A security robot intelligent inspection path optimization method

Abstract

The application relates to the technical field of robot control, and discloses a security robot intelligent inspection path optimization method, which comprises the following steps: acquiring a static two-dimensional grid map, updating network confidence flag bits according to message frequency states, and constructing a risk pheromone field and a mutual exclusion pheromone field superimposed with a space attenuation mask; combining a pan-tilt parameter to execute ray projection to generate an effective field of view polygon area, performing nonlinear attenuation update on the risk pheromone field according to a physical distance; calculating a field gradient and combining a generated orthogonal disturbance vector of a medium access control address mapping to generate a target motion vector to control chassis movement. The application solves cross-region scheduling conflicts caused by communication disconnection through network state linkage space masks; effectively perceives boundaries through field of view geometric calculation and optical distance attenuation real quantization; and breaks symmetry deadlock by using a disturbance component generated by a hardware unique identifier, so that the smoothness and continuity of multi-machine cooperative inspection movement are ensured.

Description

Technical Field

[0001] This invention relates to the field of robot control technology, specifically to a method for optimizing the intelligent inspection path of a security robot. Background Technology

[0002] With the development of automation technology, the collaborative execution of large-area space inspection tasks by multiple security robots has been widely applied. In existing multi-robot collaborative path planning schemes, the robot's movement route usually relies on a central scheduling system, or requires continuous and stable communication between each device node. However, in actual building environments such as underground parking garages and large factories, physical obstructions and electromagnetic interference often cause local communication network oscillations or partial interruptions. Existing path planning mechanisms lack effective autonomous fault-tolerant degradation strategies when facing communication anomalies. Robots are prone to scheduling conflicts based on outdated global coordinate data, leading to overlapping inspection areas or multi-robot congestion.

[0003] When evaluating the inspection coverage of robots, traditional methods often simplify the perception range of vehicle-mounted cameras to a fixed-radius circular area centered on the robot. This data processing approach fails to incorporate static obstacles in the spatial environment for field-of-view geometry calculations, nor does it consider the natural attenuation of image sharpness from optical sensors as physical observation distance increases. This leads the system to easily classify areas that are actually in visual blind spots or too far away to be effectively identified as having been inspected, failing to accurately reflect the actual perception coverage of the environment and reducing the effectiveness of security inspections.

[0004] Furthermore, existing distributed path guidance algorithms based on artificial potential fields or virtual pheromones generally face the problem of symmetric deadlock. When multiple robots meet in structurally symmetrical terrains such as narrow corridors, the forward guiding forces generated by each robot and the obstacle avoidance repulsion forces between them can easily cancel each other out, causing the calculated motion control vector to approach zero. Existing conventional processing mechanisms mostly rely on communication negotiations between robots to reallocate right-of-way. This mechanism is often difficult to execute under unstable network conditions, leading to robots getting stuck in a state of stagnation or repetitive oscillation for a long time, affecting the coherence of the underlying physical motion and the overall task execution efficiency. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for optimizing intelligent inspection paths for security robots. This method solves the problems of scheduling conflicts caused by communication interruptions, blind spot misjudgments due to mismatched field of view coverage calculations, and motion deadlock caused by multiple robots encountering each other in symmetrical environments.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for optimizing the intelligent inspection path of a security robot, comprising the following steps:

[0007] Acquire a static two-dimensional grid map of the inspection area, send and receive parameter messages in the local area communication network, and update the network confidence flag based on the received frequency status;

[0008] Based on the static two-dimensional raster map and the network confidence flag, a risk pheromone field with superimposed spatial decay mask is constructed, and a mutually exclusive pheromone field is constructed.

[0009] Extract the PTZ camera parameters from the parameter message, combine them with the static two-dimensional grid map to perform ray projection, and calculate and generate an effective field of view polygon region.

[0010] For the coordinate grid within the effective field of view polygonal region, the risk pheromone field is updated nonlinearly based on the physical distance between the coordinate grid and the PTZ camera.

[0011] The basic motion guidance vector is generated by calculating the positive gradient vector of the risk pheromone field and the negative gradient vector of the mutually exclusive pheromone field. The medium access control address identifier is extracted and mapped to generate an orthogonal disturbance vector, which is then combined with the basic motion guidance vector to generate a target motion vector. The target motion vector is then weighted and superimposed with the chassis kinematic smoothing vector. A speed command is generated through the inverse kinematics of the chassis to control the chassis of the security robot to move.

[0012] Preferably, updating the network confidence flag based on the received frequency status specifically includes:

[0013] Set the message reception timeout threshold to the number of communication synchronization cycles in which no messages are received consecutively;

[0014] If the dynamic cache queue is polled at a fixed frequency and no valid parameter message of the external neighboring node is updated within the time span corresponding to the message reception timeout threshold, the network confidence flag is overwritten to the first state value corresponding to the isolated operation mode outside the cluster.

[0015] If a valid parameter message is obtained during the polling process, the network confidence flag is set to the second state value corresponding to the multi-machine collaborative operation mode.

[0016] Preferably, the construction of the risk pheromone field with superimposed spatial attenuation mask specifically includes:

[0017] Based on the environmental basic risk constant, the static environmental volatility coefficient, and the integral term of the dynamic external trigger weight with the time without inspection, the basic risk pheromone concentration is calculated and generated.

[0018] A spatial decay mask is dynamically generated based on the state value of the network confidence flag. The basic risk pheromone concentration is multiplied by the spatial decay mask to generate the risk pheromone field.

[0019] Preferably, the dynamic generation of the spatial attenuation mask specifically includes:

[0020] When the network confidence flag is the second state value representing the multi-machine collaborative operation mode, the spatial attenuation mask value is always one.

[0021] When the network confidence flag is the first state value representing the operation mode of leaving the cluster island, a spatial decay mask is generated by using the Gaussian decay function based on the Euclidean distance between the current physical coordinates of the security robot and the coordinate grid and the set spatial tolerance radius. The value of the mask decreases exponentially as the Euclidean distance increases.

[0022] Preferably, the construction of the mutually exclusive pheromone field specifically includes:

[0023] When the network confidence flag is the state value corresponding to the multi-machine collaborative operation mode, the global physical coordinates of all surrounding security robots are extracted as the obstacle coordinate set;

[0024] When the network confidence flag is the state value corresponding to the isolated operation mode, the two-dimensional coordinate points of the dynamic obstacle clustering center detected by the three-dimensional lidar are projected onto the plane coordinate system as the obstacle coordinate set.

[0025] Based on the planar Euclidean distance between the target coordinate grid in the static two-dimensional grid map and the two-dimensional coordinate points of each entity in the obstacle coordinate set, combined with the mutual exclusion peak constant and the mutual exclusion influence radius, the cumulative value of the repulsion concentration is calculated using a Gaussian function to generate the mutual exclusion pheromone field.

[0026] Preferably, the step of extracting the pan-tilt camera parameters from the parameter message, combining them with the static two-dimensional grid map to perform ray casting, and calculating and generating an effective field-of-view polygon region specifically includes:

[0027] The relative yaw angle of the pan-tilt camera is extracted and added to the global heading angle of the chassis to establish the angular reference for ray projection, and a virtual ray is emitted along the central optical axis to the static two-dimensional grid map;

[0028] By combining the pitch angle, physical installation height, maximum sensing distance parameters of the PTZ camera and the planar Euclidean distance to the coordinate grid of the first static obstacle, the final effective projection length of a single virtual ray is calculated, and the coordinates of the cutoff point are obtained.

[0029] Connect adjacent cut-off points sequentially, and connect the first and last cut-off points to the physical coordinates of the chassis center to close and generate the effective field-of-view polygon region.

