Multi-modal bionic perception-based air-ground cooperative control method and system for denial environment

Abstract

The invention relates to a denial environment air-ground cooperative control method and system based on multi-modal bionic perception, belongs to the technical field of unmanned system control, and solves the problems of high complexity and poor reliability of denial environment air-ground cooperative control in the prior art. Comprising the following steps: acquiring unmanned system information, an initial environment map and path information based on task information; wherein the task information comprises a target area which needs to be reached, and the unmanned system information comprises the number of selected unmanned platforms of various types; based on the unmanned system information, dividing each unmanned platform into an unmanned aerial vehicle group and an unmanned vehicle group; wherein the unmanned aerial vehicles and the unmanned vehicles can sense environment information; the environment information comprises a local map and a position; based on the obtained initial environment map and path information, the unmanned aerial vehicle group and the unmanned aerial vehicle group respectively use a bionic formation method to travel from the initial area to the target area; and when the unmanned vehicle group and the unmanned aerial vehicle group both travel to the target area, each unmanned platform blocks and attacks the target.

Description

Technical Field

[0001] The present invention relates to the field of unmanned system control technology, and in particular to an air-ground collaborative control method and system in a denied environment based on multimodal bionic perception. Background Art

[0002] An air-ground collaborative control system primarily refers to a system composed of functionally distinct aerial and ground unmanned systems. This system can achieve a one-plus-one-is-greater-than-two effect through collaborative control and task allocation. In today's rapidly developing information and intelligent world, more requirements are being placed on air-ground collaborative control systems to cope with complex and changing denial environments. For example, in the face of complex and changing denial environments, including communication interference, loss of satellite positioning, and failure-to-kill, it is necessary to build an anti-interference, self-organizing data link network that can effectively cope with communication interference and ensure effective mission execution in complex environments. Traditional air-ground collaborative unmanned systems use a traditional communication link + satellite communication and positioning method, as well as a neural network-based drone-unmanned vehicle joint formation controller. This system is highly complex and does not fully consider the positions of each unmanned platform when forming a formation. This makes it difficult to effectively execute missions in complex and changing denial environments. Summary of the Invention

[0003] In view of the above analysis, the embodiments of the present invention aim to provide a method and system for air-ground collaborative control in a denied environment based on multimodal bionic perception, so as to solve the problems of high complexity and poor reliability of existing air-ground collaborative control in a denied environment.

[0004] On the one hand, an embodiment of the present invention provides an air-ground collaborative control method in a denied environment based on multimodal bionic perception, comprising the following steps:

[0005] Based on the mission information, unmanned system information, initial environment map, and path information are obtained. The mission information includes the target area to be reached, and the unmanned system information includes the number of selected unmanned platforms of each type. The path information includes the starting location and target location, and the types of unmanned platforms include unmanned vehicles and drones.

[0006] Based on the unmanned system information, each unmanned platform is divided into a drone group and an unmanned vehicle group; wherein each drone and unmanned vehicle can perceive environmental information; the environmental information includes a local map and location;

[0007] Based on the acquired initial environment map and path information, the UAV group and the unmanned vehicle group respectively use the bionic formation method to move from the initial area to the target area;

[0008] When both the unmanned vehicle group and the unmanned aerial vehicle group reach the target area, each unmanned platform will surround and attack the target.

[0009] Further, each unmanned platform in the unmanned aerial vehicle group and the unmanned vehicle group is sequentially numbered from 1 to the total number of unmanned platforms in each group; the unmanned platforms travel from an initial area to a target area by adopting a bionic formation method, including:

[0010] S31, constructing a hierarchical formation structure based on the current group and performing initialization configuration; wherein the hierarchical formation structure is constructed based on a controlled superior set of each unmanned platform, and the unmanned platform with an empty controlled superior set is taken as a leading unmanned platform, and the remaining unmanned platforms are follower unmanned platforms;

[0011] S32, obtaining a local map and a position of each unmanned platform at a current time, a historical control input of the leading unmanned platform, and then based on an environment map and path information at a previous time and the hierarchical formation structure, adopting a trained leading platform control prediction model to obtain a control input of the leading unmanned platform at a next time; wherein the control input includes control amounts in each direction of the unmanned platform;

[0012] S33, obtaining actual state quantities of the leading unmanned platform and each follower unmanned platform at the next time based on the control input of the leading unmanned platform at the next time and the controlled superior set of each unmanned platform; wherein the actual state quantities include position quantities and output control quantities of the unmanned platform;

[0013] S34, judging whether the leading unmanned platform reaches a target position, if yes, ending the formation travel; if not, returning to step S32 for continuous execution;

[0014] Further, the initial environment map includes an initial three-dimensional grid global map and an initial two-dimensional grid global map, and the target position includes an unmanned aerial vehicle target position and an unmanned vehicle target position; the local map includes a three-dimensional grid local map and a two-dimensional grid local map; the position in the environment information of the unmanned aerial vehicle is a two-dimensional position in the two-dimensional grid global map, and the position in the environment information of the unmanned vehicle is a three-dimensional position in the three-dimensional grid global map.

[0015] Further, if the current group is an unmanned aerial vehicle group, the unmanned platform is an unmanned aerial vehicle, and the control input includes a horizontal axis control amount, a vertical axis control amount and a height direction control amount; the control input of the leading unmanned aerial vehicle at the next time is obtained by the following way:

[0016] modifying the three-dimensional grid global map at the previous time based on the three-dimensional grid local map and the three-dimensional position of each unmanned aerial vehicle at the current time to obtain the three-dimensional grid global map at the current time;

[0017] The current 3D grid global map, the 3D position and path information of each UAV, and the historical control input and hierarchical formation structure of the leading UAV are preprocessed to obtain 3D global map data, leading UAV status data, following UAV status data, historical control data, hierarchical relationship data, and expected distance data.

[0018] Based on the three-dimensional global map data, the status data of the leading UAV, the status data of the following UAVs, the historical control data, the hierarchical relationship data and the expected distance data, the trained leading UAV control prediction model is used to obtain the control input of the leading UAV at the next moment.

[0019] Furthermore, the leading UAV control prediction model includes:

[0020] A global map feature extraction module, comprising a first convolution structure and a second convolution structure connected in sequence, for extracting spatial accessibility features from input three-dimensional global map data;

[0021] The time series modeling module includes a first long short-term memory network layer, which is used to extract control quantity change features from the input historical control data;

[0022] The hierarchical formation module includes a first feature construction layer, a first graph attention network layer, and a first mean pooling layer, which are connected in sequence. It is used to extract dynamic topological features from the input leading drone status data, following drone status data, hierarchical relationship data, and expected distance data.

