Driving method and system of quadruped robot based on neural network model

Abstract

The invention provides a driving method and system of a quadruped robot based on a neural network model, and relates to the technical field of robot driving, and the method comprises the steps: determining a plurality of preliminary monitoring points based on a panorama of a poison gas leakage site; determining a plurality of supplementary monitoring points based on the laser radar information of the continuous time points of each preliminary monitoring point and the panorama of the poison gas leakage site; generating a toxic gas diffusion subsequent simulation video based on the laser radar information of the continuous time point of each preliminary monitoring point, the laser radar information of the continuous time point of each supplementary monitoring point and the panorama of the toxic gas leakage site; determining a plurality of diffusion dangerous points and a plurality of diffusion safety points based on the poison gas diffusion follow-up simulation video and the panorama of the poison gas leakage site; the target safety rescue path is determined based on the multiple diffusion safety points and the multiple diffusion dangerous points. The method can quickly and accurately determine the optimal rescue path of the poison gas leakage site.

Description

Technical Field

[0001] The present invention relates to the technical field of robot driving, and particularly relates to a driving method and system for a quadruped robot based on a neural network model. Background Art

[0002] With the acceleration of the industrialization and urbanization processes, toxic gas leakage accidents such as chemical plant leaks and underground pipeline ruptures occur frequently, seriously threatening personnel safety and environmental health. At the site of a toxic gas leakage accident, it is crucial to quickly and accurately assess the trend of toxic gas diffusion and plan a safe rescue path. Traditional toxic gas monitoring methods mainly rely on fixed sensors or manual monitoring, but there are problems such as slow response speed, insufficient prediction accuracy, and the dependence of rescue path planning on manual experience. The deployment range of fixed sensors is limited and it is difficult to cover the entire leakage area. Manual monitoring is inefficient, unable to obtain dynamic diffusion data in real time, and difficult to adapt to complex terrains and dynamically changing airflow conditions, resulting in large prediction errors. Moreover, rescue personnel usually plan paths based on limited information and may enter dangerous areas by mistake, thus increasing the rescue risk.

[0003] Therefore, how to quickly and accurately determine the optimal rescue path for a toxic gas leakage site is an urgent problem to be solved currently. Summary of the Invention

[0004] The main technical problem to be solved by the present invention is how to quickly and accurately determine the optimal rescue path for a toxic gas leakage site.

[0005] According to a first aspect, the present invention provides a driving method for a quadruped robot based on a neural network model, including: obtaining a panoramic view of a toxic gas leakage site; determining a plurality of preliminary monitoring points based on the panoramic view of the toxic gas leakage site; controlling the quadruped robot to each preliminary monitoring point and obtaining lidar information at consecutive time points for each preliminary monitoring point; determining a plurality of supplementary monitoring points based on the lidar information at consecutive time points for each preliminary monitoring point and the panoramic view of the toxic gas leakage site; controlling the quadruped robot to each supplementary monitoring point and obtaining lidar information at consecutive time points for each supplementary monitoring point; generating a subsequent simulation video of toxic gas diffusion based on the lidar information at consecutive time points for each preliminary monitoring point, the lidar information at consecutive time points for each supplementary monitoring point, and the panoramic view of the toxic gas leakage site; determining a plurality of diffusion danger points and a plurality of diffusion safety points based on the subsequent simulation video of toxic gas diffusion and the panoramic view of the toxic gas leakage site; and determining a target safe rescue path based on the plurality of diffusion safety points and the plurality of diffusion danger points.

[0006] In a possible implementation manner, determining a safe rescue path based on the multiple diffusion safety points and the multiple diffusion danger points includes: generating multiple preliminary safe rescue paths based on the multiple diffusion safety points and the multiple diffusion danger points; determining information of each preliminary safe rescue path based on the subsequent simulation video of the poisonous gas diffusion and the multiple preliminary safe rescue paths; constructing a graph structure, where the graph structure includes multiple nodes and multiple edges between the nodes. Each node represents a preliminary safe rescue path, the node attribute of each node is the information of the preliminary safe rescue path, and the edge between the nodes is the similarity between the information of two preliminary safe rescue paths; processing the graph structure based on a graph neural network to determine the target safe rescue path.

[0007] In a possible implementation manner, determining multiple preliminary monitoring points based on the panoramic view of the poisonous gas leakage site includes: using a convolutional neural network to determine multiple preliminary monitoring points based on the panoramic view of the poisonous gas leakage site.

[0008] In a possible implementation manner, the input of the graph neural network is the graph structure, and the output of the graph neural network is the target safe rescue path.

