Robot dog formation collaborative operation method and system in disaster environment

Abstract

The invention relates to the field of robot navigation, in particular to a robot dog formation collaborative operation method and system in a disaster environment. The method comprises the following steps: acquiring future intention information of a pilot dog and a local environment model sensed by a relay dog, and fusing to construct a prospective environment model; calculating a link quality score of a relay dog to the intention of a pilot dog based on the model, constructing a potential field, and solving a gradient to obtain a communication gradient impetus; switching the operation mode of the relay dog according to the operation state of the pilot dog; in the corresponding mode, the communication gradient driving force, the communication attraction pointing to the pilot dog and the obstacle repulsive force are fused, the navigation resultant force is calculated, and the relay dog is controlled to work cooperatively. Through the prospective communication quality gradient field, the relay dog prejudges and avoids the communication blind area, the modal switching adaptation requirement is combined, the problem that formation communication is prone to being interrupted in the disaster environment is solved, the communication continuity and robustness are improved, and support is provided for efficient collaborative operation of the robot dog.

Description

Technical Field

[0001] This invention relates to the field of robot navigation, and in particular to a method and system for collaborative operation of robot dogs in disaster environments. Background Technology

[0002] In disaster environments such as earthquakes, fires, or building collapses, the use of multiple robots, especially highly mobile robot dogs, in coordinated search and rescue operations has become a key technology for ensuring the safety of rescue personnel and improving search and rescue efficiency. In a typical lead-relay operation mode, the lead robot dog autonomously explores deep into the disaster site, while the relay robot dog follows and establishes a dynamic communication link to ensure that the lead dog's video and sensor data can be transmitted back to the command center in real time.

[0003] To achieve this goal, the navigation algorithm of the relay robot dog is crucial. Currently, the industry generally adopts a navigation strategy based on the Artificial Potential Field (APF) method. This method treats the target point, such as the navigator dog, as an attractive force source and obstacles as repulsive force sources, and the relay robot dog moves along the direction of the resultant force of the potential field.

[0004] However, traditional APF (Advanced Persistent Function) methods have serious shortcomings in complex and unstructured environments such as disaster sites. APF is a reactive algorithm that makes decisions solely based on current gravitational and repulsive forces, lacking forward-looking predictions of future environmental changes. This strategy easily traps relay robots in local optima. In disaster sites, this trap manifests as follows: the relay robot might be navigated to a physically safe location with good signal, but this location might be a communication dead zone in the topology. As the lead robot continues deeper, bypassing load-bearing walls or heavy obstacles, the communication link between the relay and lead robots is instantly lost. At this point, if the relay robot wants to re-establish the connection, it often needs to retreat significantly and relocate, severely wasting valuable rescue time and potentially causing the lead robot to lose contact altogether.

[0005] Therefore, how to solve the local optimal trap problem in the navigation process of relay robot dogs, so that they can not only follow the navigator dog, but also proactively avoid future communication dead zones and achieve robust and continuous communication relay, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] To address the technical problems of traditional artificial potential field methods being prone to communication dead zones and local optima, this invention provides a method and system for collaborative operation of robot dogs in disaster environments.

[0007] In a first aspect, the present invention provides a method for collaborative operation of robot dogs in a disaster environment, employing the following technical solution: A method for collaborative operation of robot dogs in a disaster environment includes the following steps: The system acquires the future intent information broadcast by the navigating robot dog and the local environment model perceived by the relay robot dog itself, and integrates them to construct a forward-looking environment model. Based on the aforementioned forward-looking environment model, the link quality score of the relay robot dog at the current location for the future intention information of the navigating robot dog is calculated, and the potential field of the link quality score is constructed. Calculate the gradient of the potential field of the link quality score to obtain the communication gradient driving force; The operation mode is switched according to the operation status of the navigation robot dog. In the operation mode, the final navigation resultant force is calculated by integrating the communication gradient driving force, the communication attraction based on the current position of the navigation dog, and the obstacle repulsion force, and the relay robot dog is controlled to perform cooperative operation.

