A distributed autonomous agent cluster coordination system and method based on dynamic security roots

Abstract

This invention discloses a distributed autonomous intelligent agent cluster collaborative system and method based on dynamic security. Each node in the cluster shares a group DSF and independently generates a logical timeline defined by the logical decision moments output by the protocol security state machine. Each decision moment is associated with a monotonically increasing logical counter. Each node measures the distance between nodes by exchanging secure pilot signals, constructing a relative spatial coordinate system for the cluster; it associates its own sensor data with the logical decision moments; and it performs collaborative decision-making (including distributed AI inference, consensus algorithms, or rule-based decision-making) based on the fused state aligned with the logical timeline, outputting control commands. The system supports dynamically adding new nodes, dynamically updating the DSF to achieve forward key security, deriving individual keys to achieve cryptographic isolation, decentralized election, and millisecond-level seamless takeover. This scheme enables the cluster to achieve fully autonomous collaborative positioning, perception, and decision-making in environments without GPS, base stations, beacons, or prior maps, providing a distributed intelligent foundation independent of any external infrastructure for extreme scenarios such as military swarms, deep space exploration, and disaster relief.

Description

Technical Field

[0001] This invention belongs to the field of distributed artificial intelligence and autonomous systems technology, specifically relating to a system and method that, without any external infrastructure (including GPS, ground base stations, beacons, or prior maps), achieves relative positioning, environmental perception, decision-making, planning, and collaborative control through a distributed intelligent agent swarm synchronized via cryptography, relying on internal swarm collaboration. It is particularly suitable for extreme environments such as military drone swarms, deep space exploration swarms, disaster relief robots, and underwater unmanned vehicles.

[0002] Cross-reference to related applications This application is a divisional application of patent application No. 2026106192775, filed on May 8, 2026. The entire contents of the parent patent application are incorporated herein by reference. Background Technology

[0003] Distributed autonomous systems such as drone swarms and robot clusters have broad application prospects in fields such as military reconnaissance, coordinated strikes, disaster relief, and environmental monitoring. Current technologies for achieving swarm collaboration typically rely on one or more of the following infrastructures: 1. Global Navigation Satellite System (GNSS): The vast majority of drone swarms rely on GPS / BeiDou to obtain absolute position information and cannot operate in GPS-denied environments (electronic interference, indoors, tunnels, deep space).

[0004] 2. Ground beacons or base stations: Some solutions deploy ground beacons or base stations to provide positioning references, but these are unusable in scenarios deep behind enemy lines or without prior deployment.

[0005] 3. Prior map: Visual / LiDAR simultaneous localization and mapping (SLAM) requires a pre-built environmental map or relies on loop closure detection, and cannot be started in an unknown environment.

[0006] 4. Centralized scheduling: Most solutions rely on ground stations or queen bees for centralized decision-making. If the queen bee is destroyed, the system will be paralyzed, and there is a risk of signaling storms.

[0007] 5. Continuous full connectivity communication: Nodes need to frequently exchange status information, which cannot be guaranteed in a strong electromagnetic interference environment.

[0008] Furthermore, existing technologies have significant shortcomings in the following aspects: - Inconsistent time references: The sensing, communication, and decision-making modules of each node maintain independent time references, making integration difficult. - Inability to scale dynamically: Adding a new node to the cluster requires complex handshake signaling, and may even require restarting the entire system. - Lack of forward security: Once the key is leaked, all historical communications and flight paths are exposed. - Asynchronous AI decision-making: Each AI model infers independently, lacking a unified time anchor, making coordination difficult. Therefore, there is an urgent need for a collaborative solution for autonomous intelligent agent clusters that is zero-external-dependency, fully distributed, and intrinsically secure. Summary of the Invention

[0010] Key concept definition This application adopts the core concepts defined in the parent application, including but not limited to: Dynamic Security Foundation (DSF), Protocol Security State Machine (PSSM), logical time axis, logical decision moment, logical counter n, physical anchor moment T_anchor(n), and integrated sensing and computing. Detailed definitions of the above concepts can be found in the parent application's specification and will not be repeated here.

[0011] To facilitate understanding of this invention, the following concepts related to incremental innovation in this application are further defined: - Distributed autonomous intelligent agent cluster: A distributed system composed of multiple intelligent agents (drones, robots, vehicles, etc.), without a central node, without relying on any external infrastructure, and relying on internal collaboration within the cluster to complete the overall task.