[0030] Preferably, after closing and generating the effective field-of-view polygon region, the method further includes:

[0031] Extract the maximum and minimum values ​​of the x-coordinate and the maximum and minimum values ​​of the y-coordinate from the coordinates of all ray cutoff points to construct a two-dimensional minimum bounding rectangle;

[0032] For a coordinate grid located inside the two-dimensional minimum bounding rectangle, if the coordinate grid belongs to the effective field of view polygon region and the static obstacle occupancy attribute is no obstacle, then the corresponding field of view coverage binarization mask value is set to one; otherwise, it is set to zero, and the field of view coverage binarization mask matrix is ​​output.

[0033] Preferably, the nonlinear decay update of the risk pheromone field specifically includes:

[0034] Filter the coordinate grids whose field of view covers a binary mask value of one, and calculate the Euclidean physical distance between the center point of the coordinate grid and the optical center of the pan-tilt camera projected onto the static two-dimensional grid map.

[0035] Based on the Euclidean physical distance and optical resolution attenuation constant, a nonlinear exponential function is used to calculate the pheromone elimination ratio. Based on the pheromone elimination ratio, a nonlinear attenuation calculation is performed on the risk pheromone concentration before the update to obtain the updated risk pheromone concentration.

[0036] Based on the absolute timestamp of the current moment and the binarized mask value of the field of view coverage, reset the starting point of the uninspected time integration corresponding to the coordinate grid.

[0037] Preferably, the step of calculating the positive gradient vector of the risk pheromone field and the negative gradient vector of the mutually exclusive pheromone field to generate a basic motion guidance vector, extracting the medium access control address identifier mapping to generate an orthogonal perturbation vector and combining it with the basic motion guidance vector to generate a target motion vector specifically includes:

[0038] Within the local window of the current coordinates of the security robot, the concentration difference between adjacent grids is extracted to calculate the positive spatial gradient vector of the risk pheromone field and the spatial gradient vector of the mutually exclusive pheromone field, and weighted difference is performed to generate the basic motion guidance vector.

[0039] Read the Media Access Control address that is factory-set on the network interface, extract the low address byte and convert it into a decimal integer, and use a hash modulo algorithm to convert the decimal integer into a tangential deflection disturbance angle;

[0040] Using a two-dimensional rotation matrix containing the tangential deflection disturbance angle, a deflection transformation is performed on the basic motion guide vector to generate the target motion vector superimposed with the orthogonal disturbance vector.

[0041] Preferably, the step of weighted superposition of the target motion vector and the chassis kinematics smoothing vector, and generating a speed command through inverse kinematics of the chassis, specifically includes:

[0042] Extract the historical velocity vector from the previous control cycle and perform first-order low-pass filtering and weighted fusion with the target motion vector to generate the desired velocity vector;

[0043] The magnitude of the desired velocity vector is obtained as the desired linear velocity of the chassis. The target heading angle corresponding to the desired velocity vector is extracted using the arctangent function. The desired angular velocity is calculated by combining the current actual heading angle of the vehicle head with the angular velocity ratio gain.

[0044] The desired linear velocity and the desired angular velocity are decomposed and solved using the Jacobian inverse matrix of the two-wheel differential kinematic model, which is composed of the physical wheel track and the physical radius of the drive wheel, to obtain the desired rotational angular velocity commands of the left and right drive motors.

[0045] This invention provides a method for optimizing the intelligent inspection path of a security robot. It has the following beneficial effects:

[0046] 1. By acquiring the status of local communication network messages and updating the network confidence flag, and dynamically generating a spatial attenuation mask based on the flag and superimposing it onto the risk pheromone field and the mutual exclusion pheromone field, the system can achieve multi-machine collaboration and collision avoidance based on global parameters when the network is connected. When the network oscillates or disconnects from the cluster, the spatial attenuation mask reduces the risk attraction of the remote area as the physical distance increases, prompting the security robot to smoothly switch to local space patrol and avoid obstacles based on its own sensor data. This solves the network dependency problem in complex electromagnetic environments and avoids cross-regional path scheduling conflicts caused by communication loss.

[0047] 2. This invention extracts physical parameters such as the pitch and yaw angles of the pan-tilt camera, combines them with a static two-dimensional grid map, and uses a ray casting algorithm to generate an effective field-of-view polygonal region. Furthermore, it updates the risk pheromone nonlinearly based on the Euclidean distance between the target coordinate grid and the camera's optical center, eliminating visual obstruction areas caused by static environmental obstacles. The pheromone elimination ratio is matched to the objective law of optical resolution degradation due to increased distance. This corrects the misjudgments caused by traditional fixed-radius coverage algorithms, quantifies the true visual perception boundary, and improves the effectiveness and accuracy of inspection coverage.

[0048] 3. This invention calculates the weighted difference between the positive gradient of the risk pheromone and the negative gradient of the mutually exclusive pheromone within a local window, and uses the underlying fixed medium access control address to perform hash mapping to generate a tangential deflection angle, thereby constructing a target motion command with superimposed orthogonal perturbation vectors. When multiple robots meet in a symmetrical narrow channel, causing the conventional guiding gradient to cancel out to zero, the specific orthogonal perturbation component generated based on the hardware unique identifier can break the vector force balance, enabling different robots to automatically generate differentiated offsets to different sides without complex communication negotiation, effectively eliminating deadlock and ensuring the smoothness and continuity of the underlying motion. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the architecture of a security robot intelligent inspection path optimization system according to an embodiment of the present invention;

[0050] Figure 2 This is a flowchart illustrating a method for optimizing the intelligent inspection path of a security robot according to an embodiment of the present invention.

[0051] Figure 3 This is a flowchart of the sub-steps for calculating and generating the effective field-of-view polygon region in an embodiment of the present invention;

[0052] Figure 4 This is a flowchart of the sub-steps for updating the nonlinear decay of risk pheromones and resetting the uninspected timestamp in an embodiment of the present invention.

[0053] Figure 5 This is a graph showing the response curves of the chassis's desired linear velocity and the target heading angle in an embodiment of the present invention.

[0054] Figure 6 This is a dynamic waveform diagram of the expected rotational angular velocity of the left and right drive motors in an embodiment of the present invention. Detailed Implementation

[0055] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] See attached document Figure 1This invention provides an intelligent patrol path optimization system for security robots, applicable to collaborative patrol scenarios involving multiple security robots. The system adopts a decentralized architecture, relying on multiple security robots deployed within the patrol area to construct communication and distributed computing nodes. No central scheduling server is required between the multiple security robots; they directly establish a local communication network through physically mounted wireless self-organizing network communication modules to perform message exchange between nodes.

[0057] The physical architecture of a single security robot includes a perception unit, an execution unit, a communication unit, and an edge computing control unit as the core of its operation. The perception unit comprises a 3D LiDAR and a pan-tilt camera. The 3D LiDAR is used to collect dynamic and static spatial point cloud data of the inspection environment; the pan-tilt camera includes an electromechanical pan-tilt unit, used to output real-time pitch and yaw angle parameters and visual images. The execution unit includes the underlying drive chassis and corresponding motor drivers. The communication unit uses wireless networking hardware supporting the Mesh protocol.

[0058] The edge computing control unit is connected to the sensing unit, execution unit, and communication unit via an internal data bus. The edge computing control unit includes a processor and memory. It receives raw data from the 3D LiDAR and PTZ camera, extracts external messages from the wireless ad hoc network communication module, and sends the processed motor drive commands to the motor driver.