[0023] The feature fusion prediction module includes the first splicing layer, the first fully connected layer, the first Dropout layer and the first output layer connected in sequence. It is used to fuse and infer the input spatial accessibility features, control quantity change features, dynamic topology features and leading UAV status data, and output the predicted control input of the leading UAV at the next moment.

[0024] Furthermore, if the current group is an unmanned vehicle group, the unmanned platform is an unmanned vehicle, and the control input includes the horizontal axis control amount and the vertical axis control amount. The control input of the leading unmanned vehicle at the next moment is obtained by the following method:

[0025] Based on the two-dimensional grid local map and two-dimensional position of each unmanned vehicle at the current moment, the two-dimensional grid global map at the previous moment is corrected to obtain the two-dimensional grid global map at the current moment;

[0026] The current 2D grid global map, the 2D position and path information of each unmanned vehicle, and the historical control input and hierarchical formation structure of the leading unmanned vehicle are preprocessed to obtain 2D global map data, leading unmanned vehicle status data, following unmanned vehicle status data, leading unmanned vehicle historical control data, unmanned vehicle hierarchical relationship data, and unmanned vehicle expected distance data;

[0027] Based on two-dimensional global map data, leading unmanned vehicle status data, following unmanned vehicle status data, leading unmanned vehicle historical control data, unmanned vehicle hierarchical relationship data, and unmanned vehicle expected distance data, a trained leading unmanned vehicle control prediction model is used to obtain the control input of the leading unmanned vehicle at the next moment.

[0028] Furthermore, the control prediction model of the leading unmanned vehicle includes:

[0029] The unmanned vehicle global map feature extraction module includes a first convolutional neural network and a second convolutional neural network connected in sequence, and is used to extract traversability features from the input two-dimensional global map data;

[0030] The autonomous vehicle time series modeling module includes a second long short-term memory network layer, which is used to extract control quantity regularity features from the input historical control data of the leading autonomous vehicle;

[0031] The autonomous vehicle hierarchical formation module includes a second feature construction layer, a second graph attention network layer, and a second mean pooling layer, which are connected in sequence. It is used to extract the dynamic topological features of the autonomous vehicles based on the input leading autonomous vehicle status data, the following autonomous vehicle status data, the autonomous vehicle hierarchical relationship data, and the autonomous vehicle expected distance data.

[0032] The unmanned vehicle feature fusion prediction module includes a second splicing layer, a second fully connected layer, a second Dropout layer, and a second output layer connected in sequence. It is used to fuse and infer the input trafficability features, control quantity regularity features, unmanned vehicle dynamic topology features, and leading unmanned vehicle status data, and output the predicted control input of the leading unmanned vehicle at the next moment.

[0033] Furthermore, the actual state of the leading unmanned platform and each following unmanned platform at the next moment is obtained by the following method:

[0034] Based on the control input of the leading unmanned platform at the next moment and the actual state quantity at the current moment, the actual state quantity of the leading unmanned platform at the next moment is obtained;

[0035] Based on the actual state of the leading unmanned platform at the next moment, the formation state of the leading unmanned platform at the next moment is obtained; wherein the formation state includes the position vector and velocity vector of the unmanned platform;

[0036] Based on the formation state of the leading unmanned platform at the next moment, the controlled superior set of each unmanned platform, and the formation state of each following unmanned platform at the current moment, the control input of each following unmanned platform is obtained in sequence;

[0037] Based on the control input of each following unmanned platform and the actual state quantity at the current moment, the actual state quantity of each following unmanned platform at the next moment is obtained, and the formation state quantity of each following unmanned platform at the next moment is obtained.

[0038] Furthermore, if the current group is a drone group, the unmanned platform is a drone, and in the actual state quantity of the drone, the position quantity includes the horizontal coordinate, vertical coordinate and height direction coordinate of the drone, and the output control quantity includes the speed, heading angle and altitude change rate of the drone; if the current group is an unmanned vehicle group, the unmanned platform is an unmanned vehicle, and in the actual state quantity of the unmanned vehicle, the position quantity includes the horizontal coordinate and vertical coordinate of the unmanned vehicle, and the output control quantity includes the speed and steering angle of the unmanned platform.

[0039] On the other hand, the present invention also provides an air-ground collaborative control system for a denied environment based on multimodal bionic perception, comprising:

[0040] An information acquisition module is used to obtain unmanned system information, an initial environment map, and path information based on the mission information; wherein the mission information includes the target area to be reached, the unmanned system information includes the number of selected unmanned platforms of each type; and the path information includes the starting location and the target location;

[0041] A grouping module is used to divide each unmanned platform into a drone group and an unmanned vehicle group based on the unmanned system information; wherein each drone and unmanned vehicle can perceive environmental information; the environmental information includes a local map and location;

[0042] In the formation movement module, based on the acquired initial environment map and path information, the UAV group and the unmanned vehicle group respectively use the bionic formation method to move from the initial area to the target area;

[0043] Collaborative control module: when both the unmanned vehicle group and the unmanned aerial vehicle group move to the target area, each unmanned platform will encircle and attack the target.

[0044] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0045] The present invention provides a method and system for air-ground collaborative control in a denied environment based on multimodal bionic perception. By obtaining unmanned system information, initial environment map, and path information based on mission information, each unmanned platform is divided into an unmanned aerial vehicle group and an unmanned vehicle group. Based on the obtained initial environment map and path information, the unmanned aerial vehicle group and the unmanned vehicle group respectively adopt a bionic formation method to move from the initial area to the target area. When the unmanned vehicle group and the unmanned aerial vehicle group both move to the target area, each unmanned platform encircles and attacks the target, solving the problems of high complexity and poor reliability of existing air-ground collaborative control in a denied environment. In addition, a prediction model is used to fully consider the inherent connection between the position of each unmanned platform and the environmental characteristics for formation. When forming a formation, not only the path planning of the leading unmanned platform is considered, but also the position of each following unmanned platform is considered. The control amount of the leading unmanned platform given can better provide leadership control for the subsequent formation movement, providing a more reliable formation method, so that the unmanned platform can move to the target area. In addition, this method also reduces the complexity of the formation in a denied environment, providing a more effective collaborative control strategy.

[0046] In the present invention, the above-mentioned technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of the present invention will be described in the following description, and some advantages will become apparent from the description or be learned through practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like parts throughout the drawings.

[0048] Figure 1 A flow chart of the air-ground collaborative control method in a denied environment based on multimodal bionic perception provided in Example 1 of the present invention;

[0049] Figure 2 A schematic diagram of a hierarchical formation structure provided in Example 1 of the present invention; DETAILED DESCRIPTION

[0050] The preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, and are not used to limit the scope of the present invention.

