Unmanned cluster distributed cooperative control system

Abstract

The invention belongs to the technical field of intelligent transportation, and discloses an unmanned cluster distributed cooperative control system, which comprises a seed unmanned aerial vehicle, a sub-level unmanned aerial vehicle node, a satellite-network hybrid communication module, a dynamic synchronous control module, a security check module and a plurality of handheld cooperative control terminals. A compressed large language model inference engine is arranged in the handheld cooperative control terminal, the multi-mode instruction is analyzed, and task parameters are generated; recursively releasing sub-level nodes by the seed unmanned aerial vehicle according to energy, airspace and communication constraints, and constructing a layered bee colony; high-speed communication in clusters and cross-cluster convergence are realized through prefix address coding and layered spectrum slicing; dynamic synchronous control adopts Isin-Hamiltonian quantity minimization to quickly achieve consistency; and the security check monitors multi-source risks in real time and degrades or fuses in a grading manner. The system provided by the invention improves the capabilities of rapid deployment, robust communication and security coordination of the unmanned aerial vehicle cluster, and is suitable for scenes of security, rescue, monitoring and the like.

Description

Technical Field

[0001] This invention belongs to the fields of intelligent unmanned systems, swarm collaborative control and distributed computing technology, specifically relating to an unmanned swarm distributed collaborative control system, which falls under the cross-application category of large-scale UAV swarm scheduling, dynamic routing and autonomous collaborative control. Background Technology

[0002] In recent years, with the rapid development of computer technology, information processing technology, new materials, sensors, and artificial intelligence, the integration level of unmanned aerial vehicles (UAVs) has been continuously improving, and their applications in military and civilian fields have become increasingly widespread and in-depth. Especially in complex, dynamic, and highly adversarial combat environments or application scenarios, unprecedented demands have been placed on the autonomy, intelligence, and collaborative capabilities of unmanned systems, particularly unmanned swarms (UAV swarms). However, existing UAV swarm technologies still face severe challenges and significant limitations in the following key aspects:

[0003] Insufficient efficiency and capability in large-scale deployment: Deployment bottlenecks exist. Traditional large-scale cluster deployment methods (such as "one-time launch") are strictly limited by factors such as aircraft space, airspace blockade, takeoff and landing windows, and synchronous command capabilities, making it difficult to quickly and covertly deploy a large number of platforms (such as hundreds to thousands of aircraft) to the mission airspace in a short period of time. This severely restricts the cluster response speed and operational timeliness.

[0004] Poor environmental adaptability: In complex terrains and electromagnetic environments such as urban canyons, mountains, and areas with strong electromagnetic interference, the communication, navigation, and control during deployment face the risks of interruption, interference, and obstruction. Existing technologies cannot guarantee the stable and reliable deployment of large-scale clusters.

[0005] Heterogeneous cluster deployment difficulties: How to effectively manage and deploy heterogeneous unmanned platforms with different functions, performance, communication standards and missions, and form a unified and coordinated combat formation, is a problem that existing technologies have not yet fully solved.

[0006] High dependence on carrier platforms: Cluster deployment often relies on the centralized deployment of large carrier platforms (such as transport aircraft and ships), which limits the flexibility and concealment of deployment.

[0007] The challenge of coordinated control in large-scale clusters:

[0008] Limitations of centralized control: Traditional centralized command or global synchronous control modes, when dealing with large-scale clusters (such as 2000 nodes), experience a non-linear increase in the computational load for communication topology generation, data fusion, spectrum modeling, and task interaction, making high-fidelity simulations and real-time control difficult. Furthermore, in battlefield environments with limited bandwidth and easily distorted links, the reliability and sustainability of centralized control are difficult to guarantee.

[0009] Heterogeneity and Asynchronicity: Large-scale unmanned swarms contain heterogeneous platforms with varying performance, functions, communication capabilities, and task requirements. These differences lead to asynchronous states and complex control laws. Existing technologies struggle to achieve a unified time base and precise collaborative control under heterogeneous and asynchronous conditions, ensuring the real-time nature and consistency of information sharing and collaborative decision-making.

[0010] Robustness and Adaptability in Dynamic Environments: Clusters need to cope with dynamic changes such as random node addition / exit, communication link interruption, electronic warfare, and node failure. Maintaining the overall stability of the cluster, mission continuity, and achieving self-organization, self-healing, and situational adaptability are significant challenges currently facing technology.