[0008] This invention introduces a communication gradient impetus based on the future intentions of the navigator, replacing the traditional gravity that relies solely on the current position. This gives the relay dog ​​the ability to predict future communication quality. This allows the relay dog ​​to proactively avoid impending communication dead zones, completely solving the problem of traditional APF methods easily falling into local optima traps, and significantly improving the communication continuity and navigation robustness of collaborative operations at disaster sites.

[0009] Preferably, obtaining the future intent information broadcast by the navigation robot dog includes: Obtain the probabilistic path cone broadcast by the navigation robot dog, which consists of a central path and a set of variance parameters that expand over time.

[0010] This invention employs a probabilistic path cone, rather than a single predicted path, to more realistically model trajectory deviations caused by the navigator dog's temporary obstacle avoidance. This quantification of the uncertainty of future intentions makes the calculated link quality score more robust, avoiding drastic fluctuations in the relay dog's navigation strategy caused by minor trajectory changes in the navigator dog.

[0011] Preferably, the fusion and construction of the forward-looking environment model further includes: The relay robot dog uses sensors to build a 3D occupancy grid map in real time; Semantic assignment is performed on the voxels in the three-dimensional occupied grid map to obtain the material category of each voxel, and a semantic voxel map is constructed, which is denoted as the prospective environment model.

[0012] This invention, by constructing a semantic voxel map, enables the navigation model to perceive the material categories of obstacles for the first time. This transforms subsequent communication quality calculations from being based on simplified geometric distances to calculating penetration loss based on physical reality, providing crucial physical input for the accuracy of link quality scores.

[0013] Preferably, the method for obtaining the semantic attribution includes: By fusing the coordinates of voxels, the intensity of LiDAR reflection, and color information, geometric, physical, and visual features are extracted to form a feature vector; The feature vector is input into a pre-trained random forest classifier, which outputs the material category for each voxel.

[0014] Preferably, the calculation of the link quality score of the relay robot dog at the current location for the future intention information of the navigating robot dog includes: In the aforementioned forward-looking environment model, ray tracing is performed from the current position of the relay robot dog to multiple future path points in the future intention information of the navigator robot dog; For any obstacle element that the ray passes through, calculate the linear path loss based on its material type and penetration thickness; The link quality of the relay robot dog from its current location to the future path points is calculated based on the linear path loss, and the link quality of all future path points is fused to obtain the link quality score.

[0015] This invention describes a complex, forward-looking environment model as a single, scalarized link quality score, providing a clear, differentiable optimization objective for subsequent gradient calculations.

[0016] Preferably, the formula for calculating the linear path loss is:

[0017]

[0018] in, Indicates the first ray that passes through Linear path loss of an obstacle, and Representing materials Fixed loss and loss per meter, N represents the relay dog ​​position. To future path points The number of obstacles between them Indicates the first ray that passes through The penetration thickness of an obstacle; Indicates the location of the relay dog For future path points The quality of linear links, For linear transmit power, Based on linear path loss.

[0019] This invention employs a linear path loss formula based on a physical model to accurately describe the attenuation of signals as they penetrate different materials and thicknesses. By combining fixed loss and loss per meter, and finally performing product attenuation on a linear scale, it ensures that the calculation of link quality conforms to the propagation laws of radio signals in the physical world, greatly improving the realism of the potential field.

[0020] Preferably, calculating the gradient of the potential field of the link quality score to obtain the communication gradient driving force includes: The gradient of the link quality score potential field is estimated by employing the finite difference method and repeatedly calling the calculation of the link quality score at the current position of the relay robot dog and its immediate neighbors' virtual sampling points. ; The formula for calculating the communication gradient driving force is:

[0021] in, As a driving force for communication gradient, This represents the gradient gain coefficient.

[0022] Preferably, the step of switching the operating mode according to the operating status of the navigation robot dog includes: If the lead robot dog is in a moving exploration state, the relay robot dog switches to dynamic following mode; If the navigator robot dog is in a stationary operation state, the relay robot dog will switch to the static anchoring mode.

