Machine dog cooperative networking control method based on environment model in public network limited environment

Abstract

The application relates to the fields of mobile robots and wireless ad hoc networks, and discloses a robot dog cooperative networking control method based on an environment model in a public network limited environment. The method acquires a three-dimensional environment model with object attributes constructed in a patrol task, a robot dog state and a positioning navigation process, and generates a candidate networking position; the wireless propagation quality is predicted according to the object geometry, material and shielding relationship in the propagation path, the irregular effective coverage area is determined, and a communication connection graph is constructed; under the constraints of route coverage, link capacity, positioning credibility, energy and residence safety, a relay robot dog, a target networking position and a forwarding path are determined, a single-hop or multi-hop communication network is established; the propagation parameters are corrected according to actual link measurement results, and the networking position or the forwarding path is dynamically adjusted. Therefore, the accuracy of networking position selection in a complex environment and the continuity and reliability of patrol communication are improved.

Description

Technical Field

[0001] This application relates to the fields of mobile robots, wireless ad hoc networks, and cooperative positioning and navigation technologies, and in particular to a robot dog cooperative networking control method based on an environment model in a public network-restricted environment. Background Technology

[0002] Robot dogs are characterized by strong obstacle-crossing capabilities, the ability to carry multiple sensors, and the capacity to enter confined spaces. They have been applied in scenarios such as power plant, factory, mine, underground utility tunnel inspection, and emergency reconnaissance. When performing inspection tasks, robot dogs need to continuously acquire positioning and navigation information and transmit control commands, status data, maps, images, and detection results to the inspection station.

[0003] Existing robotic dogs typically use Real-Time Kinematic (RTK) positioning technology for outdoor positioning, and Simultaneous Localization and Mapping (SLAM) technology for indoor, underground, or areas with obstructed satellite signals, utilizing inertial measurement units for short-term state extrapolation. However, simply switching positioning methods based on the presence of RTK signals can easily lead to frequent switching or coordinate jumps in critical areas; and maintaining large-scale, high-resolution SLAM maps over a long period also increases the computational, storage, and communication burden.

[0004] In terms of communication, 4G / 5G public networks may be unavailable underground, inside factories, or in remote areas. The actual coverage of long-distance image and data transmission equipment is also affected by factors such as walls, metal equipment, terrain, and dynamic obstacles, and is therefore usually irregularly distributed, making it difficult to accurately determine based on the nominal communication distance.

[0005] Existing multi-robot networking methods often rely on node distance, fixed location, or current signal strength to deploy relay nodes. This makes it difficult to reflect the material of objects in the propagation path, penetration thickness, diffraction, and Fresnel zone obstruction. It also fails to fully consider the robot dog's reachability, dwell safety, positioning reliability, remaining battery power, and return capability, which can easily lead to problems such as unusable relay locations, unstable links, or node redundancy.

[0006] Furthermore, existing solutions typically lack mechanisms to correct coverage predictions and dynamically adjust relay locations and forwarding paths based on actual link measurements. When both public network links and data transmission links are interrupted, it is difficult to make security decisions considering location, energy, storage space, and route risks, and communication interruptions may occur due to improper relay node adjustments or withdrawal sequences.

[0007] Therefore, how to accurately predict the actual wireless coverage of candidate locations in a public network-restricted environment, jointly determine the relay robot, network location and forwarding path, and dynamically adjust the network based on the measured link results has become a technical problem that needs to be solved. Summary of the Invention

[0008] The purpose of this application is to provide a robot dog cooperative networking control method based on an environment model in a public network-restricted environment, in order to solve the problems that when relay nodes are arranged according to nominal distance or fixed location, the actual occlusion relationship cannot be accurately reflected, the networking location is disconnected from navigation and energy conditions, the network cannot be corrected according to the measured link closed loop after the network is established, and the positioning and data transmission services are difficult to maintain continuously in public network-restricted areas.

[0009] To achieve the above objectives, the first aspect of this application provides a method for collaborative networking control of robot dogs based on an environment model in a public network-restricted environment, applicable to a collaborative networking system including an inspection station and multiple robot dogs. The method includes the following steps.

[0010] Acquire inspection task information, status information of each robot dog, and a 3D environment model with object attributes constructed by at least one robot dog based on environmental perception data during the positioning and navigation process.

[0011] Candidate network locations are generated based on the passable areas and safe conditions for staying in the 3D environment model.

[0012] For each candidate network location, the wireless propagation quality is predicted based on the geometric and material information of objects along the propagation path between the candidate network location and the evaluation points. The set of evaluation points that meet the preset service quality threshold is determined as the effective coverage area of ​​the candidate network location. Evaluation points include at least one of the following: inspection route sampling points, task points, inspection station locations, and other candidate network locations.

[0013] Based on the effective coverage areas, a communication connection map is constructed. Under the constraints of inspection route coverage, end-to-end connectivity, link capacity, positioning reliability, robot dog energy, and stationary security, the relay robot dog, the target network location of the relay robot dog, and the forwarding path are determined from multiple robot dogs.

[0014] Control the relay robot dog to move to the corresponding target networking location, and establish a single-hop or multi-hop communication network between the inspection station and the robot dog performing the inspection task through a long-distance image and data transmission link.

[0015] Obtain the actual link measurement results of the communication network, update the wireless propagation prediction parameters based on the difference between the actual link measurement results and the predicted wireless propagation quality, redetermine the corresponding effective coverage area based on the updated wireless propagation prediction parameters, and adjust at least one of the target network location and forwarding path according to the redetermined effective coverage area and the robot dog's operating status.

[0016] Optionally, environmental perception data includes lidar point clouds, millimeter-wave radar echo data, and multispectral images acquired by a penta-lens camera. The environmental units of the 3D environment model record occupancy status, 3D geometric location, object category, material category, equivalent thickness, static or dynamic attributes, wireless attenuation parameters, identification confidence level, and last update time.

[0017] Optionally, the wireless propagation quality is determined based on free space propagation loss, object penetration loss, diffraction loss, first Fresnel zone occlusion loss, dynamic occlusion loss, and communication margin; the evaluation points in the effective coverage area need to ensure that both the forward and reverse links meet the corresponding service's received power, communication success probability, available rate, latency, and packet loss rate thresholds.

[0018] Optionally, candidate network locations are obtained through spatial sampling after eliminating areas from the passable area that do not meet the requirements for slope, habitable area, hazard level, distance from obstacles, or return energy. Candidate network status may also include aircraft altitude, pitch attitude, heading angle, image and data transmission antenna height, and image and data transmission antenna orientation. The height or orientation of the image and data transmission antenna can be adjusted using an adjustable bracket or the robotic arm of a robot dog.

[0019] Optionally, actual link measurement results include transmit location, receive location, environmental model version, operating frequency or channel, transmit power, antenna status, received signal strength, signal-to-noise ratio, packet loss rate, actual data rate, and latency. Based on the propagation residual and the material type and equivalent thickness of the object through which the propagation path passes, the material attenuation coefficient, regional additional loss, and propagation uncertainty are updated.

[0020] Optionally, dynamic network reconfiguration can be triggered when link quality, link load, remaining battery power of the robot dog, robot dog pose uncertainty, or changes in the 3D environment model reach corresponding conditions. The dynamic network reconfiguration adopts a build-then-dismantle approach, first establishing an alternative link, then switching the forwarding path, and finally releasing the original relay node.