[0011] The complexity of human-computer interaction and command decision-making: Operator overload. In large-scale cluster operations, a single operator position needs to process massive amounts of node status information, perform real-time perception, filtering, decision-making and intervention, resulting in huge cognitive load on operators, which can easily lead to command delays and situational misjudgments.

[0012] Inefficiency in expressing and transmitting intent: Existing human-computer interaction modes are mostly command-based or task-planning-based, requiring operators to remotely control via precise instructions. There is a lack of efficient, intuitive, and immersive human-computer interfaces to express high-level operational intent. Furthermore, limited edge computing and intent recognition capabilities negatively impact command efficiency.

[0013] Lack of flexible dynamic adjustment capabilities: Operators find it difficult to adjust cluster task planning, formation, communication strategies in real time and intuitively, as well as to dynamically intervene in the deployment process according to changes in the battlefield.

[0014] Given the numerous technical problems and limitations mentioned above, existing unmanned swarm technologies are insufficient to fully meet the high demands placed on unmanned systems by future intelligent, systematic, and high-intensity warfare. Therefore, it is necessary to develop a new type of large-scale, heterogeneous unmanned swarm distributed collaborative control system to overcome the shortcomings of existing technologies and achieve intelligent, autonomous, large-scale, and highly efficient collaborative warfare within the swarm. Summary of the Invention

[0015] In view of this, the purpose of this invention is to provide a distributed collaborative control system for unmanned swarms, which aims to solve the problems of "slow response, rigid routing, slow collaborative convergence and decentralized security verification" in existing unmanned aerial vehicle swarm control.

[0016] This invention provides an unmanned cluster distributed collaborative control system, including a seed UAV, sub-level UAV nodes, a star-network hybrid communication module, a dynamic synchronization control module, a safety verification module, and multiple handheld collaborative control terminals;

[0017] Handheld co-control terminal: Built-in compressed large language model inference engine, used to parse the operator's intention into parameterized task instructions;

[0018] Seed drone: Equipped with a multi-mode delivery mechanism, used to recursively release sub-stage drones after satisfying space, energy and communication constraints;

[0019] Sub-level drone clusters: Based on the task instructions issued by the handheld co-control terminal, they self-organize into several sub-clusters under the control of the hierarchical fission deployment algorithm;

[0020] Star-Network Hybrid Communication Module: Prefix address encoding and hierarchical spectrum slicing are used for nodes at different levels to achieve high-speed communication within local clusters and cross-cluster data aggregation;

[0021] Dynamic synchronization control module: Based on the Hamiltonian minimization algorithm, it adaptively adjusts the node coupling weights to minimize the overall network collaborative potential.

[0022] Safety verification module: Real-time monitoring of flight airspace, power threshold, link interference and energy margin to trigger circuit breaker or degradation strategies in a timely manner;

[0023] Each handheld co-control terminal is responsible for a local cluster domain consisting of 100-300 drones. The handheld co-control terminal keeps synchronized with the drones through a low-latency link. The handheld co-control terminals exchange "terminal-to-terminal collaborative frames" through UWB or 5G-NR direct links to maintain a consistent consensus on task boundaries, spectrum occupancy and circuit breaker status.

[0024] Furthermore, the handheld co-control terminal includes:

[0025] Touchscreen display: Used to display geographic information of the mission area, enemy and friendly situation, and the deployment and flight path of the UAV swarm. Operators can input parameters, plan routes and issue commands via touch screen.

[0026] Voice acquisition unit: used to collect the operator's spoken instructions;

[0027] Embedded GPU: Runs a large language model inference engine to complete intent parsing, risk assessment, and task inference in sub-second time.

[0028] Joint Positioning Module: Used to integrate data collected by various UAV sensors to achieve precise positioning and navigation.

[0029] Furthermore, the hierarchical fission deployment algorithm is set in the embedded GPU, and the settings of the hierarchical fission deployment algorithm include:

[0030] I. Using the task parameter vector θ t This serves as the quantitative basis for the hierarchical fission deployment algorithm;

[0031] θ t =g ψ (f φ (m t ))

[0032] In the formula, g ψ Indicates instruction decoder; f φ Indicates encoder; m t Represents high-dimensional multimodal input; g ψ (f φ (m t )) Input high-dimensional multimodal data into m t Compressed into an interpretable task density ρ tar Energy threshold E tar and link threshold Q tar ;

[0033] II. Calculate the replication degree;

[0034] Seed drone receives θ t Then, based on the local task density ρ t Remaining energy E l With link quality Q l Calculate the replication degree k l Recursive fission follows:

[0035] N l+1 =k l N l ,

[0036] In the formula, N l This indicates the number of drones in phase l;

[0037] To prevent global resource overdraft, an energy-bandwidth budget is set up:

[0038]

[0039] In the formula, N unit B represents the energy of a unit of drone; unit E represents the bandwidth per unit of unmanned aerial vehicle (UAV) bandwidth. tot Indicates total energy budget; B tot L represents the total bandwidth budget; L represents the total number of stages in the UAV recursive fission.