[0009] According to a second aspect, the present invention provides a driving system for a quadruped robot based on a neural network model, including: an acquisition module for acquiring a panoramic view of a poisonous gas leakage site; a preliminary monitoring point determination module for determining multiple preliminary monitoring points based on the panoramic view of the poisonous gas leakage site; a first control module for controlling the quadruped robot to each preliminary monitoring point and acquiring lidar information at consecutive time points of each preliminary monitoring point; a supplementary monitoring point determination module for determining multiple supplementary monitoring points based on the lidar information at consecutive time points of each preliminary monitoring point and the panoramic view of the poisonous gas leakage site; a second control module for controlling the quadruped robot to each supplementary monitoring point and acquiring lidar information at consecutive time points of each supplementary monitoring point; a generation module for generating a subsequent simulation video of the poisonous gas diffusion based on the lidar information at consecutive time points of each preliminary monitoring point, the lidar information at consecutive time points of each supplementary monitoring point, and the panoramic view of the poisonous gas leakage site; a dangerous area determination module for determining multiple diffusion danger points and multiple diffusion safety points based on the subsequent simulation video of the poisonous gas diffusion and the panoramic view of the poisonous gas leakage site; a path planning module for determining a target safe rescue path based on the multiple diffusion safety points and the multiple diffusion danger points.

[0010] In a possible implementation, the path planning module is further configured to: generate multiple preliminary safe rescue paths based on the multiple diffusion safety points and the multiple diffusion danger points; determine the information of each preliminary safe rescue path based on the subsequent simulation video of the gas diffusion and the multiple preliminary safe rescue paths; construct a graph structure, which includes multiple nodes and multiple edges between the nodes, each node represents a preliminary safe rescue path, the node attribute of each node is the information of the preliminary safe rescue path, and the edge between the nodes is the similarity between the information of two preliminary safe rescue paths; and determine the target safe rescue path by processing the graph structure based on a graph neural network.

[0011] In a possible implementation, the preliminary monitoring point determination module is specifically configured to: determine multiple preliminary monitoring points based on the panoramic view of the gas leakage site using a convolutional neural network.

[0012] In a possible implementation, the input of the graph neural network is the graph structure, and the output of the graph neural network is the target safe rescue path.

[0013] A driving method and system for a quadruped robot based on a neural network model provided by the present invention. The method includes obtaining a panoramic view of a gas leakage site; determining multiple preliminary monitoring points based on the panoramic view of the gas leakage site; controlling the quadruped robot to each preliminary monitoring point and obtaining lidar information at consecutive time points of each preliminary monitoring point; determining multiple supplementary monitoring points based on the lidar information at consecutive time points of each preliminary monitoring point and the panoramic view of the gas leakage site; controlling the quadruped robot to each supplementary monitoring point and obtaining lidar information at consecutive time points of each supplementary monitoring point; generating a subsequent simulation video of gas diffusion based on the lidar information at consecutive time points of each preliminary monitoring point, the lidar information at consecutive time points of each supplementary monitoring point, and the panoramic view of the gas leakage site; determining multiple diffusion danger points and multiple diffusion safety points based on the subsequent simulation video of gas diffusion and the panoramic view of the gas leakage site; and determining a target safe rescue path based on the multiple diffusion safety points and the multiple diffusion danger points. This method can quickly and accurately determine the optimal rescue path for the gas leakage site. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic flowchart of a driving method for a quadruped robot based on a neural network model provided by an embodiment of the present invention;

[0015] Figure 2 It is a schematic flowchart of a process for determining a target safe rescue path provided by an embodiment of the present invention;

[0016] Figure 3Schematic diagram of a driving system for a quadruped robot based on a neural network model provided by an embodiment of the present invention. Detailed implementation manners

[0017] In an embodiment of the present invention, there is provided a Figure 1 driving method for a quadruped robot based on a neural network model as shown. The driving method for the quadruped robot based on the neural network model includes steps S1 to S8:

[0018] Step S1, obtaining a panoramic view of the gas leakage site.

[0019] The gas leakage site is a physical area where toxic or harmful gases leak, such as industrial facilities like chemical plants and warehouses, laboratories or other enclosed and semi-enclosed environments. Gas leakage includes, for example, leakage of dangerous gases such as ammonia and chlorine due to the rupture of a gas storage tank in a chemical plant, and diffusion of toxic gases such as hydrogen sulfide and carbon monoxide caused by a malfunction of the ventilation system in a laboratory.

[0020] The panoramic view of the gas leakage site is a high-resolution image covering the complete visual information of the gas leakage site area, obtained by a multi-camera array image acquisition device.

