Behavior simulation method of mobile agent applied to emergency rescue scene

Abstract

The embodiment of the invention discloses a behavior simulation method of a mobile agent applied to an emergency rescue scene. A specific embodiment of the method comprises the following steps: generating an initial simulation environment according to scene environment information and analogue simulation configuration information; according to the agent description information corresponding to the mobile agent, environment data local extraction is carried out in the initial simulation environment, and a real-time environment information set is obtained; generating environment state characteristics and agent state characteristics according to the real-time environment information set and the agent real-time state corresponding to the mobile agent; generating behavior planning information for the mobile intelligent agent according to the environment state characteristics and the intelligent agent state characteristics; and performing behavior simulation on the mobile intelligent agent according to the behavior planning information and the real-time orientation of the intelligent agent. According to the embodiment, the robustness and accuracy of behavior planning of the mobile robot in different emergency rescue scenes are improved.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, particularly the field of behavioral simulation, and specifically to a behavioral simulation method for mobile intelligent agents applied in emergency rescue scenarios. Background Technology

[0002] Due to the limitations of the rescue environment, there are areas in emergency rescue scenarios that rescuers cannot directly cover. Therefore, mobile robots are often needed for emergency rescue (such as material transportation and environmental reconnaissance). However, because the rescue environment is often harsh, it is difficult to effectively plan the robot's behavior in real-world conditions. Summary of the Invention

[0003] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0004] Some embodiments of this disclosure propose a behavior simulation method for mobile intelligent agents applied in emergency rescue scenarios to solve the technical problems mentioned in the background section above.

[0005] In a first aspect, some embodiments of this disclosure provide a behavior simulation method for a mobile intelligent agent applied in an emergency rescue scenario. The method includes: generating an initial simulation environment based on scenario environment information and simulation configuration information, wherein the scenario environment information represents the regional environment information corresponding to the emergency rescue scenario, and the initial simulation environment contains a mobile intelligent agent whose position has been initialized and whose behavior simulation is to be performed; extracting environmental data locally within the initial simulation environment based on the agent description information corresponding to the mobile intelligent agent to obtain a real-time environment information set, wherein the agent description information includes: the agent's real-time location and sensor specification information, the sensor specification information representing the sensor specifications of the virtual environment sensors corresponding to the mobile intelligent agent; generating environmental state features and agent state features based on the real-time environment information set and the agent's real-time state; generating behavior planning information for the mobile intelligent agent based on the environmental state features and the agent state features; and performing behavior simulation on the mobile intelligent agent based on the behavior planning information and the agent's real-time location.

[0006] Secondly, some embodiments of this disclosure provide a behavior simulation device for mobile intelligent agents applied in emergency rescue scenarios. The device includes: a first generation unit configured to generate an initial simulation environment based on scene environment information and simulation configuration information, wherein the scene environment information represents the regional environment information corresponding to the emergency rescue scenario, and the initial simulation environment contains a mobile intelligent agent whose position has been initialized and whose behavior simulation is to be performed; and an environmental data local extraction unit configured to perform local environmental data extraction within the initial simulation environment based on the intelligent agent description information corresponding to the mobile intelligent agent, to obtain a real-time environmental information set. The aforementioned agent description information includes: the agent's real-time location and sensor specification information, wherein the sensor specification information characterizes the sensor specifications of the virtual environment sensors corresponding to the mobile agent; the second generation unit is configured to generate environmental state features and agent state features based on the aforementioned real-time environmental information set and the real-time state of the mobile agent; the third generation unit is configured to generate behavior planning information for the mobile agent based on the aforementioned environmental state features and the aforementioned agent state features; and the simulation unit is configured to perform behavior simulation on the mobile agent based on the aforementioned behavior planning information and the aforementioned real-time location of the agent.

[0007] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0008] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0009] The above embodiments of this disclosure have the following beneficial effects: The behavior simulation method for mobile intelligent agents applied to emergency rescue scenarios, through some embodiments of this disclosure, effectively simulates behavior in different emergency rescue scenarios by combining mobile intelligent agents, greatly improving the robustness and accuracy of behavior planning for mobile robots in different emergency rescue scenarios. In practice, common rescue robots mainly use optical fibers as the control medium for robot control. However, due to the often harsh rescue environment, situations such as fiber bending can lead to a decrease in robot control stability. Therefore, sensor-based autonomous rescue robots have become a new research direction. However, because the rescue environment differs from the conventional environment and is more severe, directly reusing existing machine learning models makes it difficult to guarantee control accuracy. Furthermore, the collection of training samples is also difficult, further increasing the difficulty of model updates and iterations. Therefore, this disclosure, from a simulation perspective, realizes the behavior simulation of mobile intelligent agents. Specifically, this disclosure first generates an initial simulation environment based on scene environment information and simulation configuration information. The scene environment information represents the regional environment information corresponding to the emergency rescue scenario, and the initial simulation environment contains a mobile intelligent agent whose position has been initialized and whose behavior simulation is to be performed. By combining different emergency rescue scenarios, a rapid simulation environment is built in the initial state. Secondly, based on the agent description information corresponding to the mobile agent, local environmental data is extracted within the initial simulation environment to obtain a real-time environmental information set. The agent description information includes the agent's real-time location and sensor specifications, where the sensor specifications characterize the sensor specifications of the virtual environment sensors corresponding to the mobile agent. In practice, conventional behavior simulation is mainly based on rule-based behavior control. However, in practice, sensor-based autonomous rescue robots primarily use sensors as data acquisition sources and perform corresponding analysis. Therefore, this disclosure uses environmental sensor simulation to achieve data acquisition and perception of the surrounding environment from a sensor perspective, thereby realizing a full-link simulation from data acquisition to behavior planning. Next, based on the real-time environmental information set and the real-time state of the mobile agent, environmental state features and agent state features are generated. This characterizes the environmental state and agent state from a feature perspective. Furthermore, based on the environmental state features and agent state features, behavior planning information for the mobile agent is generated. This quantifies the behavior control for the mobile agent. Finally, based on the aforementioned behavior planning information and the real-time location of the agent, behavioral simulation is performed on the mobile agent. In summary, this method achieves effective behavior simulation in different emergency rescue scenarios, significantly improving the robustness and accuracy of behavior planning for mobile robots in various emergency rescue situations. Attached Figure Description

