Unmanned aerial vehicle cluster adaptive mode switching method, device, equipment, medium and product

Abstract

The application discloses a UAV cluster adaptive mode switching method and device, equipment, medium and product, and relates to the field of UAV control. The method comprises the following steps: acquiring obstacle data and attitude information of a UAV; performing obstacle extraction and UAV self-state estimation on the obstacle data and the attitude information to obtain an environment perception result; performing neural pulse conversion on the environment perception result to obtain a pulse input vector signal; using a brain-like pulse neural network to output a flight mode with the maximum discharge frequency according to the pulse input vector signal; controlling the UAV to execute the flight mode with the maximum discharge frequency and returning to the step of acquiring the obstacle data and the attitude information of the UAV until reaching a final task point. The application improves the flexibility, adaptability and task completion rate of the UAV cluster in a complex environment.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) control, and in particular to a method, apparatus, device, medium, and product for adaptive mode switching of UAV swarms. Background Technology

[0002] With the widespread application of unmanned aerial vehicles (UAVs) in formation flying, environmental monitoring, disaster relief, and military missions, multi-UAV cooperative control and trajectory planning have gradually become a research hotspot. In existing technologies, common formation control methods are mostly based on PID controllers or model predictive control (MPC), which can maintain formation stability to a certain extent; trajectory planning often employs... Path search algorithms such as RRT are used to enable UAVs to generate paths and avoid obstacles in complex environments.

[0003] However, in practical applications, the flight environment of drone swarms is often complex and changeable, with both static obstacles and dynamic interference present. Single-formation maintenance or fixed trajectory planning methods have the following shortcomings: 1. Lack of flexibility in mode: Traditional methods typically treat formation keeping and trajectory planning as independent tasks, lacking the ability to flexibly switch between different flight modes. For example, when the distance between obstacles is less than the current formation width, relying solely on PID to maintain formation can easily lead to collisions or failure to pass.

[0004] 2. Switching relies on manual rules: Most existing mode switching relies on manually set thresholds (such as obstacle distance and speed limits), which lacks adaptability and is difficult to cope with dynamically changing environments.

[0005] 3. Insufficient coordination efficiency: In complex scenarios, if the drone swarm cannot intelligently switch between "formation maintenance - formation reconstruction - independent trajectory planning", the overall efficiency and safety of the drone swarm will be significantly reduced.

[0006] Therefore, there is an urgent need for an intelligent method that can simulate human decision-making mechanisms to enable UAV swarms to adaptively select between three modes: formation maintenance, formation reconstruction, and trajectory planning, thereby improving the flexibility, adaptability, and mission completion rate of UAV swarms in complex environments. Summary of the Invention

[0007] The purpose of this application is to provide a method, apparatus, device, medium, and product for adaptive mode switching of unmanned aerial vehicle (UAV) swarms, enabling UAV swarms to adaptively select between three modes: formation maintenance, formation reconstruction, and trajectory planning, thereby improving the flexibility, adaptability, and mission completion rate of UAV swarms in complex environments.

[0008] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for adaptive mode switching in a drone swarm, including: Acquire obstacle data and attitude information of the drone; The obstacle data and attitude information are used to extract obstacles and estimate the UAV's own state to obtain environmental perception results. The environmental perception results are converted into neural impulses to obtain impulse input vector signals. Based on the pulse input vector signal, a brain-like pulse neural network is used to output the flight mode with the highest discharge frequency. Control the drone to execute the flight mode with the highest discharge frequency and return to "acquire obstacle data and attitude information of the drone" until the final mission point is reached.

[0009] In one embodiment, acquiring obstacle data and attitude information of the UAV specifically includes: Point cloud data is acquired using an airborne lidar window, and the UAV’s own attitude quaternion information is acquired using an IMU sensor. The relative positions of obstacles are determined based on the point cloud data; The relative positions of the obstacles and the quaternion information of the UAV's own attitude are transformed into coordinate systems to obtain the obstacle data and attitude information of the UAV.

