A method and system for pedestrian patrol command and dispatch based on a robot dog
By constructing a cross-modal spatiotemporal graph parameter model and multi-constraint path optimization technology, the problems of insufficient data processing capabilities and inaccurate path planning of robot dogs in sidewalk patrols were solved, and efficient and safe patrol command and dispatch were achieved.
Patent Information
- Application Number
- CN202510312419.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-17
- Publication Date
- 2025-09-23
- Estimated Expiration
- 2045-03-17
AI Technical Summary
Existing robot dogs have problems with insufficient data processing capabilities and low path planning accuracy when patrolling sidewalks, resulting in poor patrol results and path deviation or collision.
By collecting multi-source heterogeneous data, building a cross-modal spatiotemporal graph parameter model, performing online graph reasoning calculations, generating decision-making instruction data, and performing multi-constraint path optimization and conflict detection, collision-free coordinated motion trajectory data is generated.
It has achieved efficient command and dispatch of robot dogs in sidewalk patrols, improved the intelligence level and management efficiency, and ensured the accuracy and safety of patrol routes.
Smart Images

Figure CN120235398B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of intelligent robots, and in particular to a method and system for realizing pedestrian patrol command and dispatch based on a robot dog. Background Art
[0002] With the acceleration of urbanization, the management and maintenance of sidewalks has become a crucial component of urban governance. Intelligent robotics technology is widely used in urban infrastructure inspections, and robot dogs offer advantages in complex environments. Currently, robot dogs are being used in areas such as power inspections and security patrols, but their application in sidewalk inspection and dispatch is still in the exploratory stage.
[0003] Existing robot dogs used in sidewalk patrols suffer from insufficient data processing capabilities and path planning accuracy. On the one hand, the integration and analysis of heterogeneous multi-source data collected during the patrol process is difficult, hindering patrol effectiveness. On the other hand, the complex and ever-changing sidewalk environment makes it difficult for traditional path planning methods to generate an optimal path in real time, making collisions and path deviations more likely to occur. This invention effectively addresses these issues by constructing a cross-modal spatiotemporal graph parameter model and employing multi-constraint path optimization technology, enabling efficient command and dispatch of robot dogs in sidewalk patrols. Summary of the Invention
[0004] In view of the above existing problems, the present invention is proposed.
[0005] Therefore, the present invention provides a method for implementing pedestrian patrol command and dispatch based on a robot dog to solve the problems of inaccurate path planning and how to achieve efficient command and dispatch in the existing robot dog pedestrian patrol method.
[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:
[0007] In a first aspect, the present invention provides a method for implementing pedestrian patrol command and dispatch based on a robot dog, which comprises collecting multi-source heterogeneous data of pedestrian infrastructure, environment, and robot dogs and preprocessing them to obtain a standardized data stream;
[0008] Construct a cross-modal spatiotemporal graph parameter model, and input multi-source heterogeneous data into the cross-modal spatiotemporal graph parameter model to obtain an initialized graph;
[0009] Perform online graph reasoning on the initialized graph to obtain the priority scores of each node of the sidewalk infrastructure, environment, and robot dog;
[0010] The priority score of each node is integrated with multi-source heterogeneous data, and decision instruction data is generated through the decision tree algorithm;
[0011] Perform multi-constraint path optimization and conflict detection on decision instruction data to obtain collision-free coordinated motion trajectory data;
[0012] The collision-free coordinated motion trajectory data is translated into action-specifying codes through a motion control instruction mapping algorithm and transmitted to the robotic dog for sidewalk patrol command.
[0013] As a preferred solution of the method for realizing pedestrian patrol command and dispatch based on a robot dog according to the present invention, wherein: by collecting multi-source heterogeneous data of pedestrian infrastructure, environment and robot dogs and pre-processing them, a standardized data stream is obtained, and the specific steps are as follows:
[0014] The multi-source heterogeneous data includes pedestrian infrastructure distribution data, environmental parameters and robot dog motion status data;
[0015] Classify and arrange the collected multi-source heterogeneous data and transmit them to the data processing center;
[0016] The classified and arranged multi-source heterogeneous data are cleaned using the Bayesian algorithm and filled with missing values using the KNN algorithm to obtain complete multi-source heterogeneous data;
[0017] The complete multi-source heterogeneous data is uniformly mapped to a fixed interval through the data normalization method to obtain a standardized data stream.
[0018] As a preferred solution of the method for realizing pedestrian patrol command and dispatch based on a robot dog according to the present invention, a cross-modal spatiotemporal graph parameter model is constructed, and multi-source heterogeneous data is input into the cross-modal spatiotemporal graph parameter model to obtain an initialized graph. The specific steps are as follows:
[0019] Generate adversarial networks to mine and enhance standardized data streams to obtain feature data;
[0020] The characteristic data includes sidewalk infrastructure distribution data, environmental parameters and robot dog motion status data;
[0021] A cross-modal spatiotemporal graph parameter model is constructed based on the feature data, and the feature data is substituted into the calculation formula of the cross-modal spatiotemporal graph parameter model to generate a node priority score;
[0022] The model parameters of the sidewalk infrastructure, environment and robot dog are respectively introduced into the calculation formula of the cross-modal spatiotemporal graph parameter model to generate the initialization graph of the cross-modal spatiotemporal graph.
[0023] As a preferred solution of the method for realizing pedestrian patrol command and dispatch based on a robot dog according to the present invention, the following steps are performed: online graph reasoning calculation is performed on the initialized graph to obtain the priority scores of each node of the pedestrian infrastructure, environment and robot dog.
[0024] The graph neural network is used to calculate the spatiotemporal correlation characteristics of each node in the initial graph, including the sidewalk infrastructure, environment, and robot dog, and to generate preliminary calculation results for the priority score of each node;
[0025] Based on the preliminary calculation results and combined with multi-source heterogeneous data, the priority score of each node is calculated by weighted summation;
[0026] Based on the calculation results of the priority score of each node, the reinforcement learning algorithm is used to adjust the parameters α, β and γ in the cross-modal spatiotemporal graph parameter model;
[0027] According to the adjusted α, β, and γ in the cross-modal spatiotemporal graph parameter model, the priority score of each node is recalculated to generate the priority scores of each node of the sidewalk infrastructure, environment, and robot dog.
[0028] As a preferred solution of the method for realizing pedestrian patrol command and dispatch based on robot dogs described in the present invention, the priority score of each node and multi-source heterogeneous data are integrated, and decision instruction data are generated through a decision tree algorithm. The specific steps are as follows:
[0029] Add node priority scores to multi-source heterogeneous data to form a fused dataset;
[0030] Perform feature selection on the fused dataset to filter out characteristic parameters of sidewalk infrastructure distribution data, environmental parameters, and the robot dog's motion state;
[0031] Using the filtered sidewalk infrastructure distribution data, environmental parameters, and characteristic parameters of the robot dog's motion state as input, a decision tree algorithm is used to calculate the decision instruction data.
