and energy capabilities of the unmanned system nodes. The mask reconstruction inference results can include the reconstructed task characteristics, the potential semantic information formed during the reconstruction process, and other inference information that can reflect the degree of matching of the corresponding task cognitive dimensions.

[0059] S106. Based on the mask reconstruction deduction results of each task cognition dimension, the task cognition results of each task cognition dimension are obtained.

[0060] Specifically, the mask reconstruction results of each task cognition dimension can be analyzed separately to determine the degree of matching between the current capability state of the unmanned system node and the corresponding task requirements. The task cognition results can be used to characterize the feasibility, credibility, risk level, or semantic state of the unmanned system node in executing the task to be executed under the corresponding task cognition dimension.

[0061] By generating task cognition results for each task cognition dimension, we can avoid judging whether a task is executable based solely on the performance of a single device, the result of a single path, or a single task indicator. This allows subsequent collaborative decision-making to comprehensively consider the execution status of unmanned system nodes under different task constraints.

[0062] S107. The cognitive results of each task are fused to obtain candidate collaborative information of unmanned system nodes.

[0063] Specifically, the task cognition results of each task cognition dimension can be mapped to a unified feature representation space, and the results of different task cognition dimensions can be fused by combining the task intent of the task to be executed and the operational status of the unmanned system nodes. During the fusion process, the influence of each task cognition dimension in the fusion result can be adjusted according to different task stages, environmental states, and node capability states.

[0064] Candidate collaboration information is used to characterize the overall suitability and collaborative decision-making tendency of unmanned system nodes to participate in a task to be performed. It can include one or more of the following: node task feasibility, task priority, node capability status, collaboration urgency, credibility, and suggested collaboration methods. Candidate collaboration information can be represented using compact features, information tuples, feature vectors, or other data formats that are easy to transmit and compare between unmanned system nodes.

[0065] S108. Perform distributed consensus calculation on the candidate collaborative information generated by each unmanned system node to obtain the global consensus result.

[0066] Specifically, each unmanned system node can form a local collaborative decision based on its own generated candidate collaborative information, and exchange candidate collaborative information, local decision results, or intermediate variables for consensus solving through communication connections between unmanned system nodes. Each unmanned system node updates its own local collaborative decision according to the received information, so that the local collaborative decisions of different nodes gradually tend to be consistent.

[0067] The global consensus result is used to characterize the unified collaborative decision-making formed by the unmanned system cluster for the task to be executed. It can reflect which unmanned system nodes execute the task, what role or function each unmanned system node undertakes, and what kind of collaborative relationship is adopted between the unmanned system nodes. Distributed consensus solution can be completed jointly by all unmanned system nodes, without having to centrally upload all the original environmental data and ontology operation data collected by each unmanned system node to a single central node.

[0068] S109. When the global consensus result meets the preset reconstruction triggering conditions, generate a collaborative reconstruction instruction based on the global consensus result.

[0069] Among them, the collaborative reconfiguration command is used to adjust the roles, functions, or topological relationships of each unmanned system node.

[0070] The preset reconfiguration trigger conditions can be pre-set based on the task requirements of the task to be executed, the capability boundaries of the unmanned system nodes, the task cognition results of each task cognition dimension, and the operational status of the unmanned system cluster. For example, when the global consensus result indicates that the actual capability of the current execution node is insufficient to continue completing the task to be executed, the existing task division cannot meet the task timeliness requirements, the existing communication topology has the risk of interruption, some unmanned system nodes experience functional abnormalities, or the task can be taken over by other unmanned system nodes to achieve higher task feasibility, it can be determined that the global consensus result meets the preset reconfiguration trigger conditions.

[0071] When the preset reconfiguration trigger conditions are met, the unmanned system nodes that need to be adjusted, the adjustment objects, and the adjustment methods can be determined based on the global consensus results, and corresponding collaborative reconfiguration instructions can be generated. Role adjustments may include redefining task execution nodes, reconnaissance nodes, relay nodes, collaborative nodes, or standby nodes; function adjustments may include enabling, disabling, downgrading, or switching the perception, computing, communication, control, or mission payload functions of unmanned system nodes, or instructing other unmanned system nodes to perform function compensation for nodes with degraded capabilities; topology adjustments may include establishing or disconnecting communication connections between nodes, changing relay relationships, adjusting control relationships, or reorganizing the network topology of the unmanned system cluster.

[0072] The collaborative reconfiguration command can also be used to instruct unmanned system nodes to perform independent tasks, collaborative tasks, degrade and wait, take over roles, switch functions, or reorganize topology. After generating the collaborative reconfiguration command, it can be sent to the relevant unmanned system nodes, which will then adjust their own operating status according to the command. The adjusted unmanned system nodes continue to collect multi-dimensional status data and re-execute S101 to S109, thus forming a closed-loop processing procedure of continuous perception, dynamic evaluation, task judgment, and collaborative reconfiguration.

[0073] In this embodiment, multi-dimensional state data, including environmental perception, physical operation, computing resources, communication, and energy, of each unmanned system node are acquired and then spatiotemporally registered and feature standardized to form embodied state data. This allows for comprehensive analysis of previously dispersed node state information under a unified state representation. Based on this, capabilities are measured for hardware, software operation, and energy states, and the nonlinear coupling relationship between hardware variability, software adaptability, and energy affordability is dynamically solved. This yields an ontological capability boundary that reflects the current actual capability limit of the node, thereby improving the accuracy of the assessment of the actual capability state of the unmanned system node. Furthermore, the ontological capability boundary is used to constrain the tasks to be executed. The multi-dimensional task cognitive simulation matches the obtained task cognitive results with the current dynamics, computing power and energy conditions of the nodes, thereby enabling timely judgment of task executability. Finally, by fusing the cognitive results of each task to form candidate collaborative information, and performing distributed consistency solution on the candidate collaborative information of each node, collaborative reconstruction instructions for adjusting node roles, functions or topological relationships are generated when the reconstruction trigger conditions are met. This enables the unmanned system cluster to adjust the division of labor and collaborative relationships in a timely manner according to the dynamic changes in node capabilities and task conditions, solving the problems of difficulty in accurately assessing the actual capabilities of nodes, difficulty in timely judging task executability and difficulty in dynamically adjusting cluster collaborative relationships in complex dynamic environments.

[0074] In one possible embodiment, the method steps shown in S105 are implemented by S1051 to S1055, which are described in detail below.

[0075] S1051. Define multiple task cognition dimensions, including basic task capabilities, spatial accessibility, time constraints, access security, information and communication connectivity, action coordination, and organizational restructuring and fault tolerance.

[0076] Specifically, based on the task requirements involved in unmanned system (UMS) mission execution, such as perception and recognition, motion arrival, time constraints, access control, communication assurance, node coordination, and fault reorganization, the tasks to be executed can be divided into the seven task cognition dimensions mentioned above. Among them, basic task capability characterizes the ability of UMS nodes to perceive and recognize task objects under the current environment and load conditions; spatial accessibility characterizes the feasibility of UMS nodes reaching the target location or completing the predetermined motion process under the current capability boundary constraints; time constraint characterizes whether UMS nodes can complete the task to be executed within the specified time range; access control security characterizes whether the task execution process complies with area access rights, electronic fences, or operation authorization requirements; information communication connectivity characterizes the ability of UMS nodes to maintain effective communication with other UMS nodes during task execution; motion coordination characterizes whether the motion states and motion processes of multiple UMS nodes can be coordinated with each other; and organizational restructuring and fault tolerance characterize the ability of the UMS to reorganize node relationships and maintain task continuity when some UMS nodes or communication connections between nodes are abnormal.

[0077] In some embodiments, mask autoencoders corresponding to the seven task cognition dimensions can be set respectively, so that different mask autoencoders can perform task cognition inference in parallel according to the task feature data, masking strategy and reconstruction target corresponding to each task cognition dimension.

[0078] S1052. Obtain task feature data corresponding to each task cognitive dimension.

[0079] Specifically, task feature data corresponding to each task cognitive dimension can be extracted from the task description data, task environment data, task constraint data, and the embodied state data of the unmanned system nodes.

[0080] For example, task feature data corresponding to basic task capabilities may include image features of the target area, target semantic features, target contour features, environmental visibility, and task payload parameters; task feature data corresponding to spatial accessibility may include environmental occupancy grid, obstacle distribution, target location, candidate motion trajectory, body motion state, and available time; task feature data corresponding to time constraints may include historical task time, remaining distance, task time window, and estimated execution time for each task stage; and task feature data corresponding to access control security may include controlled areas, electronic fence boundaries, node access permissions, and operation authorization information.

[0081] Task characteristic data corresponding to information and communication connectivity may include the relative positions of each unmanned system node, communication link quality, signal-to-noise ratio, obstacle distribution, and network connection relationships; task characteristic data corresponding to action coordination may include the predicted intentions, relative speeds, action sequences, and joint motion states of coordinating nodes; and task characteristic data corresponding to organizational restructuring and fault tolerance may include the network topology of the unmanned system, node capability status, relative relationships between nodes, and communication or control connection relationships between nodes.

[0082] By organizing task feature data separately for different task cognitive dimensions, each mask autoencoder can perform targeted reconstruction and deduction based on the corresponding task requirements, without having to input all heterogeneous task data into the same cognitive processing process.

[0083] In some embodiments, these can be mask autoencoders corresponding to the configurations of basic task capabilities, spatial reachability, time constraints, access security, information communication connectivity, action coordination, and organizational restructuring and fault tolerance. Each mask autoencoder can adopt a Transformer-based network structure and perform mask reconstruction deduction in parallel. Here, Transformer refers to a transformer network.

[0084] The mask autoencoder, corresponding to the basic mission capability, can acquire multispectral images of the target area and mission payload parameters of the unmanned system node, mask key areas in the target image, and reconstruct the target semantics and target contour based on the unmasked image area to evaluate the ability of the unmanned system node to complete target perception and recognition under complex weather, lighting changes, smoke obscuration or visual degradation conditions.

[0085] The masked autoencoder corresponding to spatial reachability can acquire the environmental occupancy grid, the motion state of unmanned system nodes, the target location, and candidate motion trajectories. It masks some future spatiotemporal nodes in the candidate motion trajectories and reconstructs the candidate motion trajectories under dynamic dynamic envelope, energy decay boundary, and available time constraints to determine whether the unmanned system nodes can reach the target location within their current practical capabilities. The specific masking and reconstruction process for spatial reachability is detailed in subsequent embodiments.

[0086] The masked autoencoder, corresponding to the time constraint, can obtain the historical task time, remaining distance and task time window, mask the time consumption characteristics of each future task stage or future time step, and reconstruct the life cycle time consumption curve of the task to be executed, so as to determine whether the unmanned system node can complete the task to be executed within the specified time range.

[0087] The masked autoencoder, corresponding to access control security, can obtain the topology of controlled areas and electronic fence boundaries, as well as node authorization sequences. It can mask the access or operation authorization attributes of some areas and reconstruct legal and secure operation boundaries that meet the access and operation authorization requirements of the areas. This allows the system to determine whether unmanned system nodes will enter restricted areas or violate corresponding access constraints when executing tasks.

[0088] The mask autoencoder, corresponding to information communication connectivity, can obtain the relative distance matrix, obstacle distribution, and signal-to-noise ratio spatial distribution between each unmanned system node. It can mask part of the channel state when the unmanned system node moves along the candidate motion path and reconstruct the communication link quality at the corresponding position or time to determine whether the unmanned system node can still maintain an effective communication connection with other unmanned system nodes after performing the corresponding action.

