Quadruped robot substation autonomous and teleoperation dual-mode inspection system and method

Abstract

The application discloses a quadruped robot substation autonomous and teleoperation dual-mode inspection system and method, and belongs to the technical field of power system intelligent operation and maintenance and robot control. In the autonomous operation mode, the device semantics, voltage level and power safety insulation distance are fused to construct a rule semantic safety map, and from geometric obstacle avoidance to deep understanding of electrical regulations is realized. A double-layer local path planning strategy is adopted, the absolute safety insulation distance is taken as a constraint, and a safe reference path conforming to the power safety regulations is searched out. Then, the lower layer is solved under the constraint of the upper layer reference path boundary constraint and other constraints. In the remote teleoperation mode, global rule safety evaluation is carried out on the continuous virtual escape trajectory, and then the continuous virtual escape trajectory is issued and executed. The problems that the quadruped robot fails to inspect in the strong electromagnetic interference and the structure height similar scene, lacks semantic understanding of electrical safety regulations, and lacks autonomous and teleoperation smooth switching when it is trapped in the dead zone and cannot make autonomous decisions are solved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance and robot control technology of power systems, and particularly relates to a dual-mode inspection system and method for quadruped robot substations with autonomous and remote operation. Background Technology

[0002] In the daily operation and maintenance of substations, staff need to conduct routine inspections of the operating status of various equipment within the station to promptly identify and address potential hazards. Daily tasks mainly include close-range observation of the operating status of high-voltage electrical equipment, accurate recording of various instrument data, detection of abnormal equipment overheating, and safe passage within designated high-voltage danger zones. These practical requirements place stringent demands on the path planning safety, inspection coverage stability, and continuity of inspection operations for inspection robots.

[0003] In the current engineering field, quadruped robots are increasingly being used to replace manual labor or wheeled robots for inspection work. Substations are subject to strong electromagnetic interference, making conventional GPS positioning signals susceptible to interference, with some areas experiencing complete signal loss. Substations also contain switchgear arrays, cable corridors, and isolating switch array areas, whose structural layouts are highly similar. When traditional vision or laser SLAM algorithms are used, if the environment lacks clearly distinguishable features, point cloud mismatches and path confusion can occur, preventing the output of stable and accurate positioning results over extended periods. Substation sites often have temporary storage of maintenance materials or temporary obstacles. These dynamic uncertainties can cause robots to enter planning dead zones in narrow passages or complex obstacle scenarios, preventing autonomous inspection operations. Once trapped in a planning dead zone, most existing systems can only stop on the spot and issue a warning, lacking an effective path to smoothly switch from autonomous inspection to remote takeover. This leads to frequent interruptions in inspection tasks, making it impossible to simultaneously meet the requirements for operational continuity and overall inspection efficiency.

[0004] In terms of navigation control and path planning, existing path planning methods mostly rely on pure geometric distance for obstacle avoidance, lacking semantic understanding of substation electrical safety regulations and the ability to actively identify safety boundaries. This can easily lead to robots accidentally touching high-voltage hazardous areas in complex passageways. Meanwhile, when autonomous planning fails and remote operation is required, existing solutions often use two-dimensional image transmission and joysticks, offering limited support for three-dimensional spatial perception. This operation method is prone to collisions in dead zones or blind spots due to the operator's visual blind spots, making it difficult to ensure the robot's safe escape from complex environments. These problems collectively limit the ability of quadruped robots to achieve safe, continuous, and intelligent inspection goals in complex substation scenarios characterized by strong electromagnetic interference, highly similar scene structures, and high-voltage hazardous obstacles. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a dual-mode inspection system and method for quadruped robots in substations, combining autonomous and remote operation. This addresses issues such as inspection failures of existing quadruped robots in scenarios with strong electromagnetic interference and highly similar structures, a lack of semantic understanding of electrical safety regulations in path planning, and a lack of smooth switching between autonomous and remote operation when the robot is trapped in dead zones or other situations where it cannot make autonomous decisions.

[0006] The technical solution of this invention is as follows: On the one hand, the present invention provides a quadruped robot substation autonomous and remote operation dual-mode inspection system, including an on-site inspection execution terminal and a remote operation terminal.

[0007] The remote control terminal is deployed in the substation monitoring room. In autonomous operation mode, it is used to visually monitor the operating status of the on-site inspection execution terminal based on multi-source sensing data transmitted back from the on-site inspection execution terminal. In remote control mode, the operator generates a continuous virtual escape trajectory and performs a global procedure safety assessment on the continuous virtual escape trajectory. After the assessment is passed, the continuous virtual escape trajectory is sent to the on-site inspection execution terminal for execution. The multi-source sensing data includes 3D point cloud data, color image frames, depth image frames, and inertial measurement data. The global procedure safety assessment is based on collision detection and absolute safety distance constraints of power regulations. It performs time-by-time spatial geometric safety verification on the continuous virtual escape trajectory in each discrete state and traverses the entire continuous virtual escape trajectory to generate the assessment result.

[0008] The on-site inspection execution terminal is set up at the substation inspection site, acquires multi-source sensing data in real time and transmits it back to the remote operation terminal. In autonomous operation mode, it autonomously completes the safety path planning and autonomous inspection tasks within the substation inspection site; in remote operation mode, it responds to the virtual escape trajectory issued by the remote operation terminal and executes escape actions.

[0009] Furthermore, the on-site inspection execution terminal includes a quadruped robot, a main industrial control computer, a 3D LiDAR, a depth camera, and network communication equipment; the 3D LiDAR has a built-in inertial measurement unit.

[0010] The quadruped robot is used to receive motion control commands from the master industrial control computer to move at the substation inspection site and transmit its own status information back to the master industrial control computer; the status information includes pose and velocity.

[0011] The three-dimensional lidar is mounted on the quadruped robot using a first mounting connector formed by 3D printing. It is used to acquire three-dimensional point cloud data and inertial measurement data of the substation inspection site in real time and transmit them to the main industrial control computer.

[0012] The depth camera is fixed to the front end of the quadruped robot using a second mounting connector formed by 3D printing. It is used to acquire RGB-D video data streams at the substation inspection site and transmit them to the main industrial control computer. The RGB-D video data stream includes color image frames and depth image frames at different times.

[0013] In autonomous operation mode, the master industrial control computer receives 3D point cloud data, inertial measurement data, and RGB-D video data streams, and sends the color image frames in the processed RGB-D video data streams to the network communication device. At the same time, it performs device semantic recognition based on the color image frames and depth image frames acquired by the depth camera, and fuses the recognized device semantics with the 3D point cloud data acquired by the LiDAR to construct a global procedure semantic safety map. Based on the global procedure semantic safety map, it performs path planning, generates executable quadruped robot motion control commands, and sends them to the quadruped robot.

[0014] When switching from autonomous operation mode to remote control mode, the master industrial control computer sends a takeover request message to the network communication device and simultaneously uploads the quadruped robot's own status information and global procedure semantic safety map.

[0015] In remote control mode, the master industrial control computer receives color frames acquired by the depth camera and three-dimensional point cloud data acquired by the lidar, and sends them to the network communication device; it also receives motion control commands sent by the network communication device and forwards them to the quadruped robot.

[0016] The network communication device is used to transmit the color frame image output by the master industrial control computer and the three-dimensional point cloud data acquired by the lidar to the remote teleoperation terminal; to forward the quadruped robot's own status information and global procedure semantic safety map to the remote teleoperation terminal; and to receive motion control commands sent by the remote teleoperation terminal and forward them to the master industrial control computer.

[0017] Furthermore, the remote control terminal includes an independent display screen and a remote industrial control computer.

[0018] In both autonomous operation mode and remote control mode, the remote industrial control computer receives color image frames sent by the on-site inspection execution terminal and sends them to an independent display screen.

[0019] In remote operation mode, the remote industrial control computer receives takeover request messages sent by the on-site inspection execution terminal, and simultaneously receives the global procedure semantic safety map and 3D point cloud data; based on the global procedure semantic safety map and 3D point cloud data, it constructs a virtual substation inspection site and a virtual quadruped robot; it sends the virtual substation inspection site including the virtual quadruped robot to an independent display screen; it responds in real time to the planar motion commands sent by the independent display screen, driving the quadruped virtual robot to perform prior simulation motion in the virtual substation inspection site in the visual interactive interface, and simultaneously records the continuous virtual escape trajectory generated by this prior simulation motion; it performs a global procedure safety assessment on the continuous virtual escape trajectory generated by the operator, and obtains the assessment result; based on the assessment result, if the continuous virtual escape trajectory is safe along its entire length, it sends the continuous virtual escape trajectory as a motion control command to the network communication device; if a collision is detected or the absolute safety distance constraint of the power procedure is not met, it sends a replanning prompt to the independent display screen.

[0020] The independent display screen is used to visualize color image frames and a virtual substation inspection site containing a virtual quadruped robot; the operator can directly input planar motion commands to the remote industrial control computer through the visual interactive interface on the independent display screen; the operator is prompted to replan.

[0021] On the other hand, the present invention also provides a quadruped robot substation dual-mode inspection method with autonomous and remote operation modes, including a quadruped robot navigation inspection method in autonomous operation mode and a quadruped robot navigation inspection method in remote operation mode.

[0022] Furthermore, the quadruped robot navigation and inspection method in the autonomous operation mode includes the following steps: A1: Real-time acquisition of multi-source sensing data at the substation inspection site via 3D LiDAR and depth camera, and transmission of the acquired multi-source sensing data to the master industrial control computer. The master industrial control computer performs timestamp alignment processing on the multi-source sensing data to generate synchronized multi-source sensing data.

[0023] A2: The main industrial control computer uses the laser-inertial fusion odometer to estimate the state of the quadruped robot based on the 3D point cloud data and inertial measurement data from the synchronous multi-source sensing data. This gives the real-time pose matrix of the quadruped robot in the world coordinate system at the current moment. At the same time, the color image frames are compressed and encoded, and then transmitted back to the independent display screen of the remote control terminal in real time through network communication equipment to realize the visualization display of the substation inspection site.

[0024] The world coordinate system uses the starting position of the quadruped robot as the origin, with the Z-axis perpendicular to the horizontal plane and pointing upwards as the positive direction, the X-axis pointing directly in front of the quadruped robot as the positive direction, and the Y-axis pointing horizontally to the left as the positive direction.

[0025] A3: The main industrial control computer uses a pre-trained YOLO instance segmentation network to perform target detection and instance segmentation on various types of live equipment at the substation inspection site captured by color image frames, outputs recognition results and detection confidence scores, and removes invalid recognition results based on the detection confidence scores; the recognition results include live equipment category labels, bounding box coordinates, and pixel-level instance masks.

[0026] A4: Extracting multiple target pixels of a charged device in a two-dimensional image coordinate system from a pixel-level instance mask. Then, pixel-level instance masks are used to extract the depth image frames, thereby obtaining the spatial position vector of the charged device in the 3D camera coordinate system. .

