A method and system for autonomous operation of unmanned equipment driven by an edge-side intelligent agent

By deploying intelligent agents on the edge of unmanned equipment, local perception, decision-making, and execution are achieved, solving the problems of human dependence and network latency in the operation control of unmanned equipment, improving the real-time performance and safety of operations, and adapting to various types of unmanned equipment.

CN123067435APending Publication Date: 2026-11-10JUNERDA (SHANGHAI) DIGITAL TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611303197.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-26
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Among the existing unmanned equipment operation control methods, remote control relies on high human costs, while cloud control suffers from large data transmission delays, high network dependence, and high computing resource pressure, which cannot meet the operational needs of high-speed movement and communication blind spots.

Method used

Deploy intelligent agents on the edge of unmanned equipment to achieve local perception, decision-making and execution, and support autonomous switching in case of communication failure. The agents include a perception module, a decision-making module, an execution module and a state management module, and have autonomous perception, decision-making and execution capabilities, as well as an autonomous mode switching mechanism.

Benefits of technology

It reduces end-to-end decision latency, improves the real-time performance and robustness of operations, ensures the continuity and security of operations, and reduces system development and maintenance costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure FT_1
    Figure FT_1
  • Figure FT_2
    Figure FT_2
  • Figure FT_3
    Figure FT_3
Patent Text Reader

Abstract

This application discloses an edge-side intelligent agent-driven autonomous operation method and system for unmanned equipment. The method includes: deploying an edge-side intelligent agent on the unmanned equipment, the edge-side intelligent agent including a perception module, a decision-making module, an execution module, a communication module, and a state management module; receiving task instructions from a cloud platform through the communication module; acquiring real-time perception data through the perception module; making local decisions based on the task instructions and perception data through the decision-making module to generate control instructions; controlling the unmanned equipment to execute the operation task through the execution module; and recording and reporting the task execution status through the state management module. When the communication link is abnormal, the edge-side intelligent agent switches to edge autonomous mode to independently complete the decision-making. This application achieves local perception and decision-making by deploying an intelligent agent on the edge, significantly reducing decision latency and ensuring operational continuity when communication is abnormal, while also supporting unified adaptation to various types of unmanned equipment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and unmanned equipment control technology, specifically to an end-side intelligent agent-driven autonomous operation method and system for unmanned equipment. Background Technology

[0002] With the rapid development of artificial intelligence and robotics technologies, unmanned equipment (including quadrupedal robot dogs, drones, and unmanned cleaning vehicles) is increasingly being used in inspection, security, cleaning, and logistics scenarios. Currently, the operation control of unmanned equipment mainly employs the following two methods: The first method is remote control, where operators directly control the movement and operation of unmanned equipment via a remote control or remote control terminal. This method requires full operator involvement, resulting in high labor costs. Furthermore, the operator's reaction speed and operational precision directly affect the operational results, making it difficult to meet the needs of large-scale, long-term operations. The second approach is centralized cloud control, where a cloud platform receives the perception data from unmanned equipment, performs decision-making calculations in the cloud, and then sends control commands to the unmanned equipment for execution. This approach has the following problems: First, data transmission latency is significant; the end-to-end latency from uploading perception data to issuing control commands is typically between 200 and 500 milliseconds, which cannot meet the real-time control requirements of high-speed motion scenarios. Second, it is highly dependent on communication networks; when there are communication blind spots or weak network signals in the unmanned equipment's operating area, an interruption of the cloud control link will cause the unmanned equipment to stop operating or even lead to safety accidents. Third, the centralized uploading of perception data from multiple unmanned equipment units to the cloud for processing puts significant pressure on cloud computing resources and network bandwidth, limiting system scalability. Therefore, how to achieve autonomous perception, decision-making and execution at the edge of unmanned equipment, while maintaining coordination with the cloud platform, is a technical problem that urgently needs to be solved in the field of unmanned equipment operation control. Summary of the Invention

