Large-scale unmanned aerial vehicle task planning method based on pigeon flock bifurcation consensus topology optimization

Abstract

The application discloses a large-scale unmanned aerial vehicle task planning method based on pigeon group bifurcation consensus topology optimization, which comprises the following steps: step 1, initializing a large-scale unmanned aerial vehicle cluster dynamic navigation task scene; step 2, generating a channel centralization bifurcation hub trajectory cluster based on pigeon group topology bifurcation search; step 3, model prediction contour control parallel optimization based on a pigeon group cooperative progress consistency mechanism; step 4, optimal hub trajectory cluster decision based on pigeon group bifurcation and consensus behavior; and step 5, outputting a current unmanned aerial vehicle cluster trajectory tracking state in the task scene. The application has the following advantages: 1) supporting cluster planning control in a large-scale, obstacle-dense and dynamic scene, and having the advantages of high efficiency and good robustness; 2) ensuring that a planner can fully analyze an environment topology; and 3) ensuring that a large-scale unmanned aerial vehicle cluster can tightly cooperate and make robust decisions in an obstacle-dense environment, and the application has strong implementability, good real-time performance and meets actual demands.

Description

Technical Field

[0001] This invention is a large-scale UAV mission planning method based on pigeon flock bifurcation consensus topology optimization, belonging to the field of UAV swarm autonomous control. Background Technology

[0002] In recent years, unmanned aerial vehicle (UAV) swarms have played a crucial role in complex industrial and military scenarios. However, achieving autonomous navigation for large-scale swarms while balancing time efficiency and safety remains a significant challenge in dynamic, chaotic, and unknown environments. When swarms navigate in complex and obstacle-dense spaces, it is essential to decouple high-level and low-level decision-making processes. High-level decisions assess and select the optimal topology for the swarm's propulsion in free space, while low-level decisions focus on trajectory optimization and control within the selected topology. However, some existing frameworks limit high-level planning to static or known environments, restricting their applicability in unknown and dynamic industrial applications. Failure to effectively separate these decision layers can degenerate swarm planning into a non-convex trajectory optimization problem that is highly susceptible to local minima, ultimately severely compromising the system's navigation efficiency and safety. This invention aims to propose a large-scale UAV mission planning method based on pigeon flock bifurcation consensus topology optimization to enhance the dynamic adaptability of UAV swarms in complex and unknown environments.

[0003] Most existing biological swarm planning and control methods draw on the cooperative mechanisms of ant colonies, bee colonies, or fish schools. Ant colony methods emphasize pheromone deposition and indirect communication, making them suitable for incremental path search. However, their decision-making relies on environmental marking and multiple rounds of cumulative feedback, making them ill-suited for high-speed, unknown, and highly dynamic airspace. Bee colony methods typically achieve collective selection through candidate evaluation and group voting, forming global preferences. However, their core remains biased towards static or low-frequency decision-making, failing to adequately respond to instantaneous conflicts in continuous movement. Fish school methods primarily rely on local rules such as alignment, aggregation, and separation to maintain formation consistency, producing smooth swarm movement, but lack explicit exploration and selection mechanisms for multi-homopy channels. Therefore, while these biological swarm methods can explain a certain degree of cooperative movement, they struggle to characterize the dynamic transitions between individual intention divergence, topological path competition, and group reunification in complex obstacle environments.

[0004] In contrast, pigeon flocks exhibit a conflict resolution mechanism more suited to aerial swarm navigation in complex environments. When facing intruding predators (such as eagles), the flock often experiences brief divergences in intent due to limited local perception, differences in individual movement capabilities, and conflicting trajectory preferences. However, these divergences do not lead to loss of control, collisions, or permanent splits. Instead, the flock first bifurcates on the potential energy surface, allowing multiple homotopic paths to be explored spontaneously in parallel. Subsequently, after a brief spatial divergence, the flock rapidly converges to the set of homotopic trajectories with the lowest overall cost through spontaneous symmetry breaking, thus avoiding the predator from the same side, such as... Figure 1 As shown. This mechanism retains both topology exploration capabilities and group connectivity, enabling the cluster to traverse complex airspace as a safe and unified whole.