[0089] The masked autoencoder corresponding to action coordination can obtain the predicted intention, relative velocity and action primitive sequence of other unmanned system nodes, mask part of the intersection dynamic state in the joint state space of multiple unmanned system nodes, and reconstruct the joint action state between multiple unmanned system nodes to determine whether there is spatial interference, velocity mismatch or action timing conflict when multiple unmanned system nodes perform cooperative tasks.

[0090] The masked autoencoder, corresponding to organizational restructuring and fault tolerance, can acquire the communication and control network topology formed by each unmanned system node, apply anomaly masks to the nodes or connections between nodes in the network topology, and reconstruct backup network topologies, backup communication paths, or functional compensation paths based on the capability status of the remaining unmanned system nodes and the relationships between nodes. This allows for the assessment of the unmanned system's ability to maintain mission continuity after node damage, communication interference, or connection interruption. Specific masking and reconstruction processes for organizational restructuring and fault tolerance are detailed in subsequent embodiments.

[0091] S1053. Under the constraints of the ontology capability boundary, mask some features in the feature data of each task.

[0092] Specifically, based on the task state changes, future task states, or potential anomalies to be simulated for each task cognitive dimension, the mask object and mask position corresponding to each task cognitive dimension can be determined, and the corresponding task features can be replaced with mask identifiers or temporarily removed from the input features.

[0093] Different masking objects can be used for different task cognitive dimensions. For example, the target image region or target semantic features corresponding to the basic capabilities of the task can be masked, the future time consumption features corresponding to time constraints can be masked, the area access attributes or operation authorization attributes corresponding to access security can be masked, the future communication link state corresponding to information communication connectivity can be masked, and the joint motion state corresponding to action coordination can be masked. The specific masking methods corresponding to spatial accessibility and organizational reshaping and fault tolerance can be further determined according to subsequent embodiments.

[0094] S1054. The masked features are reconstructed using a masked autoencoder corresponding to each task cognitive dimension to obtain the reconstructed features of each task cognitive dimension, and the latent semantic features of each task cognitive dimension are extracted.

[0095] Specifically, each masked autoencoder may include an encoder and a decoder. The encoder is used to encode the unmasked task features and ontology capability boundaries to obtain the latent semantic features of the corresponding task cognitive dimension; the decoder is used to predict and recover the masked features based on the latent semantic features to obtain the reconstructed features of the corresponding task cognitive dimension.

[0096] Different masked autoencoders corresponding to different cognitive dimensions of tasks can employ different input data, masking strategies, and reconstruction targets to simulate different constraints affecting the completion of the task. Each masked autoencoder can execute in parallel or in a preset order based on the availability of computing resources.

[0097] S1055. Based on the reconstruction features and latent semantic features of each task's cognitive dimension, form the mask reconstruction deduction results of each task's cognitive dimension.

[0098] Specifically, the reconstructed features, latent semantic features, and task cognitive dimension identifiers corresponding to any task cognitive dimension can be associated to form a corresponding masked reconstruction inference result. The masked reconstruction inference result may also include the original task features, the positions of the masked features, and association information used to compare the original task features and the reconstructed features, so as to subsequently determine the task cognitive result of the corresponding task cognitive dimension.

[0099] In this embodiment, the task to be executed is reconstructed and deduced through masking from the aspects of basic task capabilities, spatial accessibility, time constraints, access security, information communication connectivity, action coordination, and organizational restructuring and fault tolerance. The current ontological capability boundary of the unmanned system node is used as the constraint for the deduction process of each dimension. This ensures that the reconstruction results of each task cognition dimension are based on the node's actual usable dynamic capabilities, computing capabilities, and energy capabilities. This avoids judging task executability based solely on a single task indicator or equipment rated parameters and provides a deduction basis that includes the reconstruction state and potential semantic information for the subsequent formation of task cognition results for each task cognition dimension. This improves the comprehensiveness of task cognition in complex dynamic environments and the degree of matching with the actual capability state of the node.

[0100] In one possible embodiment, the processing of spatial accessibility and organizational remodeling and fault tolerance in the method steps shown in S1053 and S1054 is implemented by S10531 to S10535, which are described in detail below.

[0101] S10531. Obtain candidate motion trajectories corresponding to spatial reachability, and determine the dynamic envelope surface, energy decay boundary, and available time constraints of the task to be executed corresponding to the candidate motion trajectories.

[0102] Specifically, one or more candidate motion trajectories leading from the current position to the target position can be generated based on the environmental grid, obstacle distribution, target location, current position and motion state of the unmanned system nodes. The candidate motion trajectory includes multiple spatiotemporal nodes arranged in chronological order, and each spatiotemporal node can include the position, attitude, velocity, acceleration or other motion state at the corresponding moment.

[0103] Spatial reachability is not only used to determine whether there is a geometric path between the current position and the target position, but also to determine whether there are candidate motion trajectories that can be actually executed by unmanned system nodes under the current dynamic envelope, energy decay boundary and available time constraints.

[0104] S10532. Based on the dynamic dynamic envelope surface, energy decay boundary and available time constraint, some spatiotemporal nodes in the candidate motion trajectory are masked.

[0105] Specifically, continuous or discontinuous future spatiotemporal nodes can be selected from candidate motion trajectories, and one or more features of the selected spatiotemporal nodes, such as position, attitude, velocity, acceleration, expected power, or expected arrival time, can be replaced with mask identifiers. The masked spatiotemporal nodes are used to simulate future motion states that have not yet actually occurred, enabling the masked autoencoder to infer subsequent motion processes based on the unmasked spatiotemporal nodes and the body's capability boundaries.

[0106] S10533. Using a mask autoencoder corresponding to spatial reachability, candidate motion trajectories that satisfy the dynamic dynamic envelope, energy decay boundary and available time constraints are reconstructed to obtain the reconstructed features of spatial reachability.

[0107] Specifically, the process of reconstructing spatial accessibility can be conceived as the following optimal control problem: And satisfy: ;in, Indicates candidate motion trajectories; This represents the control sequence corresponding to the candidate motion trajectory; Indicates the available time constraint for the task to be executed; This represents the rate of change of the candidate trajectory at time t; This represents the power required to perform the corresponding motion at time t; This represents the environmental risk cost corresponding to the candidate motion trajectory; , and These represent the weighting coefficients corresponding to motion changes, energy consumption, and environmental risks, respectively. This indicates the motion state of nodes in an unmanned system. Indicates the dynamic state transition relationship; Represents the dynamic envelope surface; This represents the remaining usable energy defined by the energy decay boundary. This indicates the area occupied by the obstacle.

[0108] The masked autoencoder, corresponding to spatial accessibility, can reconstruct the masked future spatiotemporal nodes based on the unmasked spatiotemporal nodes, environmental occupancy status, target location, and the aforementioned constraints, forming candidate motion trajectories that meet the requirements of dynamic executability, energy affordability, task timeliness, and obstacle avoidance.

[0109] When the feasible solution space is empty, the blocking judgment and zero probability signal are output; when the feasible solution space is not empty, the reachability assessment result, the corresponding candidate motion trajectory and the corresponding optimal energy cost are output, and these are used as the reconstruction feature of spatial reachability.

[0110] S10534. To address organizational restructuring and fault tolerance, alternately mask the nodes and connections between nodes in the network topology formed by the nodes of each unmanned system.

[0111] Specifically, a network topology can be constructed based on the communication, control, or mission collaboration relationships between the nodes of each unmanned system, and node masks and connection relationship masks can be used alternately in different mask simulation rounds. Node masks are used to simulate the failure, damage, or offline status of critical mission nodes, relay nodes, or collaborative nodes; connection relationship masks are used to simulate interference or interruption of communication links, control connections, or mission collaboration connections.

[0112] S10535. Using a masked autoencoder corresponding to organizational restructuring and fault tolerance, based on the unmasked node capability information and inter-node relationship information, the backup network topology or functional compensation path is reconstructed to obtain the reconstruction characteristics of organizational restructuring and fault tolerance.

[0113] Specifically, node capability information may include the computing power redundancy, task payload capacity, energy status, and communication relay capability of unmasked unmanned system nodes; inter-node relationship information may include the relative position, relative distance, communication link status, control relationship, and task cooperation relationship between nodes.

[0114] When critical nodes are masked, alternative nodes with corresponding sensing, computing, communication relay, or task execution capabilities can be identified from the remaining unmanned system nodes, and functional compensation paths can be established whereby these alternative nodes assume the corresponding roles or functions. When the connections between nodes are masked, backup communication or control connections can be established through one or more remaining unmanned system nodes, forming a backup network topology that bypasses the masked connections. Functional compensation paths can include one or more of the following: sensing function compensation, communication relay compensation, computing task migration, or task execution role replacement.

[0115] Organizational restructuring and fault-tolerant reconfiguration features can include one or more of the following: reconfigured backup network topology, alternative node relationships, functional compensation paths, and re-established connections between nodes. By using node failures and connection interruptions as masking conditions to perform reconfiguration simulations in advance, it is possible to assess whether the remaining unmanned system nodes can maintain mission continuity before actual failures occur, and to provide a basis for generating candidate coordination information corresponding to subsequent node role adjustments, functional compensation, or topology reconfiguration.

[0116] In this embodiment, during the spatial reachability simulation, the dynamic dynamic envelope, energy decay boundary, and available time constraint of the task to be executed are jointly applied to the mask reconstruction of the candidate motion trajectory. This allows the obtained reconstruction features to reflect whether the unmanned system node has an executable trajectory under the current actual capability conditions, avoiding the judgment of task reachability based solely on the existence of geometric paths. Simultaneously, during the organizational restructuring and fault tolerance simulation, by alternately applying node masks and connection relationship masks, and reconstructing backup network topology or functional compensation paths based on the remaining node capabilities and inter-node relationships, the unmanned system can pre-simulate the collaborative reorganization mode under node failure and link interruption conditions. This improves the matching degree between the spatial reachability and organizational restructuring and fault tolerance cognitive results and the actual task execution conditions, and provides a more reliable simulation basis for the role succession, functional compensation, and topology adjustment of the unmanned system cluster.

[0117] In one possible embodiment, the method steps shown in S106 are implemented by S1061 to S1063, which are described in detail below.

[0118] S1061. Based on the task feature data and reconstruction features corresponding to each task cognitive dimension, determine the reconstruction error of each task cognitive dimension.

[0119] Specifically, for any task cognitive dimension, the task feature data before masking can be mapped to the reconstructed features output by the corresponding masked autoencoder according to feature position, feature type, or time order, and the corresponding reconstruction error can be calculated according to the following formula: ;in, This represents the reconstruction error of the cognitive dimension of the k-th task. ; Corresponding basic abilities for the task Corresponding spatial accessibility, Corresponding time-bound constraints Corresponding permission security, Corresponding information and communication connectivity, Corresponding action coordination and Corresponding to organizational restructuring and fault tolerance; This represents the number of masked features involved in the reconstruction error calculation in the k-th task cognitive dimension; This represents the original task feature corresponding to the i-th masked position in the k-th task cognitive dimension; This represents the corresponding reconstructed feature.

[0120] The smaller the reconstruction error, the more accurately the mask autoencoder can recover the corresponding task state under the constraints of the ontological capability boundary, and the higher the degree of matching between the current capability of the unmanned system node and the corresponding task requirements; the larger the reconstruction error, the higher the uncertainty or execution risk in the cognitive dimension of the corresponding task.

[0121] S1062. Based on the preset noise tolerance baseline and the adjustment parameters corresponding to the urgency of the task to be executed, each reconstruction error is mapped to the corresponding task feasibility confidence level.

[0122] Specifically, the reconstruction error of the k-th task cognitive dimension can be mapped to the corresponding task feasibility confidence level according to the following formula: ;in, Let represent the task feasibility confidence level corresponding to the k-th task cognitive dimension, and Located within the interval [0,1]; This represents the preset noise tolerance baseline corresponding to the cognitive dimension of the k-th task; This represents the adjustment parameter corresponding to the urgency of the task to be performed, and serves as the adaptive decay rate for the cognitive dimension of the k-th task.