[0010] In one embodiment, the obstacle data and attitude information are used to extract obstacles and estimate the UAV's own state to obtain environmental perception results, specifically including: Key features of obstacles are extracted from obstacle data; these key features include the distance to the UAV, relative azimuth angle, and altitude difference. The FAST-LIO2 algorithm is used to determine the UAV's own state based on the attitude and positioning information. The environmental perception results are determined based on the key features of the obstacles and the state of the UAV itself.

[0011] In one embodiment, based on the pulse input vector signal, a neuromorphic spiking neural network is used to output the flight mode with the highest discharge frequency, specifically including: The pulse input vector signal is divided into different neuron clusters and input into the input layer of the brain-like spiking neural network; the neuron clusters include distance clusters, orientation clusters, altitude clusters and state clusters; For each obstacle, distance and orientation are detected using the hidden layer of a neuromorphic spiking neural network based on the distance cluster and the orientation cluster to obtain a first subset; height adaptability is determined using the hidden layer based on the height cluster and the state cluster to obtain a second subset. The firing frequency of each output neuron in the output layer of the neuromorphic spiking neural network is determined based on the first subset and the second subset; The flight pattern corresponding to the neuron with the highest firing frequency is determined as the flight pattern with the highest firing frequency.

[0012] In one embodiment, the weight updates of the brain-like spiking neural network follow the pulse timing-dependent plasticity rule.

[0013] In one embodiment, controlling the drone to execute the flight mode with the highest discharge frequency and return to "acquire obstacle data and attitude information of the drone" until reaching the final mission point specifically includes: Obtain the drone's own position coordinates; The drone's own position coordinates are compared with the destination target to obtain the comparison result; Determine whether the comparison result is within the error threshold; If so, the task is considered successful, and the final task point has been reached; If not, return "Get obstacle data and attitude information of the drone".

[0014] Secondly, this application provides an adaptive mode switching device for unmanned aerial vehicle (UAV) swarms, comprising: The acquisition module is used to acquire obstacle data and attitude information of the drone; The extraction and estimation module is used to extract obstacles and estimate the UAV's own state from the obstacle data and attitude information to obtain environmental perception results; The pulse conversion module is used to convert the environmental perception results into neural pulses to obtain a pulse input vector signal. The flight mode determination module is used to output the flight mode with the highest discharge frequency based on the pulse input vector signal using a brain-like pulse neural network. The flight mode execution module is used to control the UAV to execute the flight mode with the highest discharge frequency and return "acquire UAV obstacle data and attitude information" until the final mission point is reached.

[0015] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the aforementioned drone swarm adaptive mode switching method.

[0016] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned UAV swarm adaptive mode switching method.

[0017] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned UAV swarm adaptive mode switching method.

[0018] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a method, apparatus, device, medium, and product for adaptive mode switching of unmanned aerial vehicle (UAV) swarms. The method involves extracting obstacles from obstacle data and attitude information, estimating the UAV's own state, and obtaining environmental perception results. These results are then converted into neural pulses to obtain pulse input vector signals. These signals are input into a brain-like spiking neural network (SNN), which determines the final flight mode to be executed based on the discharge frequency. Utilizing the competitive mechanism and adaptive learning capabilities of the SNN, the system intelligently learns flight modes, enabling flexible switching between three modes: formation holding, formation reconfiguration, and trajectory planning. This improves the flexibility, adaptability, and mission completion rate of the UAV swarm in complex environments. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A schematic diagram of an adaptive mode switching method for unmanned aerial vehicle (UAV) swarms; Figure 2 A diagram illustrating a drone formation; Figure 3 This is a schematic diagram of the SNN decision module; Figure 4 for Path search algorithm flowchart; Figure 5 A schematic diagram of the adaptive mode switching method for drone swarms; Figure 6 A schematic diagram of the functional modules of a device for an adaptive mode switching method for unmanned aerial vehicle (UAV) swarms; Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] In existing UAV swarm flight control methods, formation keeping and trajectory planning are often implemented separately. In complex real-world environments, a single approach is insufficient to meet diverse mission requirements. 1. When the environment is open, the drone swarm can move forward efficiently by maintaining a fixed formation.