[0010] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0011] Figure 1 This is a flowchart of some embodiments of the behavior simulation method for mobile intelligent agents applied in emergency rescue scenarios according to the present disclosure; Figure 2 This is a schematic diagram showing the locations of the environmental sensors; Figure 3 This is a schematic diagram of another deployment location for environmental sensors; Figure 4 This is a schematic diagram of the voxel block tree generation process; Figure 5 This is a structural schematic diagram of some embodiments of the behavior simulation device for mobile intelligent agents applied in emergency rescue scenarios according to the present disclosure; Figure 6 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0012] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0013] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0014] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0015] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0016] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0017] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] refer to Figure 1 The diagram illustrates a flow 100 of some embodiments of a behavior simulation method for mobile intelligent agents applied in emergency rescue scenarios according to the present disclosure. This behavior simulation method for mobile intelligent agents applied in emergency rescue scenarios includes the following steps: Step 101: Generate the initial simulation environment based on the scene environment information and simulation configuration information.

[0019] In some embodiments, the execution subject (e.g., a computing device) of the behavior simulation method of a mobile intelligent agent applied to an emergency rescue scenario can generate an initial simulation environment based on scenario environment information and simulation configuration information.

[0020] The scenario environment information represents the regional environmental information corresponding to the emergency rescue scenario. The initial simulation environment includes a mobile intelligent agent whose position has been initialized and whose behavior is to be simulated. The simulation configuration information represents the environmental parameters corresponding to the initial simulation environment.

[0021] Optionally, the scene environment information includes: a set of scene type and scene element information. Scene element information includes: scene element type, scene element location, and scene element specification parameters. The scene type represents the type of emergency rescue scenario. For example, the scene type may include: earthquake rescue scenario type, fire rescue scenario type, and flood rescue scenario type. The scene element type represents the scene elements within the corresponding area of ​​the scene environment information. For example, the scene element type may include: building type, road surface type, and vegetation type. Specifically, based on the simulation accuracy, scene elements can be classified into more fine-grained types. The scene element location represents the position of the scene element. For example, the scene element location can be represented by three-dimensional coordinates. Scene element specification parameters represent the specification parameters of the scene element. For example, scene element specification parameters may include: scene element size and scene element material type.

[0022] Optionally, the simulation configuration information includes: agent type, simulation accuracy parameters, and simulation scale parameters. The agent type represents the type of rescue machine corresponding to the mobile agent. The simulation accuracy parameters represent the simulation accuracy corresponding to the initial simulation environment. The simulation scale parameters represent the simulation scale during the initial simulation environment setup.

[0023] In practice, firstly, the aforementioned implementing entity can obtain a high-precision 3D map corresponding to the target area. The target area can be the region to be simulated for emergency rescue. Then, the 3D high-precision map is segmented to obtain a set of scene element information. Next, the corresponding scene type is selected based on the emergency rescue simulation requirements. Further, based on the computing power and timeliness requirements of the computing equipment used in the simulation, corresponding simulation accuracy and scale parameters are set. Finally, based on the scene type, a 3D simulation is performed using the scene element information set and simulation configuration information to obtain the initial simulation environment. For example, taking a fire rescue scene as an example, at least one scene element can be selected as the starting fire point, and the fire spread can be simulated through 3D simulation, with the result of the fire spread serving as the initial simulation environment.

[0024] It should be noted that the aforementioned computing devices can be either hardware or software. When the computing device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device. When the computing device is software, it can be installed on the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here.

[0025] In some optional implementations of certain embodiments, the aforementioned execution entity generates an initial simulation environment based on scene environment information and simulation configuration information, including: Step S1: Generate the simulation space based on the simulation scale parameters mentioned above.

[0026] The simulation space can be a three-dimensional space.

[0027] In practice, a three-dimensional space can be constructed based on the simulation scale parameters to serve as the simulation space.

[0028] Step S2: Determine the sensing range based on the above-mentioned agent type and sensor specifications.

[0029] The aforementioned perception range represents the perception range of the virtual environment sensors corresponding to the mobile intelligent agent. Sensor specification information represents the sensor specifications of the virtual environment sensors corresponding to the mobile intelligent agent. Specifically, sensor specification information may include: the number of sensors, sensor placement locations, and sensor specification parameters. The number of sensors represents the number of environmental sensors installed on the rescue robot. The sensor placement locations represent the deployment locations of the environmental sensors on the rescue robot. Sensor specification parameters represent the specific sensor parameters. To ensure the effectiveness of subsequent behavioral simulation, the constructed mobile intelligent agent must maintain consistency with the real rescue robot.

[0030] Example 1, see Figure 2 The diagram shows the deployment of environmental sensors, where the rescue robot can be equipped with four environmental sensors 201. Each environmental sensor 201 can be a lidar sensor. Each sensor 201 has a corresponding fan-shaped detection area 202, depending on its orientation and detection range. From a top-down perspective, the execution entity can use the center of the rescue robot as the center of its sensing range and determine the length of the sensing range by magnifying a square frame 203. Specifically, during the magnification process, the square frame 203 will intersect with the fan-shaped detection areas 202. The magnification stops when the area of ​​the four fan-shaped detection areas 202 within the square frame 203 reaches its maximum value. The side length of the square frame 203 at this point is then taken as the length of the sensing range.