[0123] When the reconstruction error is not greater than the preset noise tolerance baseline, the mission feasibility confidence remains at a high level, allowing normal environmental changes, sensor noise, and normal reconstruction deviations to be tolerated; when the reconstruction error exceeds the preset noise tolerance baseline, the mission feasibility confidence decreases exponentially as the reconstruction error increases.

[0124] For tasks with high urgency, the adjustment parameter can be increased to make the task feasibility confidence more sensitive to reconstruction errors exceeding the preset noise tolerance baseline; for tasks with a larger fault tolerance range, the adjustment parameter can be decreased to make the decrease in task feasibility confidence relatively gradual.

[0125] like Figure 4 As shown, firstly, task feature data corresponding to any task cognitive dimension is obtained, and then, according to the masking strategy of the corresponding task cognitive dimension, some features in the task feature data are randomly masked or directionally masked.

[0126] The encoder of the masked autoencoder encodes the unmasked task features to obtain the latent semantic features of the corresponding task cognitive dimension. The decoder of the masked autoencoder predicts and recovers the masked task features based on the latent semantic features to obtain the reconstructed features of the corresponding task cognitive dimension. In the above mask reconstruction process, the ontology capability boundary serves as a task cognitive constraint, used to limit the reconstructed features output by the masked autoencoder from exceeding the currently available dynamic capabilities, computational capabilities, and energy capabilities of the unmanned system node.

[0127] The task feature data before masking is compared with the corresponding reconstructed features to determine the reconstruction error of the corresponding task cognitive dimension. The smaller the reconstruction error, the more accurately the masked autoencoder can recover the corresponding task state under the constraints of the ontology capability boundary, and the higher the degree of matching between the current capability of the unmanned system node and the corresponding task requirements; the larger the reconstruction error, the higher the uncertainty or execution risk of the corresponding task cognitive dimension.

[0128] Furthermore, based on a preset noise tolerance baseline and adjustment parameters corresponding to the urgency of the task to be performed, the reconstruction error is mapped to the task feasibility confidence level. When the reconstruction error is not greater than the preset noise tolerance baseline, normal environmental fluctuations, sensor noise, and normal reconstruction deviations are tolerated; when the reconstruction error exceeds the preset noise tolerance baseline, the task feasibility confidence level decreases as the reconstruction error increases.

[0129] thus, Figure 4 The process shown can quantify the degree of matching between the current capability status of unmanned system nodes and the corresponding task requirements through mask reconstruction results, and provide task feasibility confidence and potential semantic features for forming task cognitive results of each task cognitive dimension.

[0130] S1063. Based on the feasibility confidence of each task and the corresponding latent semantic features, the task cognition results of each task cognition dimension are obtained.

[0131] Specifically, the task feasibility confidence level corresponding to any task cognition dimension can be associated with the latent semantic features extracted by the corresponding masked autoencoder to form the task cognition result for the corresponding task cognition dimension. The task feasibility confidence level is used to quantify the feasibility of unmanned system nodes meeting the corresponding task requirements, while the latent semantic features are used to retain the task state relationships, environmental constraint relationships, and node capability constraint relationships on which the task feasibility judgment is based.

[0132] For example, regarding spatial reachability, task feasibility confidence can reflect the credibility of forming an executable motion trajectory under the current ontology capability boundaries and task time conditions. Latent semantic features can retain the correlation information between obstacle distribution, dynamic constraints, energy constraints, and motion states. For information communication connectivity, task feasibility confidence can reflect the credibility of maintaining communication connections during task execution. Latent semantic features can retain the correlation information between node locations, link quality, obstacle distribution, and network connection relationships. Task cognition results for other task cognition dimensions can be formed in the same way.

[0133] Task cognition results can include identifiers for the corresponding task cognition dimensions, task feasibility confidence levels, and latent semantic features. They can also include feasible, risky, or infeasible states determined based on the task feasibility confidence levels. By simultaneously retaining task feasibility confidence levels and latent semantic features in the task cognition results, a unified quantitative judgment basis can be provided while preserving the individual task semantics and constraint information of different task cognition dimensions, thus providing input for subsequent fusion of the various task cognition results.

[0134] In this embodiment, the reconstruction error of each task cognition dimension is first determined based on the difference between the task feature data and the reconstruction features. Then, combined with a preset noise tolerance baseline and the urgency of the task to be executed, the reconstruction error is nonlinearly mapped to the task feasibility confidence level. This allows small errors caused by normal environmental fluctuations and sensor noise to be reasonably tolerated, while larger errors caused by a significant mismatch between task requirements and the actual capabilities of nodes can result in a more significant decrease in confidence level. Furthermore, the task feasibility confidence level is combined with the corresponding latent semantic features to form the task cognition result. This ensures that each task cognition dimension has comparable quantitative results of task feasibility while retaining the corresponding task status and constraint relationships. This improves the accuracy, sensitivity, and information completeness of the task executability judgment and provides a reliable basis for subsequent fusion of the task cognition results and the formation of candidate collaborative information.

[0135] In one possible embodiment, the method steps shown in S107 are implemented by S1071 to S1074, which are described in detail below.

[0136] S1071. Based on the feasibility confidence of each task, the corresponding latent semantic features are weighted to obtain the task perception features of each task cognitive dimension.

[0137] Specifically, the task perception feature corresponding to the k-th task cognition dimension can be determined according to the following formula: ;in This represents the task perception feature corresponding to the kth task cognition dimension; This represents the task feasibility confidence level corresponding to the cognitive dimension of the k-th task; This represents the latent semantic feature corresponding to the cognitive dimension of the k-th task; This indicates element-wise multiplication.

[0138] When the task feasibility confidence is a scalar, it can be used to uniformly weight each feature value in the latent semantic features; when the task feasibility confidence is expanded into a weight vector with the same dimension as the latent semantic features, the feature values ​​at the corresponding positions can be weighted element by element.

[0139] S1072. Map the perception features of each task to a shared feature space, and determine the context features based on the task intent of the task to be executed and the running status of the unmanned system nodes.

[0140] Specifically, the feature dimensions of each task-aware features can be transformed and semantically aligned using fully connected layers, mapping the task-aware features corresponding to different task cognitive dimensions to a unified shared feature space.

[0141] The task objective, phase, priority, and expected completion state of the task to be performed can be encoded and fused with one or more of the following: the embodied state, capability boundaries, task execution progress, and node role of the unmanned system node, to obtain contextual features. Contextual features employ... This indicates the relative importance of each task's cognitive dimension in reflecting the current task intent and node operating status.

[0142] S1073. Attention fusion is performed on the task-aware features in the shared feature space using contextual features to obtain fused features with task priority attributes.

[0143] Specifically, context features can be used As the query vector, the features formed by mapping the perceptual features of each task to the shared feature space are used as the key vector and value vector, and the attention weights and fused features corresponding to the cognitive dimensions of each task are determined according to the following formula: , ;in, This represents the attention weight corresponding to the cognitive dimension of the k-th task; Represents contextual features; This represents the key vector formed after the task-aware features of the k-th task cognitive dimension are mapped to the shared feature space. This represents the key vector corresponding to the cognitive dimension of the j-th task; The feature dimension represents the shared feature space; Indicates fusion characteristics; This represents the task-perceived feature after the k-th task cognition dimension is mapped to the shared feature space.

[0144] Through the aforementioned attention fusion, the contribution of different task cognitive dimensions to the fused features can dynamically change according to the task intent and the operational status of the unmanned system nodes. For example, in task phases with strict arrival time requirements, the attention weights corresponding to spatial reachability and time constraints can be increased; when there are risks of communication interference, node failure, or connection interruption, the attention weights corresponding to information communication connectivity, organizational restructuring, and fault tolerance can be increased; and when multiple unmanned system nodes need to perform joint actions, the attention weights corresponding to action coordination can be increased.

[0145] like Figure 5 As shown, the task cognition result fusion process includes task-aware feature construction, shared feature space mapping, attention fusion based on context features, and candidate collaborative information generation.

[0146] In the process of constructing task-aware features, the corresponding latent semantic features are weighted according to the task feasibility confidence level corresponding to each task cognitive dimension, thus obtaining the task-aware features for each task cognitive dimension. Through the above weighting, the task state and constraint information contained in the latent semantic features can be combined with the feasibility level of the corresponding task cognitive dimension.

[0147] During the shared feature space mapping process, the feature mapping layer performs feature dimension transformation and semantic alignment on the task perception features corresponding to different task cognitive dimensions, so that the task perception features can be compared and fused in a unified shared feature space.

[0148] Based on the task objectives, stages, priorities, and expected completion status of the task to be executed, and in conjunction with the embodied state, capability boundaries, task execution progress, and node role of the unmanned system node, contextual features are determined. These contextual features reflect the relative importance of each cognitive dimension of the task in the current task intent and node operating state.

[0149] During attention fusion, the contextual features are used as the query basis to dynamically weight and fuse the task-perceptual features in the shared feature space, resulting in fused features with task priority attributes. The contribution of different task cognitive dimensions to the fused features can dynamically change with task stage, environmental state, and node capability state.

[0150] During the candidate collaboration information generation process, the fused features are subjected to feature compression, data encapsulation, or format conversion to generate candidate collaboration information for the unmanned system nodes. This candidate collaboration information can characterize collaborative decision-making tendencies such as independent execution, request collaboration, degraded waiting, role succession, or topology adjustment, and serves as input for subsequent distributed consensus solving.

[0151] S1074. Generate candidate collaborative information for unmanned system nodes based on fusion features.

[0152] Specifically, the fused features can be compressed, data encapsulated, or format converted to generate candidate collaboration information corresponding to the unmanned system nodes. The candidate collaboration information may include the node identifier, fused features, task priority information, and collaboration trustworthiness of the unmanned system nodes. It may also include decision-making tendency information such as continuing to execute the current task, collaborating with other nodes to execute tasks, assuming a substitute role, providing functional compensation, or adjusting existing collaboration relationships.

[0153] By compressing the results of multidimensional task cognition into candidate collaborative information, each node of the unmanned system can provide information related to task feasibility, potential task semantics, and task priority to the subsequent distributed consensus solution process without directly exchanging original images, point clouds, or complete high-dimensional task feature data.

[0154] In this embodiment, the latent semantic features are first weighted using the task feasibility confidence corresponding to each task cognitive dimension, combining task semantic information with the corresponding feasibility. Then, the task perception features of different task cognitive dimensions are mapped to a shared feature space, and context features are determined based on the task intent of the task to be executed and the operating state of the unmanned system nodes. This enables the identification of task cognitive dimensions with higher importance in the current task stage based on a unified feature representation. Furthermore, attention fusion is performed using context features, so that the resulting fused features have a task priority attribute that dynamically changes with the task state. Finally, the fused features are compressed and encapsulated into candidate collaborative information, thereby reducing the amount of data transmitted and processed for subsequent inter-node collaborative solving while retaining multi-dimensional task cognitive semantics and task feasibility information, and providing reliable input for forming a global consensus result that matches the current task intent and node operating state.

[0155] In one possible embodiment, the method steps shown in S108 are implemented by S1081 to S1084, which are described in detail below.

[0156] S1081. Based on the candidate collaborative information generated by each unmanned system node, determine the local decision variables of each unmanned system node, and construct the local objective function corresponding to each local decision variable.

[0157] Specifically, for the i-th unmanned system node in the unmanned system, the candidate collaborative information generated by the unmanned system node can be parsed, and the fusion features, task priority information, task feasibility information, node capability status, and collaborative decision-making tendency can be converted into local decision variables of the unmanned system node. .