[0021] Optionally, the communication network is also used to transmit RTK differential data, anchored sub-maps, local keyframes, and coordinate transformation relationships. The robot dog switches or adjusts the fusion weights of RTK, SLAM, and inertial navigation based on the RTK positioning quality and SLAM positioning quality, and selects the transmission link based on the service quality of the 4G / 5G link and the long-distance map and data transmission link; when both types of links are unavailable, it continues to execute the pre-stored route and cache inspection data if the offline execution conditions are met, and resumes transmission after the link is restored.

[0022] A second aspect of this application provides a robot dog, including a body, a motion actuator, a processor, a memory, and an industry expansion module disposed on the body. The industry expansion module includes an RTK positioning component, an inertial measurement component, at least one of a lidar and millimeter-wave radar for SLAM mapping and environmental perception, a long-range map and data transmission component, a 4G / 5G communication component, and a multispectral imaging component composed of five lenses. The memory stores a computer program; when the processor executes the computer program, it controls the motion actuator and the industry expansion module, enabling the robot dog to collaboratively perform the aforementioned methods with inspection stations and other robot dogs.

[0023] A third aspect of this application provides a robot dog collaborative networking system, including an inspection station and multiple robot dogs. The inspection station is used to issue inspection tasks, maintain a 3D environment model, and aggregate inspection data. At least one of the processors of the inspection station and each robot dog is configured to construct a 3D environment model with object attributes, predict the effective coverage area of ​​candidate network locations, determine the relay robot dog and its target network location and forwarding path, control the relay robot dog to establish a single-hop or multi-hop communication network, and correct the effective coverage area and adjust the target network location or forwarding path based on actual link measurement results.

[0024] The fourth aspect of this application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the aforementioned robot dog cooperative networking control method.

[0025] Compared with the prior art, this application has at least the following beneficial effects.

[0026] First, the 3D environment model with object attributes built during the robot dog's positioning and navigation process is reused for wireless propagation analysis. The nominal communication distance of the long-distance image and data transmission component is converted into an irregular effective coverage area corresponding to specific objects, materials, obstructions, and service thresholds, thereby improving the accuracy of relay location judgment.

[0027] Second, by combining effective coverage area, inspection route, end-to-end connectivity, link capacity, positioning reliability, mobile energy consumption, return capability, and dwell security conditions to determine the location of the relay robot dog and the network, it is possible to reduce invalid relay nodes and unnecessary mobile energy consumption.

[0028] Third, by using the actual link measurement results after networking, the material attenuation coefficient, regional additional loss and propagation uncertainty are corrected, and the effective coverage area is recalculated, which can adapt to material identification errors, on-site changes and deviations between nominal communication capabilities and actual propagation environment.

[0029] Fourth, by using a dynamic reconfiguration method of building first and then dismantling the network, as well as capacity scheduling of the main path and auxiliary path, the risk of communication interruption caused by relay node relocation can be reduced, and interference caused by unconditionally increasing co-frequency relay nodes can be avoided.

[0030] Fifth, by using quality-based RTK, SLAM and inertial navigation fusion, coordinate transformation of overlapping areas and hierarchical sub-map mechanisms, the computational burden caused by positioning jumps and large-scale complete SLAM maps can be reduced.

[0031] Sixth, by selecting the quality of service (QoS) of 4G / 5G links and long-distance image and data transmission links, determining offline task security, resuming data transmission at the data block level, and suppressing duplicate data, the continuity of inspection tasks and data transmission can be improved.

[0032] Seventh, by deploying relay nodes from near to far and retrieving relay nodes from far to near, necessary positioning and communication services can be maintained when the robot dog enters restricted areas of the public network and returns home. Attached Figure Description

[0033] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0034] Figure 1 A flowchart illustrating the overall process of the robot dog collaborative networking control method provided in this application embodiment;

[0035] Figure 2 This is a schematic diagram of the structure of the robot dog collaborative networking system provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the robot dog and industry expansion module provided in the embodiments of this application; Figure 4 A schematic diagram of the three-dimensional environment model, candidate network locations, and irregular effective coverage areas provided in the embodiments of this application; Figure 5 A flowchart for wireless propagation quality prediction provided in an embodiment of this application; Figure 6 The communication connection diagram and relay location selection diagram provided for the embodiments of this application; Figure 7 This is a flowchart of the measured link feedback correction process provided in the embodiments of this application; Figure 8 A flowchart of a dynamic reconfiguration network built before being dismantled is provided for an embodiment of this application; Figure 9Flowchart of RTK, SLAM and inertial navigation cooperative localization provided in the embodiments of this application; Figure 10 The flowchart illustrates the multi-link selection, offline caching, and breakpoint resumption process provided in the embodiments of this application. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] Public network restricted environments refer to environments where 4G / 5G public networks are unavailable, quality is unstable, available speeds are insufficient, latency or packet loss rates do not meet current service requirements, and the expected connection duration is too short. Long-range image and data transmission links are wireless links established by a long-range image and data integrated transmission component mounted on the robot dog. These can be point-to-point links or support Ad Hoc, Mesh, or other multi-hop forwarding methods. The stated 10 km range only represents an example capability under ideal line-of-sight conditions and is not a fixed radius of the effective coverage area.

[0038] Evaluation points are spatial points used to determine whether a candidate network location can provide communication services. These can include discrete sampling points along the inspection route, inspection task points, inspection station locations, the current location of the robot dog, other candidate network locations, and locations where there is a risk of connection loss. The effective coverage area is a set of evaluation points that, starting from a candidate network state, meet the corresponding service thresholds for both forward and reverse link received power, communication success probability, available data rate, latency, and packet loss rate.

[0039] Candidate network locations represent positions that the robot dog can reach and stably reside in. Candidate network states are formed by further incorporating factors such as aircraft altitude, pitch attitude, heading angle, image / data transmission antenna height, and image / data transmission antenna orientation into the candidate network locations. Target network locations are assigned to the relay robot dog after joint screening based on route coverage, connectivity, capacity, positioning, energy, and security conditions, rather than being absolutely optimal locations unaffected by constraints.

[0040] like Figure 2 As shown, the robot dog collaborative networking system includes an inspection station 100, multiple robot dogs 200, and a collaborative networking control device 300 implemented by one or more processors in the inspection station 100 and robot dogs 200. The inspection station 100 can be a fixed inspection station or a vehicle-mounted mobile inspection station, and can simultaneously connect to the public network, a local private network, and a long-distance image and data transmission network.

[0041] The inspection station 100 includes a task management module 110, an environment model management module 120, a network decision module 130, and a data aggregation module 140. The task management module 110 is used to issue inspection routes, task points, prohibited areas, estimated task duration, and service quality requirements. The environment model management module 120 is used to store the global environment model, sub-maps, model versions, and historical link measurement records. The network decision module 130 is used to calculate or assist in calculating the relay robot dog, target network location, and forwarding path. The data aggregation module 140 is used to receive and verify the maps, status, images, videos, alarms, and operation results uploaded by each robot dog.

