Heterogeneous multi-agent microsecond-level synchronous control system and method based on low-delay radio frequency

Abstract

The invention discloses a multi-agent synchronous control system and method. The system comprises a single controller radio frequency broadcast module which delays a broadcast instruction at a microsecond level through a 2.4 GHz / 915MHz frequency band; the heterogeneous platform adapter interface is used for uniformly mapping the unmanned aerial vehicle / robot / aircraft into a position-speed-attitude triple; the task semantic analysis module is used for converting high-level instructions such as translation / rotation of a virtual rigid body into relative motion constraints; the dynamic relay management module is used for selecting relay nodes based on a signal quality scoring formula; and the intelligent planning module analyzes a natural language through a pre-training language model to generate an anti-collision task sequence. According to the method, a timestamp synchronization mechanism is adopted, all agents receive and execute instructions at the moment, and the geometric configuration is kept unchanged. The method is suitable for scenes of cluster performance, air-ground cooperation and the like, and solves the technical problems of high delay, poor heterogeneous compatibility and non-visual interaction of traditional communication.

Description

Technical Field

[0001] This invention relates to the field of multi-agent cooperative control technology, specifically to a heterogeneous multi-agent microsecond-level synchronous control system and method based on low-latency radio frequency communication, applicable to cooperative control scenarios of UAV swarms, ground robots, underwater vehicles and other mobile intelligent agents. Background Technology

[0002] With the rapid development of unmanned systems technology, multi-agent collaborative control has been widely used in fields such as commercial performances of drones or robots, logistics and transportation, and environmental monitoring.

[0003] In the process of implementing this invention, the inventors discovered that the existing technology mainly suffers from the following technical defects: excessively high communication latency (Wi-Fi / 4G / 5G latency >10ms), leading to asynchronous actions; complex architecture, with a high risk of single-point failure in master-slave mode; low instruction level, lacking geometric semantic expression capabilities such as "virtual rigid bodies"; strong platform specificity, with inconsistent cross-platform interfaces; and insufficient intelligence, with large language models not effectively integrated with microsecond-level synchronization. Therefore, a multi-agent synchronization control system with low communication latency, simple architecture, high instruction level, and strong versatility is needed. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-agent synchronous control system with low latency, support for heterogeneous platforms, and compatibility with natural language interaction.

[0005] The first aspect of this invention provides a heterogeneous multi-agent microsecond-level synchronization control system, comprising:

[0006] A single-controller RF broadcast module broadcasts via the unlicensed 2.4GHz / 915MHz frequency band with microsecond-level communication latency;

[0007] Heterogeneous platform adapter interface maps rotary-wing drones, fixed-wing aircraft, ground robots, quadruped robots, and underwater vehicles to a unified position-velocity-attitude interface;

[0008] The task semantic parsing module identifies translation, rotation, scaling, and deformation commands of virtual rigid bodies and generates relative position constraints.

[0009] The timestamp synchronization module implements distributed clock synchronization based on 64-bit microsecond precision timestamps;

[0010] The dynamic relay management module enables inter-agent relay forwarding when the direct signal strength is less than a preset threshold strength.

[0011] The intelligent planning integration module calls a pre-trained language model to convert natural language into a collision avoidance task sequence.

[0012] Optionally, the transmit power of the single-controller RF broadcast module is adjustable, ranging from 100mW to 2W.

[0013] Optionally, the microsecond-level communication delay of the single-controller RF broadcast module refers to a communication delay of <200μs.

[0014] Optionally, in the dynamic relay management module, the preset threshold strength can be preset to... When the direct signal strength RSSI < At that time, inter-agent relay forwarding is enabled.

[0015] A second aspect of the present invention provides a heterogeneous multi-agent microsecond-level synchronization control method based on low-latency radio frequency, comprising:

[0016] Controller in Simultaneously broadcast synchronization command frames containing timestamps;

[0017] All intelligent agents in Receive at all times ( (Less than the communication delay of a single-controller RF broadcast module), extract timestamps to correct the local clock;

[0018] Execute instructions based on a unified interface to maintain the geometric relationships of the virtual rigid body;

[0019] The relay forwarding mechanism is triggered when the signal is lost;

[0020] time out If no command is received, the system will enter hover / stationary safety mode.

