An autonomous explosive detection system based on edge full-stack cluster and a control method thereof

Abstract

The application discloses an autonomous explosive disposal detection system and control method based on an edge full-stack cluster, aiming to solve the computing power bottleneck of the explosive disposal robot in a weak network environment and the reliability problem of grabbing in a complex scene. The system is composed of a four-legged mobile chassis and an execution mechanical arm carrying an airborne edge computing node, and the computing node integrates video processing, Web business hub and comprehensive control layer. The method constructs a full life cycle tactical closed loop: precise searching is realized through spatial tactical calibration, target tracing and macroscopic approach; a dual-mode fault-tolerant architecture based on confidence is innovatively proposed, autonomous grabbing confidence is calculated in real time after tactical anchoring, visual servo grabbing is performed when the confidence is high, the process is automatically suspended and seamlessly switched to a low-delay master-slave synchronous remote control mode when the confidence is low, and finally the safe transfer of explosives is completed. The application greatly improves the operation survival rate and safety of special robots in a broken network and unstructured environment.

Description

Technical Field

[0001] This invention relates to the field of special robot control systems and embodied intelligence technology. Specifically, it relates to an autonomous bomb disposal reconnaissance system and control method based on edge full-stack clusters suitable for extreme unnetworked and unstructured environments. Background Technology

[0002] The detection and disposal of unexploded ordnance (UXO) and other high-risk explosive devices are highly dangerous special tasks. Currently, the mainstream bomb disposal methods still heavily rely on manual handling by personnel wearing heavy bomb suits at close range, or on tracked / wheeled bomb disposal robots remotely controlled by operators via traditional radio. However, manual bomb disposal is not only inefficient but also poses an absolute threat to the lives of personnel; traditional remote-controlled bomb disposal robots generally suffer from limited mobility, high image transmission latency (usually higher than 500ms), and a lack of near-field depth perception. Because operators lack accurate three-dimensional spatial intuition on the screen, they are prone to accidental touches due to screen lag or spatial misalignment when performing critical "micro-operations" such as stripping camouflage and precise grasping, which can lead to irreversible and catastrophic consequences.

[0003] In recent years, with the development of embodied intelligence and computer vision technologies, intelligent robots with autonomous target detection and servo grasping capabilities have gradually been introduced into this field. However, existing intelligent bomb disposal systems suffer from two significant technical bottlenecks: First, the systems generally adopt a "cloud computing power + local execution" architecture. In extreme combat environments such as wilderness ruins, underground tunnels, or areas with strong electromagnetic interference, once the communication link is interrupted or bandwidth is limited, the visual perception and path planning modules, which heavily rely on cloud computing power, will immediately fail, leading to system paralysis. Second, in real unstructured environments, unexploded ordnance often exhibits complex characteristics such as partial burial, severe corrosion, incomplete shapes, or obscuration by vegetation. Existing purely autonomous visual grasping algorithms show extremely poor robustness in feature extraction when facing such "long-tail scenarios," making it difficult to guarantee grasping confidence. Therefore, how to balance the autonomous operation efficiency of special robots in environments without network access and solve the pain point of high-reliability and fault-tolerant grasping in complex scenarios is a core technical challenge that urgently needs to be addressed in this field. Summary of the Invention

[0004] The purpose of this invention is to provide an autonomous bomb disposal reconnaissance system and control method based on an edge full-stack cluster, which aims to solve the computing power bottleneck of existing bomb disposal robots in weak network environments and construct a full life cycle tactical closed loop covering target tracing, autonomous reconnaissance, dual-modal grasping and secure destruction and transfer.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: an autonomous bomb disposal reconnaissance system based on an edge full-stack cluster, comprising an airborne edge computing node, front-line physical equipment, and a remote command center connected via wireless local area network communication; the front-line physical equipment includes a quadrupedal mobile chassis, an execution robotic arm mounted on the chassis, and an airborne vision acquisition module; the airborne edge computing node is mounted on the quadrupedal mobile chassis and serves as the edge full-stack control hub of the entire system, including: a video processing and streaming media routing hub, a chassis and execution arm integrated control layer, a Web service hub, and a motor drive and master-slave synchronization module.

[0006] Preferably, the system supports confidence-based dual-modal control, and the motor drive and master-slave synchronization module supports seamless hot switching between autonomous visual servo mode and manual master-slave synchronization fault-tolerant mode.

[0007] This invention also provides an autonomous bomb disposal reconnaissance and control method, comprising: S1 initialization and spatial tactical calibration, S2 target tracing and macroscopic autonomous approach, S3 edge autonomous homing, S4 dynamic ranging and autonomous tactical anchoring, S5 confidence-based autonomous operation and degraded takeover, and S6 safe transfer and autonomous return closed loop.

