Anti-falling early warning method, system, medium and product for patrol robot protection

Abstract

A kind of anti-falling early warning method, system, medium and product for inspection robot protection, relate to data processing technical field.In the method, the morphology contact surface of robot, environment geometric topology and dynamics state and other multi-modal perception data are acquired;Secondly, the matching support polygon is constructed by fusing data, and the three-dimensional falling boundary field of edge is extracted;Then, the trajectory of dynamic equivalent zero moment point is calculated and projected into the support polygon, to obtain the relative stable margin;Then, the spatiotemporal convergence rate of the trajectory approaching the falling boundary field is extracted, and the stable margin is combined to construct the embodied anti-falling risk tensor reflecting the instability falling trend;Finally, the tensor is mapped to a comprehensive early warning index, to trigger the corresponding embodied active anti-falling response instruction.Implementation of the technical solution provided in the present application can improve the accuracy of anti-falling early warning of inspection robot.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a fall prevention warning method, system, medium, and product for the protection of inspection robots. Background Technology

[0002] With the rapid development of artificial intelligence and the Internet of Things (IoT) technologies, machine vision-based positioning technology has been widely applied in mobile robots, drone navigation, augmented reality devices, and autonomous driving. These intelligent devices require accurate autonomous positioning and navigation in complex indoor and outdoor environments. To adapt to the needs of different application scenarios, machine vision positioning technology is showing diversified development trends, and its positioning accuracy and reliability in complex dynamic environments are of great significance for ensuring the stability of autonomous operation of intelligent devices.

[0003] Currently, machine vision localization methods primarily rely on visual information acquired from image sensors for position calculation. Image processing algorithms extract scene features and match them with reference data to determine the device's location. This image analysis-based localization method can obtain basic device position information and has been widely used in the field of machine vision localization.

[0004] However, due to the increasing complexity of application scenarios, the operating environment of intelligent devices is becoming increasingly dynamic and changeable. In practical applications, image analysis-based positioning methods struggle to suppress the interference of factors such as motion blur on positioning accuracy, and the accuracy of the positioning results is difficult to guarantee. Especially when the device is in rapid motion or in complex scenarios such as drastic changes in lighting or dynamic occlusion, the deterioration of image data quality leads to inaccurate extracted feature information, easily causing significant deviations in the positioning results or even positioning failure, thus reducing the accuracy of machine vision positioning in complex environments. Summary of the Invention

[0005] This application provides a fall prevention warning method, system, medium, and product for the protection of inspection robots, which can improve the accuracy of fall prevention warnings for inspection robots.

[0006] The first aspect of this application provides a fall prevention warning method for the protection of inspection robots, comprising: Acquire multimodal perception data of the inspection robot, wherein the inspection robot's form includes humanoid, quadrupedal, wheeled, or tracked, and the multimodal perception data includes morphological contact surface data, environmental geometric topology data, and dynamic state data; By integrating the morphological contact surface data with the environmental geometric topology data, a support polygon matching the morphology of the inspection robot is constructed, and the three-dimensional drop boundary field at the edge of the support polygon is extracted. Based on the dynamic state data, the dynamic equivalent zero-moment point trajectory of the robot is calculated, and the dynamic equivalent zero-moment point trajectory is projected into the support polygon to obtain the relative stability margin reflecting the attitude anti-overturning capability. Extract the spatiotemporal convergence rate of the dynamic equivalent zero-moment point trajectory approaching the three-dimensional fall boundary field, and combine it with the relative stability margin to construct an embodied fall protection risk tensor that reflects the unstable fall trend. The embodied fall protection risk tensor is mapped to a comprehensive early warning index, and an embodied active fall protection response command corresponding to the form of the inspection robot is triggered based on the comprehensive early warning index.

[0007] By employing the aforementioned technical solution and acquiring multimodal perception data, including morphological contact surface data, environmental geometric topology data, and dynamic state data, it is possible to comprehensively capture multi-dimensional state information of inspection robots of different shapes in complex environments. Based on this, by fusing morphological contact surface data and environmental geometric topology data to construct a support polygon matching the robot's shape, and extracting the three-dimensional fall boundary field at the edges of the support polygon, a precise characterization of fall hazard areas in the robot's actual operating environment is achieved. This enables the protection system to accurately identify the spatial boundary constraints faced by robots of different shapes. Furthermore, based on the dynamic state data, the robot's dynamic equivalent zero-moment trajectory is calculated, and this trajectory is projected into the support polygon to obtain a relative stability margin reflecting the robot's anti-tipping capability. This not only achieves a quantitative assessment of the robot's current stable state, but more importantly, by extracting the spatiotemporal convergence rate of the dynamic equivalent zero-moment trajectory approaching the three-dimensional fall boundary field, a dynamic evolutionary characterization mechanism for instability trends is established. By constructing an embodied fall protection risk tensor by combining the spatiotemporal convergence rate with the relative stability margin, this application overcomes the limitations of existing technologies that rely solely on state monitoring and lack trend prediction. It can proactively capture the evolution of stability and the trend of instability during robot movement. Thus, when the robot approaches the edge of a fall at a high speed or performs dynamic tasks in complex terrain, by mapping the embodied fall protection risk tensor to a comprehensive early warning index and triggering an embodied active fall protection response command corresponding to the robot's form, it achieves early intervention in the timing of protective measures, effectively solves the technical problem of delayed protective response, and improves the accuracy of fall protection early warning for inspection robots.

[0008] Optionally, the shape contact surface data is classified and extracted according to the shape of the inspection robot; when the shape is humanoid or quadrupedal, the foot contact coordinates of discrete landing points are extracted; when the shape is wheeled or tracked, the contour envelope coordinates of continuous ground imprints are extracted; the foot contact coordinates or the contour envelope coordinates are spatially intersected with the passable plane in the environmental geometric topology data to generate the supporting polygon; the suspended region or steep slope edge adjacent to the supporting polygon in the environmental geometric topology data is identified, and the three-dimensional spatial coordinate set of the suspended region or the steep slope edge is determined as the three-dimensional drop boundary field.

[0009] Optionally, the center of mass position coordinates, center of mass line acceleration vector, and body angular acceleration vector are extracted from the dynamic state data; the first projection component of the inertial force in the horizontal plane is calculated based on the center of mass line acceleration vector, and the second projection component of the inertial torque on the vertical axis is calculated based on the body angular acceleration vector; the first projection component and the second projection component are superimposed on the center of mass position coordinates to obtain the dynamic equivalent zero-moment point trajectory; the shortest normal distance from the current instantaneous point on the dynamic equivalent zero-moment point trajectory to each boundary of the supporting polygon is calculated; the minimum value among the shortest normal distances is taken as the relative stability margin at the current moment, and the relative stability margin is positively correlated with the anti-overturning capability.

[0010] Optionally, the projection component of the velocity vector of the dynamic equivalent zero-moment point trajectory in the direction corresponding to the three-dimensional fall boundary field is calculated and used as the spatial approximation velocity; the relative distance between the three-dimensional fall boundary field and the current equivalent zero-moment point is divided by the spatial approximation velocity to obtain the remaining collision time; the spatial approximation velocity and the reciprocal of the remaining collision time are weighted and fused to obtain the spatiotemporal convergence rate; the spatiotemporal convergence rate, the reciprocal of the relative stability margin, and the ground friction coefficient in the environmental geometric topology data are used as feature dimensions to construct the embodied fall protection risk tensor.

[0011] Optionally, the norm of the embodied fall protection risk tensor is calculated, and the norm is used as a warning index; the comprehensive warning index is compared with a preset critical fall protection threshold; when the comprehensive warning index exceeds the critical fall protection threshold, the current form of the inspection robot is detected; if the current form is humanoid or quadrupedal, a first embodied active fall protection response command is generated, and the inspection robot is controlled to perform a leg-stepping operation in a direction away from the three-dimensional fall boundary field according to the first embodied active fall protection response command, so as to reconstruct the support polygon; if the current form is wheeled or tracked, a second embodied active fall protection response command is generated, and the wheels or tracks close to the three-dimensional fall boundary field are controlled to perform differential reverse braking operation according to the second embodied active fall protection response command, and the height of the inspection robot's suspension system is reduced according to a preset ratio to reduce the height value of the center of mass position coordinate.

[0012] Optionally, the spatiotemporal convergence rate, the reciprocal of the relative stability margin, and the ground friction coefficient are extracted from the embodied fall protection risk tensor; the current working scenario of the inspection robot is detected (omitted); when the working scenario is performing inspection on the surface of a wind turbine blade, it is detected whether the ambient wind speed exceeds a preset wind speed threshold; if it does, the weight coefficient of the reciprocal of the relative stability margin is increased; when the working scenario is performing inspection on the top of an oil storage tank, it is detected whether there is oil or water on the surface of the tank; if so, the weight coefficient of the ground friction coefficient is increased; when the working scenario is performing inspection on a high-altitude platform of a substation, it is detected whether the distance between the edge of the platform and the supporting polygon is less than a preset safety distance; if so, the weight coefficient of the spatiotemporal convergence rate is increased; the spatiotemporal convergence rate, the reciprocal of the relative stability margin, and the ground friction coefficient are multiplied by their respective weight coefficients, squared, summed, and squared to obtain the norm of the embodied fall protection risk tensor.

[0013] Optionally, the inspection robot uses its visual and depth sensors to scan a path area within a preset distance range ahead, acquiring point cloud data of the environment ahead. The point cloud data is then segmented to extract the boundaries of areas with a height difference greater than a preset threshold, marking them as potential fall risk areas. The spatial coordinates of these potential fall risk areas are stored in a hazard map. During the inspection robot's path planning process, the shortest distance between the currently planned path and each potential fall risk area in the hazard map is calculated. When the shortest distance is less than a preset safety buffer distance, the boundary coordinates of the potential fall risk areas are added to the environmental geometric topology data for identification.

[0014] Secondly, embodiments of this application provide a fall prevention warning system for the protection of inspection robots. The fall prevention warning system for the protection of inspection robots includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the fall prevention warning system for the protection of inspection robots to perform the method described in the first aspect and any possible implementation thereof.

[0015] Thirdly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a fall protection warning system for the protection of an inspection robot, cause the fall protection warning system for the protection of the inspection robot to perform the method described in the first aspect and any possible implementation thereof.

[0016] Fourthly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a fall prevention warning system for the protection of an inspection robot, cause the fall prevention warning system for the protection of the inspection robot to perform the method described in the first aspect and any possible implementation thereof.