[0031] Example 2, see Figure 3 The diagram shows another possible deployment location for the environmental sensors, where the rescue robot can be equipped with four environmental sensors 201. Each environmental sensor 201 can be a full-color camera. Each environmental sensor 201 has a corresponding fan-shaped detection area 202, depending on its orientation and detection range. From a top-down perspective, the aforementioned execution entity can use the center of the rescue robot as the center of its sensing range, and determine the length of the sensing range by magnifying a square frame 203. Specifically, during the magnification process, the square frame 203 will intersect with the fan-shaped detection areas 202. The magnification stops when the area of ​​the four fan-shaped detection areas 202 within the square frame 203 reaches its maximum value. The side length of the square frame 203 at this point is then taken as the length of the sensing range.

[0032] Step S3: Based on the above perception range, perform spatial voxelization on the above simulation space to obtain the voxelized simulation space.

[0033] The aforementioned voxelized simulation space includes a set of voxel blocks. Each voxel block can be a cube. The side length of each voxel block is the same as the length of the sensing range.

[0034] In practice, the simulation space can be spatially voxelized by using the length of the perception range as the side length of the voxel block through octree voxelization, thus obtaining the voxelized simulation space.

[0035] Step S4: Using the initial loading position of the mobile intelligent agent as the center, perform eight-way spatial division on the voxelized simulation space to obtain a set of voxel blocks.

[0036] Each voxel in the voxel block group corresponds to the same segmentation direction.

[0037] In practice, firstly, since the simulation space after voxelization is a three-dimensional cubic space, eight quadrangular pyramid spaces can be constructed by taking the initial loading position of the mobile agent as the vertex and the four spatial vertices of the simulation space located on the same face after voxelization. Next, the voxel blocks within the quadrangular pyramid spaces are used as voxel block groups along the segmentation direction, thus obtaining a set of voxel block groups containing eight voxel block groups. Specifically, voxel blocks located in two adjacent quadrangular pyramid spaces can repeatedly be located within the corresponding two voxel block groups of the two quadrangular pyramid spaces.

[0038] Step S5: Generate a voxel block tree set based on the initial loading position and voxel block group set mentioned above.

[0039] The voxel block tree is rooted at the initial loading position.

[0040] In practice, a voxel block tree is constructed based on the distance between the block center and the initial loading position of the voxel block. Voxel blocks with the same positional distance correspond to the same order.

[0041] As an example: see Figure 4 The diagram shown illustrates the generation process of the voxel block tree, in which... Figure 4 The voxel block group shown consists of 3×3×3 voxel blocks (27 in total). For each layer of 3×3 voxel blocks, the voxel block whose initial loading position is closest to the pivot position is the closest, and the other 8 voxel blocks are equidistant from their initial loading positions. Therefore, a voxel block tree with a depth of 7 is constructed. In particular, to facilitate subsequent breadth-first traversal, all voxel blocks in lower layers have the first voxel block in the upper layer as their parent node.

[0042] Step S6: Generate the initial simulation environment based on the above voxel block tree set, the above scene environment information, and the above simulation configuration information.

[0043] Optionally, the execution entity generates the initial simulation environment based on the voxel block tree set, the scene environment information, and the simulation configuration information, including: Step S61: For each voxel block tree in the above voxel block tree set, perform the following processing steps: Step S611: Perform a breadth-first traversal on the above voxel block tree to obtain the tree node group sequence.

[0044] Within each tree node group, the tree nodes correspond to the same order, and each tree node corresponds to a voxel block. Specifically, each tree node in a tree node group corresponds to the same tree depth.

[0045] As an example, see further. Figure 4 ,by Figure 4Taking the voxel block tree shown as an example, the sequence of tree node groups can include: 6 tree node groups. Specifically, the first tree node group includes 1 tree node, the second tree node group includes 7 tree nodes, the third tree node group includes 1 tree node, the fourth tree node group includes 7 tree nodes, the fifth tree node group includes 1 tree node, and the sixth tree node group includes 7 tree nodes. The order of the first tree node group is 1, the second tree node group is 2, the third tree node group is 3, the fourth tree node group is 4, the fifth tree node group is 5, and the sixth tree node group is 6.

[0046] Step S612: Based on the tree node group sequence and the scene element information set, perform the following scene loading steps: Step S6121: Extract the tree node group corresponding to the target order from the tree node group sequence and use it as the target tree node group.

[0047] The initial order of the objective is 1.

[0048] As an example, continuing from the example in step S611, the target tree node group selected when the scene loading step is executed for the first time can be the first tree node group.

[0049] Step S6122: Based on the scene element positions included in the scene element information, perform element matching between the target tree nodes in the target tree node group and the scene element information in the scene element information set to obtain a matching information group.

[0050] In practice, since tree nodes correspond to voxel blocks, the scene element information is matched with the corresponding target tree node by determining the voxel block into which the scene element's position falls, thus obtaining the corresponding matching information. Specifically, the matching information can be a tuple, including: the matched target tree node and the corresponding scene element information.

[0051] Step S6123: Load the local scene according to the matching information group and the above simulation accuracy parameters to obtain the target simulation environment.

[0052] Using the aforementioned simulation accuracy parameters as accuracy constraints, and based on the scene element information included in the matching information group, local element rendering is performed on the corresponding target voxel block to obtain the target simulation environment.

[0053] Step S6124: In response to the target order being equal to the tree depth corresponding to the voxel block tree, the above scene loading steps are terminated.