[0008] The beneficial effects of this invention are as follows: (1) Full stack computing power sinking and communication self-organizing network: Abandoning the pure cloud architecture, the sensing, control, streaming media and communication gateway are integrated into an airborne single node to achieve high availability of the system in a network-free environment. (2) Closed loop of tactical operation throughout the entire life cycle: The integrated command and control flow from historical video tracing, grid coordinate mapping, preset point scheduling to final transfer and destruction is realized through the WebUI interface. (3) Innovative dual-modal fault-tolerant architecture: A degradation takeover mechanism based on confidence assessment is constructed. In normal scenarios, visual servoing is used to efficiently capture data, while in complex scenarios, the process is suspended and switched to master-slave synchronous remote control with extremely low latency (less than 150ms) to ensure absolute safety in bomb disposal. Attached Figure Description

[0009] Figure 1 This is a panoramic architecture diagram of an autonomous bomb disposal reconnaissance system based on an edge full-stack cluster, provided for embodiments of the present invention.

[0010] Figure 2 A flowchart of the autonomous bomb disposal reconnaissance and control method provided in an embodiment of the present invention. Detailed Implementation

[0011] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the underlying motion control and network communication logic in this embodiment is also applicable to virtual verification and implementation deployment in a ROS2 physics simulation environment (such as Gazebo).

[0012] Combination Figure 1 This system uses airborne edge computing nodes to carry all core services. The Web service hub uses the FastAPI framework to build a disguised communication node, receiving remote control commands via the WebSocket protocol. The video processing and streaming media routing hub utilizes FFmpeg to call hardware-accelerated encoding, combined with the WebRTC protocol to achieve concurrent distribution of low-latency composite video streams.

[0013] This embodiment provides a WebUI tactical control console deployed on an operating terminal. The interface is divided into a video monitoring area (with a calibration grid overlaid), a status monitoring area, and a task operation area (including command entry points such as navigation, search, capture, transfer, and release).

[0014] The control method described in this invention executes the following SOP process closed loop:

[0015] Step S1 (Initialization and Spatial Tactical Calibration): After system startup, the operating terminal acquires the image from the high-point camera. The system completes the homography matrix calibration of the mapping between two-dimensional video pixel coordinates and three-dimensional physical world coordinates by extracting the intersection features of the preset calibration grid in the monitoring image. At the same time, the operator sets the coordinates of the mission's standby point and detonation point through the WebUI.

[0016] Step S2 (Target Tracing and Macro-level Autonomous Approach): When a suspected UXO anomaly occurs, the operator manually locates the target's probabilistic position via WebUI video playback, clicks on the target coordinates in the gridded monitoring screen, and issues a "navigation" command. The FastAPI gateway receives the command and sends it to the chassis and actuator arm integrated control layer, driving the four-legged mobile chassis to autonomously move from the standby point to the target area.

[0017] Step S3 (Autonomous Edge Target Finding): Upon receiving the "search" command, the chassis performs a high-density spiral coverage search. The onboard YOLO target detection algorithm runs in real time, performing high-frequency frame scanning on concealed targets.

[0018] Step S4 (Dynamic Ranging and Autonomous Tactical Anchoring): When the detection algorithm locks onto the target for multiple consecutive frames, the edge computing nodes estimate the physical distance by combining the target bounding box size and camera intrinsic parameters. When the distance reaches the set optimal threshold (e.g., 30 cm), the tactical anchoring mechanism is triggered: the quadruped chassis automatically lowers its center of gravity and locks the motors to provide a stable working base.

[0019] Step S5 (Confidence-Based Autonomous Operation and Degraded Takeover): Upon receiving the "Fetch" command, the system calculates the overall confidence level of autonomous fetching in real time. Confidence level Detection box confidence and environmental shading rate Decide: when ( (For safety thresholds), the system maintains autonomous servo-based grasping; when At this point, the system triggers a fault-tolerance protection mechanism, suspending the current autonomous capture process and sending a takeover request to the operating terminal. After switching to manual master-slave synchronization mode, the operator generates motion data through the master robotic arm, and the underlying layer drives the robotic arm to perform 1:1 spatial synchronous remote control micro-operation based on the UDP / TCP direct connection protocol.

[0020] Step S6 (Safe Transfer and Autonomous Return Closed Loop): After the grab is completed and the arm position is locked, the operator clicks the "Transfer" command, and the quadruped mobile chassis carrying the explosive navigates to the detonation point preset in Step S1; upon arrival, the "Release" command is issued to complete the destruction and deployment, and then the chassis autonomously navigates back to the standby point, realizing the closed loop of the entire mission.