[0051] Example 1

[0052] A specific embodiment of the present invention discloses an air-ground collaborative control method in a denial environment based on multimodal bionic perception, such as Figure 1 As shown, the following steps are included:

[0053] S1. Based on the mission information, obtain unmanned system information, initial environment map, and path information; wherein the mission information includes the target area to be reached, the unmanned system information includes the number of selected unmanned platforms of each type; the path information includes the starting location and target location, and the types of unmanned platforms include drones and unmanned vehicles.

[0054] Specifically, drones and unmanned vehicles are equipped with lidar, cameras, and IMUs, which can acquire information about their surroundings through the drones and unmanned vehicles themselves and other drones and unmanned vehicles. It should be noted that the number of drones and unmanned vehicles selected can be determined based on specific needs.

[0055] Specifically, the initial environment map includes an initial 3D grid global map and an initial 2D grid global map. The target locations include the UAV target location and the unmanned vehicle target location. The 3D grid global map is a 3D grid map of the area between the UAV's starting position and the UAV's target location, and the 2D grid global map is a 2D grid map of the area between the UAV's starting position and the unmanned vehicle's target location. In the 3D grid global map and the 2D grid global map, if a 3D grid or a 2D grid is marked as "1", it indicates that it is passable, and if it is marked as "0", it indicates that it is not passable. It should be noted that passable means that the grid can accommodate an unmanned vehicle.

[0056] It should be noted that the unmanned platform can realize the construction of a local three-dimensional grid map according to the set equipment. This process can be realized using existing technology and will not be repeated here.

[0057] Specifically, the target location in the path information is determined according to the target area in the mission information, and can be selected in the target area according to needs; the starting location is set according to the initial area of ​​the parking position of each unmanned platform.

[0058] S2. Based on the unmanned system information, each unmanned platform is divided into a drone group and an unmanned vehicle group; wherein each drone and unmanned vehicle can perceive environmental information; the environmental information includes a local map and location.

[0059] Specifically, the local map includes a three-dimensional grid local map and a two-dimensional grid local map; the position of the unmanned vehicle in the environmental information is the two-dimensional position in the entire two-dimensional grid map, and the position of the drone in the environmental information is the three-dimensional position in the three-dimensional grid global map.

[0060] Specifically, each drone and unmanned vehicle constructs a local map by being equipped with a laser radar and a camera; wherein the size of the local map is set according to actual needs.

[0061] S3. Based on the acquired initial environment map and path information, the UAV group and the unmanned vehicle group respectively use the bionic formation method to move from the initial area to the target area.

[0062] Specifically, the unmanned platforms in the unmanned aerial vehicle group and the unmanned vehicle group are numbered sequentially from 1 to the total number of unmanned platforms in each group.

[0063] During implementation, the bionic formation method is used to move from the initial area to the target area, including:

[0064] S31. Construct a hierarchical formation structure based on the current group and perform initial configuration. The hierarchical formation structure is constructed based on the controlled superior set of each unmanned platform. The unmanned platform with an empty controlled superior set is used as the leading unmanned platform, and the remaining unmanned platforms are followed unmanned platforms.

[0065] It should be noted that the controlled superior set of an unmanned platform is the unmanned platforms that lead and control the unmanned platform.

[0066] Specifically, the initialization configuration includes the initial position, initial speed, initial altitude change rate of each unmanned platform and the expected distance between each follower and its controlled superior.

[0067] Specifically, the controlled superior set is expressed as:

[0068]

[0069] Where N i Represents the controlled superior set of unmanned platform i in the current group.

[0070] For example, taking the current group with 9 unmanned platforms as an example, the hierarchical formation structure is constructed as follows: Figure 2 As shown in the figure, the circles represent unmanned platforms, and the directed edges between the circles represent the leadership relationship between the unmanned platforms.

[0071] It can be understood that in this embodiment, it is expected that the unmanned platform group will fly in a V-shaped formation, and the leadership relationship of the unmanned platforms can play a role in gathering, collision avoidance and stability at a specified distance.

[0072] S32. Obtain the local map and position of each unmanned platform at the current moment, the historical control input of the leading unmanned platform, and then, based on the environmental map and path information at the previous moment and the hierarchical formation structure, use the trained leading platform control prediction model to obtain the control input of the leading unmanned platform at the next moment; wherein, the control input includes the control quantity of each direction of the unmanned platform.

[0073] In specific implementation, if the current group is a drone group, the unmanned platform is a drone, and the control input includes the horizontal axis control amount, the vertical axis control amount, and the altitude direction control amount. The control input of the leading drone at the next moment is obtained by the following method:

[0074] a1. Based on the 3D grid local map and 3D position of each UAV at the current moment, the 3D grid global map at the previous moment is corrected to obtain the 3D grid global map at the current moment.

[0075] Specifically, based on the 3D position of each drone at the current moment, the corresponding 3D grid local map is superimposed on the 3D grid global map at the previous moment. The 3D grid global map at the previous moment is corrected according to the set rules to obtain the 3D grid global map at the current moment.

[0076] More specifically, the set rule is to perform an AND operation on each three-dimensional grid at the same grid position to perform three-dimensional grid correction on the grid position, wherein each three-dimensional grid AND operation indicates that if there is a three-dimensional grid marked as "1", then the three-dimensional grid at the grid position is marked as "1", otherwise it is marked as "0".

[0077] a2. Preprocess the current 3D grid global map, the 3D position and path information of each UAV, and the historical control input and hierarchical formation structure of the leading UAV to obtain 3D global map data, leading UAV status data, following UAV status data, historical control data, hierarchical relationship data, and expected distance data;

[0078] Specifically,

[0079] The three-dimensional global map data is represented as a four-dimensional tensor (depth, height, width, 1). 1 represents a channel, and whether it is passable is represented by binary. If it is "1", it means it is passable, and if it is "0", it means it is not passable.

[0080] The leading UAV status data is one-dimensional data consisting of the three-dimensional position of the leading UAV and the target position of the UAV in the path information;

[0081] The following drone status data is a data sequence consisting of the three-dimensional position of each following drone;

[0082] Historical control data is the time series data consisting of the historical control input of the leading UAV, and its quantity is set according to specific needs;

[0083] Hierarchical relationship data is two-dimensional data generated by the controlled superior sets of each drone in the hierarchical formation structure. The number of rows is the number of drones, and the number of columns is the maximum number of data in each controlled superior set. Each element represents the controlled superior drone of the current drone. If the number of controlled superior drones is insufficient, it is filled with -1. It should be noted that since the number of controlled superior drones of each drone is not the same, the vacant positions in the two-dimensional data are filled with -1.

[0084] The expected distance data is two-dimensional data corresponding to the hierarchical relationship data, wherein each element is the expected distance between the drone and each drone in the controlled upper set.