[0158] Local decision variables The local collaborative decision made by the i-th unmanned system node in response to the task to be executed can include one or more of the following: task assignment status, candidate node role, candidate functional status, communication relay relationship, or candidate topology connection relationship. Local decision variables of different unmanned system nodes use the same variable dimension and data representation to facilitate comparison and updating during subsequent distributed consensus solving.

[0159] Based on candidate collaborative information, a model can be constructed that relates to local decision variables. Corresponding local objective function The local objective function is used to evaluate the degree of matching between local decision variables and the current candidate collaborative information of the i-th unmanned system node. The smaller the value of the local objective function, the more the corresponding local decision aligns with the current task feasibility, task priority, and node operating status of the unmanned system node.

[0160] In some embodiments, the local objective function can comprehensively characterize one or more of the following: feature deviation between local decisions and candidate collaborative information, task execution cost, role switching cost, communication cost, and capability boundary constraint cost. For local decisions that do not meet the capability boundaries or task feasibility requirements of unmanned system nodes, a larger cost can be set in the local objective function, thereby reducing the likelihood that the corresponding local decision will be adopted in the subsequent global consensus results.

[0161] S1082. Based on the consistency constraints among local decision variables and the local objective functions, construct a distributed consensus optimization problem.

[0162] Specifically, the local decision variables of each unmanned system node can be constrained to be consistent with the same global consensus variable, and the following distributed consensus optimization problem can be constructed: ;in, This represents the number of unmanned system nodes participating in the distributed consensus solution in the unmanned system. Indicates the sequence number of the unmanned system node; Represents a global consensus variable; This represents the consistency constraint between the local decision variables and the global consensus variables of the i-th unmanned system node.

[0163] The aforementioned distributed consensus optimization problem aims to minimize the sum of the local objective functions while requiring that the local decision variables of each unmanned system node converge to the same global consensus variable. Thus, each unmanned system node can retain its local decision-making basis formed by local candidate collaborative information, and can also form a unified collaborative decision for the entire unmanned system through consistency constraints.

[0164] To solve the distributed consensus optimization problem using the alternating direction multiplier method, the following augmented Lagrangian function can be constructed: ;in, This represents the augmented Lagrangian function corresponding to the distributed consensus optimization problem. Let represent the dual variable corresponding to the consistency constraint of the i-th unmanned system node; This indicates the transpose of the dual variable; Indicates a penalty parameter that is greater than zero; The L2 norm squared represents the difference between local decision variables and global consensus variables.

[0165] in, This is used to ensure that the local decisions of the i-th unmanned system node match its candidate collaborative information. Used to introduce consistency constraints between local decision variables and global consensus variables. It is used to penalize the degree to which local decision variables deviate from global consensus variables.

[0166] S1083. Using the alternating direction multiplier method, iteratively update each local decision variable, global consensus variable, and dual variable, and determine the original residual based on the updated local decision variables and global consensus variables, and determine the dual residual based on the global consensus variables of adjacent iterations.

[0167] Specifically, the Alternating Direction Method of Multipliers (ADMM) is used. Let t represent the current iteration number. The local decision variables, global consensus variables, and dual variables can be updated sequentially as follows.

[0168] First, update the local decision variables of the i-th unmanned system node according to the formula: ;in, This represents the local decision variable obtained by the i-th unmanned system node in the (t+1)-th iteration; This represents the dual variable corresponding to the i-th unmanned system node in the t-th iteration; This represents the global consensus variable corresponding to the t-th iteration; This represents the local decision variable that minimizes the objective function within the parentheses.

[0169] Each unmanned system node can execute the local decision variable update process locally. During the update, the i-th unmanned system node only needs to use its own local objective function, current dual variable, and current global consensus variable, without needing to obtain the original environmental perception data or complete task feature data from other unmanned system nodes.

[0170] Then, according to the formula Update the global consensus variables; where, Let represent the global consensus variable obtained in the (t+1)th iteration. By summarizing the updated local decision variables and corresponding dual corrections of each unmanned system node, the global consensus variable can simultaneously reflect the local collaborative decisions of different unmanned system nodes.

[0171] Furthermore, according to the formula Update the dual variable corresponding to the i-th unmanned system node; where, Let represent the dual variable obtained by the i-th unmanned system node in the (t+1)-th iteration. The dual variable is adjusted based on the difference between the updated local decision variables and the global consensus variable, and is used to further constrain the local decision variables to converge to the global consensus variable in the next iteration.

[0172] After updating the local decision variables, global consensus variables, and dual variables for the current iteration, the original residuals can be determined based on the differences between the updated local decision variables and the global consensus variables. The original residuals characterize the degree of consistency between the local decision variables of each unmanned system node and the current global consensus variables; the smaller the difference between the local decision variables and the global consensus variables, the smaller the original residuals, indicating that the local collaborative decisions of the unmanned system nodes tend to be consistent.

[0173] The dual residuals can be determined based on the changes in the global consensus variables between adjacent iterations. The dual residuals characterize the stability of the global consensus variables between adjacent iterations; the smaller the changes in the global consensus variables between adjacent iterations, the smaller the dual residuals, indicating that the global consensus variables tend to be more stable.

[0174] During the iteration process, each unmanned system node only needs to exchange low-dimensional information such as local decision variables, dual variables, or global consensus variables, without exchanging original images, point clouds, or complete task feature data. This reduces the amount of communication data between nodes and adapts to collaborative solution processes under conditions of limited bandwidth, link interference, or weak communication.

[0175] S1084. Stop the iteration when the original residual and the dual residual satisfy the corresponding preset convergence conditions, and obtain the global consensus result based on the global consensus variables at the time of stopping the iteration.

[0176] Specifically, preset convergence accuracies can be set for the original residuals and dual residuals respectively. After each iteration, the current original residuals and dual residuals are compared with the corresponding preset convergence accuracies. When the original residuals meet the corresponding preset convergence accuracies, it indicates that the local decision variables of each unmanned system node have reached a preset level of consistency with the current global consensus variables; when the dual residuals meet the corresponding preset convergence accuracies, it indicates that the global consensus variables have reached a preset level of stability between adjacent iterations.

[0177] When both the original residual and the dual residual meet the corresponding preset convergence accuracy, the iterative updates of the local decision variables, global consensus variables, and dual variables are stopped, and the global consensus variable at the point of stopping iteration is taken as the global consensus result. The global consensus result is used to characterize the unified collaborative decision made by each unmanned system node in the unmanned system for the task to be executed, and may include one or more of the following: unified task allocation result, node role selection result, functional collaboration result, or topology relationship selection result.

[0178] When the original residual or dual residual has not yet met the corresponding preset convergence accuracy, continue to iterate and update each local decision variable, global consensus variable and dual variable in the manner described above until the original residual and dual residual both meet the corresponding preset convergence accuracy, or reach the preset maximum number of iterations.

[0179] like Figure 6 As shown, the collaborative reconstruction process includes state and task cognition processing, candidate collaborative information generation, distributed consistency solution, and collaborative reconstruction execution.

[0180] During the state and task cognition processing, each unmanned system node constructs embodied state data based on the acquired multidimensional state data, determines hardware variability, software adaptability, and energy affordability, and obtains the ontological capability boundary through nonlinear coupling dynamic calculation. Using the ontological capability boundary as a task cognition constraint, masked reconstruction deduction is performed on multiple task cognition dimensions to obtain the task cognition results for each dimension.

[0181] During the candidate collaborative information generation process, the cognitive results of each task are fused to obtain candidate collaborative information corresponding to each unmanned system node. Each unmanned system node determines local decision variables based on its own candidate collaborative information and constructs a local objective function corresponding to the local decision variables.

[0182] In the distributed consensus solution process, a distributed consensus optimization problem is constructed based on the consistency constraints among local decision variables and the local objective functions. The alternating direction multiplier method is then used to iteratively update the local decision variables, global consensus variables, and dual variables. The original residuals are determined based on the updated local decision variables and global consensus variables, and the dual residuals are determined based on the global consensus variables from adjacent iterations.

[0183] The iteration stops when both the original residual and the dual residual satisfy their respective preset convergence conditions, and the global consensus variable at the time of iteration stop is taken as the global consensus result. When the global consensus result satisfies the preset reconstruction trigger condition, a collaborative reconstruction instruction is generated based on the global consensus result.

[0184] During the collaborative reconfiguration process, the collaborative reconfiguration instructions are sent to the relevant unmanned system nodes. Each unmanned system node performs role takeover, function switching, degraded operation, communication relationship adjustment, or network topology reconfiguration according to the collaborative reconfiguration instructions. The adjusted node status is fed back into the multi-dimensional status data acquisition process, enabling each unmanned system node to continue performing capability assessment, task recognition, distributed consistency solution, and collaborative reconfiguration, thereby forming a continuously updated closed-loop collaborative reconfiguration process.

[0185] In this embodiment, candidate collaborative information formed by each unmanned system node based on its local task cognition is first converted into local decision variables and local objective functions. Then, a distributed consensus optimization problem is constructed through consistency constraints between local decision variables and global consensus variables. The alternating direction multiplier method is used to alternately update local decision variables, global consensus variables, and dual variables, so that the local collaborative decisions of different unmanned system nodes gradually converge while preserving their local capability status and task judgment basis. Furthermore, the original residuals are used to determine the degree of consistency between each local decision variable and the global consensus variable, and the dual residuals are used to determine the stability of the global consensus variable between adjacent iterations. The global consensus result is obtained when both types of residuals meet the preset convergence conditions. Thus, each unmanned system node can form a convergent unified collaborative decision under conditions of bandwidth limitation, communication interference, or partial link anomalies without relying on a single central node to centrally process all the original data. This reduces the communication burden, decision latency, and single point of failure risk caused by centralized decision-making methods, and provides a reliable basis for subsequent reconstruction of node roles, functions, or topology relationships based on the global consensus results.

[0186] In one possible embodiment, the method steps shown in S104 are implemented by S1041 to S1045, which are described in detail below.

[0187] S1041. Based on hardware variability, software adaptability, and energy affordability, construct an ontology capability state tensor that includes hardware variability, software adaptability, and energy affordability.

[0188] Specifically, hardware variability, software adaptability, and energy affordability at the same moment can be organized according to a unified data scale to construct the following ontology capability state tensor: ;in, Represents the ontology capability state tensor at time t; This indicates the hardware variability at time t; This represents the software adaptability at time t; This represents the energy affordability at time t.

[0189] The ontological capability state tensor is used to jointly represent the current hardware capabilities, software operational capabilities, and energy availability capabilities of unmanned system nodes in the same state space. Since hardware variability, software adaptability, and energy affordability can have different numerical ranges, these three can be normalized or scaled separately before constructing the ontological capability state tensor, so that each capability component can be dynamically solved in a unified state space.

[0190] The ontological capability state tensor can be continuously updated with the embodied state data of unmanned system nodes. For example, when the performance of the actuator degrades, the computing task queue becomes congested, or the remaining state of charge of the battery decreases, the corresponding hardware variability, software adaptability, or energy affordability change accordingly, thus forming a sequence of ontological capability state tensors that are updated over time.

[0191] like Figure 3 As shown, the hardware variability, software adaptability, and energy affordability at the same moment are first scaled and aligned, and then the ontological capability state tensor is constructed according to a preset arrangement. Among them, hardware variability is used to characterize the available motion capability of the physical carrier and actuator in the current healthy operating state, software adaptability is used to characterize the real-time support capability of the current computing resources and task scheduling state for the task to be executed, and energy affordability is used to characterize the ability of the current energy state to ensure the continuous execution of the task.