[0023] Preferably, in the dynamic following mode, the formula for calculating the final navigation resultant force is:

[0024] in, This is a dynamic following resultant force, also known as the final navigation resultant force. The communication gravity generated by the navigator's current location, Repulsive force from obstacles For base station repulsion, , , This is the corresponding gain coefficient. As a driving force for communication gradient, This represents the gradient gain coefficient.

[0025] Secondly, the present invention provides a robot dog formation collaborative operation system for disaster environments, which adopts the following technical solution: A robot dog formation collaborative operation system under disaster conditions includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned robot dog formation collaborative operation method under disaster conditions is implemented.

[0026] By adopting the above technical solution, a computer program is generated for the above-mentioned method of robot dog team collaboration in a disaster environment, and stored in a memory for loading and execution by a processor. Terminal devices are then made based on the memory and processor for convenient use.

[0027] The present invention has the following technical effects: This invention introduces a gradient field based on the navigator's future intentions, enabling the relay dog ​​to predict future changes in communication quality. When the navigator is about to enter a communication dead zone, this gradient force will proactively push the relay dog ​​to move towards an area with better signal, thereby actively avoiding communication dead zones and solving the local optima problem caused by reactive decision-making in traditional APF methods.

[0028] Furthermore, this invention not only considers the geometry of obstacles but also accurately identifies their materials through semantic attribution. Combined with ray tracing from a physical model, it achieves precise calculation of linear path loss. This makes the assessment of communication quality far more accurate and robust than traditional, simple models based on signal strength. Attached Figure Description

[0029] Figure 1 This is a flowchart of a method for collaborative operation of robot dogs in a disaster environment, provided by an embodiment of the present invention. Figure 2 A comparison chart of navigation effects provided for embodiments of the present invention. Detailed Implementation

[0030] This invention discloses a method for collaborative operation of robot dogs in a disaster environment, referring to... Figure 1 This includes steps S1-S4: S1: Obtain the future intent information broadcast by the navigator robot dog, as well as the local environment model perceived by the relay robot dog itself, and integrate them to construct a forward-looking environment model.

[0031] It is important to note that any effective navigation decision must be based on a complete environmental model. In the navigator-relay collaborative scenario, this model must include information in two dimensions: first, the relay dog's own static physical environment, such as obstacles and walls; second, the navigator dog's dynamic mission objective, i.e., where it intends to go. Without the former, navigation cannot guarantee physical safety; without the latter, navigation will lose its foresight and degenerate into reactive following.

[0032] S10: Obtain information about the future intentions of the navigating robot dog.

[0033] Preferably, as an example, the information obtained regarding the future intent of the navigation robot dog includes: The relay robot dog acquires future intent information broadcast by the lead robot dog, processes this information using a probabilistic modeling method, and obtains a probabilistic path cone. This probabilistic path cone consists of a central path and variance parameters. For example, the probabilistic modeling method could be an extended Kalman filter algorithm.

[0034] Understandably, the advantage of using a path cone instead of a single path is that this model is more robust to trajectory deviations caused by the navigator dog temporarily avoiding small obstacles.

[0035] S11: The relay robot dog constructs a local environment model.

[0036] Preferably, as an example, the relay robot dog constructs a local environment model, including: The relay robot dog uses its onboard LiDAR and depth camera to collect raw sensor data in real time. The system uses the OctoMap algorithm to process the collected raw sensor data into a map, resulting in a 3D occupancy grid map, denoted as the local environment model.

[0037] The map divides the space into discrete voxels.

[0038] Optionally, as an example, the relay robot dog builds a local environment model, including: The original sensor data is processed using a truncated directed range field fusion algorithm to construct a continuous three-dimensional surface mesh model in real time, which is denoted as the local environment model.

[0039] It should be noted that this model can more accurately describe the surface topology and geometry of obstacles.

[0040] S12: Develop a forward-looking environmental model.

[0041] Preferably, as an example, constructing a forward-looking environment model includes: First, a 3D occupancy grid map is acquired and fused with LiDAR reflection intensity and RGB color collected by sensors.

[0042] Next, geometric, physical, and visual features are extracted from the fused information of voxels to form a feature vector.

[0043] Subsequently, a pre-trained random forest classifier is used to classify the material of the feature vector to obtain the material category of each voxel, and finally a semantic voxel map is constructed, denoted as the prospective environment model.