[0027] A back-projection mathematical model is established to back-project the target pixels of the powered device in the two-dimensional image coordinate system, thereby obtaining the spatial position vector of the powered device in the three-dimensional camera coordinate system. : (5); The mathematical model for back projection is: (6); The three-dimensional camera coordinate system uses the optical center of the depth camera as its origin. The axis is positive when it is horizontal to the right. The axis is vertically downwards as positive. The axis is positive when it faces forward. In the spatial position vector axis, shaft and The components of the axis; the two-dimensional image coordinate system uses the upper left corner of the color image frame as the origin, The positive direction is to the right along the horizontal axis of the color image frame. The positive direction is the vertical direction downwards along the color image frame. For the target pixel of the charged device in shaft and The components of the axis; This is the inverse of the depth camera intrinsic parameter matrix. This is a transpose.

[0028] A5: Based on the spatial position vector of the charged equipment in the 3D camera coordinate system The real-time pose matrix and recognition results of the quadruped robot in the world coordinate system are used to construct a local procedure semantic safety map.

[0029] A6: The main industrial control computer uses the real-time pose matrix of the quadruped robot in the world coordinate system at each moment to perform incremental data fusion on the local procedure semantic safety map constructed at multiple moments, and processes the semantic consistency under multi-view observation, and finally constructs the global procedure semantic safety map in the world coordinate system.

[0030] A7: The main industrial control computer performs real-time assessment of the environmental structure of the current substation inspection site based on the color image frame. If the assessment is successful, it performs path planning based on the global procedure semantic safety map, obtains control commands, and sends them to the quadruped robot for execution.

[0031] Furthermore, A5 specifically includes the following steps: A5.1: Using the real-time pose matrix of the quadruped robot in the world coordinate system at the current moment, the acquired single-frame 3D point cloud data is processed. Unified conversion to the world coordinate system, and within the set time window The system registers and stitches together multiple frames of 3D point cloud data to construct a local geometric point cloud that can realistically reflect the substation inspection site.

[0032] The local geometric point cloud is represented as follows: (7); in, This represents a local geometric point cloud.

[0033] A5.2: Spatial position vectors of all charged devices in the 3D camera coordinate system Mapping to the world coordinate system yields the device point cloud; assigning semantic attribute labels to the device point cloud yields the device semantic point cloud. .

[0034] The semantic attribute tag is represented as , For equipment category, Voltage level; This is the minimum safe distance.

[0035] A5.3: Transform the requirements of power safety regulations into computable mathematical constraints to construct a local regulation semantic cost field.

[0036] Local procedure semantic cost field : (8); (9); (10); in, For the local procedure semantic cost field, Let be any free point that can move within the space. For the cost of geometric barriers, This indicates the spatial occupancy status of the free point, where 0 represents free passage and 1 represents occupancy by physical obstacles, represented by a local geometric point cloud. Discretization mapping yields the result; The cost of procedural risks; Indicates the exponential decay coefficient; Representing free point The minimum Euclidean distance to the outer contour of the electrical equipment in the device semantic point cloud; Indicates the maximum distance affected by the cost; Points in the device semantic point cloud; This serves as the base for penalties.

[0037] A5.4: The main industrial control computer spatially fuses the local geometric point cloud, the equipment semantic point cloud, and the local procedure semantic cost field to generate a local procedure semantic safety map.

[0038] For any free point in the three-dimensional space of this local procedure semantic safety map Each has an attribute tuple containing multidimensional state information. Its data structure is defined as follows: (11); in, For equipment category; Voltage level.

[0039] Furthermore, A7 specifically includes the following steps: A7.1: Calculate the current color image frame Reference image frames from historical inspections The visual structural similarity index SSIM between the historical inspection reference image frames; To analyze current color image frames in the historical process Color image frames taken from the same area.

[0040] (14); in, It is a visual structural similarity index. Color image frames and reference image frame The average pixel brightness; It is the stability constant; Color image frames and reference image frame The pixel variance; denoted as pixel covariance.

[0041] A7.2: Calculate the current 3D point cloud data Historical reference point cloud of the previous moment Scan matching error between The historical reference point cloud is the three-dimensional point cloud data acquired at the position of the quadruped robot at the previous moment.

[0042] (15); in, For the number of points in the point cloud, For the current 3D point cloud data Point cloud points in the middle, Historical reference point cloud for the previous moment Point cloud points in the middle, The relative transformation matrix for registration output.

[0043] A7.3: The visual structural similarity index (SSIM) is weighted and fused with the scanning matching error to calculate the overall environmental similarity. .

[0044] (16); in, It is a visual structural similarity index; The weights of visual and laser features are respectively assigned during the fusion process, and satisfy the following conditions: ; This represents the maximum upper limit of the scan matching error.

[0045] A7.4: Utilizing comprehensive environmental similarity Conduct an environmental assessment, if If the quadruped robot has entered a region with similar structural heights, then A7.5 is executed to activate the enhanced navigation strategy; otherwise, A7.6 is executed directly. This is the threshold for determining environmental similarity.

[0046] A7.5: Based on the overall environmental similarity Calculate the adaptive heading penalty weight .

[0047] (17); in, It is the positive gain proportionality constant; A7.6: Based on the global procedure semantic safety map, a two-layer local path planning strategy is used to realize path planning and obtain control commands.

[0048] A7.6 specifically includes the following steps: A7.6.1: Establish the first-level spatial path cost integral function And under the absolute safety distance constraint of the power regulations, by minimizing the first-level spatial path cost integral function. A safe reference path was found. .

[0049] First-layer spatial path cost integral function for: (18); The absolute safety distance constraint stipulated in the power regulations is: (19); in, For path points, the value is expressed as the arc length of the trajectory. Parametric equations; Let the arc length of the trajectory be . The total arc length of the reference path, , These are the path length weight and the risk cost weight, respectively. As a consequence of procedural risks, path point The standard safety distance value at the location is equal to the path point. To the corresponding device semantic point cloud The minimum Euclidean distance of the profile of the electrical equipment.

[0050] A7.6.2: Establishing a continuously optimized objective functional In the reference path Trajectory optimization is performed under the constraints of the kinematic differential hard constraints of the quadruped robot to obtain a safe and smooth trajectory; the adaptive heading penalty weight is introduced into the objective functional as a heading deviation constraint term.

[0051] The continuously optimized objective functional for: (20); The kinematic differential hard constraints of the quadruped robot: (twenty one); in, To represent the state equation of the actual motion trajectory of the quadruped robot. , For three-dimensional position, Yaw angle; Plan the time range for the trajectory; , These are the trajectory smoothing cost weighting coefficient and the safety risk cost weighting coefficient, respectively. For quadruped robots The actual linear velocity vector at time t. For quadruped robots The actual linear acceleration vector at time t, Let be the square of the acceleration term of the quadruped robot; This is the cost of the safety field risk at the current three-dimensional location; The physical limits of linear velocity and acceleration allowed by the chassis controller of the quadruped robot; Reference path exist The expected yaw angle at any given moment.

[0052] A7.6.3: To ensure that the trajectory optimization of the second layer does not violate the safety constraints of the first layer, after obtaining the safe and smooth trajectory, a safety check is forcibly performed on the generated safe and smooth trajectory. If the check passes, A7.6.4 is executed; otherwise, the circuit is immediately broken and the safe and smooth trajectory is determined to be invalid. At this time, the safe and smooth trajectory is automatically abandoned and replanned. If the replanning fails continuously, it is determined to be trapped in a dead zone, and the master industrial control computer ends the navigation and inspection of the quadruped robot in the autonomous operation mode.

[0053] The security verification specifically involves: extracting a secure smooth trajectory. The trajectory points on the curve are substituted back into the local procedure semantic cost field. To obtain the actual procedural risk costs, if trajectory optimization causes a segment of the safe, smooth trajectory to deviate excessively from the reference path. This ensures that any trajectory point intrudes into the minimum safe distance. If the absolute safety distance constraint of the power regulations is not met, the verification will fail; otherwise, the verification will pass.

[0054] A7.6.4: Only after the safe and smooth trajectory passes the verification can it be converted into the control commands of the quadruped robot through the inverse kinematics solution of the underlying layer, so as to complete the highly stable autonomous inspection in the complex environment of the substation. The quadruped robot will then send its own status information back to the main industrial control computer.

[0055] Furthermore, in autonomous operation mode, if the quadruped robot cannot make a decision during autonomous inspection in the following situations, the abnormal interruption trigger condition is determined to be met, and the master industrial control computer terminates the quadruped robot's navigation and inspection in autonomous operation mode, controlling the quadruped robot to stop running in place: (1) The quadruped robot cannot find a safe and smooth trajectory that meets the absolute safety distance constraint of the power regulations in the global procedure semantic safety map. The quadruped robot is trapped in the dead zone or cannot find a safe and smooth trajectory that meets the absolute safety distance constraint of the power regulations within the set evaluation time window of 20 seconds.

[0056] (2) Within the set monitoring period, the absolute value of the deviation between the expected motion displacement corresponding to the control command of the quadruped robot and the actual physical displacement exceeds the set maximum displacement deviation threshold or the absolute value of the tilt angle or pitch angle of the quadruped robot, and exceeds the set limit safety attitude angle to prevent the equipment from tipping over.

[0057] Furthermore, the operational steps of the quadruped robot navigation and inspection method in the remote teleoperation mode are as follows: B1: The master industrial control computer sends a takeover request to the remote industrial control computer and simultaneously uploads the real-time pose matrix of the actual quadruped robot, the global procedure semantic safety map, and the 3D point cloud data collected in real time by the 3D LiDAR.

[0058] B2: The remote industrial control computer receives the takeover request message, reconstructs the virtual substation inspection site containing the safety constraints of the procedure according to the global procedure semantic safety map, and establishes a virtual quadruped robot in the virtual substation inspection site, and maps the three-dimensional point cloud data to the virtual substation inspection site to achieve pose alignment between the virtual quadruped robot and the real quadruped robot.

[0059] B3: The remote industrial control computer sends the virtual substation inspection site, which includes the virtual quadruped robot, to the visual interactive interface on the independent display screen. The operator can directly input planar motion commands through the visual interactive interface on the independent display screen to drive the quadruped virtual robot to perform prior simulation motion in the virtual substation inspection site in the visual interactive interface until the virtual robot escapes the predicament. The remote industrial control computer simultaneously records the continuous virtual escape trajectory generated by the prior simulation motion.

[0060] B4: Conduct a global procedure safety assessment on the continuous virtual escape trajectory generated by the operator and obtain the assessment results.

[0061] B5: Based on the evaluation results, if the continuous virtual escape trajectory is safe along the entire route, the continuous virtual escape trajectory will be sent to the real quadruped robot for action execution. If a collision is detected or the absolute safety distance constraint of the power regulations is not met, the continuous virtual escape trajectory will be stopped and the operator will be prompted to replan the trajectory in the visual interactive interface.

[0062] B5.1: Based on the Boolean simulation safety exit verification identifier Generate the final escape strategy for a realistic quadruped robot; (29); in, The ultimate escape strategy for a real quadruped robot; This indicates that the continuous virtual escape trajectory will be sent to the real quadruped robot to perform the escape operation; This indicates that the actual quadruped robot will remain stationary and will issue a prompt to the operator.

[0063] B6: After the real quadruped robot successfully escapes its predicament, the main industrial control computer reacquires multi-source perception data. If it determines that the current environment has met the conditions for generating a safe reference path, it sends a prompt message indicating that the escape is complete to the remote control terminal and switches back to the autonomous operation mode.