[0040] When the remaining energy of the parent node is lower than the preset threshold or the communication tension exceeds the threshold, further fission will stop and neighbor cluster resource scheduling will be initiated.

[0041] Furthermore, the star-network hybrid communication module employs TDMA or frequency hopping communication within the UAV cluster, uses lateral link multiplexing based on geographical proximity between UAV clusters, and aggregates routes through variable-length prefix codes to suppress broadcast storms.

[0042] Furthermore, the dynamic synchronization control module abstracts the UAV node as a spin variable of the Ising model, and dynamically adjusts the coupling coefficient according to the mission relevance, link quality and energy state to achieve rapid phase transition synchronization under event triggering.

[0043] Furthermore, the safety verification module performs airspace restricted area detection, power compatibility check and electromagnetic interference co-test before issuing the task. If the detection fails, it generates a restricted state execution scheme or switches to simulation mode.

[0044] Furthermore, the sub-level UAV has a dynamic role-flipping function, which can switch between multiple mission roles such as perception, communication relay, electronic jamming and strike according to the Ising-Hamilton role-flipping mechanism.

[0045] Beneficial effects:

[0046] 1. This invention proposes an unmanned cluster distributed collaborative control system. The system integrates lightweight large model, hierarchical fission deployment, star-network hybrid communication, adaptive synchronization control and multi-source risk circuit breaking. It can complete task orchestration and real-time closed loop at both the terminal and air levels, which reduces command link latency and ensures the collaborative consistency and security robustness of a large number of nodes.

[0047] 2. This invention proposes a distributed collaborative control system for unmanned swarms, comprising seed UAVs, sub-level UAV nodes, a satellite-network hybrid communication module, a dynamic synchronization control module, a safety verification module, and multiple handheld collaborative control terminals. The handheld collaborative control terminals have a built-in compressed large language model inference engine to parse multimodal commands and generate task parameters; the seed UAVs recursively release sub-level nodes according to energy, airspace, and communication constraints to construct a hierarchical swarm; prefix address encoding and hierarchical spectrum slicing enable high-speed intra-cluster communication and cross-cluster convergence; dynamic synchronization control uses Ising-Hamiltonian minimization to quickly achieve consensus; and the safety verification module monitors multi-source risks in real time and performs graded degradation or circuit breaking. This invention enables rapid deployment, robust communication, and unified security management of thousands of UAVs without the need for large command and control stations, making it suitable for high-dynamic scenarios such as remote security, emergency rescue, and mobile operations.

[0048] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the user interaction and control interface of the unmanned cluster distributed collaborative control system of the present invention;

[0050] Figure 2 A flowchart of the closed-loop process for human-computer collaborative feedback and strategy correction based on a large language model;

[0051] Figure 3 A schematic diagram of a combat system in which different functional groups interact based on collaborative information;

[0052] Figure 4 This is a detailed diagram showing the components and data flow of a digital prototype network simulation system. Detailed Implementation

[0053] To make the technical solutions, advantages, and objectives of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the protection scope of this application.

[0054] This invention provides an unmanned cluster distributed collaborative control system, including a seed UAV, sub-level UAV nodes, a star-network hybrid communication module, a dynamic synchronization control module, a safety verification module, and multiple handheld collaborative control terminals.

[0055] Handheld co-control terminal: Built-in compressed large language model inference engine, used to parse the operator's intention into parameterized task instructions;

[0056] Each handheld co-control terminal is responsible for a local cluster of 100-300 drones. The handheld co-control terminal keeps synchronized with the drones through a low-latency link. The handheld co-control terminals exchange "terminal-to-terminal collaborative frames" through UWB or 5G-NR direct links to maintain a consistent consensus on task boundaries, spectrum occupancy and circuit breaker status.