[0003] To address the aforementioned technical issues, this application provides an edge-driven autonomous operation method and system for unmanned equipment. By deploying an intelligent agent on the edge of the unmanned equipment, local perception, decision-making, and execution are achieved, and the system supports autonomous switching to edge autonomous mode in the event of communication failure, ensuring operational continuity and safety. In a first aspect, this application provides an autonomous operation method for unmanned equipment driven by an edge-side intelligent agent, the method comprising the following steps: S1: Deploy an edge-side intelligent agent on unmanned equipment. The edge-side intelligent agent includes a perception module, a decision-making module, an execution module, a communication module, and a state management module. The edge-side intelligent agent is a software system deployed on the unmanned equipment itself, running on an embedded computing platform carried by the unmanned equipment. The perception module is responsible for data acquisition, the decision-making module is responsible for local decision calculation, the execution module is responsible for converting the decision results into motion control signals for the unmanned equipment, the communication module is responsible for bidirectional data interaction with the cloud platform, and the state management module is responsible for recording and managing the task state. S2: The edge-side intelligent agent receives task instructions from the cloud platform through the communication module. The task instructions include task type, target location, and task constraints. The task types include inspection tasks, cleaning tasks, material handling tasks, and monitoring tasks. The target location is represented by coordinates in a geographic coordinate system or a local coordinate system. The task constraints include the operation time window, operation area boundaries, and safety limit parameters. S3: The edge-side intelligent agent acquires real-time perception data of the unmanned equipment through the perception module. The real-time perception data includes environmental perception data and equipment status data. The environmental perception data includes at least one of visual image data, LiDAR point cloud data, inertial measurement unit (IMU) data, satellite positioning data, and ultrasonic ranging data. Visual image data is acquired by the onboard camera and used for target recognition and scene understanding; LiDAR point cloud data is acquired by LiDAR and used for 3D environment mapping and obstacle detection; IMU data includes acceleration and angular velocity and is used for attitude estimation; satellite positioning data is used for global positioning; and ultrasonic ranging data is used for near-range obstacle detection. The equipment status data includes at least one of the following: battery level information, attitude information, speed information, load status, and fault codes. Battery level information reflects the remaining battery percentage of the unmanned equipment; attitude information includes pitch angle, roll angle, and yaw angle; speed information includes linear velocity and angular velocity; load status reflects the operating status of the onboard mission payload; and fault codes are used to identify equipment malfunctions. S4: The edge-side intelligent agent makes local decisions based on the task instructions and the real-time sensing data through the decision module, and generates control instructions. The local decision-making process includes the following sub-steps: The task instructions are decomposed into a sequence of subtasks. Taking an inspection task as an example, it can be decomposed into a path planning subtask, a navigation and movement subtask, a data collection subtask after reaching the checkpoint, and an anomaly reporting subtask. Based on the real-time perception data, an execution path and action sequence are generated for each subtask. The execution path is generated based on the global map and real-time obstacle information, and the path search is performed using the A* algorithm or Dijkstra's algorithm; the action sequence includes actions such as forward movement, turning, hovering, and data acquisition, along with their parameters. The execution path undergoes a safety check. If obstacles or risk areas are detected on the path, the execution path is replanned. The safety check is based on lidar point cloud data and ultrasonic ranging data to determine whether there are obstacles ahead of the path, and on slope data to determine whether there are risk areas with excessively steep slopes. The decision-making module also adaptively switches between cloud-guided mode, edge autonomous mode, and cloud-edge collaborative mode based on communication quality indicators fed back by the communication module. These communication quality indicators include signal strength, packet loss rate, and end-to-end latency. In cloud-guided mode, the decision-making module uploads the perceived data to the cloud platform and receives the decision results from the cloud platform. This mode is suitable for scenarios with good communication quality and can utilize the powerful computing capabilities of the cloud platform to handle complex decision-making tasks. In edge autonomy mode, the decision-making module makes decisions independently. This mode is suitable for scenarios with abnormal communication links or poor communication quality, ensuring that unmanned equipment can continue to perform tasks even in a network outage environment. In the cloud-edge collaborative mode, the decision-making module completes some decision-making tasks locally and uploads the remaining tasks to the cloud platform for collaborative completion. This mode is suitable for scenarios with average communication quality, balancing real-time performance and decision quality by reasonably allocating the computing load between local and cloud environments. S5: The edge-side intelligent agent controls the unmanned equipment to perform operational tasks according to the control instructions through the execution module. The execution module converts control commands into motion control signals for the unmanned equipment. For a quadruped robot dog, the control commands are converted into angle and torque commands for each joint; for a drone, the control commands are converted into rotor speed and attitude control commands; for an unmanned sweeper, the control commands are converted into driving speed, steering angle, and working parameters of the sweeping mechanism. S6: The edge-side intelligent agent records the task execution status through the status management module and reports the task execution status to the cloud platform through the