[0042] Multiple Robot Dog 200 units can dynamically assume the roles of gateway node, relay node, inspection node, or backup node. Gateway nodes connect to inspection stations or the public network and act as the root node of the ad hoc network; relay nodes reside in or move to the target network location and forward data; inspection nodes move along the task route and perform inspections; backup nodes establish alternative or auxiliary links when link quality deteriorates, capacity is insufficient, node power is low, or a node fails. The same Robot Dog can switch roles at different task stages.

[0043] like Figure 3 As shown, the robot dog 200 includes a body 210, a motion actuator 220, an industry expansion module 230, a processor 239, and a memory 240. The industry expansion module 230 can be detachably installed on the back of the body 210 or other load-bearing locations, and includes corresponding components in an RTK positioning component 231, an inertial measurement component 232, a lidar 233, a millimeter-wave radar 234, a long-range image and data transmission component 235, a 4G / 5G communication component 236, a five-lens camera 237, and a robotic arm 238.

[0044] It should be noted that in some cases, the processor 239 and memory 240 can also be located in the industry expansion module 230.

[0045] The penta-light lens 237 can be understood as a multispectral imaging component with five optical or spectral channels configured according to the device. Each channel can be configured from visible light, near-infrared, short-wave infrared, thermal infrared, ultraviolet, or other channels suitable for inspection and material identification. In addition to acquiring inspection images, the penta-light lens 237 can also assist in identifying metals, concrete, glass, wood, vegetation, and other materials based on the differences in the response of objects in different channels.

[0046] The robotic arm 238 can be used to operate equipment, sample, or handle anomalies. In one implementation, the antenna of the long-range image and data transmission component 235 is mounted on an adjustable bracket; in another implementation, the antenna or antenna carrier is located at the end of the robotic arm 238, and the robotic arm 238 can be used to change the antenna height or orientation to improve the line-of-sight conditions of the propagation path or the first Fresnel zone clearance. If the equipment does not support antenna adjustment, only the position and body attitude of the robot dog are considered as candidate variables.

[0047] Processor 239 can perform one or more of the following: environmental modeling, candidate location generation, propagation prediction, path planning, link measurement, data caching, and role control. The collaborative networking control device 300 can be centrally deployed at inspection station 100 or distributed among inspection station 100 and multiple robot dogs 200; the different computing locations do not change the data relationships between the various processing steps.

[0048] This embodiment provides a detailed description of the robot dog collaborative networking control method as defined in claim 1.

[0049] like Figure 1 As shown, the cooperative networking control method of this embodiment includes steps S110 to S220. Each step can be executed in the following order, or it can be executed in parallel or cyclically if data dependencies allow.

[0050] Step S110: Obtain inspection task information and robot dog status information.

[0051] The inspection task information should at least include the inspection station location, inspection route, task points, prohibited areas, estimated data generation, and the transmission rate, latency, reliability, and priority requirements for control commands, positioning maps, alarms, key images, and continuous video. The robot dog's status information should at least include its current location, pose uncertainty, RTK positioning quality, SLAM positioning quality, remaining battery power, available storage space, current role, available communication interfaces, and current link quality.

[0052] Step S120: Obtain an existing 3D environment model, and control at least one robot dog to collect lidar point clouds, millimeter-wave radar echoes and multispectral images for unknown or changed areas, in order to establish or update a 3D environment model with object attributes.

[0053] This 3D environment model is used simultaneously for positioning and navigation, accessible area identification, and wireless propagation analysis.

[0054] Step S130: Generate candidate network locations from the passable areas in the 3D environment model, and generate one or more candidate network states for each candidate network location based on the device capabilities.

[0055] Step S140: Based on the geometry, material, penetration thickness, diffraction edge, Fresnel zone intrusion degree, and dynamic occlusion probability of objects on the propagation path between the candidate network state and each evaluation point, predict the wireless propagation quality of the forward and reverse links to obtain the irregular effective coverage area of ​​each candidate network state.

[0056] Step S150: Read historical link measurement records or collect new link measurement results online, compare the actual link measurement results with the predicted results, update the material attenuation parameters, regional additional loss and propagation uncertainty, and recalculate the affected effective coverage area.

[0057] If there are no historical measurement records for the initial task, a conservative prediction can be made using preset material parameters and a large communication margin.

[0058] Step S160: Using inspection stations, candidate network status, inspection route sampling points, and task points as graph nodes, establish connection edges based on whether the nodes meet the bidirectional service quality threshold to form a communication connection graph.

[0059] Each connection edge records the link margin, expected capacity, latency, packet loss rate, probability of successful communication, propagation uncertainty, and forwarding energy consumption.

[0060] Step S170: Under the constraints of inspection route coverage, end-to-end connectivity, link capacity, positioning reliability, robot dog energy, and stationary security, determine the relay robot dog, target network location, and forwarding path.

[0061] For high-risk tasks or control operations, it may be further required that there be redundant paths with nodes that do not completely overlap.

[0062] Step S180: Deploy relay robot dogs in a step-by-step manner from the inspection station to the task area. After each relay robot dog reaches the target network location, it first establishes a bidirectional link with the upstream node and performs a test verification. After successful verification, the next relay robot dog or inspection robot dog continues to move outward. If the tested link does not meet the threshold, the position, attitude, or antenna direction is finely adjusted within the neighborhood of the current candidate location.

[0063] In step S190, the inspection robot dog performs its tasks along the inspection route and outputs continuous pose based on RTK positioning quality, SLAM positioning quality, and inertial navigation status. The cooperative communication network synchronously transmits RTK differential data, anchored sub-maps, local keyframes, coordinate transformation relationships, and route updates.

[0064] Step S200: Select a transmission link based on the service quality of the 4G / 5G link and the long-distance image and data transmission link. If both types of links are unavailable, decide whether to continue the inspection, return to the most recently connected location, suspend the task, or enter a safe dwell state, based on the location reliability, remaining battery power, available storage space, and route risk.

[0065] Step S210: Monitor link quality, link load, node power, node pose uncertainty, storage space, changes in the environmental model, and changes in inspection tasks. When any trigger condition is met, update the coverage model and perform dynamic reconfiguration of the network, building first and then dismantling later.

[0066] Step S220: After the task is completed, retransmit any unconfirmed data and retrieve the relay robot dogs in order from farthest to nearest. Before the remote relay node leaves, confirm that downstream data has been synchronized to upstream, an alternative path has been established, or the downstream node has entered a directly communicable area.

[0067] In some embodiments, the 3D environment model may employ point clouds, voxel grids, occupancy grids, triangular meshes, semantic maps, or combinations thereof. Each environment cell, in addition to recording occupancy status and 3D geometric position, also records object category, material category, equivalent thickness, surface normal, static or dynamic attributes, wireless attenuation parameters, identification confidence, last update time, and data source.

[0068] Before building the 3D environment model, the processor performs time synchronization and coordinate alignment on the LiDAR point cloud, millimeter-wave radar echo, multispectral image, RTK pose, SLAM pose, and inertial navigation data. The LiDAR point cloud is mainly used to determine object boundaries, ground slope, passage width, and habitable area; the millimeter-wave radar echo is used to supplement object information under low visibility, dust, or partial obstruction conditions; and the multispectral image is used to identify object categories and material cues.

[0069] In one implementation, the processor extracts reflection intensity, texture, and temperature response features from each spectral channel of the multispectral image, maps these features to object regions in the point cloud, and outputs the material category and confidence level using a pre-defined material classifier. The material classifier can be a rule-based model, a statistical classification model, a machine learning model, or a combination thereof. For metals, concrete, glass, wood, and vegetation, an initial unit attenuation coefficient corresponding to the operating frequency can be configured respectively.