[0005] In summary, this invention proposes a large-scale UAV mission planning method based on pigeon flock bifurcation consensus topology optimization. This method draws on the collective intelligent behavior of pigeon flocks, successfully decoupling high-level topology decision-making from low-level trajectory optimization. By introducing a multi-homotopy parallel optimization mechanism to avoid local minima, it effectively integrates the advantages of optimization-based local and global planners, achieving a balance between cluster topology adaptability and formation robustness in dynamic and unknown environments. Summary of the Invention

[0006] The purpose of this invention is to propose a large-scale UAV mission planning method based on pigeon flock bifurcation consensus topology optimization, aiming to solve the problem of autonomous planning and navigation of UAV swarms in unknown and dynamic environments. By drawing on the bifurcation behavior of pigeon flocks in nature, a pigeon flock topology bifurcation search and topology consensus decision model is established and mapped to the autonomous planning and control of UAV swarms, providing a new solution to the collaborative planning problem of UAV swarms in unknown and obstacle-dense environments.

[0007] This invention proposes a large-scale UAV mission planning method based on pigeon flock bifurcation consensus topology optimization, the overall process of which is as follows: Figure 2 As shown, the specific implementation steps are as follows: Step 1: Initialize the dynamic navigation mission scenario for a large-scale UAV swarm S11. Initialize the three-dimensional workspace and obstacle set for the UAV swarm dynamic navigation task. This invention first establishes a continuous three-dimensional workspace for dynamic planning tasks of unmanned aerial vehicle (UAV) swarms. ,in Let be a real number space. Assume the 3D navigation environment of the UAV swarm consists of both static and dynamic obstacles. Specifically, we establish a set containing all static obstacles, denoted as . ,in This represents the total number of static obstacles. Simultaneously, a set of dynamic obstacles is established. Since the area occupied by dynamic obstacles in space changes over time, this will be considered at discrete time steps. The set of dynamic obstacles occupying a specific area is represented as ,in This indicates the total number of dynamic obstacles.

[0008] To ensure the navigation safety of drone swarms in complex and dynamic environments, it is necessary to remove the areas occupied by obstacles from the three-dimensional workspace and extract free space that does not collide with any obstacles. Therefore, at any given discrete time step... The collision-free space of a drone swarm is represented as a time-varying set. Subsequent cooperative trajectory planning and control solutions will be strictly limited to this moment-varying collision-free space. Internally, thus providing a basis A drone swarm composed of homogeneous individuals provides underlying security.

[0009] S12. Set the starting and target areas for the drone swarm movement. Before performing navigation planning tasks, it is first necessary to determine the boundary constraints of the UAV swarm in the mission space. Assume the swarm consists of... It consists of a series of isomorphic drones, defining the start time of each drone in the mission. The initial state is and the end time of the mission The initial state is Based on the state distribution of individuals in the cluster, this invention constructs two mutually disjoint bounded convex sets, namely the initial convex hull and the terminal convex set. and target convex hull terminal These two terminal regions are used to characterize the overall distribution range of the drone swarm. They are defined as the convex hulls of individual state points, and their mathematical expressions are as follows: (1) in, This represents the topological convex hull operation. By setting the convex hull boundaries of the starting and target regions, this invention transforms the high-dimensional constraints of a single terminal into a topological constraint space for the entire cluster, providing a geometric benchmark for subsequent extraction of central trajectories and construction of central trajectory clusters. During navigation, the trajectories of all UAVs will be constrained by the topological constraints from the starting and target regions. Extend to Within the topological space, this ensures that the cluster maintains the preset formation robustness during both the start and end phases.

[0010] S13. Establish an individual nonlinear kinematics and dynamics model for the UAV. This invention treats all individuals in a drone swarm as isomorphic nonlinear models. ,in Let UAV state vector be... For location, It is a linear velocity vector. To describe the unit quaternion for rotation from the body coordinate system to the inertial coordinate system, The angular velocity in the body coordinate system. Given the system input, the UAV isomorphic model can be represented as: (2) in, Hamiltonian quaternion multiplication, Represents a quaternion rotation matrix. For the quality of drones, For rotational inertia, For the resultant force of thrust, This refers to the fuselage torque. (From...) and The system inputs are further decoupled as follows: (3) in, For drones The throttle of the motor, The length of the drone's arm. Let be the constant torque of the motor. It can be seen that the in the drone swarm... An individual can be represented by the following discrete-time nonlinear system: (4) in, and These are the feasible regions for state and control inputs, respectively. and These are the dimensions of state and control input, respectively. k Represents the discrete time step. This represents the state transition function.

[0011] Step 2: Generate a cluster of channel-centered bifurcation hub trajectories based on pigeon flock topological bifurcation search. S21. Generate a set of bifurcation hub trajectories based on the centralized visibility probability route map of pigeon flock intelligent channels. When a flock of pigeons encounters a rapidly approaching dynamic threat, such as an eagle, the flock does not immediately detour along a single path. Instead, it briefly bifurcates in the early stages of the threat. Some individuals tend to pass through the threatened side, while others explore feasible paths on the other side, thus conducting parallel exploration of different topological spaces around the threat within a limited time. As the flock develops a more stable assessment of the threat's location, movement trends, and passable areas, the initially dispersed flight trends gradually converge, the directional differences between individuals decrease, and the flock reconverges onto a shared escape route or detour. This bifurcating-then-converging behavior demonstrates that in dynamic risk environments, pigeon flocks can improve their route exploration capabilities through short-term spatial differentiation and maintain overall cohesion and flight stability through subsequent consensus.