[0064] Furthermore, B3 specifically includes the following steps: B3.1: The remote industrial control computer initializes the virtual substation inspection site, which includes a virtual quadruped robot, and displays it on a visual interactive interface on an independent display screen.

[0065] B3.2: The operator inputs a planar motion command in the interactive coordinate system through a visual interactive interface. The planar motion command is then forwarded to a remote industrial control computer, which converts the planar motion command in the interactive coordinate system into the expected speed command in the virtual machine coordinate system.

[0066] The interactive coordinate system has its origin at the center of the visual interactive interface. The positive direction is along the vertical axis pointing upwards towards the top of the visual interactive interface. The positive direction is the horizontal axis pointing towards the left side of the visual interactive interface. The direction of counterclockwise rotation of the axis along the visual interactive interface is the positive direction; the virtual machine coordinate system has its origin at the geometric center of the virtual quadruped robot chassis. The positive direction is defined as the horizontal direction directly forward along the axis of the virtual quadruped robot. The positive direction is the horizontal direction to the left of the virtual quadruped robot. The positive direction is the axis perpendicular to the virtual quadruped robot and pointing upwards.

[0067] Set the planar motion command in the interactive coordinate system as follows Then its corresponding expected speed command The expression is as follows: (twenty two); in, These are the set linear velocity and angular velocity mapping ratio coefficients, respectively.

[0068] B3.3: Convert the expected velocity command in the virtual machine coordinate system from a discrete control quantity to a virtual pose sequence of chassis kinematics to obtain a continuous virtual escape trajectory.

[0069] Specifically: Let the single-step deduction period be... The virtual quadruped robot in the first The two-dimensional plane position of the step is Yaw angle is Then the first Step Two-Dimensional Plane Position With yaw angle The update model is expressed as follows: (twenty three); (twenty four); Let the state vector The operator will attempt to escape the predicament by taking a total of steps. By concatenating all the internally generated state vectors sequentially, a continuous virtual escape trajectory is obtained. The expression is as follows: (25).

[0070] Furthermore, B4 specifically includes the following steps: B4.1: Utilizing the outer contour of the energized equipment in the semantic point cloud, perform 3D expansion according to the minimum safe distance corresponding to the energized equipment to generate a virtual insulation envelope as a high-voltage obstacle, thereby obtaining a set of high-voltage obstacles. .

[0071] B4.2: Calculate the shortest geometric distance between the surface of the virtual quadruped robot and the high-pressure obstacle on a continuous virtual escape trajectory.

[0072] Specifically, the virtual quadruped robot on the continuous virtual escape trajectory is designated as the collision detection entity. The system reads the URDF description file of the virtual quadruped robot, extracts its collision geometry model, and discretizes the surface of the collision geometry model to generate a discrete coordinate point. The constructed robot surface point cloud model is used as a collision detection entity. The expression; in the continuous virtual escape trajectory At each discrete moment, the point cloud model of the robot surface is spatially transformed according to the robot's real-time pose matrix, and the shortest geometric distance between the virtual quadruped robot surface and the high-pressure obstacle on the continuous virtual escape trajectory is calculated in real time: (26); in, State vector The shortest geometric distance between the surface of the virtual quadruped robot and the high-pressure obstacle on the next continuous virtual escape trajectory. For discrete points on the surface of the virtual quadruped robot body, For collision detection entities in the state vector The robot surface point cloud model below; These are discrete points on the virtual insulation envelope corresponding to the high-voltage obstacle; Representing discrete points respectively Coordinate components in the three-dimensional simulation space; Representing discrete points respectively Coordinate components in the three-dimensional simulation space.

[0073] By traversing all state vectors in a continuous virtual escape trajectory, the global shortest spatial distance of the entire continuous virtual escape trajectory is found. The expression is as follows: (27); B4.3: Convert the shortest geometric distance between the surface of the virtual quadruped robot and the high-pressure obstacle on the continuous virtual escape trajectory from numerical form to a Boolean simulation safety escape verification identifier as the evaluation result.

[0074] (28); in, This is a Boolean-type simulation security exit check identifier. To compensate for the safety margin.

[0075] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes a quadruped robot substation dual-mode inspection system and method combining autonomous and remote operation. Through the coordinated use of multi-source perception, semantic attributes, digital twins, and low-level motion control, the reliability and decision-making efficiency of the quadruped robot in complex substation environments are improved. In autonomous operation mode, the system does not simply rely on traditional three-dimensional spatial geometric information for physical obstacle avoidance. Instead, it innovatively integrates equipment semantic tags, voltage levels, and electrical safety insulation distances to construct a procedure semantic safety map, achieving a shift from geometric obstacle avoidance to a deep understanding of electrical procedures. In autonomous operation mode, the quadruped robot completes autonomous navigation based on the established procedure semantic safety map. This map, based on three-dimensional environmental geometric information, combines physical obstacle information within the substation with the necessary safety insulation distance constraints between electrical equipment. During path planning, in addition to spatial feasibility, compliance with electrical safety regulations is also required.

[0076] This invention employs a two-layer local path planning strategy with strong procedural constraints, effectively overcoming the shortcomings of traditional path planning strategies that rely solely on geometric distance obstacle avoidance and fail to understand electrical safety regulations. The upper layer is a safety guidance layer, which does not search within ordinary geometric space but solves within a procedural semantic safety map with electrical semantic attributes. It uses absolute safety insulation distance as a hard constraint to find a safe reference path that fully complies with electrical safety regulations. The lower layer is a local trajectory optimization layer, which does not simply smooth the path but performs optimal control under the combined influence of upper-layer reference path boundary constraints, real-time procedural risk costs, adaptive heading penalties in similar regions, and robot dynamics constraints. To ensure that the lower-layer trajectory optimization does not violate the spatial safety constraints established by the upper layer, this invention introduces a post-verification mechanism for trajectory procedural safety in the two-layer architecture. If the trajectory optimized by the lower layer deviates from the safe path, resulting in excessive risk costs, the system will immediately shut down the current trajectory and trigger replanning. This allows quadruped robots to safely complete inspection tasks in substation environments with dense high-voltage equipment and narrow passages. Substations often contain areas with highly similar structures, such as switchgear and disconnectors. This invention establishes a collaborative mechanism between environmental degradation perception and underlying motion control. The system assesses environmental similarity in real time through multimodal feature recognition and automatically activates an enhanced navigation mechanism when similar areas are detected. It quantifies the degree of environmental feature degradation as an adaptive heading penalty weight, which is directly input into the functional solution for underlying trajectory optimization. This collaborative fusion of perception and control enables the robot to be forcibly constrained in scenarios lacking salient features, suppressing trajectory deviations and significantly improving positioning stability and the continuity of inspection operations. In extreme cases, such as when the robot is blocked by a physical dead zone or experiences severe sideslip, rendering the autonomous operation mode inoperable, the system automatically switches to remote control mode. In this case, the real inspection environment information and the robot's pose are mapped to a 3D simulation platform in real time, allowing operators to take over control and perform auxiliary operations from the front-end interface. By leveraging a virtual environment constructed using a 3D simulation system, operators can pre-plan and demonstrate escape routes within scenarios constrained by high-voltage safety regulations. After completing a global safety check and confirming compliance with safety requirements, the trajectory is then transmitted to the real robot chassis for execution, effectively avoiding safety risks associated with communication delays and accidental equipment contact. Through this method, the system achieves continuous synchronization between the real inspection environment and the virtual scene, combining autonomous operation with remote control. This allows the robot to efficiently complete autonomous inspections under normal conditions, while in abnormal situations, operators can remotely intervene for auxiliary control, preventing task interruption. Overall, this improves the navigation stability, task continuity, and system safety of the quadruped robot in complex substation environments. Attached Figure Description

[0077] Figure 1 This is an overall structural diagram of the quadruped robot substation autonomous and remote operation dual-mode inspection system in an embodiment of the present invention.

[0078] Figure 2 This is an interactive interface diagram of the remote teleoperation mode in an embodiment of the present invention.

[0079] Figure 3 This is a schematic diagram of the workflow of the quadruped robot substation dual-mode inspection method in an embodiment of the present invention, which combines autonomous and remote operation.

[0080] Figure 4 This is a schematic diagram illustrating the principle of constructing a procedural semantic safety map and planning a two-layer local path in an embodiment of the present invention.

[0081] Figure 5 This is a schematic diagram illustrating the navigation principle of similar layout areas in an embodiment of the present invention.

[0082] Figure 6 This is a schematic diagram of the operation process of the remote teleoperation control method in an embodiment of the present invention.

[0083] Figure 7 This is an interactive scene diagram in a three-dimensional simulation system environment according to an embodiment of the present invention. Detailed Implementation

[0084] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0085] Example 1: This invention proposes a quadruped robot substation autonomous and remote operation dual-mode inspection system for inspecting substations in complex environments. This system can simultaneously map real-world work scenarios and 3D simulated virtual scenarios, integrating autonomous operation and remote operation control modes. Autonomous operation refers to the robot automatically completing path planning and inspection tasks within the substation environment under normal conditions. Remote operation refers to the operator remotely intervening to take over tasks when the robot faces complex dead zones or other obstacles. When the robot cannot autonomously complete the inspection task, it can smoothly switch from autonomous operation mode to remote operation mode, ensuring the continuous and safe conduct of the inspection task. Figure 1 As shown, the system includes an on-site inspection execution terminal and a remote operation terminal.

[0086] The remote control terminal is deployed in the substation monitoring room. In autonomous operation mode, it is used to visually monitor the operating status of the on-site inspection execution terminal based on multi-source sensing data transmitted back from the on-site inspection execution terminal. In remote control mode, the operator generates a continuous virtual escape trajectory and performs a global procedure safety assessment on the continuous virtual escape trajectory. After the assessment is passed, the continuous virtual escape trajectory is sent to the on-site inspection execution terminal for execution. The multi-source sensing data includes 3D point cloud data, color image frames, depth image frames, and inertial measurement data. The global procedure safety assessment is based on collision detection and absolute safety distance constraints of power regulations. It performs time-by-time spatial geometric safety verification on the continuous virtual escape trajectory in each discrete state and traverses the entire continuous virtual escape trajectory to generate the assessment result.

[0087] The on-site inspection execution terminal is set up at the substation inspection site, acquires multi-source sensing data in real time and transmits it back to the remote operation terminal. In autonomous operation mode, it autonomously completes the safety path planning and autonomous inspection tasks within the substation inspection site. In remote operation mode, it responds to the virtual escape trajectory issued by the remote operation terminal and executes safe escape actions for complex dead zones or obstacles.

[0088] The on-site inspection execution terminal includes a quadruped robot, a main industrial control computer, a 3D LiDAR, a depth camera, and network communication equipment; the 3D LiDAR has a built-in inertial measurement unit.

[0089] The quadruped robot, serving as the motion carrier for on-site inspection execution, receives motion control commands from the main industrial control computer (using the quadruped robot's chassis controller to convert the motion control commands into motion parameters of each joint, as well as linear and angular velocities), enabling movement at the substation inspection site and simultaneously transmitting its own status information back to the main industrial control computer; the status information includes pose and velocity.