[0032] Convert the obtained decision instruction data into JSON format.
[0033] As a preferred solution of the pedestrian patrol command and dispatch method based on the robot dog described in the present invention, wherein: multi-constraint path optimization and conflict detection are performed on the decision instruction data to obtain collision-free coordinated motion trajectory data, the specific steps are as follows:
[0034] Parse the decision instruction data from JSON format into a sequence of waypoints containing location information and timestamps;
[0035] The spatial segmentation technology is used to divide the path point sequence into multiple sub-path segments, and conflict detection is performed on each sub-path segment;
[0036] Use the A algorithm to perform multi-constraint path optimization on each sub-path segment, evaluate the quality of the path through the heuristic function, and optimize each sub-path segment;
[0037] Convert the optimized sub-path point sequence into collision-free coordinated motion trajectory data.
[0038] As a preferred solution of the method for realizing pedestrian patrol command and dispatch based on a robot dog according to the present invention, the collision-free coordinated motion trajectory data is translated into action-specifying codes through a motion control instruction mapping algorithm and transmitted to the robot dog for pedestrian patrol command. The specific steps are as follows:
[0039] Parse the collision-free coordinated motion trajectory data into a sequence of waypoints containing position information and timestamps;
[0040] Use motion control command mapping algorithm to convert the waypoint sequence into action-specifying code for the robot dog to execute;
[0041] The action-specifying code is encapsulated according to the communication protocol of the robot dog and sent to the controller of the robot dog using a wireless remote control device.
[0042] In a second aspect, the present invention provides a pedestrian patrol command and dispatch system based on a robot dog, comprising:
[0043] The data processing module collects and preprocesses multi-source heterogeneous data from sidewalk infrastructure, the environment, and the robot dog to obtain a standardized data stream;
[0044] The modeling module constructs a cross-modal spatiotemporal graph parameter model and inputs multi-source heterogeneous data into the cross-modal spatiotemporal graph parameter model to obtain an initialized graph;
[0045] The scoring module performs online graph reasoning on the initialized graph to obtain the priority scores of each node of the sidewalk infrastructure, environment, and robot dog;
[0046] The instruction generation module integrates the priority scores of each node with multi-source heterogeneous data and generates decision instruction data through a decision tree algorithm;
[0047] The path detection module performs multi-constraint path optimization and conflict detection on the decision instruction data to obtain collision-free coordinated motion trajectory data;
[0048] The control module translates the collision-free coordinated motion trajectory data into action-specifying codes through the motion control instruction mapping algorithm, and transmits them to the robotic dog for sidewalk patrol command.
[0049] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the method for realizing pedestrian patrol command and dispatch based on a robot dog as described in the first aspect of the present invention is implemented.
[0050] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the method for implementing pedestrian patrol command and dispatch based on a robot dog as described in the first aspect of the present invention is implemented.
[0051] The beneficial effects of the present invention are as follows: by collecting and preprocessing multi-source heterogeneous data, constructing a cross-modal spatiotemporal graph parameter model, combining graph neural networks and reinforcement learning algorithms to perform online graph reasoning calculations, and accurately evaluating the priority scores of each node. At the same time, a decision tree algorithm is used to generate decision instruction data, and multi-constraint path optimization and conflict detection are performed to ensure that the motion trajectory is collision-free. Finally, the optimized trajectory data is converted into action-specifying codes through a motion control instruction mapping algorithm to achieve efficient command and dispatch of the robot dog. This series of innovative measures has significantly improved the intelligence level and management efficiency of robot dogs in sidewalk patrols, effectively solved the shortcomings of existing technologies, and provided more efficient and accurate technical support for the management and maintenance of urban sidewalks. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0053] Figure 1 This is a flow chart of the pedestrian patrol command and dispatch method based on a robot dog in Example 1.
[0054] Figure 2 This is a flow chart of path detection for the pedestrian patrol command and dispatch method based on a robot dog in Example 1. DETAILED DESCRIPTION
[0055] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.
[0056] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0057] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.
[0058] Example 1, with reference to Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a method for implementing pedestrian patrol command and dispatch based on a robot dog, including the following steps:
[0059] S1. By collecting and preprocessing multi-source heterogeneous data from sidewalk infrastructure, environment, and robot dogs, a standardized data stream is obtained.
[0060] Specifically, the multi-source heterogeneous data includes sidewalk infrastructure distribution data, environmental parameters, and robot dog motion status data.
[0061] It should be noted that after the patrol mission begins, the robot dog's LiDAR scans the sidewalk at a frequency of 10Hz, generating point cloud data that accurately captures the location and shape of facilities. The camera captures images at a frame rate of 30fps, which are then processed by a convolutional neural network to extract facility status information. The two are combined to form a complete facility profile. Simultaneously, various sensors monitor environmental parameters such as temperature, humidity, light, and noise at a frequency of 1Hz, and the robot dog's own sensors collect motion status data, including posture, position, and speed, at a frequency of 100Hz. All data is transmitted to the data processing center via wireless or wired networks, encrypted using AES technology to ensure data security, and stored in a distributed file system or relational database. This process comprehensively captures the data required for the patrol, providing rich information for subsequent processing and analysis, and improving patrol efficiency and quality.
[0062] Specifically, the collected multi-source heterogeneous data are classified, arranged and transmitted to the data processing center.
[0063] It should be noted that the collected data is categorized at the data processing center. Sidewalk infrastructure distribution data is categorized by facility type and arranged in time series. Environmental parameters and robot dog motion status data are categorized by type and arranged in time series. Infrastructure distribution data is also spatially arranged. The categorized and arranged data is packaged in JSON or XML format and transmitted to the data processing center via wireless or wired networks. AES encryption ensures data security. After decompression, it is stored in a distributed file system or relational database. This process effectively manages transmitted data, improves data processing efficiency and reliability, and supports efficient command and dispatch of the robot dogs.
[0064] Specifically, the classified and arranged multi-source heterogeneous data are cleaned using the Bayesian algorithm and the missing values are filled using the KNN algorithm to obtain complete multi-source heterogeneous data.
[0065] It should be noted that after categorization and arrangement, outliers are detected using a Bayesian algorithm and removed through median or mean filtering to ensure data consistency. Missing values are filled using the KNN algorithm, which finds the nearest neighbor data points and calculates a weighted average. The filled data is then checked for completeness. This process improves data integrity and quality, providing accurate and reliable data for subsequent processing and analysis, and enhancing the intelligence level of inspections and management efficiency.