[0044] Thus, through the above processing, the relay robot dog not only obtained a high-precision three-dimensional geometric map, but also the future dynamics of the navigator dog and the material of each obstacle in the environment, providing a data foundation for subsequent analysis.

[0045] Optionally, as an example, construct a forward-looking environment model, including The collected point cloud is directly input into a pre-trained 3D semantic segmentation deep neural network, which directly outputs the semantic label of each point or voxel. A prospective environment model is constructed based on the semantic label, which is denoted as the prospective environment model.

[0046] It should be noted that this method does not require manual feature design and typically has higher classification accuracy.

[0047] S2: Based on the aforementioned forward-looking environment model, calculate the link quality score of the relay robot dog at its current location for the future intention information of the navigator robot dog, and construct the potential field of the link quality score.

[0048] It's important to note that traditional navigation optimizes geometric distance. However, in disaster sites filled with highly attenuating media like reinforced concrete, the strength of radio signals is not determined by geometric distance, but by the penetration loss through obstacles. Therefore, optimizing only geometric distance is an incorrect optimization objective. Thus, it's necessary to establish an optimization objective that truly reflects the physical propagation reality and replaces geometric distance.

[0049] Preferably, as an example, based on the aforementioned forward-looking environment model, the link quality score of the relay robot dog at its current location regarding the future intention information of the navigating robot dog is calculated, and the potential field of the link quality score is constructed, including: First, a sampling method is used to select from the probabilistic path cone. Discrete future path points Get the current location of the relay robot dog. .

[0050] Next, the 3D ray tracing algorithm is used to simulate the environment from a forward-looking environment model. arrive The signal propagation path was determined, and the materials of each obstacle along the path were statistically analyzed. and the penetration thickness of each obstacle .

[0051] Subsequently, the linear path loss reflecting the signal attenuation characteristics is calculated based on the obstacle material and penetration thickness.

[0052] For example, the linear path loss satisfies the following relationship:

[0053] in, The linear path loss of a ray penetrating the j-th obstacle. and Representing materials Fixed losses and losses per meter, This represents the penetration thickness of the ray through the j-th obstacle; Thus, by utilizing material properties and penetration thickness, which are strongly correlated with signal loss, the signal loss under traditional obstacles can be accurately described.

[0054] Then, link quality, which reflects the communication quality on the path, is calculated by combining link quality and linear path loss.

[0055] For example, link quality satisfies the following relationship:

[0056] in, Indicates the location of the relay dog For future path points The quality of linear links; For linear transmit power, Based on the linear path loss, N represents the relay dog ​​position. To future path points The number of obstacles on the ray between them.

[0057] Understandable This reflects the signal condition on the path when there is no loss. This reflects the loss along the path, and thus the signal quality of the path can be accurately calculated using these two combinations.

[0058] Finally, the method of averaging is used for all The linear link quality obtained from each future path point is fused to obtain the link quality score of the relay dog ​​at the current location point, and the link quality score is denoted as the potential field.

[0059] In this way, the above steps can describe a complex, multi-dimensional environmental model as a single, scalar potential field that reflects the quality of communication, thus providing a data foundation for subsequent analysis.

[0060] Optionally, as an example, based on the aforementioned forward-looking environment model, the link quality score of the relay robot dog at its current location regarding the future intention information of the navigating robot dog is calculated, and the potential field of the link quality score is constructed, including: First, the 3D ray tracing algorithm is also used to simulate the environment in a forward-looking environment model. arrive The signal propagation path, but only used to statistically analyze the ray's passage through each material. voxel count .

[0061] Then, calculate the total linear path loss:

[0062] in, This represents the total linear path loss of the path. and For standard logarithmic distance loss parameters, Indicates the total path length; For materials The attenuation weight is used to control the degree of material attenuation. For rays to pass through materials The number of voxels, It represents the set of all materials through which the ray passes.

[0063] Next, calculate the link quality:

[0064] in, Indicates the location of the relay dog For future path points The quality of linear links; This represents the linear transmit power.