[0023] 2. When encountering narrow or irregularly shaped obstacles, maintaining the original formation may not allow for smooth passage.

[0024] 3. Switching directly to independent trajectory planning mode will lead to a decrease in group coordination and an increase in communication and computing burden.

[0025] In existing technologies, mode switching mostly relies on manual rules or fixed thresholds, making it difficult to make flexible adjustments based on environmental dynamics and mission objectives, thus resulting in reduced flight efficiency or even flight failure.

[0026] To address the aforementioned issues, this application utilizes the competitive mechanism and adaptive learning capability of spiking neural networks to process perceived environmental and formation information, intelligently select flight modes, and switch between three modes: formation maintenance, formation reconstruction, and trajectory planning.

[0027] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0028] In one exemplary embodiment, such as Figure 5 As shown, an adaptive mode switching method for unmanned aerial vehicle (UAV) swarms is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method includes the following steps.

[0029] Step 501: Obtain obstacle data and attitude information of the drone.

[0030] Step 502: Extract obstacles and estimate the UAV's own state from the obstacle data and attitude information to obtain environmental perception results.

[0031] Step 503: Perform neural pulse conversion on the environmental perception results to obtain pulse input vector signals.

[0032] Step 504: Based on the pulse input vector signal, use a brain-like spiking neural network to output the flight mode with the highest discharge frequency.

[0033] Step 505: Control the drone to execute the flight mode with the highest discharge frequency and return to "Acquire drone obstacle data and attitude information" until the final mission point is reached.

[0034] In an exemplary embodiment, acquiring obstacle data and attitude information of a UAV specifically includes: acquiring point cloud data using an onboard LiDAR window and acquiring the UAV's own attitude quaternion information using an IMU sensor; determining the relative position of obstacles based on the point cloud data; and performing coordinate system transformation on the relative position of obstacles and the UAV's own attitude quaternion information to obtain the UAV's obstacle data and attitude information.

[0035] In an exemplary embodiment, obstacle extraction and UAV self-state estimation are performed on the obstacle data and attitude information to obtain environmental perception results. Specifically, this includes: extracting key features of obstacles based on obstacle data; the key features include the distance, relative azimuth angle, and altitude difference between the obstacle and the UAV; determining the UAV self-state using the FAST-LIO2 algorithm based on the attitude information and positioning information; and determining the environmental perception results based on the key features of the obstacles and the UAV self-state.

[0036] In an exemplary embodiment, the process of outputting the flight mode with the highest firing frequency using a neuromorphic spiking neural network based on the pulse input vector signal specifically includes: dividing the pulse input vector signal into different neuron clusters and inputting them into the input layer of the neuromorphic spiking neural network; the neuron clusters include distance clusters, orientation clusters, altitude clusters, and state clusters; for each obstacle, performing distance and orientation detection using the hidden layer of the neuromorphic spiking neural network based on the distance cluster and the orientation cluster to obtain a first subset; performing altitude adaptability judgment using the hidden layer based on the altitude cluster and the state cluster to obtain a second subset; determining the firing frequency of each output neuron in the output layer of the neuromorphic spiking neural network based on the first subset and the second subset; and determining the flight mode corresponding to the neuron with the highest firing frequency as the flight mode with the highest firing frequency.

[0037] In one exemplary embodiment, the weight updates of the brain-like spiking neural network follow the pulse timing-dependent plasticity rule.

[0038] In an exemplary embodiment, controlling the drone to execute the flight mode with the highest discharge frequency and returning "Acquire drone obstacle data and attitude information" until reaching the final mission point specifically includes: acquiring the drone's own position coordinates; comparing the drone's own position coordinates with the target endpoint to obtain a comparison result; determining whether the comparison result is within an error threshold; if yes, the mission is considered successful and the final mission point has been reached; otherwise, returning "Acquire drone obstacle data and attitude information" is executed. In another exemplary embodiment, such as Figure 1 As shown, the specific working process of the UAV swarm adaptive mode switching method in practical applications is also provided, including the following steps.