[0021] Step S2, determining a plurality of preliminary monitoring points based on the panoramic view of the gas leakage site.

[0022] In some embodiments, a plurality of preliminary monitoring points can be determined based on the panoramic view of the gas leakage site using a convolutional neural network. The input of the convolutional neural network is the panoramic view of the gas leakage site, and the output of the convolutional neural network is a plurality of preliminary monitoring points.

[0023] A convolutional neural network (CNN) is a deep learning model that is good at processing data with a grid structure such as videos or images. Its core idea is to automatically extract the spatial hierarchical features of the input data through a local receptive field and a weight sharing mechanism. A convolutional neural network consists of a convolutional layer, a pooling layer, and a fully connected layer. Among them, the convolutional layer captures local features through a sliding filter (convolution kernel), the pooling layer can reduce the data dimension and enhance translation invariance, and the fully connected layer can finally complete classification or regression tasks.

[0024] The preliminary monitoring points are key positions in the gas leakage site output by the convolutional neural network, and the preliminary monitoring points can be used to deploy a quadruped robot for environmental data collection.

[0025] The panoramic view of the gas leakage site contains rich environmental information, such as the location of the leakage source, the distribution of obstacles, the direction of ventilation openings, and the visual characteristics of gas diffusion. By analyzing the panoramic view, key areas near the leakage source, key nodes of the ventilation system, and the distribution of obstacles that may affect the gas diffusion path can be identified. These locations have high monitoring value. For example, the area around the leakage source can reflect the initial diffusion state, the vicinity of the ventilation opening can capture the air flow direction, and open areas are helpful for evaluating the overall diffusion trend. Selecting preliminary monitoring points from these key locations can ensure the representativeness of the data collected subsequently.

[0026] Convolutional neural networks can extract multi-level spatial features from the panoramic view of the gas leakage site. For example, shallow convolutional kernels can identify basic geometric features such as the edges of obstacles and the shapes of ventilation openings in the site, while deep networks can combine context information to distinguish semantic units such as open spaces, narrow channels, and enclosed areas. Through semantic segmentation of the panoramic view, convolutional neural networks can locate key spatial elements such as the physical location of the leakage source, the spatial structure of the obstacle distribution, and the orientation of the ventilation openings. These spatial features are highly correlated with the potential paths of gas diffusion. For example, the orientation of the ventilation opening determines the initial diffusion direction of the gas driven by the air flow, and the obstacle distribution may form eddies resulting in gas retention. By learning the correlation rules between the spatial layout and diffusion patterns in historical scenarios, convolutional neural networks can automatically identify positions of key value for gas diffusion monitoring, such as directly below the leakage source, upstream and downstream of the ventilation opening, and gaps between obstacles, and then determine the preliminary monitoring points where quadruped robot monitoring equipment needs to be preferentially deployed.

[0027] Step S3: Control the quadruped robot to each preliminary monitoring point and obtain the lidar information at consecutive time points for each preliminary monitoring point.

[0028] The lidar information is obtained in the following way: When the quadruped robot arrives at the preset monitoring point, it emits a laser beam with a specific wavelength through the gas monitoring lidar carried by it, and then receives the reflected signal to analyze the three-dimensional spatial distribution information of the gas. The lidar information includes information such as the gas concentration gradient distribution and the geometric characteristics of the gas diffusion boundary.

[0029] The lidar information at consecutive time points is that the quadruped robot maintains a stable pose and continuously collects lidar information data at a fixed sampling rate (such as 5Hz) to form a time series monitoring record. Through this lidar information at consecutive time points, the dynamic change process of the gas concentration and the temporal evolution of the diffusion direction can be captured.

[0030] Step S4: Determine multiple supplementary monitoring points based on the lidar information at consecutive time points for each preliminary monitoring point and the panoramic view of the gas leakage site.

[0031] In some embodiments, multiple supplementary monitoring points can be determined based on the lidar information at consecutive time points of each preliminary monitoring point and the panoramic view of the gas leakage site using a supplementary point analysis model, where the supplementary point analysis model is a convolutional neural network. The input of the supplementary point analysis model is the lidar information at consecutive time points of each preliminary monitoring point and the panoramic view of the gas leakage site, and the output of the supplementary point analysis model is multiple supplementary monitoring points.