[0057] Handheld control terminals include:

[0058] Touchscreen display: Used to display geographic information of the mission area, enemy and friendly situation, and the deployment and flight path of the UAV swarm. Operators can input parameters, plan routes and issue commands via touch screen.

[0059] Voice acquisition unit: used to collect the operator's spoken instructions;

[0060] Embedded GPU: Runs a large language model inference engine to complete intent parsing, risk assessment, and task inference in sub-second time.

[0061] Joint Positioning Module: Used to integrate data collected by various UAV sensors to achieve precise positioning and navigation.

[0062] Seed drone: Equipped with a multi-mode delivery mechanism, used to recursively release sub-stage drones after satisfying space, energy and communication constraints;

[0063] The handheld collaborative control terminal supports multi-operator collaborative management, employs lightweight situation summary exchange and arbitration protocols to resolve resource conflicts, and distinguishes cluster domains managed by different operators through permission tags. When a communication link is detected to be disconnected, the handheld collaborative control terminal can enter predictive hold mode, extrapolating the node status for several future time steps based on historical trajectories and autonomous strategies, and performing differential alignment after the link is restored. It monitors the computational load between the handheld collaborative control terminal and cluster head nodes, dynamically offloading computationally intensive tasks to idle cluster heads or command and control centers to reduce average queuing latency and balance edge computing load.

[0064] Sub-level drone clusters: Based on the task instructions issued by the handheld co-control terminal, they self-organize into several sub-clusters under the control of the hierarchical fission deployment algorithm.

[0065] The sub-stage UAV possesses a dynamic role-switching capability, enabling it to switch between multiple mission roles—sensing, communication relay, electronic jamming, and strike—using an Ising-Hamiltonian role-switching mechanism. During descent, the sub-stage UAV employs a spiral diffusion trajectory to avoid vertical airflow crosstalk and transmits airflow disturbance parameters back to its parent node before landing.

[0066] The hierarchical fission deployment algorithm is set up in the embedded GPU. The settings of the hierarchical fission deployment algorithm include:

[0067] I. Using the task parameter vector θ t This serves as the quantitative basis for the hierarchical fission deployment algorithm;

[0068] θ t =g ψ (f φ (m t )) (1)

[0069] In the formula, g ψ Indicates instruction decoder; f φ This refers to the encoder, where the operator inputs voice, touch, or gesture commands via a handheld controller, which are then processed by the large-scale encoder. φ Generate intent embedding v t m t Represents high-dimensional multimodal input; g ψ (f φ (m t )) Input high-dimensional multimodal data into m tCompressed into an interpretable task density ρ tar Energy threshold E tar and link threshold Q tar ;

[0070] II. Calculate the replication degree;

[0071] Seed drone receives θ t Then, based on the local task density ρ t Remaining energy E l With link quality Q l Calculate the replication degree k l Recursive fission follows:

[0072]

[0073] In the formula, N l This indicates the number of drones in phase l;

[0074] The above formula reflects the product effect of the three factors "demand-resources-communication": if any one factor is insufficient, k l This means it's compressed to avoid over-deployment. To prevent global resource overdraft, an energy-bandwidth budget is also set:

[0075]

[0076] In the formula, N unit B represents the energy of a unit of drone; unit E represents the bandwidth per unit of unmanned aerial vehicle (UAV) bandwidth. tot Indicates total energy budget; B tot L represents the total bandwidth budget; L represents the total number of stages in the UAV recursive fission.

[0077] The above formula provides a hard upper limit for the fission process, allowing the hierarchical colony to expand within a tolerable energy and spectrum range. That is, when the remaining energy of the parent node falls below a preset threshold or the communication stress exceeds the threshold, further fission stops and neighbor cluster resource scheduling is initiated.

[0078] Star-Network Hybrid Communication Module: Prefix address coding and hierarchical spectrum slicing are adopted for nodes at different levels to achieve high-speed communication within local clusters and cross-cluster data aggregation; Among them, the Star-Network Hybrid Communication Module uses TDMA or frequency hopping communication within UAV clusters, adopts lateral link multiplexing based on geographical proximity between UAV clusters, and uses variable-length prefix code aggregation routing to suppress broadcast storms.

[0079] Dynamic synchronization control module: Based on the Hamiltonian minimization algorithm, it adaptively adjusts the node coupling weights to minimize the overall network collaborative potential energy. The module abstracts UAV nodes as spin variables of the Ising model and dynamically adjusts the coupling coefficients according to task relevance, link quality, and energy state to achieve rapid phase-change synchronization triggered by events.