[0017] In summary, one or more technical solutions provided in this application have at least the following technical effects or advantages: By employing the aforementioned technical solution and acquiring multimodal perception data, including morphological contact surface data, environmental geometric topology data, and dynamic state data, it is possible to comprehensively capture multi-dimensional state information of inspection robots of different shapes in complex environments. Based on this, by fusing morphological contact surface data and environmental geometric topology data to construct a support polygon matching the robot's shape, and extracting the three-dimensional fall boundary field at the edges of the support polygon, a precise characterization of fall hazard areas in the robot's actual operating environment is achieved. This enables the protection system to accurately identify the spatial boundary constraints faced by robots of different shapes. Furthermore, based on the dynamic state data, the robot's dynamic equivalent zero-moment trajectory is calculated, and this trajectory is projected into the support polygon to obtain a relative stability margin reflecting the robot's anti-tipping capability. This not only achieves a quantitative assessment of the robot's current stable state, but more importantly, by extracting the spatiotemporal convergence rate of the dynamic equivalent zero-moment trajectory approaching the three-dimensional fall boundary field, a dynamic evolutionary characterization mechanism for instability trends is established. By constructing an embodied fall protection risk tensor by combining the spatiotemporal convergence rate with the relative stability margin, this application overcomes the limitations of existing technologies that rely solely on state monitoring and lack trend prediction. It can proactively capture the evolution of stability and the trend of instability during robot movement. Thus, when the robot approaches the edge of a fall at a high speed or performs dynamic tasks in complex terrain, by mapping the embodied fall protection risk tensor to a comprehensive early warning index and triggering an embodied active fall protection response command corresponding to the robot's form, it achieves early intervention in the timing of protective measures, effectively solves the technical problem of delayed protective response, and improves the accuracy of fall protection early warning for inspection robots. Attached Figure Description

[0018] Figure 1 This is a schematic flowchart of a fall prevention warning method for the protection of an inspection robot disclosed in an embodiment of this application; Figure 2 This is another schematic flowchart of a fall prevention warning method for the protection of inspection robots disclosed in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a system provided in an embodiment of this application.

[0019] Explanation of reference numerals in the attached drawings: 301, Central Processing Unit; 302, Read-Only Memory; 303, Random Access Memory; 304, Bus; 305, Input / Output Interface; 306, Input Section; 307, Output Section; 308, Storage Section; 309, Communication Section; 310, Driver; 311, Removable Media. Detailed Implementation

[0020] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0021] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0022] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0023] This application provides a fall prevention early warning method for the protection of inspection robots, referring to... Figure 1 , Figure 1 This is a flowchart illustrating a fall prevention warning method for inspection robot protection provided in an embodiment of this application. The method is applied to a system, which refers to a hardware and software integrated platform capable of executing a fall prevention warning program for inspection robot protection. The system can execute a fall prevention warning program for inspection robot protection. The method includes steps 101 to 105, as follows: Step 101: Acquire multimodal perception data of the inspection robot. The inspection robot's form includes humanoid, quadrupedal, wheeled, or tracked. The multimodal perception data includes morphological contact surface data, environmental geometric topology data, and dynamic state data.

[0024] Multimodal perception data refers to a heterogeneous collection of data collected by various types of sensors, reflecting the physical characteristics of the robot itself and its environment; for example, a data stream synchronously output by a vision sensor and an inertial measurement unit. Inspection robot morphology refers to the mechanical configuration of the robot body and the type of its motion actuators; for example, a quadruped configuration relying on jointed legs or a wheeled configuration relying on axles. Morphological contact surface data refers to the geometric positional information of the robot's actuators in physical contact with the external environment; for example, the discrete coordinates of the contact points between the soles of a quadruped robot's feet and the ground. Environmental geometric topology data refers to the three-dimensional spatial structure and connectivity characteristics of the robot's surrounding working environment; for example, a point cloud map of elevation differences and steps obtained through LiDAR scanning. Dynamic state data refers to physical quantities describing the robot's motion state and forces in space; for example, the triaxial acceleration of the robot's center of mass and the triaxial angular acceleration of the robot body.

[0025] Specifically, acquiring multimodal perception data relies on an underlying hardware sensor network and data synchronization bus. Contact torque sensors and joint encoders mounted on the robot's chassis or feet collect the pose parameters of the actuators in real time, thereby calculating the morphological contact surface data. Simultaneously, a 3D LiDAR and RGB-D depth camera mounted on the robot's head perform high-frequency scanning of the terrain in front and around, acquiring a point cloud set containing spatial coordinates, normal vectors, and curvature. After filtering and plane segmentation algorithms, environmental geometric topology data is generated. For dynamic state data, a built-in high-precision inertial measurement unit (IMU) measures the angular velocity and linear acceleration of the robot in real time, and combines this with forward kinematics algorithms to derive the real-time pose of the center of mass in the global coordinate system. All sensor data undergoes strict timestamp alignment via hardware trigger signals to ensure absolute consistency of multimodal data in the time domain, providing a high-precision data source for subsequent fusion calculations.

[0026] In one possible implementation, the process of acquiring the multimodal perception data of the inspection robot includes steps 1011-1014 prior to this step, as follows: Step 1011: Use the vision and depth sensors on the inspection robot to scan the path area within a preset distance range ahead to obtain point cloud data of the environment ahead.

[0027] A vision sensor is a hardware device that captures optical images of the environment; that is, an electronic eye that acquires two-dimensional color or grayscale images; for example, an industrial-grade high-resolution color camera mounted on top of a robot. A depth sensor is a detection device that measures the physical distance from an object's surface to the sensor; that is, hardware that acquires three-dimensional spatial depth information; for example, multi-line LiDAR or a depth camera based on the time-of-flight principle. A preset distance range refers to a pre-defined effective detection space boundary; that is, the maximum scanning radius and angle that the sensor can guarantee data accuracy; for example, a fan-shaped area from 0.5 meters to 5 meters directly in front of the robot. A path area refers to the physical space covered by the robot's planned route; that is, the ground area that the robot must traverse when moving; for example, a corridor floor with a width of 1.5 meters. Point cloud data refers to a massive collection of points representing the geometry of an object's surface in three-dimensional space; that is, a spatial data structure containing the three-dimensional coordinates and color attributes of each point; for example, a stepped-shaped model composed of tens of thousands of points with three-dimensional coordinate values.

[0028] Specifically, in implementing this step, the data acquisition and processing relies on a multi-sensor hardware synchronization triggering mechanism and a spatial coordinate system registration algorithm. First, a hardware clock synchronization signal simultaneously triggers the vision sensor and depth sensor, ensuring that the acquired 2D color image and 3D depth image are perfectly aligned in timestamps. Next, the raw depth values ​​obtained by the depth sensor scanning a path region within a preset distance range are extracted. Using a pre-calibrated camera intrinsic parameter matrix, each pixel in the 2D image is converted into a coordinate point in 3D space. The specific calculation process is as follows: subtract the horizontal coordinate value of the camera's principal point from the horizontal coordinate value of the current pixel, divide the difference by the camera's horizontal focal length, and multiply the quotient by the corresponding depth value to obtain the horizontal physical coordinates of the point in 3D space; similarly, subtract the vertical coordinate value of the camera's principal point from the vertical coordinate value of the current pixel, divide the difference by the camera's vertical focal length, and multiply the quotient by the depth value to obtain the vertical physical coordinates in 3D space; the vertical physical coordinates in 3D space are directly obtained from the depth value. After performing the multiplication, division, addition, and subtraction operations by traversing all valid pixels, the color information from the visual sensor is appended to the corresponding three-dimensional coordinate points, ultimately generating point cloud data of the foreground environment containing rich geometric and texture information, which is then stored in a memory buffer for subsequent steps.

[0029] Step 1012: Perform ground segmentation on the point cloud data and extract the boundaries of areas with a height difference greater than a preset threshold, marking them as potential fall risk areas.

[0030] Ground segmentation refers to the algorithmic process of dividing point cloud data into ground points and non-ground points; that is, the operation of extracting a drivable plane from complex 3D data; for example, using the random sampling consensus algorithm to extract a set of points on a flat concrete surface. Height difference refers to the vertical elevation difference between adjacent physical areas; that is, the distance between two points in space along a vertical line; for example, the vertical distance of 0.2 meters between the upper and lower edges of a step. Preset threshold refers to a pre-set numerical limit used to determine the degree of danger; that is, the standard for distinguishing between safe and dangerous states; for example, a set vertical height difference limit of 0.1 meters. Area boundary refers to the geometric outline at the intersection of spaces with different physical attributes; that is, the edge line dividing safe and dangerous areas; for example, a series of continuous spatial points at the edge of a cliff. Potential fall risk area refers to the spatial range where there is a risk of the robot falling and being damaged; that is, a geographical area where the height difference exceeds the safety tolerance; for example, the area around a deep pit without guardrails.

[0031] Specifically, firstly, the point cloud data acquired in the previous step is read, and a voxel filtering algorithm is used to downsample the massive point cloud to reduce computational load. Then, a principal plane model is fitted to the point cloud using a random sample consensus algorithm, and the vertical distance from all points to this fitted plane is calculated. Points with a vertical distance less than a specific tolerance value are classified as ground point clouds, and the rest are classified as non-ground point clouds, thus completing ground segmentation. Next, for the segmented ground point cloud, Euclidean clustering is used to divide it into multiple independent spatial blocks. The average vertical elevation of adjacent spatial blocks is calculated. A subtraction operation is performed, subtracting the average elevation of one block from the average elevation of the adjacent blocks, and the absolute value of the result is taken to obtain the height difference between the two blocks. This height difference is compared with a preset threshold. If the height difference is greater than the preset threshold, it is determined that there is a fall hazard at the boundary between the two blocks. Using edge detection operators or convex hull algorithms, the edge point cloud set on the side with higher elevation at the intersection is extracted, the three-dimensional coordinate values ​​of these edge points are obtained, the spatial range surrounded or delineated by these boundary points is formally marked as a potential fall risk area, and the boundary point set data of the area is output.

[0032] Step 1013: Store the spatial coordinates of potential fall risk areas in the danger zone map; during the inspection path planning process of the inspection robot, calculate the shortest distance between the current planned path and each potential fall risk area in the danger zone map.