[0012] - Relative spatial coordinate system: A relative position reference system established with a node in the cluster as the origin, used within the cluster. The position, velocity, and orientation of all nodes are represented in this coordinate system.

[0013] - Dynamic addition: During cluster operation, a new node can obtain the group DSF and current logical state by performing a single security handshake with any node already in the cluster, automatically synchronize the logical timeline and join the cluster without interrupting cluster operation.

[0014] - Dynamic DSF Update (Key Rotation): During operation, the cluster negotiates and generates a new group DSF according to a preset strategy or through a distributed consensus algorithm, and smoothly switches it at the specified future logical decision time to ensure forward security.

[0015] - Derived DSF (Individual Key): Each node generates an individual DSF based on the group DSF root key and its own unique identifier through cryptographic derivation functions. This individual DSF is used to generate the pilot sequence, encryption key, etc. of that node, thereby achieving cryptographic isolation between nodes.

[0016] - Distributed perception fusion: Each node in the cluster independently collects environmental data. After being aligned with the logical timeline, the nodes selectively share perception features to form a cluster-level environmental cognition.

[0017] Regarding the state transition rules: The state transition rule Rule_ID in this application is not limited to any specific form. The rule can be any deterministic state evolution function predefined by the network (such as a linear counter, hash chain, physical clock sampling, etc.), or it can be a user-defined custom rule through a programmable script. As long as both communicating parties share the same Rule_ID definition, deterministic synchronization of the logical state S(t) can be guaranteed. This application uses broadcast clock-driven rules (Rule A), hash chain-driven rules (Rule B), and logical epoch-driven rules (Rule C) as examples for illustration, but the scope of protection of this invention is not limited thereto.

[0018] Secure pilot frame (see parent specification): A physical layer signal frame containing secure pilots. It should be noted that the core of this invention lies in the generation and use of secure pilots; the secure pilot frame is merely one preferred form of transmission. Unless otherwise specified, "sending secure pilots" or "pilot transmission and reception" mentioned in this application include both sending an independent pilot sequence and sending a secure pilot frame containing that sequence. Secure pilot: Refers to a pilot sequence generated by a node based on the DSF and the current logical state using a cryptographic deterministic function.

[0019] Origin and Evolution of Inventive Ideas The parent patent (integrated sensing and computing) first proposed using a logical time axis as a unified foundation for communication, sensing, and computing, realizing the deep integration of sensing and computing within a single node.

[0020] Based on the parent application, this application extends the application of the logical timeline from individual intelligence to swarm intelligence. It proposes for the first time a distributed autonomous intelligent agent cluster collaborative system that does not rely on any external infrastructure, covering the entire lifecycle from cluster initialization, relative positioning, distributed perception, collaborative decision-making to dynamic expansion and key rotation.

[0021] Core Paradigm: Distributed Autonomous Collaboration This invention discloses and defines a fundamental operational paradigm for distributed autonomous intelligent agent clusters. Its core lies in: All agents within the cluster share the same logical timeline, relying solely on internal cluster collaboration to achieve localization, perception, decision-making, and control without depending on any external infrastructure.

[0022] This paradigm includes the following key mechanisms: 1. Logical Timeline Synchronization: All nodes operate PSSM independently based on a shared DSF, with strictly synchronized logical counters, and do not rely on GPS. 2. Relative Spatial Coordinate System Construction: A relative spatial coordinate system is constructed within the cluster through mutual transmission of secure pilot ranging. 3. Distributed Sensing Fusion: Aligning the sensing data from each node along a logical timeline to form a cluster-level environmental awareness. 4. Distributed collaborative decision-making: Using the logical steps as the time anchor, each node reasons independently but maintains consistency in decision-making time. 5. Dynamic self-organization: Supports dynamic node joining, dynamic DSF updates, decentralized election, and seamless takeover. Logical timeline synchronization This application fully inherits the mechanisms from the parent design regarding PSSM operation, logic counter synchronization, and silent calibration (see the parent design specification). All nodes share the same group DSF root key K_sec_root and the agreed initial logical state.