[0065] Finally, the method of averaging is used for all The linear link quality obtained from each future path point is fused to obtain the link quality score of the relay dog ​​at the current location point, and the link quality score is denoted as the potential field.

[0066] It should be noted that this method is faster in calculation and easier to calibrate online using field data.

[0067] S3: Calculate the gradient of the potential field of the link quality score to obtain the communication gradient driving force.

[0068] It should be noted that the link quality score calculated above only tells the robot dog the communication quality at its current location, but it does not tell the robot dog which direction to move in to improve communication quality. To obtain this directional guidance, gradient analysis of this indicator must be performed.

[0069] Preferably, as an example, calculating the gradient of the link quality score potential field to obtain the communication gradient driving force includes: First, the spatial gradient of the potential field at the current position of the relay dog ​​is estimated using the finite difference method.

[0070] Subsequently, a communication gradient driving force that reflects the direction of communication quality improvement is calculated based on the spatial gradient.

[0071] For example, the communication gradient driving force satisfies the following relationship:

[0072] in, For communication gradient driving force; This is the gradient gain coefficient, a preset value used to control the thrust intensity. for Spatial gradient of a point.

[0073] Thus, through the above calculations, the relay robot dog gains a powerful navigation force. This force is not directed to the current position of the navigator dog, but to a direction that optimizes the overall quality of its future communication links.

[0074] S4: Switch the operation mode according to the operation status of the navigation robot dog. In the operation mode, integrate the communication gradient driving force, the communication attraction based on the current position of the navigation dog, and the obstacle repulsion force to calculate the final navigation resultant force, and control the relay robot dog to perform cooperative operation.

[0075] S40: Switch the operating mode according to the operating status of the navigation robot dog. In the operating mode, the final navigation resultant force is calculated by integrating the communication gradient driving force, the communication attraction based on the current position of the navigation dog, and the obstacle repulsion force.

[0076] It should be noted that the relay dog's operating mode is closely related to the navigator robot dog's operating status, therefore the relay dog's operating mode needs to be adjusted in real time based on the navigator robot dog's operating status.

[0077] Preferably, as an example, the operating mode is switched according to the operating state of the navigation robot dog. In the operating mode, the communication gradient driving force, the communication attraction force based on the current position of the navigation dog, and the obstacle repulsion force are fused to calculate the final navigation resultant force, including: The system acquires the operational status broadcast by the navigation robot dog in real time. A finite state machine is used to determine this operational status and switch the relay dog's operational mode accordingly. The relay dog's operational modes include dynamic following and static anchoring.

[0078] If the relay dog's operating mode is dynamic following, then dynamic following will be executed.

[0079] If the relay dog's operating mode is static anchoring, then static anchoring will be performed.

[0080] The specific implementation methods for performing dynamic following and static anchoring are explained below.

[0081] The specific methods for performing dynamic following include: It should be noted that the communication gradient driving force calculated in the above steps only represents the direction of communication optimization. However, in the real world, relay dogs also need to meet the following three hard constraints: 1. Physical safety, such as not colliding with obstacles; 2. Current connectivity, such as not being too far from the navigator dog; 3. Link scalability, such as not being too close to the base station. Furthermore, to better communicate with the navigator dog, the following strategy needs to be adjusted in real time based on the navigator dog's state. Therefore, it is necessary not only to consider the direction of communication optimization but also to combine other requirements to jointly adjust the relay dog's future navigation strategy.

[0082] First, based on the current location of the navigation dog, the 3D occupied grid map, and the location of the base station, the traditional APF algorithm is used to calculate the communication attraction, obstacle repulsion, and base station repulsion respectively.

[0083] Subsequently, the calculated forces were fused together.

[0084] Preferably, as an example, methods for fusing the calculated multiple forces include:

[0085] in, To dynamically follow the combined force, As a driving force for communication gradient, For communication gravity, Repulsive force from obstacles Repulsive force for base stations; , , This is the corresponding gain coefficient, used to adjust the priority of different navigation targets.

[0086] Understandably, through Constraints can guide the relay dog ​​to the area with the best future communication quality. Through... Constraints can maintain the current link connectivity with the navigation dog. Through Restraints ensure the physical safety of the relay dog ​​and prevent collisions. Through... Constraints can prevent the relay dog ​​from getting too close to the base station and maintain the scalability of the link.