[0033] Spatial coordinates refer to the numerical combination that determines the absolute position of a physical point in three-dimensional space; that is, positioning data that includes positional information in the horizontal, vertical, and longitudinal directions; for example, the three-dimensional positional value with the robot's starting point as the origin. A hazard map is a digital layer that specifically records impassable areas in the environment where there is a risk of falling; that is, a blacklist spatial database used for safe navigation; for example, a two-dimensional grid map marking the locations of all stairwells and deep pits. Inspection path planning refers to the algorithmic process of calculating the optimal route from the current starting point to the target endpoint; that is, finding a trajectory that avoids obstacles and satisfies kinematic constraints; for example, a smooth curve connecting various inspection points generated using a heuristic search algorithm. A planned path refers to the robot's expected trajectory calculated by an algorithm; that is, a movement route composed of a series of continuous spatial waypoints; for example, a polyline formed by connecting fifty waypoints spaced 0.1 meters apart. The shortest distance is the minimum length among all lines connecting two spatial geometric entities; that is, the most stringent indicator for measuring the proximity of a path to a hazard area; for example, the straight-line length of 0.5 meters from a point on the path to the edge of a pit.

[0034] Specifically, this step mainly involves the persistent storage of spatial data and the calculation of distance metrics between discrete geometric point sets. First, the boundary point set of potential fall risk areas extracted in the previous step is received. Its three-dimensional spatial coordinate values ​​are serialized according to a predetermined data structure format and written to the danger zone map database of the robot navigation system. This map is independent of the conventional static obstacle map and is specifically used for high-priority fall prevention logic. After the inspection path planning module generates a currently planned path consisting of multiple discrete waypoints, distance verification calculation is initiated. Each discrete waypoint on the planned path is traversed, and for the currently selected waypoint, all boundary points of potential fall risk areas recorded in the danger zone map are traversed again. The spatial straight-line distance between waypoints and boundary points is calculated as follows: Subtract the lateral coordinate of the boundary point from the lateral coordinate of the waypoint, and square the difference; subtract the longitudinal coordinate of the boundary point from the longitudinal coordinate of the waypoint, and square the difference; subtract the perpendicular coordinate of the boundary point from the perpendicular coordinate of the waypoint, and square the difference; sum these three squared values, and finally take the square root of the sum to obtain the spatial distance between the waypoint and the boundary point. Record the distance values ​​calculated from all waypoints and all boundary points. Using a sorting or extreme value search algorithm, extract the minimum value among these values; this minimum value is the shortest distance between the currently planned path and each potential fall risk area.

[0035] Step 1014: When the shortest distance is less than the preset safety buffer distance, the boundary coordinates of the potential fall risk area are added to the environmental geometric topology data as identification objects.

[0036] Preset safety buffer distance refers to an additional safety distance set in advance to prevent accidents caused by control errors; that is, the virtual protective thickness defined outside the physical boundary; for example, a collision avoidance warning distance set at 0.8 meters. Boundary coordinates refer to the spatial point position data that constitutes the outline edge of the danger zone; that is, the set of coordinates that accurately describes the geometry of the fall edge; for example, the three-dimensional point set that outlines the four edges of a rectangular pit. Environmental geometric topology data refers to structured information describing the terrain undulations and connectivity of the robot's environment; that is, a digital map containing physical environmental attributes and spatial relationships; for example, a grid elevation map containing slope, roughness, and node connection relationships. Identified objects refer to target entities that need to be given special attention and processing in environmental perception and navigation algorithms; that is, the trigger source that triggers specific obstacle avoidance or fall prevention logic; for example, the edge of a high-drop staircase that the system marks as something that must be avoided.

[0037] Specifically, the shortest distance calculated in the previous step is extracted first, and the preset safety buffer distance is read from the system configuration file. A direct comparison operation is performed to determine if the shortest distance is less than the preset safety buffer distance. If the shortest distance is greater than or equal to the preset safety buffer distance, it indicates that the currently planned inspection path is far enough from the fall edge and is in an absolutely safe state, requiring no additional operation. If the shortest distance is less than the preset safety buffer distance, it indicates that the robot is highly likely to cross the safety boundary and fall due to physical factors such as wheel slippage or control delay while traveling along the current path. In this case, the environmental topology data update mechanism is immediately triggered. All boundary coordinates of the potential fall risk area are extracted and mapped to the coordinate system of the environmental geometric topology data. Attribute labels representing extremely high risk are written to the corresponding grid or node positions in the environmental geometric topology data, and these boundary coordinates are materialized as insurmountable virtual obstacle walls. Through this data supplementation operation, the potential fall risk area officially becomes a mandatory identification target for the underlying motion control and local obstacle avoidance algorithms, forcing the navigation module to recalculate and generate a new path away from the boundary, thereby fundamentally preventing the occurrence of fall accidents.

[0038] Step 102: Integrate the morphological contact surface data with the environmental geometric topology data to construct a support polygon that matches the shape of the inspection robot, and extract the three-dimensional drop boundary field at the edge of the support polygon.

[0039] A support polygon is a convex hull region formed by the horizontal projection of all effective contact points between the robot and the ground. This region determines the robot's gravitational stability range under static conditions; for example, the quadrilateral region formed by connecting the contact points of the four legs of a quadruped robot. A three-dimensional fall boundary field refers to the set of spatial coordinates outside the support polygon that contains abrupt elevation changes or unsupported suspended areas exceeding the robot's traversal capabilities; for example, the cliff-like outline of the edge of a high-altitude inspection platform.

[0040] Specifically, the contact surface data is first extracted based on the robot's specific shape: for humanoid or quadrupedal shapes, the foot contact coordinates of discrete landing points are extracted; for wheeled or tracked shapes, the contour envelope coordinates of continuous ground imprints are extracted. Then, the extracted foot contact coordinates or contour envelope coordinates are spatially intersected with the passable planes in the environmental geometry and topology data, and a convex hull algorithm (such as Graham's scan) is used to generate the support polygon at the current moment. When extracting the three-dimensional fall boundary field, a preset detection distance is extended outward from each edge of the support polygon, and the elevation gradient of the environmental geometry and topology data within this extended area is calculated. Areas with downward elevation gradients and absolute gradient values ​​greater than a preset safety threshold are identified as steep slope edges, or areas with no point cloud reflection signals are identified as suspended areas. Finally, the three-dimensional spatial coordinates of these suspended areas or steep slope edges are clustered and smoothed to establish a continuous three-dimensional fall boundary field, serving as an absolute danger reference system for subsequent fall prevention warnings.

[0041] In one possible implementation, morphological contact surface data and environmental geometric topology data are fused to construct a support polygon that matches the shape of the inspection robot, and the three-dimensional drop boundary field at the edge of the support polygon is extracted. Specifically, steps 1021-1023 are included, as follows: Step 1021: Classify and extract the shape contact surface data according to the shape of the inspection robot; when the shape is humanoid or quadrupedal, extract the foot contact coordinates of discrete foot points; when the shape is wheeled or tracked, extract the contour envelope coordinates of continuous ground imprints.

[0042] The morphology of an inspection robot refers to the mechanical configuration of its body; that is, the type of mechanical structure the robot relies on for movement; for example, a quadrupedal configuration or a wheeled configuration. Morphological contact surface data refers to the geometric positional information of the robot's actuators in physical contact with the external environment; that is, data reflecting the specific location where the robot steps on or presses against the ground; for example, the discrete coordinate points of the foot's contact with the ground. Foot contact coordinates refer to the discrete three-dimensional spatial points of the end effector of a legged robot in contact with the supporting surface; that is, the specific position of the mechanical leg's footplate in the global coordinate system; for example, the X, Y, and Z coordinates of the four footpads of a quadruped robot. Contour envelope coordinates refer to the set of outer boundary points of the grounding imprint of a continuous contact robot; that is, the outer edge contour of the continuous contact surface formed by the wheels or tracks pressing against the ground; for example, the coordinates of the four vertices of the grounding rectangle of the two tracks of a tracked vehicle.

[0043] Specifically, the underlying control module identifies the current inspection robot's shape category based on a preset hardware identifier and executes differentiated data extraction algorithms. When a humanoid or quadrupedal shape is identified, the extraction module calls the forward kinematics solver to read the real-time encoder angles of each joint. Combined with pre-calibrated link length parameters, it calculates the relative position of each foot in the body coordinate system. Subsequently, based on the contact torque values ​​fed back by the multi-dimensional force sensors installed on the soles of the feet, it filters out the feet whose normal contact force is greater than a set load-bearing threshold and extracts their corresponding three-dimensional coordinates as the foot contact coordinates of discrete foot placement points. When a wheeled or tracked shape is identified, the extraction module reads fixed mechanical parameters such as the chassis wheelbase, wheel track, and wheel radius or track ground contact length. Combined with the compression displacement sensor data of the suspension system, it calculates the actual contact area between the wheel or track and the ground in the current posture. Then, it samples at equal intervals along the geometric outer edge of the contact area to extract a series of edge points surrounding the ground imprint, forming the contour envelope coordinates of the continuous ground imprint. This classification and extraction mechanism ensures that the data source for subsequent stability analysis can accurately match the physical grounding characteristics of robots with different configurations.

[0044] Step 1022: Spatial intersection of the foot contact coordinates or contour envelope coordinates with the passable plane in the environmental geometric topology data to generate a supporting polygon.

[0045] Environmental geometric topology data refers to the three-dimensional spatial structure and connectivity characteristics of the robot's surrounding operating environment; that is, a three-dimensional digital model of the external physical world constructed through sensor scanning; for example, a point cloud map with normal vector information acquired by LiDAR. A traversable plane refers to a continuous geometric surface in the environmental geometric topology data whose slope and roughness meet the requirements for safe robot movement; that is, a local ground surface where the robot can stand or move stably; for example, a flat cement surface with an inclination angle of less than 15 degrees. Spatial intersection refers to the spatial analytical process of calculating the overlapping or intersecting parts between different geometric entities in a three-dimensional coordinate system; that is, solving for the mathematical intersection between the robot's contact points and the environmental ground; for example, calculating the coordinates of the intersection point of a vertical line and a horizontal plane. A supporting polygon refers to the convex hull region formed by all effective contact points between the robot and the ground on the horizontal projection plane; that is, the geometric boundary that determines the robot's gravitational stability range; for example, the quadrilateral formed by the lines connecting the contact points of the four legs of a quadruped robot.

