Multi-unmanned aerial vehicle path planning method and system based on cooperative task

Abstract

The invention provides a multi-unmanned aerial vehicle path planning method and system based on cooperative tasks, and relates to the technical field of unmanned aerial vehicle path planning. The method comprises the following steps: firstly, determining unmanned aerial vehicle global data, and performing embedding processing on the unmanned aerial vehicle global data to form an unmanned aerial vehicle embedding vector; secondly, performing deep mining on the unmanned aerial vehicle embedded vector to form an unmanned aerial vehicle depth vector; then, a target position of each unmanned aerial vehicle is generated based on the unmanned aerial vehicle depth vector and the work area vectors of the uncompleted work areas of the plurality of unmanned aerial vehicles. Based on the method, the problem of relatively low reliability of multi-unmanned aerial vehicle path planning in the prior art can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle path planning, and in particular, to a multi-unmanned aerial vehicle path planning method and system based on collaborative tasks. Background Art

[0002] The goal of multi-unmanned aerial vehicle collaborative path planning is usually to efficiently complete tasks in the target area through multiple unmanned aerial vehicles, such as crop growth monitoring, precise spraying of pesticides or fertilizers, and soil analysis. Each unmanned aerial vehicle needs to perform tasks in a complex environment and cooperate with other unmanned aerial vehicles to ensure the efficient and conflict-free completion of tasks. However, in the prior art, generally, the corresponding staff allocates the operation areas of each unmanned aerial vehicle according to experience, and then each unmanned aerial vehicle performs corresponding operations in the allocated corresponding areas. However, through the research of the inventor, it is found that on the one hand, it is limited by the limitations of the staff's experience, and on the other hand, since the unmanned aerial vehicle will be interfered by other factors during the operation process, therefore, performing operations according to the path planning formed before operation may be difficult to meet the actual operation requirements. That is to say, in the prior art, there is a problem that the reliability of multi-unmanned aerial vehicle path planning is relatively low. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a multi-unmanned aerial vehicle path planning method and system based on collaborative tasks to improve the problem of relatively low reliability of multi-unmanned aerial vehicle path planning existing in the prior art.

[0004] To achieve the above purpose, the embodiments of the present invention adopt the following technical solutions: A multi-unmanned aerial vehicle path planning method based on collaborative tasks, comprising: Determine the global data of the unmanned aerial vehicle, and perform embedding processing on the global data of the unmanned aerial vehicle to form an unmanned aerial vehicle embedding vector, wherein the global data of the unmanned aerial vehicle is at least used to reflect the current position, current power, and current material reserve of each unmanned aerial vehicle among multiple unmanned aerial vehicles during the operation process, and the unmanned aerial vehicle embedding vector is used to reflect the shallow semantic information of the global data of the unmanned aerial vehicle; Perform in-depth mining on the unmanned aerial vehicle embedding vector to form an unmanned aerial vehicle depth vector, wherein the unmanned aerial vehicle depth vector is used to reflect the deep semantic information of the global data of the unmanned aerial vehicle; Generate the target position of each unmanned aerial vehicle based on the unmanned aerial vehicle depth vector and the operation area vectors of the uncompleted operation areas of the multiple unmanned aerial vehicles, wherein each target position belongs to the uncompleted operation area, and the target position is used as the movement target for the unmanned aerial vehicle to move from the current position to complete the path planning, and the unmanned aerial vehicle completes the corresponding operation tasks based on the stored materials during the process of moving from the current position to the target position.

[0005] In some preferred embodiments, in the above-mentioned multi-UAV path planning method based on collaborative tasks, the step of deeply mining the UAV embedding vector to form a UAV deep vector includes: In the first mapping subunit included in the semantic encoding unit, perform a first mapping operation on the UAV embedding vector to form an initial deep vector, and perform a second mapping operation on the initial deep vector to form an intermediate deep vector, and perform a third mapping operation on the intermediate deep vector to form the encoding vector of the first mapping subunit, where the first mapping operation, the second mapping operation, and the third mapping operation are different from each other; Load the encoding vector of the previous mapping subunit into the next mapping subunit for multi-level mapping operations until the encoding vector of the last mapping subunit is formed, and use this encoding vector as the UAV encoding vector of the semantic encoding unit, where the semantic encoding unit is configured with multiple successively connected mapping subunits; Use the semantic decoding unit to perform a semantic decoding operation on the UAV encoding vector of the semantic encoding unit to form a UAV decoding vector, and based on the UAV decoding vector, obtain a UAV deep vector, where the semantic decoding unit is configured with multiple successively connected mapping subunits.

[0006] In some preferred embodiments, in the above-mentioned multi-UAV path planning method based on collaborative tasks, the step of performing a first mapping operation on the UAV embedding vector to form an initial deep vector includes: Map the vector parameters in the UAV embedding vector to a target interval to form a UAV normalized vector, where the lower limit value of the target interval is equal to 0 and the upper limit value is equal to 1; Perform a first mapping operation on the UAV normalized vector to form a first mapping vector, where the first mapping operation includes multiple mapping stages, and a skip connection link is configured between adjacent two mapping stages, and a transmission control parameter is set on this link, and the transmission control parameter is used to reflect whether the output of the previous mapping stage is connected to the output of the next mapping stage; Perform a fully connected operation on the first mapping vector to form an initial deep vector.

[0007] In some preferred embodiments, in the above-mentioned multi-UAV path planning method based on collaborative tasks, the step of performing a first mapping operation on the UAV normalized vector to form a first mapping vector includes: Perform a sliding window process on the UAV normalized vector to form multiple UAV sliding window vectors; In the first mapping stage, perform self-attention operation on the first drone sliding window vector to form a corresponding drone attention vector; In the second and each subsequent mapping stage, based on the drone attention vector of the previous mapping stage, perform cross-attention operation on the drone sliding window vector corresponding to the current mapping stage to form a corresponding drone cross vector; Based on the similarity between the drone cross vector corresponding to the current mapping stage and the drone attention vector of the previous mapping stage, determine the transmission control parameter of the current stage; When the transmission control parameter is equal to 0, perform bitwise addition operation on the drone attention vector of the previous mapping stage and the drone cross vector of the current mapping stage to form the drone attention vector corresponding to the current stage; when the transmission control parameter is equal to 1, determine the drone cross vector of the current mapping stage as the drone attention vector corresponding to the current stage; Take the drone attention vector corresponding to the last mapping stage as the first mapping vector.

[0008] In some preferred embodiments, in the above multi-drone path planning method based on collaborative tasks, the step of determining the transmission control parameter of the current stage based on the similarity between the drone cross vector corresponding to the current mapping stage and the drone attention vector of the previous mapping stage includes: Determine the similarity between the drone cross vector corresponding to the current mapping stage and the drone attention vector of the previous mapping stage, and determine the quantity ratio between the transmission control parameters equal to 0 and the transmission control parameters equal to 1 in the transmission control parameters of each previous mapping stage; Based on the quantity ratio, update the similarity to form an updated similarity; Perform a rounding operation on the updated similarity to form the transmission control parameter of the current stage.