[0070] When the confidence level of material identification is lower than the material confidence threshold, instead of directly using the result of a single material, a set of candidate materials, the probability of each candidate material, and the larger propagation uncertainty are saved for the corresponding environmental unit. Wireless propagation prediction can select the most unfavorable candidate material, or calculate the expected loss and confidence interval based on the candidate probabilities, thereby avoiding overestimation of coverage due to a single misidentification.

[0071] The equivalent thickness can be determined based on the intersection of the propagating ray and the object's surface, as well as the object's three-dimensional boundaries. If the object's internal structure is not visible, the equivalent thickness can be estimated based on the object category, the distance between visible surfaces, or historical link data, thus increasing the corresponding uncertainty. Static or dynamic attributes can be determined based on continuous frame position changes, radar velocity information, and historical existence time. Building walls and fixed equipment are written as static objects in the long-term model, while vehicles and personnel are written as dynamic objects in the short-term dynamic layer.

[0072] The 3D environment model uses version identification. Each time a fixed object changes, material parameters are updated, or a submap is stitched together, a new model version is generated. This ensures that the link measurement records can be associated with the environment model used at that time, preventing the incorrect updating of propagation parameters using an environment structure that has changed.

[0073] In some embodiments, such as Figure 4 and Figure 5 As shown, the antenna positions corresponding to the candidate network states are taken as the transmitting end, and the antenna positions corresponding to the evaluation points are taken as the receiving end. For each transmitting and receiving end combination, the propagation distance, operating frequency, direct line-of-sight state, object types and equivalent thicknesses traversed by the propagation path, diffraction edge, terrain undulation, vegetation distribution, and dynamic occlusion probability are extracted from the 3D environment model. The predicted propagation loss can be expressed as:

[0074] Where Lij represents the predicted propagation loss between candidate network state i and evaluation point j; Lfs represents the free space propagation loss related to propagation distance dij and operating frequency f; αm(f) represents the unit attenuation coefficient of the m-th type of object at operating frequency f; lm represents the equivalent penetration length of the propagation path in the m-th type of object; Ldif represents the diffraction loss; Lfre represents the Fresnel zone occlusion loss; and Ldyn represents the dynamic obstacle additional loss.

[0075] Free-space propagation loss can be calculated based on propagation distance and operating frequency. Penetration loss can be calculated by multiplying the corresponding unit attenuation coefficient by the equivalent penetration length for each material area traversed by the propagation path and then summing the results. Diffraction loss can be determined based on the geometry of building edges, door frames, tunnel corners, or equipment edges. Dynamic occlusion loss can be determined based on the probability and duration of vehicles, personnel, or mobile devices entering the propagation path.

[0076] To identify cases where there is a line-of-sight between endpoints but an obstacle intrudes into the first Fresnel zone, the radius of the first Fresnel zone can be calculated at the location along the propagation path:

[0077] Where r1(x) represents the radius of the first Fresnel zone at position x on the propagation path; λ represents the operating wavelength of the graph data transmission link; d1(x) and d2(x) represent the distances from position x to the transmitter and receiver, respectively.

[0078] The processor samples along the propagation path and calculates the intrusion ratio of obstacles into the first Fresnel zone at each sampling location. When the intrusion ratio exceeds a preset clearance threshold, the Fresnel zone occlusion loss is increased. This allows for the differentiation of propagation states such as complete line-of-sight, unobstructed line of sight but intrusion into the Fresnel zone, partial occlusion, and complete non-line-of-sight.

[0079] After determining the predicted propagation loss, the predicted received power at the evaluation point can be expressed as:

[0080] Where Pijrx represents the predicted received power; Ptx represents the transmitted power; Gi and Gj represent the antenna gain of the transmitter and receiver in the corresponding elevation and azimuth directions, respectively; and Mij represents the communication margin determined based on the environmental model confidence level, material identification error, positioning error, and historical propagation residual.

[0081] The processor further determines the probability of successful communication, the expected available data rate, latency, and packet loss rate based on the predicted received power, noise level, modulation and coding scheme, and propagation uncertainty. The effective coverage area for a candidate network configuration can be represented as:

[0082] Where Ci represents the effective coverage area of ​​candidate network state i; qj represents the evaluation point; Pmin represents the received power threshold; η represents the coverage probability threshold; Rmin represents the minimum available rate; Tmax represents the maximum allowable delay; PERij represents the predicted packet loss rate; and ε represents the maximum allowable packet loss rate.

[0083] The aforementioned set also requires the same judgment to be performed on the reverse link. Different thresholds can be configured for control commands, location data, alarms, critical images, and continuous video. For example, control and alarm services can be prioritized with lower latency and higher success probability thresholds, while continuous video services can be prioritized with higher minimum rate thresholds. Therefore, the same candidate network state can have different effective coverage areas for different services.

[0084] When there is significant uncertainty in the environmental model or propagation parameters, coverage can be determined based on the lower confidence bound of the received power distribution, rather than simply using the predicted mean. This approach allows the system to adopt conservative coverage results when the material is unknown or the positioning error is large, and gradually reduce the communication margin after accumulating measured data.

[0085] In some embodiments, the processor first extracts the traversable area of ​​the robot dog from the 3D environment model, calculates the ground slope, habitable area, ground stability, distance from obstacles, and hazard level, and combines this with the estimated travel energy consumption from the current location to the candidate area and the return energy to eliminate areas that do not meet the conditions. Subsequently, sampling is performed at preset spatial intervals or adaptive intervals based on the environmental structure to form candidate network locations.

[0086] For each candidate network location, the processor generates a candidate network state based on the adjustable range of the robot dog's body and antenna. If the antenna direction is fixed, states corresponding to different heading angles can be generated; if the antenna is mounted on an adjustable bracket or robotic arm, states corresponding to different antenna heights and orientations can be further generated. Each candidate network state is associated with reachable path, dwell attitude, positioning reliability, expected coverage area, deployment energy consumption, and return-to-home energy.

[0087] To obtain the initial candidate selection order, the unit cost gain of the candidate network states relative to the current network can be calculated:

[0088] Where Si represents the selection score of candidate network state i; ΔCi represents the new coverage of the previously uncovered route segment; ΔKi represents the new connectivity relative to the current network; Qloc,i represents the location reliability of the candidate network state; Ei represents the estimated energy consumption for deployment and return; Ii represents the interference cost; β1 to β5 represent the weights; and δ represents a positive number to avoid a zero denominator.

[0089] like Figure 6 As shown, ( Figure 6 In the graph, double borders indicate selected relays; dashed circles indicate unselected candidate states. The processor uses inspection stations, candidate network states, route sampling points, and task points as graph nodes. For two candidate nodes, a connection edge is established only if both the forward and reverse links meet the corresponding service thresholds. The capacity of the connection edge is determined based on the expected available rate and reserved bandwidth, and the edge weight can simultaneously represent latency, energy consumption, link margin, and propagation uncertainty.