[0046] Specifically, the data processing unit first performs a coordinate system alignment operation. Using the pose estimation results from the inertial measurement unit and odometry, it transforms the extracted foot contact coordinates or contour envelope coordinates from the local coordinate system of the robot to a global world coordinate system unified with the environmental geometric topology data. Next, in the global coordinate system, it extracts a subset of point clouds located directly below and around the robot from the environmental geometric topology data, and fits the local traversable plane equation using the Random Sample Consensus (RANSAC) algorithm. Subsequently, it performs spatial intersection calculation: the transformed foot contact coordinates or contour envelope coordinates are projected perpendicularly along the direction of gravity onto the geometric surface determined by the traversable plane equation to obtain the coordinates of the projected intersection points. Finally, all projected intersection point coordinates are input into a convex hull generation algorithm (such as the Graham scan algorithm or the QuickHull algorithm). This algorithm uses these intersection points as the input set, finds the boundary of the smallest convex polygon that can enclose all intersection points, and connects these peripheral extreme points in sequence to finally generate a support polygon that strictly matches the current robot shape and terrain, serving as the absolute geometric benchmark for subsequent evaluation of whether the robot has overturned.

[0047] Step 1023: Identify the suspended regions or steep slope edges adjacent to the supporting polygons in the environmental geometric topology data, and determine the three-dimensional spatial coordinate set of the suspended regions or steep slope edges as the three-dimensional drop boundary field.

[0048] Suspended areas refer to blank spaces in the environmental geometry and topology data that lack effective physical support; that is, areas where sensors cannot detect reflective surfaces or where depth values ​​exceed safe ranges; for example, the bottomless space outside an aerial work platform. Steep slope edges refer to geometric boundaries where the environmental elevation changes drastically and the slope exceeds the robot's maximum traversal capability; that is, the boundary between flat ground and a dangerous abyss; for example, the edge of the first step downwards from the top of a staircase. A three-dimensional spatial coordinate set refers to a data group composed of multiple discrete points with X, Y, and Z axis position information; that is, a mathematical lattice used to describe the shape of a specific physical area; for example, a series of three-dimensional coordinate points depicting the outline of a cliff. A three-dimensional fall boundary field refers to a continuous spatial boundary model existing around the supporting polygon that is highly likely to cause the robot to fall; that is, an absolutely insurmountable safety red line area defined by the system; for example, the coordinate set of a virtual fall arrest fence around a substation maintenance platform.

[0049] Specifically, the environment perception module uses the generated supporting polygon as a reference and extends horizontally outward by a preset detection radius (e.g., 1.5 meters) to delineate the Region of Interest (ROI) for edge detection. Within this ROI, dense point clouds from the environmental geometric topology data are extracted, and the local normal vector and elevation gradient value of each point cloud are calculated. For the identification of steep slope edges, the algorithm traverses the point clouds within the ROI, calculates the elevation difference between adjacent points, and extracts continuous point clouds with vertically downward elevation gradients and absolute gradient values ​​greater than a preset drop threshold (e.g., a drop greater than 0.3 meters). For the identification of suspended regions, the algorithm analyzes the ray projection model of the depth camera or LiDAR, identifies regions where the ray return distance reaches infinity or exceeds the sensor's maximum range, and marks these regions as suspended boundaries by back-projecting the intersection points of the ray and the edge of the passable plane. Subsequently, the extracted steep slope edge point clouds and the intersection points of the suspended boundaries are spatially clustered to remove isolated noise points. A B-spline curve fitting algorithm is used to smoothly connect the clustered effective point sets, generating one or more continuous three-dimensional spatial curves. Finally, the set of three-dimensional spatial coordinates contained in these curves is established as a three-dimensional drop boundary field, providing accurate dangerous target points for subsequent calculation of spatiotemporal convergence rate.

[0050] Step 103: Based on the dynamic state data, calculate the dynamic equivalent zero-moment point trajectory of the robot, and project the dynamic equivalent zero-moment point trajectory into the support polygon to obtain the relative stability margin reflecting the attitude anti-overturning capability.

[0051] The dynamic equivalent zero-moment point (ZMP) trajectory refers to the continuous movement path along the time axis of the point where the sum of the horizontal overturning moments generated by the ground reaction forces is zero during the robot's dynamic motion. For example, when the robot accelerates forward, the ZMP will move towards the rear edge of the supporting polygon due to inertial forces. The relative stability margin refers to the shortest geometric distance from the dynamic equivalent zero-moment point to each physical boundary of the supporting polygon, used to quantify the robot's ability to resist external disturbances without tipping over. For example, if the ZMP is 10 centimeters from the edge of the platform, this 10 centimeters represents the current stability margin.

[0052] Specifically, firstly, the center of mass position coordinates, center of mass linear acceleration vector, and body angular acceleration vector are extracted from the dynamic state data. Based on the Newton-Euler equations, the first projection component of the inertial force in the horizontal plane is calculated by multiplying the center of mass linear acceleration vector by the robot's total mass; the second projection component of the inertial torque in the vertical axis is calculated by multiplying the body angular acceleration vector by the robot's moment of inertia matrix. The equivalent offsets generated by the first and second projection components are superimposed on the static center of mass position coordinates to calculate the dynamic equivalent zero-moment point coordinates at the current moment, and recorded in a time series to form a ZMP trajectory. Next, this ZMP trajectory is projected in real-time onto the plane containing the supporting polygon constructed in step 102. Using the shortest distance formula from a point to a line segment, the normal distance from the current instantaneous ZMP coordinates to each boundary line segment of the supporting polygon is calculated. Comparing all calculated normal distances, the minimum value is extracted as the relative stability margin at the current moment. The larger the margin value, the closer the ZMP is to the center of the supporting polygon, and the stronger the robot's resistance to tipping over; conversely, the smaller the margin value, the closer the robot is to the center of the supporting polygon, and the stronger its resistance to tipping over.

[0053] In one possible implementation, based on dynamic state data, the dynamic equivalent zero-moment point trajectory of the robot is calculated, and the dynamic equivalent zero-moment point trajectory is projected onto the support polygon to obtain the relative stability margin reflecting the attitude's anti-tipping capability. Specifically, this includes steps 1031-1034, as follows: Step 1031: Extract the center of mass position coordinates, center of mass line acceleration vector, and body angular acceleration vector from the dynamic state data.

[0054] Dynamic state data refers to the set of physical quantities describing the robot's motion state and forces in space; that is, the readings and calculation results of the underlying sensors reflecting the robot's dynamic characteristics; for example, the three-axis acceleration and angular velocity flow output by the inertial measurement unit. Center of mass position coordinates refer to the three-dimensional spatial position of the robot's overall center of mass in the global reference coordinate system; that is, the geometric point representing the robot's translational inertial center; for example, the center of mass point with coordinates (1.2, 0.5, 0.8) in the world coordinate system. Center of mass linear acceleration vector refers to the rate of change and direction of the robot's center of mass's linear velocity per unit time; that is, a three-dimensional vector describing the intensity of the center of mass's translational state; for example, an acceleration vector with a magnitude of 2.0 m / s² along the positive X-axis. Body angular acceleration vector refers to the rate of change and direction of the robot's overall angular velocity when rotating around its center of mass; that is, a three-dimensional vector describing the intensity of the body's rotational attitude; for example, an angular acceleration vector with a magnitude of 1.5 radians / s² around the Z-axis.

[0055] Specifically, the data extraction and processing relies on a multi-sensor fusion filtering algorithm and a robot dynamics model. First, the main control unit reads the raw data stream from the high-precision inertial measurement unit (IMU) installed at the core of the robot's torso via a high-speed real-time bus. A pre-defined Kalman filter algorithm is used to perform high-frequency noise reduction and zero-bias compensation on the raw accelerometer and gyroscope data, obtaining high-precision three-axis accelerations and three-axis angular velocities. Then, the first-order time-difference derivative of the three-axis angular velocities is performed to directly calculate the body angular acceleration vector. For the extraction of the center of mass coordinates, the extraction module calls the mass distribution parameter matrix of each link and the current angle values ​​of each joint encoder, which are pre-stored in memory. Using a multibody kinematics forward solving algorithm, the local coordinates of each link's center of mass in the robot's base coordinate system are calculated, and a weighted average is performed based on the mass of each link to obtain the local position of the overall robot's center of mass in the base coordinate system. Next, combining the real-time pose matrix (including rotation and translation vectors) of the aircraft base in the global coordinate system output by visual odometry or laser SLAM algorithms, the local centroid position is mapped to the global coordinate system through homogeneous coordinate transformation to obtain the absolute centroid position coordinates. Simultaneously, the second-order time-difference derivative of these centroid position coordinates is calculated, or coordinate system rotation projection is performed using IMU linear acceleration data to accurately extract the centroid linear acceleration vector in the global coordinate system.

[0056] Step 1032: Calculate the first projection component of the inertial force in the horizontal plane based on the acceleration vector along the center of mass, and calculate the second projection component of the inertial torque on the vertical axis based on the angular acceleration vector of the body.

[0057] Inertial force refers to the reaction force generated by an object's mass against changes in its state of motion; that is, the virtual force caused by the accelerated motion of a robot's center of mass; for example, the forward thrust generated by a robot during sudden braking. The first projection component in the horizontal plane refers to the vector projection of the aforementioned inertial force onto a two-dimensional plane parallel to the gravitational equipotential surface; that is, the effect of the inertial force in the X and Y axis directions; for example, a force component of 50 Newtons in magnitude and directed along the horizontal X-axis. Inertial torque refers to the reaction torque generated by an object's moment of inertia against changes in its rotational state; that is, the virtual torque caused by the angular acceleration of a robot's body rotation; for example, the torque resisting rotation generated when a robot makes a sharp turn. The second projection component on the vertical axis refers to the vector projection of the aforementioned inertial torque onto a one-dimensional axis parallel to the direction of gravity; that is, the effect of the inertial torque in the Z-axis direction; for example, a torque component of 10 Newton-meters in magnitude about the vertical Z-axis.

[0058] Specifically, firstly, the first projection component of the inertial force in the horizontal plane is calculated, and the obtained centroid acceleration vector is extracted and represented as a three-dimensional column vector [a x a y a z] T Read the robot's total mass scalar m. According to Newton's second law, correlate the total mass m with the components a of the acceleration vector along the horizontal X and Y axes. x and a y Performing scalar multiplication, we obtain the horizontal inertial force component Fx = −m⋅a. x And Fy=−m⋅a y (The negative sign indicates that the direction of the inertial force is opposite to the direction of acceleration). These two components together constitute the first projection component of the inertial force in the horizontal plane. Next, the second projection component of the inertial torque on the vertical axis is calculated, and the obtained body angular acceleration vector is extracted and represented as a three-dimensional column vector [α]. x α y α z ] T Read the third-order rotational inertia tensor matrix I of the robot relative to its center of mass in its current posture. According to Euler's equations, calculate the product of the angular acceleration vector and the rotational inertia tensor matrix to obtain the three-dimensional inertial torque vector M = −I⋅[α]. x α y α z ] T Extract the component M of the three-dimensional moment of inertia vector along the vertical Z-axis. z This component is the second projection component of the inertial torque on the vertical axis. These two projection components accurately quantify the horizontal shearing tendency and the torsional tendency around the vertical axis exerted by the robot on the ground during dynamic motion, providing basic physical parameters for subsequent calculation of the zero-moment point offset.