[0192] A nonlinear evolution model of ontological capabilities is established based on the ontological capability state tensor to characterize the cross-domain coupling relationship between hardware variability, software adaptability, and energy affordability. For example, a decline in actuator performance may lead to an increase in control compensation calculations, and an increase in computational load may accelerate energy consumption. Therefore, changes in any capability component may have a cascading effect on other capability components.

[0193] Using the current ontological capability state tensor as the initial state, and combining it with control inputs, task loads, and state influence quantities, the nonlinear evolution model of ontological capabilities is solved, yielding an ontological capability evolution trajectory composed of ontological capability state tensors at multiple prediction times. Through this ontological capability evolution trajectory, the combined changing trends of hardware variability, software adaptability, and energy affordability over time can be predicted.

[0194] Further stability analysis of the capability evolution trajectory is conducted to determine the capability dissipation safety boundary. The capability dissipation safety boundary is used to distinguish between stable capability regions that can maintain mission execution and risky regions where rapid capability decay or instability may occur.

[0195] Finally, based on the ontological capability evolution trajectory and capability dissipation safety boundary, the dynamic dynamic envelope, computational redundancy boundary, and energy decay boundary are determined respectively. The dynamic dynamic envelope defines the range of motion states that unmanned system nodes can safely reach; the computational redundancy boundary defines the range of computational resources that can be further allocated; and the energy decay boundary defines the energy, power, endurance, or mission execution range that can be safely used. These three elements together constitute the ontological capability boundary and serve as task cognitive constraints for subsequent task cognitive deduction.

[0196] S1042. Based on the ontological capability state tensor, establish an ontological capability nonlinear evolution model. The ontological capability nonlinear evolution model is used to characterize the cross-domain coupling relationship between hardware variability, software adaptability and energy affordability.

[0197] Specifically, the following ontological evolution equation can be established: ;in, This represents the rate of change of the ontological capability state tensor at time t; This indicates the evolutionary effect of external control and task stimuli on the state of the individual's capabilities; This represents the control or task input applied to the unmanned system node at time t; This represents the cross-domain coupling perturbation between hardware state, software operating state, and energy state. This represents the state information used to determine cross-domain coupled disturbances.

[0198] Cross-domain coupled disturbances are used to describe the cascading effects of a change in one capability component on other capability components. For example, when hardware variability is reduced due to actuator damage, the control program may increase the calculation frequency or increase the amount of compensation calculation in order to maintain attitude stability or motion accuracy, thereby crowding out computing resources and changing software adaptability; the increase in computing load and execution output may also accelerate energy consumption, further reducing energy affordability.

[0199] In some embodiments, the evolution of energy affordability satisfies: ;in, This represents the rate of change in energy affordability at time t; This represents the current at time t; This represents the voltage at time t; This represents the basic electrical energy consumption generated during the current operation of an unmanned system node; This represents the coefficient indicating the impact of hardware variability on additional energy consumption. This represents the non-linear growth coefficient of additional energy consumption as hardware variability decreases. This represents the coefficient indicating the impact of software adaptability on additional energy consumption, and , and All are greater than zero.

[0200] As can be seen from the above energy evolution formula, when hardware variability When it decreases, (1- As hardware performance increases, the additional energy consumption caused by declining hardware performance increases exponentially; the impact of software performance on energy consumption is determined by... The corresponding terms are characterized. Therefore, changes in hardware performance, software resources, and energy consumption can be incorporated into the same ontological capability nonlinear evolution model to characterize the cross-domain cascading relationship between hardware capability decline, increased computational compensation, and accelerated energy consumption.

[0201] S1043. Based on the nonlinear evolution model of the ontological capability, the coupled change trend of hardware variability, software adaptability and energy affordability is predicted to obtain the ontological capability evolution trajectory.

[0202] Specifically, the current state tensor of the ontological capability can be used as the initial state of the ontological capability nonlinear evolution model. The current control input, the task load corresponding to the task to be executed, and the state influence quantity within the prediction time range are input into the ontological capability nonlinear evolution model to solve the change process of the ontological capability state tensor over time.

[0203] The prediction timeframe can be determined based on the expected duration of the task to be executed, the control cycle, or the collaborative decision-making cycle. By continuously solving or discretely iterating the ontology evolution equation within the prediction timeframe, the hardware variability, software adaptability, and energy affordability corresponding to multiple prediction moments can be obtained, thereby forming the ontology capability evolution trajectory.

[0204] The evolution trajectory of an unmanned system's capabilities can be represented as a sequence of trajectories composed of multiple capability state tensors arranged in chronological order. This sequence reflects the combined changing trends of hardware capabilities, software operational capabilities, and energy availability as unmanned system nodes continue executing their current tasks or receive new control inputs. For example, the evolution trajectory can reflect whether software adaptability continues to decline after hardware variability decreases, and whether energy affordability experiences accelerated decay.

[0205] In some embodiments, a nonlinear observer can be used in conjunction with real-time power consumption, battery temperature, and remaining state of charge to make online corrections to the state variables or model parameters of the nonlinear evolution model of the capability, so that the capability evolution trajectory can be dynamically updated with the real-time operating status of the unmanned system nodes.

[0206] S1044. Perform stability analysis on the evolution trajectory of the entity's capabilities to determine the capability dissipation safety boundary corresponding to the evolution trajectory of the entity's capabilities.

[0207] Specifically, the Lyapunov second method can be used to perform stability analysis on the evolution trajectory of the capability. A Lyapunov function that characterizes the degree to which the capability state tensor deviates from the safe operating state can be selected, and the variation trend of the Lyapunov function along the capability evolution trajectory can be determined based on the nonlinear evolution model of the capability.

[0208] When the Lyapunov function and its changing trend indicate that the entity capability state can be maintained within the preset safe state range, or can gradually return to the safe working state, it can be determined that the corresponding entity capability evolution trajectory is in the stable region; when the changing trend of the Lyapunov function indicates that the entity capability state continues to move away from the safe working state, or that the cascading changes between hardware variability, software adaptability and energy affordability may lead to rapid dissipation of comprehensive capabilities, it can be determined that the corresponding entity capability evolution trajectory approaches or crosses the capability dissipation safety boundary.

[0209] The safety boundary for capability dissipation can be represented by a safety threshold surface. The safety threshold surface separates the stable capability region, which can maintain mission execution, from the risk region where capability instability or rapid degradation may occur. The safety threshold surface can be predetermined based on the unmanned system node's equipment type, mission type, historical operational data, and safety operation requirements, or it can be updated in real-time based on changes in the system's capability evolution trajectory.

[0210] When the predicted trajectory of the system's capability evolution crosses the capability dissipation safety boundary, it can form hard constraint information guided by survival and self-preservation, indicating that the unmanned system node should not continue to operate according to the current task load or control mode; when the trajectory of the system's capability evolution does not cross the capability dissipation safety boundary, the current capability margin that the unmanned system node can safely use can be determined based on the distance between the trajectory and the capability dissipation safety boundary.

[0211] S1045. Based on the ontological capability evolution trajectory and capability dissipation safety boundary, determine the dynamic dynamic envelope, computing power redundancy boundary and energy decay boundary to obtain the ontological capability boundary.

[0212] Specifically, the predicted changes in hardware variability can be extracted from the evolution trajectory of the system's capabilities. Combined with the capability dissipation safety boundary, the range of motion states that an unmanned system node can safely reach under the current hardware health condition can be determined, thereby obtaining the dynamic dynamic envelope. The dynamic dynamic envelope can limit the range of motion parameters such as the maximum bank angle, maximum level flight speed, limiting yaw acceleration, dynamic turning radius, or other motion parameters adapted to the type of unmanned system node.

[0213] The predictive change process of software adaptability can be extracted from the evolution trajectory of the ontology's capabilities. Combined with the current task queue, remaining computing resources, and task timeliness requirements, the range of computing resources that can continue to be allocated while ensuring the real-time performance of control instructions and the timeliness of task computation can be determined, thus obtaining the computing power redundancy boundary. When the predicted computing load reaches or exceeds the computing power redundancy boundary, the addition of new computing tasks can be restricted, or insufficient computing power constraint information can be provided for the subsequent task cognitive process.

[0214] The predictive change process of energy affordability can be extracted from the evolution trajectory of the system's capabilities. Combined with the capability dissipation safety boundary, the remaining energy, endurance time, or power range that the unmanned system node can safely use under the current power consumption, battery temperature, and remaining charge state can be determined, thereby obtaining the energy decay boundary.

[0215] The dynamic dynamic envelope, computational redundancy boundary, and energy decay boundary together constitute the ontological capability boundary. This boundary is dynamically updated along with the ontological capability evolution trajectory and the capability dissipation safety boundary, serving as an insurmountable task cognitive constraint during subsequent task cognitive deduction. When the ontological capability evolution trajectory crosses the capability dissipation safety boundary, the dynamic dynamic envelope, computational redundancy boundary, or energy decay boundary can be correspondingly contracted, ensuring that subsequent task executability judgments are based on the currently available physical, computational, and energy capabilities of the unmanned system node.

[0216] In this embodiment, hardware variability, software adaptability, and energy affordability are constructed as a unified ontological capability state tensor. A nonlinear evolution model of ontological capability is used to characterize the cross-domain coupling relationship among these three elements, ensuring that changes in software resources and energy consumption caused by hardware capability variations are reflected in the same dynamic evolution process. Based on this, the ontological capability evolution trajectory is predicted, and stability analysis is performed using Lyapunov's second method to determine the capability dissipation safety boundary, thereby forming a dynamic dynamic envelope, a computational redundancy boundary, and an energy decay boundary. Thus, the resulting ontological capability boundary is no longer a static capability result determined based on rated performance or a fixed threshold, but can be dynamically updated with the coupled changes in hardware state, software operating state, and energy state. This improves the accuracy of the actual capability state assessment of unmanned system nodes and provides hard constraints that match the actual capabilities of nodes for subsequent task cognitive inference.

[0217] In one possible embodiment, the method steps shown in S103 are implemented by S1031 to S1033, which are described in detail below.

[0218] S1031. Based on the execution instructions and corresponding execution feedback of the unmanned system node, determine the operational health status of the actuator, and based on the operational health status and the dynamic constraints of the unmanned system node, determine the range of available motion capabilities of the unmanned system node to obtain hardware variability.

[0219] Specifically, execution commands and feedback can be collected from each actuator in the unmanned system node. Execution commands can include target rotational speed, target torque, target position, or target attitude control parameters for each actuator, while execution feedback can include the actual rotational speed, actual torque, actual position, vibration status, temperature status, or output response status of each actuator.

[0220] When the actuators are multi-rotor, motor, servo, or hub-driven mechanisms, the real-time speed feedback of each actuator can be organized into a real-time speed feedback vector. The corresponding system-given instructions are organized into a given instruction vector. Using an extended Kalman filter (EKF) and a residual observer, the response deviation between the real-time speed feedback vector and the given command vector is estimated to obtain the health status of each actuator, and an actuator health status matrix is ​​constructed. ;in, Represents the health matrix of the implementing agency; This represents the operation of constructing a diagonal matrix; These represent the health status of the 1st to nth actuators, respectively; n represents the number of actuators participating in the hardware variability measurement. The closer the actual response of an actuator is to the corresponding execution instruction, the higher its health status; when an actuator experiences wear, jamming, output attenuation, or local damage, its health status decreases.

[0221] The actuator health matrix can be applied to the nominal Jacobian matrix of the unmanned system node under normal hardware conditions to reconstruct the damaged Jacobian matrix: ;in, Represents the damaged Jacobian matrix; The nominal Jacobian matrix represents the node of an unmanned system in the normal state of its actuator; This represents matrix operations. The damaged Jacobian matrix is ​​used to characterize the mapping relationship between the actuator output and the overall motion state of the unmanned system nodes after the actuator's operational health status changes.