[0009] In some preferred embodiments, in the above multi-drone path planning method based on collaborative tasks, the step of performing a second mapping operation on the initial depth vector to form an intermediate depth vector includes: Extract a first feature vector and a second feature vector from the initial depth vector; Perform a non-linear activation operation on the first feature vector to form a non-linear activation vector, where each vector parameter in the non-linear activation vector is less than or equal to 1 and greater than or equal to 0; Perform a bitwise multiplication operation on the non-linear activation vector and the second feature vector, and perform a fully connected operation on the result of the bitwise multiplication operation to form an intermediate depth vector.

[0010] In some preferred embodiments, in the above-mentioned multi-UAV path planning method based on collaborative tasks, the step of extracting the first characterization vector and the second characterization vector from the initial depth vector includes: Map the vector parameters in the initial depth vector to a target interval to form a normalized depth vector, where the lower limit value of the target interval is equal to 0 and the upper limit value is equal to 1; Perform convolution operations on the normalized depth vector through two convolutional network layers respectively to form a first characterization vector and a second characterization vector with the same size, where the two convolutional network layers have the same convolutional kernel architecture and different convolutional kernel parameters.

[0011] In some preferred embodiments, in the above-mentioned multi-UAV path planning method based on collaborative tasks, the step of performing a third mapping operation on the intermediate depth vector to form the encoded vector of the first mapping sub-unit includes: Perform a normalization operation on the intermediate depth vector to form an intermediate normalized vector; Perform pooling operations on the intermediate normalized vector through two pooling network layers respectively to form a first pooling vector and a second pooling vector with the same size, where the two pooling network layers have the same pooling window and pooling stride and different pooling methods; Perform a non-linear activation operation on the first pooling vector, and based on the result of the non-linear activation operation, perform a bitwise multiplication operation on the second pooling vector to form the encoded vector of the first mapping sub-unit; Wherein, in the semantic decoding unit, each mapping sub-unit is used to perform a first mapping operation, a second mapping operation and a third mapping operation in sequence, and the first mapping operation and the second mapping operation have the same mapping method as the first mapping operation and the second mapping operation in the semantic encoding unit, and the third mapping operation performs a fully connected operation on the result of the bitwise multiplication operation on the basis of the third mapping operation in the semantic encoding unit.

[0012] In some preferred embodiments, in the above-mentioned multi-UAV path planning method based on collaborative tasks, the step of generating the target position of each UAV based on the UAV depth vector and the operation area vectors of the uncompleted operation areas of the multiple UAVs includes: Determine the operation area vectors of the uncompleted operation areas of the multiple UAVs, where in the operation area vector, the positions represented by the coordinates corresponding to the parameters equal to 1 belong to the uncompleted operation areas, and the positions represented by the coordinates corresponding to the parameters equal to 0 belong to the completed operation areas; Fuse the UAV depth vector and the operation area vector to form a UAV operation vector; For the first drone, based on the mapping of the drone operation vector, the corresponding position probability distribution is obtained, and based on this position probability distribution, the target position corresponding to the drone is determined, where the position probability distribution is used to reflect the probability that each position belongs to the target position of the drone; For the second and each subsequent drone, based on the embedding vector of the target position corresponding to the previous drone and the output vector corresponding to the previous drone, the output vector corresponding to the current drone is determined, and based on this output vector, the corresponding position probability distribution is mapped, and based on this position probability distribution, the target position corresponding to the drone is determined, where the output vector corresponding to the first drone is the drone operation vector.

[0013] An embodiment of the present invention further provides a multi-drone path planning system based on a collaborative task, including a processor and a memory, where the memory is used to store a computer program, and the processor is used to execute the computer program to implement the above-mentioned multi-drone path planning method based on a collaborative task.

[0014] The multi-drone path planning method and system based on a collaborative task provided by the embodiment of the present invention, first, determine the global drone data, and perform embedding processing on the global drone data to form a drone embedding vector; secondly, deeply mine the drone embedding vector to form a drone depth vector; then, based on the drone depth vector and the operation area vectors of the uncompleted operation areas of multiple drones, generate the target position of each drone. Based on the above, on the one hand, due to the potential semantic information in the global drone data can be obtained through embedding processing and deep mining, compared with the conventional operations of artificial personnel based on experience, more reliable path planning can be performed (the basis for path planning can be more reliable and rich). On the other hand, since the global drone data is used to reflect the current position, current power, and current material reserve of the drone during the operation process, that is, the dynamic information during the operation process is fully considered. For example, the current power and current material reserve will affect whether the drone can perform effective operations, so that the determined target position is the position where the drone can actually perform effective operations, ensuring the reliability of the path planning, and thus improving the problem of relatively low reliability of multi-drone path planning existing in the prior art.

[0015] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following specifically enumerates preferred embodiments and, in conjunction with the accompanying drawings, makes a detailed description as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a structural block diagram of a multi-drone path planning system based on a collaborative task provided by an embodiment of the present invention.

[0017] Figure 2 A schematic flowchart of the steps included in the multi-UAV path planning method based on collaborative tasks provided by the embodiments of the present invention.

[0018] Figure 3 A schematic diagram of in-depth mining provided by the embodiments of the present invention.

[0019] Figure 4 A schematic diagram of the first mapping operation provided by the embodiments of the present invention.

[0020] Figure 5 Another schematic diagram of the first mapping operation provided by the embodiments of the present invention.

[0021] Figure 6 A schematic diagram of the operation area provided by the embodiments of the present invention. Detailed implementation manners

[0022] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Usually, the components of the embodiments of the present invention described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed present invention, but merely represents the selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts fall within the scope of protection of the present invention.

[0024] As Figure 1 shown, the embodiments of the present invention provide a multi-UAV path planning system based on collaborative tasks. Among them, the multi-UAV path planning system based on collaborative tasks may include a memory and a processor.

[0025] Specifically, the memory and the processor are electrically connected directly or indirectly to realize data transmission or interaction. For example, they can be electrically connected through one or more communication buses or signal lines. The memory may store at least one software functional module (computer program) that can exist in the form of software or firmware. The processor can be used to execute the executable computer program stored in the memory, so as to implement the multi-UAV path planning method based on collaborative tasks provided by the embodiments of the present invention (as described later).

[0026] That is to say, the multi-UAV path planning system based on collaborative tasks can be used for: Determine the global UAV data, and perform embedding processing on the global UAV data to form a UAV embedding vector. Among them, the global UAV data is at least used to reflect the current position, current power, and current material reserve of each UAV among multiple UAVs during the operation process, and the UAV embedding vector is used to reflect the shallow semantic information of the global UAV data; Deeply mine the UAV embedding vector to form a UAV deep vector, where the UAV deep vector is used to reflect the deep semantic information of the global UAV data; Based on the UAV deep vector and the operation area vector of the unfinished operation area of the multiple UAVs, generate the target position of each UAV. Each target position belongs to the unfinished operation area, and the target position is used as the movement target for the UAV to start moving from the current position to complete the path planning. Moreover, the UAV completes the corresponding operation tasks based on the stored materials during the process of moving from the current position to the target position.