[0059] Step 1033: Superimpose the first projection component and the second projection component onto the centroid position coordinates to obtain the dynamic equivalent zero-moment point trajectory.

[0060] The first projection component refers to the force vector of the inertial force calculated above on the horizontal X and Y axes; that is, the mechanical component that causes the robot to have a tendency to translate and slide horizontally; for example, a 30 Newton inertial force along the negative X-axis. The second projection component refers to the torque vector of the inertial moment calculated above on the vertical Z-axis; that is, the torque component that causes the robot to have a tendency to yaw and rotate around the vertical axis; for example, a 5 Newton-meter inertial moment around the Z-axis. The center of mass position coordinates refer to the three-dimensional spatial position of the robot's overall center of mass in the global reference coordinate system; that is, the geometric point representing the robot's translational inertial center; for example, three-dimensional coordinates (1.2, 0.5, 0.8). The dynamic equivalent zero-moment point trajectory refers to the continuous movement path of the point of action on the time axis where the sum of the horizontal overturning moments generated by the ground reaction force is zero during the robot's continuous movement; that is, the line connecting the ground force center points reflecting the robot's dynamic equilibrium state; for example, the curve formed by the ground force center point moving forward continuously under the footplate when the robot walks.

[0061] Specifically, the calculation in this step aims to map the inertial forces and moments in space onto the ground contact surface using the principle of mechanical equivalence. Let the coordinates of the center of mass be (x... c y c , z c The scalar of gravitational acceleration is g. First, calculate the offset of the zero-moment point coordinates caused by the horizontal acceleration of the center of mass (i.e., the first projected component). According to the principle of torque balance, the horizontal inertial force F at the center of mass is... x and F y This will generate an overturning moment on the ground, which requires ground support (mainly composed of gravity m⋅g and vertical inertial force m⋅a). z The components generate a counter-torque at the offset position to balance it. Therefore, the formula for calculating the offset in the X-axis direction is Δx1 = −(z) c ⋅a x ) / (g+a z The formula for calculating the offset in the Y-axis direction is Δy1 = −(z) c ⋅a y ) / (g+a z Next, the offset of the zero-moment point coordinates caused by the body's angular acceleration (i.e., the second projection component) is calculated. The rate of change of angular momentum (i.e., the torque M) generated by angular acceleration is also calculated. x and M y It also requires ground support for balance. The formula for calculating the additional offset in the X-axis direction is Δx2 = −M y / m⋅(g+a z The formula for calculating the additional offset in the Y-axis direction is Δy2=M. x / m⋅(g+a z Finally, perform the superposition operation: combine the horizontal component (x, y) of the centroid position coordinates. c y c Adding the two sets of offsets algebraically to obtain the planar coordinates of the current dynamic equivalent zero moment point (ZMP): p x =x c +Δx1+Δx2, p y =y c +Δy1+Δy2. The control system executes the above calculation cyclically according to a fixed control cycle (e.g., every 10 milliseconds), connects the discrete ZMP coordinate points calculated at different timestamps in chronological order, and performs smoothing using a cubic spline interpolation algorithm to finally generate a continuous dynamic equivalent zero-moment point trajectory.

[0062] Step 1034: Calculate the shortest normal distance from the current instantaneous point on the dynamic equivalent zero-moment point trajectory to each boundary of the supporting polygon; take the minimum value among the shortest normal distances as the relative stability margin at the current moment, and the relative stability margin is positively correlated with the overturning resistance.

[0063] The current instantaneous point on the dynamic equivalent zero-moment trajectory refers to the exact two-dimensional coordinates of the robot's equivalent ground force center within the current calculation cycle; that is, the unique physical point representing the robot's dynamic equilibrium state at this moment; for example, the ZMP point with coordinates (1.3, 0.6) at the current moment. The boundaries of the supporting polygon refer to the set of outer line segments connecting the robot's effective contact points with the ground; that is, the geometric border defining the robot's safe and stable area; for example, the four sides of the rectangle formed by the lines connecting the four legs of a quadruped robot. The shortest normal distance refers to the length of the perpendicular line segment drawn from the current instantaneous point to a certain boundary line segment of the supporting polygon, where the foot of the perpendicular falls on the line segment; that is, the vertical spatial distance from the ZMP point to a certain boundary; for example, the vertical distance from the ZMP point to the right boundary is 0.15 meters. Relative stability margin refers to the global minimum of the shortest normal distances from the current instantaneous point to all boundaries of the supporting polygon; that is, a scalar value that quantifies how much leeway the robot has before it overturns and becomes unstable; for example, if the calculated distances to the four sides are 0.2 meters, 0.3 meters, 0.1 meters, and 0.4 meters respectively, then the relative stability margin is 0.1 meters. Anti-tipping capability refers to the physical limit of a robot's ability to resist external disturbances or internal inertial forces without tipping over; that is, the ability to maintain an upright position; for example, the robot's ability to withstand a lateral thrust of 50 Newtons without tipping over.

[0064] Specifically, first, extract the calculated coordinates P of the current instantaneous ZMP point. zmp (p x p y Simultaneously, the vertex coordinate sequence of the supporting polygon is extracted, and adjacent vertices are connected in counterclockwise order to form N boundary segments. Let the coordinates of the two endpoints of the i-th boundary segment be A and B, respectively. i (x a y a ) and B i (x b y b For each boundary segment, calculate vector A. i B i Sum vector A i P zmp Calculate the projection scale parameter t = (A / T) using the dot product formula. i B i ⋅A i P zmp ) / (A i B i ⋅A i Bi ). Determine the position of the foot of the perpendicular according to the value of parameter t: If t ≤ 0, it means that the foot of the perpendicular is outside the A end of the extended line segment. At this time, the shortest distance is the Euclidean distance from point P to endpoint A i ; If t ≥ 1, it means that the foot of the perpendicular is outside the B end of the extended line segment. The shortest distance is the Euclidean distance from point P to endpoint B zmp ; If 0 < t < 1, it means that the foot of the perpendicular strictly falls on the line segment AB i . At this time, use the cross - product formula to calculate the shortest normal distance di = ∣AP × AB∣ / ∣AiBi∣. Traverse all N boundaries of the support polygon to calculate N sets of distance values {d1, d2,..., d i}. Sort this set and extract the minimum value d zmp , and assign it as the relative stability margin at the current moment. This relative stability margin has a strictly positive correlation with the anti - tipping ability of the robot: when d i is large, the ZMP point is deep inside the support polygon, the robot's posture is extremely stable, and the anti - tipping ability is strong; when d i approaches zero, the ZMP point approaches the edge of the support polygon, the robot is in a critical state of about to tip over, and the anti - tipping ability is extremely weak; if d i becomes negative (i.e., the ZMP point moves out of the support polygon), it indicates that the robot has undergone irreversible physical tipping. i P zmp ×A i B i ∣ / ∣AiBi∣. Traverse all N boundaries of the support polygon to calculate N sets of distance values {d1, d2,..., d N}. Sort this set and extract the minimum value d min , and assign it as the relative stability margin at the current moment. This relative stability margin has a strictly positive correlation with the anti - tipping ability of the robot: when d min is large, the ZMP point is deep inside the support polygon, the robot's posture is extremely stable, and the anti - tipping ability is strong; when d min approaches zero, the ZMP point approaches the edge of the support polygon, the robot is in a critical state of about to tip over, and the anti - tipping ability is extremely weak; if d min becomes negative (i.e., the ZMP point moves out of the support polygon), it indicates that the robot has undergone irreversible physical tipping.

[0065] Step 104: Extract the spatio - temporal convergence rate at which the dynamic equivalent zero - moment point trajectory approaches the three - dimensional fall boundary field, and combine it with the relative stability margin to construct an embodied anti - fall risk tensor reflecting the instability and fall trend.

[0066] The spatio - temporal convergence rate refers to the comprehensive urgency of the dynamic equivalent zero - moment point approaching the dangerous fall boundary in both the spatial distance and time dimensions; for example, the robot is not only only 0.5 meters away from the edge and is rushing towards the edge at a speed of 2 meters per second. This combination of high speed and short distance represents a very high spatio - temporal convergence rate. The embodied anti - fall risk tensor is a multi - dimensional mathematical structure that structurally encapsulates the internal stability state related to the robot's body shape and the degree of external environmental danger; for example, a one - dimensional matrix containing the convergence rate, the reciprocal of the stability margin, and the ground friction coefficient.

[0067] Specifically, when calculating the spatiotemporal convergence rate, the first time derivative of the dynamic equivalent zero-moment point trajectory is first performed to obtain the velocity vector of the ZMP. The projection component of this velocity vector in the direction pointing to the nearest point of the 3D fall boundary field is calculated and defined as the spatial approximation velocity. Subsequently, the Euclidean distance between the nearest point of the 3D fall boundary field and the current ZMP coordinates is calculated, and this relative distance is divided by the spatial approximation velocity to calculate the time remaining after impact (TTC). To eliminate dimensional differences, the spatial approximation velocity is normalized, and the reciprocal of the time remaining after impact is taken (the larger the reciprocal, the more urgent the time). The two are then linearly weighted and fused according to a preset fixed weight coefficient to obtain the spatiotemporal convergence rate. When constructing the embodied fall protection risk tensor, the spatiotemporal convergence rate calculated above is extracted as the first feature dimension, the reciprocal of the relative stability margin is extracted as the second feature dimension, and the ground friction coefficient estimated by visual material recognition or contact force feedback from environmental geometric topology data is extracted as the third feature dimension. By arranging these three feature dimensions in a fixed order, a three-dimensional embodied fall risk tensor is constructed, which comprehensively and quantitatively characterizes the physical trend of current robot instability and fall.

[0068] In one possible implementation, the spatiotemporal convergence rate of the dynamic equivalent zero-moment point trajectory approaching the three-dimensional fall boundary field is extracted, and combined with the relative stability margin, an embodied fall protection risk tensor reflecting the unstable fall trend is constructed. Specifically, steps 1041-1043 are included, as follows: Step 1041: Calculate the projection component of the velocity vector of the dynamic equivalent zero-moment point trajectory in the direction corresponding to the three-dimensional drop boundary field, and use it as the spatial approximation velocity.