[0085] a3. Based on the three-dimensional global map data, the leading UAV status data, the following UAV status data, the historical control data, the hierarchical relationship data, and the expected distance data, the trained leading UAV control prediction model is used to obtain the control input of the leading UAV at the next moment.

[0086] Specifically, the leading UAV control prediction model includes:

[0087] A global map feature extraction module, comprising a first convolution structure and a second convolution structure connected in sequence, for extracting spatial accessibility features from input three-dimensional global map data;

[0088] Specifically, the first convolutional structure is used to extract local obstacle features, including the first convolutional layer, batch normalization layer, and maximum pooling layer connected in sequence; the second convolutional structure is used to capture high-level semantic features, including the second convolutional layer and global average pooling layer connected in sequence;

[0089] The time series modeling module includes a first long short-term memory network layer, which is used to extract control quantity change features from the input historical control data;

[0090] The hierarchical formation module includes a first feature construction layer, a first graph attention network layer, and a first mean pooling layer, which are connected in sequence. It is used to extract dynamic topological features from the input leading drone status data, following drone status data, hierarchical relationship data, and expected distance data.

[0091] More specifically, the first feature construction layer constructs the topological features of each drone and then obtains the overall topological features, which are input into the connected graph attention network layer. The topological features of each drone are composed of the drone's three-dimensional position, the relative position difference with each of its controlled superior drones, and the expected distance difference. This can be obtained by using the leading drone's status data, the following drone's status data, the hierarchical relationship data, and the expected distance data.

[0092] The feature fusion prediction module includes the first splicing layer, the first fully connected layer, the first Dropout layer and the first output layer connected in sequence. It is used to fuse and infer the input spatial accessibility features, control quantity change features, dynamic topology features and leading UAV status data, and output the predicted control input of the leading UAV at the next moment.

[0093] Specifically, the training sample data includes feature data and target data, wherein the feature data includes the three-dimensional global map data at the current moment, the leading drone status data, the following drone status data, historical control data, hierarchical relationship data and expected distance data; the target data is the control input of the leading drone at the next moment corresponding to the feature data.

[0094] Specifically, the loss function L during training is expressed as:

[0095] L=L MSE +λ1L ob +λ2L f +λ||W|| 2

[0096] Among them, λ1 and λ2 represent the first and second hyperparameters respectively; L MSE , L ob , L f They represent the mean square error loss, obstacle avoidance penalty and follow-up machine constraint penalty respectively, λ represents the regularization coefficient, and W represents the network weight.

[0097] More specifically, the mean square error loss, the obstacle avoidance penalty term, and the following machine constraint penalty term L MSE , L ob , L f Respectively expressed as:

[0098]

[0099] Where, They represent the predicted horizontal axis control amount, vertical axis control amount and height direction control amount of the leading UAV at the next moment, They respectively represent the horizontal axis control amount, vertical axis control amount and altitude direction control amount of the real leading UAV at the next moment.

[0100]

[0101] In the formula, α represents the penalty coefficient, β represents the attenuation coefficient, represents the distance from the predicted 3D position of the leading UAV at the next moment to the nearest traversable area, and p′ represents the channel value at the predicted 3D position of the leading UAV at the next moment in the 3D grid global map; wherein, is obtained by calculating the Euclidean distance, which is obtained by comparing the 3D position of the leading UAV at the current moment with the control input of the predicted leading UAV at the next moment.

[0102]

[0103] Where D ij represents the actual distance between UAV i and the controlled superior UAV j, d ijrepresents the expected distance between UAV i and its controlled superior UAV j, and || represents the number of elements in the set.

[0104] It can be understood that by constructing sample data based on historical data and training the leading UAV control prediction model, a trained leading UAV control prediction model can be obtained.

[0105] In specific implementation, if the current group is an unmanned vehicle group, the unmanned platform is an unmanned vehicle, and the control input includes the horizontal axis control amount and the vertical axis control amount. The control input of the leading unmanned vehicle at the next moment is obtained through the following method:

[0106] b1. Based on the 2D grid local map and 2D position of each unmanned vehicle at the current moment, the 2D grid global map at the previous moment is corrected to obtain the 2D grid global map at the current moment.

[0107] Specifically, based on the two-dimensional position of each unmanned vehicle at the current moment, the corresponding two-dimensional grid local map is superimposed on the two-dimensional grid global map at the previous moment. The two-dimensional grid global map at the previous moment is corrected according to the set rules to obtain the two-dimensional grid global map at the current moment.

[0108] More specifically, the set rule is that each two-dimensional grid at the same grid position is subjected to an AND operation to perform a two-dimensional grid correction on the grid position, wherein each two-dimensional grid AND operation indicates that if there is a two-dimensional grid marked as "1", then the two-dimensional grid at the grid position is marked as "1", otherwise it is marked as "0".

[0109] b2. Preprocess the current two-dimensional grid global map, the two-dimensional position and path information of each unmanned vehicle, and the historical control input and hierarchical formation structure of the leading unmanned vehicle to obtain two-dimensional global map data, leading unmanned vehicle status data, following unmanned vehicle status data, leading unmanned vehicle historical control data, unmanned vehicle hierarchical relationship data, and unmanned vehicle expected distance data.

[0110] Specifically,

[0111] The two-dimensional global map data is represented as two-dimensional data; wherein the rows and columns represent the map size, and each element represents whether the two-dimensional grid is passable, if it is "1", it means it is passable, and if it is "0", it means it is not passable;

[0112] The leading unmanned vehicle state data is one-dimensional data consisting of the two-dimensional position of the leading unmanned vehicle and the target position of the unmanned vehicle in the path information;

[0113] The state data of the following unmanned vehicles is a data sequence consisting of the two-dimensional positions of each following unmanned vehicle;

[0114] The historical control data of the leading unmanned vehicle is time series data composed of historical control inputs of the leading unmanned vehicle, and the number thereof is set according to specific requirements;

[0115] The hierarchical relationship data of the unmanned vehicle is two-dimensional data generated by the controlled superior set of each unmanned vehicle in the hierarchical formation structure; the number of rows is the number of unmanned vehicles, and the number of columns is the maximum value of the number of data in each controlled superior set; each element represents the controlled superior unmanned vehicle of the current unmanned vehicle, and if insufficient, -1 is supplemented; it should be noted that the number of controlled superiors of each unmanned vehicle is not the same, so -1 is filled in the vacant position in the two-dimensional data;

[0116] The expected distance data of the unmanned vehicle is two-dimensional data corresponding to the hierarchical relationship data, wherein each element is the expected distance between the unmanned vehicle and each unmanned vehicle in the controlled superior set.