[0066] Specifically, the complete multi-source heterogeneous data is uniformly mapped to a fixed interval through the data normalization method to obtain a standardized data stream.
[0067] It should be noted that the cleaned and padded data is mapped to the [0, 1] interval using the min-max normalization method. The minimum and maximum values for each data type are calculated and processed using normalization formulas to generate a standardized data stream. The normalized data is then validated for performance and distribution to ensure that the data falls within the specified interval and that its distribution is consistent with the original data. This process standardizes the data, improving processing efficiency and accuracy, providing reliable data support for subsequent processing and analysis, and enhancing the intelligence level of inspections and management efficiency.
[0068] S2. Construct a cross-modal spatiotemporal graph parameter model and input multi-source heterogeneous data into the cross-modal spatiotemporal graph parameter model to obtain an initialized graph.
[0069] Specifically, generative adversarial networks are used to mine and enhance standardized data streams.
[0070] It should be noted that the normalized data stream is input into the generator of a generative adversarial network. The generator extracts features and transforms the data through a multi-layer neural network structure. The generator extracts features from the input normalized data and generates augmented data with a distribution similar to the real data. The generator's network structure typically includes multiple convolutional layers and fully connected layers, which perform nonlinear transformations using activation functions (such as ReLU). The discriminator receives the data generated by the generator and the real data, and classifies the data using a multi-layer neural network structure to determine whether the data is generated or real. The discriminator's network structure typically includes multiple convolutional layers and fully connected layers, which output classification results using activation functions (such as Sigmoid). The generator and discriminator are continuously optimized through adversarial training. The goal of the generator is to generate data with a distribution similar to the real data, while the goal of the discriminator is to distinguish between the generated data and the real data. Through continuous iterative optimization, the data generated by the generator gradually approaches the real data distribution. After adversarial training, the augmented data generated by the generator not only retains the characteristics of the original data but also enhances the data's robustness and generalization ability. The enhanced data serves as input for the subsequent cross-modal spatiotemporal graph parameter model. By utilizing a generative adversarial network (GAN) to mine and enhance the standardized data stream, deep feature extraction and enhancement are achieved. Through adversarial training of the generator and discriminator, the GAN effectively captures the complex distribution of data and generates enhanced data that resembles the real data distribution. This not only enhances the robustness and generalization capabilities of the data, but also improves its expressiveness in the cross-modal spatiotemporal graph parameter model. The enhanced data better reflects the characteristics of the sidewalk infrastructure, environment, and the robot dog's motion state, providing more accurate and reliable data support for the subsequent node priority scoring calculations, thereby improving the intelligence level and management efficiency of the entire system.
[0071] Specifically, the characteristic data includes sidewalk infrastructure distribution data, environmental parameters and robot dog motion state data.
[0072] It should be noted that the enhanced standardized data stream contains sidewalk infrastructure distribution data, environmental parameters, and robot dog motion data, defined as M, E, and D, respectively. The enhanced data stream is categorized by data type: sidewalk infrastructure distribution data includes the location, shape, and status of facilities; environmental parameters include environmental information such as temperature, humidity, light, and noise; and robot dog motion data includes motion information such as posture, position, velocity, and acceleration. These categorized data are defined as M, E, and D, respectively. M represents sidewalk infrastructure distribution data, E represents environmental parameters, and D represents robot dog motion data. The defined data sets M, E, and D are structured to generate a structured dataset. This structured dataset facilitates the subsequent construction and calculation of a cross-modal spatiotemporal graph parameter model. This definition not only simplifies the data processing process but also improves data operability and manageability. In the subsequent construction of the cross-modal spatiotemporal graph parameter model, different types of feature data can be more easily incorporated into the model to calculate node priority scores. Furthermore, this definition provides a clear basis for model parameter adjustment and optimization, improving model accuracy and reliability, thereby further enhancing the system's intelligence and management efficiency.
[0073] Specifically, a cross-modal spatiotemporal graph parameter model is constructed based on the feature data, and the feature data is substituted into the calculation formula of the cross-modal spatiotemporal graph parameter model to generate a node priority score.
[0074] It should be noted that the enhanced feature data M, E, and D are introduced into the cross-modal spatiotemporal graph parameter model to construct the calculation formula, which is as follows:
[0075]
[0076] Among them, G is the node priority score, M(t) is the feature data of the sidewalk infrastructure distribution data at time t, E(t) is the feature data of the environmental parameters at time t, D(t) is the feature data of the robot dog at time t, W is the length of the time interval, α is the model parameter for adjusting the weight of the municipal feature data in the spatiotemporal dimension, β is the model parameter for adjusting the weight of the environmental feature data in the spatiotemporal dimension, γ is the model parameter for adjusting the weight of the robot dog feature data in the spatiotemporal dimension, dt is the increment of time t, and the enhanced feature data M, E and D are input into the cross-modal spatiotemporal graph parameter model. The input data of the model include the sidewalk infrastructure distribution data M, the environmental parameters E and the robot dog motion state data D, and the input feature data M, E and D are weighted. The weighting coefficients are α, β and γ, which represent the model parameters for adjusting the weight of the municipal feature data, environmental feature data and robot dog feature data in the spatiotemporal dimension, respectively, and the weighted feature data are subjected to time decay processing. The time decay coefficients are e -αt 、e-βt and e -γt , represents the degree of decay of the feature data at time t. The time-decayed feature data is integrated to generate the numerator of the node priority score, with the integration interval being [0, W], where W is the length of the time interval. The time-decayed feature data is normalized to generate the denominator of the node priority score. Normalization is achieved through integration over the interval [0, W]. The numerator and denominator are divided to generate the node priority score G. The calculation formula integrates different types of feature data through a weighted summation to calculate the priority score for each node. This comprehensive assessment not only considers the characteristics of various data types but also highlights the importance of different feature data in both temporal and spatial dimensions through the adjustment of model parameters. This makes the node priority score more accurate and reliable, better reflecting the actual situation during sidewalk inspections and providing strong data support for subsequent decision-making instructions, thereby improving the system's intelligence and management efficiency.
[0077] Specifically, the model parameters of the sidewalk infrastructure, environment, and robot dog are respectively introduced into the calculation formula of the cross-modal spatiotemporal graph parameter model to generate an initialization graph of the cross-modal spatiotemporal graph.