[0090] When selecting relay nodes from candidate network states and determining forwarding paths for inspection nodes, the following comprehensive objective function can be used:

[0091] Where J represents the overall target value; Nrelay represents the number of relay robot dogs; Emove represents the expected mobile energy consumption for deploying, adjusting, and recycling relay robot dogs; Ddelay represents the end-to-end latency; Rblind represents the communication blind spot risk of the inspection route; Qloc represents the overall positioning reliability of the selected relay node and inspection route; w1 to w5 represent the weights configured according to the task type.

[0092] The determination process must meet at least the following hard constraints: there must be an end-to-end path that meets the threshold between the inspection station and each inspection node that needs online communication; the maximum number of hops, total latency, minimum rate, and cumulative success probability of the end-to-end path must meet the corresponding business requirements; the data rate carried by each link must not exceed its available capacity, and bandwidth must be reserved for control, alarm, and positioning data; the relay robot dog must be able to reach and stay stably, and the remaining power after deployment, stay, adjustment, and return must not be lower than the safety threshold; the pose uncertainty of the relay node must not be higher than the positioning threshold; and the candidate location must not be located in a restricted area, equipment movement area, water accumulation area, or collapse risk area.

[0093] In the specific solution process, candidate network states can be added from high to low selection scores until the route coverage and end-to-end connectivity constraints are met. Then, selected states are sequentially deleted or replaced. If deleting a state still satisfies all constraints, that state is deleted. Finally, a local search is performed within the target location's neighborhood for position, attitude, or antenna orientation. Mixed-integer programming, maximum coverage optimization, constrained shortest path, genetic algorithms, reinforcement learning, or other solution methods can also be used. This application does not limit the specific optimization algorithm, as long as the algorithm outputs the relay dog, target network location, and forwarding path based on the above inputs and constraints.

[0094] In some embodiments, such as Figure 7 As shown, the robot dog periodically sends probe data packets during movement and stationary operation, or uses actual service data packets to form link measurement records. Each record is associated with at least the transmitting robot dog identifier, receiving node identifier, transmitting location, receiving location, location uncertainty, environment model version, operating frequency or channel, transmitting power, antenna status, received signal strength, signal-to-noise ratio, packet loss rate, actual data rate, latency, and recording time.

[0095] For a single link, the difference between the actual link quality and the predicted link quality can be used as the propagation residual, and the propagation parameters can be updated as follows:

[0096] Where re(t) represents the propagation residual of link e at time t; qemeas(t) represents the measured link quality; qepred(t) represents the predicted link quality; θt represents the parameter vector consisting of the material attenuation coefficient, regional additional loss, and propagation uncertainty to be corrected; and Kt represents the update gain determined based on sample confidence and historical variance.

[0097] Link quality can be selected from received power, signal-to-noise ratio, packet loss rate, available rate, or normalized combinations thereof. Parameter updates can employ recursive least squares, Kalman update, Bayesian update, exponential moving average, or regression models trained based on historical samples. To avoid a link residual being incorrectly assigned to all objects in the propagation path, update weights can be assigned based on the equivalent penetration length, material identification confidence, and historical identifiability of each object.

[0098] The system distinguishes between short-term dynamic occlusion and continuous fixed deviation. When the residual duration is less than the abnormal duration threshold and the environmental perception component detects a vehicle, person, or mobile device entering the propagation path, the dynamic occlusion loss or propagation uncertainty is mainly updated. When the residual duration reaches the abnormal duration threshold and corresponds to a fixed wall, fixed equipment, or a newly added fixed object, the material attenuation coefficient or regional additional loss of the fixed environment is updated.

[0099] After parameter updates, only the propagation paths and candidate network states intersecting with the updated object or region are recalculated, avoiding the need to recalculate all candidate locations for each measurement. The recalculated effective coverage area is written into the corresponding environmental model version and used for current topology verification and initial predictions for subsequent tasks.

[0100] When the sample size is insufficient or the measurement location is unreliable, the system limits the parameter update amplitude and maintains a large propagation uncertainty. Only when multiple measurements produce consistent residuals at different locations and directions is the fixed material parameter significantly adjusted, thereby avoiding over-correction of the model due to occasional interference.

[0101] In some embodiments, in a known environment, the inspection station pre-determines the target network location based on an existing environmental model and the inspection task. The relay robots are deployed in a forward-to-backward order, and each node is only allowed to enter the restricted public network area after each node has completed upstream bidirectional link verification. This method ensures that the robots do not simultaneously enter areas where connectivity is lost before the network is established.

[0102] In an unknown environment, a leading robot dog constructs a local environment model as it moves. When the link margin between the leading robot dog and the upstream node drops to the warning threshold, the coverage area of ​​nearby candidate network locations is predicted based on the constructed local model, and subsequent robot dogs are dispatched to locations that can simultaneously cover the upstream node and the leading robot dog. After the new relay link is verified, the leading robot dog continues to move; multiple local models are then stitched together into a regional environment model using overlapping features, RTK anchor points, or known landmarks.

[0103] Topology re-evaluation is triggered when received power, signal-to-noise ratio, or communication success probability continuously fall below a threshold; end-to-end latency or packet loss rate continuously exceeds a threshold; link load is too high; node remaining power or storage space is insufficient; node pose uncertainty is too high; new fixed obstructions appear in the environment; the inspection route or data volume changes; or a node fails. To avoid frequent network reassemblies caused by instantaneous fluctuations, each triggering indicator can be configured with a duration threshold and a recovery hysteresis threshold.

[0104] like Figure 8 As shown, the dynamic reconfiguration network based on the principle of "build first, then dismantle" includes: selecting an alternative network state according to the updated effective coverage area; controlling the backup robot dog or mobile relay robot dog to reach the alternative position; enabling the alternative robot dog to establish bidirectional links with the upstream and downstream nodes respectively and completing link verification and synchronization of data to be forwarded; switching the forwarding path to the alternative robot dog and confirming that the key data has been confirmed by the upstream; and finally releasing the original relay robot dog to move to the next position, take over other nodes, or return to the inspection station.

[0105] For each forwarding link, the link load rate can be determined based on the expected data rate to be forwarded and the available capacity:

[0106] Where ρe represents the load rate of link e; λe represents the expected data rate that needs to be forwarded through link e; and Ce represents the available capacity allocated to link e.

[0107] When the link load rate reaches the first load threshold, the bitrate of low-priority continuous video is reduced or ordinary file data is delayed; when it reaches the second load threshold, which is higher than the first load threshold, a robot dog capable of establishing an auxiliary path with partially overlapping nodes is selected from the backup robot dogs. The auxiliary path is only counted in the newly added effective capacity when the primary path and the auxiliary path use different channels, different time resources, or have sufficient space for multiplexing.

[0108] The primary path prioritizes control commands, location data, and alarm data, while the secondary path carries critical images, continuous video, or batch inspection files. The receiving end rearranges and deduplicates data blocks based on their unique identifiers. If no mutually exclusive resources exist, priority is given to reducing the bitrate, delaying data, or throttling tasks, rather than simply adding co-frequency relay nodes.

[0109] When the number of robot dogs is insufficient and the inspection route is long, a rolling relay method can be used to move relay nodes. Before each move, the system verifies that the remaining nodes can maintain the connection between the downstream nodes and the inspection station, or confirms that the downstream nodes have saved enough offline task information. After the task is completed, the relay nodes are generally recycled in order from farthest to nearest.