[0059] The memory stores executable logic modules, specifically including an environment synchronization module, a field calculation module, a field-of-view cancellation module, and a motion control module. The environment synchronization module maintains a static two-dimensional grid map and parses the status of local communication network messages; the field calculation module performs matrix integration operations on virtual pheromones and hardware data source switching mapping; the field-of-view cancellation module performs spatial ray projection and nonlinear geometric calculations based on gimbal mechanical parameters; and the motion control module couples multi-dimensional virtual field gradients with hardware identifier features, ultimately converting them into chassis physical control parameters. These hardware entities and logic modules together constitute the implementation environment of the method of this invention.

[0060] See attached document Figure 2 This invention provides a method for optimizing the intelligent inspection path of a security robot, comprising the following steps:

[0061] S100: Obtain a static two-dimensional grid map of the inspection area. The security robot broadcasts its own pose and gimbal parameters in the local communication network at a set period, receives parameter messages from neighboring nodes, and updates the network confidence flag based on the network message reception frequency status.

[0062] S200, based on the aforementioned static two-dimensional grid map and network confidence flag, constructs a risk pheromone field and a mutual exclusion pheromone field. The risk pheromone field is generated by integrating environmental basic parameters, IoT external trigger signals, and the integral of the uninspected time, and then superimposed with a spatial attenuation mask. The mutual exclusion pheromone field is generated by calculating and generating anti-collision repulsion force based on the adjacent node coordinates in the aforementioned parameter message or the obstacle clustering coordinates output by the three-dimensional LiDAR.

[0063] S300: Extract the pitch angle and yaw angle of the pan-tilt camera from the aforementioned pan-tilt parameters, and combine them with the aforementioned static two-dimensional grid map to perform a ray projection algorithm to calculate and generate an effective field of view polygon area.

[0064] S400, for the coordinate grid located within the aforementioned effective field of view polygon area, perform nonlinear decay update on the aforementioned risk pheromone field based on the physical distance between the corresponding coordinate grid and the PTZ camera;

[0065] S500 calculates the positive gradient vector of the risk pheromone field and the negative gradient vector of the aforementioned mutually exclusive pheromone field at the current coordinates of the security robot after nonlinear decay and update. It extracts the medium access control address identifier feature mapping of the security robot to generate an orthogonal disturbance vector. It then weights and superimposes the aforementioned vectors with the chassis kinematics smoothing vector. Through the inverse kinematics solution of the chassis, it generates the desired linear velocity and desired angular velocity commands and sends them to the underlying drive chassis to execute physical movement.

[0066] In this embodiment, the step S100 of acquiring the static two-dimensional grid map of the inspection area and updating the network confidence flag can be completed through the following sub-steps:

[0067] S101: Acquire a static two-dimensional grid map of the inspection area and perform parameterized discretization definition on the environmental space. The edge computing control unit of the security robot receives environmental point cloud data scanned by the three-dimensional LiDAR and processes it through synchronous localization and mapping algorithms to generate a two-dimensional environmental contour of the inspection area.

[0068] After generating the base map, the system extracts the boundary dimensions of the physical plane and divides the continuous physical space into a two-dimensional grid array of equal size, thereby constructing a static two-dimensional grid map. In this embodiment, by dividing the environment into discrete grids, the continuous physical space can be mapped into a discrete coordinate system that facilitates matrix calculations by the computer.

[0069] The side length of the discrete grid is determined based on the physical chassis dimensions of the security robot and the scanning resolution of the 3D LiDAR, with a range of 0.05m to 0.2m. The set of discretized coordinate grids is defined as follows: ,in and Integer coordinate index of the raster in a two-dimensional plane. and These correspond to the maximum number of grid cells at the horizontal and vertical boundaries of the inspection area, respectively. The value is obtained by dividing the total length of the environmental physical boundary in the corresponding direction by the grid cell side length and rounding down. Each coordinate grid cell records the static obstacle occupancy attribute at the corresponding physical location.

[0070] S102 broadcasts its own pose and gimbal parameters at a set period within the local communication network, while simultaneously receiving parameter messages from neighboring nodes. Multiple security robots establish communication nodes using a wireless ad hoc network communication module. A unified communication synchronization period time constant is set internally within the system.

[0071] In this embodiment, the time constant is set based on the security robot's maximum speed and the bandwidth capacity of the wireless ad hoc network communication module to ensure that the physical displacement of the security robot during the two broadcasts is within a safe tolerance distance, with a setting range of 0.1s to 1.0s. Each time a communication synchronization cycle is completed, the security robot packages its hardware perception data into a heartbeat message and broadcasts it. The data payload of the heartbeat message includes the current global physical coordinates of the security robot's chassis, the chassis heading angle, and the real-time pitch and yaw angles of the gimbal camera. The edge computing control unit opens its listening port to continuously capture parameter messages sent by other neighboring nodes in the local communication network.

[0072] The edge computing control unit performs protocol parsing on the received parameter messages, extracts the pose and gimbal status data of the surrounding security robots, and the message interaction process provides a global location distribution data foundation for multi-machine collaboration. It also pushes the status data into a local dynamic cache queue for subsequent pheromone calculation.

[0073] S103: Determine network connectivity based on the receiving frequency of network packets and update the network confidence flag. Due to physical obstructions and electromagnetic interference in the security environment, the local area communication network may experience temporary packet loss or connection interruption.

[0074] In this embodiment, by monitoring network status and dynamically updating flag bits, the system can smoothly switch working modes in the event of network disconnection caused by complex building structures, avoiding path planning conflicts based on outdated coordinate data. The edge computing control unit is equipped with state machine monitoring logic and sets a message reception timeout threshold.

[0075] The message reception timeout threshold is quantized as the number of communication synchronization cycles in which no messages are received consecutively, and is denoted as a constant. The value range is set to 3 to 10 times based on the intensity of environmental interference. The edge computing control unit polls the local dynamic cache queue at a fixed frequency. If in continuous Within the time span corresponding to each communication synchronization cycle, if the dynamic cache queue does not update any valid parameter messages from external neighboring nodes, the edge computing control unit determines that a network oscillation or communication loss has been triggered and overwrites the network confidence flag in the memory with the first state value (e.g., a value of 0).

[0076] If a valid message is successfully obtained during the polling process, the edge computing control unit determines that the network connectivity is normal and sets the network confidence flag to the second state value (e.g., value 1). The first state value indicates that the security robot is currently in an isolated operation mode, detached from the cluster, while the second state value indicates that the security robot is in a normal multi-machine collaborative operation mode. The network confidence flag is used to control parameter switching and fault tolerance degradation during subsequent virtual pheromone field calculations.

[0077] In this embodiment, step S200, which involves constructing a risk pheromone field based on a static two-dimensional grid map and network confidence flags, can be implemented through the following sub-steps:

[0078] S201 integrates environmental fundamental constants, external trigger signals from the Internet of Things, and the integral term of uninspected time to construct a basic risk pheromone field.