[0039] Step 1) Environmental perception: (1) Sensor data acquisition: Each UAV acquires point cloud data of the surrounding space through an onboard lidar sensor to detect the relative position of obstacles. ,in This represents the position of the obstacle point in the lidar coordinate system. The coordinates of the obstacle along the X-axis. The coordinates of the obstacle along the Y-axis. The coordinates of the obstacle along the Z-axis are used; simultaneously, the attitude quaternion information of the UAV itself is obtained through the IMU sensor. ,in, The scalar part of the quaternion is used to represent the weight or magnitude of the rotation; The vector part of the quaternion represents the component of the rotation axis in the x-direction; The vector part of the quaternion represents the component of the rotation axis in the y-direction; The vector part of the quaternion represents the component of the rotation axis in the z-direction. For example, in a drone swarm formation... Figure 2 As shown.

[0040] (2) Coordinate alignment and data fusion: The obstacle data measured by the lidar and the attitude information provided by the IMU are transformed into coordinate systems to unify the obstacle information into the UAV's own body coordinate system. ,in Let be the rotation matrix calculated using quaternions. , It is a translation vector. Let the position vector of the obstacle be in the UAV's body coordinate system. The converted data is then fused. The relative positions of the obstacles are the result of obstacle data processing; the lidar cannot directly obtain the obstacle coordinates.

[0041] (3) Obstacle information extraction: Key features of obstacles are extracted from the fused environmental data, including: the distance between the obstacle and the UAV (i.e., the relative distance); the relative azimuth angle (the offset angle of the obstacle relative to the front of the UAV); and the height difference. The fused environmental data includes obstacle data and the coordinates of other UAVs in the cluster.

[0042] Relative distance: .

[0043] Relative azimuth (in the horizontal plane): .

[0044] Height difference: .

[0045] (4) UAV self-state estimation: Based on IMU data and positioning information, the FAST-LIO2 algorithm is used to estimate the UAV's position and velocity in the global coordinate system. ,in, For the drone's own status, The x-axis coordinate in the global coordinate system. The Y-axis coordinate in the global coordinate system. The Z-axis coordinate in the global coordinate system. The velocity along the X-axis in the global coordinate system. The velocity along the Y-axis in the global coordinate system. This represents the Z-axis velocity in the global coordinate system.

[0046] (5) Perception result output: The final output is the perception information used for pattern decision-making, including: the distance between the current UAV and the obstacle. Relative azimuth The results include altitude difference, the drone's own position, attitude, and velocity. These results will serve as a basis for subsequent... Figure 3 The input to the decision-making module of the brain-like spiking neural network shown.

[0047] The structured information output by the environmental perception module is as follows: .

[0048] Indicates at time The set of obstacle information is the result of environmental perception; For drones and obstacles The distance; The azimuth angle of the obstacle; The difference in height; This refers to the drone's own status, including its position, attitude, and speed.

[0049] Step 2) Input the code: 2.1) The results of environmental perception This is converted into neural impulse input. N represents the number of obstacles.

[0050] 2.2) Distance to each obstacle Applicable encoding functions Convert to pulse frequency; azimuth angle and height difference Similar encoding: .

[0051] 2.3) Encode the UAV's own state as a pulse input: .

[0052] , , , For encoding functions, continuous numerical values ​​are mapped to pulse frequencies or pulse time sequences. This mapping can be achieved through linear mapping, normalized frequency modulation, or a threshold-based firing mechanism. The pulse frequency after encoding the distance to the obstacle. The pulse frequency after azimuth encoding. The pulse frequency after height difference encoding. The pulse frequency is encoded as the state of the UAV itself. This step ensures that the SNN can perform competitive calculations based on environmental information and achieve mode selection.

[0053] The role of the encoding function is to convert the continuous environmental quantities (distance, angle, altitude, speed, etc.) sensed by the UAV into pulse signals that can be processed by the SNN. This step ensures that the SNN can perform competitive calculations based on environmental information and achieve mode selection. The encoding function can be adjusted according to the specific implementation, such as using: linear frequency encoding (the closer the distance, the higher the pulse frequency); time-to-first-spike encoding; Gaussian modulation encoding (segmented activation for angle or altitude).