[0069] The velocity vector of the dynamic equivalent zero-moment point trajectory refers to the rate of displacement change and direction of the robot's equivalent center of force per unit time; that is, a physical quantity describing the speed and orientation of the dynamic equilibrium point's movement on the ground; for example, a movement speed of 0.5 meters per second in a forward direction. The three-dimensional fall boundary field refers to the set of spatial geometric boundaries in the robot's operating environment that characterize dangerous areas such as height differences or cliffs; that is, a three-dimensional virtual protective wall defining the robot's safe operating range; for example, the boundary surface formed by the extension of a one-dimensional line segment formed by the edge of a staircase in three-dimensional space. The projection component refers to the orthogonal projection length of a vector onto another vector in a specific direction; that is, the effective magnitude of the velocity vector in a specific direction; for example, the effective component of a velocity in the northeast direction in the east direction of 0.3 meters per second. The spatial approach velocity refers to the substantial rate at which the robot's dynamic equilibrium point approaches the dangerous boundary; that is, a scalar indicator measuring the degree of increase in the robot's fall risk; for example, approaching the edge of a cliff at an effective rate of 0.4 meters per second.

[0070] Specifically, in implementing this step, the calculation process relies entirely on kinematic difference and spatial vector dot product operations under discrete-time series. First, the three-dimensional coordinates of the dynamic equivalent zero-moment point in the current control cycle and the dynamic equivalent zero-moment point in the previous control cycle are extracted. The differences between the current coordinates and the previous cycle coordinates in each dimension are divided by the fixed time interval of the control cycle, thus obtaining the velocity vector of the dynamic equivalent zero-moment point trajectory containing three-dimensional components. Next, in the three-dimensional drop boundary field, a spatial nearest neighbor search algorithm is used to find the boundary point with the shortest spatial straight-line distance to the current equivalent zero-moment point. A spatial direction vector is constructed from the current equivalent zero-moment point, pointing towards this shortest-distance boundary point. Each dimensional component of this direction vector is divided by the total spatial length of the vector to complete the normalization process, obtaining the unit direction vector. Finally, using the dot product operation of spatial vectors, the previously obtained velocity vector is multiplied dimension-by-dimensionally with the unit direction vector and summed to calculate the projection component of the velocity vector onto the unit direction vector. This projection component is the spatial approximation velocity. If the calculated value is less than or equal to zero, it indicates that the dynamic equivalent zero moment point is moving away from the fall boundary or parallel to the boundary. In this case, the spatial approach velocity is directly assigned to zero. If the value is greater than zero, it indicates that it is approaching the danger boundary. The calculated value is directly retained and entered into the subsequent risk assessment steps.

[0071] Step 1042: Divide the relative distance between the three-dimensional drop boundary field and the current equivalent zero moment point by the spatial approximation velocity to obtain the remaining collision time; weight and fuse the spatial approximation velocity and the reciprocal of the remaining collision time to obtain the spatiotemporal convergence rate.

[0072] Relative distance refers to the straight-line spatial length between the current equivalent zero-moment point and the nearest point on the three-dimensional fall boundary field; that is, the physical distance between the robot's current position and the danger edge; for example, the distance from the equivalent center of force to the edge of the step is 1.2 meters. Collision remaining time refers to the time required for the equivalent zero-moment point to reach the fall boundary while maintaining the current spatial approach velocity; that is, the time margin for predicting a robot's fall risk; for example, predicting that it will cross the safety boundary in 2.5 seconds. The reciprocal is a non-zero value whose product is one; that is, the inverse proportional quantification form in mathematics; for example, the reciprocal of the collision remaining time of two seconds is 0.5 Hz. Weighted fusion refers to the linear combination of multiple variables with specific weight coefficients; that is, a mathematical processing method that comprehensively considers the importance of multiple factors; for example, adding the velocity and the reciprocal of time at a ratio of 60% and 40%, respectively. Spatiotemporal convergence rate refers to a composite risk index that integrates the speed of spatial approach and the degree of time urgency; that is, a numerical value that comprehensively quantifies the rate of deterioration of the dangerous state; for example, a calculated comprehensive risk score of 4.8.

[0073] Specifically, firstly, the spatial straight-line distance between the current equivalent zero-moment point and the nearest point on the fall boundary, obtained in the previous steps, is extracted as the relative distance. Simultaneously, the spatial approach velocity calculated in the previous steps is extracted. To calculate the remaining collision time, a basic division operation is performed, using the relative distance as the dividend and the spatial approach velocity as the divisor. In the program logic implementation, a minimum error-prevention threshold needs to be set. When the spatial approach velocity is less than this threshold, the remaining collision time is forcibly set to a preset maximum value to prevent division by zero errors. Then, the reciprocal of the remaining collision time is calculated, i.e., the value is one divided by the remaining collision time. This reciprocal directly reflects the urgency of the time; a larger value indicates a faster fall. Next, two pre-calibrated dimensionless weight coefficients are read from the configuration file. The first weight coefficient corresponds to the spatial approach velocity, and the second weight coefficient corresponds to the time urgency, and the two weight coefficients must add up to one. A weighted fusion operation is performed, multiplying the spatial approach velocity by the first weight coefficient and the reciprocal of the remaining collision time by the second weight coefficient, and finally adding these two products. The final scalar result obtained from the calculation is the spatiotemporal convergence rate. The larger this rate value, the more it indicates that the robot is not only rushing towards the boundary at an extremely high speed, but also that the reaction time left for the underlying control module is extremely short, and the risk of falling is in a state of sharp increase.

[0074] Step 1043: Construct the embodied fall protection risk tensor by using the spatiotemporal convergence rate, the reciprocal of the relative stability margin, and the ground friction coefficient in the environmental geometric topology data as feature dimensions.

[0075] The reciprocal of the relative stability margin refers to the reciprocal form of the stability margin value obtained in the preceding steps; that is, the mathematical expression that amplifies the impact of a small stability margin on risk; for example, the reciprocal of a stability margin of 0.1 meters is 10. Environmental geometric topology data refers to structured information describing the terrain undulations, surface materials, and connectivity of the robot's environment; that is, a digital map containing physical environmental attributes; for example, a grid elevation map containing information such as slope and roughness. The ground friction coefficient is a physical parameter that hinders relative sliding between the robot's foot and the ground it is currently in contact with; that is, a constant reflecting the degree of slipperiness or grip; for example, the friction coefficient of rough concrete is 0.8. Feature dimensions refer to independent attribute variables used to describe a complex object or state; that is, the basic coordinate axes for constructing multidimensional mathematical models; for example, using speed, distance, and friction as three independent dimensions for risk assessment. Embodied fall risk tensor refers to a high-order mathematical structure that combines the risk features of the above multiple dimensions; that is, a dataset that comprehensively characterizes the robot's fall hazard state; for example, a first-order tensor containing three feature elements.

[0076] Specifically, first, the spatiotemporal convergence rate calculated in the previous step is extracted. Next, the relative stability margin obtained in the previous step is extracted, and its reciprocal is calculated by dividing the value by the relative stability margin. During this process, a zero-reduction protection mechanism is also required; when the relative stability margin approaches zero, its reciprocal is given a very large penalty value. Then, by querying the grid node corresponding to the robot's current position coordinates in the environmental geometric topology data map, the pre-stored ground friction coefficient of that node is read. Since the physical dimensions and numerical ranges of these three feature dimensions (spatiotemporal convergence rate, reciprocal stability margin, and friction coefficient) differ greatly, maximum and minimum value normalization is required. Each value is subtracted from its historical minimum value and then divided by the difference between the historical maximum and minimum values, thus mapping all feature values ​​to the interval between zero and one. In particular, for the friction coefficient, its normalized value needs to be subtracted from one, because the smaller the friction coefficient, the higher the risk of slipping and falling; it must be converted into a positive risk indicator. Finally, these three processed normalized eigenvalues ​​are used as independent elements of a tensor and arranged in a fixed order to construct a one-dimensional array-like first-order embodied fall risk tensor. In more complex engineering implementations, these three features can be projected onto the robot's three orthogonal motion axes to construct a 3x3 second-order risk tensor matrix, thereby accurately describing the fall risk distribution in different spatial directions.

[0077] Step 105: Map the embodied fall protection risk tensor to a comprehensive early warning index, and trigger the embodied active fall protection response command corresponding to the form of the inspection robot based on the comprehensive early warning index.

[0078] The comprehensive early warning index is a scalar value obtained by reducing the dimensionality of a multi-dimensional risk tensor through mathematical norm operations. It is used to intuitively determine the absolute level of the current fall risk. For example, if the calculated index is 85, it exceeds the set danger threshold of 80. Embodied active fall prevention response commands refer to hardware action control signals issued directly by the underlying control system to the motors or hydraulic actuators based on the robot's specific mechanical configuration, aimed at eliminating the risk of fall. For example, issuing an emergency reverse braking current command to the drive wheels of a wheeled robot.

[0079] Specifically, the execution process of this step is divided into two stages: index mapping and command triggering. First, the spatiotemporal convergence rate, the reciprocal of the relative stability margin, and the ground friction coefficient are extracted from the embodied fall protection risk tensor. These are then multiplied by the weighting coefficients corresponding to the current operational scenario. The sum of the squares of the weighted values ​​of each dimension is then taken to calculate the second norm of the tensor. This norm value is strictly defined as the comprehensive warning index. The comprehensive warning index is compared with the system's preset critical fall protection threshold in real time. When the comprehensive warning index exceeds the critical fall protection threshold, the control system immediately detects the current form identifier code of the inspection robot. If the form identifier code indicates a humanoid or quadrupedal shape, the first embodied active fall protection response command is generated. A new landing point is planned through inverse kinematics calculation, and the robot is controlled to perform a forced leg-stepping operation in the opposite direction away from the three-dimensional fall boundary field, thereby reconstructing and expanding the support polygon in physical space. If the shape identifier indicates wheeled or tracked, a second active anti-fall response command is generated, which sends a differential reverse braking signal with maximum torque to the wheel or track motor on the side closest to the fall boundary. At the same time, the active suspension system is controlled to reduce the suspension travel according to the preset compression ratio, forcibly reducing the overall center of gravity height of the robot. By increasing downforce and changing the motion vector, the tendency to become unstable and fall is completely blocked.

[0080] In the above embodiments, basic fall protection warning functions for inspection robots are achieved by supplementing environmental geometric topology data and identifying hazardous areas. To further enhance the active safety protection capabilities of inspection robots in complex high-altitude environments and reduce the impact of sudden environmental factors on robot stability, this application also provides a fall protection warning method for inspection robots. This method dynamically adjusts risk assessment weights based on specific operating scenarios to accurately calculate the norm of the embodied fall protection risk tensor, constructs an adaptive active fall protection response mechanism for different robot forms, and performs differentiated posture and motion interventions, enabling the system to more efficiently handle the highly dynamic fall protection control requirements under complex and extreme working conditions.