[0079] In this embodiment, the edge computing control unit of the security robot updates the virtual scalar field in real time on a locally maintained static two-dimensional grid map. This applies to the coordinate grid set. Any target coordinate grid within The target coordinate grid contains risk pheromones that characterize the degree of risk accumulated over time in physical areas not covered by security robots. The concentration of the basic risk pheromone field monotonically increases over time. The equation for calculating the concentration of the basic risk pheromone field is defined as:

[0080] ;

[0081] In the formula, Represents the target coordinate grid timestamp The baseline risk pheromone concentration at that time; This represents the basic environmental risk constant, which corresponds to a fixed value of the default baseline risk distribution of the environment, and its value range is set from 0 to 10. This represents the absolute timestamp of the last time a risk pheromone removal action occurred on the target coordinate grid, for physical areas that have never been inspected. Initialize to system startup time; This represents the static environment volatility coefficient. The value is positively correlated with the security level of the physical area corresponding to the target coordinate grid. For example, the value set for the vault entrance area is greater than the value set for the ordinary corridor. The value range is set from 0.1 to 1.0. Indicates the integral variable The dynamic external trigger weight is set at any time. When an external IoT sensor deployed near the corresponding physical coordinate generates an alarm signal, the value becomes a preset high-level activation value (e.g., set to 100). When there is no abnormality, the value is zero. and These represent the system-preset environmental proportional gain coefficient and the external trigger proportional gain coefficient, respectively, used to adjust the weight distribution between the normal growth rate of environmental volatiles and the sudden alarm growth rate. Their values ​​are determined through normalization, and both are set within a range of 0 to 1. In the actual operational logic of the edge computing control unit, the aforementioned continuous-time integral term is implemented by discretely accumulating the environmental fundamental constant and the trigger signal within each communication synchronization cycle.

[0082] In this embodiment, by introducing time integration, the concentration of risk pheromones corresponding to uninspected physical areas continuously accumulates and increases over time. The mathematical integration mechanism forms a dynamic attraction field in two-dimensional physical space, which can utilize naturally formed physical gradients to prompt the security robot to actively move towards high-risk areas that have not been covered for a long time.

[0083] S202, an adaptive spatial attenuation mask is generated based on the network confidence flag, and this mask is superimposed on the basic risk pheromone field to calculate and generate a dynamic risk pheromone field. In this embodiment, to address the multi-machine scheduling conflict caused by the security robot's inability to obtain global coordinates during network oscillations or network disconnections, a degradation and fault tolerance mechanism linked to the network state is introduced. The edge computing control unit extracts the updated network confidence flag in step S100 and dynamically generates a spatial attenuation mask based on the state value of the network confidence flag. The piecewise function expression for a spatial attenuation mask is defined as follows:

[0084] ;

[0085] In the formula, Indicates the network confidence flag; Indicates the current physical coordinates of the security robot With target coordinate grid The Euclidean distance between them is calculated from the planar relative position of the center coordinates of the security robot chassis and the center point coordinates of the target coordinate grid; This indicates the spatial tolerance radius in island mode, which is set according to the single-pass endurance of the security robot and the area of ​​the local area, with a setting range of 10m to 50m.

[0086] When the network confidence flag is at the second state value (corresponding to...) When the security robot is in a multi-robot collaborative operation mode, the spatial attenuation mask is constantly set to 1, indicating that the security robot responds to the complete risk gradient within the global static two-dimensional grid map. When the network confidence flag is at the first state value (corresponding to...), the security robot is in a multi-robot collaborative operation mode. When the security robot operates in an isolated, detached mode, the spatial attenuation mask is transformed into a Gaussian attenuation function based on Euclidean distance. The spatial attenuation mask value decreases exponentially and approaches zero for coordinate grids farther from the security robot. The final dynamic risk pheromone field concentration is calculated by multiplying the base risk pheromone concentration by the spatial attenuation mask.

[0087] In this embodiment, through a multiplicative and superimposed calculation method, the system can exponentially reduce the risk pheromone concentration in the remote area as physical distance increases when the local communication network is disconnected. Security robots attracted by the remote high-risk area switch their operating state to respond to the local risk gradient of the near area, thereby confining the security robots to perform high-frequency patrols within the local physical space when the network disconnection occurs, avoiding patrol trajectory conflicts caused by cross-regional location scheduling. The dynamic risk pheromone field concentration calculated and output in this step is used as the dynamic risk pheromone concentration before the update in subsequent step S401. Perform the calculation.

[0088] In this embodiment, for step S200 of constructing a mutually exclusive pheromone field based on a static two-dimensional grid map and network confidence flags, in addition to the aforementioned calculation of the basic risk pheromone field and spatial attenuation mask, the specific implementation process also completes the construction of the mutually exclusive pheromone field through the following sub-steps:

[0089] S203 dynamically switches data sources and calculates and generates a mutually exclusive pheromone field based on the adjacent node coordinates in the parameter message or the obstacle clustering coordinates output by the 3D LiDAR.

[0090] In this embodiment, the edge computing control unit determines the data source for the obstacle coordinate set required for the mutually exclusive pheromone field based on the network confidence flag state. When the network confidence flag is at the second state value (corresponding to...), When the system is in a multi-machine collaborative operation mode, the edge computing control unit extracts the global physical coordinates of all surrounding security robots from the local dynamic cache queue as the obstacle coordinate set, which is used to construct anti-collision repulsion force.

[0091] When the network confidence flag is at the first state value (corresponding to...) When the system is in an isolated operation mode, detached from the cluster, the edge computing control unit stops reading cached messages and instead reads real-time point cloud data from local 3D LiDAR scanning, extracting the dynamic obstacle cluster coordinates in the physical environment as the obstacle coordinate set.

[0092] In this embodiment, the edge computing control unit filters out the height axis data of the three-dimensional center coordinates, and projects the three-dimensional center coordinates of the dynamic obstacle vertically into the plane coordinate system of the static two-dimensional grid map, generating corresponding two-dimensional coordinate points that are included in the obstacle coordinate set.

[0093] After determining the set of obstacle coordinates, the edge computing control unit calculates the mutually exclusive pheromone field on a static two-dimensional raster map. The equation for calculating the concentration of the mutually exclusive pheromone field is defined as:

[0094] ;

[0095] In the formula, Represents the target coordinate grid The concentration of mutually exclusive pheromones at the location; This represents the total number of entities in the current obstacle coordinate set; This represents the peak repulsion constant, which indicates the maximum repulsive force at the center of the entity, and its value range is set from 50 to 200. Represents the first element in the set of obstacle coordinates. The central physical coordinates of each entity; Represents the target coordinate grid With the Physical coordinates of the center of each entity The planar Euclidean distance between them is calculated by taking the square root of the sum of the squares of the differences between the x and y coordinates of the target coordinate grid center point and the x and y coordinates of the entity's two-dimensional coordinate point. This indicates the mutual exclusion radius, and the value is set based on the physical chassis size of the security robot and the maximum safe braking distance, with a setting range of 2m to 5m.

[0096] In this embodiment, the purpose of constructing a mutually exclusive pheromone field is to simulate a repulsive force field in physical space. The mutually exclusive pheromone field generates a high-concentration repulsive region with Gaussian decay around adjacent nodes or physical obstacles. The fault-tolerant switching mechanism of the data source ensures that when the local communication network is connected, the system can use global node coordinates to plan avoidance paths in advance; when the local communication network is disconnected, it can dynamically switch to relying on the body perception hardware to detect local obstacles.

[0097] The attenuation calculation method based on Gaussian function makes the repulsion concentration higher the closer to the center of the obstacle, thus transforming environmental perception data into a quantitative indicator for chassis obstacle avoidance control and improving the physical fault tolerance performance in the collaborative inspection process of multiple security robots.

[0098] See attached document Figure 3In this embodiment, regarding step S300, which involves extracting gimbal parameters and calculating the effective field-of-view polygon region using a static two-dimensional raster map, the edge computing control unit ultimately outputs the geometric boundary of the two-dimensional region generated by the closure of the cutoff point. The specific implementation process is completed through the following sub-steps:

[0099] S301, extract the pitch and yaw angles of the gimbal camera and calculate the absolute orientation of the central optical axis by combining it with the chassis heading angle. In this embodiment, the basic perception logic and yaw parameters of the electromechanical gimbal component have been verified. The edge computing control unit directly extracts the global heading angle of the underlying drive chassis and adds it to the relative yaw angle of the gimbal camera, establishing the angular reference for ray projection through spatial dimension transformation.