[0027] Optionally, the memory can be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electric Erasable Programmable Read-Only Memory (EEPROM), etc.

[0028] Optionally, the processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), a System on Chip (SoC), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0029] And, Figure 1 The structure shown is only for illustration. The multi-UAV path planning system based on collaborative tasks may also include more or fewer components than those shown, or have the same as Figure 1 shown more or less components, or have the same asFigure 1 The different configurations shown, for example, may include a communication unit for information interaction with other devices (such as drones and other devices) to obtain the current position, current battery level, and current material reserve of the drone during operation, etc.

[0030] Among them, in an alternative example, the multi - drone path planning system based on collaborative tasks may be a server or a server cluster with data - processing capabilities.

[0031] Combined with Figure 2 , the embodiment of the present invention also provides a multi - drone path planning method based on collaborative tasks, which can be applied to the above - mentioned multi - drone path planning system based on collaborative tasks. Among them, the method steps defined by the process related to the multi - drone path planning method based on collaborative tasks can be implemented by the multi - drone path planning system based on collaborative tasks (hereinafter simply referred to as the path planning system). The following will Figure 2 elaborate in detail on the specific process shown.

[0032] Step S110: Determine the global data of the drone, and perform embedding processing on the global data of the drone to form a drone embedding vector.

[0033] In the embodiment of the present invention, the path planning system can determine the global data of the drone and perform embedding processing on the global data of the drone to form a drone embedding vector. Among them, the global data of the drone is at least used to reflect the current position (such as longitude, latitude, and altitude), current battery level (i.e., the remaining battery power of the battery), and current material reserve (such as the remaining fertilizer or pesticide, etc.) of each drone among multiple drones during operation. The drone embedding vector is used to reflect the shallow semantic information of the global data of the drone. That is to say, through preliminary embedding processing, the global data of the drone can be embedded into the vector space for representation.

[0034] Step S120: Perform in - depth mining on the drone embedding vector to form a drone deep vector.

[0035] In the embodiment of the present invention, after forming the drone embedding vector, the path planning system can perform in - depth mining on the drone embedding vector to form a drone deep vector. Among them, the drone deep vector is used to reflect the deep semantic information of the global data of the drone. That is to say, since the drone embedding vector belongs to the shallow semantic information formed through embedding processing, therefore, in order to further improve the representation accuracy of semantic information, further in - depth mining can be performed to obtain high - level and abstract semantic information.

[0036] Step S130: Generate the target position of each drone based on the drone depth vector and the operation area vectors of the uncompleted operation areas of the multiple drones.

[0037] In an embodiment of the present invention, after forming the drone depth vector, the path planning system may generate the target position of each drone based on the drone depth vector and the operation area vectors of the uncompleted operation areas of the multiple drones. Among them, each target position belongs to the uncompleted operation area, and the target position serves as the movement target for the drone to start moving from the current position to complete the path planning. Moreover, the drone completes the corresponding operation tasks based on the stored materials during the process of moving from the current position to the target position. That is to say, when performing path planning, not only the current position, current power, and current material reserve of the drone need to be considered, but also the actual uncompleted operation area, that is, effectively constrain the target position of the drone, thereby improving the reliability of the determined target position.

[0038] Based on the above content, on the one hand, since potential semantic information in the global drone data can be obtained through embedding processing and in-depth mining, compared with the conventional operations of manual personnel based on experience, more reliable path planning can be performed (the basis for path planning can be more reliable and rich). On the other hand, since the global drone data is used to reflect the current position, current power, and current material reserve of the drone during the operation process, that is, fully consider the dynamic information during the operation process. For example, the current power and current material reserve will both affect whether the drone can perform effective operations, making the determined target position a position where the drone can actually perform effective operations, ensuring the reliability of the path planning, and thus improving the problem of relatively low reliability of multi-drone path planning existing in the prior art.

[0039] The first part: Regarding step S110, it should be further noted that the specific method of forming the drone embedding vector is not limited and can be selected according to actual needs.

[0040] For example, in a specific implementation manner, the global drone data may belong to text data, such as the current position, current power, and current material reserve of drone 1, and the current position, current power, and current material reserve of drone 2. Based on this, after obtaining the global drone data, the trained word embedding model (such as the Word2Vec model, etc.) can be used to perform word embedding processing on the global drone data (Word Embedding, by mapping words to a continuous vector space, so that words with similar semantics are closer in the vector space) to obtain the corresponding drone embedding vector.

[0041] For the second part, regarding step S120, it should be further noted that the specific method for deeply mining the drone embedding vector is not limited and can be selected according to actual needs.

[0042] For example, in a specific implementation, the drone embedding vector can be processed through a Deep Convolutional Neural Networks (DCNN) to achieve deep mining, thereby forming a corresponding drone deep vector.

[0043] For another example, in another specific implementation, considering that there is no direct semantic relationship among the current position, current battery level, and current material reserve of the drone during operation, however, there is a potential correlation. For example, as the drone moves continuously, the position will change continuously, and the corresponding battery level and material storage will also decrease. Therefore, it is necessary to capture these potential correlations. Based on this, step S120 above can further include step S121, step S122, and step S123. The specific implementation process of each step is as follows (in combination with Figure 3 shown).

[0044] Step S121, in the first mapping subunit included in the semantic encoding unit, perform a first mapping operation on the drone embedding vector to form an initial deep vector, and then perform a second mapping operation on the initial deep vector to form an intermediate deep vector, and then perform a third mapping operation on the intermediate deep vector to form the encoding vector of the first mapping subunit.

[0045] In an embodiment of the present invention, the in-depth mining of the drone embedding vector can be achieved through a semantic encoding unit and a semantic decoding unit. Among them, during the encoding process of the semantic encoding unit, as the network depth gradually increases, the size of the corresponding output vector can gradually decrease. For example, it can be successively 1 / 2, 1 / 4, 1 / 8, 1 / 16, 1 / 32, etc. of the size of the drone embedding vector. During the decoding process of the semantic decoding unit, as the network depth gradually increases, the size of the corresponding output vector can gradually increase. For example, it can be successively 1 / 16, 1 / 8, 1 / 4, 1 / 2, 1 times, etc. of the size of the drone embedding vector. That is to say, during the encoding process, the vector is gradually compressed, and during the decoding process, the vector is gradually expanded. Specifically, in the first mapping sub-unit included in the semantic encoding unit, the drone embedding vector is subjected to a first mapping operation to form an initial depth vector, and the initial depth vector is subjected to a second mapping operation to form an intermediate depth vector, and the intermediate depth vector is subjected to a third mapping operation to form the encoding vector of the first mapping sub-unit. Among them, the first mapping operation, the second mapping operation, and the third mapping operation are different from each other, that is, three different mapping operations are used to capture or fit some potential non-linear correlation relationships in the drone embedding vector.

[0046] Step S122: Load the encoding vector of the previous mapping sub-unit into the next mapping sub-unit for multiple levels of mapping operations until the encoding vector of the last mapping sub-unit is formed, and use this encoding vector as the drone encoding vector of the semantic encoding unit.