[0054] 2.4) Output: After encoding, a set of pulse input vector signals is output. : .

[0055] This vector serves as the input to the SNN mode decision module, driving subsequent mode selection. After input, it enters the corresponding hidden layer and finally the output layer, where weights are adjusted according to the reward function to update the next input.

[0056] Step 3) Pattern Decision: 3.1) The input signal in step 2 is used as a pulse signal input. Based on their physical meaning, they are divided into four groups, each mapped to a different cluster of neurons in the SNN input layer. , , , , representing the distance cluster, orientation cluster, altitude cluster, and state cluster respectively, and the cluster with the closest distance is selected. The obstacles are sorted in ascending order of distance and then input. Information on obstacles that are outside the range is discarded, and zero pulses are added if necessary. This represents the number of obstacles within the distance threshold.

[0057] Hidden layers contain Each LIF neuron performs spatiotemporal integration of the output pulses. The specific processing flow is as follows: (1) Distance and orientation detection (processing) and For each obstacle Distance from cluster neurons and directional cluster neurons The pulses converge in the hidden layer to the same subset, the first subset. .

[0058] .

[0059] in: , These represent the pulse firing times of the distance and orientation neurons, respectively. For synaptic kernel functions; , These are the synaptic weights learned by STDP. This corresponds to the membrane potential of the neuron. For different times.

[0060] When an obstacle is close ( High frequency) and located directly in front ( When it corresponds to around 0°, The neurons in the brain produce a strong response and output high-frequency pulses.

[0061] (2) High adaptability judgment (processing) and ): High-level cluster With state clusters The velocity and attitude information in the middle is converged into the second subset. : .

[0062] in: For the generality function: . This represents the membrane potential of the k-th neuron in the second subset. The current flight altitude of the drone. To the state cluster neurons Ns The l From one neuron to a subset of adaptability H pass No. k Synaptic weights of individual neurons Input the drone status. The high security threshold is represented by l, where l is the state cluster. Ns Neuron index in the database.

[0063] (3) Comprehensive threat assessment: Hidden layer output and The pulse signal is sent to the model output layer. Signals with a large number of pulses within the time window are selected for the model output layer.

[0064] 3.2) Spiking neuron state update: Spiking neural networks include Each neuron has a membrane potential. The evolution equation is: .

[0065] in: The membrane potential time constant; For input To neurons Synaptic weights; The input pulse signal.

[0066] When the membrane potential reaches the threshold : .

[0067] Then neurons After a spike discharge, the membrane potential is reset to... . This is the resting potential.

[0068] After input, the pulse input is accumulated as membrane potential until a threshold is reached. After a pulse is fired, the current neuron updates its state. The current spiking neural network is a multi-layer structure. Input data directly affects the first layer of the network and indirectly affects the subsequent layers. The learning rules in 3.3 update the connection weights between the networks. As a weight, it affects the accumulation and update of membrane potential. The larger the weight of the connection, the faster the membrane potential accumulates, thus relatively increasing the importance of the current link.

[0069] 3.3) Learning Rules: Weight updates follow the impulse-time-dependent plasticity rule: .

[0070] in: , which is the time difference between the presynaptic and postsynaptic pulses; The learning rate; Let be the time constant, where the subscript + indicates increase and the subscript - indicates decrease. For input To neurons The change in synaptic weight. This refers to the timing of the pulse firing of the postsynaptic neuron. This refers to the pulse firing time of the presynaptic neuron.

[0071] 3.4) Mode Output Layer: Three competing neurons are set in the output layer, each corresponding to one of the three flight modes: To maintain formation, For formation reconfiguration, For trajectory planning.

[0072] The firing frequency of each output neuron Defined as: .

[0073] in Indicates within the time window Inner neurons The number of pulses. k is the number of output neurons.

[0074] 3.5) Mode Selection: Select , , The flight pattern corresponding to the neuron with the highest firing frequency among the three neurons is used as the final mode selection: .