Claims

1. An autonomous explosive detection system based on edge full stack cluster, characterized by, It includes a remote command center, airborne edge computing nodes, and frontline physical equipment connected via wireless local area network communication; the frontline physical equipment includes a quadruped mobile chassis, an execution robotic arm mounted on the quadruped mobile chassis, and an airborne vision acquisition module; The airborne edge computing node, mounted on the quadrupedal mobile chassis, serves as the edge full-stack control hub for the entire system. It includes: a video processing and streaming media routing hub, used to receive image data from the airborne visual acquisition module, execute edge target detection algorithms, and perform hardware encoding and streaming of the video stream; a web service hub, used to receive control commands from the remote command center and convert these commands into underlying node communication protocol data; a chassis and actuator arm integrated control layer, used to receive the underlying node communication protocol data, control the motion posture of the quadrupedal mobile chassis, and incorporate a visual servo grasping algorithm; a motor drive and master-slave synchronization module, used to control the movement of the underlying motors of the quadrupedal mobile chassis and the actuator arm, and to support seamless switching between automatic servo mode and manual master-slave synchronization mode; and a remote command center, including a master control actuator arm and an operating terminal. The operating terminal communicates with the airborne edge computing node via a network, and the master control actuator arm is directly connected to the motor drive and master-slave synchronization module via a protocol to achieve synchronous remote operation control of the actuator arm.

2. The edge full stack cluster based autonomous explosive detection system as claimed in claim 1, wherein, The video processing and streaming media routing hub integrates a hardware-accelerated encoder and a WebRTC streaming media distribution service to ensure extremely low latency transmission of end-to-end first-person perspective images when fault-tolerant mode is triggered.

3. The edge full stack cluster based autonomous explosive detection system as claimed in claim 1, wherein, The Web service hub includes an edge communication backend built on FastAPI, which processes WebSocket commands from the operating terminal and bridges them to the integrated control layer of the chassis and execution arm.

4. A method of autonomous EOD investigation control based on the system of any of claims 1-3, characterized in that, The system includes the following steps: S1. Initialization and Spatial Tactical Calibration: After system startup, the system extracts the intersection features of the preset calibration grid in the monitoring screen to complete the homography matrix calibration of the mapping between two-dimensional video pixel coordinates and three-dimensional physical world coordinates. The operator presets the standby point and detonation point coordinates of the task through the operation terminal; S2. Target Tracing and Macro-Autonomous Approach: In case of an anomaly, the operator locks the probability position of the target through video playback on the operation terminal and issues navigation commands. The chassis and actuator arm integrated control layer drives the quadrupedal mobile chassis to start from the standby point and autonomously move to the target area; S3. Edge Autonomous Locating: After reaching the target area, the chassis and actuator arm integrated control layer drives the quadrupedal mobile chassis to execute a high-density spiral coverage search route, while the airborne target detection algorithm scans for concealed targets. S4. Dynamic Ranging and Autonomous Tactical Anchoring: When the detection algorithm locks onto the target for multiple consecutive frames, the airborne edge computing node estimates the physical distance by combining the target bounding box and camera intrinsic parameters. When the distance reaches the set optimal threshold, the tactical anchoring mechanism is triggered, controlling the quadrupedal mobile chassis to lower its center of gravity and lock the chassis motor. S5. Confidence-Based Autonomous Operation and Degradation Takeover: After tactical anchoring, the system calculates the overall confidence level of autonomous grasping in real time. If the confidence level is greater than or equal to a preset safety threshold, the system maintains autonomous mode, and the execution robotic arm completes the target grasping. If the confidence level is less than the preset safety threshold, the system triggers a fault-tolerant protection mechanism to suspend the autonomous grasping process, sends a takeover request to the remote command center, enters master-slave synchronization mode, and the operator controls the master robotic arm to perform spatial synchronous remote control operation on the execution robotic arm. S6. Safe transfer and autonomous return closed loop: After the grab is completed, the quadrupedal mobile chassis carries the explosive to the detonation point preset in step S1, releases and destroys it, and then autonomously navigates back to the standby point.

5. The control method according to claim 4, characterized by, In step S5, the calculation of the overall confidence level of autonomous grasping includes the confidence level of the target detection box and the environmental occlusion rate; wherein, the confidence level of the target detection box is determined by the overlap degree of the bounding boxes output by the edge target detection algorithm and the category probability, and the environmental occlusion rate is calculated by extracting the depth information obtained by the airborne vision acquisition module.

Images