[0047] In an embodiment of the present invention, in the mapping sub-unit after the first one included in the semantic encoding unit, the encoding vector of the previous mapping sub-unit can be loaded into the subsequent mapping sub-unit for mapping operations at multiple levels until the encoding vector of the last mapping sub-unit is formed, and this encoding vector is used as the UAV encoding vector of the semantic encoding unit. Among them, the semantic encoding unit is configured with multiple sequentially connected mapping sub-units. That is to say, in the second mapping sub-unit, the encoding vector of the first mapping sub-unit can be sequentially subjected to the first mapping operation, the second mapping operation, and the third mapping operation as described above (the logic of each mapping operation is the same as that in the first mapping sub-unit, and specific mapping parameters can be formed during training), so as to obtain the encoding vector of the second mapping sub-unit, and so on, until the encoding vector of the last mapping sub-unit is obtained to complete the encoding. Among them, the size of the encoding vector of the first mapping sub-unit can be 1 / 2 of the size of the UAV embedding vector, the size of the encoding vector of the second mapping sub-unit can be 1 / 4 of the size of the UAV embedding vector, the size of the encoding vector of the third mapping sub-unit can be 1 / 8 of the size of the UAV embedding vector, the size of the encoding vector of the fourth mapping sub-unit can be 1 / 16 of the size of the UAV embedding vector, and the size of the encoding vector of the fifth mapping sub-unit can be 1 / 32 of the size of the UAV embedding vector.

[0048] Step S123: Use the semantic decoding unit to perform a semantic decoding operation on the UAV encoding vector of the semantic encoding unit to form a UAV decoding vector, and based on the UAV decoding vector, obtain a UAV depth vector.

[0049] In an embodiment of the present invention, after encoding is completed, that is, after obtaining the UAV encoding vector of the semantic encoding unit, a semantic decoding unit can be used to perform a semantic decoding operation on the UAV encoding vector of the semantic encoding unit to form a UAV decoding vector, and based on the UAV decoding vector, a UAV depth vector can be obtained. Among them, the semantic decoding unit is configured with a plurality of sequentially connected mapping sub-units, that is, the first mapping sub-unit performs a semantic decoding operation on the UAV encoding vector of the semantic encoding unit to obtain a corresponding decoding vector, and the second mapping sub-unit performs a semantic decoding operation on the decoding vector corresponding to the first mapping sub-unit to obtain a corresponding decoding vector, and so on, until obtaining the decoding vector corresponding to the last mapping sub-unit, and taking this UAV decoding vector as the corresponding UAV decoding vector. Then, this UAV decoding vector can be directly used as the corresponding UAV depth vector, or this UAV decoding vector can also be further processed, such as attention processing, etc., to obtain the UAV depth vector. Among them, the size of the decoding vector of the first mapping sub-unit can be 1 / 16 of the size of the UAV embedding vector, the size of the decoding vector of the second mapping sub-unit can be 1 / 8 of the size of the UAV embedding vector, the size of the decoding vector of the third mapping sub-unit can be 1 / 4 of the size of the UAV embedding vector, the size of the decoding vector of the fourth mapping sub-unit can be 1 / 2 of the size of the UAV embedding vector, and the size of the decoding vector of the fifth mapping sub-unit can be 1 times the size of the UAV embedding vector, that is, decoded to the size of the UAV embedding vector.

[0050] Optionally, in the above step S121, the specific manner of performing the first mapping operation on the UAV embedding vector is not limited either. In order to take into account the full mining of deep semantic information and consider the problems of semantic distortion and computational resource consumption during the deep mining process, the above step S121 can further include step S121a, step S121b, and step S121c. The specific implementation processes of each step are as follows (in combination with Figure 4 shown).

[0051] Step S121a: Map the vector parameters in the UAV embedding vector to a target interval to form a UAV normalized vector.

[0052] In an embodiment of the present invention, the vector parameters in the UAV embedding vector can be first mapped (i.e., normalized) to a target interval to form a UAV normalized vector. Among them, the lower limit value of the target interval is equal to 0, and the upper limit value is equal to 1. In this way, the computational complexity of subsequent processing can be reduced to a certain extent.

[0053] Step S121b: Perform a first mapping operation on the UAV normalized vector to form a first mapping vector.

[0054] In an embodiment of the present invention, after obtaining the normalized vector of the drone, the normalized vector of the drone may be subjected to a first mapping operation to form a first mapping vector. Wherein, the first mapping operation includes multiple mapping stages, a jump connection link is configured between two adjacent mapping stages, and a transmission control parameter is set on this link. The transmission control parameter is used to reflect whether the output of the previous mapping stage is connected to the output of the next mapping stage. For example, after processing the normalized vector of the drone in the first mapping stage to obtain a first processing vector, the first processing vector may be processed in the second mapping stage. Among them, if the transmission control parameter between the first mapping stage and the second mapping stage is a first value, it can be determined that the first processing vector and the processing result of the second mapping stage are added to obtain a second processing vector to improve the problem of semantic distortion. If the transmission control parameter between the first mapping stage and the second mapping stage is a second value, it can be determined that the first processing vector is not connected to the processing result of the second mapping stage. In this way, the processing result of the second mapping stage can be directly used as the second processing vector to improve the problem of excessive consumption of computing resources.

[0055] Step S121c: Perform a fully connected operation on the first mapping vector to form an initial depth vector.

[0056] In an embodiment of the present invention, after obtaining the first mapping vector, the first mapping vector may be subjected to a fully connected operation to form an initial depth vector. It should be noted that in step S121c, after multiple stages of mapping processing, the size of the vector may change accordingly, such as decreasing. Based on this, in order to make the size of the formed initial depth vector equal to the size of the drone embedding vector, it can be achieved by performing a fully connected operation. For example, according to the rule of matrix multiplication, a vector of (m*n) multiplied by a vector of (n*p) can obtain a vector of (m*p). Therefore, the first mapping vector can be multiplied by a fully connected matrix (the size can be (m*p), and the specific parameters can be formed during the training process of the semantic encoding unit. Initially, it can be a randomly generated matrix) to obtain the initial depth vector.

[0057] Optionally, in the above step S121b, the specific manner of performing the first mapping operation on the normalized vector of the drone is not limited. For example, in a specific implementation manner, in order to fully capture potential correlation relationships through the first mapping operation, the above step S121b may further include steps b1, b2, b3, b4, b5, and b6. The specific implementation process of each step is as follows (in combination with Figure 5 shown).

[0058] Step b1: Perform a sliding window process on the normalized vector of the drone to form multiple drone sliding window vectors.

[0059] In the embodiment of the present invention, a sliding window process can be performed on the normalized vector of the drone to form multiple drone sliding window vectors. Thus, the size of each drone sliding window vector is equal to the window size of the sliding window process, such as a*b. Additionally, the step size of the sliding window process can be equal to 1 to fully capture potential correlation relationships that may exist, or the step size of the sliding window process can also be greater than 1 to reduce the number of formed drone sliding window vectors, thereby reducing the consumption of computing resources.