[0078] It should be noted that the model parameters α, which adjust the spatiotemporal weighting of municipal feature data, β, which adjusts the spatiotemporal weighting of environmental feature data, and γ, which adjusts the spatiotemporal weighting of the robot dog feature data, are input into the calculation formula. By adjusting the values of the model parameters α, β, and γ, the weighting coefficients in the calculation formula are optimized. Parameter adjustment can be achieved using gradient descent or other optimization algorithms. Substituting the adjusted model parameters into the calculation formula generates an initialization graph for the cross-modal spatiotemporal graph. In the initialization graph, the priority score of each node reflects the combined characteristics of the sidewalk infrastructure, the environment, and the robot dog's motion state. This initialization graph of the generated cross-modal spatiotemporal graph is output as the basis for subsequent online graph inference calculations. In the initialization graph, the priority score of each node comprehensively considers the characteristics of the sidewalk infrastructure, the environment, and the robot dog's motion state. By adjusting the model parameters, the importance of different feature data in the spatiotemporal dimensions is highlighted. This makes the initialization graph more accurate and reliable, better reflecting the actual situation during sidewalk inspections. The initialization graph provides a solid foundation for subsequent online graph reasoning calculations, enabling the system to more accurately evaluate the priority of each node, thereby improving the system's intelligence level and management efficiency, and providing more efficient and accurate technical support for the management and maintenance of urban sidewalks.
[0079] S3. Perform online graph reasoning calculations on the initialized graph to obtain the priority scores of each node of the sidewalk infrastructure, environment, and robot dog.
[0080] Specifically, the graph neural network is used to calculate the spatiotemporal correlation characteristics of each node of the sidewalk infrastructure, environment and robot dog in the initialization graph, and generate preliminary calculation results of the priority score of each node.
[0081] It should be noted that the initial cross-modal spatiotemporal graph is input into the graph neural network. Each node in the graph contains feature data for sidewalk infrastructure, the environment, and the robot dog, as well as the spatiotemporal relationships between nodes. The graph neural network extracts and transforms node features through a multi-layer neural network structure. Each layer of the neural network aggregates the feature information of the node and its neighbors to generate a new node feature representation. This process captures the node's associated features in the spatiotemporal dimensions. During feature extraction, the graph neural network pays particular attention to the spatiotemporal relationships between nodes. By aggregating the feature information of neighboring nodes, the network learns the temporal and spatial dependencies between nodes, thereby generating more accurate node feature representations. Based on the extracted node features, the graph neural network generates preliminary priority scores for each node. These preliminary scores reflect the importance of the node in the spatiotemporal dimensions and provide a basis for the subsequent weighted summation calculation. The preliminary results are output as input data for the subsequent weighted summation calculation. This process not only considers the node's own characteristics but also the spatiotemporal relationships between nodes, making the preliminary scoring more accurate and reliable. This provides high-quality input data for the subsequent weighted summation calculation, thereby improving the intelligence level and management efficiency of the entire system.
[0082] Specifically, based on the preliminary calculation results and combined with multi-source heterogeneous data, the priority score G of each node is calculated by weighted summation.
[0083] It should be noted that the preliminary calculation results are fused with multi-source heterogeneous data. Multi-source heterogeneous data includes sidewalk infrastructure distribution data, environmental parameters, and robot dog motion status data, and weighted coefficients are set for each type of data. The weighted coefficients reflect the importance of different types of data in the node priority score. For example, the weighted coefficient of the sidewalk infrastructure distribution data is set to α, the weighted coefficient of the environmental parameters is set to β, and the weighted coefficient of the robot dog motion status data is set to γ. The preliminary calculation results are fused with multi-source heterogeneous data through weighted summation to calculate the priority score G of each node. The calculation formula is as follows:
[0084] G=α·G 初步 +β·E+γ·D;
[0085] Among them, G 初步As a preliminary calculation result, the calculated priority score G for each node is output as input data for subsequent reinforcement learning algorithm model parameter adjustment. By combining multi-source heterogeneous data and using a weighted summation method to calculate each node's priority score G, a comprehensive assessment of the node in both temporal and spatial dimensions is achieved. This process not only considers the preliminary calculation results but also incorporates multi-source heterogeneous data on sidewalk infrastructure, the environment, and the robot dog's motion status, making the priority scoring more comprehensive and accurate. The setting of weighting coefficients allows the importance of different types of data to be reflected, further improving the reliability and practicality of the scoring results. This provides high-quality input data for subsequent reinforcement learning algorithm model parameter adjustment, thereby improving the overall intelligence level and management efficiency of the entire system.
[0086] Specifically, according to the calculation result of the priority score G of each node, the reinforcement learning algorithm is used to adjust the parameters α, β and γ in the cross-modal spatiotemporal graph parameter model.
[0087] It should be noted that the priority score G of each node is used as input data and input into the reinforcement learning algorithm.
[0088] Model parameter initialization: Initialize α, β, and γ in the cross-modal spatiotemporal graph parameter model. These parameters represent the weights of the sidewalk infrastructure distribution data, environmental parameters, and the robot dog's motion state data in the spatiotemporal dimension.
[0089] Reinforcement Learning Training: Using a reinforcement learning algorithm, the model parameters α, β, and γ are adjusted based on the calculated node priority score G. Through interaction with the environment, the reinforcement learning algorithm learns the optimal parameter adjustment strategy, enabling the model to better adapt to different node priority scores. Based on the training results of the reinforcement learning algorithm, α, β, and γ in the cross-modal spatiotemporal graph parameter model are updated. The updated parameters better reflect the importance of different data types in node priority scores. The updated model parameters are output as input for subsequent node priority score recalculations. By using the reinforcement learning algorithm to adjust α, β, and γ in the cross-modal spatiotemporal graph parameter model, dynamic optimization of the model parameters is achieved. This process enables the model to automatically adjust the weights of different data types based on the calculated node priority score G, thereby better adapting to different node priority scores. Through interaction with the environment, the reinforcement learning algorithm learns the optimal parameter adjustment strategy, improving the model's adaptability and robustness. This provides more accurate and reliable model parameters for subsequent node priority score recalculations, thereby enhancing the overall intelligence and management efficiency of the system.
[0090] Specifically, according to the adjusted α, β, and γ in the cross-modal spatiotemporal graph parameter model, the priority score G of each node is recalculated to generate the priority scores of each node of the sidewalk infrastructure, environment, and robot dog.
[0091] It should be noted that the adjusted model parameters α, β, and γ, along with preliminary calculation results and multi-source heterogeneous data, are used as input. The adjusted model parameters are used to recalculate the priority score G for each node. This recalculated priority score G is then output to generate node priority scores for the sidewalk infrastructure, environment, and robot dog. These node priority scores are then used in subsequent decision-making and scheduling, providing data support for sidewalk patrol and control. By recalculating the priority score G for each node based on the adjusted model parameters, more accurate and reliable node priority scores are generated. This process enables the node priority scores to better reflect the comprehensive characteristics of the sidewalk infrastructure, environment, and robot dog's motion state, providing high-quality data support for subsequent decision-making and scheduling. The adjusted model parameters are more adaptable to different node priority scores, improving the accuracy and reliability of the scoring results. This provides more efficient and accurate data support for sidewalk patrol and control, thereby enhancing the overall intelligence level and management efficiency of the system.