[0021] Optionally, the Within the communication delay range of a single-controller RF broadcast module, when the communication delay of the single-controller RF broadcast module is <200μs, <200μs.

[0022] A third aspect of the present invention provides a computer-readable storage medium and an electronic device for performing the above-described method. Attached Figure Description

[0023] Figure 1 This is a diagram of the overall system architecture.

[0024] Figure 2 This is a timing diagram for communication between a single controller and multiple agents.

[0025] Figure 3 The flowchart for task semantic parsing and virtual rigid body control;

[0026] Figure 4 Abstract the interface block diagram for heterogeneous platforms;

[0027] Figure 5This is a state transition diagram for a dynamic relay mechanism.

[0028] Figure 6 Flowchart of data processing for the intelligent planning integration module;

[0029] Figure 7 This is a schematic diagram illustrating a cluster light show application scenario.

[0030] Figure 8 This is a schematic diagram of an air-ground collaborative control application scenario. Detailed Implementation

[0031] The present invention will be further described below with reference to specific embodiments.

[0032] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. The terms "and" and "or" as used herein include any and all combinations of one or more of the associated listed items.

[0033] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0034] Embodiment 1 of this invention is the overall system architecture; this embodiment is a specific implementation of the overall system architecture, such as... Figure 1 .

[0035] This embodiment consists of the following parts: the system includes a ground control unit and an intelligent agent cluster.

[0036] The handheld controller has a built-in radio frequency transmission module (supports 2.4GHz and 915MHz, power 100mW-2W).

[0037] The intelligent agent cluster is equipped with an RF receiving module, an embedded processor, and a sensor fusion unit;

[0038] The communication link is divided into the main link (direct connection delay <200μs) and the relay link (forwarding delay <5ms).

[0039] Embodiment 2 of the present invention is a microsecond-level synchronous communication mechanism. This embodiment is the implementation of the communication mechanism, such as... Figure 2 .

[0040] The controller in this second embodiment is Time broadcast command frame:

[0041] [Frame header 0xA5A5 | 64-bit timestamp | Address code mask | Task data | CRC32 checksum], frame length 32 bytes;

[0042] Agent A (RSSI > -80dBm) in = +130μs reception;

[0043] Agent B (RSSI < -85dBm) transmits data via relay node. ≈ +5ms reception;

[0044] The difference in the speed of light propagation is less than 3.3 μs within a 1000m range, which is negligible.

[0045] Embodiment 3 of this invention is a virtual rigid body task instruction parsing. This embodiment describes the task instruction processing flow, as follows: Figure 3 .

[0046] The processing flow for the input command "move the formation forward 5 meters" in this embodiment three is as follows:

[0047] The intelligent planning module is parsed as "{type:"TRANSLATION", vector:[5,0,0], frame:"WORLD"}";

[0048] Constraint calculation is ;

[0049] Collision avoidance verification ,in A safe distance of 1.5m is required.

[0050] Broadcast execution.

[0051] Embodiment 4 of this invention is an abstract interface for heterogeneous platforms. This embodiment unifies the communication interface, such as... Figure 4 .

[0052] The abstraction layer in this embodiment four converts the native control signals of each platform into standard triples:

[0053] In rotary-wing UAVs, the thrust T can be obtained by inputting a PWM signal, and the mapping function is... ;

[0054] In ground robots, the speed v can be obtained by inputting the encoder signal, and the speed can be obtained by using the differential encoder signal.

[0055] In a quadruped robot, the center of mass pose can be obtained by inputting joint angles. Through inverse kinematics, the state feedback protocol of all platforms is unified as "[ID | Timestamp | x,y,z | vx,vy,vz | qw,qx,qy,qz]".

[0056] Embodiment 5 of the present invention is a dynamic relay mechanism. This embodiment describes the dynamic selection of relay agents, such as... Figure 5 .

[0057] In this fifth embodiment, agent i scores the relay of j:

[0058] ,

[0059] Where α=0.5, β=200, and γ=0.1 are the weights. Based on historical average delay, This represents the percentage of battery power. The node with the highest score is selected as the relay node.