communication module. The task execution status includes the currently executing subtask, task completion progress, device location, and device status. The status management module records the task execution status at a preset frequency (e.g., once per second) and reports it to the cloud platform via the communication module at a preset period. When the communication link between the edge-side intelligent agent and the cloud platform is abnormal, the edge-side intelligent agent switches to edge autonomy mode. In edge autonomy mode, the decision-making module independently makes local decisions and controls the unmanned equipment to continue performing its tasks. The conditions for determining communication link abnormality include: three consecutive failed heartbeat detections, signal strength below a preset threshold, or packet loss rate exceeding a preset ratio. Furthermore, the method also includes a safety handling step: the sensing module monitors the environmental safety status around the unmanned equipment in real time during operation; when an emergency safety event is detected, the edge-side intelligent agent triggers a safety protection action directly through the execution module without going through the cloud platform. The safety protection action is determined according to the type of unmanned equipment: for a quadruped robot dog, it triggers emergency braking and grounds; for a drone, it triggers hovering or automatic return; for an unmanned sweeping vehicle, it triggers emergency braking and shuts down the sweeping mechanism. Furthermore, the edge-side intelligent agent supports various types of unmanned equipment, including quadrupedal robot dogs, drones, and unmanned cleaning vehicles. The edge-side intelligent agent loads the corresponding device capability description file according to the type of unmanned equipment. This device capability description file defines the kinematic model, sensor configuration, and executable task types of the unmanned equipment. Through the dynamic loading mechanism of the device capability description file, the same edge-side intelligent agent software framework can be adapted to different types of unmanned equipment. Furthermore, the task instruction also includes a task priority and a task timeout. The status management module queues multiple tasks to be executed according to the task priority, with higher-priority tasks being executed first. When a task execution time exceeds the task timeout, the status management module triggers a task timeout handling process, which includes pausing the current task, reporting the timeout event, and switching to the next task. Furthermore, during the execution of the task, the edge-side intelligent agent collects scene data of the work area through the sensing module, associates and stores the scene data with the task execution status, and forms a work record. When the communication link is restored, the work record is synchronized to the cloud platform, realizing the complete transmission of work data during the network outage. Secondly, this application provides an autonomous operation system for unmanned equipment driven by an edge-side intelligent agent. The system includes an edge-side intelligent agent module deployed on the unmanned equipment. The edge-side intelligent agent module includes a perception unit, a decision-making unit, an execution unit, a communication unit, and a state management unit, with the functions of each unit corresponding to the functions of each module in the above-described method steps. The sensing unit includes at least one of a visual sensor, a lidar, an inertial measurement unit, a satellite positioning module, and an ultrasonic sensor, and is used to acquire environmental perception data and equipment status data. The decision-making unit includes a task decomposition subunit, a path planning subunit, a security verification subunit, and a mode switching subunit. The task decomposition subunit decomposes task instructions into a sequence of subtasks; the path planning subunit generates execution paths and action sequences based on real-time sensing data; the security verification subunit performs security verification on the execution paths; and the mode switching subunit adaptively switches between cloud-guided mode, edge autonomous mode, and cloud-edge collaborative mode based on communication quality indicators. The system also includes a safety processing unit, configured to monitor the environmental safety status in real time during operation, and to directly trigger safety protection actions when an emergency safety event is detected, without waiting for a response from the cloud platform. The edge-side intelligent agent module also includes a device adaptation unit, configured to load the corresponding device capability description file according to the type of unmanned equipment, so as to realize the adaptation of the same software framework to multiple types of unmanned equipment. Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. The beneficial effects of this application include: (1) By deploying end-side intelligent agents on the end of unmanned equipment, local perception, decision-making and execution can be realized. Data can be used to complete decisions without uploading to the cloud. The end-to-end decision delay is reduced from 200-500 milliseconds to 10-50 milliseconds, which significantly improves the real-time control performance. (2) When the communication link is abnormal, the end-side intelligent agent automatically switches to edge autonomous mode, and the unmanned equipment can continue to perform the operation in the network-off environment, avoiding operation interruption or safety accidents caused by communication interruption, and improving the robustness of the system. (3) Through the mode adaptive switching mechanism, the cloud-guided mode, edge autonomous mode and cloud-edge collaborative mode are dynamically switched according to the communication quality, which takes into account both real-time performance and decision quality, and realizes the rational allocation of computing resources. (4) When the safety processing unit detects an emergency safety event, it directly triggers the safety protection action without going through the cloud platform. The safety response delay is reduced from 200-500 milliseconds to less than 10 milliseconds, which effectively ensures the operational safety of unmanned equipment. (5) Through the dynamic loading mechanism of the device capability description file, the same set of end-side intelligent agent software framework can be adapted to various types of unmanned equipment such as quadruped robot dogs, drones, and unmanned cleaning vehicles, reducing the system development and maintenance costs. Attached Figure Description