[0012] In the dynamic navigation scenario of UAV swarms based on pigeon flock bifurcation and consensus behavior, in order to explore the topological space around obstacles, it is necessary to first plan multiple bifurcation paths with different bifurcation points, i.e., bifurcation center trajectories. Currently, the most common bifurcation path planning method is the visibility probability roadmap, but the paths generated by the traditional visibility probability roadmap method tend to be close to the obstacle edges, which is not conducive to the swarm's passage. When avoiding obstacles, pigeon flocks usually maintain the swarm centerline in the middle of the gaps between obstacles to ensure sufficient safety margin. Inspired by this, this invention proposes a pigeon flock intelligent channel-centered visibility probability roadmap algorithm to generate bifurcation center trajectories with sufficient safety margin. First, it traverses any obstacle pair in the environment. Extract the shortest connecting line segment between them. ,in and Next, using the midpoint of the line segment... Centered on, Construct an isotropic detection sphere with diameter as its diameter. If the probe sphere does not contain any other obstacles, then the condition is met. If the line segment is found to be a valid channel, it will be added to the set of valid channels. In order to simulate the intelligent passage behavior of pigeon flocks, this invention makes the central trajectory as close as possible to the central axis of the local free space. Specifically, for each effective channel... Identify its left and right vertices and Calculate the geometric center of the channel. and half width Before performing graph expansion, the above-mentioned intelligent geometric center anchor point of the pigeon flock is set. Pre-allocate and initialize to the node set and the set of guard nodes In the middle, using the initialized set of nodes, a visibility probability roadmap expansion is performed to construct a connected graph containing all homotopy classes. ,in Let be the set of undirected edges. For a graph... Each undirected edge in Define and assign a pigeon-centered communication cost function that simultaneously penalizes the total path length and centripetal deviation: (5) in, Indicates the length cost weight. The weight for the centripetal deviation penalty. Indicates the edge of the line With corresponding channel The intersection points. Then, a depth-first search algorithm is used to search the connected graph. Extract all feasible path sets Calculate the total cost of each path in the set. The paths are then sorted in ascending order of cost, from smallest to largest. Subsequently, the sorted path set is traversed, and the 3D paths are analyzed using the 2D H-Signature method. Homotopy comparisons are performed on the projections on the horizontal plane. Only the optimal paths belonging to different homotopy classes are retained, while suboptimal homotopy redundant paths are eliminated. This process continues until a predetermined number of bifurcation paths are extracted, ultimately outputting a set of intelligent bifurcation central trajectories for the pigeon flock. , which contains a finite number of lines starting from the origin To the finish line bifurcation center trajectory .

[0013] S22. Generate a set of linear transition boundary trajectories based on the pigeon flock hierarchical interaction mechanism. In a pigeon flock, there are both leaders and followers. The leader is responsible for planning and decision-making, transmitting the planned strategy to the followers through a circular attractor network. This hierarchical interaction provides a solution for lightweight, large-scale UAV swarm navigation. First, based on the intelligent bifurcation center trajectory of the pigeon flock generated in step S21... Construct the linear transition boundary trajectories. Define the set of linear transition boundary trajectories as follows: For any point on the terminal boundary of the initial convex hull. any point on the boundary of the target convex hull terminal By performing deformable linear transfer on the central trajectory, connection point pairs are generated. linear transition boundary trajectory Its mathematical expression is: (6) in, For path progress parameters, and These are the spatial translation vectors at the starting and target terminals, respectively.