[0032] Supplementary monitoring points are key position points optimized and determined by the supplementary point analysis model on the basis of preliminary monitoring points for blind spots in the coverage of the monitoring network, data-weak areas, or high-risk mutation areas. Supplementary monitoring points include, but are not limited to, eddy current areas behind obstacles, branch nodes of the ventilation system, and abnormal areas where the concentration gradient change rate exceeds a preset threshold. For example, when the diffusion speed in a certain direction is significantly higher than expected, supplementary monitoring points are added on the path in that direction to capture the mutation trend in real time.

[0033] The convolutional neural network has unique spatio-temporal feature analysis and dynamic pattern recognition capabilities. Through a three-dimensional convolutional structure, the convolutional neural network can synchronously process the static spatial features of the panoramic view and the dynamic temporal lidar information, so as to be able to extract information such as sudden changes in the diffusion direction and changes in the concentration gradient during the gas diffusion process. At the same time, the deep semantic understanding ability of the convolutional neural network enables it to identify blind spots in the distribution of preliminary monitoring points or potential high-risk areas, and then combine the influence law of the ventilation system on the air flow to generate multiple supplementary monitoring points.

[0034] In some embodiments, the supplementary point analysis model includes a diffusion analysis layer, an anomaly monitoring layer, and a point location planning layer. The diffusion analysis layer, the anomaly monitoring layer, and the point location planning layer all include convolutional neural network structures. The input of the diffusion analysis layer is the lidar information at consecutive time points of each preliminary monitoring point, and the output of the diffusion analysis layer is a sequence of gas temporal concentration fields and the gas regional diffusion trend. The input of the anomaly monitoring layer is the sequence of gas temporal concentration fields, the gas regional diffusion trend, and the panoramic view of the gas leakage site, and the output of the anomaly monitoring layer is the marked abnormal high-concentration areas and the mutation points of the diffusion path. The input of the point location planning layer is the marked abnormal high-concentration areas, the mutation points of the diffusion path, and the panoramic view of the gas leakage site, and the output of the point location planning layer is multiple supplementary monitoring points.

[0035] The sequence of gas temporal concentration fields is a sequence of gas concentration distribution images indexed by time. Each frame of the image corresponds to a moment and can fully display the gas concentration distribution within the entire gas leakage site area and its change over time. The concentration values are visualized through color gradients and numerical annotations.

[0036] The gas area diffusion trend includes the gas diffusion direction and velocity distribution within the gas leakage site area.

[0037] The abnormally high concentration area marker is the spatial area where the gas concentration exceeds the safety threshold and the duration reaches or exceeds the set standard.

[0038] The diffusion path mutation point is the spatial position where the gas diffusion direction or velocity changes significantly within a short period of time.

[0039] Different layers can be responsible for information processing in different dimensions. For example, the diffusion analysis layer focuses on analyzing the concentration distribution and temporal diffusion characteristics in the lidar data. The anomaly monitoring layer combines the panoramic view to locate high-risk areas and diffusion mutation points. The point location planning layer can generate specific supplementary monitoring point plans based on the results of the former two. Through hierarchical processing, each module performs its own functions to avoid the inefficiency caused by the mixed processing of complex data, which not only improves the analysis accuracy of gas diffusion characteristics but also can accurately deploy supplementary monitoring points, thus enhancing the integrity and effectiveness of monitoring.

[0040] Step S5, control the quadruped robot to each supplementary monitoring point and obtain the lidar information at consecutive time points for each supplementary monitoring point.

[0041] Regarding Step S5, reference can be made to Step S3, which will not be elaborated here.

[0042] Step S6, generate a subsequent simulation video of gas diffusion based on the lidar information at consecutive time points for each preliminary monitoring point, the lidar information at consecutive time points for each supplementary monitoring point, and the panoramic view of the gas leakage site.

[0043] In some embodiments, a subsequent simulation video of gas diffusion can be generated based on the lidar information at consecutive time points for each preliminary monitoring point, the lidar information at consecutive time points for each supplementary monitoring point, and the panoramic view of the gas leakage site using a simulation generation model. The simulation generation model is a generative adversarial network. The input of the simulation generation model is the lidar information at consecutive time points for each preliminary monitoring point, the lidar information at consecutive time points for each supplementary monitoring point, and the panoramic view of the gas leakage site, and the output of the simulation generation model is the subsequent simulation video of gas diffusion.

[0044] The Generative Adversarial Network (GAN) is a deep learning model composed of a Generator and a Discriminator. The Generator is responsible for learning the distribution law from the input data and generating realistic new data, while the Discriminator attempts to distinguish between the generated data and the real data. The two are continuously optimized through adversarial training. The Generative Adversarial Network can generate high-fidelity and diverse outputs, and is applicable to the modeling of complex dynamic processes such as the spatio-temporal evolution of gas diffusion.