[0100] S302, Set the maximum sensing distance parameter, emit a ray along the central optical axis of the static two-dimensional grid map, and execute a ray projection algorithm that truncates upon encountering obstacles. In this embodiment, the initial ray generation rules and discretized coordinate sampling traversal adopt standard processing. The edge computing control unit skips intermediate bridging variables based on the pitch angle state and directly executes the calculation of the ray truncation length and polygon conversion. Considering that the line of sight of the PTZ camera cannot intersect with the ground when it is in a level or upward view state, the final effective projection length calculation equation for a single virtual ray is defined as:

[0101] ;

[0102] In the formula, Indicates the final effective projection length of a single virtual ray; This indicates the maximum sensing distance parameter to be set, with a setting range of 5m to 20m. This indicates the physical installation height of the PTZ camera's optical center from the ground. The value is set based on the robot's mechanical structure drawings, and the setting range is 1m to 2m. This indicates the pitch angle of the pan-tilt camera. The value is positive when looking down, and non-positive when looking at eye level or up. This represents the planar Euclidean distance from the starting point coordinates to the first static obstacle coordinate grid along the ray direction.

[0103] After the traversal calculation is completed, the edge computing control unit extracts the coordinates of the ray cutoff points corresponding to all virtual rays, connects adjacent cutoff points sequentially according to the order in which the rays were generated, and connects the first and last ray cutoff points to the physical coordinates of the chassis center. This directly closes the polygonal region of the effective field of view on the static two-dimensional grid map. In this embodiment, the polygon closure process transforms the single-line ranging result into a continuous two-dimensional region boundary, providing a definite geometric contour for subsequent determination of whether the coordinate grid belongs to the effective field of view coverage.

[0104] In this embodiment, regarding step S300, which calculates the specific conditions for generating the boundary of the effective field of view polygon region based on the effective penetration range of rays, the edge calculation control unit finally outputs a field of view coverage binarized mask matrix to the system for updating the pheromone field state. The specific implementation process is completed through the following sub-steps:

[0105] S303: Boundary conditions are established based on the effective penetration range of the ray, and a binary mask for the field of view coverage is generated. In this embodiment, the edge computing control unit performs point-to-surface inclusion determination on the spatial coordinates within the static two-dimensional raster map. To reduce the computational power consumption of the edge computing control unit and meet real-time requirements, the system pre-extracts the maximum and minimum values ​​of the abscissa and ordinate of all ray cutoff point coordinates, thereby constructing a two-dimensional minimum bounding rectangle for the effective field of view polygonal region. The edge computing control unit only traverses and determines the coordinate raster located inside the two-dimensional minimum bounding rectangle.

[0106] Determine any coordinate grid The specific conditions for determining whether a region belongs to the boundary of the effective field of view polygon region are determined by directly solving the following Boolean logic equation:

[0107] ;

[0108] In the formula, Represents coordinate grid The corresponding field of view coverage binarization mask value, a value of 1 indicates that it is within the effective field of view of the physical camera, and a value of 0 indicates that it is outside the field of view or is blocked; It represents the extent of a polygonal geometric space formed by the set of ray truncation points closed within a two-dimensional plane; This indicates the obstacle occupancy attribute of the coordinate grid in a static 2D grid map. A value of 0 represents no obstacles, and a value of 1 represents the presence of obstacles.

[0109] After batch mesh determination, the edge computing control unit outputs a complete field-of-view coverage binarized mask. In this embodiment, Boolean logic equations are used for filtering and judgment. The system can map continuous geometric polygons back to a discrete two-dimensional grid coordinate system and use the obstacle properties of the environment itself to remove impassable areas inside the polygons. The field of view coverage binarization mask matrix defines the physical visible coverage boundary under the current gimbal posture and chassis position conditions of the security robot. Areas outside the mask coverage area do not participate in subsequent processing calculations, so that the boundary of the finally extracted effective field of view polygon region is limited by the optical physical penetration range and environmental occlusion conditions.

[0110] See attached document Figure 4In this embodiment, the pheromone nonlinear anisotropy elimination step S400 for coupling physical distance is specifically implemented through the following sub-steps:

[0111] S401 calculates the Euclidean physical distance from the center point of the coordinate grid to the optical center of the PTZ camera for the coordinate grid within the effective field of view polygon area, and performs risk pheromone nonlinear exponential decay update by configuring the optical resolution attenuation constant.

[0112] In this embodiment, the edge computing control unit filters out target coordinate grids with a mask value of 1 based on the field of view coverage binarized mask matrix output from the previous sub-step. For each target coordinate grid, the edge computing control unit extracts the horizontal and vertical coordinates of the center point of the target coordinate grid and the horizontal and vertical coordinates of the projection point of the optical center of the PTZ camera on the static two-dimensional grid map, and calculates the planar Euclidean physical distance between the two.

[0113] After obtaining the planar Euclidean physical distance, the edge computing control unit performs a nonlinear attenuation calculation of the risk pheromone field based on the planar Euclidean physical distance and the set optical resolution attenuation constant. The update equation for the risk pheromone concentration is defined as:

[0114] ;

[0115] In the formula, Indicates the target coordinate raster at the timestamp Updated risk pheromone concentration; This indicates the dynamic risk pheromone concentration before the update; This represents the optical resolution attenuation constant, the value of which is determined by the pixel density of the image sensor of the pan-tilt camera, and the value range is set to 0.1 to 0.5; This represents the Euclidean physical distance between the center point of the target coordinate grid and the projection point of the optical center of the pan-tilt camera onto the static two-dimensional grid map plane, in meters.

[0116] After completing the concentration decay calculation, in this embodiment, the edge computing control unit synchronously updates the uninspected time integration start point corresponding to the target coordinate grid, i.e., the absolute timestamp of the last elimination action in the aforementioned basic risk pheromone field calculation, to the absolute timestamp of the current moment, so as to re-accumulate the risk in subsequent patrol cycles.

[0117] In this embodiment, considering that the image clarity acquired by the camera gradually decreases with increasing physical distance, the control unit's confidence in identifying distant environmental anomalies also decreases. A nonlinear exponential attenuation calculation logic is introduced to convert the spatial resolution of the optical sensor into a pheromone elimination ratio. Physical areas closer to the PTZ camera have higher image resolution and can achieve a larger pheromone elimination margin, with the updated concentration approaching zero. In more distant areas, due to decreased pixel resolution, the control unit retains a higher proportion of residual risk pheromones. Through this distance attenuation mechanism, the system can objectively reflect the actual impact of physical optical sensing characteristics on the inspection coverage quality.

[0118] In this embodiment, regarding the reset logic for the uninspected timestamp corresponding to the target coordinate grid pheromone concentration update, the edge computing control unit finally outputs the updated absolute timestamp matrix and writes it into local memory as the time base for the next cycle. The specific implementation process is completed through the following sub-steps:

[0119] S402, after the target coordinate grid pheromone concentration is updated, the corresponding uninspected timestamp reset logic is executed. In this embodiment, the basic judgment conditions for system clock reading and field coverage binarized mask matrix have been established. The edge computing control unit skips the intermediate matching judgment step and directly updates the conditions based on the current absolute timestamp and the mask matrix to determine the starting point of time integration.

[0120] The mathematical expression for the timestamp reset logic is defined as follows:

[0121] ;

[0122] In the formula, Indicates the coordinate grid after reset The absolute timestamp of the last time a risk pheromone removal action occurred; This represents the absolute timestamp of the current system moment extracted by the edge computing control unit, in seconds. This represents the field-of-view coverage binarization mask value generated in the preceding steps; Indicates the coordinate grid before resetting. The history retained in system memory has its absolute timestamps removed.