[0080] The safety verification module performs real-time monitoring of flight airspace, power thresholds, link interference, and energy margins to promptly trigger circuit breaker or degradation strategies. Before issuing a mission, the module performs airspace exclusion zone detection, power compatibility checks, and electromagnetic interference (EMI) co-testing. If the detection fails, it generates a restricted-state execution plan or switches to simulation mode.

[0081] The principle behind rapid deployment of a Hierarchical Cascading Self-Replication (HCSR) cluster is as follows:

[0082] HCSR represents the cluster as a rooted hierarchical graph with lateral reconnection. The initial set enters the target airspace by manual launch, vehicle-mounted catapult, or airdrop; any upper-level node can derive k child node sets after meeting the space, energy, communication, and mission triggering conditions to form a new layer. Each child node completes the following upon activation: (1) identity registration (Parent-ID inheritance + Local Child-Index); (2) local coordinate and altitude allocation; (3) communication channel / time slot or frequency hopping subdomain application; (4) initial role (sensing / communication / interference) binding for mission functions; (5) minimum autonomous control power-on. Through this progressive recursive replication mechanism, the total number of nodes in the cluster will grow exponentially. Its final size is determined by the number of initial "seed" nodes, the average replication ratio, and the depth of the deployment hierarchy. This mode enables us to achieve rapid and exponential expansion of the cluster size without relying on centralized launch facilities. HCSR distinguishes between macro-deployment beats and micro-local initialization windows. After the upper-level seed completes coarse-grained distribution, it can determine the perimeter environment and release its children within the local clock window without waiting for network-wide synchronization. Each child inherits the deployment timestamp of its parent while using random jitter offset to avoid synchronization peaks caused by cross-layer collisions. This forms a "waterfall-like" parallel deployment: multi-layer releases can partially overlap, and the actual total deployment time can be significantly shorter than serial accumulation. To ensure safe intervals, communication visibility, and task coverage, each parent node performs a local 3D environment scan (airborne laser ranging, visual dense reconstruction, terrain database matching + electromagnetic mapping) before fission. The scan output is converted into a local deployable volume and divided into subdomains under constraints (Voronoi, sphere filling, or terrain-fitting block algorithms can be used). The parent node assigns an initial deployment position, height layer (to avoid spiral fall crosstalk), inter-machine safety radius, and spectrum subband suggestions to each subdomain; after approval, the child node enters the target subdomain and establishes a local mesh formation. This "environmental awareness - subdomain shaping" link enables HCSR to still achieve controllable fission in restricted airspace such as urban canyons and mountain valleys.

[0083] HCSR uses hierarchical path-based address encoding: Node-ID=<Layer,ParentPath,LocalIndex> It can be compressed into variable-length prefix codes to support fast domain broadcast filtering; cluster domain routing can complete local aggregation forwarding based on prefix matching, reducing cross-layer broadcast storms. Spectrum management adopts the principle of "layered initial allocation + adaptive multiplexing": the parent allocates several sub-bands or frequency hopping modes from the globally licensed frequency band to sub-clusters; sub-clusters are further subdivided or time-divided; adjacent sub-clusters dynamically exchange bandwidth based on interference measures; in case of disconnection, it can fall back to the parent's backup channel. This mechanism effectively compresses link interference during large-scale expansion.

[0084] Each newly activated child node first joins the parent control beacon to complete authentication and key derivation; then, neighbor detection (RSSI / ToF) is performed within the subdomain to establish a local connectivity graph; the optimal sub-cluster coordinator (LocalCoordinator) is selected to form a cluster-level star-network hybrid backbone; the parent and child cluster coordinators form an uplink aggregation tree; lateral links are added according to a geographical proximity strategy to improve redundancy. After the backbone is formed, intra-cluster control flow and data flow are forwarded along separate paths: control uplink is flattened, and data locality is prioritized, reducing the load on seed-level nodes from an architectural perspective.

[0085] To avoid disorderly expansion, HCSR sets multi-dimensional replication triggering criteria: insufficient spatial coverage (detection density < threshold), excessive communication link strain (hop count > threshold, bandwidth saturation), mission payload requirements (reconnaissance zone expansion, jamming front advancement), energy reallocation (high remaining energy in the parent and vacant child domains), and damage replenishment (cluster loss > ratio). Child classes can only be derived after any triggering condition is met and a security check is passed; in the event of multiple concurrent triggers, the replication sequence is determined by mission priority.