[0090] The three-dimensional lidar is mounted on the quadruped robot using a first mounting connector formed by 3D printing. It is used to acquire high-precision three-dimensional point cloud data and inertial measurement data of the substation inspection site in real time and transmit them to the main industrial control computer.

[0091] The depth camera is fixed to the front end of the quadruped robot using a second mounting connector formed by 3D printing. It is used to acquire close-range RGB-D video data streams at the substation inspection site and transmit them to the main industrial control computer. The RGB-D video data stream includes color image frames and depth image frames at different times.

[0092] In this embodiment, the first and second mounting connectors are used to improve the stability and reliability of the 3D point cloud data and RGB-D video data stream.

[0093] In autonomous operation mode, the master industrial control computer receives 3D point cloud data, inertial measurement data, and RGB-D video data streams, and sends the color image frames in the processed RGB-D video data streams to the network communication device. At the same time, it performs device semantic recognition based on the color image frames and depth image frames acquired by the depth camera, and fuses the recognized device semantics with the 3D point cloud data acquired by the LiDAR to construct a global procedure semantic safety map. Based on the global procedure semantic safety map, it performs path planning, generates executable quadruped robot motion control commands, and sends them to the quadruped robot.

[0094] When switching from autonomous operation mode to remote control mode, the master industrial control computer sends a takeover request message to the network communication device and simultaneously uploads the quadruped robot's own status information and global procedure semantic safety map.

[0095] In remote control mode, the master industrial control computer receives color frames acquired by the depth camera and three-dimensional point cloud data acquired by the lidar, and sends them to the network communication device; it also receives motion control commands sent by the network communication device and forwards them to the quadruped robot.

[0096] The network communication device is used to realize high-frequency, low-latency data interaction between the on-site inspection execution terminal and the remote teleoperation terminal. It transmits the color frame image output by the main industrial control computer and the three-dimensional point cloud data acquired by the lidar to the remote teleoperation terminal to provide first-person perspective and local spatial information; it forwards the quadruped robot's own state information and global procedure semantic safety map to the remote teleoperation terminal; and it receives motion control commands sent by the remote teleoperation terminal and forwards them to the main industrial control computer.

[0097] In this embodiment, the data transmission rate of the network communication device is not less than 200Mbps, and the end-to-end communication latency is not more than 100ms.

[0098] In this embodiment, the main industrial control computer runs on an Ubuntu 22.04 operating system and deploys the ROS2 distributed communication framework. The multi-source sensing, map building, and path planning modules on the field execution end all run as independent C++ nodes within the ROS2 framework. These modules communicate asynchronously through a data distribution service to ensure low latency and reliability of the data. The depth camera uses an Intel RealSense D435i, which connects to the main industrial control computer via a USB interface, while the LiDAR is connected via an Ethernet interface.

[0099] The remote control terminal includes an independent display screen and a remote industrial control computer.

[0100] like Figure 2As shown, in both autonomous operation mode and remote control mode, the remote industrial control computer receives color image frames sent by the on-site inspection execution terminal and sends them to an independent display screen.

[0101] In remote operation mode, the remote industrial control computer receives takeover request messages from the on-site inspection execution terminal and simultaneously receives the global procedure semantic safety map and 3D point cloud data. Based on the global procedure semantic safety map and 3D point cloud data, it constructs a virtual substation inspection site and a virtual quadruped robot. The virtual substation inspection site, including the virtual quadruped robot, is sent to an independent display screen. It responds in real-time to planar motion commands sent by the independent display screen, driving the quadruped virtual robot to perform prior simulation motion in the virtual substation inspection site within the visual interactive interface, simultaneously recording the continuous virtual escape trajectory generated by this prior simulation motion. A global procedure safety assessment is performed on the continuous virtual escape trajectory generated by the operator, obtaining the assessment result. Based on the assessment result, if the continuous virtual escape trajectory is entirely safe, it is sent as a motion control command to the network communication device, realizing visual monitoring and virtual verification of the complex dead zone operation process on-site. If a collision is detected or the absolute safety distance constraint of the power regulations is not met, a replanning prompt is sent to the independent display screen. It avoids the blind spot risks inherent in traditional video remote control, and significantly improves the intuitiveness and operational efficiency of the remote takeover system while ensuring safety.

[0102] The independent display screen is used to visualize color image frames and a virtual substation inspection site including a virtual quadruped robot; operators can directly input planar motion commands to a remote industrial control computer through the visual interactive interface on the independent display screen; the operator is prompted to replan the operation. The quadruped robot substation autonomous and remote operation dual-mode inspection method includes a quadruped robot navigation inspection method in autonomous operation mode and a quadruped robot navigation inspection method in remote operation mode.

[0103] like Figure 3 As shown, the navigation and inspection method for a quadruped robot in autonomous operation mode includes the following steps: A1: Real-time acquisition of multi-source sensing data at the substation inspection site via 3D LiDAR and depth camera, and transmission of the acquired multi-source sensing data to the master industrial control computer. The master industrial control computer performs timestamp alignment processing on the multi-source sensing data to generate synchronized multi-source sensing data.

[0104] A2: The main industrial control computer uses the laser-inertial fusion odometer to estimate the state of the quadruped robot based on the 3D point cloud data and inertial measurement data from the synchronous multi-source sensing data. This gives the real-time pose matrix of the quadruped robot in the world coordinate system at the current moment. At the same time, the color image frames are compressed and encoded, and then transmitted back to the independent display screen of the remote control terminal in real time through network communication equipment to realize the visualization display of the substation inspection site.

[0105] The world coordinate system uses the starting position of the quadruped robot as the origin, with the Z-axis perpendicular to the horizontal plane and pointing upwards as the positive direction, the X-axis pointing directly in front of the quadruped robot as the positive direction, and the Y-axis pointing horizontally to the left as the positive direction.

[0106] This embodiment employs a fusion framework based on tightly coupled iterative extended Kalman filtering. This framework can directly and tightly couple the 3D point cloud data acquired by 3D LiDAR with the inertial measurement data to obtain the real-time pose matrix of the quadruped robot in the world coordinate system at the current moment. (1); (2); in, Let be the state vector of the quadruped robot. Indicates time, and Let represent the rotation matrix and translation vector of the quadruped robot relative to the world coordinate system, respectively. Let be the velocity vector of the quadruped robot in the world coordinate system. and These are the accelerometer bias and gyroscope bias of the inertial measurement unit, respectively; This represents the real-time pose matrix of the quadruped robot in the world coordinate system at the current moment.

[0107] In this embodiment, the color image frame is compressed and encoded into JPEG format.

[0108] A3: The main industrial control computer uses a pre-trained YOLO instance segmentation network to perform target detection and instance segmentation on various types of live equipment at the substation inspection site captured by color image frames, outputs recognition results and detection confidence scores, and removes invalid recognition results based on the detection confidence scores; the recognition results include live equipment category labels, bounding box coordinates, and pixel-level instance masks.

[0109] No. One identification result Represented as: (3); in, For electrical equipment category labels, For bounding box coordinates, For pixel-level instance masks, This refers to the number of the equipment being energized.

[0110] Then, based on the set confidence threshold In this embodiment, the confidence threshold is set to 0.75. Invalid identification results with a detection confidence level lower than the confidence threshold are discarded to obtain a valid set of live devices. for: (4); in, Represents the initial set of charged devices. To test the confidence level.

[0111] A4: Extracting multiple target pixels of a charged device in a two-dimensional image coordinate system from a pixel-level instance mask. Then, pixel-level instance masks are used to extract the depth image frames, thereby obtaining the spatial position vector of the charged device in the 3D camera coordinate system. .

[0112] Based on the principle of reverse perspective from a pinhole camera, a back-projection mathematical model is established to back-project the target pixels of the electrical device in the two-dimensional image coordinate system, thereby obtaining the spatial position vector of the electrical device in the three-dimensional camera coordinate system. : (5); The mathematical model for back projection is: (6); The three-dimensional camera coordinate system uses the optical center of the depth camera as its origin. The axis is positive when it is horizontal to the right. The axis is vertically downwards as positive. The axis is positive when it faces forward. In the spatial position vector axis, shaft and The components of the axis; the two-dimensional image coordinate system uses the upper left corner of the color image frame as the origin, The positive direction is to the right along the horizontal axis of the color image frame. The positive direction is the vertical direction downwards along the color image frame. For the target pixel of the charged device in shaft and The components of the axis; This is the inverse of the depth camera intrinsic parameter matrix. This is a transpose.

[0113] A5: Based on the spatial position vector of the charged device in the 3D camera coordinate system The real-time pose matrix and recognition results of the quadruped robot in the world coordinate system are used to construct a local procedure semantic safety map.

[0114] The terminology related to maps and cost fields involved in this invention is explained as follows: To achieve safe navigation under the constraints of power safety regulations, this invention adopts a map construction method that integrates geometric information, equipment semantic information, and regulatory risk constraints layer by layer. First, the system constructs a local geometric point cloud reflecting the site based on multi-source sensing data, and then obtains an equipment semantic point cloud including equipment category and voltage level labels. Next, the system transforms the power safety regulations into mathematical constraints and calculates the local regulatory semantic cost field. Subsequently, the local geometric point cloud, equipment semantic point cloud, and local regulatory semantic cost field are fused to generate a local regulatory semantic safety map. Finally, the master industrial control computer uses the real-time pose of the quadruped robot to perform incremental data fusion and semantic consistency processing on the local regulatory semantic safety maps constructed at multiple moments, thereby ultimately constructing a global regulatory semantic safety map.

[0115] like Figure 4 The diagram shown illustrates the principle of constructing a semantic safety map and planning a two-layer local path. The specific steps are as follows: A5.1: Utilizing the real-time pose matrix of the quadruped robot in the world coordinate system at the current moment. The collected single-frame 3D point cloud data Unified conversion to the world coordinate system, and within the set time window The system registers and stitches together multiple frames of 3D point cloud data to construct a local geometric point cloud that can realistically reflect the substation inspection site.

[0116] The local geometric point cloud is represented as follows: (7); in, This represents a local geometric point cloud.

[0117] A5.2: Spatial position vectors of all charged devices in the 3D camera coordinate system Mapping to the world coordinate system yields the device point cloud; assigning semantic attribute labels to the device point cloud yields the device semantic point cloud. .

[0118] To describe the spatial transformation relationship between the 3D camera, the quadruped robot body, and the world coordinate system, a quadruped robot body coordinate system is defined. This quadruped robot body coordinate system uses the geometric center of the quadruped robot's chassis as its origin. The positive direction is the horizontal direction directly forward along the axis of the quadruped robot. The positive direction is the horizontal direction to the left of the quadruped robot. The positive direction is the axis perpendicular to the quadruped robot and pointing upwards.

[0119] The transformation relationship between coordinates can be expressed as: (12); (13); in, Let be the position vector of the energized equipment in the world coordinate system. Let be the position vector of the electrical equipment in the coordinate system of the quadruped robot; and and represent the calibration rotation matrix and translation vector of the depth camera relative to the quadruped robot body, respectively.

[0120] The semantic attribute tag is represented as , For equipment category, Voltage level, unit is kV; The minimum safe distance is measured in meters.