[0222] Based on the damaged Jacobian matrix and the dynamic constraints of the unmanned system node, a system of differential-algebraic equations containing six-degree-of-freedom motion constraints can be solved to redetermine the range of motion parameters that the unmanned system node can safely achieve under the current hardware conditions. The range of motion parameters may include one or more of the following: maximum bank angle, maximum level flight speed, limiting yaw acceleration, and dynamic turning radius. It may also include, depending on the equipment type of the unmanned system node, maximum climb rate, maximum braking acceleration, maximum load, or other motion capability parameters.

[0223] The redefined ranges of motion parameters are combined to obtain the available motion capability range of the unmanned system node, and hardware variability is derived based on this range. Hardware variability can be represented by a capability vector containing the upper and lower limits of each motion parameter, or by normalized values ​​or motion capability regions. Compared to the fixed motion capability determined based on the equipment's rated parameters, hardware variability reflects the actual usable motion capability of the unmanned system node after minor damage, performance degradation, or local anomalies to the actuator.

[0224] S1032. Based on the task queue status, computing resource occupancy status, and resource requirements and timeliness requirements of the unmanned system node for the tasks to be executed, determine the real-time execution capability and computing resource reserve of the unmanned system node for the tasks to be executed, and obtain the software adaptability based on the real-time execution capability and computing resource reserve.

[0225] Specifically, the task queue status can be obtained at the kernel layer of the Real-Time Operating System (RTOS) of the unmanned system node. The task queue status can include the number of tasks in the ready and blocked states, task priorities, task waiting times, and task queue depth. The computing resource occupancy status can include one or more of the following: CPU utilization, GPU utilization, memory utilization, number of computing threads, and algorithm execution latency.

[0226] A spatiotemporal distribution matrix of computing resources can be constructed based on the task queue status and computing resource occupancy status to represent the allocation of computing resources across different processors, task threads, and time periods. This is applicable to the set of algorithms running in unmanned system nodes. A queuing model can be established, and the utilization rate of computing resources and the probability of completing the task within the time limit can be determined according to the following formula: , ;in, Indicates the utilization rate of computing resources; Indicates the task completion rate; Indicates the task service rate; Indicates the estimated execution time of the task to be executed; This indicates the deadline for the timeliness requirement of the task to be executed; This represents the probability that the task to be executed will be completed before the deadline; e represents the natural constant.

[0227] Based on the processor resources, graphics processing resources, storage resources, and algorithm execution time required by the task to be executed, the resource requirements of the task can be determined, and the resource requirements can be matched with the current spatiotemporal distribution matrix of computing resources. The higher the probability that the task to be executed will be completed before the deadline, the stronger the real-time execution capability of the unmanned system node for the task to be executed; the larger the difference between the currently available computing resources and the computing resources required by the task to be executed, the larger the computing resource margin.

[0228] In some embodiments, a preset safety threshold can be set corresponding to the probability that a task will be completed before the deadline. When the probability is lower than the preset safety threshold, it indicates that the current task queue and computing resource occupancy status cannot guarantee the timeliness requirements of the tasks to be executed, and the software adaptability can be reduced accordingly. Software degradation contingency plans can also be generated, such as switching high-computing-power sensing threads to low-frequency inertial navigation threads or lightweight obstacle avoidance threads, to release computing resources and ensure the real-time performance of control commands. Software degradation contingency plans can be used as part of software adaptability to characterize the likelihood that unmanned system nodes can restore real-time execution capabilities by adjusting algorithm configurations.

[0229] Software adaptability can be obtained by comprehensively considering real-time execution capabilities, remaining computing resources, and available software degradation methods. Software adaptability can be represented by probability values, normalized values, resource reserve vectors, or software configuration levels, and is used to characterize the ability of unmanned system nodes to execute tasks under the current software operating state and adapt to changes in computing resources.

[0230] S1033. Based on the real-time power consumption, battery temperature and remaining state of charge of the unmanned system node, determine the available range of the unmanned system node, and obtain the energy affordability based on the available range.

[0231] Specifically, real-time voltage and current can be obtained through the energy management interface of the unmanned system node, and real-time power consumption can be determined based on the real-time voltage and current; cell temperature or battery pack temperature can be obtained through battery temperature sensors; and the remaining state of charge (SOC) can be obtained through the battery management system. The SOC is used to characterize the ratio of the current remaining battery capacity to the rated capacity.

[0232] Real-time power consumption, battery temperature, and remaining state of charge can be input into a nonlinear observer to estimate the remaining available energy and remaining operating time under the current continuous task load conditions. Specifically, the nonlinear observer can adjust the energy consumption rate based on real-time power consumption changes, adjust the battery's effective capacity and discharge efficiency based on battery temperature, and determine the energy that can continue to be released based on the remaining state of charge.

[0233] When determining the available endurance range, the current speed of the unmanned system node, the expected mission power consumption, the energy required for return or withdrawal, and the preset safety energy margin can be considered to determine the time range, distance range, or mission phase range within which the unmanned system node can continue to operate. For example, for a drone, the available endurance range can include the remaining safe flight time, the remaining safe flight distance, and the maximum mission depth required to meet the return requirements; for an unmanned vehicle, the available endurance range can include the remaining driving time, the remaining driving range, and the range of mission load it can continue to undertake.

[0234] When the battery temperature is outside the normal range, real-time power consumption continues to rise, or the remaining state of charge continues to decrease, the usable driving range can be reduced accordingly. When the real-time power consumption is stable, the battery temperature is within the normal range, and the remaining state of charge is sufficient, the usable driving range can be increased accordingly. Therefore, the usable driving range is not only statically calculated based on the remaining state of charge, but also comprehensively reflects the combined impact of real-time power consumption, battery temperature, and remaining state of charge on driving range.

[0235] Energy affordability can be obtained based on available endurance. Energy affordability can be represented by one or more of the following: remaining safe operating time, remaining safe operating distance, remaining available energy, manageable task load range, or normalized capability value, to characterize the ability of an unmanned system node to continue executing pending tasks under the current energy condition.

[0236] It should be noted that this step is used to determine the energy affordability corresponding to the current energy state. The nonlinear coupling evolution process between energy affordability and hardware variability and software adaptability, as well as the dynamic change process of the energy state affected by hardware performance degradation and software operating load, are solved in the embodiment corresponding to S104, and will not be repeated here.

[0237] In this embodiment, the operational health status of each actuator is first determined by the response relationship between execution instructions and execution feedback. The available mobility range of the unmanned system node is then recalculated using the actuator health matrix and the damaged Jacobian matrix, ensuring that the resulting hardware variability reflects the impact of local hardware damage on actual mobility. Secondly, the probability of task completion within the deadline and the remaining computing resources are determined by utilizing task queue status, computing resource occupancy status, and the resource and timeliness requirements of pending tasks. This ensures that the resulting software adaptability reflects the real-time execution capability under the current computing load and the ability to release computing resources through software configuration adjustments. Finally, the available range is predicted by comprehensively considering real-time power consumption, battery temperature, and remaining state of charge, ensuring that the resulting energy affordability reflects the actual task support capability under the current energy state. Therefore, the current actual capability of the unmanned system node can be dynamically measured from three aspects: hardware, software, and energy. This avoids judging node capability solely based on rated hardware parameters, static computing resource indicators, or a single remaining battery power, providing accurate basic capability data for subsequent nonlinear coupling calculations between the three types of capabilities and the determination of the entity's capability boundaries.

[0238] In one possible embodiment, the method steps shown in S102 are implemented by S1021 to S1023, which are described in detail below.

[0239] S1021. Based on the acquisition time corresponding to each multidimensional state data, time synchronization is performed on each multidimensional state data to obtain time synchronization data.

[0240] Specifically, the multidimensional state data can be generated by heterogeneous sensor networks, device control buses, energy management interfaces, and real-time operating systems located in the nodes of the unmanned system. The heterogeneous sensor network may include LiDAR (Light Detection and Ranging), depth cameras, millimeter-wave radar, inertial measurement units (IMU), and Global Navigation Satellite System (GNSS); the device control bus may include a Controller Area Network (CAN) bus; and the operating system may be a Real-Time Operating System (RTOS).

[0241] The aforementioned data acquisition components have different sampling frequencies and data output periods. For example, environmental perception data can be generated according to image frames, point cloud frames, or radar scan cycles; body operation data can be generated according to motion control cycles; computing resource data can be generated according to task scheduling cycles; and communication status data and energy status data can also be generated according to their respective monitoring cycles. Therefore, when acquiring various multi-dimensional status data, a corresponding acquisition time can be attached to each data frame or data record.

[0242] A unified time base can be determined from the acquisition time of various multidimensional state data, and data from different data sources can be sorted and matched according to the unified time base. Data with the same acquisition time or within the preset time tolerance range can be grouped into the same time synchronization data group; for data with different sampling frequencies, the data with the closest time can be selected according to the target sampling time, or the data corresponding to the target sampling time can be determined based on the data at adjacent sampling times.

[0243] When no new sampled values ​​are generated for some data at the target sampling time, the most recent valid data can be used within a preset valid duration. When the data loss time exceeds the preset valid duration, the corresponding data can be marked as missing or low-confidence. Through the above processing, environmental perception data, system operation data, computing resource data, communication status data, and energy status data can be synchronized according to a unified collection time.

[0244] S1022. Based on the unified spatial reference system of unmanned system nodes, spatial alignment is performed on the data with spatial attributes in the time synchronization data, and according to the corresponding acquisition time, the data without spatial attributes is associated with the spatially aligned data to obtain spatiotemporal registration data.

[0245] Specifically, a unified spatial reference system can be established based on the equipment type and mission scenario of the unmanned system nodes. The unified spatial reference system can be the unmanned system node's body coordinate system, local mission coordinate system, or global geographic coordinate system.

[0246] For data with spatial attributes, data from different data sources can be converted to a unified spatial reference frame based on the installation location and attitude of the corresponding data acquisition components on the unmanned system node, as well as the pose of the unmanned system node at the corresponding acquisition time. Data with spatial attributes can include lidar point clouds, spatial points in depth images, millimeter-wave radar target positions, attitude information determined by the inertial measurement unit, positions determined by the global navigation satellite system, and the motion state of dynamic targets.

[0247] Spatial alignment enables traversable areas, obstacle boundaries, target positions, target semantics, and dynamic target attributes detected by different sensors to correspond to each other within the same spatial reference frame. For example, LiDAR point clouds and depth camera data can be aligned to the body coordinate system of unmanned system nodes, and further transformed to the local task coordinate system based on the node pose at the corresponding acquisition time, thus forming a unified environmental geometry and spatial topology.

[0248] For data lacking spatial attributes, its state can be correlated with spatially aligned data according to the corresponding acquisition time. Data without spatial attributes can include task queue status, computing resource occupancy status, communication link quality, actuator health status, battery temperature, real-time power consumption, and remaining state of charge. After state correlation, the environmental spatial state, body motion state, computing resource state, communication state, and energy state at the same acquisition time are organized into the same spatiotemporal state record, thereby obtaining spatiotemporal registration data.

[0249] S1023. Perform noise suppression, feature extraction and normalization on the spatiotemporal registration data, and fuse the processed features to obtain the embodied state data of the unmanned system node.

[0250] Specifically, noise suppression can be performed on spatiotemporal registration data based on the noise characteristics of different types of data. For environmental perception data, outliers, invalid points, duplicate targets, or obviously abnormal measurement results can be removed; for system operation data, abnormal fluctuations caused by mechanical vibration, sensor drift, or transient interference can be suppressed; for computing resource data, communication status data, and energy status data, abnormal jump data can be identified, and valid data that reflects the true state changes of nodes can be retained.