[0086] The star-network hybrid communication and adaptive synchronization method is as follows:

[0087] To balance high speed within a cluster and robustness across clusters, node addresses use a prefix + suffix encoding. i =p0p1…p l ||s i Link capacity is affected by signal-to-noise ratio γ ij The constraint, its Shannon limit, is written as:

[0088] Ri j =B ij log2(1+γ ij (4)

[0089] Equation (4) guides the spectral slice: if γ ij Decrease, dynamic reduction of B ij Greater bandwidth is reserved for critical links to achieve load balancing. Node collaboration employs the Ising-Hamiltonian model. Overall potential energy:

[0090]

[0091] Where σ i ∈{±1} represents the node role (e.g., perception / interference), J ij The system adaptively updates based on link quality and task relevance. The node collaborative potential calculation cycle is no more than 500ms, and the system achieves ≥90% node state consistency within two iterations.

[0092] Gradient descent iteration:

[0093] J ij ←Jij +ασ i σ j (6)

[0094] By causing H to decrease monotonically, node state consistency can be achieved in two iterations, realizing rapid adaptive operation of the "communication-control-role" trinity. The system normalizes and weights four types of risks—spatial, power, link, and energy—to form a comprehensive index.

[0095]

[0096] When R(t) > R max This triggers the circuit breaker (FUSE(level)). If the limit is slightly exceeded, a degradation measure is implemented: power is reduced and auxiliary tasks are suspended. If the limit is severely exceeded, the sub-drone hovers and the seed drone returns to base, ensuring a safe closed-loop system. This invention eliminates the need for large ground command and control stations, facilitating rapid deployment by individual soldiers and making it suitable for various scenarios such as security patrols, disaster relief, precision agriculture, and tactical reconnaissance.

[0097] In the application scenario of thousands of drones in this invention, the handheld co-controller is not a single terminal, but is deployed in groups according to the concept of "multi-terminal-multi-cluster": each operator is equipped with a portable co-controller, with a typical scale of 4-10 units; each terminal is responsible for a local cluster domain consisting of 100-300 drones, and maintains a "co-control-co-control" mesh interconnection with other terminals through a 5G-NR / UWB direct link. The terminal has a built-in compressed large language model inference engine, integrates 5G / Wi-Fi / UWB multi-mode communication, GNSS-RTK positioning and dual battery hot-swappable power supply, and can independently complete task input, policy distribution and safety circuit breaking without ground stations or backbone networks.

[0098] The process of implementing unmanned cluster distributed collaborative control using the system of this invention is as follows:

[0099] S1: Mission planning and human-machine collaborative interaction. The operator first uses a handheld coordinator to access the "Unmanned Aerial Vehicle (UAV) Swarm Intelligent System" interface, inputting macro-level parameters such as the operational area, node size, and communication radius via voice or touch. The terminal uses a local large language model to parse the spoken commands into structured task templates, while simultaneously projecting a real-time mirrored map and situational awareness onto a large screen at the fixed command and control station, ensuring synchronization between individual operator operations and rear-end decision-making. When the operator selects a target area or drags a flight path node on the handheld screen, all modifications are immediately transmitted back to the backend database via WebRTC stream, achieving end-to-end consistent visual collaboration.

[0100] Reference Figure 1 Operators define and monitor tasks through the user interface of the "Intelligent Drone Swarm System." This interface is mainly divided into three areas:

[0101] Control Panel (left side): Here, the operator enters macro-level parameters for the mission, such as the operational area, the number of nodes to be generated (number of drones), node speed, and communication range. The operator can also set generation parameters, such as generation time and formation status, and set mission constraints by checking options such as "Display Track" and "Communicate Flight".

[0102] Situation Display Area (Center): This area displays real-time geographic information of the mission area, enemy and friendly situation, and the deployment and flight paths of the UAV swarm in the form of a two-dimensional or three-dimensional map. Operators can intuitively monitor the entire mission execution process.

[0103] Handheld controller (right): Operators can use a handheld terminal to make fine-grained real-time interventions on individual or part of the drones, such as manually planning routes and issuing specific commands (such as "guard").