[0117] a3, based on the two-dimensional global map data, the leading unmanned vehicle state data, the following unmanned vehicle state data, the leading unmanned vehicle historical control data, the unmanned vehicle hierarchical relationship data and the unmanned vehicle expected distance data, the leading unmanned vehicle control prediction model is trained, and the control input of the leading unmanned vehicle at the next moment is obtained.

[0118] Specifically, the leading unmanned vehicle control prediction model comprises:

[0119] The unmanned vehicle global map feature extraction module comprises a first convolutional neural network and a second convolutional neural network connected in sequence, and is used for performing passability feature extraction on the input two-dimensional global map data;

[0120] The unmanned vehicle time series modeling module comprises a second long short-term memory network layer, and is used for performing control quantity law feature extraction on the input leading unmanned vehicle historical control data;

[0121] The unmanned vehicle hierarchical formation module comprises a second feature construction layer, a second graph attention network layer and a second mean pooling layer connected in sequence, and is used for performing unmanned vehicle dynamic topology feature extraction on the input leading unmanned vehicle state data, following unmanned vehicle state data, unmanned vehicle hierarchical relationship data and unmanned vehicle expected distance data;

[0122] More specifically, the second feature construction layer is used to construct the topology feature of each unmanned vehicle to obtain the global topology feature, which is input into the connected second graph attention network layer; wherein the topology feature of each unmanned vehicle is composed of the two-dimensional position of the unmanned vehicle, the relative position difference and the expected distance difference of each controlled superior unmanned vehicle of the unmanned vehicle, which can be obtained through the leading unmanned vehicle state data, the following unmanned vehicle state data, the unmanned vehicle hierarchical relationship data and the unmanned vehicle expected distance data;

[0123] The unmanned vehicle feature fusion prediction module includes a second splicing layer, a second fully connected layer, a second Dropout layer, and a second output layer connected in sequence. It is used to fuse and infer the input trafficability features, control quantity regularity features, unmanned vehicle dynamic topology features, and leading unmanned vehicle status data, and output the predicted control input of the leading unmanned vehicle at the next moment.

[0124] Specifically, the training sample data includes feature data and target data, wherein the feature data includes the two-dimensional global map data at the current moment, the leading unmanned vehicle status data, the following unmanned vehicle status data, the leading unmanned vehicle historical control data, the unmanned vehicle hierarchical relationship data and the unmanned vehicle expected distance data; the target data is the control input of the leading unmanned vehicle at the next moment corresponding to the feature data.

[0125] Specifically, the loss function L during training is UGV Expressed as:

[0126] L UGV =L MSE,UGV +λ 1,,UGV L ob,UGV +λ 2,UGV L f,UGV

[0127] Among them, λ 1,,UGV ,λ 2,UGV Represent the third and fourth hyperparameters respectively; L MSE,UGV , L ob,UGV , L f,UGV They represent the mean square error loss of the unmanned vehicle, the obstacle avoidance penalty of the leading unmanned vehicle, and the constraint penalty of the following vehicle, respectively.

[0128] More specifically, the mean square error loss of the autonomous vehicle, the obstacle avoidance penalty of the leading autonomous vehicle, and the constraint penalty of the following vehicle L MSE,UGV , L ob,UGV , L f,UGV Respectively expressed as:

[0129]

[0130] Where, They represent the predicted horizontal axis control amount, vertical axis control amount and height direction control amount of the leading unmanned vehicle at the next moment, They respectively represent the horizontal axis control amount, vertical axis control amount and height direction control amount of the actual leading unmanned vehicle at the next moment.

[0131]

[0132] In the formula, α1 represents the penalty coefficient of the unmanned vehicle, β1 represents the attenuation coefficient of the unmanned vehicle, represents the distance from the predicted two-dimensional position of the leading unmanned vehicle at the next time to the nearest passable region, p' UGV represents the numerical value of the two-dimensional position in the two-dimensional grid global map at the predicted next time of the leading unmanned vehicle; wherein the numerical value is obtained by calculating the Euclidean distance, and the three-dimensional position of the leading unmanned vehicle at the current time and the control input of the predicted leading unmanned vehicle at the next time are obtained.

[0133]

[0134] In the formula, D ij,UGV represents the actual distance between the unmanned vehicle i and the controlled superior unmanned vehicle j, d ij,UGV represents the expected distance between the unmanned vehicle i and the controlled superior unmanned vehicle j, and || represents the number of elements in the set.

[0135] It can be understood that the sample data is constructed according to the historical data, the leading unmanned vehicle control prediction model is trained, and the trained leading unmanned vehicle control prediction model is obtained.

[0136] S33, based on the control input of the leading unmanned platform at the next time, the controlled superior set of each unmanned platform, obtaining the actual state quantity of the leading unmanned platform and each following unmanned platform at the next time; wherein the actual state quantity includes the position quantity and the output control quantity of the unmanned platform;

[0137] In specific implementation, the actual state quantity of the leading unmanned platform and each following unmanned platform at the next time is obtained by the following way:

[0138] S331, based on the control input of the leading unmanned platform at the next time and the actual state quantity at the current time, obtaining the actual state quantity of the leading unmanned platform at the next time;

[0139] S332, based on the actual state quantity of the leading unmanned platform at the next time, obtaining the formation state quantity of the leading unmanned platform at the next time; wherein the formation state includes the position vector and the velocity vector of the unmanned platform;

[0140] S333, based on the formation state quantity of the leading unmanned platform at the next time, the controlled superior set of each unmanned platform, and the formation state quantity of each following unmanned platform at the current time, obtaining the control input of each following unmanned platform in turn;

[0141] Specifically, the control input of each following unmanned platform is obtained in turn by the following way:

[0142] The controlled superior set of the unmanned platform i is obtained in turn based on the number sequence of the unmanned platform in the current group, and the following is performed:

[0143] Obtain the next moment's formation state of each unmanned platform in the currently controlled superior set, and then, based on the current moment's formation state of unmanned platform i, obtain the potential function between unmanned platform i and each unmanned platform in its controlled superior set;

[0144] Based on the potential function of unmanned platform i and each unmanned platform in the controlled superior set, the control input of unmanned platform i is obtained;

[0145] Wherein, i is greater than 1.

[0146] S334. Based on the control input of each following unmanned platform and the actual state quantity at the current moment, the actual state quantity of each following unmanned platform at the next moment is obtained, and the formation state quantity of each following unmanned platform at the next moment is obtained.

[0147] Specifically, if the current group is a drone group, the unmanned platform is a drone. Among the actual state quantities of the drone, the position quantity includes the horizontal coordinate, vertical coordinate and altitude direction coordinate of the drone, and the output control quantity includes the speed, heading angle and altitude change rate of the drone.