[0060] The state transition conditions in Example 5 are as follows:

[0061] Normal → Detection, RSSI < -85dBm for 3 cycles (15ms);

[0062] Detection → Activation, Request Received and > Threshold;

[0063] Detection → Timeout: =500ms without receiving a command;

[0064] Timeout → Normal: Signal restored or new instructions received.

[0065] Embodiment 6 of the present invention is an intelligent planning integration module. This embodiment describes the process from natural language input to intelligent agent execution, such as... Figure 6 .

[0066] The processing flow of this embodiment six is ​​as follows:

[0067] Voice / text input → NLP parsing (latency <100ms);

[0068] Call a pre-trained language model (such as LLaMA-3-8B), and the prompt words include API instruction sets and security constraints;

[0069] Output the pseudocode task sequence, verified by collision detection (delay <50ms);

[0070] Radio frequency broadcast execution.

[0071] The working mode of this embodiment six:

[0072] Real-time mode, end-to-end latency <1 second;

[0073] Pre-programmed mode generates scripts and triggers execution.

[0074] Embodiment 7 of this invention is an intelligent agent cluster light show. This embodiment demonstrates the specific implementation of the system based on this invention in a large-scale outdoor light show performance, such as... Figure 7 .

[0075] The system in this embodiment seven is deployed in an open square airspace of 500m×300m, with a total of 200 heterogeneous intelligent agents, including 180 rotary-wing UAVs (open-source UAVs using Ardupilot flight control, with a takeoff weight of 6.3kg and a maximum flight altitude of 120m) and 20 ground mobile robots (model Clearpath Husky A200, with a maximum speed of 1.0m / s).

[0076] Each of the heterogeneous intelligent agents is equipped with a 2.4GHz radio frequency receiver module (chip: nRF24L01P, receiver sensitivity). ), embedded processor (STM32H750VBT6, main frequency 480MHz) and RGB-LED lighting effect device (single LED power 15W, refresh rate ≥60Hz).

[0077] The control command generation and semantic parsing in this embodiment seven are as follows:

[0078] The operator inputs a natural language command through the ground station software: "Display the word 'HELLO' and move forward 20 meters at a speed of 2 m / s."

[0079] The intelligent planning integration module calls the pre-trained language model (LLaMA-3-8B) for semantic parsing.

[0080] The prompt word engineering of the pre-trained language model includes the following constraints:

[0081] Character dot matrix resolution: 5×7 pixel matrix / character; character height: 15 meters.

[0082] Virtual rigid body constraints ensure that the formation topology retention rate is >99.5%.

[0083] Safe distance threshold, =1.5 meters (minimum clearance for collision avoidance);

[0084] The entire translation is completed within a 3-second execution time window.

[0085] The pre-trained language model outputs a standard task instruction format:

[0086] "{type: "TRANSLATION", pattern: "HELLO", vector: [20, 0, 0], speed:2.0, frame: "WORLD_FRAME", duration: 10.0}".

[0087] The spatiotemporal coordinate calculation and collision avoidance verification in this embodiment seven are as follows:

[0088] The task semantic parsing module determines the semantic parsing based on the current formation geometry (relative positions of each agent). Using virtual rigid body constraints, the target position sequence for each agent is calculated;

[0089] For the agent at the i-th pixel of the character 'H', its target position is:

[0090] ,

[0091] Where the rotation matrix Maintaining the identity matrix (without rotation), the velocity vector V = [2,0,0] m / s. ;

[0092] The collision avoidance verification module performs temporal geometric deduction to calculate the minimum distance between any two agents during a 10-second execution period:

[0093] ;

[0094] The verification results show that all pairs (i,j) satisfy... >1.5 meters, passed safety inspection.

[0095] The microsecond-level synchronous broadcast mechanism in this embodiment seven is as follows:

[0096] The handheld controller generates a 32-byte instruction frame at time T0:

[0097] "[Frame header 0xA5A5 (2 bytes) | 64-bit timestamp (8μs precision) | Address code mask (4 bytes) | Task data (16 bytes) | CRC32 (2 bytes)]";

[0098] The radio frequency transmission module broadcasts at a frequency of 2.4 GHz and a power of 500 mW, with an air transmission time of 128 μs.

[0099] All 200 intelligent agents in Real-time reception, including direct-connected agents (RSSI> ) =130±50μs, received by intelligent agents (15 in the edge region) through relay nodes. ≈5ms;

[0100] The timestamp synchronization module extracts a 64-bit timestamp and corrects the local clock, with a clock synchronization error of <5μs.