[0251] After noise suppression is completed, feature extraction can be performed based on data type. Specifically, semantic segmentation can be performed on image data to extract passable areas, obstacle boundaries, target semantics, and dynamic target attributes; point cloud data can be voxelized to convert spatially irregular point clouds into voxel features with a uniform spatial scale; node pose, velocity, acceleration, executed instructions, and execution feedback features can be extracted from ontology operation data; task queue, resource usage, and algorithm execution latency features can be extracted from computing resource data; communication link quality, transmission latency, and network connectivity features can be extracted from communication status data; and real-time power consumption, battery temperature, and remaining state of charge features can be extracted from energy status data.

[0252] Since different types of features have different dimensions and numerical ranges, the extracted features can be normalized. As one possible approach, energy state is normalized using the following formula: ;in, Indicates the normalized energy state; This represents the energy state value currently collected by the unmanned system node; This represents the minimum value corresponding to the energy state; This represents the maximum value corresponding to the energy state.

[0253] By performing extreme value normalization as described above, energy state can be mapped to a unified numerical range, reducing the dimensional differences between energy state and environmental characteristics, system operational characteristics, computing resource characteristics, and communication state characteristics. For other numerical characteristics, extreme value normalization can also be performed based on the value range of the corresponding characteristic, or standardization processing appropriate to the characteristic type can be adopted.

[0254] After noise suppression, feature extraction and normalization are completed, environmental semantic features, geometric topological features, dynamic state features, computing resource features, communication state features and energy state features can be spliced ​​or correlated and fused according to a unified feature arrangement order to form the embodied state data of unmanned system nodes.

[0255] Embodied state data can be represented using embodied state vectors, feature matrices, or state tensors, and associated with the corresponding unmanned system node identifier and acquisition time. Embodied state data is used not only to represent targets and obstacles in the external environment of the unmanned system node, but also to simultaneously represent the unmanned system node's own motion state, computational state, communication state, and energy state, thereby providing a unified data input for subsequent capability measurements of hardware variability, software adaptability, and energy affordability.

[0256] In this embodiment, multidimensional state data with different sampling frequencies and acquisition periods are first synchronized according to a unified time reference. Then, data with spatial attributes are converted to a unified spatial reference system, and data without spatial attributes are associated with the state according to the data after spatial alignment at the acquisition time. This allows environmental states, ontological states, computational states, communication states, and energy states from different sources to form corresponding spatiotemporal state records. Furthermore, noise suppression, semantic segmentation, point cloud voxelization, feature extraction, and normalization processes are used to eliminate data noise, differences in representation formats, and numerical scales. The processed features are then fused into embodied state data. Thus, the originally asynchronous, heterogeneous, and dispersed multidimensional state data can be converted into a unified node state representation, providing a reliable data foundation for accurately measuring the current actual hardware capabilities, software operation capabilities, and energy security capabilities of unmanned system nodes.

[0257] This application also provides an embodied cognition assessment and collaborative reconstruction system for unmanned systems. The system includes a data acquisition module, a state data processing module, an ontology capability measurement module, a capability boundary calculation module, a task cognition deduction module, a cognition result fusion module, a distributed consistency solution module, and a collaborative reconstruction module.

[0258] The data acquisition module is used to implement the multi-dimensional state data acquisition step in the aforementioned embodiments; the state data processing module is used to implement the spatiotemporal registration and feature standardization processing steps; the ontology capability measurement module is used to implement the capability measurement steps for hardware variability, software adaptability, and energy affordability; the capability boundary calculation module is used to implement the dynamic calculation step for the ontology capability boundary; the task cognition inference module is used to implement the mask reconstruction inference of the task cognition dimension and the task cognition result generation steps; the cognition result fusion module is used to implement the task cognition result fusion and candidate collaborative information generation steps; the distributed consistency solution module is used to implement the distributed consistency solution step for candidate collaborative information; and the collaborative reconstruction module is used to implement the collaborative reconstruction instruction generation step.

[0259] The specific implementation methods and technical effects of the above modules can be found in the corresponding content of the aforementioned method embodiments, and will not be repeated here.

[0260] In one possible implementation, the system is distributed across nodes of a heterogeneous unmanned system cluster. Each node is equipped with a processor, memory, sensor interface, and communication interface. Each node can perform embodied state construction, ontological capability measurement, and task cognitive inference based on its acquired multidimensional state data. It also exchanges candidate collaborative information and local decision variables, global consensus variables, or dual variables generated during the distributed consensus solution process via communication connections. Thus, each node can collectively form a global consensus result and execute corresponding collaborative reconstruction instructions without needing to centrally transmit original images, point clouds, or complete task feature datasets to a single central node.

[0261] In one specific application embodiment, the unmanned system consists of a high-computing-power, high-payload ground unmanned vehicle, a first UAV, and a second UAV. The first UAV serves as the initial primary reconnaissance node, the second UAV serves as a standby and communication relay node, and the ground unmanned vehicle possesses high computing power and mission payload capacity. Each unmanned system node runs an Ubuntu operating system and a real-time operating system, and communicates asynchronously through a robot operating system middleware.

[0262] This embodiment uses the Gazebo physics engine to construct a complex 3D simulation environment of urban blocks. The 3D simulation environment includes buildings, roads, dynamic obstacles, and different types of occlusion areas, and can simulate multipath electromagnetic attenuation, sudden smoke, and communication link interruption. Ground unmanned vehicles, a first UAV, and a second UAV jointly perform target reconnaissance and information transmission tasks in the 3D simulation environment.

[0263] After the mission begins, each unmanned system node acquires environmental perception data, body operation data, computing resource data, communication status data, and energy status data. Specifically, image semantic segmentation and point cloud voxelization can be used to extract traversable areas, obstacle boundaries, target semantics, and dynamic target attributes. Furthermore, the position, attitude, velocity, actuator feedback, processor task queue status, battery temperature, and remaining charge status of each unmanned system node are spatiotemporally registered and feature-normalized to obtain the embodied state data corresponding to each unmanned system node.

[0264] During mission execution, the first UAV collided with a building, causing a decrease in the operational health of one rotor. Based on the rotor execution commands and corresponding feedback from the first UAV, the actuator health matrix was determined, and the damaged Jacobian matrix was reconstructed based on this matrix. The usable motion capability range of the first UAV was then redefined based on the damaged Jacobian matrix and the dynamic constraints of the first UAV, resulting in the reduced hardware variability. The reduction in usable motion capability range manifests as a decrease in the maximum permissible tilt angle, a decrease in the maximum level flight speed, and an increase in the dynamic turning radius.

[0265] Because the flight control program requires increased compensation calculations to maintain the attitude stability of the first UAV, the high-frequency control thread usage of the first UAV increases, the task queue congestion worsens, and the computing resources available for target identification, path planning, and communication processing tasks decrease, thus reducing the software adaptability of the first UAV. Simultaneously, the increased actuator compensation output and computational load lead to higher real-time power consumption, significantly reducing the usable range predicted based on the current battery temperature and remaining state of charge, thereby decreasing the energy affordability of the first UAV.

[0266] After performing nonlinear coupled calculations on the hardware variability, software adaptability, and energy affordability of the first UAV, the contracted dynamic dynamic envelope, computing power redundancy boundary, and energy decay boundary are obtained. The dynamic dynamic envelope, computing power redundancy boundary, and energy decay boundary are then used as task cognitive constraints for the first UAV to perform subsequent task cognitive inference.

[0267] Subsequently, the mask autoencoders corresponding to each task's cognitive dimension perform mask reconstruction deduction in parallel. Due to the sudden appearance of smoke in the 3D simulation environment, the mask autoencoders corresponding to the task's basic capabilities, after masking some key areas in the target image, are unable to accurately reconstruct the target contour and target semantics based on the remaining image information, thus increasing the reconstruction error corresponding to the task's basic capabilities.

[0268] The masked autoencoder corresponding to spatial reachability masks some future spatiotemporal nodes in the original candidate motion trajectory of the first UAV, and reconstructs the candidate motion trajectory under the constraints of the contracted dynamic envelope, energy decay boundary, and available time. Since the motion capability and energy capability of the first UAV have been significantly reduced, the feasible solution space corresponding to the original candidate motion trajectory is empty or the feasible range is significantly reduced, which increases the reconstruction error corresponding to spatial reachability.

[0269] The masking autoencoder corresponding to the time constraint extrapolates based on the first UAV's current mobility, remaining distance, and mission time window, determining that it would be difficult for the first UAV to reach the target area within the specified time if it continued to independently perform the main reconnaissance mission. The masking autoencoder corresponding to the information and communication connectivity extrapolates based on the first UAV's candidate movement path, building obstructions, and current communication transmission power, determining that if the first UAV continues to penetrate deeper into the target area, the quality of the communication link between it and other unmanned system nodes may significantly degrade.

[0270] The masked autoencoder corresponding to action coordination further simulates the joint actions of the ground unmanned vehicle, the first UAV, and the second UAV. It determines that the joint actions of the ground unmanned vehicle moving to a favorable communication position and undertaking communication relay and main reconnaissance tasks, the second UAV crossing the obstacle area to perform forward positioning, and the first UAV reducing its workload and performing near-field hovering demonstrate high action coordination. The masked autoencoder corresponding to organizational restructuring and fault tolerance reconstructs a backup coordination relationship based on the remaining node capabilities and inter-node relationships, whereby the ground unmanned vehicle and the second UAV perform role replacement and functional compensation for the first UAV.

[0271] Based on the task feature data and reconstruction features of each task cognitive dimension, the task feasibility confidence level of each task cognitive dimension is obtained. Since the task feasibility confidence levels corresponding to basic task capabilities, spatial accessibility, and information communication connectivity are significantly reduced, the attention weight of each task cognitive dimension during the task cognitive result fusion process is adjusted from focusing on spatial accessibility and time constraints during the normal cruise phase to focusing on information communication connectivity, organizational restructuring and fault tolerance, and action coordination. This forms candidate coordination information that represents the needs of role switching and topology adjustment.

[0272] The ground-based unmanned vehicle, the first UAV, and the second UAV each form local decision variables based on their own candidate collaborative information, and use the alternating direction multiplier method for distributed consensus solution. During the solution process, each node of the unmanned system exchanges local decision variables, global consensus variables, or dual variables, but does not exchange original video, point cloud, or complete task feature data.

[0273] Once the original residual and the dual residual satisfy the corresponding preset convergence conditions, a global consensus result is obtained. The global consensus result shows that: the first UAV is adjusted from the main reconnaissance node to a near-field hovering and communication relay node; the ground unmanned vehicle moves to a favorable communication position and assumes the functions of main reconnaissance reception and communication relay; the second UAV crosses the obstacle area to perform a forward replenishment mission.

[0274] Based on the global consensus results, collaborative reconfiguration instructions are generated and sent to the ground-based unmanned vehicle, the first UAV, and the second UAV. Each unmanned system node adjusts its role, function, and communication connection according to the collaborative reconfiguration instructions. After the adjustment is completed, each unmanned system node continues to acquire multi-dimensional state data and re-executes embodied state construction, ontology capability measurement, task cognition inference, task cognition result fusion, and distributed consistency solution, thereby forming a closed-loop processing process from state perception, capability assessment, task judgment to collaborative reconfiguration.

[0275] During the simulation verification process, a communication interruption rate of 40% was set, and highly dynamic unstructured terrain, complex building occlusion, dynamic obstacles, and strong communication interference were introduced into the 3D simulation environment. A method based solely on preset rules and geometric path planning for task allocation and motion control was used as a comparison.

[0276] Simulation results show that the task success rate of the comparison method is 23.5%, and the collision and crash rate is 18.7%. By using the method provided in this application embodiment, nonlinear coupling calculations are performed on hardware variability, software adaptability and energy affordability. After combining seven-dimensional task cognition deduction, task cognition result fusion and distributed consistency solution, the task success rate is increased to 88.2% and the hardware loss rate is reduced to 1.2%.