[0054] Step S613: In response to the target order being less than the tree depth corresponding to the voxel block tree, the target order is incremented, and the target order after the increment is used as the target order. The scene element information set with the scene element information corresponding to the matching information group removed is used as the scene element information set, and the above scene loading steps are re-executed.

[0055] In practice, the order increases by 1 each time during the order increase process.

[0056] Step S614: Based on the obtained target simulation environment set, generate a local simulation environment in the initial simulation environment that corresponds to the segmentation direction of the voxel block tree.

[0057] In practice, since the target simulation environment is a partial simulation environment for voxel block groups of the same order, it is necessary to stitch together the obtained target simulation environment set to obtain the local simulation environment corresponding to the segmentation direction of the voxel block tree. In this way, layer-by-layer and direction-by-direction simulation environment rendering centered on the mobile intelligent agent is realized, and the rendering granularity is constrained by the sensor specifications. Especially under low computing power conditions, it can ensure the effective operation of the simulation.

[0058] Step 102: Based on the agent description information corresponding to the mobile agent, perform local extraction of environmental data in the initial simulation environment to obtain a set of real-time environmental information.

[0059] In some embodiments, the aforementioned execution entity can extract environmental data locally within the initial simulation environment based on the agent description information corresponding to the mobile agent, thereby obtaining a set of real-time environmental information.

[0060] The agent description information includes: the agent's real-time location and sensor specifications. The real-time location represents the mobile agent's real-time position and orientation. As the simulation iterates, the mobile agent's position and orientation will change. The sensor specifications represent the sensor specifications of the virtual environment sensors corresponding to the mobile agent. Specifically, sensor specifications can include: the number of sensors, sensor placement locations, and sensor parameter specifications. The number of sensors represents the number of environmental sensors installed on the rescue robot. The sensor placement locations represent the deployment locations of the environmental sensors on the rescue robot. The sensor parameter specifications represent the specific sensor parameters. To ensure the effectiveness of subsequent behavioral simulations, the constructed mobile agent must maintain consistency with the real rescue robot.

[0061] In practice, environmental data can be collected in the initial simulation environment by centering on the real-time position of the agent and constraining the data acquisition distance corresponding to the sensor's directional information. In particular, the positions and orientations of the environmental sensors on the rescue robot are often fixed. To ensure that the positions and orientations of the mobile agent and the virtual environment sensors correspond to those of the real environment sensors, it is necessary to adjust the acquisition direction of the virtual environment sensors based on the orientation represented by the agent's real-time position, thereby further realizing the acquisition of real-time environmental information.

[0062] In some optional implementations of certain embodiments, the execution entity performs local environmental data extraction within the initial simulation environment based on the agent description information corresponding to the mobile agent, to obtain a real-time environmental information set, including: Step S1: Determine the sensor location corresponding to the virtual environment sensor based on the real-time location of the intelligent agent included in the intelligent agent description information above.

[0063] In practice, since the positions and orientations of environmental sensors on rescue robots are often fixed, to ensure that the positions and orientations of the mobile agent and virtual environmental sensors correspond to those of the real environmental sensors, it is necessary to determine the sensor orientation corresponding to the virtual environmental sensor by combining the real-time orientation of the agent. For example, assuming the virtual environmental sensor is placed directly in front of the mobile agent, as the orientation of the mobile agent changes, the virtual environmental sensor must also be rotated by the same angle as the orientation of the mobile agent to obtain the sensor orientation.

[0064] Step S2: Determine the voxel block tree in the above voxel block tree set that is associated with the above sensor orientation as a candidate voxel block tree.

[0065] In practice, based on the working principles of different sensors—for example, cameras and lidar primarily use a fan-shaped detection area, while distance sensors primarily use a point detection area—voxel block trees within the detection range corresponding to the virtual environment sensor, oriented downwards from the sensor's location, can be identified as candidate voxel block trees.

[0066] Step S3: Based on the above sensor orientation and sensor specification information, perform subtree segmentation on the above candidate voxel block tree to obtain the target voxel block set.

[0067] In practice, due to the limitations of the detection direction and detection area of ​​virtual environment sensors, not all voxel blocks in the candidate voxel block tree fall within the detection area. Therefore, it is necessary to combine the sensor orientation and the detection area corresponding to the sensor specification information mentioned above to perform subtree segmentation on the voxel block tree to obtain the voxel blocks contained within the detection area, which serves as the target voxel block set.

[0068] Step S4: Based on the above sensor orientation, extract scene elements from the target voxel blocks in the above target voxel block set under a fixed viewpoint to obtain the above real-time environment information set.

[0069] In practice, the voxels in the voxel block tree are already bound to the scene elements corresponding to the scene element information. In particular, since multiple scene elements in the same direction along the sensor orientation may be occluded in the actual environment, it is necessary to combine the sensor orientation as the acquisition viewpoint to extract features from the scene elements corresponding to the target voxel block to obtain a set of real-time environmental information.

[0070] Step 103: Generate environmental state features and agent state features based on the real-time environmental information set and the real-time state of the mobile agent.

[0071] In some embodiments, the aforementioned execution entity can generate environmental state features and agent state features based on the real-time environmental information set and the real-time state of the agent corresponding to the mobile agent.

[0072] Among them, environmental state features represent the environmental state through high-dimensional features. Agent state features represent the state of the mobile agent through high-dimensional features.

[0073] In practice, since environmental state information is collected from the perspective of virtual environment sensors—for example, taking a camera as an example—image processing models such as YOLO (You Only Look Once) can be used as a base model to extract features from real-time environmental state information, thus obtaining environmental state features. Similarly, taking a LiDAR as an example, point cloud processing models such as 3D-Unet can be used as a base model to extract features from real-time environmental state information, thus obtaining environmental state features. Furthermore, when the virtual environment sensor corresponds to multiple different types of sensors, feature fusion can be implemented based on the above examples to achieve feature fusion across different modalities, resulting in environmental state features. In addition, for agent state features, state index mapping or encoding methods can be used to obtain agent state features.