[0104] S2: Strategy generation and correction based on a large language model. Natural language data and trajectory sketches collected by the handheld terminal are directly fed into the LLM module to generate preliminary combat strategies. The strategies do not take effect immediately; instead, a rapid pre-simulation is triggered locally on the terminal: the co-controller invokes a lightweight digital sandbox to perform 3-5 second rolling simulations of key indicators such as whether the trajectory has crossed boundaries and whether the formation has broken links. If obvious defects are found, the terminal displays a pop-up prompt and allows the operator to correct them verbally. Subsequently, this "pre-screening" strategy, along with the operator's additional comments, is submitted to the high-fidelity critique-feedback closed loop. In this way, the handheld terminal integrates human-computer interaction, model self-review, and rapid simulation testing into the first quality control gate, greatly reducing the unnecessary load on the backend critique module. Figure 2 As shown, the "feedback signal" generated by the critique module is sent to the "policy correction module." This module, based on the quality of the feedback signal, uses various learning mechanisms (such as supervised learning, reinforcement learning, and post-processing) to fine-tune and optimize the LLM or its downstream policy model. The corrected "policy" is then output again, entering the next round of the "critique-feedback" loop, or, after reaching its optimal state, is formally issued to the drone swarm for execution.

[0105] S3: In the collaborative execution of multi-functional clusters, when the final strategy, optimized through a closed-loop process in the background, returns, the handheld co-controller once again plays the dual role of "broadcaster" and "scheduler." It sends hierarchical fission parameters to the seed drones while simultaneously monitoring the real-time heartbeat packets of each functional cluster. Once the reconnaissance cluster detects a high-value target, the terminal will push a vibration alert within 200 milliseconds and allow the operator to directly click on the target point on the handheld screen and add a "priority strike" tag. This instruction is broadcast to the strike cluster via a single hop from the terminal to the air, avoiding the additional latency caused by traditional central node forwarding, thus maintaining high real-time performance of distributed collaboration.

[0106] like Figure 3As shown by the dashed lines, the various factions do not communicate through a central server, but rather through point-to-point or multi-point-to-multi-point information exchange via "coordination information." For example, after the "reconnaissance" faction discovers a target, it directly sends "reconnaissance coordination information" (including target location, type, etc.) to the "decision-making" and "strike" factions. The "decision-making" faction issues attack commands based on this information, while the "strike" faction carries out precision strikes based on this information.

[0107] S4: Digital Prototype Network Simulation and Performance Evaluation. Throughout the mission, the handheld terminal continuously collects node status, link indicators, and environmental information, and transmits the data to the digital twin simulation federation via 5G / satellite backlink. The simulation system dynamically adjusts battlefield electromagnetic and terrain obstruction parameters by comparing them with real airborne logs, forming a high-fidelity digital image. After the simulation results are quantified by the performance evaluation liaison members, they are fed back to the handheld terminal in real time: if the evaluation score falls below the safety threshold, the terminal automatically triggers a first-level degradation (reducing transmission power and suspending high-definition backlink); if the score continues to deteriorate, the operator can press and hold the fuse button to perform a global shutdown or return to base. Thus, the closed loop of "handheld-air-simulation-evaluation" is fully completed.

[0108] Reference Figure 4 Each member can represent a real hardware device (such as a UAV flight control computer) or a purely software-simulated node. They input simulated or real state data (experimental point n) into the simulation network through the "Experimental Data Acquisition Subsystem" and the "Data Transmission Subsystem." The core of each member is the "Network-Level Simulation Subsystem," which contains detailed modeling of the entire battlefield environment, such as "telecommunications equipment models," "node models," "protocol models," and "service models." It is responsible for simulating communication processes between UAVs, the establishment and interruption of data links, etc. The results of the network-level simulation interact with the "Battlefield Environment Support Member." This member includes the "Virtual Battlefield Disability Classification Subsystem" and the "Complex Electromagnetic Environment Generation Subsystem," used to simulate real battlefield combat effects (such as battle damage and communication interference). All data and results from the simulation process are ultimately sent to the "Effectiveness Assessment Liaison Member," which uses the "Data Preprocessing Subsystem," "Data Analysis Subsystem," etc., to quantitatively assess the effectiveness of the entire combat mission (assessment conclusions and archived data). Ultimately, the quantitative evaluation results output by the "Performance Evaluation Liaison Member" will serve as the most critical simulation feedback signal for the "Critique Module" in S2, driving the strategy iteration and optimization of the entire intelligent system.