[0123] The updated timestamp data matrix is ​​directly written to local memory for persistent storage. In this embodiment, the edge computing control unit synchronously resets the time variables of the physical regions within the effective field of view through the mask condition equation, while areas not covered by the field of view retain their historical timestamps. The reset mechanism establishes a temporal correlation between the environmental risk re-accumulation process and the physical inspection actions corresponding to chassis movement and gimbal rotation, ensuring the continuity of the calculation process.

[0124] In this embodiment, the vector control and kinematic output step S500, which integrates hardware identifier perturbations, is implemented through the following sub-steps:

[0125] S501, within the local window of the current coordinates of the security robot, calculate the weighted difference between the gradient vector of the risk pheromone field and the gradient vector of the mutually exclusive pheromone field.

[0126] In this embodiment, the edge computing control unit uses the physical coordinates of the chassis center of the security robot as the origin and extracts a two-dimensional grid neighborhood of a set size to form a local window. Considering the balance between the noise resistance of gradient calculation and the computing power consumption of the edge computing control unit, in this embodiment, the physical size of the local window corresponds to the grid set covered by a square area with a side length ranging from 3m to 5m.

[0127] Within a localized window, risk pheromones in the environment guide the security robot to inspect high-risk areas, while mutually exclusive pheromones repel the robot from other operating equipment. The edge computing control unit extracts the spatial distribution concentrations of the two pheromone fields separately and combines their traction trends on the chassis through differential calculation.

[0128] The difference equation for the basic motion guide vector is defined as follows:

[0129] ;

[0130] In the formula, This represents the basic motion guidance vector generated in a two-dimensional planar coordinate system; This indicates the set risk pheromone traction weight coefficient, with a setting range of 1.0 to 2.0; This represents the positive spatial gradient vector of the risk pheromone field within the local window; This indicates the set mutual exclusion pheromone repulsion weight coefficient, with a setting range of 1.5 to 3.0; This represents the spatial gradient vector of the mutually exclusive pheromone field within a local window. For gradient solving in a discrete two-dimensional coordinate grid set, the edge computing control unit extracts the concentration difference between adjacent grids and uses a numerical difference algorithm to calculate the concentration change rates in the horizontal and vertical directions, synthesizing the gradient vector in the discrete space.

[0131] S502, extract the factory-fixed media access control address identifier, hash the media access control address identifier to a tangential deflection disturbance angle and generate an orthogonal disturbance vector.

[0132] In this embodiment, when multiple security robots meet in a narrow, symmetrical corridor, the positive gradient of the risk pheromone calculated within a local window is subtracted from the gradient of the mutually exclusive pheromone, resulting in a phenomenon of numerical symmetry cancellation. This numerical symmetry cancellation causes the calculated basic motion guidance vector to approach zero, causing the chassis control program to stagnate or oscillate.

[0133] To break the balance of vector computing in a symmetrical environment, in this embodiment, the edge computing control unit reads the media access control address that is factory-fixed in the network interface hardware as a unique identifier, and extracts the four bytes of the low address bits and converts them into a decimal integer for subsequent calculations to avoid data overflow.

[0134] The edge computing control unit uses a hash modulo algorithm to convert the media access control address into an independent tangential deflection perturbation angle. The mapping equation is defined as follows:

[0135] ;

[0136] In the formula, This indicates the generated tangential deflection disturbance angle; This indicates the maximum allowable deflection angle parameter, with a value setting range of 5° to 15°. This indicates the value of the low-order byte of the read media access control address after conversion to a decimal integer. This represents the set hash modulo constant, which is usually a large prime number; This indicates the remainder operation.

[0137] After obtaining the tangential deflection disturbance angle, in this embodiment, the edge computing control unit performs a deflection transformation on the basic motion guidance vector using a two-dimensional rotation matrix. Geometrically, the deflection action is equivalent to superimposing orthogonal disturbance components in the normal direction of the basic motion guidance vector, thereby generating the target motion vector after superimposing the orthogonal disturbance vector. The mathematical expression is defined as:

[0138] ;

[0139] In the formula, This represents the final calculated target motion vector. Represents the basic motion guiding vector; This indicates the tangential deflection disturbance angle.

[0140] In this embodiment, because the media access control addresses of different security robots are physically unique, the input hash variables are different. Two robots at symmetrical meeting positions will generate tiny deflection angles of different directions and magnitudes during calculation. The hardware-based perturbation mechanism avoids the complex negotiation and communication process between multiple robots, enabling the security robots to deflect to different sides, bypass each other, and continue performing their inspection tasks guided by the environmental risk gradient.

[0141] In this embodiment, the calculation steps for performing weighted vector superposition of combined attraction, repulsion, disturbance force, and kinematic inertia, and issuing low-level driving commands through the Jacobian inverse matrix, are specifically implemented through the following sub-steps:

[0142] S503, combining the chassis's kinematic inertia, a weighted vector superposition is performed on the target motion vector to generate the desired speed command. In this embodiment, the security robot chassis has mechanical inertia in the actual physical environment, making it impossible to instantly complete abrupt changes in speed direction and magnitude between adjacent control cycles. To avoid mechanical impact on the chassis motors and tire slippage, the edge computing control unit extracts the historical speed vector recorded by the system from the previous control cycle and performs a first-order low-pass filtering weighted fusion with the currently calculated target motion vector.

[0143] The mathematical expression for the weighted superposition of kinematic inertia is defined as:

[0144] ;

[0145] In the formula, This represents the desired velocity vector after inertial smoothing. This represents the set kinematic smoothness coefficient, the value of which is determined by the chassis mass and the maximum torque of the motor, and the value range is set from 0.2 to 0.6; This represents the target motion vector generated by the superposition of orthogonal perturbation vectors in the preceding steps; This represents the historical expected velocity vector output by the system in the previous control cycle.

[0146] In this embodiment, after calculating the desired velocity vector, the edge computing control unit directly extracts the desired linear velocity of the chassis by obtaining the magnitude of the desired velocity vector. The mathematical expression is defined as:

[0147] ;

[0148] Meanwhile, the edge computing control unit uses the arctangent function to solve for the longitudinal and lateral components of the desired velocity vector, and extracts the target heading angle of the desired velocity vector in a two-dimensional coordinate system. The desired angular velocity is calculated by subtracting the current heading angle of the security robot from its actual heading angle, and then combining this with the set angular velocity proportional gain. The mathematical expression is defined as:

[0149] ;

[0150] In the formula, This represents the desired linear velocity of the chassis, expressed in m / s. This represents the magnitude of the desired velocity vector; This represents the desired angular velocity of the chassis, expressed in rad / s. This indicates the set angular velocity proportional gain, with a value range of 0.5 to 1.5; This represents the target heading angle corresponding to the desired velocity vector; This indicates the current actual heading angle of the security robot.

[0151] S504: Construct a chassis kinematic model and convert the desired speed command into drive commands for the underlying motors using the Jacobian inverse matrix. This embodiment uses a two-wheel differential drive chassis as an example for specific implementation. To transform the abstract two-dimensional vector motion into the mechanical movements of the physical chassis, the edge computing control unit needs to decompose the desired planar linear velocity and angular velocity into the specific rotational angular velocities of the left and right drive motors.