[0060] Step b2: In the first mapping stage, perform a self-attention operation on the first drone sliding window vector to form a corresponding drone attention vector.

[0061] In the embodiment of the present invention, after obtaining the multiple drone sliding window vectors, a self-attention operation (Self-Attention, which allows each vector element to interact with different parts of itself during the processing to dynamically adjust the degree of attention to different parts, so as to capture potential correlation relationships) can be performed on the first drone sliding window vector in the first mapping stage to form a corresponding drone attention vector.

[0062] Step b3: In the second and each subsequent mapping stage, based on the drone attention vector of the previous mapping stage, perform a cross-attention operation on the drone sliding window vector corresponding to the current mapping stage to form a corresponding drone cross vector.

[0063] In the embodiments of the present invention, in each of the second and subsequent mapping stages, a cross-attention operation may be performed on the drone sliding window vector corresponding to the current mapping stage based on the drone attention vector of the previous mapping stage to form a corresponding drone cross vector. That is, a cross-attention operation (Cross-Attention, where the query vector comes from the drone attention vector of the first mapping stage, and the key vector and value vector come from the drone sliding window vector corresponding to the second mapping stage) may be performed on the drone sliding window vector corresponding to the second mapping stage based on the drone attention vector of the first mapping stage to form the drone cross vector of the second mapping stage; and, a cross-attention operation may be performed on the drone sliding window vector corresponding to the third mapping stage based on the drone attention vector of the second mapping stage to form the drone cross vector of the third mapping stage, and so on. In addition, it should be noted that there is a one-to-one correspondence between the multiple drone sliding window vectors and the multiple mapping stages, that is, the first drone sliding window vector is the drone sliding window vector corresponding to the first mapping stage, the second drone sliding window vector is the drone sliding window vector corresponding to the second mapping stage, the third drone sliding window vector is the drone sliding window vector corresponding to the third mapping stage, and so on.

[0064] Step b4, determine the transmission control parameter of the current stage based on the similarity between the drone cross vector corresponding to the current mapping stage and the drone attention vector of the previous mapping stage.

[0065] In the embodiments of the present invention, the transmission control parameter of the current stage may also be determined based on the similarity between the drone cross vector corresponding to the current mapping stage and the drone attention vector of the previous mapping stage. For example, when the similarity is large (such as greater than 0.5), the transmission control parameter may be equal to the second value, that is, when it is relatively similar, the possibility of semantic distortion is small, so there is no need to perform connection. When the similarity is small (such as not greater than 0.5), the transmission control parameter may be equal to the first value, that is, when it is relatively dissimilar, the possibility of semantic distortion is large, so connection is required.

[0066] Step b5, when the transmission control parameter is equal to 0, perform a bitwise addition operation on the drone attention vector of the previous mapping stage and the drone cross vector of the current mapping stage to form the drone attention vector corresponding to the current stage; when the transmission control parameter is equal to 1, determine the drone cross vector of the current mapping stage as the drone attention vector corresponding to the current stage.

[0067] In an embodiment of the present invention, when the transmission control parameter is equal to 0 (i.e., the first value), the UAV attention vector of the previous mapping stage and the UAV cross vector of the current mapping stage can be bitwise added to form the UAV attention vector corresponding to the current stage; when the transmission control parameter is equal to 1 (i.e., the second value), the UAV cross vector of the current mapping stage is determined as the UAV attention vector corresponding to the current stage.

[0068] Step b6: Use the UAV attention vector corresponding to the last mapping stage as the first mapping vector.

[0069] In an embodiment of the present invention, after obtaining the UAV attention vector corresponding to the last mapping stage, the UAV attention vector corresponding to the last mapping stage can be used as the first mapping vector. In this way, by performing cross-attention operations on adjacent UAV sliding window vectors in sequence, the correlation between adjacent UAV sliding window vectors can be captured.

[0070] Optionally, in step b4 above, the transmission control parameter of the current stage can be determined based on the similarity between vectors. In this case, the consumption of computing resources will increase to a certain extent. Therefore, in other specific embodiments, each transmission control parameter can be used as a parameter of the semantic encoding unit, that is, formed during the training process of the semantic encoding unit. In this way, in practical applications, the parameters learned from the sample data can be directly used, thereby reducing the consumption of computing resources. Among them, for the foregoing embodiments, the specific method of determining the transmission control parameter based on similarity is not limited either. For example, in a specific embodiment, it may include: The first step: Determine the similarity (such as cosine similarity, etc.) between the UAV cross vector corresponding to the current mapping stage and the UAV attention vector of the previous mapping stage, and determine the ratio of the number of transmission control parameters equal to 0 to the number of transmission control parameters equal to 1 among the transmission control parameters of each previous mapping stage. The second step: Update the similarity based on the ratio to form an updated similarity. For example, the ratio can be multiplied by the similarity. The third step: The updated similarity can be rounded (it can be rounded down, and when it is greater than 1, it can be updated to 1) to form the transmission control parameter of the current stage.

[0071] That is to say, when the number of transmission control parameters equal to 0 in the transmission control parameters of each previous mapping stage is relatively large (i.e., when the number of connections is relatively large), the obtained updated similarity may be greater, and the transmission control parameters of the current stage are more likely to be 1, that is, the possibility of not making a connection in the current mapping stage is greater, so as to reduce the number or proportion of connections.

[0072] Optionally, in the above step S121, the specific manner of performing the second mapping operation on the initial depth vector is not limited either. In order to screen important semantic information during the mapping process, the above step S121 may further include step S121d, step S121e, and step S121f, and the specific implementation processes of each step are as follows.

[0073] Step S121d: Extract a first characterization vector and a second characterization vector from the initial depth vector.

[0074] In the embodiment of the present invention, a first characterization vector and a second characterization vector can be extracted from the initial depth vector. Among them, both the first characterization vector and the second characterization vector can characterize the global semantic information in the initial depth vector. However, the key points of characterization can be different, that is, the extracted semantic information can be different.

[0075] Step S121e: Perform a non-linear activation operation on the first characterization vector to form a non-linear activation vector.

[0076] In the embodiment of the present invention, after obtaining the first characterization vector, a non-linear activation operation can be performed on the first characterization vector to form a non-linear activation vector. Among them, each vector parameter in the non-linear activation vector is less than or equal to 1 and greater than or equal to 0. Exemplarily, the Sigmoid activation function can be used to perform the non-linear activation operation on the first characterization vector.

[0077] Step S121f: Perform a bitwise multiplication operation on the non-linear activation vector and the second characterization vector, and perform a fully connected operation on the result of the bitwise multiplication operation to form an intermediate depth vector.

[0078] In an embodiment of the present invention, after obtaining the non-linear activation vector, the non-linear activation vector and the second characterization vector can be subjected to a bitwise multiplication operation, and a fully connected operation can be performed on the result of the bitwise multiplication operation to form an intermediate depth vector. Since each vector parameter in the non-linear activation vector is less than or equal to 1 and greater than or equal to 0, after the bitwise multiplication operation, it is possible to intercept the parameters at the corresponding positions in the second characterization vector that are equal to 0 in the non-linear activation vector, retain the parameters at the corresponding positions in the second characterization vector that are equal to 1 in the non-linear activation vector, and update and adjust the parameters at the corresponding positions of the parameters between 0 and 1, achieving different degrees of attention. In addition, since the first characterization vector and the second characterization vector are extracted from the initial depth vector, their sizes may decrease. Thus, through a further fully connected operation, the size can be maintained unchanged, that is, the size of the intermediate depth vector is equal to the size of the initial depth vector.