[0109] It is hereby declared that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An unmanned cluster distributed collaborative control system, characterized in that: It includes seed drones, sub-level drone nodes, star-network hybrid communication modules, dynamic synchronization control modules, security verification modules, and multiple handheld co-control terminals; Handheld co-control terminal: Built-in compressed large language model inference engine, used to parse the operator's intention into parameterized task instructions; Seed drone: Equipped with a multi-mode delivery mechanism, used to recursively release sub-stage drones after satisfying space, energy and communication constraints; Sub-level drone clusters: Based on the task instructions issued by the handheld co-control terminal, they self-organize into several sub-clusters under the control of the hierarchical fission deployment algorithm; Star-Network Hybrid Communication Module: Prefix address encoding and hierarchical spectrum slicing are used for nodes at different levels to achieve high-speed communication within local clusters and cross-cluster data aggregation; Dynamic synchronization control module: Based on the Hamiltonian minimization algorithm, it adaptively adjusts the node coupling weights to minimize the overall network collaborative potential. Safety verification module: Real-time monitoring of flight airspace, power threshold, link interference and energy margin to trigger circuit breaker or degradation strategies in a timely manner; Each handheld co-control terminal is responsible for a local cluster domain consisting of 100-300 drones. The handheld co-control terminal keeps synchronized with the drones through a low-latency link. The handheld co-control terminals exchange "terminal-to-terminal collaborative frames" through UWB or 5G-NR direct links to maintain a consistent consensus on task boundaries, spectrum occupancy and circuit breaker status.

2. The unmanned cluster distributed collaborative control system according to claim 1, characterized in that, The handheld co-control terminal includes: Touchscreen display: Used to display geographic information of the mission area, enemy and friendly situation, and the deployment and flight path of the UAV swarm. Operators can input parameters, plan routes and issue commands via touch screen. Voice acquisition unit: used to collect the operator's spoken instructions; Embedded GPU: Runs a large language model inference engine to complete intent parsing, risk assessment, and task inference in sub-second time. Joint Positioning Module: Used to integrate data collected by various UAV sensors to achieve precise positioning and navigation.

3. The unmanned cluster distributed collaborative control system according to claim 2, characterized in that, The hierarchical fission deployment algorithm is set in the embedded GPU, and the settings of the hierarchical fission deployment algorithm include: I. Using the task parameter vector θ t This serves as the quantitative basis for the hierarchical fission deployment algorithm; i t =g ψ (f φ (m t )) In the formula, g ψ Indicates instruction decoder; f φ Indicates encoder; m t Represents high-dimensional multimodal input; g ψ (f φ (m t )) Input high-dimensional multimodal data into m t Compressed into an interpretable task density ρ tar Energy threshold E tar and link threshold Q tar ; II. Calculate the replication degree; Seed drone receives θ t Then, based on the local task density ρ t Remaining energy E l With link quality Q l Calculate the replication degree k l Recursive fission follows: In the formula, N l This indicates the number of drones in phase l; To prevent global resource overdraft, an energy-bandwidth budget is set up: In the formula, N unit B represents the energy of a unit of drone; unit E represents the bandwidth per unit of unmanned aerial vehicle (UAV) bandwidth. tot Indicates total energy budget; B tot L represents the total bandwidth budget; L represents the total number of stages in the UAV recursive fission. When the remaining energy of the parent node is lower than the preset threshold or the communication tension exceeds the threshold, further fission will stop and neighbor cluster resource scheduling will be initiated.

4. The unmanned cluster distributed collaborative control system according to claim 3, characterized in that: The star-network hybrid communication module uses TDMA or frequency hopping communication within the UAV cluster, employs lateral link multiplexing based on geographical proximity between UAV clusters, and uses variable-length prefix code aggregation routing to suppress broadcast storms.

5. The unmanned cluster distributed collaborative control system according to claim 4, characterized in that: The dynamic synchronization control module abstracts the UAV node as a spin variable of the Ising model and dynamically adjusts the coupling coefficient according to the mission relevance, link quality and energy state to achieve rapid phase transition synchronization under event triggering.

6. The unmanned cluster distributed collaborative control system according to claim 5, characterized in that: Before issuing a task, the safety verification module performs airspace restricted area detection, power compatibility check and electromagnetic interference collaborative test. If the detection fails, it generates a restricted state execution scheme or switches to simulation mode.

7. The unmanned cluster distributed collaborative control system according to claim 1, characterized in that: The sub-level UAV has a dynamic role-flipping function, which can switch between multiple mission roles such as perception, communication relay, electronic jamming and strike according to the Ising-Hamilton role-flipping mechanism.

Images