[0092] S4. Integrate the priority scores of each node with multi-source heterogeneous data, and generate decision instruction data through the decision tree algorithm.
[0093] Specifically, the node priority score G is added to the multi-source heterogeneous data to form a fused dataset.
[0094] It should be noted that multi-source heterogeneous data, including sidewalk infrastructure distribution data, environmental parameters, and robot dog motion status data, is acquired. This data has been preprocessed, cleaned, padded, and normalized to form a standardized data stream. The calculated node priority score G is then fused with the multi-source heterogeneous data. Specifically, each node priority score G is added as a new feature column to the corresponding data record. For example, for sidewalk infrastructure distribution data, the corresponding node priority score G is added to the record of each facility; for environmental parameters and robot dog motion status data, the corresponding node priority score G is similarly added to the record at each time point. The multi-source heterogeneous data, with the added node priority scores G, is then integrated into a fused dataset. This dataset, which includes the original multi-source heterogeneous data and the newly added node priority scores G, forms a richer dataset, providing a foundation for subsequent feature selection and decision instruction generation. The fused dataset is stored in an appropriate data storage system, such as a distributed file system (e.g., Hadoop HDFS) or a relational database (e.g., MySQL), for subsequent processing and analysis. By adding each node priority score G to the multi-source heterogeneous data, a fused dataset is formed, achieving data enrichment and enhancement. This process not only preserves the characteristics of the original multi-source heterogeneous data but also incorporates the crucial node priority score G, making the dataset more comprehensive and representative. The fused dataset provides richer data support for subsequent feature selection and decision-making instruction generation, improving the accuracy and reliability of decision-making. Furthermore, the fused dataset facilitates subsequent data management and storage, providing a strong foundation for the efficient operation of the entire system.
[0095] Specifically, feature selection is performed on the fused dataset to screen out the characteristic parameters of the sidewalk infrastructure distribution data, environmental parameters, and the robot dog's motion state.
[0096] It should be noted that the fused dataset is loaded from the storage system and ready for feature selection. All feature columns in the fused dataset are identified, including the original multi-source heterogeneous data features and the newly added node priority score G. Based on the importance and relevance of the features, feature parameters for the sidewalk infrastructure distribution data, environmental parameters, and the robot dog's motion state are selected. Specifically, statistical methods (such as variance analysis and correlation analysis) or machine learning methods (such as feature importance assessment) can be used to determine which features are most valuable for subsequent decision instruction generation. The selected feature parameters are extracted from the fused dataset to form a new feature dataset. This feature dataset contains key features of the sidewalk infrastructure distribution data, environmental parameters, and the robot dog's motion state, as well as the node priority score G. The extracted feature data is preprocessed, such as through standardization and normalization, to ensure that the feature data is on the same scale for subsequent processing by the decision tree algorithm. Feature selection on the fused dataset selects key feature parameters for the sidewalk infrastructure distribution data, environmental parameters, and the robot dog's motion state, achieving data simplification and optimization. This process not only removes redundant and irrelevant features but also retains the most valuable features for decision instruction generation, improving data utilization efficiency and decision accuracy. The dataset after feature selection is more compact and efficient, reducing the computational complexity of subsequent processing and improving the system's operational efficiency. In addition, feature selection improves the model's generalization ability, making decision instruction generation more reliable and stable.
[0097] Specifically, the decision tree algorithm is used to calculate the decision instruction data using the screened sidewalk infrastructure distribution data, environmental parameters and characteristic parameters of the robot dog's motion state as input.
[0098] It should be noted that the filtered feature parameter dataset is used as input data for training a decision tree algorithm. Specifically, the decision tree algorithm recursively partitions the dataset to generate a series of decision rules, forming a decision tree. Each decision node represents a judgment condition for a feature, and each leaf node represents a decision outcome. During training, decision tree algorithm parameters, such as tree depth, splitting criteria (such as information gain and Gini coefficient), and minimum sample size, are set to control the complexity and generalization of the decision tree. The decision tree model is optimized through methods such as cross-validation and pruning to prevent overfitting and improve generalization and prediction accuracy. The trained decision tree model is then used to predict the input data and generate decision instruction data. This decision instruction data contains specific instructions for sidewalk inspections, such as inspection routes and key inspection areas. The generated decision instruction data is output in a suitable format, such as JSON, for subsequent path optimization and conflict detection. By using the filtered feature parameters as input and calculating the decision instruction data using the decision tree algorithm, data-driven intelligent decision-making is achieved. This process not only fully utilizes the selected key characteristic parameters but also generates accurate and reliable decision instructions through the efficiency and interpretability of the decision tree algorithm. The decision tree algorithm is capable of handling complex nonlinear relationships, and the generated decision instruction data is highly accurate and practical. Furthermore, the interpretability of the decision tree model makes the decision process more transparent, facilitating subsequent path optimization and conflict detection, and improving the overall system's intelligence and management efficiency.
[0099] Specifically, the obtained decision instruction data is converted into JSON format.
[0100] It should be noted that the generated decision instruction data is obtained and prepared for format conversion. The decision instruction data is structured into key-value pairs. For example, information such as the inspection route and key inspection areas is used as the key, and the corresponding value is the specific instruction content. JSON encoding rules are used to convert the structured data into JSON format. Specifically, the key-value pairs are enclosed in double quotes, the key and value are separated by a colon, multiple key-value pairs are separated by commas, and the entire data set is enclosed in curly braces. The converted JSON data is validated to ensure data integrity and correctness. JSON validation tools or scripts can be used for validation, and the validated JSON-formatted decision instruction data is output for subsequent path optimization and conflict detection. By converting the obtained decision instruction data into JSON format, data standardization and normalization are achieved. JSON is lightweight, easy to read, and easy to parse, making it easier to process in the subsequent path optimization and conflict detection modules. This process not only improves data readability and maintainability, but also ensures data compatibility and consistency across different modules. JSON-formatted data facilitates storage and transmission, reduces data processing complexity, and improves system efficiency. In addition, the standardization of the JSON format makes the entire system more modular and extensible, facilitating subsequent maintenance and upgrades.
[0101] S5. Perform multi-constraint path optimization and conflict detection on the decision instruction data to obtain collision-free coordinated motion trajectory data.