[0079] Optionally, in step S121d above, the specific manner of extracting the first characterization vector and the second characterization vector from the initial depth vector is not limited. For example, in a specific embodiment, to ensure that the first characterization vector and the second characterization vector have the same size and can both represent global semantic information, enabling control of the semantic information of one based on the semantic information of the other, step S121d above may further include: In the first step, the vector parameters in the initial depth vector can be mapped to a target interval to form a normalized depth vector, where the lower limit value of the target interval is equal to 0 and the upper limit value is equal to 1. In the second step, the normalized depth vector can be subjected to convolution operations through two convolutional network layers respectively to form a first characterization vector and a second characterization vector with the same size, where the two convolutional network layers have the same convolutional kernel architecture and different convolutional kernel parameters. For example, the first convolutional network layer may include convolutional kernel 1, and the second convolutional network layer may include convolutional kernel 2. The sizes of convolutional kernel 1 and convolutional kernel 2 can both be 3*3, the stride can be 1, and no padding is performed at the edges. Exemplarily, convolutional kernel 1 can be: ; Convolutional kernel 2 can be: .

[0080] Optionally, in the above step S121, the specific manner of performing the third mapping operation on the intermediate depth vector is not limited either. In order to further screen important semantic information during the mapping process, the above step S121 may further include step S121g, step S121h, and step S121i. The specific implementation processes of each step are as follows.

[0081] Step S121g: Normalize the intermediate depth vector to form an intermediate normalized vector.

[0082] In the embodiment of the present invention, the intermediate depth vector may be normalized first to form an intermediate normalized vector, so that the computational complexity of subsequent processing can be reduced.

[0083] Step S121h: Perform pooling operations on the intermediate normalized vector through two pooling network layers respectively to form a first pooling vector and a second pooling vector with the same size.

[0084] In the embodiment of the present invention, after forming the intermediate normalized vector, pooling operations may be performed on the intermediate normalized vector through two pooling network layers respectively to form a first pooling vector and a second pooling vector with the same size. Among them, the two pooling network layers have the same pooling window and pooling stride, and different pooling methods. For example, the size window corresponding to the first pooling network layer may be 3*3, the pooling stride may be 2, and the pooling method may be average pooling. The size window corresponding to the second pooling network layer may be 3*3, the pooling stride may be 2, and the pooling method may be max pooling.

[0085] Step S121i: Perform a non-linear activation operation on the first pooling vector, and based on the result of the non-linear activation operation, perform a bitwise multiplication operation on the second pooling vector to form an encoded vector of the first mapping sub-unit.

[0086] In the embodiment of the present invention, after obtaining the first pooling vector and the second pooling vector, a non-linear activation operation may be performed on the first pooling vector, and based on the result of the non-linear activation operation, a bitwise multiplication operation may be performed on the second pooling vector to form an encoded vector of the first mapping sub-unit, as described above. It should be noted that in a mapping sub-unit, both the first mapping operation and the second mapping operation can be implemented by a fully connected operation to keep the input and output sizes consistent. The third mapping operation may not perform a fully connected operation. In this way, the size compression can be achieved through the pooling operation in step S121h. The specific compression ratio can be configured according to the requirements by adjusting the pooling stride. Alternatively, a further fully connected operation may be performed on the result of the bitwise multiplication operation in step S121i to adjust the size.

[0087] Optionally, in the above step S123, the specific manner of performing semantic decoding operation on the UAV encoding vector of the semantic encoding unit is also not limited. For example, in a specific embodiment, in the semantic decoding unit, each mapping sub-unit is used to perform a first mapping operation, a second mapping operation, and a third mapping operation in sequence, and the first mapping operation and the second mapping operation have the same mapping manner as the first mapping operation and the second mapping operation in the semantic encoding unit. The third mapping operation is based on the third mapping operation in the semantic encoding unit, and performs a fully connected operation on the result of the bitwise multiplication operation to achieve the expansion of the vector size, such as expanding from 1 / 32 of the size of the UAV embedding vector to 1 / 16, 1 / 8, 1 / 4, 1 / 2, 1 times the size of the UAV embedding vector, etc. For the specific processing process, reference can be made to the relevant descriptions in the previous text, and details will not be elaborated here.

[0088] In the second part, it should be further noted that for step S120, after deeply mining the UAV embedding vector, the result of the deep mining can be further processed to obtain a UAV deep vector with higher semantic representation accuracy.

[0089] Optionally, through research, it is found that if the power change between adjacent positions is abnormal or the material reserve is abnormal, it may affect the prediction result of the subsequent target position. Therefore, the results of the historical deep mining can be further fused to obtain the UAV deep vector. For example, the above step S120 may include: In the first step, the UAV embedding vector can be deeply mined to obtain the corresponding deep mining result (the deep mining method can be achieved through the relevant methods in the previous text); In the second step, based on the historical deep mining result in the previous target position prediction (i.e., when the current position is the target position) (i.e., the result of deep mining the UAV embedding vector formed by embedding the global UAV data of the previous position), the deep mining result is updated to form the corresponding UAV deep vector.

[0090] Among them, in a specific embodiment, the manner of updating the deep mining result can be cross-attention processing. In another specific embodiment, in order to analyze the relationship between the historical deep mining result and the deep mining result from multiple perspectives, so as to provide a reliable basis for the updated result of the deep mining result, the update process may include: In the first step, bitwise multiplication, bitwise addition, and bitwise subtraction operations can be respectively performed on the historical in-depth mining result and the in-depth mining result to form corresponding multiplication analysis results, addition analysis results, and subtraction analysis results; In the second step, adjacent difference calculations can be respectively performed on the multiplication analysis result, the addition analysis result, and the subtraction analysis result to obtain a multiplication difference result, an addition difference result, and a subtraction difference result. For example, the absolute difference between the first parameter and the first parameter in the multiplication analysis result can be calculated to obtain the first parameter in the multiplication difference result, and the absolute difference between the second parameter and the first parameter in the multiplication analysis result can be calculated to obtain the second parameter in the multiplication difference result, and the absolute difference between the third parameter and the second parameter in the multiplication analysis result can be calculated to obtain the third parameter in the multiplication difference result; In the third step, based on the volatility among the multiplication difference result, the addition difference result, and the subtraction difference result, the corresponding key point parameter distribution can be determined, where the value of the key point parameter corresponding to the position with greater volatility is larger; In the fourth step, the key point parameter distribution and the in-depth mining result are bitwise multiplied to obtain the corresponding UAV depth vector.