[0110] In some embodiments, such as Figure 9 As shown, RTK positioning quality can be determined based on the state of the fixed or floating-point solution, the number of available satellites, the timeliness of differential data, the horizontal accuracy factor, and the position covariance; SLAM positioning quality can be determined based on the environmental feature density, point cloud matching residuals, loop closure detection reliability, the number of consecutive matching failures, and the pose covariance. The system does not use whether or not satellite signals are received as the sole switching condition.

[0111] When the RTK positioning quality is below the first RTK threshold and the SLAM positioning quality meets the SLAM threshold, the fusion weights of SLAM and inertial navigation are gradually increased. When the RTK positioning quality is continuously higher than the second RTK threshold for a set time, the RTK weights are gradually restored, where the second RTK threshold is higher than the first RTK threshold. Similarly, entry and exit thresholds can be set for SLAM quality to form a hysteresis interval, avoiding frequent switching at building entrances or semi-obscured areas.

[0112] In the overlapping region where RTK and SLAM are simultaneously effective, multiple sets of time-synchronized RTK global positions and SLAM local positions are saved, and the coordinate transformation relationship from the SLAM local coordinate system to the RTK global coordinate system is determined:

[0113] Where pG represents the position in the global coordinate system; pL represents the position in the SLAM local coordinate system; RLG represents the rotation matrix from the local coordinate system to the global coordinate system; and tLG represents the translation vector from the local coordinate system to the global coordinate system.

[0114] The rotation matrix and translation vector can be obtained through least-squares registration, robust registration, or filtering estimation of multiple sets of synchronized positions. When switching the localization source, the SLAM pose is first transformed to the global coordinate system using the above coordinate transformation relationship, and then the fusion weights are gradually adjusted; the inertial navigation system performs state extrapolation during short-term occlusion and switching to maintain the continuity of the output pose.

[0115] To address the issue of excessively large datasets for complete SLAM maps of large outdoor sites, the area is divided into multiple spatial sub-regions, generating sub-maps and connections between them. The robot dog only loads the sub-map belonging to its current location and adjacent sub-maps; the raw point cloud after leaving is filtered and compressed using keyframes before being transmitted to the inspection station. Anchored sub-maps with global coordinates are generated within the effective RTK area, while local sub-maps within the invalid RTK area establish coordinate relationships with the anchored sub-maps through overlap features.

[0116] The cooperative communication network is used to transmit RTK differential data, anchored sub-maps, local keyframes, coordinate transformation relationships, route updates, and pose status. When the inspection robot dog enters an area without RTK, it can match the local scan results with the anchored sub-map obtained through the network to recover the global coordinates based on the SLAM local pose. The pose and uncertainty of the relay robot dog itself are used as topology determination inputs, and nodes with unreliable positions are not used as stable spatial anchors.

[0117] In some embodiments, such as Figure 10 As shown, the system calculates the link quality based on the received signal strength, available rate, latency, packet loss rate, stabilization duration, and expected hold time of the 4G / 5G link and the long-distance image data transmission link, and selects the transmission link in combination with the service quality requirements of the current data type.

[0118] For control commands and emergency alarms, links with lower latency and higher stability can be prioritized. Alternatively, redundant transmission can be performed and deduplication can be carried out at the receiving end when both public network links and image / data transmission links are available. Fixed bandwidth is reserved for positioning, map, and status data. Key images and detection results can be compressed and segmented. Continuous video and batch files are transmitted with adaptive bitrate or delay based on remaining bandwidth.

[0119] To prevent frequent switching when public network signals are momentarily available, services will only be migrated to a candidate link if the quality of the candidate link remains above the switching threshold for a preset duration, and the expected hold time covers the shortest effective transmission duration. The original link remains available until the switching is complete and confirmation is received, in order to reduce the risk of data interruption.

[0120] When both the 4G / 5G link and the long-distance image and data transmission link are unavailable, the robot dog will continue to execute the pre-stored inspection route only if the following conditions are met simultaneously: the current location reliability is not lower than the safety threshold; the remaining battery power is sufficient to complete the next task segment and return to the nearest safe point; the remaining storage space is sufficient to save the expected data; and the risk level of the route ahead is not higher than the offline allowable level. If any condition is not met, the robot dog will pause the task, return to the most recently connected location, or enter a safe dwell state.

[0121] Inspection data is stored in data blocks. Each data block is associated with at least the task identifier, robot dog identifier, unique data block identifier, acquisition time, acquisition location and positioning quality, data type, priority, data length, checksum, and transmission confirmation status. Larger video files can be divided into multiple sequential segments, and each segment can be further divided into data blocks of fixed or variable length.

[0122] After any communication link is restored, the robot dog first sends a list of data to be transmitted and the local status of each data block. The inspection station or upstream node returns the identifiers of the data blocks that have been fully received, as well as the last confirmed offset position of the incomplete data blocks. The robot dog continues transmission from the stated offset position, prioritizing the retransmission of any missing alarms, key images, and positioning status. The receiving end uses the unique identifier and checksum of each data block to perform integrity verification and deduplication suppression.

[0123] After the connection is restored, the system re-evaluates the estimated completion time and reliability of the 4G / 5G link and the long-distance image and data transmission link, and selects one link to perform the resume transmission; when different channels, different time resources or other non-interference conditions are available, different data blocks can also be transmitted in parallel through the two types of links.

[0124] The specific inspection implementation is as follows: The following example uses an inspection area including an open-air area, a factory building with dense metal equipment, and underground passages to illustrate the complete execution process of the above method. The inspection station is located in an open-air area. The inspection route starts from the inspection station, passes through the factory entrance, internal factory intersections, underground passage corners, and the target equipment, and then returns along the same route. The system is configured with five robot dogs: one acts as a gateway node near the inspection station, two are used for initial relay, one performs the inspection, and one serves as a backup or auxiliary relay. This number is for illustrative purposes only.

[0125] Before the mission begins, the inspection station issues the inspection route, task points, prohibited areas, and data service requirements. Control and alarm data are configured with the highest priority, positioning and map data with high priority, key images and detection results with medium-high priority, and continuous multispectral video with normal priority. Each robot dog reports its remaining battery power, current location, positioning quality, available storage space, and communication interface status.

[0126] The inspection robot and the advance relay robot use LiDAR to construct geometric models of the factory and underground passages, millimeter-wave radar to fill in areas partially obscured by dust or equipment, and multispectral images acquired by a penta-lens lens to assist in the identification of concrete walls, metal equipment, and ordinary partitions. The model records the material type of the walls and equipment, estimated thickness, and identification confidence level.

[0127] The system generates candidate network locations near the factory entrance, internal intersections, and corners of underground passages. Although the straight-line distance from the factory entrance to the corner of the underground passage is less than the nominal distance of the image and data transmission components, the propagation path passes through metal equipment and concrete walls, and the factory corner intrudes into the first Fresnel zone. The predicted received power and available rate cannot meet the video threshold, therefore this straight-line path is not considered a valid connection edge.

[0128] After constructing the communication connection map based on the effective coverage area, the system selects the factory entrance as the target networking location for the first relay robot dog, and selects the internal cross passage of the factory as the target networking location for the second relay robot dog. The first relay location can connect the inspection station and the factory interior, while the second relay location can bypass metal equipment and connect to the entrance of the underground passage. The backup robot dog is kept in an area where it can quickly reach multiple alternative locations.