[0148] More specifically, the actual state of the drone at the next moment is obtained by:

[0149] Based on the control input of the UAV at the next moment and the actual state quantity at the current moment, the control values ​​of the UAV's speed, heading angle and altitude at the next moment are obtained, which are expressed as:

[0150]

[0151] Where, They represent the control values ​​of the speed, heading angle and altitude of UAV i at the next moment, They represent the horizontal axis control value, vertical axis control value and height direction control value of UAV i at the next moment, h i represents the height direction coordinate of UAV i at the current moment, τ v , τ ψ , τ h , τ λ They represent the speed time constant, heading time constant, first altitude time constant and second high speed time constant of the UAV respectively, V i , ψ i ,λ i They represent the speed, heading angle, and altitude change rate of UAV i at the current moment, respectively. If i is 1, it indicates the leading UAV.

[0152] Based on the control values ​​of the speed, heading angle and altitude of the UAV at the next moment, the time derivatives of the speed, heading angle and altitude change rate of the UAV as well as the time derivatives of the horizontal coordinate, vertical coordinate and altitude direction coordinate are obtained, which are expressed as follows:

[0153]

[0154] Where, They represent the horizontal coordinate x of drone i at the current moment. i , vertical coordinate y i and height coordinate h i The time derivative of They represent the time derivatives of the velocity, heading angle and altitude change rate of UAV i at the current moment respectively;

[0155] Based on the time derivatives of the speed, heading angle, and altitude change rate of the UAV, as well as the time derivatives of the horizontal, vertical, and altitude coordinates, the actual state quantity at the current moment, and the constraints, the actual state quantity of the UAV at the next moment is obtained. The constraints are expressed as:

[0156]

[0157] Where, (V i )′、(ψ i )′、(λ i )′ represent the speed, heading angle and altitude change rate of the actual state of UAV i at the next moment, V min 、V max They represent the maximum and minimum speed of the drone respectively, and g represents the acceleration due to gravity, which is 10m / s 2 ,λ min ,λ max Represent the maximum and minimum values ​​of the altitude change rate of the UAV, n max Indicates the maximum permissible lateral overload.

[0158] More specifically, the UAV formation state at the next moment is expressed as:

[0159]

[0160] In the formula, (X i )′、(v i )′ represent the position vector and velocity vector of UAV i at the next moment, (x i )′、(y i )′、(h i )′ respectively represent the horizontal coordinate, vertical coordinate and height coordinate of UAV i at the next moment, (V i )′、(ψ i )′ represent the speed and heading angle of UAV i at the next moment respectively.

[0161] More specifically, the potential function between UAV i and each UAV in its controlled superior set is expressed as:

[0162]

[0163] Where j∈N i ;

[0164] In the formula, || || represents the Euclidean distance, P ij represents the potential function of UAV i and its controlled superior UAV j, X i Represents the current position vector of UAV i, (X j )′ represents the position vector of UAV j at the next moment, d ij represents the expected distance between UAV i and its controlled superior UAV j.

[0165] More specifically, the control input of drone i at the next moment is expressed as:

[0166]

[0167] Where K p Represents the artificial potential field gain factor, which is greater than 0; K h Represents the height feedback gain factor; K v Indicates the speed feedback gain factor, which is greater than 0; Represents the gradient component of the potential function in the horizontal direction; They represent the horizontal axis speed, vertical axis speed and altitude speed of the velocity vector of UAV i at the current moment respectively; They represent the horizontal axis velocity, vertical axis velocity and height direction velocity of the velocity vector of UAV j at the next moment respectively; They represent the horizontal axis speed, vertical axis speed and altitude speed of the leading UAV’s velocity vector at the next moment respectively; Represents the coordinate of the current position vector of UAV i in the height direction; Represents the coordinate of the next moment position vector of UAV j in the height direction; m i represents the mass of UAV i; ω1 and ω2 represent the adjustment factors of horizontal control and height direction control respectively; k i represents the velocity damping factor of UAV i.

[0168] Specifically, if the current group is an unmanned vehicle group, the unmanned platform is an unmanned vehicle. In the actual state quantity of the unmanned vehicle, the position quantity includes the horizontal coordinate and vertical coordinate of the unmanned vehicle, and the output control quantity includes the speed and steering angle of the unmanned platform.

[0169] More specifically, the actual state of the autonomous vehicle at the next moment is obtained by:

[0170] Based on the control input of the unmanned vehicle at the next moment and the actual state quantity at the current moment, the control values ​​of the speed and steering angle of the unmanned vehicle at the next moment are obtained, which are expressed as:

[0171]

[0172] Where, They represent the speed and steering angle control values ​​of the unmanned vehicle i at the next moment, They represent the horizontal and vertical control quantities of the unmanned vehicle i at the next moment respectively; if i is 1, it represents the leading unmanned vehicle.

[0173] Based on the control values ​​of the speed and steering angle of the unmanned vehicle at the next moment, the time derivatives of the speed and steering angle of the unmanned vehicle and the time derivatives of the horizontal and vertical coordinates are obtained, which are expressed as:

[0174]

[0175] Where, Represents the horizontal coordinates of the unmanned vehicle i at the current moment vertical axis The time derivative of They represent the current speed and time derivative of the steering angle of the unmanned vehicle i, They represent the speed and steering angle of the unmanned vehicle i at the current moment, τ v,UGV , τ ψUGV They represent the speed time constant and steering angle time constant of the unmanned vehicle respectively;

[0176] Based on the speed of the unmanned vehicle, the time derivative of the steering angle, the time derivative of the horizontal and vertical coordinates, the actual state quantity at the current moment and the constraints, the actual state quantity of the unmanned vehicle at the next moment is obtained. The constraints are expressed as:

[0177]

[0178] Where, They represent the speed and steering angle of the actual state of the unmanned vehicle i at the next moment, V min,UGV 、V max,UGV They represent the maximum and minimum speeds of the unmanned vehicle, ψ max,UGV Indicates the maximum permissible lateral overload of the unmanned vehicle.

[0179] More specifically, the state of the unmanned vehicle formation at the next moment is expressed as:

[0180]

[0181] Where, They represent the position vector and velocity vector of the unmanned vehicle i at the next moment, They represent the horizontal and vertical coordinates of the unmanned vehicle i at the next moment, They represent the speed and steering angle of the autonomous vehicle i at the next moment respectively.

[0182] More specifically, the potential function between an autonomous vehicle i and each autonomous vehicle in its controlled superior set is expressed as:

[0183]

[0184] Where j∈N i ;

[0185] In the formula, || || represents the Euclidean distance, represents the potential function of the unmanned vehicle i and its controlled superior unmanned vehicle j, represents the current position vector of the unmanned vehicle i, represents the position vector of the autonomous vehicle j at the next moment, d ij,UGV represents the expected distance between unmanned vehicle i and its controlled superior unmanned vehicle j.