[0121] In this embodiment, a substation equipment ledger and a preset safety distance table are pre-stored in the main industrial control computer. The equipment ledger includes equipment number, equipment category, voltage level, and corresponding spatial installation area. The preset safety distance table is a structured configuration file, and its values ​​are based on the safety distance requirements for equipment without power interruption in the national standard GB26860 "Electrical Safety Operation Procedures for Power Plants and Substations". It includes a mapping relationship field between voltage level and minimum safety distance, specifically set to 0.7m for 10kV, 1.0m for 35kV, and 1.5m for 110kV, etc. This table can be generated by power safety operation procedures or on-site management rules and supports dynamic adjustment and rapid adaptation according to procedure updates or changes in application scenarios. After identifying the equipment category, the spatial location of the equipment point cloud is matched with the spatial installation area in the equipment ledger; if the match is successful, the voltage level of the corresponding equipment in the ledger is directly read. If a match fails, the highest risk voltage level corresponding to that equipment category is used as a conservative safety constraint to ensure safety. Ultimately, the master industrial control computer determines the voltage level... The preset safety distance table is used as an index to retrieve the corresponding value, which is then used as the minimum safety distance for the device in the procedure semantic safety map. .

[0122] A5.3: Transform the requirements of power safety regulations into computable mathematical constraints to construct a local regulation semantic cost field.

[0123] Local procedure semantic cost field : (8); (9); (10); in, For the local procedure semantic cost field, Let be any free point that can move within the space. For the cost of geometric barriers, This indicates the spatial occupancy status of the free point, where 0 represents free passage and 1 represents occupancy by physical obstacles, represented by a local geometric point cloud. Discretization mapping yields the result; The cost of procedural risks; This represents the exponential decay coefficient, which is set to 2.5. As the free point gradually approaches the safety boundary of the energized equipment, the procedural risk cost increases rapidly in an exponential manner, thereby guiding the robot away from the high-voltage area in advance. Representing free point The minimum Euclidean distance to the outer contour of the electrical equipment in the device semantic point cloud; Indicates the maximum distance affected by the cost; Points in the device semantic point cloud; The penalty base is set to 100 to increase the total cost weight of paths near high-pressure hazard areas.

[0124] Impassable area setting: when the minimum Euclidean distance is reached. At this point, it means that the space has encroached upon the high-voltage insulation red line defined by the power safety regulations. This directly translates to the regulatory risks and consequences. Set to infinity At the underlying system level, this forces the area's properties to become impassable (equivalent to a physical obstacle), thus completely preventing the robot from accidentally entering a high-voltage danger zone at the algorithm level.

[0125] Risk smoothing decay zone: when At this time, the grid is located outside the safety boundary but close to the high-voltage equipment. To guide the robot to actively move away from the high-voltage source, a penalty threshold is set. And introduce an exponential decay coefficient. ( At this point, the risk cost decays exponentially with increasing distance in a non-linear manner. The numerator of the formula calculates the exponential decay difference at the current distance, while the denominator is used to globally normalize the decay function. This ensures the continuity of the cost at the boundary transition, avoiding severe jitter in the robot's trajectory.

[0126] Absolutely safe zone: when At that time, it was determined that the discharge or electromagnetic threat from the high-voltage equipment to the area was negligible, and the risk cost of the procedure was reduced to 0.

[0127] A5.4: The main industrial control computer spatially fuses the local geometric point cloud, the equipment semantic point cloud, and the local procedure semantic cost field, and maps them according to a unified grid coordinate system to generate a local procedure semantic safety map containing space occupancy information, equipment semantic attribute information, and procedure risk cost information.

[0128] For any free point in the three-dimensional space of this local procedure semantic safety map Each has an attribute tuple containing multidimensional state information. Its data structure is defined as follows: (11); in, For equipment category; Voltage level.

[0129] A6: The main industrial control computer uses the real-time pose matrix of the quadruped robot in the world coordinate system at each moment to perform incremental data fusion on the local procedure semantic safety map constructed at multiple moments, and processes the semantic consistency under multi-view observation, and finally constructs the global procedure semantic safety map in the world coordinate system.

[0130] A7: The main industrial control computer performs real-time assessment of the environmental structure of the current substation inspection site based on the color image frame. If the assessment is successful, it performs path planning based on the global procedure semantic safety map, obtains control commands, and sends them to the quadruped robot for execution.

[0131] Real-time assessment, such as Figure 5 As shown, it includes the following steps: A7.1: Calculate the current color image frame Reference image frames from historical inspections The visual structural similarity index SSIM between the historical inspection reference image frames; To analyze current color image frames in the historical process Color image frames taken from the same area.

[0132] (14); in, It is a visual structural similarity index. Color image frames and reference image frame The average pixel brightness; It is the stability constant; Color image frames and reference image frame The pixel variance; denoted as pixel covariance.

[0133] A7.2: Calculate the current 3D point cloud data Historical reference point cloud of the previous moment Scan matching error between The historical reference point cloud is the three-dimensional point cloud data acquired at the position of the quadruped robot at the previous moment.

[0134] The scanning matching error is the error obtained by using a 3D point cloud registration algorithm on the current 3D point cloud data. Historical reference point cloud of the previous moment The sum of squared residuals during the registration process.

[0135] (15); in, For the number of points in the point cloud, For the current 3D point cloud data Point cloud points in the middle, Historical reference point cloud for the previous moment Point cloud points in the middle, The relative transformation matrix for registration output.

[0136] A7.3: The visual structural similarity index (SSIM) is weighted and fused with the scanning matching error to calculate the overall environmental similarity. .

[0137] (16); in, It is a visual structural similarity index; The weights of visual and laser features are respectively assigned during the fusion process, and satisfy the following conditions: In this embodiment, visual weights are set. =0.4, laser weight =0.6; Considering the strong electromagnetic interference and numerous similar structures in the substation environment, laser point clouds have higher stability in terms of geometric constraints, and therefore are given a greater weight; at the same time, visual structural features are retained for auxiliary verification; This represents the maximum upper limit of the scan matching error.

[0138] A7.4: To accurately identify areas with highly similar structures within a substation inspection site, comprehensive environmental similarity is utilized. Conduct an environmental assessment, if If the quadruped robot has entered a region with similar structural heights, then A7.5 is executed to activate the enhanced navigation strategy; otherwise, A7.6 is executed directly. The environmental similarity threshold is set to 0.85. When the overall environmental similarity is higher than 0.85, it is determined that the current area has highly similar structural features, and the enhanced navigation strategy is activated. This threshold can effectively identify similar structural areas such as switch cabinet arrays and isolation switch areas, while avoiding the enhanced navigation mechanism from being triggered erroneously in ordinary scenarios.

[0139] A7.5: Based on the overall environmental similarity Calculate the adaptive heading penalty weight .

[0140] (17); in, The positive gain scaling constant is set to 1.5, which ensures that the higher the geometric similarity of the environment, the greater the corrective penalty applied to the robot's heading, thereby suppressing trajectory deviation.

[0141] A7.6: Based on the global procedure semantic safety map, a two-layer local path planning strategy is used to realize path planning and obtain control commands.

[0142] A7.6.1: Establish the first-level spatial path cost integral function And under the absolute safety distance constraint of the power regulations, by minimizing the first-level spatial path cost integral function. A safe reference path was found (within an area with dense electrical equipment). .

[0143] First-layer spatial path cost integral function for: (18); The absolute safety distance constraint stipulated in the power regulations is: (19); in, For path points, the value is expressed as the arc length of the trajectory. Parametric equations; Let the arc length of the trajectory be . The total arc length of the reference path, , The path length weight and risk cost weight are set to 0.4 and 0.6 respectively. By increasing the risk cost weight, the path planning prioritizes meeting the safety constraints of the power regulations. As a consequence of procedural risks, path point The standard safety distance value at the location is equal to the path point. To the corresponding device semantic point cloud The minimum Euclidean distance of the profile of the electrical equipment.

[0144] In the first-level safety-guided local path planning, the first-level spatial path cost integral function Calculating the arc length of the trajectory When assessing the risks and costs of procedures, the system will query the global procedure semantic security map in real time. Because the global procedure semantic security map satisfies The procedural risks and costs in the region have been immeasurably high. First-level spatial path cost integral function Evaluating any candidate path traversing the region yields an infinite total cost, driving the underlying search algorithm to automatically discard illegal paths, thus resulting in a final reference path. Strictly adhering to the absolute safety distance constraints of power regulations, the quadruped robot is guided to travel safely in areas with dense electrical equipment.

[0145] A7.6.2: Establishing a continuously optimized objective functional In the reference path Trajectory optimization is performed under the constraints of the kinematic differential hard constraints of the quadruped robot to obtain a safe and smooth trajectory. The adaptive heading penalty weight is introduced as a heading deviation constraint term into the continuously optimized objective functional to enhance the trajectory direction stability in the region of structural similarity and to constrain the quadruped robot's actual yaw angle to be close to the expected yaw angle of the reference path.

[0146] The continuously optimized objective functional for: (20); The kinematic differential hard constraints of the quadruped robot: (twenty one); in, To represent the state equation of the actual motion trajectory of the quadruped robot. , For three-dimensional position, Yaw angle; Plan the time range for the trajectory; , The trajectory smoothing cost weighting coefficient and the safety risk cost weighting coefficient are set to 0.3 and 0.7 respectively. For quadruped robots The actual linear velocity vector at time t. For quadruped robots The actual linear acceleration vector at time t, The acceleration square term of the quadruped robot is used as a kinematic smoothing penalty term in the objective functional to suppress the violent acceleration and deceleration during the movement of the quadruped robot, thereby ensuring the smoothness of the generated trajectory. This is the cost of the safety field risk at the current three-dimensional location; The permissible physical limits for linear velocity and acceleration of the chassis controller for the quadruped robot are 1.2 m / s² and 0.8 m / s², respectively. 2 ; Reference path exist The expected yaw angle at any given moment.

[0147] A7.6.3: To ensure that the trajectory optimization of the second layer does not violate the safety constraints of the first layer, after obtaining the safe and smooth trajectory, a safety check is forcibly performed on the generated safe and smooth trajectory. If the check passes, A7.6.4 is executed; otherwise, the circuit is immediately broken and the safe and smooth trajectory is determined to be invalid. At this time, the safe and smooth trajectory is automatically abandoned and replanned. If the replanning fails continuously, it is determined to be trapped in a dead zone, and the master industrial control computer ends the navigation and inspection of the quadruped robot in the autonomous operation mode.

[0148] The security verification specifically involves: extracting a secure smooth trajectory. The trajectory points on the curve are substituted back into the local procedure semantic cost field. To obtain the actual procedural risk costs, if trajectory optimization causes a segment of the safe, smooth trajectory to deviate excessively from the reference path. This ensures that any trajectory point intrudes into the minimum safe distance. If the absolute safety distance constraint of the power regulations is not met, the verification will fail; otherwise, the verification will pass.

[0149] A7.6.4: Only after the safe and smooth trajectory passes the verification can it be converted into the control commands of the quadruped robot through the inverse kinematics solution of the underlying layer, so as to complete the highly stable autonomous inspection in the complex environment of the substation. The quadruped robot will then send its own status information back to the main industrial control computer.