[0004] Figure 1 A schematic diagram of the architecture of an unmanned equipment autonomous operation system driven by an edge-side intelligent agent provided in an embodiment of this application; Figure 2This is a schematic diagram of the module composition of the edge-side intelligent agent provided in the embodiments of this application; Figure 3 A flowchart of an autonomous operation method for unmanned equipment driven by an edge-side intelligent agent provided in an embodiment of this application; Figure 4 This is a flowchart illustrating the adaptive switching of decision modes provided in an embodiment of this application. Figure 5 A security processing flowchart provided for embodiments of this application; Figure 6 This is a schematic diagram illustrating how the edge-side intelligent agent adapts to various types of unmanned equipment, as provided in the embodiments of this application. Detailed Implementation

[0005] The technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments. Example 1: Inspection work of a quadruped robot dog driven by an edge-side intelligent agent This embodiment uses a quadruped robot dog performing inspection work in an industrial park as an example to illustrate the technical solution of this application. like Figure 1 As shown, the system in this embodiment includes a cloud platform and an edge-side intelligent agent deployed on a quadruped robot dog. The cloud platform is responsible for global task scheduling and status monitoring, while the edge-side intelligent agent is responsible for local perception, decision-making, and execution. like Figure 2 As shown, the edge-side intelligent agent includes the following modules: The perception module includes a binocular camera (1920x1080 resolution, 30fps), a 16-line lidar (100-meter detection radius, 0.2-degree angular resolution), an inertial measurement unit (200Hz sampling rate), a BeiDou / GPS dual-mode positioning module (1-meter positioning accuracy), and an ultrasonic sensor (5-meter detection range) mounted on the quadruped robot dog. Decision-making module: Running on the embedded computing platform (NVIDIA Jetson OrinNano, 8GB memory) of the quadruped robot dog, implemented using the ROS2 (Robot Operating System 2) framework. The decision-making module includes a task decomposition subunit, a path planning subunit, a safety verification subunit, and a mode switching subunit. Execution module: Converts the control commands output by the decision module into angle and torque commands for the 12 joint motors of the quadruped robot dog, and sends them to the joint controllers via the CAN bus. Communication module: It adopts a 5G communication module (downlink speed up to 1Gbps, uplink speed up to 200Mbps), while retaining a Wi-Fi module as a backup communication link. Status Management Module: Records task execution status at a frequency of 1Hz, including currently executing subtask, task completion progress (percentage), device location (latitude and longitude coordinates), and device status (battery percentage, attitude angle, speed, fault code). like Figure 3 As shown, the workflow of this embodiment is as follows: Step S1: Deploy the edge agent on the quadruped robot dog, initialize each module, and load the device capability description file corresponding to the quadruped robot dog. This device capability description file defines the kinematic model of the quadruped robot dog (including gait parameters, maximum moving speed of 1.5m / s, and maximum climbing angle of 30 degrees), sensor configuration (parameters of the above sensors), and executable task types (inspection, monitoring, and handling). Step S2: The edge agent receives the inspection task instruction from the cloud platform through the communication module. The task instruction includes: the task type is inspection, the target location is the geographic coordinate sequence of 10 inspection points in Area A of the industrial park, and the task constraints are the operation time window (8:00-10:00), the operation area boundary (the fence range of Area A of the park), and safety restriction parameters (maintain a distance of more than 3 meters from people, and prohibit entry into the power distribution room area). Step S3: The edge-side intelligent agent acquires real-time perception data through the perception module. Environmental perception data includes: visual image data collected by the binocular camera, used to identify inspection targets (such as abnormal equipment appearance, leaks, etc.); point cloud data collected by the lidar, used to construct a 3D map of the surrounding environment and detect obstacles; acceleration and angular velocity data collected by the inertial measurement unit, used for attitude estimation and gait control; satellite positioning data collected by the positioning module, used for global position positioning; and near-range ranging data collected by the ultrasonic sensor, used for obstacle avoidance. Equipment status data includes: battery percentage (e.g., 85%), attitude information (pitch angle 5 degrees, roll angle 2 degrees, yaw angle 180 degrees), speed (linear velocity 0.8 m / s), load status (camera working normally), and fault codes (no fault). Step S4: The edge agent makes local decisions through the decision module. The decision module first decomposes the inspection task into a sequence of sub-tasks: path planning sub-task → navigation to checkpoint 1 sub-task → data collection sub-task → navigation to checkpoint 2 sub-task → data collection sub-task → ... → return to the starting point sub-task. Then, the path planning sub-unit generates an execution path for each navigation sub-task based on the local and global maps constructed by the LiDAR and using the A* algorithm. The safety verification sub-unit performs a safety verification on the path. If it detects temporarily stacked goods on the path from checkpoint 3 to checkpoint 4, it triggers the path