[0186] More specifically, the control input of autonomous vehicle i at the next moment is expressed as:

[0187]

[0188] Where K p Represents the artificial potential field gain factor, which is greater than 0; K v Indicates the speed feedback gain factor, which is greater than 0; represents the gradient component of the potential function; They represent the horizontal axis speed and vertical axis speed of the velocity vector of the unmanned vehicle i at the current moment respectively; They represent the horizontal axis speed and vertical axis speed of the velocity vector of the unmanned vehicle j at the next moment respectively; ω1 represents the horizontal control amount adjustment factor; represents the velocity damping factor of unmanned vehicle i.

[0189] S34. Determine whether the leading unmanned platform has reached the target position. If it has, end the formation movement; if not, return to step S32 and continue execution.

[0190] Specifically, whether the target position has been reached is determined based on the current position of the leading unmanned platform. Preferably, the target position can be set separately for unmanned vehicles and drones.

[0191] S4. When both the unmanned vehicle group and the unmanned aerial vehicle group reach the target area, each unmanned platform will encircle and attack the target.

[0192] Specifically, when the leading unmanned vehicle and the leading drone arrive at the target location, it is considered that both the unmanned vehicle group and the drone group have moved to the target area.

[0193] Specifically, when the unmanned vehicle group and the unmanned aerial vehicle group both travel to the target area, the grouping is cancelled, and the existing cooperative strategy is adopted to surround and attack the target in the target area. Exemplarily, a bionic wolf pack strategy is adopted.

[0194] Compared with the prior art, the embodiment provides a denial environment air-ground cooperative control method and system based on multi-modal bionic perception. By acquiring unmanned system information, an initial environment map and path information based on task information, each unmanned platform is divided into an unmanned aerial vehicle group and an unmanned vehicle group. Based on the acquired initial environment map and path information, the unmanned aerial vehicle group and the unmanned vehicle group respectively adopt a bionic formation method to travel from an initial area to a target area. When the unmanned vehicle group and the unmanned aerial vehicle group both travel to the target area, each unmanned platform surrounds and attacks the target. The problems of high complexity and poor reliability of existing denial environment air-ground cooperative control are solved. By using a prediction model, the internal relationship between the positions of each unmanned platform and the environmental characteristics is comprehensively considered for formation. When forming, not only the path planning of the leading unmanned platform is considered, but also the positions of each following unmanned platform are considered. The control amount of the leading unmanned platform can better provide leadership control for subsequent formation travel, a more reliable formation method is provided, the unmanned platform travels to the target area, and the complexity of formation in the denial environment is also reduced, and a more effective cooperative control strategy is provided.

[0195] Embodiment 2

[0196] The embodiment provides a specific embodiment, which discloses a denial environment air-ground cooperative control system based on multi-modal bionic perception, comprising:

[0197] An information acquisition module is configured to acquire unmanned system information, an initial environment map and path information based on task information. The task information includes a target area to be reached, and the unmanned system information includes the number of selected unmanned platforms of each type. The path information includes a departure position and a target position.

[0198] A grouping module is configured to divide each unmanned platform into an unmanned aerial vehicle group and an unmanned vehicle group based on the unmanned system information. Each unmanned aerial vehicle and unmanned vehicle can perceive environmental information. The environmental information includes a local map and a position.

[0199] A formation travel module is configured to make the unmanned aerial vehicle group and the unmanned vehicle group respectively adopt a bionic formation method to travel from an initial area to a target area based on the acquired initial environment map and path information.

[0200] A cooperative control module is configured to make each unmanned platform surround and attack the target when the unmanned vehicle group and the unmanned aerial vehicle group both travel to the target area.

[0201] The specific implementation process of the embodiment of the present invention can be found in the above method embodiment, and this embodiment will not be repeated here.

[0202] Since the principles of this embodiment are the same as those of the above method embodiment, the present device also has the corresponding technical effects of the above method embodiment.

[0203] Those skilled in the art will appreciate that all or part of the process steps of the above-described embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, such as a magnetic disk, an optical disk, a read-only memory, or a random access memory.

[0204] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for air-ground collaborative control in a denial environment based on multimodal bionic perception, characterized in that: The following steps are involved: Based on the mission information, unmanned system information, initial environment map, and path information are obtained. The mission information includes the target area to be reached, and the unmanned system information includes the number of selected unmanned platforms of each type. The path information includes the starting location and target location, and the types of unmanned platforms include unmanned vehicles and drones. Based on the unmanned system information, each unmanned platform is divided into a drone group and an unmanned vehicle group; wherein each drone and unmanned vehicle can perceive environmental information; the environmental information includes a local map and location; Based on the acquired initial environment map and path information, the UAV group and the unmanned vehicle group respectively use the bionic formation method to move from the initial area to the target area; When both the unmanned vehicle group and the unmanned aerial vehicle group reach the target area, each unmanned platform will surround and attack the target.

2. The air-ground collaborative control method in a denial environment based on multimodal bionic perception according to claim 1 is characterized in that: The unmanned platforms in the unmanned aerial vehicle group and the unmanned vehicle group are numbered sequentially from 1 to the total number of unmanned platforms in each group; The bionic formation method is used to move from the initial area to the target area, including: S31. Construct a hierarchical formation structure based on the current group and perform initial configuration. The hierarchical formation structure is constructed based on the controlled parent set of each unmanned platform. The unmanned platform with an empty controlled parent set is designated as the leading unmanned platform, and the remaining unmanned platforms are designated as follower unmanned platforms. S32. Obtain the local map and position of each unmanned platform at the current moment, the historical control input of the leading unmanned platform, and then, based on the environmental map and path information at the previous moment and the hierarchical formation structure, use the trained control prediction model of the leading platform to obtain the control input of the leading unmanned platform at the next moment; wherein the control input includes the control variables of each direction of the unmanned platform; S33. Based on the control input of the leading unmanned platform at the next moment and the controlled superior set of each unmanned platform, obtain the actual state of the leading unmanned platform and each following unmanned platform at the next moment; wherein the actual state includes the position and output control of the unmanned platform; S34. Determine whether the leading unmanned platform has reached the target position. If it has, end the formation movement; if not, return to step S32 and continue execution.

3. The air-ground collaborative control method in a denial environment based on multimodal bionic perception according to claim 2 is characterized in that: The initial environmental map includes an initial three-dimensional grid global map and an initial two-dimensional grid global map, the target position includes the drone target position and the unmanned vehicle target position; the local map includes a three-dimensional grid local map and a two-dimensional grid local map; the position of the drone in the environmental information is the two-dimensional position in the entire two-dimensional grid map, and the position of the drone in the environmental information is the three-dimensional position in the three-dimensional grid global map.