[0074] In some optional implementations of certain embodiments, the execution entity generates environmental state features and agent state features based on the aforementioned real-time environmental information set and the real-time state of the mobile agent, including: Step S1: Based on the above real-time environmental information set, determine the first environmental region and the second environmental region.

[0075] The first environmental region represents the intersection acquisition area corresponding to the virtual environmental sensor, and the second environmental region represents the non-intersection acquisition area corresponding to the virtual environmental sensor.

[0076] In practice, to ensure effective detection coverage, rescue robots are often equipped with multiple environmental sensors. Therefore, to maintain consistency, mobile intelligent agents also need to be equipped with multiple virtual environmental sensors. There may be overlap in the detection areas between adjacent virtual environmental sensors. Conventionally, independent feature extraction is performed based on the environmental sensors. However, this disclosure, in order to improve the response speed of simulation, divides the region corresponding to the real-time environmental information through intersection region partitioning, obtaining a first environmental region and a second environmental region.

[0077] Step S2: Extract cross-linking features from the real-time environmental information in the set of real-time environmental information that corresponds to the first environmental region mentioned above, and obtain the first environmental features.

[0078] In practice, the first environmental region is an intersection area, corresponding to real-time environmental information from at least two virtual environment sensor perspectives. This inherently involves some feature redundancy. Furthermore, extracting features independently from a single virtual environment sensor ignores feature perception from multiple virtual environment sensor perspectives. Therefore, this disclosure employs a cross-linking feature extraction method to extract cross-linking features from the real-time environmental information corresponding to the first environmental region, obtaining the first environmental features. Specifically, this disclosure uses a multi-path convolutional neural network to extract features from the real-time environmental information corresponding to the first environmental region in parallel, obtaining the first environmental features. The multi-path convolutional neural network includes at least two parallel convolutional neural networks, and the parameters are shared among the convolutional neural networks.

[0079] Step S3: Extract features from the real-time environmental information in the above-mentioned real-time environmental information set that corresponds to the above-mentioned second environmental region to obtain the second environmental features.

[0080] In practice, since the second environmental region is a non-overlapping region, corresponding to only one virtual sensor, it is only necessary to extract features independently from the real-time environmental information corresponding to the second environmental region. Specifically, to ensure the consistency of the feature space between the second environmental features and the first environmental features, a single convolutional neural network in a multi-path convolutional neural network can be activated to extract features from the real-time environmental information corresponding to the second environmental region, thus obtaining the second environmental features.

[0081] Step S4: Based on the real-time location of the intelligent agent, perform bird's-eye view feature projection on the first environmental feature and the second environmental feature to obtain the bird's-eye view environmental features.

[0082] In practice, the aforementioned executing entity can take the real-time location of the intelligent agent as the center and orientation, and use the BEV (Bird's Eye View) feature projection method to project the first environmental feature and the second environmental feature to obtain the bird's eye environmental feature, thereby achieving a unified mapping of different environmental features from a bird's-eye view.

[0083] Step S5: Perform multi-scale feature extraction on the above-mentioned bird's-eye view environmental features to obtain environmental state features.

[0084] In practice, the above-mentioned bird's-eye view environmental features can be extracted at multiple scales using the FPN (Feature Pyramid Networks) model to obtain environmental state features. These environmental state features include features at multiple scales.

[0085] Step S6: Perform state mapping on the real-time state of the above-mentioned agent to obtain the state features of the agent.

[0086] In practice, since the real-time state of an agent is mainly represented by index values, the real-time state of the agent can be mapped by value mapping (such as normalization mapping) to obtain the agent's state features.

[0087] Step 104: Generate behavioral planning information for the mobile intelligent agent based on the environmental state characteristics and the intelligent agent state characteristics.

[0088] In some embodiments, the aforementioned executing entity can generate behavioral planning information for the mobile intelligent agent based on environmental state characteristics and intelligent agent state characteristics.

[0089] Among them, behavior planning information represents the behavior control instructions for the mobile intelligent agent at the next moment. Specifically, behavior planning information for the mobile intelligent agent can be generated by a reward model (RM) based on environmental state characteristics and intelligent agent state characteristics.

[0090] In some optional implementations of certain embodiments, the aforementioned executing entity generates behavior planning information for the mobile intelligent agent based on environmental state characteristics and agent state characteristics, including: Step S1: Perform feature segmentation on the above environmental state features to obtain a set of sub-environmental state features.

[0091] In practice, conventional feature processing methods primarily use the entire system as input. While this approach is relatively simple, effective feature extraction often requires a deep model, increasing the model's processing load. The behavior planning method for mobile intelligent agents disclosed in this paper, however, is mainly based on the perception of the surrounding environment. Therefore, by using feature segmentation, the environmental state features corresponding to the overall environment are divided into sub-environmental state features corresponding to independent small regions. By sacrificing some correlation between features, independent feature extraction is performed on these small regions, thereby reducing model complexity.

[0092] Step S2: Perform parallel feature encoding on the sub-environmental state features in the above sub-environmental state feature set to obtain the encoded environmental state feature set.

[0093] In practice, the sub-environmental state features in the sub-environmental state feature set are encoded in parallel. Specifically, multiple parallel weak feature extractors (e.g., a convolutional neural network consisting of 3-5 convolutional layers) are set up to encode each sub-environmental state feature in parallel, resulting in the encoded environmental state feature set.

[0094] Step S3: Perform coarse environmental state classification on each sub-environmental state feature in the above encoded environmental state feature set to obtain an environmental classification type set.