[0150] In autonomous operation mode, task status monitoring and anomaly detection are performed in real time. When the quadruped robot can make autonomous decisions, the inspection task is executed autonomously. If the quadruped robot cannot make a decision during autonomous inspection in the following situations, the abnormal interruption trigger condition is determined to be met, the master industrial control computer terminates the quadruped robot's navigation and inspection in autonomous operation mode, and controls the quadruped robot to stop running in place: (1) Due to factors such as temporary material storage or dynamic obstacles at the substation inspection site, the safe and smooth trajectory that meets the absolute safety distance constraint of the power regulations cannot be searched in the global procedure semantic safety map. The quadruped robot is trapped in the dead zone or cannot search for a safe and smooth trajectory that meets the absolute safety distance constraint of the power regulations within the set evaluation time window of 20 seconds. This time window corresponds to the average time required for the quadruped robot to complete local escape or replanning at normal inspection speed. It can avoid misjudgment caused by dynamic environmental interference and prevent long-term stagnation from affecting the continuity of inspection.

[0151] (2) When the quadruped robot encounters physical obstruction, severe sideslip, or other situations that restrict its autonomous operation and prevent it from escaping: within the set monitoring period, if the absolute value of the deviation between the expected motion displacement corresponding to the control command of the quadruped robot and the actual physical displacement exceeds the set maximum displacement deviation threshold of 0.15 meters, or if the absolute value of the side tilt angle of the quadruped robot exceeds 5° or the absolute value of the pitch angle exceeds 10°, that is, if the set limit safety attitude angle for preventing the equipment from tipping over is exceeded. like Figure 6 As shown, the workflow steps of the quadruped robot navigation and inspection method in remote teleoperation mode are as follows: B1: The master industrial control computer sends a takeover request to the remote industrial control computer and simultaneously uploads the real-time pose matrix of the actual quadruped robot, the global procedure semantic safety map, and the 3D point cloud data collected in real time by the 3D LiDAR.

[0152] When the quadruped robot's autonomous operation is restricted, the quadruped robot stops moving, but the on-site inspection execution end continues to operate to maintain real-time perception. The 3D LiDAR maintains high-frequency scanning, and the main industrial control computer continuously outputs the quadruped robot's real-time pose matrix to provide data support for subsequent collision-free simulations.

[0153] B2: The remote industrial control computer receives the takeover request message, reconstructs the virtual substation inspection site containing the safety constraints of the procedure according to the global procedure semantic safety map, and establishes a virtual quadruped robot in the virtual substation inspection site, and maps the three-dimensional point cloud data to the virtual substation inspection site to achieve pose alignment between the virtual quadruped robot and the real quadruped robot.

[0154] B3: The remote industrial control computer sends the virtual substation inspection site, which includes the virtual quadruped robot, to the visual interactive interface on the independent display screen. The operator can directly input planar motion commands through the visual interactive interface on the independent display screen to drive the quadruped virtual robot to perform prior simulation motion in the virtual substation inspection site in the visual interactive interface until the virtual robot escapes the predicament. The remote industrial control computer simultaneously records the continuous virtual escape trajectory generated by the prior simulation motion.

[0155] Specifically, the steps include: B3.1: For example Figure 7 As shown, the remote industrial control computer initializes a virtual substation inspection site including a virtual quadruped robot and displays it on a visual interactive interface on an independent display screen.

[0156] B3.2: The operator inputs a planar motion command in the interactive coordinate system through a visual interactive interface. The planar motion command is then forwarded to a remote industrial control computer, which converts the planar motion command in the interactive coordinate system into the expected speed command in the virtual machine coordinate system.

[0157] The interactive coordinate system has its origin at the center of the visual interactive interface. The positive direction is along the vertical axis pointing upwards towards the top of the visual interactive interface. The positive direction is the horizontal axis pointing towards the left side of the visual interactive interface. The direction of counterclockwise rotation of the axis along the visual interactive interface is the positive direction; the virtual machine coordinate system has its origin at the geometric center of the virtual quadruped robot chassis. The positive direction is defined as the horizontal direction directly forward along the axis of the virtual quadruped robot. The positive direction is the horizontal direction to the left of the virtual quadruped robot. The positive direction is the axis perpendicular to the virtual quadruped robot and pointing upwards.

[0158] In this embodiment, as shown Figure 2 As shown, the longitudinal motion control of the virtual quadruped robot corresponds to the interactive coordinate system. The axis refers to the interactive coordinate system when the operator presses the forward key alone. Axial component A positive value generates the positive longitudinal linear velocity of the virtual quadruped robot. When the back button is pressed alone, the interactive coordinate system... Axial component Negative values. Lateral motion control corresponds to the interactive coordinate system. Axis. When the left mouse button is pressed alone, the interactive coordinate system is activated. Axial component A positive value generates the leftward lateral linear velocity of the virtual quadruped robot. When you press the right mouse button alone, It is a negative value.

[0159] Steering control corresponding interactive coordinate system shaft and The axis is coupled to the output. When the forward and left keys are pressed simultaneously, the interactive coordinate system... Axial component Positive values ​​and interactive coordinate system Axial component A positive value generates the positive yaw rate of the virtual quadruped robot. With longitudinal linear velocity This drives the virtual robot to turn left; when the forward and right keys are pressed simultaneously, the interactive coordinate system... Axial component Negative values ​​and interactive coordinate system Axial component A positive value drives the virtual quadruped robot to turn right.

[0160] Set the planar motion command in the interactive coordinate system as follows Then its corresponding expected speed command The expression is as follows: (twenty two); in, These are the set linear velocity and angular velocity mapping ratios, with the linear velocity mapping ratio set to 0.8 and the angular velocity mapping ratio set to 0.6.

[0161] B3.3: Convert the expected velocity command in the virtual machine coordinate system from a discrete control quantity to a virtual pose sequence of chassis kinematics to obtain a continuous virtual escape trajectory.

[0162] Specifically: Let the single-step simulation cycle of the 3D simulation engine be... =100ms, the virtual quadruped robot in the... The two-dimensional plane position of the step is Yaw angle is Its corresponding first Step Two-Dimensional Plane Position With yaw angle The update model is expressed as follows: (twenty three); (twenty four); Let the state vector The operator will attempt to escape the predicament by taking a total of steps. By concatenating all the internally generated state vectors sequentially, a continuous virtual escape trajectory is obtained. The expression is as follows: (25); B4: Conduct a global procedure safety assessment on the continuous virtual escape trajectory generated by the operator and obtain the assessment results.

[0163] The specific implementation method is as follows: B4.1: Utilizing the outer contour of the energized equipment in the semantic point cloud, perform 3D expansion according to the minimum safe distance corresponding to the energized equipment to generate a virtual insulation envelope as a high-voltage obstacle, thereby obtaining a set of high-voltage obstacles. .

[0164] Specifically, devices of different voltage levels will trigger different expansion radii, allowing the generated virtual insulation package to accurately map the insulation boundaries required by power regulations, thus serving as a set of high-voltage obstacles in the virtual scene. .

[0165] B4.2: Calculate the shortest geometric distance between the surface of the virtual quadruped robot and the high-pressure obstacle on a continuous virtual escape trajectory.

[0166] Specifically, the virtual quadruped robot on the continuous virtual escape trajectory is designated as the collision detection entity. To balance computational efficiency and detection accuracy, the URDF description file of the virtual quadruped robot is read, its collision geometry model is extracted, and the surface of the collision geometry model is discretized and sampled to generate a discrete coordinate point. The constructed robot surface point cloud model is used as a collision detection entity. The expression; in the continuous virtual escape trajectory At each discrete moment, the point cloud model of the robot surface is spatially transformed according to the robot's real-time pose matrix. The shortest geometric distance between the virtual quadruped robot surface and the high-pressure obstacle on the continuous virtual escape trajectory is calculated in real time, and the spatial relationship between the robot and the obstacle is converted from high-dimensional geometric features into scalar distance form. (26); in, State vector The shortest geometric distance between the surface of the virtual quadruped robot and the high-pressure obstacle on the next continuous virtual escape trajectory. For discrete points on the surface of the virtual quadruped robot body, For collision detection entities in the state vector The robot surface point cloud model below; These are discrete points on the virtual insulation envelope corresponding to the high-voltage obstacle; Representing discrete points respectively Coordinate components in the three-dimensional simulation space; Representing discrete points respectively Coordinate components in the three-dimensional simulation space.

[0167] By traversing all state vectors in a continuous virtual escape trajectory, the global shortest spatial distance of the entire continuous virtual escape trajectory is found. The expression is as follows: (27); B4.3: Convert the shortest geometric distance between the surface of the virtual quadruped robot and the high-pressure obstacle on the continuous virtual escape trajectory from numerical form to a Boolean simulation safety escape verification identifier as the evaluation result; (28); in, This is a Boolean-type simulation security exit check identifier. The compensation safety margin is set to overcome the physical tracking error of the machine body.

[0168] B5: Based on the evaluation results, if the continuous virtual escape trajectory is safe along the entire route, the continuous virtual escape trajectory will be sent to the real quadruped robot for action execution. If a collision is detected or the absolute safety distance constraint of the power regulations is not met, the continuous virtual escape trajectory will be stopped and the operator will be prompted to replan the trajectory in the visual interactive interface.

[0169] Based on the Boolean simulation safety exit check identifier Generate the final escape strategy for a realistic quadruped robot; (29); in, The final escape strategy for a real quadruped robot; when The value is This indicates that the continuous virtual escape trajectory rehearsed by the operators has successfully achieved the goal of escaping the predicament. This means that the continuous virtual escape trajectory is sent to the real quadruped robot to perform the escape operation, thereby fully avoiding the risk of loss of control on site due to communication delays. The value is This indicates that there may be safety issues with the continuous virtual escape trajectory rehearsed by the operators. This indicates that the real quadruped robot will remain in a stopped state and will prompt the operator to replan a continuous virtual escape trajectory within a safe range.

[0170] B6: After the real quadruped robot successfully escapes its predicament, the main industrial control computer reacquires multi-source perception data. If it determines that the current environment has met the conditions for generating a safe reference path, it sends a prompt message indicating that the escape is complete to the remote control terminal and switches back to the autonomous operation mode.

[0171] To mitigate the collision risks associated with network latency during the entire remote operation process, a cross-platform asynchronous communication mechanism and an emergency stop protection mechanism were added to the control architecture. During the execution of the virtual escape trajectory, a reliable TCP-based transmission link was established between the remote operator and the on-site master industrial control computer to transmit the continuous virtual escape trajectory, including discrete pose sequences. The data is packaged into lightweight packets. The master industrial control computer receives this data and sends it to the quadruped robot, driving each joint to perform actions. This asynchronous data delivery and local execution mechanism reduces the robot's reliance on real-time network conditions during execution, preventing loss of robot control or collisions due to network latency.

[0172] When the quadruped robot performs an action, an emergency stop protection mechanism is activated. The main industrial control computer is responsible for acquiring perception data of the surrounding environment. If a dynamic obstacle suddenly intrudes into the absolute safe distance during the trajectory execution process, the robot will take immediate action. If the robot's tilt angle exceeds a safe threshold, an emergency stop command will be triggered, interrupting the current action and reporting the status to the remote control terminal, informing the operator to replan the route. This ensures the robot's operational safety in unexpected or emergency situations.