[0045] The subsequent simulation video of gas diffusion is a dynamic simulation video generated by a simulation generation model to simulate the diffusion path, concentration change, and evolution of dangerous areas of gas in the future.

[0046] The Generator of the Generative Adversarial Network can extract features from the input data distribution and synthesize new samples based on these features. First, the lidar information of consecutive time points of the preliminary monitoring points and supplementary monitoring points is fused with the panoramic view of the gas leakage site and input into the Generator. The Generator can learn the gas diffusion law through a spatio-temporal attention mechanism and add the Fick diffusion equation constraint term "∇²C = ∂C / ∂t" and the mass conservation regularization term to its loss function to ensure that the evolution of the concentration field conforms to physical laws. The Discriminator can compare the statistical distribution differences between the generated video frames and the real monitoring data, including the single-frame concentration field distribution and the change gradient between adjacent frames. Through adversarial games, the prediction ability of the Generator for the diffusion behavior in the unmonitored area can be improved, so as to ensure that the model can finally generate a subsequent simulation video of gas diffusion that meets physical accuracy.

[0047] Step S7, determining a plurality of diffusion danger points and a plurality of diffusion safety points based on the subsequent simulation video of gas diffusion and the panoramic view of the gas leakage site.

[0048] In some embodiments, a plurality of diffusion danger points and a plurality of diffusion safety points can be determined through a Gated Recurrent Unit based on the subsequent simulation video of gas diffusion and the panoramic view of the gas leakage site. The input of the Gated Recurrent Unit is the subsequent simulation video of gas diffusion and the panoramic view of the gas leakage site, and the output of the Gated Recurrent Unit is a plurality of diffusion danger points and a plurality of diffusion safety points.

[0049] The Gated Recurrent Unit (GRU) is a variant of the Recurrent Neural Network, which can be used to process sequence data and temporal information and is good at capturing dependencies in long sequences. The Gated Recurrent Unit includes three components: a memory unit, an update gate, and a reset gate.

[0050] Multiple diffusion risk points are spatial locations where the toxic gas concentration output by the gated recurrent unit exceeds the safety threshold and lasts for a relatively long time, such as the center of a high-concentration gas mass, low-wind-speed retention areas on the diffusion path, etc.

[0051] Diffusion safety points are locations where the toxic gas concentration output by the gated recurrent unit is always lower than the safety threshold and has a certain environmental safety, such as high upwind locations, enclosed areas with good airtightness, upstream areas of the ventilation system, etc.

[0052] The gated recurrent unit (GRU) has the ability of temporal dynamic analysis and spatial semantic fusion. The concentration field change sequence in the subsequent simulation video of toxic gas diffusion can be captured by the gated recurrent unit through the gating mechanism to identify the time-dependent relationship, so as to identify the areas where the concentration continuously exceeds the limit. The panoramic view can extract spatial features such as obstacles and ventilation openings. The gated recurrent unit can combine dynamic temporal features with static spatial features to analyze the impact of the environment on toxic gas diffusion, so as to locate high-risk areas and low-risk areas, and then determine multiple diffusion risk points and multiple diffusion safety points.

[0053] Step S8, determine the target safe rescue path based on the multiple diffusion safety points and the multiple diffusion risk points.

[0054] In some embodiments, it can be through Figure 2 the following process to determine the target safe rescue path, Figure 2 which is a schematic flow diagram of a process for determining the target safe rescue path provided by an embodiment of the present invention. The determination of the target safe rescue path includes steps S21 to S24:

[0055] Step S21, generate multiple preliminary safe rescue paths based on the multiple diffusion safety points and the multiple diffusion risk points.

[0056] In some embodiments, multiple preliminary safe rescue paths can be generated by a variational autoencoder based on the multiple diffusion safety points and the multiple diffusion risk points. The input of the variational autoencoder is the multiple diffusion safety points and the multiple diffusion risk points, and the output of the variational autoencoder is multiple preliminary safe rescue paths.

[0057] The variational autoencoder (VAE) is a generative deep learning model. The variational autoencoder compresses the input data into probability distribution parameters in a low-dimensional latent space through an encoder, and the decoder can sample from this distribution to generate new data. Its core is to use variational inference to balance the reconstruction accuracy and the regularization of the latent distribution to achieve data generation and feature decoupling.

[0058] The preliminary safe rescue path is a candidate passage route generated by a variational autoencoder based on safe points and dangerous points, which consists of a series of continuous coordinate points. The preliminary safe rescue path satisfies the connectivity from the starting point to multiple spreading safe points in the middle and then to the ending point, and avoids spreading dangerous points throughout the whole process.