4. The air-ground collaborative control method in a denial environment based on multimodal bionic perception according to claim 3 is characterized in that: If the current group is a drone group, the unmanned platform is a drone, and the control input includes the horizontal axis control amount, the vertical axis control amount, and the altitude direction control amount. The control input of the leading drone at the next moment is obtained by the following method: Based on the 3D grid local map and 3D position of each UAV at the current moment, the 3D grid global map at the previous moment is corrected to obtain the 3D grid global map at the current moment; The current 3D grid global map, the 3D position and path information of each UAV, and the historical control input and hierarchical formation structure of the leading UAV are preprocessed to obtain 3D global map data, leading UAV status data, following UAV status data, historical control data, hierarchical relationship data, and expected distance data. Based on the three-dimensional global map data, the status data of the leading UAV, the status data of the following UAVs, the historical control data, the hierarchical relationship data and the expected distance data, the trained leading UAV control prediction model is used to obtain the control input of the leading UAV at the next moment.

5. The air-ground collaborative control method in a denial environment based on multimodal bionic perception according to claim 4 is characterized in that: The leading UAV control prediction model includes: A global map feature extraction module, comprising a first convolution structure and a second convolution structure connected in sequence, for extracting spatial accessibility features from input three-dimensional global map data; The time series modeling module includes a first long short-term memory network layer, which is used to extract control quantity change features from the input historical control data; The hierarchical formation module includes a first feature construction layer, a first graph attention network layer, and a first mean pooling layer, which are connected in sequence. It is used to extract dynamic topological features from the input leading drone status data, following drone status data, hierarchical relationship data, and expected distance data. The feature fusion prediction module includes the first splicing layer, the first fully connected layer, the first Dropout layer and the first output layer connected in sequence. It is used to fuse and infer the input spatial accessibility features, control quantity change features, dynamic topology features and leading UAV status data, and output the predicted control input of the leading UAV at the next moment.

6. The air-ground collaborative control method in a denial environment based on multimodal bionic perception according to claim 3 is characterized in that: If the current group is an unmanned vehicle group, the unmanned platform is an unmanned vehicle, and the control input includes the horizontal axis control amount and the vertical axis control amount; The control input of the leading unmanned vehicle at the next moment is obtained in the following way: Based on the two-dimensional grid local map and two-dimensional position of each unmanned vehicle at the current moment, the two-dimensional grid global map at the previous moment is corrected to obtain the two-dimensional grid global map at the current moment; The current 2D grid global map, the 2D position and path information of each unmanned vehicle, and the historical control input and hierarchical formation structure of the leading unmanned vehicle are preprocessed to obtain 2D global map data, leading unmanned vehicle status data, following unmanned vehicle status data, leading unmanned vehicle historical control data, unmanned vehicle hierarchical relationship data, and unmanned vehicle expected distance data; Based on two-dimensional global map data, leading unmanned vehicle status data, following unmanned vehicle status data, leading unmanned vehicle historical control data, unmanned vehicle hierarchical relationship data, and unmanned vehicle expected distance data, a trained leading unmanned vehicle control prediction model is used to obtain the control input of the leading unmanned vehicle at the next moment.

7. The air-ground collaborative control method in a denial environment based on multimodal bionic perception according to claim 6 is characterized in that: The leading unmanned vehicle control prediction model includes: The unmanned vehicle global map feature extraction module includes a first convolutional neural network and a second convolutional neural network connected in sequence, and is used to extract traversability features from the input two-dimensional global map data; The autonomous vehicle time series modeling module includes a second long short-term memory network layer, which is used to extract control quantity regularity features from the input historical control data of the leading autonomous vehicle; The autonomous vehicle hierarchical formation module includes a second feature construction layer, a second graph attention network layer, and a second mean pooling layer, which are connected in sequence. It is used to extract the dynamic topological features of the autonomous vehicles based on the input leading autonomous vehicle status data, the following autonomous vehicle status data, the autonomous vehicle hierarchical relationship data, and the autonomous vehicle expected distance data. The unmanned vehicle feature fusion prediction module includes a second splicing layer, a second fully connected layer, a second Dropout layer, and a second output layer connected in sequence. It is used to fuse and infer the input trafficability features, control quantity regularity features, unmanned vehicle dynamic topology features, and leading unmanned vehicle status data, and output the predicted control input of the leading unmanned vehicle at the next moment.

8. The air-ground collaborative control method in a denial environment based on multimodal bionic perception according to claim 2 is characterized in that: The actual state of the leading unmanned platform and each following unmanned platform at the next moment is obtained by the following method: Based on the control input of the leading unmanned platform at the next moment and the actual state quantity at the current moment, the actual state quantity of the leading unmanned platform at the next moment is obtained; Based on the actual state of the leading unmanned platform at the next moment, the formation state of the leading unmanned platform at the next moment is obtained; wherein the formation state includes the position vector and velocity vector of the unmanned platform; Based on the formation state of the leading unmanned platform at the next moment, the controlled superior set of each unmanned platform, and the formation state of each following unmanned platform at the current moment, the control input of each following unmanned platform is obtained in sequence; Based on the control input of each following unmanned platform and the actual state quantity at the current moment, the actual state quantity of each following unmanned platform at the next moment is obtained, and the formation state quantity of each following unmanned platform at the next moment is obtained.

9. The air-ground collaborative control method in a denial environment based on multimodal bionic perception according to claim 8 is characterized in that: If the current group is a drone group, the unmanned platform is a drone. Among the actual state quantities of the drone, the position quantity includes the horizontal coordinate, vertical coordinate and altitude coordinate of the drone, and the output control quantity includes the speed, heading angle and altitude change rate of the drone; if the current group is an unmanned vehicle group, the unmanned platform is an unmanned vehicle. Among the actual state quantities of the unmanned vehicle, the position quantity includes the horizontal coordinate and vertical coordinate of the unmanned vehicle, and the output control quantity includes the speed and steering angle of the unmanned platform.

10. An air-ground collaborative control system for a denied environment based on multimodal bionic perception, characterized in that: include: An information acquisition module is used to obtain unmanned system information, an initial environment map, and path information based on the mission information; wherein the mission information includes the target area to be reached, the unmanned system information includes the number of selected unmanned platforms of each type; and the path information includes the starting location and the target location; A grouping module is used to divide each unmanned platform into a drone group and an unmanned vehicle group based on the unmanned system information; wherein each drone and unmanned vehicle can perceive environmental information; the environmental information includes a local map and location; In the formation movement module, based on the acquired initial environment map and path information, the UAV group and the unmanned vehicle group respectively use the bionic formation method to move from the initial area to the target area; Collaborative control module: when both the unmanned vehicle group and the unmanned aerial vehicle group move to the target area, each unmanned platform will encircle and attack the target.

Images