[0129] After the first relay robot dog arrives at the factory entrance, it completes forward and reverse link verification with the inspection station and uploads the measured link results. The system fine-tunes the additional losses in the entrance area based on the measured results. After successful verification, the second relay robot dog continues to move to the internal cross passage of the factory; after the second relay link verification is successful, the inspection robot dog enters the underground passage. This deployment sequence ensures that the inspection robot dog is always located after an established or recoverable communication path.

[0130] In the open-air area, RTK positioning quality is high, and the robot dog primarily uses RTK and inertial navigation as its positioning sources. Within the overlapping area at the factory entrance, both RTK global pose and SLAM local pose are simultaneously stored, and coordinate transformation relationships are calculated. Upon entering the factory and underground passages, the robot dog gradually increases the fusion weights of SLAM and inertial navigation, and receives anchored sub-maps and RTK differential data through a cooperative network. Therefore, the local SLAM pose within the underground passages can still be transformed to the global coordinate system.

[0131] After the inspection robot dog activates continuous multispectral video, the load rate of the second relay link reaches the second load threshold. The system first ensures the reserved bandwidth for control, positioning, and alarm data, then dispatches a backup robot dog to another candidate location that can simultaneously connect to the upstream node and the inspection node, and allocates different channel or time resources to this robot dog. Control, positioning, and alarm data continue to be transmitted through the main path, while some video and batch files are transmitted through the auxiliary path.

[0132] A mobile device briefly remained in the channel, causing a decrease in the received power of the second relay link. The environmental sensing component detected that the object was moving, and since the propagation residual duration was less than the abnormal duration threshold, the system increased the dynamic occlusion loss and propagation uncertainty without immediately and permanently modifying the attenuation coefficient of the concrete wall. After the mobile device left, the link quality recovered, and the system reduced the dynamic risk.

[0133] If a metal device is moved to a new fixed location and continues to block the link, the system updates the object to the fixed environment layer and recalculates the effective coverage area of ​​the affected candidate location. The backup robot dog first moves to the alternative location, establishes a bidirectional link with the upstream and downstream, and completes data synchronization; after the forwarding path switch is completed, the original relay robot dog leaves the original location, thereby avoiding communication interruption during the network reconfiguration process.

[0134] After the inspection robot dog enters the short-term blind spot at the end of the underground passage, both the 4G / 5G and long-distance image and data transmission links become unavailable. The system confirms that its positioning reliability, remaining battery power, available storage space, and the risks of the route ahead meet the offline execution conditions, and therefore allows it to continue executing the pre-stored route, saving the images, detection results, robotic arm operation results, and status data in blocks.

[0135] After the inspection robot dog returns to the position where the image and data transmission signal has been restored, it sends a list of data to be transmitted to the upstream node. The upstream node returns the identifiers of the data blocks that have been fully received and the last confirmed offset positions of the incomplete data blocks. The inspection robot dog continues transmission from the corresponding offset positions, prioritizing the transmission of alarms and key images, and then transmitting continuous video; the receiving end performs deduplication and integrity verification according to the unique identifier and check value of the data blocks.

[0136] After the inspection mission is completed and the remote data is synchronized, the inspection robot dog returns along the planned route. The second relay robot dog returns to the coverage area of ​​the first relay dog ​​after confirming that there is no unconfirmed critical data downstream. The first relay robot dog is the last to be withdrawn to the inspection station. Through the order of retrieval from far to near, necessary positioning maps and communication services are maintained during the return phase.

[0137] This application also provides a robot dog, including a body, a motion actuator, a processor, a memory, and an industry expansion module disposed on the body; The industry expansion module includes an RTK positioning component, an inertial measurement component, at least one of a lidar and a millimeter-wave radar for SLAM mapping and environmental perception, a long-range map and data transmission component, a 4G / 5G communication component, and a multispectral imaging component consisting of a five-lens camera. The memory stores a computer program. When the processor executes the computer program, it controls the motion actuator and the industry expansion module, enabling the robot dog to work in concert with the inspection station and other robot dogs to perform the aforementioned robot dog collaborative networking control method.

[0138] Specifically, processor 239 can be a central processing unit, graphics processing unit, embedded processor, field-programmable gate array, or other processing unit with data processing capabilities. Memory 240 can include random access memory, read-only memory, flash memory, solid-state drive, or other non-volatile memory. Motion actuator 220 includes joint actuators, leg mechanisms, and motion controllers for moving or staying according to path and posture commands output by the processor.

[0139] The industry expansion module 230 can be connected to the main body 210 through a standard mechanical interface, power supply interface, and data interface. After the processor 239 runs the program in the memory 240, it can build an environment model and upload link measurement results during the inspection task phase. When assigned as a relay node, it can move to the target network location and establish a forwarding link. When assigned as an inspection node, it can move along the route and perform inspections. When assigned as a standby node, it can respond to dynamic network reconfiguration or auxiliary path scheduling.

[0140] This application also provides a robot dog collaborative networking system, including an inspection station and multiple robot dogs, each of which is equipped with an environmental perception component, a positioning and navigation component, a long-distance image and data transmission component, a processor and a memory. The inspection station is used to issue inspection tasks, maintain the three-dimensional environment model, and collect inspection data. At least one of the processors of the inspection station and each of the robot dogs is configured to: construct a three-dimensional environment model with object attributes based on data collected by the environmental perception component during the positioning and navigation process; predict the effective coverage area of ​​the candidate network location based on the three-dimensional environment model; determine at least one relay robot dog, the target network location, and the forwarding path based on the effective coverage area and the robot dog status; control the relay robot dog to establish a single-hop or multi-hop communication network via a long-distance image and data transmission component; and correct the effective coverage area and adjust the target network location or the forwarding path based on the actual link measurement results after the network is established.

[0141] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the above-described machine dog cooperative networking control method.

[0142] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A method for collaborative networking control of robotic dogs based on an environment model in a public network-restricted environment, applied to a collaborative networking system including an inspection station and multiple robotic dogs, characterized in that, include: Acquire inspection task information, status information of each robot dog, and a three-dimensional environment model with object attributes constructed by at least one of the robot dogs based on environmental perception data during the positioning and navigation process; Candidate network locations are generated based on the passable areas and safe conditions for staying in the three-dimensional environment model. For each candidate network location, the wireless propagation quality is predicted based on the geometric information, material information and occlusion relationship of objects on the propagation path between the candidate network location and the evaluation point. The set of evaluation points that meet the preset service quality threshold is determined as the effective coverage area of ​​the candidate network location. The evaluation points include at least one of the following: inspection route sampling points, task points, inspection station locations and other candidate network locations. A communication connection diagram is constructed based on each effective coverage area. Under the constraints of inspection route coverage, end-to-end connectivity, link capacity, positioning reliability, robot dog energy, and stationary security, a relay robot dog, the target network location of the relay robot dog, and the forwarding path are determined from multiple robot dogs. Control the relay robot dog to move to the corresponding target networking location, and establish a single-hop or multi-hop communication network between the inspection station and the robot dog performing the inspection task through a long-distance image and data transmission link; Obtain the actual link measurement results of the communication network, update the wireless propagation prediction parameters based on the difference between the actual link measurement results and the predicted wireless propagation quality, redetermine the corresponding effective coverage area based on the updated wireless propagation prediction parameters, and adjust at least one of the target network location and the forwarding path according to the redetermined effective coverage area and the robot dog's operating status.