[0152] Based on the rigid body kinematics of a two-wheel differential chassis, the Jacobian inverse matrix equation for the conversion of desired speed to bottom wheel speed is defined as:

[0153] ;

[0154] In the formula, and These represent the expected rotational angular velocities of the right and left drive motors, respectively, in rad / s. The Jacobian inverse matrix represents the kinematic model of the two-wheel differential speed; ; This represents the desired angular velocity of the chassis, expressed in rad / s. This indicates the physical radius of the security robot's drive wheels, in meters (m). This indicates the physical wheel distance between the centers of the left and right drive wheels of the security robot, in meters (m).

[0155] In this embodiment, the Jacobian inverse matrix establishes a physical mapping between the vehicle's central motion state and the independent motion states of the left and right wheels. By solving inversely, the control unit can explicitly control the speed difference between the left and right drive wheels, thereby synthesizing the desired straight-line forward movement and steering action on the physical ground.

[0156] After the calculation is completed, in this embodiment, the edge computing control unit sends the desired rotational angular velocity commands of the left and right motors to the corresponding underlying motor drivers via the controller area network bus. The motor drivers perform voltage regulation, driving the chassis to move according to the final motion vector that integrates environmental risk pheromone guidance, mutual exclusion pheromone avoidance, and hardware anti-lockdown perturbation.

[0157] Specific application examples:

[0158] The scenario involves two security robots traveling towards each other in a narrow, symmetrical corridor, facing a deadlock due to their impending collision. The physical wheelbase constants of the security robots' chassis are set. The physical radius of the drive wheel is 0.4m. The value is 0.1m. The edge computing control unit performs the following calculation steps to break the deadlock.

[0159] First, calculate the mutual exclusion pheromone field concentration. Assume the current obstacle coordinate set contains only one oncoming security robot entity, and the total number of entities is... Set the mutual exclusion peak constant to 1. The value is 100, and the mutual exclusion influence radius constant is 2m. The planar Euclidean distance from the target coordinate grid extracted by the edge computing control unit to the physical coordinates of the entity's center is 2m. Substituting this into the mutual exclusion pheromone concentration calculation equation:

[0160] ;

[0161] Substitute specific values:

[0162] ;

[0163] Faced with an opposing anti-collision repulsive force, the basic motion guidance vector approaches a standstill. The basic motion guidance vector is set to a horizontal forward state, with the horizontal component set to 1 and the vertical component to 0. At this point, the hardware identifier perturbation mechanism is triggered. The low-order byte of the media access control address preset at the factory of the local network interface is read, converted to a decimal integer, and set to this value. The value is 12500. The maximum allowable deflection angle is set to 10 degrees, and the hash modulo constant is [value missing]. The value is 10000. Substituting this into the tangential deflection perturbation angle mapping equation:

[0164] ;

[0165] Substitute specific numerical values ​​to calculate the remainder and disturbance angle:

[0166] ;

[0167] After obtaining the negative 5-degree tangential deflection disturbance angle, a deflection transformation is performed on the basic motion guide vector using a two-dimensional rotation matrix. The deflection equation is defined as follows:

[0168] ;

[0169] Substituting the cosine value of -5 degrees, 0.996, with the sine value of -0.087, into the matrix multiplication:

[0170] ;

[0171] Subsequently, a first-order low-pass filter weighted fusion is performed based on the chassis kinematics and inertia. The kinematic smoothing coefficient is set to 0.5, and the lateral component of the historical expected velocity vector output from the previous control cycle is 0.8 while the longitudinal component is 0. Substituting these values ​​into the kinematic inertia weighted superposition equation:

[0172] ;

[0173] Substitute specific values:

[0174] ;

[0175] Based on the above results, the equation for the desired linear velocity of the chassis is extracted:

[0176] ;

[0177] Substitute specific numerical values ​​to calculate the vector magnitude:

[0178] ;

[0179] At this point, the calculated target heading angle corresponding to the desired velocity vector is -0.048 radians. Assume the actual heading angle of the vehicle is 0 radians, and the angular velocity proportional gain is 1.0. Substitute these values ​​into the equation for calculating the desired angular velocity:

[0180] ;

[0181] Substitute specific values:

[0182] ;

[0183] Finally, the two-dimensional abstract vector is transformed into the physical drive commands for the underlying motor. Substituting this into the Jacobian inverse matrix equation:

[0184] ;

[0185] Substitute the wheel track and radius constants, as well as the desired linear velocity and desired angular velocity values:

[0186] ;

[0187] After receiving the command, the left drive motor's angular velocity increased to 9.086 radians per second, while the right drive motor's angular velocity decreased to 8.894 radians per second. This difference in speed caused the chassis to produce a slight and smooth deflection to the right, successfully breaking the spatial symmetry deadlock and achieving safe avoidance.

[0188] Experimental verification and effect comparison:

[0189] To verify the actual technical effect of the present invention, a construction area of ​​500m² was used. 2 A structured inspection test environment was established. The environment included multiple narrow corridors, each only wide enough for a single robot to pass through at a time. Three identical security robots were deployed to perform continuous, uninterrupted inspection tasks, with a total test period of 72 hours. The control group used a traditional artificial potential field method for path planning, while the experimental group used the intelligent inspection path optimization method proposed in this invention, which couples hardware identifier perturbation with optical attenuation masks. The performance data of the two algorithms on three core technical indicators were recorded and compared.

[0190] The first metric is the average time to resolve deadlocks during multi-aircraft encounters. In the control group, when encountering each other in a narrow corridor, the chassis experienced repeated oscillations due to the complete symmetrical cancellation of the potential field gradients, resulting in an average deadlock resolution time as high as 45 seconds. In some extreme cases, manual intervention was ultimately required. The experimental group generated orthogonal perturbation vectors using network interface physical address hash mapping. The chassis's expected linear velocity and target heading angle response curves when triggering the anti-deadlock perturbation mechanism are shown in the attached figure. Figure 5 As shown in the attached figure, the dynamic waveforms of the expected rotational angular velocities of the left and right drive motors at the bottom layer are as follows. Figure 6 As shown, the experimental group was able to break the spatial force field balance within 0.2s and achieve smooth differential separation of the left and right motors under the kinematic filtering weighted fusion effect, with an average release time of only 1.5s. The obstacle avoidance response efficiency was improved and no mechanical impact was caused.

[0191] The second metric is the convergence time for global inspection coverage. Due to the lack of a nonlinear time integration mechanism for the physical field of view, the control group exhibited inefficient random walk characteristics in open areas, requiring 120 minutes to achieve 95% environmental coverage. The experimental group, utilizing a two-dimensional camera field-of-view mask projection and a nonlinear exponential decay model, accurately eliminated residual risk pheromones in effectively observed areas, ensuring the chassis was consistently and strongly attracted to uninspected high-risk areas. Achieving the same coverage rate took only 45 minutes, improving inspection efficiency by over 60%.

[0192] The third indicator is the fault-tolerant operational stability in islanded mode. In the network outage blind zone test where local wireless network communication was artificially cut off, the control group robot experienced 12 physical collisions and severe trajectory overlap due to the loss of global node coordinate status. The experimental group's edge computing control unit quickly generated a Gaussian spatial decay mask based on the network confidence flag, smoothly downgrading global collaborative inspection to local area safe stationary patrol, achieving 0 collisions during the disconnection period and ensuring the high availability of the security group system architecture.