[0023] Initial synchronization mechanism: This application supports the following two initial synchronization modes, neither of which depends on any external infrastructure: 1. Unified Power-On Mode: When all nodes receive the DSF configuration at the same physical moment (e.g., unified power-on deployment before wartime), a common logical startup clock is agreed upon (e.g., "the 1000th local clock clock clock after receiving the configuration"). Each node independently starts the PSSM from this starting point, and the logical counter synchronously increments from 0.

[0024] 2. Asynchronous Joining Mode: When nodes start in batches, the first node to start runs PSSM and broadcasts a security pilot. Subsequent nodes listen to the pilot and utilize the silent calibration mechanism described in the parent case—matching the token generated by the candidate value of the local counter with the received token within a preset time window—to obtain an estimate of the current logic counter. They then complete precise synchronization at the agreed-upon future logic decision point, with the error converging to within ±1 step.

[0025] Regardless of the mode used, once synchronization is established, each node runs PSSM independently, and the logical counter n remains strictly synchronized across the cluster without any continuous signaling interaction.

[0026] Construction of relative spatial coordinate system Without any initial absolute position reference, this invention establishes a relative coordinate system within the cluster through the following steps: Step 1: Mutual Ranging. At each logical decision time n, all nodes broadcast a safety pilot frame. The pilot sequence is generated based on the individual DSF. After receiving the pilot from other nodes, each node measures the arrival time t_actual using the matched filtering method described in the parent case. It compares this time with the theoretical expected time t_expected to obtain the propagation delay τ = t_actual – t_expected, and calculates the distance d = c·τ.

[0027] Step 2: Distance Matrix Construction. After multiple logical decision-making moments, each node accumulates distance measurements with other nodes, forming a distributed distance matrix.

[0028] Step 3: Coordinate Calculation. Each node independently runs a distributed positioning algorithm (such as Multidimensional Scaling (MDS) or Distributed Kalman Consistent Filtering) to calculate the relative positions of all nodes within the cluster, constructing a relative spatial coordinate system with itself as the origin. The coordinate systems of each node can be unified by broadcasting its own origin position, or by selecting a global reference node through distributed consensus.

[0029] Step 4: Continuous Updates. At each logical decision point, the node updates its relative coordinate system using the new ranging results, supporting dynamic topology changes.

[0030] Distributed perception fusion Each node associates its own sensor data (vision, LiDAR, IMU, etc.) with a logical decision time n. Based on task requirements, nodes selectively broadcast perceived features (rather than raw data) to neighboring nodes. Receiving nodes align shared features with their own features using the same number of logical steps n, forming a cluster-level environmental awareness. The fusion of all perceived data uses a logical counter n as a unified time index, eliminating sampling time discrepancies between nodes.

[0031] Distributed collaborative decision-making Each node performs collaborative decision-making based on its logically time-aligned fusion state. This collaborative decision-making includes, but is not limited to, distributed AI inference, distributed consensus algorithms, or rule-based decision-making.

[0032] In a preferred embodiment, the AI ​​model (neural network, reinforcement learning, etc.) uses logical decision time n as the time anchor. At each logical decision time n, each node: - Read the current logical step number n - Read its own position in the cluster's relative coordinate system - Read the shared perceptual features of neighboring nodes - Output control commands for this node through distributed reinforcement learning or consensus algorithms. Decisions are made by all nodes at different phases of the same logical step, naturally maintaining time consistency. Updates to AI model parameters are also performed synchronously at the specified logical decision time, ensuring consistent model versions within the cluster.

[0033] Dynamic joining mechanism When a new node is added during cluster operation, the following steps are performed: Step J1: Discovery. The new node listens to the security pilot frames broadcast by nodes within the cluster and extracts group information from them.

[0034] Step J2: Handshake. The new node completes a single secure handshake with any node already in the cluster, obtaining an encrypted copy of the group DSF root key and an estimate of the current logical counter, n_est.

[0035] Step J3: Synchronization. At the agreed future logical decision time n_join, the new node starts its local PSSM, initializes the logical counter to n_est, and uses a silent calibration mechanism to converge the error to within ±1 step within several logical ticks, achieving precise synchronization.

[0036] Step J4: Broadcast. The new node broadcasts an acknowledgment pilot signal at its first logical decision moment and simultaneously begins participating in mutual ranging. Other nodes in the cluster update their neighbor lists, incorporating the new node into the relative coordinate system calculation.