[0095] In practice, different elements in the surrounding environment have varying degrees of influence on the mobile agent. Therefore, by mapping corresponding weights through coarse classification, the environmental state features are directly weighted, which effectively reduces the amount of feature processing compared to conventional multi-layer convolutional feature extraction methods. Specifically, a multi-classifier that takes sub-environmental state features as input is set to obtain the environmental classification type. The multi-classifier and weak feature extractor are trained as a whole.

[0096] Step S4: Generate a weighted feature matrix based on the above set of environmental classification types.

[0097] In practice, each different environmental classification type has a pre-set corresponding influence value. The weight value (weight value = confidence level × influence value) is obtained by combining the confidence level and influence value corresponding to the environmental classification type, thus yielding the weight feature matrix.

[0098] Step S6: Based on the above weight feature matrix, perform feature weighting on the above environmental state features to obtain the weighted environmental state features.

[0099] In practice, the weight values ​​in the weight feature matrix are multiplied by the feature values ​​at the corresponding positions of the environmental state features to obtain the weighted environmental state features.

[0100] Step S7: Generate the above-mentioned behavior planning information based on the above-mentioned weighted environmental state features, the above-mentioned agent state features, and the pre-trained behavior planning model.

[0101] In practice, the behavior planning model adopts an incentive model, which takes the weighted environmental state features corresponding to the current moment and the aforementioned agent state features as inputs to the model, and outputs the behavior plan for the mobile agent at the next moment.

[0102] Step 105: Based on the behavior planning information and the real-time location of the agent, perform behavior simulation on the mobile agent.

[0103] In practice, the aforementioned implementing entities can simulate the behavior of mobile intelligent agents based on behavior planning information and the real-time location of the intelligent agents.

[0104] Specifically, based on the agent's real-time location, the position and behavior of the mobile agent can be updated according to behavior planning information, thereby achieving the purpose of simulating the behavior of the mobile agent. In particular, to ensure the effectiveness of the simulation, multiple rounds of iterative iteration can be set to achieve continuous simulation control of the mobile agent.

[0105] In some optional implementations of some embodiments, the above method further includes: Step S1: Generate behavioral evaluation information for the simulated behavior.

[0106] In practice, the incentive values ​​corresponding to the incentive model can be used as behavioral evaluation information for simulated behaviors.

[0107] Step S2: Generate behavior samples based on the above behavior evaluation information, the above behavior planning information, the above agent description information, and the above scene environment information.

[0108] In practice, JSON (JavaScript Object Notation) format can be used to format the behavior evaluation information, the aforementioned behavior planning information, the aforementioned agent description information, and the aforementioned scenario environment information to generate behavior samples.

[0109] Step S3: Archive and store the above behavioral samples.

[0110] In practice, this method can yield a large number of behavioral samples for training and optimizing the control algorithms of subsequent rescue robots. Furthermore, the results of the entire simulation can serve as a reference for behavioral planning in real emergency rescue operations.

[0111] The above embodiments of this disclosure have the following beneficial effects: The behavior simulation method for mobile intelligent agents applied to emergency rescue scenarios, through some embodiments of this disclosure, effectively simulates behavior in different emergency rescue scenarios by combining mobile intelligent agents, greatly improving the robustness and accuracy of behavior planning for mobile robots in different emergency rescue scenarios. In practice, common rescue robots mainly use optical fibers as the control medium for robot control. However, due to the often harsh rescue environment, situations such as fiber bending can lead to a decrease in robot control stability. Therefore, sensor-based autonomous rescue robots have become a new research direction. However, because the rescue environment differs from the conventional environment and is more severe, directly reusing existing machine learning models makes it difficult to guarantee control accuracy. Furthermore, the collection of training samples is also difficult, further increasing the difficulty of model updates and iterations. Therefore, this disclosure, from a simulation perspective, realizes the behavior simulation of mobile intelligent agents. Specifically, this disclosure first generates an initial simulation environment based on scene environment information and simulation configuration information. The scene environment information represents the regional environment information corresponding to the emergency rescue scenario, and the initial simulation environment contains a mobile intelligent agent whose position has been initialized and whose behavior simulation is to be performed. By combining different emergency rescue scenarios, a rapid simulation environment is built in the initial state. Secondly, based on the agent description information corresponding to the mobile agent, local environmental data is extracted within the initial simulation environment to obtain a real-time environmental information set. The agent description information includes the agent's real-time location and sensor specifications, where the sensor specifications characterize the sensor specifications of the virtual environment sensors corresponding to the mobile agent. In practice, conventional behavior simulation is mainly based on rule-based behavior control. However, in practice, sensor-based autonomous rescue robots primarily use sensors as data acquisition sources and perform corresponding analysis. Therefore, this disclosure uses environmental sensor simulation to achieve data acquisition and perception of the surrounding environment from a sensor perspective, thereby realizing a full-link simulation from data acquisition to behavior planning. Next, based on the real-time environmental information set and the real-time state of the mobile agent, environmental state features and agent state features are generated. This characterizes the environmental state and agent state from a feature perspective. Furthermore, based on the environmental state features and agent state features, behavior planning information for the mobile agent is generated. This quantifies the behavior control for the mobile agent. Finally, based on the aforementioned behavior planning information and the real-time location of the agent, behavioral simulation is performed on the mobile agent. In summary, this method achieves effective behavior simulation in different emergency rescue scenarios, significantly improving the robustness and accuracy of behavior planning for mobile robots in various emergency rescue situations.

[0112] Further reference Figure 5As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a behavior simulation device for mobile intelligent agents applied in emergency rescue scenarios. These device embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, the behavior simulation device for mobile intelligent agents applied to emergency rescue scenarios can be specifically applied to various electronic devices.