planning sub-unit to replan the detour route. The mode switching subunit in the decision-making module determines the mode based on communication quality indicators fed back by the communication module. In this embodiment, the 5G signal strength is -75dBm, the packet loss rate is 2%, the end-to-end latency is 30ms, and the communication quality is good, so the decision-making module operates in cloud-guided mode. When the quadruped robot dog moves to the underground utility tunnel area, the 5G signal strength drops to -105dBm, the packet loss rate rises to 15%, and the mode switching subunit switches the decision-making mode to edge autonomous mode. After completing the inspection in the underground utility tunnel area and returning to the ground area, the communication quality recovers, and the mode switching subunit switches the decision-making mode to cloud-edge collaborative mode. Step S5: The edge-side intelligent agent controls the quadruped robot dog to perform tasks according to the control instructions through the execution module. The execution module converts the navigation control instructions into angle and torque instructions for 12 joint motors, controlling the quadruped robot dog to walk to each inspection point according to the planned path, and performing data collection actions (such as taking pictures with the camera and measuring infrared temperature) at each inspection point. Step S6: The edge agent records the task execution status through the state management module and reports it to the cloud platform via the communication module at a frequency of 1Hz. During the communication interruption in the underground utility tunnel area, the state management module continues to record the task execution status and stores it in local memory. After returning to the ground and communication is restored, the operation records stored during the network outage are synchronized to the cloud platform. The safety handling unit continues to operate during the inspection process. For example... Figure 5 As shown, when the quadruped robot dog detects a person passing by 2 meters ahead on its inspection path, the safety processing unit determines it to be an emergency safety event. Without going through the cloud platform, it directly triggers emergency braking and lies down in place, resuming inspection operations only after the person has left. The end-to-end latency of this safety response is 8 milliseconds, far lower than the 200-500 millisecond latency of processing through the cloud platform. Example 2: UAV security monitoring operations driven by edge-side intelligent agents This embodiment uses drones for security monitoring in a large park as an example. The edge-side intelligent agent is deployed on an embedded computing platform carried by the UAV (NVIDIA Jetson Orin NX, 16GB RAM). The perception module includes a 4K high-definition camera, millimeter-wave radar, and a BeiDou / GPS positioning module. The path planning subunit in the decision-making module uses the RRT* algorithm to generate a 3D flight path. The execution module translates control commands into rotor speed and attitude control signals for the UAV. The mission instructions include: the mission type is security monitoring, the target location is the waypoint sequence of the park perimeter patrol route, and the mission constraints are a flight altitude of 50-120 meters and a flight speed not exceeding 15 m / s. During the operation, the mode switching subunit detected severe 5G signal obstruction in the eastern area of ​​the park and automatically switched to edge autonomous mode. The drone then autonomously completed its patrol and continued monitoring mission in that area. The safety processing unit detected that the drone was less than 5 meters away from a building ahead and triggered an emergency obstacle avoidance maneuver. The drone automatically climbed to a safe altitude and continued its patrol. Example 3: Cleaning operation of an unmanned sweeper driven by an edge-side intelligent agent This embodiment uses an unmanned sweeper vehicle performing cleaning operations on urban roads as an example. The edge-side intelligent agent is deployed on an embedded computing platform mounted on the unmanned sweeper. The perception module includes a 360-degree LiDAR, multiple ultrasonic sensors, and cameras. The path planning subunit in the decision-making module generates a comprehensive cleaning path. The execution module translates control commands into vehicle speed, steering angle, and operating parameters of the sweeping mechanism (brush speed, suction power). The task instructions include: the task type is cleaning operation, the target location is the cleaning area of ​​the specified road section, and the task constraints are the operation time window (4:00-6:00 am) and the cleaning coverage rate is not less than 95%. During operation, the safety verification subunit detects illegally parked vehicles on the cleaning path, triggering the route planning subunit to replan the detour route and mark the area as a re-cleaning area. After completing the overall cleaning, the vehicle returns to the re-cleaning area. The mode switching subunit detects weak communication signals in a certain road section and switches to edge autonomous mode, allowing the unmanned sweeper to autonomously complete the cleaning operation in that section even without network connectivity. The above embodiments demonstrate that the edge-side intelligent agent of this application can be adapted to various types of unmanned equipment such as quadruped robot dogs, drones, and unmanned cleaning vehicles. It can achieve local autonomous perception, decision-making, and execution in different operating scenarios, and supports autonomous switching and safety protection when communication is abnormal, which significantly improves the real-time performance, robustness, and safety of unmanned equipment operations.