[0087] In this way, all constraints can be taken into account to set a navigation strategy that better suits the needs of the relay dog.

[0088] Optionally, as an example, ways to fuse the calculated multiple forces include: The obstacle repulsion force is defined as the highest priority, the communication gradient driving force as the second priority, and the communication attraction force and the base station repulsion force as the lowest priority.

[0089] It should be noted that collisions will damage the robot dog, so obstacle repulsion should be the highest constraint, hence the obstacle repulsion priority is set to the highest; future communication capabilities can prevent communication anomalies in the robot dog in advance, hence the communication gradient driving force is set to the secondary priority.

[0090] Calculate the highest priority force .

[0091] Calculate the force of the second priority .Will Projected onto On an orthogonal plane, we obtain .

[0092] By analogy, the forces at each level are calculated sequentially based on priority, and then projected sequentially to obtain the final resultant force. .

[0093] In this way, the relay dog ​​will never crash into obstacles in pursuit of a better signal, because the effect of the communication gradient propulsion is strictly limited to a safe direction of movement.

[0094] In addition, the specific methods for performing static anchoring include: Obtain the dwell point and base station of the navigation robot dog's broadcast.

[0095] The local gradient optimization algorithm is used to calculate the local communication quality gradients corresponding to the navigation robot dog's dwell point and the base station. The two local communication quality gradients and the obstacle repulsion force are then fused by averaging to obtain the static anchoring resultant force.

[0096] S41: Control the relay robot dog to perform collaborative operations.

[0097] Preferably, as an example, controlling the relay robot dog to perform cooperative tasks includes: The resultant force of dynamic following and the resultant force of static anchoring are collectively referred to as the final navigation resultant force.

[0098] The final navigation force is output as a control command to the underlying motion controller of the relay robot dog to drive the robot dog to perform tasks.

[0099] Thus, this invention perfectly integrates forward-looking gradient forces with reactive attraction / repulsion forces, and through a multi-modal switching mechanism, enables the relay robot dog to execute the optimal cooperative navigation strategy whether the lead dog is moving or stationary, ultimately achieving efficient, robust, and uninterrupted communication relay at disaster sites. Figure 2The diagram shows a comparison of navigation performance. Red X's represent physical obstacles. Red dots represent local optima that the traditional APF algorithm might mistakenly identify as having good signal strength. Green stars represent the communication optima calculated by this invention, i.e., the global optima. Blue squares represent the starting positions of the relay robot. Red dashed contour lines represent the potential field obtained by the traditional method. Green solid contour lines represent the potential field obtained by this invention.

[0100] As can be seen from the image, the path obtained by the traditional method, starting from the starting point, is attracted by the gravitational pull of spurious communication while simultaneously being repelled by physical repulsion. Ultimately, it fails to reach either target, instead getting stuck in a local optimum between the two forces, thus verifying the shortcomings of existing technologies.

[0101] The path obtained by this invention, also starting from the initial point, generates a powerful driving force towards the optimal solution due to the calculation of the forward-looking communication quality gradient field. This force successfully suppresses false gravitational interference, enabling the relay dog ​​to actively avoid traps and navigate directly to the global optimum.

[0102] The effect diagram powerfully demonstrates that the present invention can solve the local optima problem of traditional APF, actively avoid communication dead zones, and successfully navigate to the actual communication optimum, thereby ensuring the stability and optimization of the communication link.

[0103] This invention also discloses a robot dog formation collaborative operation system under disaster conditions, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, a robot dog formation collaborative operation method under disaster conditions according to the present invention is implemented.

[0104] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0105] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as resistive random access memory (DRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (DRAM), high-bandwidth memory, hybrid memory cube, etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device.