[0113] like Figure 5 As shown, a mobile intelligent agent behavior simulation device 500 applied to emergency rescue scenarios in some embodiments includes: a first generation unit 501, an environmental data local extraction unit 502, a second generation unit 503, a third generation unit 504, and a simulation unit 505. The system comprises the following components: a first generation unit 501, configured to generate an initial simulation environment based on scene environment information and simulation configuration information, wherein the scene environment information represents the regional environment information corresponding to the emergency rescue scenario, and the initial simulation environment contains a mobile intelligent agent whose position has been initialized and whose behavior is to be simulated; an environment data local extraction unit 502, configured to perform local extraction of environment data within the initial simulation environment based on the intelligent agent description information corresponding to the mobile intelligent agent, to obtain a real-time environment information set, wherein the intelligent agent description information includes: the real-time location of the intelligent agent and sensor specification information, wherein the sensor specification information represents the sensor specifications of the virtual environment sensors corresponding to the mobile intelligent agent; a second generation unit 503, configured to generate environment state features and intelligent agent state features based on the real-time environment information set and the real-time state of the intelligent agent corresponding to the mobile intelligent agent; a third generation unit 504, configured to generate behavior planning information for the mobile intelligent agent based on the environment state features and the intelligent agent state features; and a simulation unit 505, configured to perform behavior simulation on the mobile intelligent agent based on the behavior planning information and the real-time location of the intelligent agent.

[0114] It is understandable that the units and references described in the mobile intelligent agent behavior simulation device 500 applied to emergency rescue scenarios are similar. Figure 1 The steps in the described method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method are also applicable to the behavior simulation device 500 of mobile intelligent agents applied in emergency rescue scenarios and the units contained therein, and will not be repeated here.

[0115] The following is for reference. Figure 6 It illustrates a schematic diagram of the structure of an electronic device (e.g., a computing device) suitable for implementing some embodiments of the present disclosure. Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality or scope of the embodiments of this disclosure. Figure 6As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The memory may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system and a computer program. The computer program includes program instructions that, when executed, cause the processor to perform any of the methods described above. The processor provides computational and control capabilities to support the operation of the entire computer device. The internal memory provides an environment for the execution of the computer program in the non-volatile storage medium; when executed by the processor, the computer program causes the processor to perform any of the methods described above. The network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the computer device to which the present disclosure is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0116] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0117] In one embodiment, the processor is configured to run a computer program stored in a memory to perform the following steps: generating an initial simulation environment based on scene environment information and simulation configuration information, wherein the scene environment information represents the regional environment information corresponding to the emergency rescue scenario, and the initial simulation environment contains a mobile intelligent agent whose position has been initialized and whose behavior is to be simulated; extracting environmental data locally within the initial simulation environment based on the intelligent agent description information corresponding to the mobile intelligent agent to obtain a real-time environment information set, wherein the intelligent agent description information includes: the real-time location of the intelligent agent and sensor specification information, and the sensor specification information represents the sensor specifications of the virtual environment sensors corresponding to the mobile intelligent agent; generating environmental state features and intelligent agent state features based on the real-time environment information set and the real-time state of the intelligent agent corresponding to the mobile intelligent agent; generating behavior planning information for the mobile intelligent agent based on the environmental state features and the intelligent agent state features; and performing behavior simulation on the mobile intelligent agent based on the behavior planning information and the real-time location of the intelligent agent.

[0118] This disclosure also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, and the method implemented when the program instructions are executed can be referred to the various embodiments of the methods described above.

[0119] The aforementioned computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as a hard disk or memory of the computer device. Alternatively, the aforementioned computer-readable storage medium may be an external storage device of the computer device, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0120] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0121] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for simulating the behavior of a mobile intelligent agent applied in emergency rescue scenarios, characterized in that, include: An initial simulation environment is generated based on scene environment information and simulation configuration information. The scene environment information represents the regional environment information corresponding to the emergency rescue scene. The initial simulation environment contains a mobile intelligent agent that has been initialized in position and is to be simulated in behavior. Based on the agent description information corresponding to the mobile agent, local environmental data is extracted in the initial simulation environment to obtain a set of real-time environmental information. The agent description information includes: the real-time location of the agent and sensor specification information. The sensor specification information represents the sensor specifications of the virtual environment sensors corresponding to the mobile agent. Based on the real-time environmental information set and the real-time state of the mobile intelligent agent, environmental state features and intelligent agent state features are generated. Based on the environmental state characteristics and the agent state characteristics, behavioral planning information for the mobile agent is generated; Based on the behavior planning information and the real-time location of the agent, the behavior of the mobile agent is simulated.

2. The behavior simulation method for mobile intelligent agents applied in emergency rescue scenarios according to claim 1, characterized in that, The method further includes: Generate behavioral evaluation information for simulated behaviors; Based on the behavior evaluation information, the behavior planning information, the agent description information, and the scene environment information, a behavior sample is generated; The behavioral samples are archived and stored.

3. The behavior simulation method for mobile intelligent agents applied in emergency rescue scenarios according to claim 2, characterized in that, The scene environment information includes: a scene type and a set of scene element information. The scene element information includes: scene element type, scene element location, and scene element specification parameters. The simulation configuration information includes: agent type, simulation accuracy parameters, and simulation scale parameters. The step of generating the initial simulation environment based on the scene environment information and the simulation configuration information includes: Based on the simulation scale parameters, a simulation space is generated; Based on the agent type and sensor specifications, the perception range is determined, wherein the perception range characterizes the perception range of the virtual environment sensor corresponding to the mobile agent. Based on the perception range, the simulation space is spatially voxelized to obtain a voxelized simulation space, wherein the voxelized simulation space includes a set of voxel blocks. Centered on the initial loading position of the mobile intelligent agent, the voxelized simulation space is divided into eight directions to obtain a set of voxel blocks, wherein each voxel block in the voxel block set corresponds to the same division direction. Based on the initial loading position and the set of voxel blocks, a set of voxel block trees is generated, wherein the voxel block tree is rooted at the initial loading position; The initial simulation environment is generated based on the voxel block tree set, the scene environment information, and the simulation configuration information.