[0014] S23. Central trajectory cluster relocalization based on real-time reconstruction behavior of pigeon flock formation When pigeons fly in the air, they often adjust their formation due to the sudden appearance of obstacles. Unlike other birds such as geese and starlings, pigeons' adjustments to their formation only affect pigeons at the boundary positions, maintaining the overall formation without a complete structural change. The pigeon flock hierarchical interactive linear transfer boundary trajectories generated in step S22 may collide with obstacles, thus requiring further processing. This invention proposes a relocation method for the trajectory cluster of the pigeon flock configuration real-time reconstruction behavior execution center. This method relocates only a finite number of linear transfer boundary trajectories, and then generates an infinite number of feasible internal curves through convex combinations, thereby ensuring the method's lightweight nature. Specifically, the boundary trajectories are first extracted. The straight line through which the central trajectory passes The intersection point is used as a reference point for evaluation. Calculate the distance between the reference point and the nearest obstacle boundary point. Euclidean distance If this distance is less than the preset safe collision distance for obstacles. ,Right now This triggers the centripetal compression mechanism. The edge reference safe point is calculated as follows: (7) Thus, the centripetal compression coefficient is obtained as: (8) If this distance is less than the preset obstacle safe collision distance ,Right now Then let Finally, the reference point is moved along the trajectory pointing to the central pivot. Compress the direction to obtain a safe reference point after relocation: (9) After obtaining the safe reference point for relocation, based on the ordered set of reference parameters For each parameter range Perform deformable linear transfers to reconstruct continuous boundary trajectories, wherein the ordered set of reference parameters is composed of a series of reference points. The path progress parameters constitute the path. The reconstructed continuous boundary trajectory is represented as: (10) This yields a set of safe relocation linear transfer boundary trajectories. Then, a convex combination method is adopted, utilizing the convex combination coefficients. By linearly weighting the linear transition boundary trajectories, an infinite number of feasible and non-intersecting internal curves are generated. ,Right now: (11) This constitutes the set of internal curves for relocation. The output includes a repositioning hub trajectory cluster containing the reconstructed boundary trajectory set and the internal curve set. This provides a safe and feasible optimized local topology space for drone swarms.

[0015] Step 3: Parallel optimization of model prediction profile control based on pigeon flock coordination and progress consistency mechanism S31. Local optimization of cluster individuals based on standard model predictive contour control For traditional model-predictive contour control frameworks, the desired 3D path is assumed. Its arc length parameterized as ,in Represents discrete time steps The desired arc length of the path. The desired path is in The unit tangent vector at point is represented as ,satisfy Given a three-dimensional desired spatial path Model Predictive Contouring Control (MPCC) aims to generate a control policy. This maps the current state to the optimal control input in the rolling time domain. Specifically, the goal of MPCC is to maximize the propagation speed along the path while minimizing the contour error. Its objective function is defined as: (12) in, Indicates the prediction time domain, These are the penalty weights for contour error and propulsion progress, respectively. and They are state and input, respectively. This is the initial state. Discrete time step. Contour error Expressed as: (13) Because the accurate contour projection is obtained by solving the above formula The computational load is extremely high, making it difficult to meet the requirements of real-time online optimization. Therefore, this invention treats the arc length parameter as an additional state. Introducing the original system and augmenting it, namely: (14) in, The virtual propulsion speed is determined by the controller. The sampling interval is denoted as . The forward evolution strictly enforces a monotonically increasing tracking progress. Virtual propulsion speed. By increasing system speed Projecting onto the unit tangent vector yields the following, expressed as: Therefore, the extended system state and input can be represented as: The extended state is defined as follows: Extended control input is defined as .

[0016] at discrete time step Position error vector It can be decoupled into two orthogonal components, namely the hysteresis error. and contour error Hysteresis error Defined as the longitudinal deviation along the direction of motion, it is obtained by measuring the position error. Projected onto the unit tangent vector The above approximate calculation is expressed as: (15) Contour error The lateral deviation orthogonal to the desired path is quantified and expressed as: (16) in, Let be an identity matrix of appropriate order. It can be deduced that if... ,but The approximate values ​​will be equal to their true values.

[0017] S32. Group model prediction profile control based on pigeon flock coordination progress consistency mechanism If the aforementioned independent model-predicted contour control process is applied individually to each individual in the cluster to track its predefined path, each individual will aggressively maximize its own propulsion speed under local constraints. However, when individuals navigate in local environments with heterogeneous dynamic constraints, the optimal propulsion speeds of each individual will differ significantly. This inconsistent decision-making not only disrupts the cluster's formation topology, thereby increasing communication overhead between agents, but may also cause designated central trajectory clusters to be truncated by moving obstacles. Therefore, this situation is highly susceptible to triggering asynchronous or severely conflicting obstacle avoidance maneuvers, ultimately leading to cluster navigation failure.

[0018] Unlike other birds, pigeons share information with their neighbors through circular attractors during flock navigation. These circular attractors propagate throughout the flock, ensuring consistent path-tracking progress across all individuals and maintaining the formation's topology. Inspired by this cooperative progress consistency mechanism, this invention improves upon standard model prediction contour control by introducing a cooperative progress consistency penalty term into the cost function for each individual: (17) in, This represents the consistency penalty coefficient. The introduction of this technology forces the cluster to maintain a synchronized formation topology in a highly dynamic environment, thereby maximizing tracking progress while ensuring safe tracking of the preset internal curves within the central trajectory cluster.