[0102] Specifically, the decision instruction data is parsed from the JSON format into a sequence of path points containing location information and timestamps.
[0103] It should be noted that the generated JSON-formatted decision instruction data is loaded from the storage system and prepared for parsing. A JSON parsing library (such as Python's json module) is used to parse the JSON data. During the parsing process, the key-value pairs in the JSON data are extracted one by one and converted into a data structure (such as a dictionary or object) that can be processed by the program. Waypoint information is then extracted from the parsed data structure. Waypoint information includes location information (such as latitude and longitude or coordinates) and a timestamp. Each waypoint represents the location that the robot dog should reach at a specific time. The extracted waypoints are arranged in timestamp order to construct a waypoint sequence. The waypoint sequence is an ordered list of waypoints representing the robot dog's patrol path. The constructed waypoint sequence is verified to ensure the integrity and correctness of the waypoints. Verification includes whether the location information of the waypoints is reasonable and whether the timestamps are continuous. By parsing the decision instruction data from JSON format into a waypoint sequence containing location information and timestamps, the data is structured and operational. This process not only converts the JSON-formatted data into a form that can be directly processed by the program but also ensures the integrity and correctness of the waypoint sequence. The construction of the pathpoint sequence provides basic data for subsequent path optimization and conflict detection, improving the system's operational efficiency and reliability. In addition, the structured form of the pathpoint sequence facilitates subsequent storage and transmission, reducing the complexity of data processing.
[0104] Specifically, a space segmentation technique is used to divide the path point sequence into multiple sub-path segments, and conflict detection is performed on each sub-path segment.
[0105] It should be noted that spatial segmentation techniques (such as grid segmentation or quadtree segmentation) are used to divide the pathpoint sequence into multiple sub-path segments. Spatial segmentation divides the spatial region into multiple sub-regions and assigns points in the pathpoint sequence to corresponding sub-regions to form sub-path segments. Based on the spatial segmentation results, multiple sub-path segments are generated. Each sub-path segment contains a certain number of pathpoints, representing the robot dog's patrol path within a specific area. Collision detection is performed on each sub-path segment. Conflict detection ensures path safety by checking whether pathpoints in a sub-path segment collide with other objects (such as obstacles, other robot dogs, etc.). Conflict detection can be implemented using collision detection algorithms (such as bounding box detection and distance detection). Sub-path segments detected to be in conflict are marked for subsequent path optimization. The marked information includes the location, type, and severity of the conflict. The divided sub-path segments and the conflict detection results are output as input data for subsequent path optimization. By using spatial segmentation to divide the pathpoint sequence into multiple sub-path segments and performing conflict detection on each sub-path segment, refined path management and safety checks are achieved. This process not only breaks down complex pathpoint sequences into manageable sub-paths but also ensures path safety through conflict detection. Spatial segmentation improves path management efficiency, while conflict detection ensures the robot dog's safety during patrols. Furthermore, conflict detection results provide crucial information for subsequent path optimization, enhancing the system's intelligence and management efficiency.
[0106] Specifically, the A algorithm is used to perform multi-constraint path optimization on each sub-path segment, and the quality of the path is evaluated through a heuristic function, and each sub-path segment is optimized.
[0107] It should be noted that after obtaining the divided sub-path segments and conflict detection results, path optimization is prepared and the parameters of the A algorithm, including the heuristic function, weight coefficient, and search range, are initialized. The heuristic function is used to evaluate the quality of the path, and the weight coefficient is used to balance the influence of different constraints. For each sub-path segment, the A algorithm performs multi-constraint path optimization. The A algorithm searches the path space to find the optimal path that meets the constraints. During the optimization process, multiple constraints such as path length, time, and energy consumption are considered. During the path optimization process, the heuristic function is used to evaluate the quality of the path. The heuristic function calculates a path score based on path characteristics (such as distance, time, and obstacle distance) to guide the search direction. Based on the search results of the A algorithm, the path point sequence of the sub-path segment is updated. The updated path point sequence represents the optimized inspection path. The optimized sub-path segment is verified to ensure the feasibility and safety of the path. Verification includes whether the path meets the constraints and whether there are conflicts with other objects. By using the A algorithm to perform multi-constraint path optimization on each sub-path segment and evaluating the quality of the path through the heuristic function, efficient path optimization is achieved. This process not only considers multiple constraints such as path length, time, and energy consumption, but also improves search efficiency through heuristic functions. The use of algorithm A ensures path optimality, while the heuristic function evaluation improves the path's adaptability and robustness. The optimized path not only improves the robot dog's patrol efficiency but also ensures safety during the patrol process. Furthermore, the path optimization results provide a foundation for subsequent path coordination, enhancing the system's intelligence and management efficiency.
[0108] Specifically, the optimized sub-path point sequence is converted into collision-free coordinated motion trajectory data.
[0109] It should be noted that the optimized sub-path point sequences are integrated to form complete motion trajectory data. During the integration process, smooth connections between sub-path segments are ensured to avoid path interruptions or abrupt changes. Collision detection is performed on the integrated motion trajectory data to ensure collision-free paths. Collision detection ensures path safety by checking whether path points collide with other objects (such as obstacles, other robot dogs, etc.). The motion trajectory data is then smoothed to generate a smooth trajectory. Smoothing can be achieved using methods such as spline curves and Bezier curves to ensure smooth robot movement. The smoothed trajectory data is then converted into executable motion instructions for the robot dog. During the conversion process, the path point sequence is converted into specific motion parameters (such as speed, acceleration, steering angle, etc.). The converted collision-free coordinated motion trajectory data is output as the robot dog's final patrol path. By converting the optimized sub-path point sequence into collision-free coordinated motion trajectory data, the path is finally optimized and coordinated. This process not only ensures collision-free paths but also improves the robot dog's smoothness through trajectory smoothing. Data conversion converts the path point sequence into specific motion instructions, ensuring that the robot dog can accurately perform its patrol tasks. The resulting collision-free coordinated motion trajectory data improves the robot dog's patrol efficiency and safety, ensuring the successful completion of patrol missions. Furthermore, this process enhances the system's intelligence and management efficiency, providing efficient technical support for the management and maintenance of urban sidewalks.
[0110] S6. The collision-free coordinated motion trajectory data is translated into action-specifying codes through a motion control instruction mapping algorithm and transmitted to the robot dog for sidewalk patrol command.
[0111] Specifically, the collision-free coordinated motion trajectory data is parsed into a path point sequence containing position information and timestamps.