[0081] Step 201: Calculate the norm of the embodied fall protection risk tensor and use the norm as an early warning index.

[0082] Specifically, in implementing this step, the system uses rigorous mathematical operations to reduce the high-dimensional risk features into a single scalar indicator for rapid subsequent judgment. First, real-time embodied fall prevention risk tensor data is extracted from the robot's state estimation and environmental perception modules. Assuming this tensor is a three-dimensional data array, the Frobenius norm is calculated to obtain its overall size. The specific calculation process is as follows: the system controller iterates through each fundamental element of the multidimensional tensor, multiplying the currently extracted value by itself and squaring it; then, the squares of all elements are summed to obtain a total sum of squares; finally, the square root of the sum is taken. This non-negative scalar value is the norm of the embodied fall prevention risk tensor. The system directly assigns this norm value to the warning index variable, which accurately maps the overall fall risk of the robot in its current posture and environment. A larger value indicates a higher risk, and this value is stored in a cache for use in the next step.

[0083] In one possible implementation, the norm of the physical fall risk tensor is calculated, and the norm is used as a warning index. Specifically, steps 2011-2013 are included, as follows: Step 2011: Extract the spatiotemporal convergence rate, the reciprocal of the relative stability margin, and the ground friction coefficient from the embodied fall protection risk tensor.

[0084] Specifically, precise data extraction is achieved by directly accessing pre-allocated contiguous address blocks in memory. Within each control cycle, the underlying state estimation module packages the full amount of risk data calculated from multi-source sensor fusion into this memory region, forming a embodied fall-prevention risk tensor. The extraction process strictly follows the tensor's data structure index. First, the controller addresses a specific dimension in the tensor representing dynamic kinematic characteristics, reads the projection component of the current robot's center-of-mass velocity vector in the direction pointing to the fall boundary, extracts it as the spatiotemporal convergence rate, and stores it in the first register. Next, the controller addresses the dimension representing static characteristics, reads the shortest vertical distance from the robot's zero-moment point to the edge of the supporting polygon, performs a division operation (dividing a constant by this shortest vertical distance) to obtain the reciprocal of the relative stability margin, and stores it in the second register. Finally, the controller addresses the dimension representing environmental physical properties, reads the current contact surface friction resistance assessment value fed back by the foot-end tactile sensor or visual material recognition module, extracts it as the ground friction coefficient, and stores it in the third register. Through this fixed-index-based memory reading and basic operations, the system extracts the three most critical scalar features from the high-dimensional tensor, providing basic input data for subsequent weighted calculations. The entire extraction process is completed in microseconds, ensuring the real-time nature of the fall prevention warning.

[0085] Step 2012: Detect the current working scenario of the inspection robot; when the working scenario is to perform inspection on the surface of a wind turbine blade, detect whether the ambient wind speed exceeds the preset wind speed threshold. If it does, increase the weight coefficient of the reciprocal of the relative stability margin; when the working scenario is to perform inspection on the top of an oil storage tank, detect whether there is oil or water on the surface of the tank. If so, increase the weight coefficient of the ground friction coefficient; when the working scenario is to perform inspection on a high-altitude platform of a substation, detect whether the distance between the edge of the platform and the supporting polygon is less than the preset safety distance. If it is less, increase the weight coefficient of the spatiotemporal convergence rate.

[0086] Specifically, the system first determines the current operational scenario label by reading task scheduling instructions or matching real-time GPS coordinates with a pre-loaded high-precision 3D map. When the operational scenario label is identified as the surface of a wind turbine blade, the system calls the real-time data from the onboard anemometer and compares it with a preset wind speed threshold stored in non-volatile memory. If the current wind speed is significantly greater than the preset threshold, it indicates that strong winds could easily cause the robot to tip over. The system immediately increases the weighting coefficient of the reciprocal of the pre-allocated relative stability margin by a fixed step value, for example, by 0.5, to amplify its impact in the final risk assessment. When the operational scenario label is identified as the top of an oil tank, the system activates an infrared spectral or polarized light vision sensor array to scan the surface reflectivity along the predetermined travel trajectory. If an abnormally high reflectivity is detected or the spectral characteristics match an oil-water mixture, it indicates that the ground is highly slippery. The system immediately increases the weighting coefficient of the ground friction coefficient pre-allocated by a fixed step value to strengthen the weight of anti-slip capability in the fall prevention assessment. When the work scenario is identified as a substation high-altitude platform, the system uses LiDAR or a depth camera to calculate in real time the shortest straight-line distance from the outermost edge of the robot's support polygon to the platform's fall boundary. If this shortest straight-line distance is significantly less than the preset safety distance, it indicates that the robot is highly likely to crash off the platform due to inertia. The system immediately increases the weight coefficient pre-assigned to the spatiotemporal convergence rate by a fixed step value to emphasize the core role of speed control in emergency avoidance. Through this scene-aware dynamic weight adjustment mechanism, the system can adaptively optimize the fall risk assessment model, ensuring accurate capture of the most critical fall triggers in different hazardous environments.

[0087] Step 2013: Multiply the spatiotemporal convergence rate, the reciprocal of the relative stability margin, and the ground friction coefficient by their respective weighting coefficients, sum the squares, and take the square root to obtain the norm of the embodied fall protection risk tensor.

[0088] Specifically, in implementing this step, the central processing unit performs a series of rigorous scalar arithmetic operations to integrate the three core risk features, which have undergone dynamic weighting, into an intuitive fall prevention warning index. First, the system reads the spatiotemporal convergence rate from the register and its corresponding first weight coefficient, dynamically adjusted for the scenario, from memory. A multiplication instruction is executed to multiply the two to obtain the first weighted risk component. Next, the system reads the reciprocal of the relative stability margin and its corresponding second weight coefficient, and a multiplication instruction is executed to multiply the two to obtain the second weighted risk component. Subsequently, the system reads the ground friction coefficient and its corresponding third weight coefficient, and a multiplication instruction is executed to multiply the two to obtain the third weighted risk component. After calculating all weighted components, the system performs squaring operations on each of these three weighted risk components, multiplying each component by itself to obtain the first, second, and third squared values. Then, the system executes addition instructions to add these three squared values ​​sequentially, accumulating them to obtain a total sum of squares. Finally, the system calls a mathematical library function to perform a square root operation on this total sum of squares. The calculated non-negative scalar result is the norm of the embodied fall prevention risk tensor. The system directly assigns this norm value to the warning index variable, which accurately maps the overall fall risk of the robot in a specific working scenario. A higher value indicates a higher risk, and this value is stored in a high-speed cache for subsequent use by the fall prevention response mechanism. The entire calculation process is accelerated by a hardware floating-point unit, ensuring completion within an extremely short control cycle, providing solid underlying data support for the robot's safe inspection.

[0089] Step 202: Compare the comprehensive early warning index with the preset critical fall prevention threshold; when the comprehensive early warning index exceeds the critical fall prevention threshold, detect the current state of the inspection robot.

[0090] Specifically, the central control unit first reads the comprehensive warning index calculated in the previous step from the cache, and simultaneously reads the pre-calibrated critical fall prevention threshold from the safety configuration database. A direct numerical comparison is performed: the comprehensive warning index is subtracted from the critical fall prevention threshold. If the difference is greater than zero, meaning the comprehensive warning index is strictly greater than the critical fall prevention threshold, the robot is determined to face an extremely high risk of falling, and the active fall prevention response mechanism must be activated immediately. Once this mechanism is triggered, the system immediately executes the current form detection program. The specific process of form detection is as follows: a status query command is sent to the robot's chassis controller and joint actuators to read the encoder feedback data of each drive motor, the displacement sensor data of the suspension system, and the locking status position of the configuration switching mechanism. By parsing this low-level hardware data, the robot's current physical structure is determined. If the leg joint motors are detected to be in an enabled state and have discrete foot contact force feedback, it is determined to be in humanoid or quadrupedal form; if the hub motors or track drive wheels are detected to be in continuous rotation mode and the suspension system is in the corresponding working range, it is determined to be in wheeled or tracked form. The results of the morphological detection will be output in the form of enumerated variables, serving as the sole basis for generating targeted fall protection commands.

[0091] Step 203: If the current form is humanoid or quadrupedal, generate the first active fall protection response command and control the inspection robot to perform a leg-stepping operation in a direction away from the three-dimensional fall boundary field according to the first active fall protection response command, so as to reconstruct the support polygon.

[0092] Specifically, the first embodied active fall avoidance response command refers to a specific motion control signal generated for the leg-foot configuration; that is, the low-level data packet that triggers the leg joint motors to perform avoidance actions; for example, a control frame containing the target angle and angular velocity of each joint. The three-dimensional fall boundary field refers to a mathematical field model in space with height abrupt changes and repulsive potential energy; that is, the spatial distribution that quantifies the degree of danger at the fall edge; for example, a high potential energy gradient region formed at the edge of a step. The support polygon refers to a convex polygon formed by the projections of all the robot's foot endpoints onto a horizontal plane; that is, the geometric basis that determines static and dynamic equilibrium; for example, the rectangular area formed by the line connecting the outer edges of the two feet when standing on two feet.

[0093] The implementation of this step mainly involves potential field gradient calculation, inverse kinematics solution, and dynamic updating of geometric topology. When the shape is determined to be humanoid or quadrupedal, the system extracts the potential energy distribution function of the three-dimensional fall boundary field. The potential field gradient vector at the current robot's center of mass coordinate position is calculated. Specifically, the partial derivatives of the potential energy function are calculated with respect to the three coordinate axes in space, thus combining them into a three-dimensional vector pointing in the direction of the fastest increase in potential energy. The gradient vector is then reversed to obtain the optimal escape vector away from the fall boundary. Based on this escape vector, the gait planner calculates the three-dimensional target coordinates of the next safe landing point. Next, the inverse kinematics equation is solved using the pseudo-inverse algorithm of the Jacobian matrix, converting the target landing point coordinates into target rotation angle values ​​for each leg joint. These angle values ​​are packaged to generate the first embodied active fall prevention response command and sent to the underlying servo driver. The driver controls the mechanical leg closest to the boundary to lift up and step down along the escape vector direction. After the robotic leg completes its step and re-contacts the ground, the system reads the three-dimensional coordinates of all the feet that touched the ground and projects them onto a horizontal two-dimensional plane. Using the Graham scan algorithm, the convex hull of these projected points is calculated; the new geometric region formed by this convex hull is the reconstructed support polygon. This transfer of physical position ensures that the robot's zero-moment point falls back into the central region of the reconstructed support polygon, thereby restoring the system's dynamic balance and completely escaping the fall hazard zone.