[0101] The execution effect and performance indicators of this embodiment seven are as follows: after the LED lights are lit synchronously, the formation moves as a whole 20 meters:

[0102] The pattern is intact, with the "HELLO" character retaining 99.8% of its original value, showing no tearing or misalignment.

[0103] Synchronization accuracy, start-up time difference of 200 intelligent agents =180μs (<200μs threshold);

[0104] Positioning error, relative position error <0.1 meters (RTK-GNSS positioning);

[0105] Communication coverage, control distance within 800 meters, RSSI> Packet loss rate <0.1%;

[0106] The system capacity supports a maximum of >200 nodes, and the broadcast period can be expanded to 256 nodes within 5ms.

[0107] Compared to traditional Wi-Fi solutions (latency > 10ms), this embodiment achieves microsecond-level synchronization, avoiding pattern tearing.

[0108] Compared to the preset waypoint scheme, this seventh embodiment reduces operational complexity by 90% through natural language interaction, and virtual rigid body constraints ensure geometric invariance, eliminating the need for individual programming.

[0109] Embodiment 8 of this invention presents an application scenario of air-ground collaborative control. This embodiment demonstrates the collaborative control of a drone and a ground robot in a warehouse inspection task, such as... Figure 8 .

[0110] The system in this embodiment eight includes one quadcopter drone (an open-source drone using Ardupilot flight control, with a maximum speed of 23 m / s) and one ground mobile robot (AGV-100, with a maximum speed of 2 m / s). The drone is equipped with a GNSS RTK module (horizontal positioning accuracy ±10 mm + 1 ppm) and an IMU (update frequency 200 Hz). The ground robot is equipped with a 2D lidar (model SICK LMS151, scanning frequency 50 Hz) and a wheel encoder (resolution 0.1 mm).

[0111] In this embodiment, the coordinate system is the world coordinate system (WORLD_FRAME), with the origin set at the warehouse entrance.

[0112] The heterogeneous platform adaptation interface in this embodiment eight maps two types of platforms into a position-velocity-attitude triplet, involving ① mapping of UAV end and ② mapping of ground robot end.

[0113] ① The UAV end mapping uses an electronically controlled PWM signal (1000-2000μs) as the native control interface, which is then converted into thrust commands by the abstraction layer through a dynamic model:

[0114] ,

[0115] in =0.012 N / μs, =-12N, the position vector is obtained through RTK-GNSS, the velocity vector is calculated through IMU and GNSS fusion, and the attitude vector uses quaternions. , , , ]express.

[0116] ② Ground robot end mapping, the native control interface is differential wheel speed ( , The abstract layer is transformed into the following through a kinematic model:

[0117] ,

[0118] ,

[0119] ,

[0120] in θ is the heading angle. The lidar SLAM provides relative positioning and is updated every 100ms.

[0121] The task semantic parsing and constraint generation in Example 8 are as follows:

[0122] The operator inputs the command "Keep the drone 2 meters directly above the robot and move synchronously to target point A (50,30,0)", which the intelligent planning module interprets as:

[0123] "{type: "FORMATION_MAINTAIN",leader: "GROUND_ROBOT",follower: "DRONE",relative_offset: [0, 0, 2],target: [50, 30, 0],speed: 1.5}";

[0124] The task semantic parsing module generates relative positional constraints:

[0125] ,

[0126] Where ε(t) is the dynamic compensation amount, used to correct positioning differences.

[0127] In this embodiment eight, the multi-frequency timestamp synchronization mechanism employs a timestamp synchronization strategy due to the difference in control frequencies (100Hz for the drone and 20Hz for the robot):

[0128] Synchronization Strategy ① Ground station software generates a unified timestamp (64-bit microsecond level);

[0129] Synchronization Strategy ② broadcasts a command frame every 50ms, including: robot target pose (20Hz control) and the next 5 interpolation points of the UAV (100Hz prediction).

[0130] Synchronization Strategy ③ After the UAV receives the data, it performs linear interpolation based on the timestamp:

[0131] .