[0059] Through its encoder, the variational autoencoder can encode data such as safe point coordinates and dangerous point masks into low-dimensional latent vectors. These latent vectors imply path features such as detour tendency, etc. The decoder samples in the latent space to generate a diverse and safe-rule-compliant path sequence. The probabilistic generation mechanism of the variational autoencoder can effectively cope with the uncertainties in the rescue scenario, thus outputting multiple preliminary safe rescue paths covering different risk preferences.

[0060] Step S22, determine the information of each preliminary safe rescue path based on the subsequent simulation video of the poisonous gas diffusion and the multiple preliminary safe rescue paths.

[0061] In some embodiments, the information of each preliminary safe rescue path can be determined through a path analysis model based on the subsequent simulation video of the poisonous gas diffusion and the multiple preliminary safe rescue paths. The path analysis model is a Transformer model. The input of the path analysis model is the subsequent simulation video of the poisonous gas diffusion and the multiple preliminary safe rescue paths, and the output of the path analysis model is the information of each preliminary safe rescue path.

[0062] The Transformer model consists of two parts: an encoder and a decoder. In the Transformer model structure, the encoder is responsible for performing representation learning operations on the input sequence, and integrates a self-attention mechanism and a feed-forward neural network (Feed-Forward Network) inside. Different from the encoder, the decoder adds a multi-head attention mechanism on the basis of inheriting the encoder architecture, which can be specifically used to parse the output content of the encoder and generate the corresponding target sequence. With such an architecture design, the Transformer model can accurately mine the information of each preliminary safe rescue path.

[0063] The information of each preliminary safe rescue path is a quantitative description of the risk and feasibility of each preliminary safe rescue path output by the path analysis model in a dynamic poisonous gas environment. The information of the preliminary safe rescue path includes risk indicators, time and space indicators, environmental adaptability indicators, etc.

[0064] The risk indicator is used to measure the poisoning risk of rescue personnel on the path. The influencing factors of the risk indicator include the cumulative exposure concentration and the proportion of high-risk periods.

[0065] The cumulative exposure concentration is the integral value of the concentrations at each point along the path over time. For example, if the cumulative exposure amount of a path reaches 500 ppm·min and far exceeds the safety threshold, it means that the poisoning risk of this path is high.

[0066] The proportion of high-risk time periods reflects the time ratio of the path passing through the area where the poisonous gas exceeds the standard. The higher the proportion, the longer the time exposed to the dangerous environment during the operation of this path, and the greater the risk.

[0067] The time and space indicators include the estimated travel time and the path length to reflect the travel efficiency.

[0068] The environmental adaptability indicators include the number of alternative safe points beside the path and the risk fluctuation range of the path under different wind speed and wind direction scenarios, which are used to evaluate the reliability of the path to cope with dynamic environmental changes.

[0069] The Transformer model can process the subsequent simulation video of poisonous gas diffusion containing time series information. Using the self-attention mechanism, it can effectively capture the correlations between video frames at different time points, as well as the spatial correspondence between the preliminary safe rescue path and the risk area in the video. The Transformer model can comprehensively consider the impact of the poisonous gas diffusion situation at each moment on the path, and then calculate indicators such as the risk level and travel feasibility of each path at different times, so as to determine the information of each preliminary safe rescue path.

[0070] The encoder of the Transformer model can use the self-attention mechanism to extract features from the input subsequent simulation video of poisonous gas diffusion and the preliminary safe rescue path, so as to capture the spatio-temporal relationships such as the change of poisonous gas concentration and diffusion direction at different moments in the video, as well as the spatial connections of each point in the path. The feed-forward neural network further transforms and enhances the features. Based on the output of the encoder, the decoder focuses on the key information through the multi-head attention mechanism, deeply fuses the video features and the path features, and finally generates the path information.

[0071] Step S23, construct a graph structure. The graph structure includes multiple nodes and multiple edges between the nodes. Each node represents a preliminary safe rescue path, the node attribute of each node is the preliminary safe rescue path information, and the edge between the nodes is the similarity between the information of two preliminary safe rescue paths.

[0072] The graph structure is composed of multiple nodes and multiple edges connecting these nodes. It can abstractly express complex relationships in the form of a topological network. Each node and edge carry specific attributes and meanings, and can be used to describe objects and their associated relationships.

[0073] The node corresponds to each preliminary safe rescue path, and its node attribute is the detailed information of this path.