[0277] The above results demonstrate that by using dynamically changing ontological capability boundaries as task cognition constraints, the situation where task planning results exceed the actual execution capabilities of unmanned system nodes can be reduced. By performing masked reconstruction and deduction on multiple task cognition dimensions, risks such as spatial unreachability, task timeout, communication interruption, and node functional failure can be identified in a timely manner. By integrating the results of various task cognitions and performing distributed consistency solutions, role replacement, functional compensation, and topology adjustment results can be formed in a timely manner under conditions of communication constraints and reduced capabilities of some nodes, thereby improving the task continuity and collaborative robustness of unmanned systems in complex dynamic environments.

[0278] This application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it causes the processor to execute the unmanned system embodied cognition assessment and collaborative reconstruction method of any of the foregoing embodiments.

[0279] Computer-readable storage media may include read-only memory, random access memory, solid-state memory, magnetic disks, optical disks, or other media capable of storing computer programs. A computer program may be executed by a single processor or by multiple processors distributed across multiple unmanned system nodes in a coordinated manner.

[0280] This application also provides an electronic device, which includes a memory and at least one processor. The memory stores computer-executable instructions, and the at least one processor executes the computer-executable instructions to implement the unmanned system embodied cognitive assessment and collaborative reconstruction method of any of the foregoing embodiments. The electronic device can be located at unmanned system nodes, or it can be an edge computing device, collaborative control device, or server that is communicatively connected to each unmanned system node.

[0281] The above are merely specific embodiments of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to these embodiments, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for embodied cognitive assessment and collaborative reconstruction of unmanned systems, characterized in that, The method includes: Acquire multi-dimensional state data of each unmanned system node in the unmanned system. The multi-dimensional state data includes environmental perception data, system operation data, computing resource data, communication status data, and energy status data. For any unmanned system node, spatiotemporal registration and feature standardization are performed on the multidimensional state data of the unmanned system node to obtain the specific state data of the unmanned system node. Based on the embodied state data, the hardware state, software operating state, and energy state of the unmanned system node are measured to obtain hardware variability, software adaptability, and energy affordability. The nonlinear coupling relationship between hardware variability, software adaptability, and energy affordability is dynamically solved to obtain the ontological capability boundary; wherein, the ontological capability boundary includes a dynamic dynamic envelope surface, a computing power redundancy boundary, and an energy attenuation boundary. Using the aforementioned ontological capability boundary as a task cognition constraint, a mask autoencoder is used to perform mask reconstruction deduction on multiple task cognition dimensions of the task to be executed, thereby obtaining the mask reconstruction deduction results for each task cognition dimension. Based on the mask reconstruction results of each task cognition dimension, the task cognition results of each task cognition dimension are obtained; The cognitive results of each task are fused to obtain candidate collaborative information of the unmanned system nodes; Distributed consensus solution is performed on the candidate collaboration information generated by each of the unmanned system nodes to obtain a global consensus result; When the global consensus result meets the preset reconstruction triggering conditions, a collaborative reconstruction instruction is generated based on the global consensus result. The collaborative reconstruction instruction is used to adjust the role, function, or topology of each of the unmanned system nodes.

2. The method according to claim 1, characterized in that, The process involves using the ontological capability boundary as a task cognition constraint, and employing a mask autoencoder to perform mask reconstruction deduction on multiple task cognition dimensions of the task to be executed, to obtain the mask reconstruction deduction results for each task cognition dimension, including: The multiple task cognition dimensions are defined, including basic task capabilities, spatial accessibility, time constraints, access security, information and communication connectivity, action coordination, and organizational restructuring and fault tolerance. Obtain task feature data corresponding to each of the aforementioned task cognitive dimensions; Under the constraints of the ontology capability boundary, some features in each task feature data are masked, and the masked features are reconstructed using a mask autoencoder corresponding to each task cognitive dimension to obtain the reconstructed features of each task cognitive dimension, and the latent semantic features of each task cognitive dimension are extracted. Based on the reconstruction features and latent semantic features of each task cognitive dimension, a mask reconstruction deduction result is formed for each task cognitive dimension.

3. The method according to claim 2, characterized in that, When, under the constraints of the ontology capability boundary, a portion of the features in each of the task feature data is masked, and the masked features are reconstructed using a mask autoencoder corresponding to the cognitive dimension of each task: For the spatial reachability, based on the dynamic dynamic envelope, energy decay boundary and available time constraint of the task to be executed, some spatiotemporal nodes in the candidate motion trajectory are masked, and the candidate motion trajectory that satisfies the dynamic dynamic envelope, energy decay boundary and available time constraint is reconstructed using the mask autoencoder corresponding to the spatial reachability, so as to obtain the reconstruction feature of the spatial reachability. Regarding the organizational restructuring and fault tolerance, the nodes and their connections in the network topology formed by the unmanned system nodes are masked alternately. Then, using the mask autoencoder corresponding to the organizational restructuring and fault tolerance, the backup network topology or functional compensation path is reconstructed based on the unmasked node capability information and the relationship information between nodes, thus obtaining the reconstruction features of the organizational restructuring and fault tolerance.

4. The method according to claim 2, characterized in that, The mask reconstruction deduction results based on each of the task cognition dimensions yield the task cognition results for each task cognition dimension, including: Based on the task feature data and reconstruction features corresponding to each of the task cognition dimensions, the reconstruction error of each of the task cognition dimensions is determined; Based on a preset noise tolerance baseline and adjustment parameters corresponding to the urgency of the task to be performed, each of the reconstruction errors is mapped to the corresponding task feasibility confidence level; Based on the feasibility confidence of each task and the corresponding latent semantic features, the task cognition results of each task cognition dimension are obtained.

5. The method according to claim 4, characterized in that, The process of fusing the cognitive results of each task to obtain candidate collaborative information for the unmanned system nodes includes: Based on the feasibility confidence of each task, the corresponding latent semantic features are weighted to obtain the task perception features of each task cognitive dimension. The task-aware features are mapped to a shared feature space, and context features are determined based on the task intent of the task to be executed and the running state of the unmanned system nodes. Attention fusion is performed on each of the task-aware features in the shared feature space using the context features to obtain a fused feature with task priority attributes; Candidate collaborative information for the unmanned system nodes is generated based on the fusion features.

6. The method according to claim 5, characterized in that, The process of performing a distributed consensus solution on the candidate collaborative information generated by each of the unmanned system nodes to obtain a global consensus result includes: Based on the candidate collaborative information generated by each of the unmanned system nodes, the local decision variables of each of the unmanned system nodes are determined, and a local objective function corresponding to each of the local decision variables is constructed. Based on the consistency constraints among the local decision variables and the local objective functions, a distributed consensus optimization problem is constructed. The alternating direction multiplier method is used to iteratively update each of the local decision variables, global consensus variables and dual variables, and the original residuals are determined based on the updated local decision variables and global consensus variables, and the dual residuals are determined based on the global consensus variables of adjacent iterations. The iteration stops when the original residual and the dual residual satisfy the corresponding preset convergence conditions, and the global consensus result is obtained based on the global consensus variables at the time of stopping the iteration.

7. The method according to claim 1, characterized in that, The dynamic solution of the nonlinear coupling relationship between hardware variability, software adaptability, and energy affordability to obtain the ontological capability boundary includes: Based on the aforementioned hardware variability, software adaptability, and energy affordability, an ontology capability state tensor is constructed that includes the aforementioned hardware variability, software adaptability, and energy affordability. Based on the ontological capability state tensor, an ontological capability nonlinear evolution model is established. The ontological capability nonlinear evolution model is used to characterize the cross-domain coupling relationship between hardware variability, software adaptability and energy affordability. Based on the aforementioned nonlinear evolution model of the ontological capability, the coupled change trends of hardware variability, software adaptability, and energy affordability are predicted to obtain the ontological capability evolution trajectory. A stability analysis is performed on the evolution trajectory of the ontological capability to determine the capability dissipation safety boundary corresponding to the evolution trajectory of the ontological capability. Based on the evolution trajectory of the ontological capability and the capability dissipation safety boundary, the dynamic dynamic envelope, the computing power redundancy boundary, and the energy decay boundary are determined to obtain the ontological capability boundary.

8. The method according to claim 1, characterized in that, Based on the embodied state data, the hardware state, software operating state, and energy state of the unmanned system nodes are measured to obtain hardware variability, software adaptability, and energy affordability, including: Based on the execution instructions and corresponding execution feedback of the unmanned system node, the operational health status of the actuator is determined, and based on the operational health status and the dynamic constraints of the unmanned system node, the range of available motion capabilities of the unmanned system node is determined, thus obtaining the hardware variability; Based on the task queue status, computing resource occupancy status, and resource requirements and timeliness requirements of the unmanned system node to be executed, the real-time execution capability and computing resource reserve of the unmanned system node for the task to be executed are determined, and the software adaptability is obtained based on the real-time execution capability and computing resource reserve. Based on the real-time power consumption, battery temperature, and remaining state of charge of the unmanned system node, the available range of the unmanned system node is determined, and based on the available range, the energy affordability is obtained.

9. The method according to claim 1, characterized in that, The spatiotemporal registration and feature standardization processing of the multidimensional state data of the unmanned system node yields the specific state data of the unmanned system node, including: Based on the acquisition time corresponding to each of the multidimensional state data, time synchronization is performed on each of the multidimensional state data to obtain time synchronization data. Based on the unified spatial reference frame of the unmanned system node, the data with spatial attributes in the time synchronization data are spatially aligned, and according to the corresponding acquisition time, the data without spatial attributes are state-associated with the spatially aligned data to obtain spatiotemporal registration data. The spatiotemporal registration data is subjected to noise suppression, feature extraction and normalization, and the processed features are fused to obtain the embodied state data of the unmanned system node.

10. A system for embodied cognitive assessment and collaborative reconstruction of unmanned systems, characterized in that, The system includes: The data acquisition module is used to acquire multi-dimensional status data of each unmanned system node in the unmanned system. The multi-dimensional status data includes environmental perception data, body operation data, computing resource data, communication status data, and energy status data. The state data processing module is used to perform spatiotemporal registration and feature standardization on the multidimensional state data of any unmanned system node to obtain the specific state data of the unmanned system node. The on-body capability measurement module is used to measure the hardware status, software operation status and energy status of the unmanned system node based on the on-body status data, and to obtain hardware variability, software adaptability and energy affordability. The capability boundary calculation module is used to dynamically calculate the nonlinear coupling relationship between hardware variability, software adaptability and energy affordability to obtain the ontological capability boundary; wherein, the ontological capability boundary includes a dynamic dynamic envelope surface, a computing power redundancy boundary and an energy attenuation boundary. The task cognition deduction module is used to use the ontology capability boundary as the task cognition constraint, and to use the mask autoencoder to perform mask reconstruction deduction on multiple task cognition dimensions of the task to be executed, to obtain the mask reconstruction deduction result of each task cognition dimension, and to obtain the task cognition result of each task cognition dimension based on the mask reconstruction deduction result of each task cognition dimension. The cognitive result fusion module is used to fuse the cognitive results of each task to obtain candidate collaborative information of the unmanned system nodes; The distributed consensus solving module is used to perform distributed consensus solving on the candidate collaborative information generated by each of the unmanned system nodes to obtain a global consensus result. The collaborative reconstruction module is used to generate a collaborative reconstruction instruction based on the global consensus result when the global consensus result meets the preset reconstruction triggering condition. The collaborative reconstruction instruction is used to adjust the role, function or topology relationship of each of the unmanned system nodes.

Images