Claims

1. A method for autonomous operation of unmanned equipment driven by an edge-side intelligent agent, characterized in that, The method includes the following steps: S1: Deploy an edge-side intelligent agent on unmanned equipment, the edge-side intelligent agent including a perception module, a decision-making module, an execution module, a communication module and a state management module; S2: The edge-side intelligent agent receives task instructions from the cloud platform through the communication module. The task instructions include task type, target location, and task constraints. S3: The edge-side intelligent agent obtains real-time perception data of the unmanned equipment through the perception module. The real-time perception data includes environmental perception data and equipment status data. S4: The edge-side intelligent agent makes local decisions based on the task instructions and the real-time sensing data through the decision module, and generates control instructions; S5: The edge-side intelligent agent controls the unmanned equipment to perform operational tasks according to the control instructions through the execution module; S6: The edge-side intelligent agent records the task execution status through the status management module and reports the task execution status to the cloud platform through the communication module; When the communication link between the edge-side intelligent agent and the cloud platform is abnormal, the edge-side intelligent agent switches to edge autonomous mode. In edge autonomous mode, the decision module independently completes local decision-making and controls the unmanned equipment to continue to perform the operation.

2. The method according to claim 1, characterized in that, In step S3, the environmental perception data includes at least one of visual image data, lidar point cloud data, inertial measurement unit data, satellite positioning data, and ultrasonic ranging data; the equipment status data includes at least one of power information, attitude information, speed information, load status, and fault codes.

3. The method according to claim 1, characterized in that, In step S4, the local decision includes: The task instructions are decomposed into a sequence of subtasks; Based on the real-time sensing data, an execution path and action sequence are generated for each subtask; The execution path is subjected to security verification. If obstacles or risky areas are detected on the path, the execution path is replanned.