Claims

1. A method for collaborative operation of robot dogs in a disaster environment, characterized in that, Including the following steps: The system acquires the future intent information broadcast by the navigating robot dog and the local environment model perceived by the relay robot dog itself, and integrates them to construct a forward-looking environment model. Based on the aforementioned forward-looking environment model, the link quality score of the relay robot dog at the current location for the future intention information of the navigating robot dog is calculated, and the potential field of the link quality score is constructed. Calculate the gradient of the potential field of the link quality score to obtain the communication gradient driving force; The operation mode is switched according to the operation status of the navigation robot dog. In the operation mode, the communication gradient driving force, the communication attraction based on the current position of the navigation dog, and the obstacle repulsion force are integrated to calculate the final navigation resultant force, and the relay robot dog is controlled to perform cooperative operation.

2. The method for collaborative operation of robot dogs in a disaster environment according to claim 1, characterized in that, The acquisition of the future intent information broadcast by the navigation robot dog includes: Obtain the probabilistic path cone broadcast by the navigation robot dog, which consists of a central path and a set of variance parameters that expand over time.

3. The method for collaborative operation of robot dogs in a disaster environment according to claim 1, characterized in that, The fusion construction of the forward-looking environment model also includes: The relay robot dog uses sensors to build a 3D occupancy grid map in real time; Semantic assignment is performed on the voxels in the three-dimensional occupied grid map to obtain the material category of each voxel, and a semantic voxel map is constructed, which is denoted as the prospective environment model.

4. The method for collaborative operation of robot dogs in a disaster environment according to claim 3, characterized in that, The method for obtaining the semantic attribution includes: By fusing the coordinates of voxels, the intensity of LiDAR reflection, and color information, geometric, physical, and visual features are extracted to form a feature vector; The feature vector is input into a pre-trained random forest classifier, which outputs the material category for each voxel.

5. The method for collaborative operation of robot dogs in a disaster environment according to claim 1, characterized in that, The calculation of the link quality score of the relay robot dog at the current location for the navigation robot dog's future intention information includes: In the aforementioned forward-looking environment model, ray tracing is performed from the current position of the relay robot dog to multiple future path points in the future intention information of the navigator robot dog; For any obstacle element that the ray passes through, calculate the linear path loss based on its material type and penetration thickness; The link quality of the relay robot dog from its current location to the future path points is calculated based on the linear path loss, and the link quality of all future path points is fused to obtain the link quality score.

6. The method for collaborative operation of robot dogs in a disaster environment according to claim 5, characterized in that, The formula for calculating the linear path loss is as follows: ； ； in, Indicates the first ray that passes through Linear path loss of an obstacle, and Representing materials Fixed loss and loss per meter, N represents the relay dog ​​position. To future path point The number of obstacles on the ray between them Indicates the first ray that passes through The penetration thickness of an obstacle; Indicates the location of the relay dog For future path points The quality of linear links, For linear transmit power, Based on linear path loss.

7. The method for collaborative operation of robot dogs in a disaster environment according to claim 1, characterized in that, The calculation of the gradient of the potential field of the link quality score to obtain the communication gradient driving force includes: The gradient of the link quality score potential field is estimated by employing the finite difference method and repeatedly calling the calculation of the link quality score at the current position of the relay robot dog and its immediate neighbors' virtual sampling points. ; The formula for calculating the communication gradient driving force is: ； in, As a driving force for communication gradient, This represents the gradient gain coefficient.

8. The method for collaborative operation of robot dogs in a disaster environment according to claim 1, characterized in that, The step of switching the operating mode according to the operating status of the navigation robot dog includes: If the lead robot dog is in a moving exploration state, the relay robot dog switches to dynamic following mode; If the navigator robot dog is in a stationary operation state, the relay robot dog will switch to the static anchoring mode.

9. A method for collaborative operation of robot dogs in a disaster environment according to claim 8, characterized in that, In dynamic following mode, the formula for calculating the final navigation resultant force is: ； in, This is a dynamic following resultant force, also known as the final navigation resultant force. The communication gravity generated by the navigator's current location, Repulsive force from obstacles For base station repulsion, , , This is the corresponding gain coefficient. As a driving force for communication gradient, This represents the gradient gain coefficient.

10. A robot dog swarm cooperative operation system under disaster conditions, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a method for collaborative operation of robot dogs in a disaster environment according to any one of claims 1-9.

Images