[0075] in Indicates time Network weights For the forward propagation function of a spiking neural network, for The multidimensional feature vectors are input into the SNN at each time step. In order to be in The flight mode selected at all times This represents the complete weight matrix of the spiking neural network at time t. The formula itself does not perform control; it only acts as a counter for the competitive election. In summary, the possible input data and mode selection correspondences are shown in Table 1.

[0076] Table 1 Input Data and Mode Selection Table

[0077] The above mode selections are merely references for the neural network to output results under certain circumstances. Here, F1 represents the discharge frequency for the formation-keeping mode, F2 for the formation-reconstruction mode, and F3 for the trajectory planning mode.

[0078] Step 4) Pattern Execution: Based on the pattern output by the decision module. The drone swarm then enters the corresponding control strategy.

[0079] 1. Formation Maintenance Mode ( ).

[0080] Objective: To maintain the relative positions of the drones within the formation. (Regarding the first...) Deploy drones and set their desired formation positions. for: .

[0081] in: For the captain position; For the first The relative offset of the drones in the formation.

[0082] Calculate the control quantity using a PID controller: .

[0083] Let be the control input for the i-th UAV in the body coordinate system. It is a proportional gain matrix. Here is the integral gain matrix. This is the differential gain matrix.

[0084] .

[0085] in, For positional error, Let t be the expected formation position of the i-th UAV. Let t be the current actual position of the i-th UAV at time t.

[0086] 2. Formation Restructuring Mode ( ).

[0087] Objective: To adjust formation to ensure safe passage in narrow passages or confined environments. Desired new formation position. Defined as: .

[0088] in: This is the formation scaling factor. This indicates a reduction in formation. This indicates an enlarged formation.

[0089] When formation switching is required, matrix transformation can be used: .

[0090] in This is a rotation / mapping matrix used to switch to a new formation (such as a wedge). (Columnar)

[0091] The control law also uses PID: .

[0092] Let be the integration variable, representing the traversal time point in the time integration.

[0093] 3. Trajectory Planning Mode ( ).

[0094] Objective: Each drone independently generates a trajectory to avoid obstacles. First, a grid map is constructed, converting the point cloud... Convert to occupying grid map : .

[0095] in: Raster resolution; For raster indexes.

[0096] like Figure 4 As shown, using The algorithm searches for paths in the raster map M. The algorithm first initializes the open and closed sets, placing the starting point into the open set and assigning it an initial evaluation value, while keeping the closed set empty. Then, it enters the main loop. At the beginning of each iteration, it checks if the open set is empty. If it is empty, it means all reachable nodes have been searched and no feasible path has been found, the search fails, and the process ends. Otherwise, it selects the node with the smallest overall evaluation value from the open set as the current optimal node, removes it from the open set, and moves it into the closed set to mark it as expanded. It then determines if this node is the target endpoint. If it is, it traces backward from the endpoint, backtracking the pointers of the preceding nodes, layer by layer, back to the starting point to reconstruct the optimal path, and the algorithm terminates successfully. If the current optimal node is not the endpoint, it obtains all its subsequent adjacent nodes to form a set to be processed. It iterates through each subsequent node in this set, calculating the path from the starting point through the current optimal node to that subsequent node. The actual path length is determined, and the subsequent node is handled in three ways based on its current state in the set: First, if the subsequent node already exists in the open set, the cost of the new path is compared with that of the historical path. Only if the new path length is less than the original record is the historical node deleted from the open set, its actual cost and preceding pointer updated, and the comprehensive evaluation value obtained by adding the actual cost and the heuristically estimated cost recalculated before being added back to the open set. Second, if the subsequent node is not in the open set but already exists in the closed set, it is also deleted from the closed set and re-added to the open set to restore its search eligibility, and its actual cost and comprehensive evaluation value are updated. Third, if the subsequent node appears in neither the open set nor the closed set, it is considered a newly discovered node, directly added to the open set, and assigned the corresponding actual cost and comprehensive evaluation value. After processing each subsequent node, the remaining nodes in the set to be processed are traversed until the set is empty, then the process is repeated from the beginning of the main loop until the destination is found or the open set is exhausted.