[0037] The entire joining process does not interrupt the cluster operation, and the new node is plug-and-play.

[0038] Dynamic DSF update mechanism To prevent security risks arising from the long-term use of keys, this invention supports dynamic updating of the DSF during cluster operation: Step U1: Trigger. Triggering conditions include, but are not limited to: reaching a preset time threshold, the number of nodes changing beyond the threshold, an external security event, or a node making an active request.

[0039] Step U2: Negotiation. Negotiate a new DSF (new K_sec_root and / or new Init_Anchor) using a distributed consensus algorithm (improved Raft or PBFT, with the logical timeline as the time reference). The negotiation window covers the overlap period between the old and new DSFs.

[0040] Step U3: Distribution. The new DSF is encrypted and distributed to each node through the group secure channel. The amount of data distributed is extremely small (only tens of bytes).

[0041] Step U4: Switching. All nodes simultaneously enable the new DSF at the preset future logical decision time n_switch, and the old DSF is deactivated after n_switch-1. The logical counter n of each node is not interrupted at the switching time.

[0042] Step U5: Rollback. If some nodes do not receive the new DSF, they will still use the old DSF to send a re-request after n_switch, triggering the redistribution process.

[0043] Derived DSF hierarchy Each node generates its individual DSF based on the group DSF root key K_sec_root and its own unique identifier ID (such as MAC address, serial number, or dynamically assigned temporary ID) using cryptographic derivation functions: > K_sec_node = PRF(K_sec_root, ID || “NODE”) Where PRF stands for Pseudo-random function in cryptography (such as HMAC-SHA256). Each individual DSF can be further used to derive subkeys for various purposes: > - Communication key: K_comm = PRF(K_sec_node, “COMM”) > - Locating the pilot key: K_pilot_loc = PRF(K_sec_node, “LOC”) > - Encryption key: K_enc = PRF(K_sec_node, “ENC”) > - AI model key: K_ai = PRF(K_sec_node, “AI”) The derivation mechanism ensures cryptographic isolation between nodes. Once a single node is compromised, the attacker can only obtain that node's individual key; they cannot deduce the group root key or decrypt communications from other nodes. The group DSF root key is not transmitted directly in plaintext between nodes; it is always distributed through an encrypted channel.

[0044] Decentralized elections and seamless takeover The cluster can dynamically elect a temporary coordinating node (queen bee) to collect and summarize sensing information, distribute collaborative tasks, and communicate externally. The election mechanism is based on deterministic rules. For example, at the same logical decision moment, all nodes sort the current active neighbor list by node ID and select the node with the smallest ID as the queen bee. Since all nodes share the same neighbor list (obtained through pilot communication) and sorting rules, the calculation results are naturally consistent, requiring no additional signaling.

[0045] When the queen bee is destroyed, the remaining nodes automatically re-execute the election process at the next logical decision point to produce a new queen bee. The election result takes effect at logical step n+1, with an interruption time of only one logical tick (e.g., 0.125ms). The AI ​​decision model seamlessly switches to the new queen bee, and flight control remains uninterrupted. Beneficial effects

[0046] 1. Autonomous Collaborative Capability with Zero External Dependencies: In a preferred embodiment, all operations of this invention do not rely on any external infrastructure, including but not limited to Global Navigation Satellite Systems (GNSS), ground base stations, ground beacons, and prior maps, and can be deployed and operated in completely unknown environments without any external signals. This characteristic makes this system particularly suitable for extreme scenarios such as GPS denial, strong electromagnetic countermeasures, deep space exploration, and disaster relief.

[0047] 2. High battlefield survivability: No central node; failure of any node does not affect the overall function of the cluster; seamless takeover within milliseconds after the queen bee is destroyed. 3. Low communication overhead: Only requires broadcasting short pilot signals (such as symbols 64-128) and lightweight sensing characteristics, adaptable to strong electromagnetic warfare environments. 4. Natural time consistency: The logical timeline unifies the perception, decision-making, and control timing of all nodes, eliminating the need for additional synchronization signaling. 5. Dynamically scalable: New nodes are plug-and-play, supporting elastic scaling of the cluster, and the process of adding nodes does not interrupt cluster operation. 6. Forward Security: Dynamic DSF updates ensure that even if the key is compromised at some point, historical communications and flight paths remain secure. 7. Cryptographic Isolation: Derived DSF ensures that a single point of failure does not affect the security of other nodes in the cluster. 8. Intrinsic Security: Unpredictable pilot signals, resistant to replay attacks, and resistant to deception. 9. Integrated Sensing, Communication, and Computing: Sensing, communication, and computing are naturally aligned on a unified logical timeline, providing a stable benchmark for AI collaboration. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the system architecture of the present invention, showing a distributed intelligent agent cluster with a shared logical time axis.