[0112] It should be noted that the generated collision-free coordinated motion trajectory data is loaded from the storage system and prepared for parsing. A data parsing algorithm is used to decompose the coordinated motion trajectory data into a sequence of pathpoints. Each pathpoint contains location information (such as latitude and longitude or coordinates) and a timestamp, indicating the location the robot dog should reach at a specific time. Pathpoint information is extracted from the parsed data. The pathpoint information, including location information and timestamps, ensures the integrity and accuracy of the pathpoints. The extracted pathpoints are arranged in timestamp order to construct a pathpoint sequence. The pathpoint sequence is an ordered list of pathpoints representing the robot dog's patrol path. The constructed pathpoint sequence is verified to ensure the integrity and correctness of the pathpoints. Verification includes ensuring the reasonableness of the pathpoint location information and the continuity of the timestamps. By parsing the collision-free coordinated motion trajectory data into a pathpoint sequence containing location information and timestamps, the data is structured and operational. This process not only converts the complex motion trajectory data into a form that can be directly processed by the program but also ensures the integrity and correctness of the pathpoint sequence. The construction of the pathpoint sequence provides basic data for subsequent motion control instruction mapping, improving the system's operational efficiency and reliability. In addition, the structured form of the waypoint sequence facilitates subsequent storage and transmission, reducing the complexity of data processing.
[0113] Specifically, a motion control instruction mapping algorithm is used to convert the path point sequence into action-specifying codes for the robot dog to execute.
[0114] It should be noted that after obtaining the parsed pathpoint sequence, motion control command mapping is prepared and the parameters of the motion control command mapping algorithm, including the motion control model and mapping rules, are initialized. The motion control model is set based on the robot's kinematic characteristics, and the mapping rules are defined based on the pathpoint sequence and the robot's control commands. Based on the pathpoint sequence, the motion control command mapping algorithm is used to convert each pathpoint into an action-specific code for the robot. During the mapping process, the pathpoint's position information and timestamp are taken into account to generate corresponding motion control commands, such as speed, acceleration, and steering angle. The generated action-specific codes are then integrated to form a complete motion control command sequence. The command sequence is arranged in chronological order to ensure that the robot can perform its patrol mission according to the pathpoint sequence. The generated action-specific codes are then verified to ensure the integrity and correctness of the commands. Verification includes whether the commands conform to the robot's kinematic characteristics and whether they can achieve the motion of the pathpoint sequence. By converting the pathpoint sequence into action-specific codes for the robot to execute using the motion control command mapping algorithm, the transition from path planning to actual execution is achieved. This process not only converts the pathpoint sequence into motion control commands executable by the robot but also ensures the integrity and correctness of the commands. The use of a motion control command mapping algorithm improves the efficiency and accuracy of command generation, ensuring the robot dog can execute patrol tasks along the predetermined path. Furthermore, the generated action-specific code provides the basis for subsequent command transmission, enhancing the system's intelligence and management efficiency.
[0115] Specifically, the action-specifying code is encapsulated according to the communication protocol of the robot dog and sent to the controller of the robot dog using a wireless remote control device.
[0116] It should be noted that the generated action-specific code is obtained and prepared for encapsulation and transmission. The action-specific code is encapsulated according to the robot's communication protocol. During the encapsulation process, the action-specific code is converted into a data packet that complies with the communication protocol, including header information, data payload, and checksum. This encapsulated data packet is then sent to the robot's controller using a wireless remote control device. The wireless remote control device transmits data via a wireless network (such as Wi-Fi, Bluetooth, or Zigbee) to ensure real-time and reliable data. The robot's controller receives the encapsulated data packet, decapsulates and parses it, and extracts the action-specific code. Based on the action-specific code, the robot controls the robot to execute the corresponding motion command, such as move, turn, or stop. The robot's controller then feeds back the execution results to the sender to ensure correct command execution. The feedback information includes execution status, current position, and timestamp. By encapsulating the action-specific code according to the robot's communication protocol and sending it to the robot's controller using a wireless remote control device, closed-loop control is achieved from command generation to execution. This process not only ensures real-time transmission and correct execution of commands, but also improves system reliability and stability. The encapsulation of the communication protocol ensures data integrity and security, and the use of wireless remote control devices improves the flexibility and efficiency of command transmission. Furthermore, the feedback mechanism for execution results ensures the correct execution of commands, improves the system's intelligence level and management efficiency, and provides efficient technical support for the management and maintenance of urban sidewalks.
[0117] This embodiment also provides a pedestrian patrol command and dispatch system based on a robot dog, including:
[0118] The data processing module collects and preprocesses multi-source heterogeneous data from sidewalk infrastructure, the environment, and the robot dog to obtain a standardized data stream;
[0119] The modeling module constructs a cross-modal spatiotemporal graph parameter model and inputs multi-source heterogeneous data into the cross-modal spatiotemporal graph parameter model to obtain an initialized graph;
[0120] The scoring module performs online graph reasoning on the initialized graph to obtain the priority scores of each node of the sidewalk infrastructure, environment, and robot dog;
[0121] The instruction generation module integrates the priority scores of each node with multi-source heterogeneous data and generates decision instruction data through a decision tree algorithm;
[0122] The path detection module performs multi-constraint path optimization and conflict detection on the decision instruction data to obtain collision-free coordinated motion trajectory data;
[0123] The control module translates the collision-free coordinated motion trajectory data into action-specifying codes through the motion control instruction mapping algorithm, and transmits them to the robotic dog for sidewalk patrol command.
[0124] This embodiment also provides a computer device, which is suitable for implementing a pedestrian patrol command and dispatch method based on a robot dog, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the pedestrian patrol command and dispatch method based on a robot dog as proposed in the above embodiment.
[0125] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.
[0126] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the pedestrian patrol command and dispatch method based on a robot dog as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk.
[0127] In summary, the present invention achieves this by: collecting and preprocessing multi-source heterogeneous data, constructing a cross-modal spatiotemporal graph parameter model, combining graph neural networks and reinforcement learning algorithms to perform online graph reasoning calculations, and accurately evaluating the priority scores of each node. At the same time, a decision tree algorithm is used to generate decision instruction data, and multi-constraint path optimization and conflict detection are performed to ensure that the motion trajectory is collision-free. Finally, the optimized trajectory data is converted into action-specifying codes through a motion control instruction mapping algorithm to achieve efficient command and dispatch of the robot dog. This series of innovative measures has significantly improved the intelligence level and management efficiency of robot dogs in sidewalk patrols, effectively solved the shortcomings of existing technologies, and provided more efficient and accurate technical support for the management and maintenance of urban sidewalks.