Claims

1. A quadruped robot substation autonomous and remote operation dual-mode inspection system, characterized in that, This includes both on-site inspection execution terminals and remote control terminals; The remote control terminal is deployed in the substation monitoring room. In autonomous operation mode, it is used to visually monitor the operating status of the field inspection execution terminal based on the multi-source sensing data transmitted back from the field inspection execution terminal. In remote teleoperation mode, the operator generates a continuous virtual escape trajectory and performs a global procedural safety assessment on the continuous virtual escape trajectory; After the assessment is passed, the continuous virtual escape trajectory is sent to the on-site inspection execution terminal for execution; the multi-source perception data includes three-dimensional point cloud data, color image frames, depth image frames and inertial measurement data; the global procedure safety assessment is based on collision detection and the absolute safety distance constraint of the power procedure, and performs time-by-time spatial geometric safety verification on the continuous virtual escape trajectory in each discrete state, and traverses the entire continuous virtual escape trajectory to generate the assessment result; The on-site inspection execution terminal is set up at the substation inspection site, acquires multi-source sensing data in real time and transmits it back to the remote operation terminal. In autonomous operation mode, it autonomously completes the safety path planning and autonomous inspection tasks within the substation inspection site. In remote control mode, it responds to the virtual escape trajectory issued by the remote control terminal and executes the escape action.

2. The quadruped robot substation autonomous and remote operation dual-mode inspection system according to claim 1, characterized in that, The on-site inspection execution terminal includes a quadruped robot, a main industrial control computer, a 3D LiDAR, a depth camera, and network communication equipment; the 3D LiDAR has a built-in inertial measurement unit. The quadruped robot is used to receive motion control commands from the master industrial control computer to move at the substation inspection site and transmit its own status information back to the master industrial control computer; the status information includes pose and velocity. The three-dimensional lidar is mounted on the quadruped robot using a first mounting connector formed by 3D printing. It is used to acquire three-dimensional point cloud data and inertial measurement data of the substation inspection site in real time and transmit them to the main industrial control computer. The depth camera is fixed to the front end of the quadruped robot using a second mounting connector formed by 3D printing. It is used to acquire RGB-D video data streams at the substation inspection site and transmit them to the main industrial control computer. The RGB-D video data streams include color image frames and depth image frames at different times. In autonomous operation mode, the master industrial control computer receives 3D point cloud data, inertial measurement data, and RGB-D video data stream, and sends the color image frames in the processed RGB-D video data stream to the network communication device; at the same time, it performs device semantic recognition based on the color image frames and depth image frames acquired by the depth camera, and fuses the recognized device semantics with the 3D point cloud data acquired by the lidar to construct a global procedure semantic safety map, performs path planning based on the global procedure semantic safety map, generates executable quadruped robot motion control commands, and sends them to the quadruped robot; When switching from autonomous operation mode to remote control mode, the master industrial control computer sends a takeover request message to the network communication device and simultaneously uploads the quadruped robot's own status information and global procedure semantic safety map. In remote control mode, the master industrial control computer receives color frames acquired by the depth camera and three-dimensional point cloud data acquired by the lidar, and sends them to the network communication device; it also receives motion control commands sent by the network communication device and forwards them to the quadruped robot. The network communication device is used to transmit the color frame image output by the master industrial control computer and the three-dimensional point cloud data acquired by the lidar to the remote teleoperation terminal; to forward the quadruped robot's own status information and global procedure semantic safety map to the remote teleoperation terminal; and to receive motion control commands sent by the remote teleoperation terminal and forward them to the master industrial control computer.

3. The quadruped robot substation autonomous and remote operation dual-mode inspection system according to claim 1, characterized in that, The remote control terminal includes an independent display screen and a remote industrial control computer; In both autonomous operation mode and remote control mode, the remote industrial control computer receives color image frames sent by the on-site inspection execution terminal and sends them to an independent display screen. In remote operation mode, the remote industrial control computer receives the takeover request message sent by the field inspection execution terminal and simultaneously receives the global procedure semantic safety map and three-dimensional point cloud data. Based on the global procedure semantic safety map and 3D point cloud data, a virtual substation inspection site and a virtual quadruped robot are constructed; the virtual substation inspection site containing the virtual quadruped robot is sent to an independent display screen; the planar motion command sent by the independent display screen is responded to in real time, driving the quadruped virtual robot to perform prior simulation motion in the virtual substation inspection site in the visual interactive interface, and the continuous virtual escape trajectory generated by the prior simulation motion is recorded simultaneously. A global procedure safety assessment is performed on the continuous virtual escape trajectory generated by the operator to obtain the assessment results. If the continuous virtual escape trajectory is safe along the entire route, it is sent as a motion control command to the network communication device. If a collision is detected or the absolute safety distance constraint of the power procedure is not met, a replanning prompt is sent to the independent display screen. The independent display screen is used to visualize color image frames and a virtual substation inspection site containing a virtual quadruped robot; the operator can directly input planar motion commands to the remote industrial control computer through the visual interactive interface on the independent display screen; the operator is prompted to replan.

4. A dual-mode inspection method for substations using a quadruped robot, characterized in that... This includes navigation and inspection methods for quadruped robots in autonomous operation mode and navigation and inspection methods for quadruped robots in remote teleoperation mode.

5. The method for dual-mode inspection of a substation using a quadruped robot with both autonomous and remote operation, as described in claim 4, is characterized in that... The autonomous operation mode of the quadruped robot navigation and inspection method includes the following steps: A1: Real-time acquisition of multi-source sensing data at the substation inspection site via 3D LiDAR and depth camera, and transmission of the acquired multi-source sensing data to the master industrial control computer. The master industrial control computer performs timestamp alignment on the multi-source sensing data to generate synchronized multi-source sensing data. A2: The main industrial control computer uses the laser-inertial fusion odometer to estimate the state of the quadruped robot based on the 3D point cloud data and inertial measurement data from the synchronous multi-source sensing data. This gives the quadruped robot the real-time pose matrix in the world coordinate system at the current moment. At the same time, the color image frames are compressed and encoded, and then transmitted back to the independent display screen of the remote control terminal in real time through network communication equipment to realize the visualization display of the substation inspection site. The world coordinate system uses the starting position of the quadruped robot as the origin, with the Z-axis perpendicular to the horizontal plane and upward as the positive direction, the X-axis along the front of the quadruped robot as the positive direction, and the Y-axis horizontal to the left as the positive direction. A3: The main industrial control computer uses a pre-trained YOLO instance segmentation network to perform target detection and instance segmentation on various types of live equipment at the substation inspection site captured by color image frames, outputs recognition results and detection confidence scores, and removes invalid recognition results based on the detection confidence scores; the recognition results include live equipment category labels, bounding box coordinates, and pixel-level instance masks; A4: Extracting multiple target pixels of a charged device in a two-dimensional image coordinate system from a pixel-level instance mask. Then, pixel-level instance masks are used to extract the depth image frames, thereby obtaining the spatial position vector of the charged device in the 3D camera coordinate system. ; A back-projection mathematical model is established to back-project the target pixels of the powered device in the two-dimensional image coordinate system, thereby obtaining the spatial position vector of the powered device in the three-dimensional camera coordinate system. : （5）； The mathematical model for back projection is: （6）； The three-dimensional camera coordinate system uses the optical center of the depth camera as its origin. The axis is positive when it is horizontal to the right. The axis is vertically downwards as positive. The axis is positive when it faces forward. In the spatial position vector axis, shaft and The components of the axis; the two-dimensional image coordinate system uses the upper left corner of the color image frame as the origin, The positive direction is to the right along the horizontal axis of the color image frame. The positive direction is the vertical direction downwards along the color image frame. For the target pixel of the charged device in shaft and The components of the axis; This is the inverse of the depth camera intrinsic parameter matrix. For transpose; A5: Based on the spatial position vector of the charged device in the 3D camera coordinate system The real-time pose matrix and recognition results of the quadruped robot in the world coordinate system at the current moment are used to construct a local procedure semantic safety map; A6: The main industrial control computer uses the real-time pose matrix of the quadruped robot in the world coordinate system at each moment to perform incremental data fusion on the local procedure semantic safety map constructed at multiple moments, and processes the semantic consistency under multi-view observation, and finally constructs the global procedure semantic safety map in the world coordinate system. A7: The main industrial control computer performs real-time assessment of the environmental structure of the current substation inspection site based on the color image frame. If the assessment is successful, it performs path planning based on the global procedure semantic safety map, obtains control commands, and sends them to the quadruped robot for execution.

6. A dual-mode inspection method for a quadruped robot substation with autonomous and remote operation as described in claim 5, characterized in that, The A5 specifically includes the following steps: A5.1: Using the real-time pose matrix of the quadruped robot in the world coordinate system at the current moment, the acquired single-frame 3D point cloud data is processed. Unified conversion to the world coordinate system, and within the set time window The system registers and stitches together multiple frames of 3D point cloud data to construct a local geometric point cloud that can truly reflect the substation inspection site. The local geometric point cloud is represented as follows: （7）； in, This represents a local geometric point cloud. A5.2: Spatial position vectors of all charged devices in the 3D camera coordinate system Mapping to the world coordinate system yields the device point cloud; assigning semantic attribute labels to the device point cloud yields the device semantic point cloud. ; The semantic attribute tag is represented as , For equipment category, Voltage level; Minimum safe distance; A5.3: Transform the requirements of power safety regulations into computable mathematical constraints to construct a local regulation semantic cost field; Local procedure semantic cost field : （8）； （9）； （10）； in, For the local procedure semantic cost field, Let be any free point that can move within the space. For the cost of geometric barriers, This indicates the spatial occupancy status of the free point, where 0 represents free passage and 1 represents occupancy by physical obstacles, represented by a local geometric point cloud. Discretization mapping yields the result; The cost of procedural risks; Indicates the exponential decay coefficient; Representing free point The minimum Euclidean distance to the outer contour of the electrical equipment in the device semantic point cloud; Indicates the maximum distance affected by the cost; Points in the device semantic point cloud; As the base for punishment; A5.4: The main industrial control computer spatially fuses the local geometric point cloud, the equipment semantic point cloud, and the local procedure semantic cost field to generate a local procedure semantic safety map. For any free point in the three-dimensional space of this local procedure semantic safety map Each has an attribute tuple containing multidimensional state information. Its data structure is defined as follows: （11）； in, For equipment category; Voltage level.