[0049] Figure 2 Flowchart for constructing a relative spatial coordinate system.

[0050] Figure 3 Flowchart for dynamically adding mechanisms.

[0051] Figure 4 Update the timing diagram for dynamic DSF.

[0052] Figure 5 This is a diagram of the derived DSF hierarchy.

[0053] Figure 6 This is a sequence diagram for distributed collaborative decision-making.

[0054] Figure 7 A flowchart for decentralized elections and seamless takeover.

[0055] Figure 8 This is a sequence diagram of the combat process for Example 1 (military drone swarm). Detailed Implementation

[0056] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0057] Example 1: Complete Operational Process of Military Unmanned Aerial Vehicle Swarm A swarm of 96 drones carried out a reconnaissance and strike mission deep behind enemy lines, without GPS, ground beacons, or pre-prepared maps.

[0058] Pre-flight injection (unified power-on mode): 96 UAVs are deployed and powered on simultaneously before the operation. The ground station injects the group DSF root key K_sec_root into all UAVs via a secure channel, with the initial logic counter n(0)=0, and the start anchor point being "the 1000th local clock tick after receiving the configuration". Since all nodes receive the configuration simultaneously and their local clock start points are consistent, this anchor point corresponds to the same absolute physical time.

[0059] Ascent and Synchronization: Each UAV simultaneously activates its PSSM when its local clock reaches the 1000th tick, and the logic counter synchronously increments from 0 (T_step = 0.5ms). No inter-node synchronization signaling is required.

[0060] Note: If the cluster cannot be powered on uniformly (e.g., takeoffs in batches), the "asynchronous joining mode" described in this patent specification is adopted—the later-starting node obtains the estimated value of the logic counter by listening to the pilot signal of the earlier-starting node and using a silent calibration mechanism, and completes precise synchronization at the agreed future logic decision time. For the sake of simplicity, this embodiment uses the uniform power-on mode as an example.

[0061] Relative coordinate system establishment: During the first 50 logical cycles after takeoff, all UAVs exchange safety pilot signals. Each UAV measures its distance from the other 95 UAVs, runs a distributed MDS algorithm, and completes the relative coordinate system construction at logical cycle n=100.

[0062] GPS-free penetration: When the swarm enters the enemy's electronic jamming zone, all GPS signals are lost. Relying on DSF pilot signal transmission, the swarm maintains a relative positioning accuracy of ±0.5 meters, and the formation remains unchanged.

[0063] Dynamic addition: Four reinforcement drones take off from the rear. They listen to the swarm pilot, handshake with the edge nodes, and seamlessly join the cluster at logical tick n=15600, taking approximately 1.2 seconds.

[0064] Key rotation: After 30 minutes of flight, the swarm automatically triggers a DSF update. Each node negotiates a new key through distributed consensus, and the switch occurs simultaneously at the 5000th logical tick (2.5 seconds later). The old key is destroyed, ensuring forward security.

[0065] Queen Bee Destroyed: The queen bee (lowest ID) is shot down by the enemy. The remaining 95 drones automatically elect a new queen bee (second lowest ID) in the next logical tick (0.5ms later). The AI ​​decision-making model switches seamlessly, and the formation commands are not interrupted.

[0066] AI-driven collaborative decision-making: Edge nodes in the cluster detect ground targets, extract visual features, and broadcast these features (not the original image) to neighboring nodes at logical tick n=125000. Each node aligns the features according to the same number of logical ticks, and the fusion confirms the target's location. The queen bee distributes attack commands at logical tick n=125001, and 95 drones maneuver synchronously.

[0067] Return to home: Return to friendly area, some nodes restore GPS, share absolute position with the entire hive via group broadcast, and automatically calibrate relative coordinate system.