[0091] Among them, the calculation formula for the key point parameter distribution can include: ; On this basis, by normalizing Y i , each corresponding key point parameter in the key point parameter distribution can be obtained. Among them, Max() represents taking the maximum value, A() represents calculating the mean value, B() represents calculating the standard deviation, u i , v i and z i respectively represent the i-th parameter in the multiplication difference result, the addition difference result, and the subtraction difference result. Specifically, when the volatility is large, the standard deviation will be larger and the natural logarithm function value will be smaller, making the obtained Y i larger. In this way, the value of the corresponding key point parameter is also larger, that is, it serves as a key point. Thus, through the subsequent bitwise multiplication operation, the attention degree of the parameter at the corresponding position in the in-depth mining result can be higher.

[0092] For the third part, it should be further noted that the specific method for generating the target position of each UAV is not limited and can be selected according to actual needs.

[0093] For example, in a specific embodiment, the drone depth vector and the operation area vectors of the uncompleted operation areas of the multiple drones may be spliced, and then, the generated vector obtained by splicing is processed to generate position coordinates (for example, it may be implemented by a decoder in a Transformer model), so as to form the target position of each drone.

[0094] For another example, in another specific embodiment, in order to make the generation of the target position applicable to the entire operation process of the drone, the above step S130 may further include: In the first step, the operation area vectors of the uncompleted operation areas of the multiple drones may be determined. Among them, in the operation area vectors, the positions represented by the coordinates corresponding to the parameters equal to 1 belong to the uncompleted operation areas, and the positions represented by the coordinates corresponding to the parameters equal to 0 belong to the completed operation areas. In this way, it can be ensured that the size of the operation area vector in the prediction of the target position at each moment during the entire operation process is the same, so as to facilitate subsequent processing; in addition, in order to ensure that the solution provided by the present invention can be adapted to more operation areas, the operation area vector may be configured as a vector with a larger size. In this way, when the number of positions in an operation area is less than this size, the corresponding size can be ensured by padding 0 values at the edges, that is, the other areas that do not belong to the operation area are regarded as completed operation areas. In this way, it does not affect the constraint on the uncompleted operation area; in addition, the number of positions in the operation area may be formed by evenly dividing the operation area according to a certain unit area. As Figure 6 shown, this unit area may be determined based on the area of the effective action area of the drone at a position during operation, such as 4 square meters, etc.; based on this, both the current position and the target position of the drone belong to Figure 6 the center position of a small grid in, that is, the corresponding longitude, latitude and altitude of the position. That is to say, the drone can operate on the areas corresponding to each small grid in Figure 6 in turn. In the second step, the drone depth vector and the operation area vector may be fused to form a drone operation vector. For example, the drone depth vector and the operation area vector may be spliced, added, etc. to form a drone operation vector, or self-attention processing may be performed on the results of splicing and adding to form a drone operation vector. Thirdly, for the first drone, a corresponding position probability distribution can be obtained based on the mapping of the drone operation vector, and the target position corresponding to the drone can be determined based on this position probability distribution. The position probability distribution is used to reflect the probability of each position belonging to the target position of the drone (in this way, the position with the highest probability belongs to the target position). The size of the position probability distribution can be equal to the size of the operation area vector, so that each parameter can correspond to a position. Additionally, the drone operation vector can be fully connected, and then the size can be mapped to the size of the operation area vector, and then the result of the fully connected process can be mapped through the softmax function to obtain the corresponding position probability distribution; Fourthly, for the second and each subsequent drone, the output vector corresponding to the current drone can be determined based on the embedding vector corresponding to the target position of the previous drone (the size of the embedding vector is equal to the size of the operation area vector, where only the parameter corresponding to the target position is equal to 1, and the parameters corresponding to other positions are all equal to 0) and the output vector corresponding to the previous drone (such as splicing, adding, or the self-attention processing result of the result of splicing and adding). Additionally, a corresponding position probability distribution can be obtained based on the mapping of this output vector (as described above), and the target position corresponding to the drone can be determined based on this position probability distribution. The output vector corresponding to the first drone is the drone operation vector. It should be noted that the order among the multiple drones can be determined in advance, such as being drone a, drone b, drone c, drone d, drone e, etc. in sequence, that is, the first drone is drone a, the second drone is drone b, and the third drone is drone c. Based on this, information transmission or task collaboration in the process of determining the target positions among the drones can be achieved, that is, each drone not only independently executes tasks but also depends on the decision result of the previous drone, which can improve the collaboration of task execution.

[0095] In summary, for the multi-UAV path planning method and system based on collaborative tasks provided by the present invention, first, the global UAV data is determined and embedded to form a UAV embedding vector; secondly, the UAV embedding vector is deeply mined to form a UAV deep vector; then, based on the UAV deep vector and the operation area vectors of the unfinished operation areas of multiple UAVs, the target position of each UAV is generated. Based on the above, on the one hand, due to the potential semantic information in the global UAV data that can be obtained through embedding processing and deep mining, compared with the conventional operations based on experience by artificial personnel, more reliable path planning can be carried out (the basis for path planning can be reliable and rich). On the other hand, since the global UAV data is used to reflect the current position, current power, and current material reserve of the UAV during the operation process, that is, the dynamic information during the operation process is fully considered. For example, the current power and current material reserve will affect whether the UAV can perform effective operations, so that the determined target position is the position where the UAV can actually perform effective operations, ensuring the reliability of the path planning, and thus improving the problem of relatively low reliability of multi-UAV path planning existing in the prior art.

[0096] In several embodiments provided by the embodiments of the present invention, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device and method embodiments described above are only illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of devices, methods, and computer program products according to multiple embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, as well as the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0097] In addition, the functional modules in each embodiment of the present invention can be integrated together to form an independent part, or each module can exist separately, or two or more modules can be integrated to form an independent part.

[0098] When the above-mentioned functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, an electronic device, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs, Read-Only Memories), random access memories (RAMs, Random Access Memories), magnetic disks, or optical discs that can store program codes. It should be noted that in this article, the terms "including", "comprising", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such a process, method, article, or device. Without further limitations, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article, or device including the said element.

[0099] The foregoing are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A multi-UAV path planning method based on collaborative tasks, characterized in that Including: Determine the global data of the drones, and perform embedding processing on the global data of the drones to form drone embedding vectors. Among them, the global data of the drones is at least used to reflect the current position, current power, and current material reserve of each drone during the operation process, and the drone embedding vectors are used to reflect the shallow semantic information of the global data of the drones; Deeply mine the drone embedding vectors to form drone deep vectors. Among them, the human-machine deep vectors are used to reflect the deep semantic information of the global data of the drones; Based on the drone deep vectors and the operation area vectors of the unfinished operation areas of the multiple drones, generate the target positions of each drone. Among them, each target position belongs to the unfinished operation area, and the target position is used as the movement target for the drone to start moving from the current position to complete the path planning. Moreover, the drone completes the corresponding operation tasks based on the stored materials during the process of moving from the current position to the target position.