[0097] The algorithm first adds the starting point to the OPEN table to be queried, and initializes it. , Let be the initial position of the i-th drone. Let be the actual position of the i-th drone at the current time t. From the starting point to the current node The cost of the path already taken, The heuristic function is the current node. To the finish line The estimated cost function, To comprehensively estimate the value, repeatedly select the node with the smallest estimated value f. ;judge Whether it is the endpoint, i.e., check. , The optimal node with the smallest f value is selected from the open list. If it is the endpoint, the shortest path is found by backtracking; otherwise, all its successor child nodes are expanded and excluded. The grid; if through The path to the successor is shorter, update the parent pointer; from Backtracking along the parent pointer chain to .

[0098] After the algorithm completes the search, it outputs discrete path points: .

[0099] in: The target points are those given when the trajectory planning task is assigned; total number of points. It is determined by the path length.

[0100] The cost function is defined as: .

[0101] in: From the starting point to the node The cumulative cost; For nodes Heuristic estimates of the distance to the target (such as Euclidean distance).

[0102] The UAV performs control along the generated trajectory points, and the error is defined as: .

[0103] in This serves as the reference point on the trajectory at the corresponding moment.

[0104] The control law can be a position controller (PID or a more stable method): .

[0105] A position controller is used to reach different discrete points, thereby completing the trajectory execution.

[0106] Step 5) Feedback Update and Closed-Loop Execution: After each execution, it checks whether the drone has reached the final mission point. If it has, the mission ends, mode selection stops, and the loop exits. If it has not reached the target point, mode selection continues. The actual control input for mode execution is...

[0107] .

[0108] in, For the comprehensive control function, for The planned route.

[0109] During flight, new sensing data It is re-encoded as a pulse input The data is then fed into the SNN for iterative computation, enabling dynamic updates to the pattern decisions. .

[0110] in Indicates time Network weights.

[0111] This application utilizes a neuromorphic spiking neural network (SNN) to achieve intelligent mode switching for UAV swarms. The UAVs acquire their own state and obstacle information via lidar and IMU, encoding continuous quantities such as distance, azimuth, and altitude difference as pulse signals, which are then input into the SNN. The SNN updates its membrane potential based on these pulse signals and outputs the activation levels of three modes through a competition mechanism and learning rules: formation maintenance, formation reconstruction, or trajectory planning. The mode with the highest discharge frequency is selected as the current flight strategy. Subsequently, the UAVs execute corresponding control based on the mode: PID control to maintain formation, formation scaling or switching, or... Trajectory planning and obstacle avoidance. Environmental information is continuously fed back to the SNN to enable dynamic adaptive flight, balancing collaborative efficiency and obstacle avoidance safety.

[0112] Based on the same inventive concept, this application also provides a drone swarm adaptive mode switching device for implementing the aforementioned drone swarm adaptive mode switching method. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more drone swarm adaptive mode switching device embodiments provided below can be found in the limitations of the drone swarm adaptive mode switching method described above, and will not be repeated here.

[0113] In one exemplary embodiment, such as Figure 6 As shown, a drone swarm adaptive mode switching device is provided, comprising: The acquisition module is used to acquire obstacle data and attitude information of the drone.

[0114] The extraction and estimation module is used to extract obstacles and estimate the UAV's own state from the obstacle data and attitude information to obtain environmental perception results.

[0115] The pulse conversion module is used to convert the environmental perception results into neural pulses to obtain a pulse input vector signal.

[0116] The flight mode determination module is used to output the flight mode with the highest discharge frequency based on the pulse input vector signal using a brain-like pulse neural network.

[0117] The flight mode execution module is used to control the UAV to execute the flight mode with the highest discharge frequency and return "acquire UAV obstacle data and attitude information" until the final mission point is reached.

[0118] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data related to adaptive mode switching in UAV swarms. The I / O interfaces facilitate information exchange between the processor and external devices. The communication interface allows communication with external terminals via a network connection. When executed by the processor, the computer program implements an adaptive mode switching method for UAV swarms.

[0119] Those skilled in the art will understand that Figure 7 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method embodiments.