[0068] Example 2: Deep Space Exploration Cluster Three Mars rovers form a cluster, with a one-way communication delay of several minutes to Earth, and no GPS or ground beacons. A group DSF (Digital Sequence Message) is injected before launch. Each rovers operate their PSSM (Planetary Sequence Message Message) independently, with their logic counters synchronized. On the Martian surface, they measure distances to each other by exchanging pilot signals to establish a relative coordinate system. Due to the large distances between rovers (tens to hundreds of kilometers), the pilot propagation delay is not negligible. This system employs a two-way time comparison method: each rovers transmit a pilot signal at logic time n and record their local transmission time. The receiver, after receiving the pilot signal at logic time n, sends back its local reception time at the next logic time n+1. The transmitters exchange information, subtract signal processing delays, and accurately calculate the one-way propagation delay to obtain high-precision distance. During the mission, when one rovers discover a potential sampling point, they share its location information (relative coordinates) with the other rovers via a direct link, and all three arrive collaboratively. The DSF is dynamically updated every Martian day to ensure the safety of the long-term mission.

[0069] Example 3: Disaster Relief Robot Cluster At the site of a building collapse, there was no GPS signal or base station coverage. A swarm of rescue robots established a logical timeline through short-range communication (SRF) negotiation within a group. Each robot constructed a relative coordinate system by exchanging pilot signals for ranging. Sensor data (thermal imaging, radar) was selectively shared after being correlated with logical steps, forming a swarm-level environmental awareness. The collapse caused some robots to lose contact; the remaining robots automatically elected a new coordinating node to continue the search and rescue operation.

[0070] This invention can be widely applied to scenarios such as military drone swarms, urban counter-terrorism, disaster relief, deep space exploration, underwater unmanned vehicle collaboration, autonomous driving formations, and warehouse robot clusters. Its characteristics of zero external dependence, high survivability, low communication overhead, and inherent security make it particularly suitable for providing reliable distributed autonomous collaborative capabilities under harsh conditions such as GPS-denied environments, strong electromagnetic countermeasures, and lack of infrastructure deployment.

Claims

1. A distributed autonomous intelligent agent cluster collaborative system based on dynamic security foundation, characterized in that, include: At least two agent nodes, each configured as follows: - A logical timeline is generated based on a shared Group Dynamic Security Foundation (DSF), which is defined by logical decision moments output by the Protocol Security State Machine (PSSM), with each logical decision moment associated with a monotonically increasing logical counter n; - By exchanging safety pilot signals to measure the distance between nodes, a relative spatial coordinate system for the cluster is constructed; - Associate its own sensor data with the logical decision-making moment; - Based on the fusion state aligned with the logical time axis, perform collaborative decision-making, which includes but is not limited to distributed AI inference, consensus algorithms or rule-based decision-making, and output control instructions.

2. The system according to claim 1, characterized in that, The group DSF includes a group root key K_sec_root, an initial anchor point Init_Anchor, and a state transition rule identifier Rule_ID; the state transition rule identified by Rule_ID is a preset deterministic evolution rule, which is not limited to a specific rule type.

3. The system according to claim 2, characterized in that, The state transition rule includes at least one of the following: - Broadcast clock driving rule: The logical state S(t) is physical layer timing information, and the logical decision is triggered by the timing information update boundary; - Hash chain driven rule: The logical state S(t) is the secret hash chain value, and the logical decision is periodically triggered by a predefined logical tick. - Logic Epoch Driving Rule: The logic state S(t) is a public logic counter (Epoch_ID, Counter), and the logic decision moment is periodically triggered by the logic tick.

4. The system according to claim 1, characterized in that, Multiple nodes run PSSM independently, and the logical counters are synchronized across the cluster. The nodes absorb physical clock drift through a silent calibration mechanism. The silent calibration includes: the master node periodically broadcasts a verification token generated based on the current logical state, and the slave nodes match the token with the candidate value of their local counter within the time deviation window. After a successful match, the local counter is silently calibrated.

5. The system according to claim 1, characterized in that, The construction of the relative spatial coordinate system includes: - At each logical decision time n, all nodes broadcast a safety pilot frame; - Each node measures the arrival time of the received pilot signal and calculates the distance between nodes; - Each node independently runs a distributed positioning algorithm to calculate the relative position of each node within the cluster.

6. The system according to claim 1, characterized in that, Nodes selectively share sensing features, and receiving nodes align the shared features with their own features using the same number of logical steps n, forming a cluster-level environmental cognition.