[0094] Step 204: If the current form is wheeled or tracked, generate a second active fall protection response command, and control the wheel or track close to the three-dimensional fall boundary field to perform differential reverse braking operation according to the second active fall protection response command, and reduce the height of the inspection robot's suspension system according to a preset ratio to reduce the height value of the center of mass position coordinate.

[0095] Specifically, when the robot's form is determined to be wheeled or tracked, the system first calculates the vertical distances of the robot's two drive wheels or tracks from the three-dimensional fall boundary field, comparing them to determine the side with the smaller distance is closer to the boundary. The system generates a second active fall protection response command, sending a maximum negative torque command to the drive motor closer to the boundary, causing it to reverse; simultaneously, it sends a zero torque or positive braking torque command to the drive motor farther from the boundary. This differential reverse braking operation generates a strong yaw torque, forcing the robot chassis to rapidly rotate and yaw away from the boundary, changing its original dangerous trajectory. While performing differential braking, the command synchronously controls the active suspension system. It reads the current suspension system height value and the preset descent ratio constant. The target suspension height is calculated by multiplying the current suspension height value by one and subtracting the descent ratio constant. The hydraulic or electric actuator is controlled to compress the suspension height to the calculated target suspension height. As the chassis physical height decreases, the vertical height value of the robot's overall center of mass position coordinates decreases proportionally. According to the principle of static stability margin, the reduction of the center of mass height directly increases the robot's ability to resist overturning moment, effectively preventing rollover and fall accidents caused by centrifugal force or edge ground collapse during emergency differential turn.

[0096] The following describes a fall prevention warning system for inspection robot protection from a hardware processing perspective, according to an embodiment of this invention. Please refer to [link / reference needed]. Figure 3 This is a schematic diagram of a fall prevention warning system for the protection of an inspection robot in an embodiment of this application.

[0097] It should be noted that, Figure 3 The structure of a fall protection warning system for inspection robots shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0098] like Figure 3 As shown, a fall prevention warning system for inspection robot protection includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage section 308 into a random access memory (RAM) 303, such as executing the methods described in the above embodiments. The RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0099] The following components are connected to I / O interface 305: input section 306 including audio input devices, push-button switches, etc.; output section 307 including liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 308 including hard disks, etc.; and communication section 309 including network interface cards such as LAN (Local Area Network) cards, modems, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0100] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by central processing unit (CPU) 301, it performs the various functions defined in the present invention.

[0101] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0102] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0103] Specifically, a fall prevention warning system for inspection robot protection according to this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, it implements the fall prevention warning method for inspection robot protection provided in the above embodiment.

[0104] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the fall prevention warning system for inspection robot protection described in the above embodiments; or it may exist independently and not assembled into the fall prevention warning system for inspection robot protection. The storage medium carries one or more computer programs, which, when executed by a processor of the fall prevention warning system for inspection robot protection, cause the fall prevention warning system for inspection robot protection to implement the fall prevention warning method for inspection robot protection based on IoT data encryption transmission provided in the above embodiments.

Claims

1. A fall prevention early warning method for the protection of inspection robots, characterized in that, The method includes: Acquire multimodal perception data of the inspection robot, wherein the inspection robot's form includes humanoid, quadrupedal, wheeled, or tracked, and the multimodal perception data includes morphological contact surface data, environmental geometric topology data, and dynamic state data; By integrating the morphological contact surface data with the environmental geometric topology data, a support polygon matching the morphology of the inspection robot is constructed, and the three-dimensional drop boundary field at the edge of the support polygon is extracted. Based on the dynamic state data, the dynamic equivalent zero-moment point trajectory of the robot is calculated, and the dynamic equivalent zero-moment point trajectory is projected into the support polygon to obtain the relative stability margin reflecting the attitude anti-overturning capability. Extract the spatiotemporal convergence rate of the dynamic equivalent zero-moment point trajectory approaching the three-dimensional fall boundary field, and combine it with the relative stability margin to construct an embodied fall protection risk tensor that reflects the unstable fall trend. The embodied fall protection risk tensor is mapped to a comprehensive early warning index, and an embodied active fall protection response command corresponding to the form of the inspection robot is triggered based on the comprehensive early warning index.

2. The method according to claim 1, characterized in that, The process of fusing the morphological contact surface data with the environmental geometric topology data to construct a support polygon that matches the morphology of the inspection robot, and extracting the three-dimensional drop boundary field at the edges of the support polygon, includes: The shape contact surface data is classified and extracted according to the shape of the inspection robot; When the shape is humanoid or quadrupedal, extract the foot contact coordinates of discrete foot landing points; when the shape is wheeled or tracked, extract the contour envelope coordinates of continuous grounding imprints. The supporting polygon is generated by spatially intersecting the foot contact coordinates or the contour envelope coordinates with the passable plane in the environmental geometric topology data. Identify the suspended regions or steep slope edges adjacent to the supporting polygon in the environmental geometric topology data, and determine the three-dimensional spatial coordinate set of the suspended regions or steep slope edges as the three-dimensional drop boundary field.

3. The method according to claim 1, characterized in that, Based on the dynamic state data, the dynamic equivalent zero-moment point trajectory of the robot is calculated, and the dynamic equivalent zero-moment point trajectory is projected onto the supporting polygon to obtain a relative stability margin reflecting the attitude's anti-tipping capability, including: Extract the center of mass position coordinates, center of mass linear acceleration vector, and body angular acceleration vector from the dynamic state data; The first projection component of the inertial force in the horizontal plane is calculated based on the centroidal acceleration vector, and the second projection component of the inertial torque on the vertical axis is calculated based on the body angular acceleration vector. The first projection component and the second projection component are superimposed on the centroid position coordinates to obtain the dynamic equivalent zero moment point trajectory. Calculate the shortest normal distance from the current instantaneous point on the dynamic equivalent zero-moment point trajectory to each boundary of the supporting polygon; The minimum value among the shortest normal distances is taken as the relative stability margin at the current moment, and the relative stability margin is positively correlated with the overturning resistance.

4. The method according to claim 1, characterized in that, The extraction of the spatiotemporal convergence rate of the dynamic equivalent zero-moment point trajectory approaching the three-dimensional fall boundary field, combined with the relative stability margin, constructs an embodied fall protection risk tensor reflecting the unstable fall trend, including: Calculate the projection component of the velocity vector of the dynamic equivalent zero-moment point trajectory in the direction corresponding to the three-dimensional drop boundary field, and use it as the spatial approximation velocity; Divide the relative distance between the three-dimensional drop boundary field and the current equivalent zero moment point by the spatial approximation velocity to obtain the remaining collision time. The spacetime convergence rate is obtained by weighting and fusing the spatial approximation velocity with the reciprocal of the remaining collision time. Using the spatiotemporal convergence rate, the reciprocal of the relative stability margin, and the ground friction coefficient in the environmental geometric topology data as feature dimensions, an embodied fall protection risk tensor is constructed.

5. The method according to claim 1, characterized in that, The process of mapping the embodied fall protection risk tensor to a comprehensive early warning index, and triggering an embodied active fall protection response command corresponding to the form of the inspection robot based on the comprehensive early warning index, includes: Calculate the norm of the physical fall protection risk tensor and use the norm as an early warning index; The comprehensive early warning index is compared with a preset critical fall prevention threshold; When the comprehensive early warning index exceeds the critical fall prevention threshold, the current state of the inspection robot is detected; If the current form is humanoid or quadrupedal, a first embodied active fall protection response command is generated, and the inspection robot is controlled to perform a leg-stepping operation in a direction away from the three-dimensional fall boundary field according to the first embodied active fall protection response command, so as to reconstruct the support polygon; If the current form is wheeled or tracked, a second active fall protection response command is generated, and the wheels or tracks close to the three-dimensional fall boundary field are controlled to perform differential reverse braking operation according to the second active fall protection response command. The height of the inspection robot's suspension system is reduced according to a preset ratio to reduce the height value of the center of mass position coordinate.

6. The method according to claim 5, characterized in that, The calculation of the norm of the personal fall protection risk tensor includes: Extract the spatiotemporal convergence rate, the reciprocal of the relative stability margin, and the ground friction coefficient from the physical fall protection risk tensor; The current working environment of the inspection robot is omitted. When the operation scenario is to perform inspection on the surface of wind turbine blades, the system detects whether the ambient wind speed exceeds a preset wind speed threshold. If it does, the weighting coefficient of the reciprocal of the relative stability margin is increased. When the operation scenario involves performing an inspection on the top of an oil storage tank, the surface of the tank is checked for oil stains or water accumulation. If such stains or water accumulation are found, the weighting coefficient of the ground friction coefficient is increased. When the operation scenario is to perform inspection on a high-altitude platform in a substation, detect whether the distance between the edge of the detection platform and the supporting polygon is less than a preset safety distance. If it is less than that, increase the weighting coefficient of the spatiotemporal convergence rate. The norm of the embodied fall protection risk tensor is obtained by multiplying the spatiotemporal convergence rate, the reciprocal of the relative stability margin, and the ground friction coefficient by their respective weighting coefficients, summing their squares, and taking the square root.

7. The method according to claim 1, characterized in that, Before acquiring the multimodal perception data of the inspection robot, the process also includes: The inspection robot uses its visual and depth sensors to scan the path area within a preset distance range ahead, and obtains point cloud data of the environment ahead. The point cloud data is segmented into ground layers, and the boundaries of areas with a height difference greater than a preset threshold are marked as potential fall risk areas. Store the spatial coordinates of the potential fall risk areas in the danger zone map; During the inspection path planning process of the inspection robot, the shortest distance between the current planned path and each potential fall risk area in the danger zone map is calculated; When the shortest distance is less than the preset safety buffer distance, the boundary coordinates of the potential fall risk area are added to the environmental geometric topology data as identification objects.

8. A fall protection warning system for inspection robots, characterized in that, The fall protection warning system for inspection robot protection includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the fall protection warning system for inspection robot protection to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the fall protection warning system for the protection of the inspection robot, the fall protection warning system for the protection of the inspection robot performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the fall protection warning system for the protection of the inspection robot, the fall protection warning system for the protection of the inspection robot performs the method as described in any one of claims 1-7.

Images