[0074] The edges between nodes represent the degree of association between two paths. The edge is the similarity between the information of two preliminary safe rescue paths, which reflects the similarity or difference in aspects such as risk characteristics and spatial orientation of the paths.

[0075] In some embodiments, the similarity between the information of two preliminary safe rescue paths can be calculated through a deep neural network model.

[0076] Step S24, process the graph structure based on a graph neural network to determine the target safe rescue path.

[0077] A graph neural network (GNN) is a type of deep learning model designed for graph-structured data. Its core principle is to enable nodes to exchange information with each other through a message passing mechanism and gradually update their own feature representations. In this way, the graph neural network can effectively capture the high-order relationships between nodes in the graph structure, thereby mining the global features and potential laws of the data. The input of the graph neural network is the graph structure, and the output of the graph neural network is the target safe rescue path.

[0078] The target safe rescue path is the optimal rescue passage route generated by processing the graph structure through a graph neural network. This path is composed of continuous spatial coordinate points and needs to comprehensively meet the requirements of safety, efficiency, and environmental adaptability. Specifically, the target safe rescue path requires avoiding dangerous areas with excessive toxic gas concentration throughout the whole process to control the cumulative exposure concentration of personnel within the safety threshold, while shortening the path length or travel time as much as possible to improve rescue efficiency; in addition, the target safe rescue path also needs to have robustness to cope with the dynamic changes of toxic gas diffusion, such as having less risk fluctuation under different meteorological conditions and being close to alternative safe points to ensure emergency avoidance ability.

[0079] The information of each preliminary safe rescue path covers multi-dimensional attributes such as risk, time, and space. The design of the node attributes of the graph structure can integrate these complex information and provide rich context data for the graph neural network, enabling the model to comprehensively consider various factors for path evaluation.

[0080] The graph structure visually presents the association between preliminary safe rescue paths in the form of edges, which is convenient for the graph neural network to capture the potential connections between paths.

[0081] The path information attribute of each node is the core basis for the decision-making of the graph neural network. For example, the cumulative exposure concentration is used to evaluate the health threat of the path to the rescue personnel, and the higher the value, the greater the risk; the expected travel time directly affects the rescue efficiency, and the path with a shorter travel time has more advantages in emergency scenarios; the environmental adaptability index reflects the stability of the path in the dynamic gas diffusion environment, and the higher the index, the less affected the path is by environmental changes. These data can help the graph neural network accurately calculate the comprehensive score of each path, so as to screen out the most suitable target safe rescue path.

[0082] Based on the same inventive concept, Figure 3 The following is a schematic diagram of a driving system of a quadruped robot based on a neural network model provided by an embodiment of the present invention. The driving system of the quadruped robot based on the neural network model includes:

[0083] An acquisition module 31, configured to acquire a panoramic view of the gas leakage site;

[0084] A preliminary monitoring point determination module 32, configured to determine a plurality of preliminary monitoring points based on the panoramic view of the gas leakage site;

[0085] A first control module 33, configured to control the quadruped robot to each preliminary monitoring point and acquire lidar information at consecutive time points of each preliminary monitoring point;

[0086] A supplementary monitoring point determination module 34, configured to determine a plurality of supplementary monitoring points based on the lidar information at consecutive time points of each preliminary monitoring point and the panoramic view of the gas leakage site;

[0087] A second control module 35, configured to control the quadruped robot to each supplementary monitoring point and acquire lidar information at consecutive time points of each supplementary monitoring point;

[0088] A generation module 36, configured to generate a subsequent simulation video of gas diffusion based on the lidar information at consecutive time points of each preliminary monitoring point, the lidar information at consecutive time points of each supplementary monitoring point, and the panoramic view of the gas leakage site;

[0089] A dangerous area determination module 37, configured to determine a plurality of diffusion dangerous points and a plurality of diffusion safe points based on the subsequent simulation video of gas diffusion and the panoramic view of the gas leakage site;

[0090] A path planning module 38, configured to determine a target safe rescue path based on the plurality of diffusion safe points and the plurality of diffusion dangerous points.

[0091] Similarly, it should be noted that, in order to simplify the presentation disclosed in this specification and thus assist in the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of this specification, various features are sometimes grouped together into one embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the features required by the subject matter of this specification are more than those recited in the claims. In fact, the features of the embodiments are less than all the features of the individual embodiments disclosed above.

[0092] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, by way of example and not limitation, alternative configurations of the embodiments of this specification may be regarded as consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to those explicitly presented and described in this specification.