2. The multi-UAV path planning method based on collaborative tasks according to claim 1, characterized in that, The step of deeply mining the drone embedding vectors to form drone deep vectors includes: In the first mapping subunit included in the semantic encoding unit, perform a first mapping operation on the drone embedding vectors to form initial deep vectors, and perform a second mapping operation on the initial deep vectors to form intermediate deep vectors, and perform a third mapping operation on the intermediate deep vectors to form the encoding vectors of the first mapping subunit. Among them, the first mapping operation, the second mapping operation, and the third mapping operation are different from each other; Load the encoding vectors of the previous mapping subunit into the next mapping subunit for multiple levels of mapping operations until the encoding vectors of the last mapping subunit are formed, and use these encoding vectors as the drone encoding vectors of the semantic encoding unit. Among them, the semantic encoding unit is configured with multiple sequentially connected mapping subunits; Use the semantic decoding unit to perform semantic decoding operations on the drone encoding vectors of the semantic encoding unit to form drone decoding vectors, and based on the drone decoding vectors, obtain drone deep vectors. Among them, the semantic decoding unit is configured with multiple sequentially connected mapping subunits.

3. The multi-UAV path planning method based on collaborative tasks according to claim 2, wherein, The step of performing a first mapping operation on the drone embedding vectors to form initial deep vectors includes: Map the vector parameters in the drone embedding vectors to a target interval to form drone normalized vectors. Among them, the lower limit value of the target interval is equal to 0, and the upper limit value is equal to 1; Perform a first mapping operation on the drone normalized vectors to form first mapping vectors. Among them, the first mapping operation includes multiple mapping stages, and there are jump-connected links between adjacent two mapping stages, and transmission control parameters are set on these links. The transmission control parameters are used to reflect whether the output of the previous mapping stage is connected to the output of the next mapping stage; Perform a fully connected operation on the first mapping vectors to form initial deep vectors.

4. The multi-UAV path planning method based on collaborative tasks according to claim 3, characterized in that The step of performing a first mapping operation on the normalized vector of the UAV to form a first mapping vector includes: Performing a sliding window process on the normalized vector of the UAV to form a plurality of UAV sliding window vectors; In the first mapping stage, performing a self-attention operation on the first UAV sliding window vector to form a corresponding UAV attention vector; In each subsequent mapping stage, based on the UAV attention vector of the previous mapping stage, performing a cross-attention operation on the UAV sliding window vector corresponding to the current mapping stage to form a corresponding UAV cross vector; Determining the transmission control parameter of the current stage based on the similarity between the UAV cross vector corresponding to the current mapping stage and the UAV attention vector of the previous mapping stage; When the transmission control parameter is equal to 0, performing a bitwise addition operation on the UAV attention vector of the previous mapping stage and the UAV cross vector of the current mapping stage to form the UAV attention vector corresponding to the current stage; when the transmission control parameter is equal to 1, determining the UAV cross vector of the current mapping stage as the UAV attention vector corresponding to the current stage; Taking the UAV attention vector corresponding to the last mapping stage as the first mapping vector.

5. The multi-UAV path planning method based on collaborative tasks according to claim 4, wherein The step of determining the transmission control parameter of the current stage based on the similarity between the UAV cross vector corresponding to the current mapping stage and the UAV attention vector of the previous mapping stage includes: Determining the similarity between the UAV cross vector corresponding to the current mapping stage and the UAV attention vector of the previous mapping stage, and determining the quantity ratio between the transmission control parameters equal to 0 and the transmission control parameters equal to 1 in the transmission control parameters of each previous mapping stage; Updating the similarity based on the quantity ratio to form an updated similarity; Performing a rounding operation on the updated similarity to form the transmission control parameter of the current stage.

6. The multi-UAV path planning method based on collaborative tasks according to claim 2, characterized in that, The step of performing a second mapping operation on the initial depth vector to form an intermediate depth vector includes: Extracting a first characterization vector and a second characterization vector from the initial depth vector; Performing a non-linear activation operation on the first characterization vector to form a non-linear activation vector, where each vector parameter in the non-linear activation vector is less than or equal to 1 and greater than or equal to 0; Performing a bitwise multiplication operation on the non-linear activation vector and the second characterization vector, and performing a fully connected operation on the result of the bitwise multiplication operation to form an intermediate depth vector.

7. The multi-UAV path planning method based on collaborative tasks according to claim 6, wherein The step of extracting the first characterization vector and the second characterization vector from the initial depth vector includes: Mapping the vector parameters in the initial depth vector to a target interval to form a normalized depth vector, where the lower limit value of the target interval is equal to 0 and the upper limit value is equal to 1; Performing convolution operations on the normalized depth vector through two convolutional network layers respectively to form a first characterization vector and a second characterization vector with the same size, where the two convolutional network layers have the same convolutional kernel architecture and different convolutional kernel parameters.

8. The multi-UAV path planning method based on collaborative tasks according to claim 2, wherein, The step of performing a third mapping operation on the intermediate depth vector to form the encoding vector of the first mapping subunit includes: Performing a normalization operation on the intermediate depth vector to form an intermediate normalized vector; Respectively passing the intermediate normalized vector through two pooling network layers to perform pooling operations to form a first pooling vector and a second pooling vector with the same size, wherein the two pooling network layers have the same pooling window and pooling stride, and different pooling methods; Performing a non-linear activation operation on the first pooling vector, and based on the result of the non-linear activation operation, performing a bitwise multiplication operation on the second pooling vector to form the encoding vector of the first mapping subunit; Wherein, in the semantic decoding unit, each mapping subunit is used to sequentially perform a first mapping operation, a second mapping operation, and a third mapping operation, and the first mapping operation and the second mapping operation have the same mapping method as the first mapping operation and the second mapping operation in the semantic encoding unit. The third mapping operation performs a fully connected operation on the result of the bitwise multiplication operation based on the third mapping operation in the semantic encoding unit.

9. The multi-UAV path planning method based on collaborative tasks according to claim 1, characterized in that The step of generating the target position of each drone based on the drone depth vector and the operation area vectors of the uncompleted operation areas of the multiple drones includes: Determining the operation area vectors of the uncompleted operation areas of the multiple drones, wherein in the operation area vector, the positions represented by the coordinates corresponding to the parameter equal to 1 belong to the uncompleted operation area, and the positions represented by the coordinates corresponding to the parameter equal to 0 belong to the completed operation area; Performing a fusion operation on the drone depth vector and the operation area vector to form a drone operation vector; For the first drone, mapping based on the drone operation vector to obtain the corresponding position probability distribution, and based on the position probability distribution, determining the target position corresponding to the drone, wherein the position probability distribution is used to reflect the probability of each position belonging to the target position of the drone; For each subsequent drone, based on the embedding vector of the target position corresponding to the previous drone and the output vector corresponding to the previous drone, determining the output vector corresponding to the current drone, and based on the output vector, mapping to obtain the corresponding position probability distribution, and based on the position probability distribution, determining the target position corresponding to the drone, wherein the output vector corresponding to the first drone is the drone operation vector.

10. A multi-UAV path planning system based on collaborative tasks, characterized in that, Comprising a processor and a memory, the memory is used to store a computer program, and the processor is used to execute the computer program to implement the multi-drone path planning method based on collaborative tasks according to any one of claims 1-9.

Images