7. The system according to claim 1, characterized in that, Distributed AI inference uses the logical decision moment n as the time anchor. Each node reads the current logical step number, its own position in the cluster's relative coordinate system, and the perceptual features shared by neighboring nodes at the logical step number n. It then outputs the control command of its own node through distributed reinforcement learning or consensus algorithms. The decisions of all nodes naturally maintain temporal consistency at the same logical step number.

8. The system according to claim 1, characterized in that, New nodes are dynamically added to the cluster using the following steps: - Listen for security pilot frames broadcast by nodes within the cluster; - Complete a single security handshake with any node already in the cluster to obtain the estimated value of the group DSF and the current logical counter; - Start PSSM at the agreed-upon future logical decision time, and precisely synchronize and join the cluster through a silent calibration mechanism; - The entire joining process does not interrupt cluster operation.

9. The system according to claim 1, characterized in that, The cluster supports dynamic DSF updates during operation: - Negotiate a new group DSF through distributed consensus; - The new DSF is encrypted and distributed to each node; - All nodes simultaneously enable the new DSF at the preset future logical decision time n_switch, and the old DSF is stopped after n_switch-1; - Achieve forward security of the key, preventing the leakage of historical communications.

10. The system according to claim 1, characterized in that: - Each node derives its individual DSF based on the group DSF root key K_sec_root and its own unique identifier ID: K_sec_node = PRF(K_sec_root, ID || "NODE"); - The derived individual DSF is further used to generate the pilot sequence and encryption key of the node, achieving cryptographic isolation between nodes; - The group DSF root key is not transmitted directly in plaintext between nodes.

11. The system according to claim 1, characterized in that: - The cluster dynamically elects a temporary coordinating node (queen bee). The election is based on deterministic rules, and all nodes independently calculate the results at the same logical decision-making moment. - After the queen bee is destroyed, the remaining nodes will automatically elect a new queen bee at the next logical decision moment, with the interruption time not exceeding one logical beat.

12. The system according to claim 1 or 7, characterized in that, The AI ​​model parameters are also updated synchronously at the designated logical decision-making time to ensure that the model version is consistent across all nodes in the cluster.

13. The system according to claim 1, characterized in that, The node can be any one of a drone, robot, vehicle, satellite, or underwater vehicle; the sensor includes at least one of an inertial measurement unit, a visual sensor, a lidar, or a radar.

14. A distributed autonomous intelligent agent cluster collaboration method based on dynamic security foundation, characterized in that, include: At least two agent nodes share a group DSF and each independently generates a logical timeline. The logical timeline is defined by the logical decision moments output by the PSSM, and each logical decision moment is associated with a monotonically increasing logical counter n. Each node measures the distance between nodes by exchanging safety pilot signals, thus constructing a relative spatial coordinate system for the cluster; Each node associates its own sensor data with the logical decision-making moment; Each node performs collaborative decision-making based on the fusion state aligned with the logical timeline. The collaborative decision-making includes, but is not limited to, distributed AI inference, consensus algorithms, or rule-based decision-making, and outputs control commands.

15. The method according to claim 14, characterized in that, Also includes: New nodes dynamically join the cluster by listening to cluster pilot signals, shaking hands with nodes in the cluster, starting PSSM at agreed logical times, and achieving precise synchronization through silent calibration.

16. The method according to claim 14, characterized in that, Also includes: The cluster negotiates a new DSF through distributed consensus, and all nodes switch simultaneously at the preset logical decision time to achieve dynamic key updates.

17. The method according to claim 14, characterized in that, Also includes: Each node derives its own DSF based on the group DSF root key and its own ID, thus achieving cryptographic isolation between nodes.

18. The method according to claim 14, characterized in that, Also includes: The cluster dynamically elects a temporary coordinating node. After the queen bee is destroyed, the remaining nodes automatically elect a new queen bee at the next logical decision moment.

19. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the method of any one of claims 14 to 18.

20. An intelligent agent node, comprising a processor, a memory, a transceiver, and a sensor interface, characterized in that, When the processor executes the program in the memory, it implements the node-side steps of the method according to any one of claims 14 to 18.

21. A distributed autonomous intelligent agent cluster, characterized in that, It includes multiple intelligent agent nodes as described in claim 20.

Images