4. The method according to claim 3, characterized in that, The method further includes: The decision-making module adaptively switches between cloud-guided mode, edge autonomous mode, and cloud-edge collaborative mode based on the communication quality indicators fed back by the communication module. In the cloud-guided mode, the decision-making module uploads the perceived data to the cloud platform and receives the decision results issued by the cloud platform. In the edge autonomy mode, the decision-making module makes decisions independently; In the cloud-edge collaboration mode, the decision-making module completes some decision-making tasks locally and uploads the remaining decision-making tasks to the cloud platform for collaborative completion.

5. The method according to claim 1, characterized in that, The method also includes a security processing step: The sensing module monitors the environmental safety status around the unmanned equipment in real time during operation. When an emergency safety event is detected, the edge-side intelligent agent triggers a safety protection action directly through the execution module without going through the cloud platform. The safety protection action includes at least one of emergency braking, hovering, return to home, and obstacle avoidance.

6. The method according to claim 1, characterized in that, The edge-side intelligent agent supports multiple types of unmanned equipment, including quadrupedal robot dogs, drones, and unmanned cleaning vehicles. The edge-side intelligent agent loads the corresponding device capability description file according to the type of the unmanned equipment. The device capability description file defines the kinematic model, sensor configuration, and executable task types of the unmanned equipment.

7. The method according to claim 1, characterized in that, In step S2, the task instruction also includes a task priority and a task timeout; the status management module queues multiple tasks to be executed according to the task priority, and when the task execution time is detected to exceed the task timeout, the status management module triggers the task timeout processing procedure.

8. The method according to claim 1, characterized in that, The method further includes: During the execution of the task, the edge-side intelligent agent collects scene data of the work area through the perception module; The scene data is associated with the task execution status and stored to form a job record; Once the communication link is restored, the job records will be synchronized to the cloud platform.

9. An autonomous operation system for unmanned equipment driven by an edge-side intelligent agent, characterized in that, The system includes: The edge-side intelligent agent module, deployed on unmanned equipment, includes: The sensing unit is configured to acquire real-time sensing data of the unmanned equipment, the real-time sensing data including environmental sensing data and equipment status data; The decision-making unit is configured to make local decisions based on task instructions and the real-time sensing data, and generate control instructions. The execution unit is configured to control the unmanned equipment to perform operational tasks according to the control instructions; The communication unit is configured to receive task instructions from the cloud platform and report the task execution status to the cloud platform. The status management unit is configured to record task execution status and manage the task queue; When the communication link between the communication unit and the cloud platform is abnormal, the decision-making unit switches to edge autonomy mode, independently makes local decisions, and controls the unmanned equipment to continue to perform its tasks.

10. The system according to claim 9, characterized in that, The sensing unit includes at least one of a visual sensor, a lidar, an inertial measurement unit, a satellite positioning module, and an ultrasonic sensor.

11. The system according to claim 9, characterized in that, The decision-making unit includes: The task decomposition subunit is configured to decompose the task instruction into a sequence of subtasks. The path planning subunit is configured to generate an execution path and action sequence for each subtask based on the real-time sensing data. The security verification subunit is configured to perform security verification on the execution path. When an obstacle or risk area is detected on the path, the path planning subunit is triggered to replan the execution path.

12. The system according to claim 9, characterized in that, The decision-making unit also includes a mode switching subunit, configured to adaptively switch between cloud-guided mode, edge autonomous mode and cloud-edge collaborative mode based on the communication quality indicators fed back by the communication unit.

13. The system according to claim 9, characterized in that, The system also includes a safety processing unit configured to monitor the environmental safety status around the unmanned equipment in real time during operation, and to trigger safety protection actions directly without going through the cloud platform when an emergency safety event is detected.

14. The system according to claim 9, characterized in that, The edge-side intelligent agent module also includes a device adaptation unit, configured to load a corresponding device capability description file according to the type of the unmanned equipment. The device capability description file defines the kinematic model, sensor configuration, and executable task types of the unmanned equipment.

15. The system according to claim 9, characterized in that, The status management unit is also configured to queue multiple tasks to be executed according to task priority, and to trigger a task timeout processing flow when the task execution time is detected to exceed the task timeout time.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-8.