[0120] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the above-described method embodiments.

[0121] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the above-described method embodiments.

[0122] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0123] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0124] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0125] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0126] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for adaptive mode switching in a drone swarm, characterized in that, include: Acquire obstacle data and attitude information of the drone; The obstacle data and attitude information are used to extract obstacles and estimate the UAV's own state to obtain environmental perception results. The environmental perception results are converted into neural impulses to obtain impulse input vector signals. Based on the pulse input vector signal, a brain-like pulse neural network is used to output the flight mode with the highest discharge frequency. Control the drone to execute the flight mode with the highest discharge frequency and return to "acquire obstacle data and attitude information of the drone" until the final mission point is reached.

2. The UAV swarm adaptive mode switching method according to claim 1, characterized in that, Acquire obstacle data and attitude information for the drone, specifically including: Point cloud data is acquired using an airborne lidar window, and the UAV’s own attitude quaternion information is acquired using an IMU sensor. The relative positions of obstacles are determined based on the point cloud data; The relative positions of the obstacles and the quaternion information of the UAV's own attitude are transformed into coordinate systems to obtain the obstacle data and attitude information of the UAV.

3. The UAV swarm adaptive mode switching method according to claim 1, characterized in that, The obstacle data and attitude information are used to extract obstacles and estimate the UAV's own state to obtain environmental perception results, specifically including: Key features of obstacles are extracted from obstacle data; these key features include the distance to the UAV, relative azimuth angle, and altitude difference. The FAST-LIO2 algorithm is used to determine the UAV's own state based on the attitude and positioning information. The environmental perception results are determined based on the key features of the obstacles and the state of the UAV itself.

4. The UAV swarm adaptive mode switching method according to claim 1, characterized in that, Based on the pulse input vector signal, a neuromorphic spiking neural network is used to output the flight mode with the highest discharge frequency, specifically including: The pulse input vector signal is divided into different neuron clusters and input into the input layer of the brain-like spiking neural network; the neuron clusters include distance clusters, orientation clusters, altitude clusters and state clusters; For each obstacle, distance and orientation are detected using the hidden layer of a neuromorphic spiking neural network based on the distance cluster and the orientation cluster to obtain a first subset; height adaptability is determined using the hidden layer based on the height cluster and the state cluster to obtain a second subset. The firing frequency of each output neuron in the output layer of the neuromorphic spiking neural network is determined based on the first subset and the second subset; The flight mode corresponding to the neuron with the highest firing frequency is determined as the flight mode with the highest firing frequency.

5. The UAV swarm adaptive mode switching method according to claim 1, characterized in that, The weight updates of the brain-like spiking neural network follow the pulse timing-dependent plasticity rule.

6. The UAV swarm adaptive mode switching method according to claim 1, characterized in that, Control the drone to execute the flight mode with the highest discharge frequency and return to "acquire obstacle data and attitude information of the drone" until the final mission point is reached, specifically including: Obtain the drone's own position coordinates; The drone's own position coordinates are compared with the destination target to obtain the comparison result; Determine whether the comparison result is within the error threshold; If so, the task is considered successful, and the final task point has been reached; If not, return "Get obstacle data and attitude information of the drone".

7. An adaptive mode switching device for unmanned aerial vehicle (UAV) swarms, characterized in that, The drone swarm adaptive mode switching device includes: The acquisition module is used to acquire obstacle data and attitude information of the drone; The extraction and estimation module is used to extract obstacles and estimate the UAV's own state from the obstacle data and attitude information to obtain environmental perception results; The pulse conversion module is used to convert the environmental perception results into neural pulses to obtain a pulse input vector signal. The flight mode determination module is used to output the flight mode with the highest discharge frequency based on the pulse input vector signal using a brain-like pulse neural network. The flight mode execution module is used to control the UAV to execute the flight mode with the highest discharge frequency and return "acquire obstacle data and attitude information of the UAV" until the final mission point is reached.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the UAV swarm adaptive mode switching method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the drone swarm adaptive mode switching method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the drone swarm adaptive mode switching method as described in any one of claims 1-6.

Images