Claims

1. A driving method for a quadruped robot based on a neural network model, characterized in that, Including: Obtain a panoramic view of the gas leakage site; Determine multiple preliminary monitoring points based on the panoramic view of the gas leakage site; Control the quadruped robot to each preliminary monitoring point and obtain lidar information at consecutive time points for each preliminary monitoring point; Determine multiple supplementary monitoring points based on the lidar information at consecutive time points for each preliminary monitoring point and the panoramic view of the gas leakage site; Control the quadruped robot to each supplementary monitoring point and obtain lidar information at consecutive time points for each supplementary monitoring point; Generate a subsequent simulation video of gas diffusion based on the lidar information at consecutive time points for each preliminary monitoring point, the lidar information at consecutive time points for each supplementary monitoring point, and the panoramic view of the gas leakage site; Determine multiple diffusion danger points and multiple diffusion safety points based on the subsequent simulation video of gas diffusion and the panoramic view of the gas leakage site; Determine a target safe rescue path based on the multiple diffusion safety points and the multiple diffusion danger points.

2. The driving method of the quadruped robot based on the neural network model according to claim 1, characterized in that The determining of the safe rescue path based on the multiple diffusion safety points and the multiple diffusion danger points includes: Generate multiple preliminary safe rescue paths based on the multiple diffusion safety points and the multiple diffusion danger points; Determine the information of each preliminary safe rescue path based on the subsequent simulation video of gas diffusion and the multiple preliminary safe rescue paths; Construct a graph structure, which includes multiple nodes and multiple edges between the nodes. Each node represents a preliminary safe rescue path, the node attribute of each node is the information of the preliminary safe rescue path, and the edge between the nodes is the similarity between the information of two preliminary safe rescue paths; Process the graph structure based on a graph neural network to determine the target safe rescue path.

3. The driving method of the quadruped robot based on the neural network model according to claim 1, characterized in that The determining of multiple preliminary monitoring points based on the panoramic view of the gas leakage site includes: Use a convolutional neural network to determine multiple preliminary monitoring points based on the panoramic view of the gas leakage site.

4. The driving method of the quadruped robot based on the neural network model according to claim 2, wherein The input of the graph neural network is the graph structure, and the output of the graph neural network is the target safe rescue path.

5. A drive system for a quadruped robot based on a neural network model, characterized in that, Including: An acquisition module for obtaining a panoramic view of the gas leakage site; A preliminary monitoring point determination module for determining multiple preliminary monitoring points based on the panoramic view of the gas leakage site; A first control module for controlling the quadruped robot to each preliminary monitoring point and obtaining lidar information at consecutive time points for each preliminary monitoring point; A supplementary monitoring point determination module for determining multiple supplementary monitoring points based on the lidar information at consecutive time points for each preliminary monitoring point and the panoramic view of the gas leakage site; A second control module for controlling the quadruped robot to each supplementary monitoring point and obtaining lidar information at consecutive time points for each supplementary monitoring point; A generation module for generating a subsequent simulation video of gas diffusion based on the lidar information at consecutive time points for each preliminary monitoring point, the lidar information at consecutive time points for each supplementary monitoring point, and the panoramic view of the gas leakage site; A danger area determination module, configured to determine a plurality of diffusion danger points and a plurality of diffusion safety points based on the subsequent simulation video of the poisonous gas diffusion and the panoramic view of the poisonous gas leakage site; A path planning module, configured to determine a target safe rescue path based on the plurality of diffusion safety points and the plurality of diffusion danger points.

6. The drive system of the quadruped robot based on the neural network model according to claim 5, characterized in that, The path planning module is further configured to: Generate a plurality of preliminary safe rescue paths based on the plurality of diffusion safety points and the plurality of diffusion danger points; Determine the information of each preliminary safe rescue path based on the subsequent simulation video of the poisonous gas diffusion and the plurality of preliminary safe rescue paths; Construct a graph structure, the graph structure includes a plurality of nodes and a plurality of edges between the nodes, each node represents a preliminary safe rescue path, the node attribute of each node is the preliminary safe rescue path information, and the edge between the nodes is the similarity between the information of two preliminary safe rescue paths; Process the graph structure based on a graph neural network to determine the target safe rescue path.

7. The drive system of the quadruped robot based on the neural network model according to claim 5, characterized in that The preliminary monitoring point determination module is specifically configured to: Determine a plurality of preliminary monitoring points based on the panoramic view of the poisonous gas leakage site using a convolutional neural network.

8. The drive system of the quadruped robot based on the neural network model according to claim 6, characterized in that, The input of the graph neural network is the graph structure, and the output of the graph neural network is the target safe rescue path.

Images