4. The behavior simulation method for mobile intelligent agents applied in emergency rescue scenarios according to claim 3, characterized in that, The step of generating the initial simulation environment based on the voxel block tree set, the scene environment information, and the simulation configuration information includes: For each voxel block tree in the set of voxel block trees, perform the following processing steps: A breadth-first traversal is performed on the voxel block tree to obtain a sequence of tree node groups, wherein the tree nodes in the tree node group correspond to the same order, and the tree nodes correspond to voxel blocks; Based on the tree node group sequence and the scene element information set, perform the following scene loading steps: Extract the tree node group corresponding to the target order from the tree node group sequence, and use it as the target tree node group, where the target order is initially 1; Based on the location of scene elements included in the scene element information, element matching is performed between the target tree nodes in the target tree node group and the scene element information in the scene element information set to obtain a matching information group. Local scene loading is performed based on the matching information group and the simulation accuracy parameters to obtain the target simulation environment; The scene loading step ends when the target order is equal to the tree depth corresponding to the voxel block tree. In response to the target order being less than the tree depth corresponding to the voxel block tree, the target order is incremented, and the target order after incrementing is used as the target order. The scene element information set with the scene element information corresponding to the matching information group removed is used as the scene element information set, and the scene loading step is re-executed. Based on the obtained set of target simulation environments, a local simulation environment corresponding to the segmentation direction of the voxel block tree is generated in the initial simulation environment.

5. The behavior simulation method for mobile intelligent agents applied in emergency rescue scenarios according to claim 4, characterized in that, The step of extracting environmental data locally within the initial simulation environment based on the agent description information corresponding to the mobile agent to obtain a real-time environmental information set includes: Based on the real-time location of the intelligent agent included in the intelligent agent description information, determine the sensor location corresponding to the virtual environment sensor; The voxel block tree associated with the sensor orientation in the set of voxel block trees is determined as a candidate voxel block tree; Based on the sensor orientation and sensor specification information, the candidate voxel block tree is subtree segmented to obtain the target voxel block set; Based on the sensor orientation, scene elements corresponding to the target voxel blocks in the target voxel block set are extracted from a fixed viewpoint to obtain the real-time environment information set.

6. The behavior simulation method for mobile intelligent agents in emergency rescue scenarios according to claim 5, characterized in that, The step of generating environmental state features and agent state features based on the real-time environmental information set and the real-time state of the mobile agent includes: Based on the real-time environmental information set, a first environmental region and a second environmental region are determined, wherein the first environmental region represents the intersection acquisition region corresponding to the virtual environment sensor, and the second environmental region represents the non-intersection acquisition region corresponding to the virtual environment sensor. The first environmental feature is obtained by extracting cross-linking features from the real-time environmental information in the set of real-time environmental information that corresponds to the first environmental region. Feature extraction is performed on the real-time environmental information corresponding to the second environmental region in the real-time environmental information set to obtain the second environmental feature; Based on the real-time location of the intelligent agent, a bird's-eye view feature projection is performed on the first environmental feature and the second environmental feature to obtain the bird's-eye view environmental feature. Multi-scale feature extraction is performed on the bird's-eye view environmental features to obtain environmental state features; The real-time state of the agent is mapped to obtain the agent's state features.

7. The behavior simulation method for mobile intelligent agents applied in emergency rescue scenarios according to claim 6, characterized in that, The step of generating behavior planning information for the mobile intelligent agent based on the environmental state characteristics and the agent state characteristics includes: The environmental state features are segmented to obtain a set of sub-environmental state features; Parallel feature encoding is performed on the sub-environmental state features in the sub-environmental state feature set to obtain the encoded environmental state feature set. Perform coarse environmental state classification on each sub-environmental state feature in the encoded environmental state feature set to obtain an environmental classification type set. Generate a weighted feature matrix based on the set of environmental classification types; Based on the weight feature matrix, the environmental state features are weighted to obtain the weighted environmental state features; The behavior planning information is generated based on the weighted environmental state features, the agent state features, and the pre-trained behavior planning model.

8. A behavior simulation device for mobile intelligent agents applied in emergency rescue scenarios, characterized in that, include: The first generation unit is configured to generate an initial simulation environment based on scene environment information and simulation configuration information. The scene environment information represents the regional environment information corresponding to the emergency rescue scene, and the initial simulation environment contains a mobile intelligent agent that has been initialized in position and is to be simulated in behavior. The environmental data local extraction unit is configured to perform local environmental data extraction in the initial simulation environment based on the agent description information corresponding to the mobile agent, and obtain a set of real-time environmental information. The agent description information includes: the real-time location of the agent and sensor specification information. The sensor specification information represents the sensor specifications of the virtual environment sensor corresponding to the mobile agent. The second generation unit is configured to generate environmental state features and agent state features based on the real-time environmental information set and the real-time state of the agent corresponding to the mobile agent. The third generation unit is configured to generate behavior planning information for the mobile intelligent agent based on the environmental state characteristics and the intelligent agent state characteristics. The simulation unit is configured to perform behavioral simulation on the mobile intelligent agent based on the behavior planning information and the real-time location of the intelligent agent.

9. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 7.

10. A computer-readable medium, characterized in that, It stores a computer program thereon, wherein the computer program, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images