7. A dual-mode inspection method for a quadruped robot substation with autonomous and remote operation as described in claim 5, characterized in that, The A7 specifically includes the following steps: A7.1: Calculate the current color image frame Reference image frames from historical inspections The visual structural similarity index SSIM between the historical inspection reference image frames; To analyze current color image frames in the historical process Color image frames captured from the same area; （14）； in, It is a visual structural similarity index. Color image frames and reference image frame The average pixel brightness; It is the stability constant; Color image frames and reference image frame The pixel variance; For pixel covariance; A7.2: Calculate the current 3D point cloud data Historical reference point cloud of the previous moment Scan matching error between The historical reference point cloud is the three-dimensional point cloud data acquired at the position of the quadruped robot at the previous moment. （15）； in, For the number of points in the point cloud, For the current 3D point cloud data Point cloud points in the middle, Historical reference point cloud for the previous moment Point cloud points in the middle, The relative transformation matrix for the registration output; A7.3: The visual structural similarity index (SSIM) is weighted and fused with the scanning matching error to calculate the overall environmental similarity. ; （16）； in, It is a visual structural similarity index; The weights of visual and laser features are respectively assigned during the fusion process, and satisfy the following conditions: ; This represents the maximum upper limit of the scan matching error. A7.4: Utilizing comprehensive environmental similarity Conduct an environmental assessment, if If the quadruped robot has entered a region with similar structural heights, then A7.5 is executed to activate the enhanced navigation strategy; otherwise, A7.6 is executed directly. The threshold for determining environmental similarity; A7.5: Based on the overall environmental similarity Calculate the adaptive heading penalty weight ; （17）； in, It is the positive gain proportionality constant; A7.6: Based on the global procedure semantic safety map, a two-layer local path planning strategy is used to realize path planning and obtain control commands; A7.6 specifically includes the following steps: A7.6.1: Establish the first-level spatial path cost integral function And under the absolute safety distance constraint of the power regulations, by minimizing the first-level spatial path cost integral function. A safe reference path was found. ; First-layer spatial path cost integral function for: （18）； The absolute safety distance constraint stipulated in the power regulations is: （19）； in, For path points, the value is expressed as the arc length of the trajectory. Parametric equations; Let the arc length of the trajectory be . The total arc length of the reference path, , These are the path length weight and the risk cost weight, respectively. As a consequence of procedural risks, path point The standard safety distance value at the location is equal to the path point. To the corresponding device semantic point cloud The minimum Euclidean distance of the profile of the electrical equipment in the middle; A7.6.2: Establishing a continuously optimized objective functional In the reference path Trajectory optimization is performed under the constraints of kinematic differential hard constraints of the quadruped robot to obtain a safe and smooth trajectory; the adaptive heading penalty weight is introduced into the objective functional as a heading deviation constraint term; The continuously optimized objective functional for: （20）； The kinematic differential hard constraints of the quadruped robot: （21）； in, To represent the state equation of the actual motion trajectory of the quadruped robot. , For three-dimensional position, Yaw angle; Plan the time range for the trajectory; , These are the trajectory smoothing cost weighting coefficient and the safety risk cost weighting coefficient, respectively. For quadruped robots The actual linear velocity vector at time t. For quadruped robots The actual linear acceleration vector at time t, Let be the square of the acceleration term of the quadruped robot; This is the cost of the safety field risk at the current three-dimensional location; The physical limits of linear velocity and acceleration allowed by the chassis controller of the quadruped robot; Reference path exist The expected yaw angle at any given moment; A7.6.3: To ensure that the trajectory optimization of the second layer does not violate the safety constraints of the first layer, after obtaining the safe and smooth trajectory, a safety check is forcibly performed on the generated safe and smooth trajectory. If the check passes, A7.6.4 is executed; otherwise, the circuit is immediately broken and the safe and smooth trajectory is determined to be invalid. At this time, the safe and smooth trajectory is automatically abandoned and replanned. If the replanning fails continuously, it is determined to be trapped in a dead zone, and the master industrial control computer ends the navigation and inspection of the quadruped robot in the autonomous operation mode. The security verification specifically involves: extracting a secure smooth trajectory. The trajectory points on the curve are substituted back into the local procedure semantic cost field. To obtain the actual procedural risk costs, if trajectory optimization causes a segment of the safe, smooth trajectory to deviate excessively from the reference path. This ensures that any trajectory point intrudes into the minimum safe distance. If the absolute safety distance constraint of the power regulations is not met, the verification fails; otherwise, the verification passes. A7.6.4: Only after the safe and smooth trajectory passes the verification can it be converted into the control commands of the quadruped robot through the inverse kinematics solution of the underlying layer, so as to complete the highly stable autonomous inspection in the complex environment of the substation. The quadruped robot will then send its own status information back to the main industrial control computer.

8. A dual-mode inspection method for a quadruped robot substation, combining autonomous and remote operation, as described in claim 4, is characterized in that... In autonomous operation mode, if the quadruped robot cannot make a decision during autonomous inspection in the following situations, the abnormal interruption trigger condition is determined to be met. The master industrial control computer will end the quadruped robot's navigation and inspection in autonomous operation mode and control the quadruped robot to stop running in place: (1) The quadruped robot cannot find a safe and smooth trajectory that meets the absolute safety distance constraint of the power regulations in the global procedure semantic safety map. The quadruped robot is trapped in the dead zone or cannot find a safe and smooth trajectory that meets the absolute safety distance constraint of the power regulations within the set evaluation time window of 20 seconds. (2) Within the set monitoring period, the absolute value of the deviation between the expected motion displacement corresponding to the control command of the quadruped robot and the actual physical displacement exceeds the set maximum displacement deviation threshold or the absolute value of the tilt angle or pitch angle of the quadruped robot, and exceeds the set limit safety attitude angle to prevent the equipment from tipping over.

9. A dual-mode inspection method for a quadruped robot substation with autonomous and remote operation as described in claim 4, characterized in that, The workflow steps of the quadruped robot navigation and inspection method in the remote teleoperation mode are as follows: B1: The master industrial control computer sends a takeover request to the remote industrial control computer and simultaneously uploads the real-time pose matrix of the real quadruped robot, the global procedure semantic safety map, and the 3D point cloud data collected in real time by the 3D LiDAR. B2: The remote industrial control computer receives the takeover request message, reconstructs the virtual substation inspection site containing the safety constraints of the procedure according to the global procedure semantic safety map, and establishes a virtual quadruped robot in the virtual substation inspection site, and maps the three-dimensional point cloud data to the virtual substation inspection site to achieve pose alignment between the virtual quadruped robot and the real quadruped robot. B3: The remote industrial control computer sends the virtual substation inspection site, which includes the virtual quadruped robot, to the visual interactive interface on the independent display screen. The operator can directly input planar motion commands through the visual interactive interface on the independent display screen to drive the quadruped virtual robot to perform prior simulation motion in the virtual substation inspection site in the visual interactive interface until the virtual robot escapes the predicament. The remote industrial control computer synchronously records the continuous virtual escape trajectory generated by the prior simulation motion. The B3 specifically includes the following steps: B3.1: The remote industrial control computer initializes the virtual substation inspection site, which includes a virtual quadruped robot, and displays it on a visual interactive interface on an independent display screen; B3.2: The operator inputs a planar motion command in the interactive coordinate system through a visual interactive interface. The planar motion command is then forwarded to a remote industrial control computer, which converts the planar motion command in the interactive coordinate system into the expected speed command in the virtual machine coordinate system. The interactive coordinate system has its origin at the center of the visual interactive interface. The positive direction is along the vertical axis pointing upwards towards the top of the visual interactive interface. The positive direction is the horizontal axis pointing towards the left side of the visual interactive interface. The direction of counterclockwise rotation of the axis along the visual interactive interface is the positive direction; the virtual machine coordinate system has its origin at the geometric center of the virtual quadruped robot chassis. The positive direction is defined as the horizontal direction directly forward along the axis of the virtual quadruped robot. The positive direction is the horizontal direction to the left of the virtual quadruped robot. The positive direction is the axis perpendicular to the virtual quadruped robot and pointing upwards. Set the planar motion command in the interactive coordinate system as follows Then its corresponding expected speed command The expression is as follows: （22）； in, These are the set linear velocity and angular velocity mapping ratio coefficients, respectively; B3.3: Convert the expected velocity command in the virtual machine coordinate system from a discrete control quantity to a virtual pose sequence of chassis kinematics to obtain a continuous virtual escape trajectory; Specifically: Let the single-step deduction period be... The virtual quadruped robot in the first The two-dimensional plane position of the step is Yaw angle is Then the first Step Two-Dimensional Plane Position With yaw angle The update model is expressed as follows: （23）； （24）； Let the state vector The operator will attempt to escape the predicament by taking a total of steps. By concatenating all the internally generated state vectors sequentially, a continuous virtual escape trajectory is obtained. The expression is as follows: （25）； B4: Conduct a global procedural safety assessment of the continuous virtual escape trajectory generated by the operator, and obtain the assessment results; B5: Based on the evaluation results, if the continuous virtual escape trajectory is safe along the entire route, the continuous virtual escape trajectory will be sent to the real quadruped robot for action execution. If a collision is detected or the absolute safety distance constraint of the power regulations is not met, the continuous virtual escape trajectory will be stopped and the operator will be prompted to replan in the visual interactive interface. B5.1: Based on the Boolean simulation safety exit verification identifier Generate the final escape strategy for a realistic quadruped robot; （29）； in, The ultimate escape strategy for a real quadruped robot; This indicates that the continuous virtual escape trajectory will be sent to the real quadruped robot to perform the escape operation; This indicates that the actual quadruped robot will remain stationary and will issue a prompt to the operator; B6: After the real quadruped robot successfully escapes its predicament, the main industrial control computer reacquires multi-source perception data. If it determines that the current environment has met the conditions for generating a safe reference path, it sends a prompt message indicating that the escape is complete to the remote control terminal and switches back to the autonomous operation mode.

10. A dual-mode inspection method for a quadruped robot substation with autonomous and remote operation as described in claim 9, characterized in that, The B4 specifically includes the following steps: The B4 specifically includes the following steps: B4.1: Utilizing the outer contour of the energized equipment in the semantic point cloud, perform 3D expansion according to the minimum safe distance corresponding to the energized equipment to generate a virtual insulation envelope as a high-voltage obstacle, thereby obtaining a set of high-voltage obstacles. ; B4.2: Calculate the shortest geometric distance between the surface of the virtual quadruped robot and the high-pressure obstacle on a continuous virtual escape trajectory; Specifically, the virtual quadruped robot on the continuous virtual escape trajectory is designated as the collision detection entity. The system reads the URDF description file of the virtual quadruped robot, extracts its collision geometry model, and discretizes the surface of the collision geometry model to generate a discrete coordinate point. The constructed robot surface point cloud model is used as a collision detection entity. The expression; in the continuous virtual escape trajectory At each discrete moment, the point cloud model of the robot surface is spatially transformed according to the robot's real-time pose matrix, and the shortest geometric distance between the virtual quadruped robot surface and the high-pressure obstacle on the continuous virtual escape trajectory is calculated in real time: （26）； in, State vector The shortest geometric distance between the surface of the virtual quadruped robot and the high-pressure obstacle on the next continuous virtual escape trajectory. For discrete points on the surface of the virtual quadruped robot body, For collision detection entities in the state vector The robot surface point cloud model below; These are discrete points on the virtual insulation envelope corresponding to the high-voltage obstacle; Representing discrete points respectively Coordinate components in the three-dimensional simulation space; Representing discrete points respectively Coordinate components in the three-dimensional simulation space; By traversing all state vectors in a continuous virtual escape trajectory, the global shortest spatial distance of the entire continuous virtual escape trajectory is found. The expression is as follows: （27）； B4.3: Convert the shortest geometric distance between the surface of the virtual quadruped robot and the high-pressure obstacle on the continuous virtual escape trajectory from numerical form to a Boolean simulation safety escape verification identifier as the evaluation result; （28）； in, This is a Boolean-type simulation security exit check identifier. To compensate for the safety margin.

Images