[0128] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
Claims
1. A method for implementing pedestrian patrol command and dispatch based on a robot dog, characterized by: include, By collecting and preprocessing multi-source heterogeneous data from sidewalk infrastructure, environment, and robot dogs, a standardized data stream is obtained; Construct a cross-modal spatiotemporal graph parameter model, and input multi-source heterogeneous data into the cross-modal spatiotemporal graph parameter model to obtain an initialized graph; Perform online graph reasoning on the initialized graph to obtain the priority scores of each node of the sidewalk infrastructure, environment, and robot dog; The priority score of each node is integrated with multi-source heterogeneous data, and decision instruction data is generated through the decision tree algorithm; Perform multi-constraint path optimization and conflict detection on decision instruction data to obtain collision-free coordinated motion trajectory data; The collision-free coordinated motion trajectory data is translated into action-specifying codes through a motion control instruction mapping algorithm and transmitted to the robotic dog for sidewalk patrol command.
2. The method for implementing pedestrian patrol command and dispatch based on a robot dog as claimed in claim 1, characterized in that: The method collects multi-source heterogeneous data of sidewalk infrastructure, environment and robot dogs and pre-processes them to obtain standardized data streams. The specific steps are: The multi-source heterogeneous data includes pedestrian infrastructure distribution data, environmental parameters and robot dog motion status data; Classify and arrange the collected multi-source heterogeneous data and transmit them to the data processing center; The classified and arranged multi-source heterogeneous data are cleaned using the Bayesian algorithm and filled with missing values using the KNN algorithm to obtain complete multi-source heterogeneous data; The complete multi-source heterogeneous data is uniformly mapped to a fixed interval through the data normalization method to obtain a standardized data stream.
3. The method for implementing pedestrian patrol command and dispatch based on a robot dog as claimed in claim 2, characterized in that: The cross-modal spatiotemporal graph parameter model is constructed, and multi-source heterogeneous data are input into the cross-modal spatiotemporal graph parameter model to obtain an initialized graph. The specific steps are: Generate adversarial networks to mine and enhance standardized data streams to obtain feature data; The characteristic data includes sidewalk infrastructure distribution data, environmental parameters and robot dog motion status data; A cross-modal spatiotemporal graph parameter model is constructed based on the feature data, and the feature data is substituted into the calculation formula of the cross-modal spatiotemporal graph parameter model to generate a node priority score; The model parameters of the sidewalk infrastructure, environment and robot dog are respectively introduced into the calculation formula of the cross-modal spatiotemporal graph parameter model to generate the initialization graph of the cross-modal spatiotemporal graph.
4. The method for implementing pedestrian patrol command and dispatch based on a robot dog as claimed in claim 3, characterized in that: The online graph reasoning calculation is performed on the initialized graph to obtain the priority scores of each node of the sidewalk infrastructure, environment and robot dog. The specific steps are: The graph neural network is used to calculate the spatiotemporal correlation characteristics of each node in the initial graph, including the sidewalk infrastructure, environment, and robot dog, and to generate preliminary calculation results for the priority score of each node; Based on the preliminary calculation results and combined with multi-source heterogeneous data, the priority score of each node is calculated by weighted summation; Based on the calculation results of the priority score of each node, the reinforcement learning algorithm is used to adjust the parameters α, β and γ in the cross-modal spatiotemporal graph parameter model; According to the adjusted α, β, and γ in the cross-modal spatiotemporal graph parameter model, the priority score of each node is recalculated to generate the priority scores of each node of the sidewalk infrastructure, environment, and robot dog.
5. The method for implementing pedestrian patrol command and dispatch based on a robot dog as claimed in claim 4, characterized in that: The specific steps of integrating the priority score of each node with multi-source heterogeneous data and generating decision instruction data through the decision tree algorithm are as follows: Add node priority scores to multi-source heterogeneous data to form a fused dataset; Perform feature selection on the fused dataset to filter out characteristic parameters of sidewalk infrastructure distribution data, environmental parameters, and the robot dog's motion state; Using the filtered sidewalk infrastructure distribution data, environmental parameters, and characteristic parameters of the robot dog's motion state as input, a decision tree algorithm is used to calculate the decision instruction data. Convert the obtained decision instruction data into JSON format.
6. The method for implementing pedestrian patrol command and dispatch based on a robot dog as claimed in claim 5, characterized in that: The decision instruction data is subjected to multi-constraint path optimization and conflict detection to obtain collision-free coordinated motion trajectory data. The specific steps are: Parse the decision instruction data from JSON format into a sequence of waypoints containing location information and timestamps; The spatial segmentation technology is used to divide the path point sequence into multiple sub-path segments, and conflict detection is performed on each sub-path segment; Use the A algorithm to perform multi-constraint path optimization on each sub-path segment, evaluate the quality of the path through the heuristic function, and optimize each sub-path segment; Convert the optimized sub-path point sequence into collision-free coordinated motion trajectory data.
7. The method for implementing pedestrian patrol command and dispatch based on a robot dog as claimed in claim 6, characterized in that: The collision-free coordinated motion trajectory data is translated into action-specifying codes through the motion control instruction mapping algorithm and transmitted to the robot dog for sidewalk patrol command. The specific steps are: Parse the collision-free coordinated motion trajectory data into a sequence of waypoints containing position information and timestamps; Use motion control command mapping algorithm to convert the waypoint sequence into action-specifying code for the robot dog to execute; The action-specifying code is encapsulated according to the communication protocol of the robot dog and sent to the controller of the robot dog using a wireless remote control device.
8. A pedestrian patrol command and dispatch system based on a robot dog, based on the pedestrian patrol command and dispatch method based on a robot dog according to any one of claims 1 to 7, characterized in that: include, The data processing module collects and preprocesses multi-source heterogeneous data from sidewalk infrastructure, the environment, and the robot dog to obtain a standardized data stream; The modeling module constructs a cross-modal spatiotemporal graph parameter model and inputs multi-source heterogeneous data into the cross-modal spatiotemporal graph parameter model to obtain an initialized graph; The scoring module performs online graph reasoning on the initialized graph to obtain the priority scores of each node of the sidewalk infrastructure, environment, and robot dog; The instruction generation module integrates the priority scores of each node with multi-source heterogeneous data and generates decision instruction data through a decision tree algorithm; The path detection module performs multi-constraint path optimization and conflict detection on the decision instruction data to obtain collision-free coordinated motion trajectory data; The control module translates the collision-free coordinated motion trajectory data into action-specifying codes through the motion control instruction mapping algorithm, and transmits them to the robotic dog for sidewalk patrol command.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for implementing pedestrian patrol command and dispatch based on a robot dog as described in any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for implementing pedestrian patrol command and dispatch based on a robot dog as described in any one of claims 1 to 7 are implemented.
Citation Information
Patent Citations
Multi-robot cooperation path global planning method and system
CN116501041A
Method and system for commanding and dispatching sidewalk patrol based on robot dog
CN118192469A