2. The robot dog collaborative networking control method according to claim 1, characterized in that, The environmental perception data includes lidar point clouds, millimeter-wave radar echo data, and multispectral images acquired by a five-lens camera. The environmental unit of the three-dimensional environment model with object attributes records the occupancy status, three-dimensional geometric position, object category, material category, equivalent thickness, static or dynamic attributes, wireless attenuation parameters, identification confidence level, and last update time. Specifically, the boundary of an object is determined by using at least one of the lidar point cloud and the millimeter-wave radar echo data, the material category of the object is determined or assisted by the multispectral image, and when the material identification confidence is lower than the material confidence threshold, a material candidate set and propagation uncertainty are configured for the corresponding environmental unit.

3. The robot dog collaborative networking control method according to claim 1, characterized in that, The predicted wireless propagation quality includes: Free space propagation loss is determined based on candidate network locations, evaluation points, and operating frequencies; penetration loss is determined based on the material type, equivalent penetration length, and corresponding unit attenuation coefficient of various objects traversed by the propagation path; diffraction loss is determined based on building edges or curves in the propagation path; Fresnel zone occlusion loss is determined based on the degree of intrusion of obstacles into the first Fresnel zone; and dynamic occlusion loss is determined based on the occlusion probability of dynamic objects. The free space propagation loss, penetration loss, diffraction loss, Fresnel zone occlusion loss, dynamic occlusion loss, and communication margin determined based on environmental model confidence and positioning error are used to determine the received power, communication success probability, available rate, latency, and packet loss rate. The effective coverage area is composed of evaluation points that meet the corresponding service's received power, communication success probability, available rate, latency, and packet loss rate thresholds for both forward and reverse links.

4. The robot dog collaborative networking control method according to claim 1, characterized in that, The candidate network locations are obtained by spatial sampling after removing areas from the passable areas that do not meet the corresponding conditions for slope, dwellable area, hazard level, distance from obstacles, or return energy. For each candidate network location, at least one candidate network state is generated. The candidate network state includes the location and at least one of the following: aircraft height, pitch attitude, heading angle, image and data transmission antenna height, and image and data transmission antenna orientation. Determining the relay robot dog, target network location, and forwarding path includes: selecting candidate network states based on the new coverage of the uncovered route segment, the new connectivity relative to the current network, the location reliability, mobile energy consumption, and interference cost; and deleting, replacing, or fine-tuning the selected candidate network states to ensure that the number of relay robot dogs and mobile energy consumption meet preset optimization conditions. The height or orientation of the image data transmission antenna can be adjusted by an adjustable bracket or the robotic arm of a robot dog.

5. The robot dog collaborative networking control method according to claim 1, characterized in that, The actual link measurement results include the transmission location, the receiving location, the environmental model version, the operating frequency or channel, the transmission power, the antenna status, the received signal strength, the signal-to-noise ratio, the packet loss rate, the actual data rate, and the latency. The updated wireless propagation prediction parameters include: updating the corresponding material attenuation coefficient, regional additional loss, and propagation uncertainty based on the propagation residual between the actual link measurement results and the prediction results, as well as the material type and equivalent thickness of the objects traversed by the propagation path. The propagation residuals with durations less than the abnormal duration threshold are used to update the dynamic occlusion loss or propagation uncertainty, while the propagation residuals with durations reaching the abnormal duration threshold and corresponding to a fixed object are used to update the material attenuation coefficient or regional additional loss of the fixed environment.

6. The robot dog collaborative networking control method according to claim 1, characterized in that, Dynamic network reassembly is triggered when the link quality fails to meet the service quality threshold, the link load rate exceeds the preset threshold, the remaining battery power of the robot dog is lower than the battery threshold, the robot dog's pose uncertainty exceeds the positioning threshold, or the 3D environment model changes. The dynamic reconfiguration network includes: selecting an alternative network state based on the newly determined effective coverage area, controlling the backup robot dog or mobile relay robot dog to reach the alternative network position, first establishing and verifying bidirectional links with the upstream and downstream nodes respectively and completing data synchronization, then switching the forwarding path to the alternative robot dog, and then releasing the original relay robot dog. When the link load rate reaches the first load threshold, the low-priority video bitrate is reduced or the low-priority data is delayed; when the link load rate reaches the second load threshold which is higher than the first load threshold, an auxiliary path is established using different channels, different time resources or spatial multiplexing resources. The main path carries control commands, positioning data and alarm data, while the auxiliary path carries images, videos or batch inspection files.

7. The robot dog collaborative networking control method according to claim 1, characterized in that, The communication network is used to transmit RTK differential data, anchored sub-maps, local keyframes, and coordinate transformation relationships; Each robot dog determines the RTK positioning quality based on the RTK fixed solution state, differential data timeliness, and position covariance. It determines the SLAM positioning quality based on the point cloud matching residual, loop closure detection reliability, and pose covariance. In the overlapping area where RTK and SLAM are simultaneously effective, it determines the coordinate transformation relationship between the SLAM local coordinate system and the RTK global coordinate system. Based on the quality threshold with hysteresis interval, it switches or adjusts the fusion weights of RTK, SLAM, and inertial navigation. Each robot dog also selects a transmission link based on the available rate, latency, packet loss rate, stability duration, and expected hold time of the 4G / 5G link and the long-distance map and data transmission link. When both types of links are unavailable and the positioning reliability, remaining power, available storage space, and risks of the route ahead all meet the offline execution conditions, it continues to execute the pre-stored inspection route and caches the inspection data in blocks. After the link is restored, it performs breakpoint resumption and duplicate data suppression based on the unique identifier of the data block, the check value, and the last confirmed offset position.

8. A robotic dog, characterized in that, It includes a body, a motion actuator, a processor, a memory, and an industry expansion module disposed on the body; The industry expansion module includes an RTK positioning component, an inertial measurement component, at least one of a lidar and a millimeter-wave radar for SLAM mapping and environmental perception, a long-range map and data transmission component, a 4G / 5G communication component, and a multispectral imaging component consisting of a five-lens camera. The memory stores a computer program. When the processor executes the computer program, it controls the motion actuator and the industry expansion module, enabling the robot dog to cooperate with the inspection station and other robot dogs to perform the robot dog cooperative networking control method as described in any one of claims 1 to 7.

9. A robot dog collaborative networking system, characterized in that, The system includes an inspection station and multiple robot dogs, each of which is equipped with an environmental perception component, a positioning and navigation component, a long-distance image and data transmission component, a processor, and a memory. The inspection station is used to issue inspection tasks, maintain the three-dimensional environment model, and collect inspection data. At least one of the processors of the inspection station and each of the robot dogs is configured to: construct a three-dimensional environment model with object attributes based on data collected by the environmental perception component during the positioning and navigation process; predict the effective coverage area of ​​the candidate network location based on the three-dimensional environment model; determine at least one relay robot dog, the target network location, and the forwarding path based on the effective coverage area and the robot dog status; and control the relay robot dog to establish a single-hop or multi-hop communication network via a long-distance image and data transmission component. And based on the actual link measurement results after network establishment, the effective coverage area is corrected and the target network location or the forwarding path is adjusted.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, causes the processor to perform the robot dog cooperative networking control method according to any one of claims 1 to 7.

Images