[0132] The dynamic relay and communication guarantee in this embodiment eight are as follows:

[0133] Ground robots as relay nodes ( = ), relaying instructions to drones in areas obscured by warehouse shelves, with relay scoring as follows:

[0134] ,

[0135] in =2ms, =85%, score =89.2, which is higher than the threshold of 80, so relay mode is activated.

[0136] The execution effect and performance verification of this embodiment eight were demonstrated during a 50-meter coordinated movement:

[0137] The relative position is maintained, with the drone positioned 2.0 ± 0.15 meters directly above the robot and a horizontal deviation of < 0.2 meters;

[0138] Synchronization accuracy, action start time difference Δt = 25ms (within the robot's 20Hz cycle);

[0139] Positioning error: UAV RTK error ±0.02 meters, robot SLAM error ±0.1 meters, overall error <0.2 meters;

[0140] Communication latency: 120μs for direct links and 3.2ms for relay links;

[0141] The task completion time was 33.3 seconds (speed 1.5m / s), which is less than 2% different from the theoretical value.

[0142] The fault tolerance and security mechanism in this embodiment eight is as follows:

[0143] If the drone is If no command is received within 500ms, enter hover safety mode;

[0144] If robot communication is interrupted, the drone will maintain its current relative position and wait for it to be restored.

[0145] The system supports heterogeneous platform adaptation, eliminating the need for manual coordinate system calibration.

[0146] While preferred embodiments of the invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous variations, modifications, and substitutions will now occur to those skilled in the art without departing from the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed to practice the invention. The following claims are intended to define the scope of the invention, and methods and structures within the scope of these claims and their equivalents are also covered.

Claims

1. A heterogeneous multi-agent microsecond-level synchronous control system, characterized in that, include: The system includes a single-controller RF broadcast module configured to broadcast synchronization command frames containing 64-bit timestamps to multiple heterogeneous intelligent agents via an unlicensed 2.4GHz or 915MHz frequency band; a heterogeneous platform adapter interface configured to map the native control signals of rotary-wing UAVs, fixed-wing aircraft, ground robots, quadruped robots, and underwater vehicles into unified position-velocity-attitude triplets; a task semantic parsing module configured to identify high-level geometric semantic commands for virtual rigid body translation, rotation, scaling, and deformation, and convert them into relative motion constraints between the intelligent agents; and a timestamp synchronization module configured to achieve distributed clock synchronization based on the 64-bit microsecond precision timestamps in the command frames, ensuring that all intelligent agents are synchronized. Execute instructions at all times.

2. A heterogeneous multi-agent microsecond-level synchronization control method based on low-latency radio frequency, characterized in that, include: Controller in Synchronization command frames containing 64-bit timestamps are generated at all times; The instruction frames are broadcast via the 2.4GHz or 915MHz unlicensed frequency band; All controlled agents synchronously receive and correct their local clocks based on the timestamps; they execute instructions based on a unified position-velocity-attitude triplet interface, maintaining the virtual rigid body geometric relationships unchanged. When the RSSI of the direct communication signal is lower than the preset threshold, the inter-agent relay mechanism is triggered.

3. The system according to claim 1, characterized in that, The heterogeneous intelligent agent includes at least two of the following: airborne aircraft, ground mobile platforms, and underwater vehicles.

4. The system according to claim 1, characterized in that, It also includes a dynamic relay management module, configured to be based on a signal quality scoring formula. Select the optimal relay node, where α, β, and γ are preset weight coefficients. For historical delays between nodes, This represents the remaining battery power.

5. The system according to claim 1, characterized in that, The position-velocity-attitude triplet includes: a three-dimensional position vector [x, y, z], a three-dimensional velocity vector [x, y, z], and a three-dimensional velocity vector [x, y, z]. , , ] and quaternion attitude vector[ , , , ].

6. The system according to claim 1, characterized in that, It also includes an intelligent planning integration module, configured to call a pre-trained language model to convert natural language instructions into a collision avoidance task sequence, and execute it after verification by temporal geometric deduction.

7. The method according to claim 2, characterized in that, The high-level geometric semantic commands include virtual rigid body translation, rotation around an axis, scaling, and topology-preserving deformation.

8. The method according to claim 2, characterized in that, When the agent reaches the preset timeout threshold If no valid command is received, it will automatically enter a hover or stationary safe mode.

Images