Claims

1. A method for optimizing the intelligent inspection path of a security robot, characterized in that, Includes the following steps: Acquire a static two-dimensional grid map of the inspection area, send and receive parameter messages in the local area communication network, and update the network confidence flag based on the received frequency status; Based on the static two-dimensional raster map and the network confidence flag, a risk pheromone field with superimposed spatial decay mask is constructed, and a mutually exclusive pheromone field is constructed. Extract the PTZ camera parameters from the parameter message, combine them with the static two-dimensional grid map to perform ray projection, and calculate and generate an effective field of view polygon region. For the coordinate grid within the effective field of view polygonal region, the risk pheromone field is updated nonlinearly based on the physical distance between the coordinate grid and the PTZ camera. The basic motion guidance vector is generated by calculating the positive gradient vector of the risk pheromone field and the negative gradient vector of the mutually exclusive pheromone field. The medium access control address identifier is extracted and mapped to generate an orthogonal disturbance vector, which is then combined with the basic motion guidance vector to generate a target motion vector. The target motion vector is then weighted and superimposed with the chassis kinematic smoothing vector. A speed command is generated through the inverse kinematics of the chassis to control the chassis of the security robot to move.

2. The method for optimizing the intelligent inspection path of a security robot according to claim 1, characterized in that, The step of updating the network confidence flag based on the received frequency status specifically includes: Set the message reception timeout threshold to the number of communication synchronization cycles in which no messages are received consecutively; If the dynamic cache queue is polled at a fixed frequency and no valid parameter message of the external neighboring node is updated within the time span corresponding to the message reception timeout threshold, the network confidence flag is overwritten to the first state value corresponding to the isolated operation mode outside the cluster. If a valid parameter message is obtained during the polling process, the network confidence flag is set to the second state value corresponding to the multi-machine collaborative operation mode.

3. The method for optimizing the intelligent inspection path of a security robot according to claim 2, characterized in that, The construction of the risk pheromone field with superimposed spatial attenuation mask specifically includes: Based on the environmental basic risk constant, the static environmental volatility coefficient, and the integral term of the dynamic external trigger weight with the time without inspection, the basic risk pheromone concentration is calculated and generated. A spatial decay mask is dynamically generated based on the state value of the network confidence flag. The basic risk pheromone concentration is multiplied by the spatial decay mask to generate the risk pheromone field.

4. The method for optimizing the intelligent inspection path of a security robot according to claim 3, characterized in that, The dynamic generation of the spatial attenuation mask specifically includes: When the network confidence flag is the second state value representing the multi-machine collaborative operation mode, the spatial attenuation mask value is always one. When the network confidence flag is the first state value representing the operation mode of leaving the cluster island, a spatial decay mask is generated by using the Gaussian decay function based on the Euclidean distance between the current physical coordinates of the security robot and the coordinate grid and the set spatial tolerance radius. The value of the mask decreases exponentially as the Euclidean distance increases.

5. The method for optimizing the intelligent inspection path of a security robot according to claim 2, characterized in that, The construction of the mutually exclusive pheromone field specifically includes: When the network confidence flag is the state value corresponding to the multi-machine collaborative operation mode, the global physical coordinates of all surrounding security robots are extracted as the obstacle coordinate set; When the network confidence flag is the state value corresponding to the isolated operation mode, the two-dimensional coordinate points of the dynamic obstacle clustering center detected by the three-dimensional lidar are projected onto the plane coordinate system as the obstacle coordinate set. Based on the planar Euclidean distance between the target coordinate grid in the static two-dimensional grid map and the two-dimensional coordinate points of each entity in the obstacle coordinate set, combined with the mutual exclusion peak constant and the mutual exclusion influence radius, the cumulative value of the repulsion concentration is calculated using a Gaussian function to generate the mutual exclusion pheromone field.

6. The method for optimizing the intelligent inspection path of a security robot according to claim 1, characterized in that, The step of extracting the pan-tilt camera parameters from the parameter message, combining them with the static two-dimensional grid map to perform ray casting, and calculating and generating an effective field-of-view polygon region specifically includes: The relative yaw angle of the pan-tilt camera is extracted and added to the global heading angle of the chassis to establish the angular reference for ray projection, and a virtual ray is emitted along the central optical axis to the static two-dimensional grid map; By combining the pitch angle, physical installation height, maximum sensing distance parameters of the PTZ camera and the planar Euclidean distance to the coordinate grid of the first static obstacle, the final effective projection length of a single virtual ray is calculated, and the coordinates of the cutoff point are obtained. Connect adjacent cut-off points sequentially, and connect the first and last cut-off points to the physical coordinates of the chassis center to close and generate the effective field-of-view polygon region.

7. The method for optimizing the intelligent inspection path of a security robot according to claim 6, characterized in that, After closing and generating the effective field-of-view polygon region, the method further includes: Extract the maximum and minimum values ​​of the x-coordinate and the maximum and minimum values ​​of the y-coordinate from the coordinates of all ray cutoff points to construct a two-dimensional minimum bounding rectangle; For a coordinate grid located inside the two-dimensional minimum bounding rectangle, if the coordinate grid belongs to the effective field of view polygon region and the static obstacle occupancy attribute is no obstacle, then the corresponding field of view coverage binarization mask value is set to one; otherwise, it is set to zero, and the field of view coverage binarization mask matrix is ​​output.

8. The method for optimizing the intelligent inspection path of a security robot according to claim 7, characterized in that, The nonlinear decay update of the risk pheromone field specifically includes: Filter the coordinate grids whose field of view covers a binary mask value of one, and calculate the Euclidean physical distance between the center point of the coordinate grid and the optical center of the pan-tilt camera projected onto the static two-dimensional grid map. Based on the Euclidean physical distance and optical resolution attenuation constant, a nonlinear exponential function is used to calculate the pheromone elimination ratio. Based on the pheromone elimination ratio, a nonlinear attenuation calculation is performed on the risk pheromone concentration before the update to obtain the updated risk pheromone concentration. Based on the absolute timestamp of the current moment and the binarized mask value of the field of view coverage, reset the starting point of the uninspected time integration corresponding to the coordinate grid.

9. The method for optimizing the intelligent inspection path of a security robot according to claim 1, characterized in that, The process of calculating the positive gradient vector of the risk pheromone field and the negative gradient vector of the mutually exclusive pheromone field to generate a basic motion guidance vector, extracting the medium access control address identifier mapping to generate an orthogonal perturbation vector, and combining it with the basic motion guidance vector to generate a target motion vector specifically includes: Within the local window of the current coordinates of the security robot, the concentration difference between adjacent grids is extracted to calculate the positive spatial gradient vector of the risk pheromone field and the spatial gradient vector of the mutually exclusive pheromone field, and weighted difference is performed to generate the basic motion guidance vector. Read the Media Access Control address that is factory-set on the network interface, extract the low address byte and convert it into a decimal integer, and use a hash modulo algorithm to convert the decimal integer into a tangential deflection disturbance angle; Using a two-dimensional rotation matrix containing the tangential deflection disturbance angle, a deflection transformation is performed on the basic motion guidance vector to generate the target motion vector superimposed with the orthogonal disturbance vector.

10. The method for optimizing the intelligent inspection path of a security robot according to claim 1, characterized in that, The step of weighted superposition of the target motion vector and the chassis kinematics smoothing vector, and generating a speed command through inverse kinematics of the chassis, specifically includes: Extract the historical velocity vector from the previous control cycle and perform first-order low-pass filtering and weighted fusion with the target motion vector to generate the desired velocity vector; The magnitude of the desired velocity vector is obtained as the desired linear velocity of the chassis. The target heading angle corresponding to the desired velocity vector is extracted using the arctangent function. The desired angular velocity is calculated by combining the current actual heading angle of the vehicle head with the angular velocity ratio gain. The desired linear velocity and the desired angular velocity are decomposed and solved using the Jacobian inverse matrix of the two-wheel differential kinematic model, which is composed of the physical wheel track and the physical radius of the drive wheel, to obtain the desired rotational angular velocity commands of the left and right drive motors.

Images