[0019] S33. Parallel solution of multihomopy optimal control problem based on pigeon bifurcation driving. Combining the contour error, hysteresis error, and consistency penalty obtained in steps S31 and S32, for each individual In the prediction time domain The complete local objective function within is expressed as: (18) The penalty matrix is ​​defined as follows: ,parameter , and The penalty weights represent the contour error, hysteresis error, and propulsion speed. Positive definite matrix. Used to penalize changes in control input between adjacent time steps.

[0020] By all The summation of the local costs of each agent, the total traversal cost of the central trajectory cluster is defined as follows: Subsequently, the local planner The trajectory optimization process can be represented as the following optimal control problem: (19) In this optimal control problem, the ergodic cost objective function described above expresses the overall planning objective. The agent dynamics, initial conditions, and state / input constraints of the UAV are strictly limited by the corresponding equations and inclusion relations. Feasible state set This constitutes the central trajectory cluster constraint of topological homotopy and the dynamic obstacle avoidance constraint between agents in the optimal control problem.

[0021] Step 4: Make a decision on the optimal central trajectory cluster based on pigeon flock bifurcation and consensus behavior. Pigeon flocks in nature exhibit remarkable collective intelligence when resolving conflicts in complex environments without centralized coordination. When a flock moves at high speed through dense environments, individual pigeons often diverge in intent due to limited local perception, heterogeneous kinematic constraints, or conflicting trajectory preferences. However, the flock will optimize its decision-making in parallel, selecting a single homotopy class to execute. Based on this flock consensus decision-making behavior, this invention establishes a flock consensus decision-making weighted topology consensus model. That is, during dynamic navigation, large-scale UAV swarms only execute decisions within the candidate relocation center trajectory cluster set. Select and execute the unique optimal central trajectory cluster Since all local planners in each homotopy class minimize local cost functions with identical structures, the traversal cost can be directly obtained from their solutions. The quality of each central trajectory cluster is evaluated. However, if decisions are made solely based on the minimum traversal cost at the current moment, when the costs of multiple central trajectory clusters of different homotopy classes are similar, the UAV swarm may frequently switch between different homotopy classes, leading to swarm structure divergence or collisions with dynamic obstacles. To prevent system oscillations caused by such frequent switching and to simulate the topological consensus decision-making mechanism of a pigeon flock to avoid permanent structural failure, this invention introduces a consensus decision weight. The index of the optimal central trajectory cluster is determined by the following formula: (20) The specific weight allocation rules are as follows: If the current central trajectory cluster has been selected for execution in the previous iteration cycle, priority compensation is given to it, and its weight is set to... Otherwise, treat it as a candidate central trajectory cluster and set its weight to... For priority compensation coefficient The physical meaning of its extreme case is as follows: when When this decision-making strategy degenerates into a strict minimum cost decision, it completely tends to explore new and better topological paths; when In this case, the local planner will maintain the central trajectory cluster of the same homotopy class as the previous iteration, abandoning the exploration of new topologies. Through the above evaluation model based on consensus decision weights, this invention enables UAV swarms to dynamically balance topological bifurcation search and structural cohesion maintenance in complex dynamic environments. Once the parallel solvers reach a consensus in a certain topological homotopy class, this weight mechanism will protect that homotopy class from the influence of minor external perturbations, thereby ensuring that the swarm safely traverses obstacle-dense areas in a unified cooperative state.

[0022] Step 5: Output the current drone swarm trajectory tracking status in the mission scenario. In static and dynamic task scenarios of UAV swarm dynamic navigation based on pigeon flock bifurcation and consensus behavior, the real-time position of each individual UAV in the swarm (e.g., ...) is output. Figure 3 As shown), virtual progress speed and actual speed (e.g.) Figure 4 , Figure 6 As shown), motor thrust (such as) Figure 5 , Figure 7 (As shown). The simulation ends once all drones have reached the designated area.

[0023] The advantages and beneficial effects of this invention are as follows: 1) A topology bifurcation-driven model prediction contour control parallel optimization collaborative planning framework based on pigeon flock intelligence was established, which successfully decoupled high-level topology decision-making from low-level trajectory optimization, effectively overcoming the shortcomings of traditional single homotopy optimization planners that are prone to getting trapped in local minima and are difficult to adapt to dynamic environments. 2) Based on the pigeon flock topological bifurcation search mechanism, a channel center visibility probability route map and central trajectory cluster relocation method are proposed, which significantly reduces the computational overhead of large-scale cluster collaboration and ensures the algorithm's lightweight and real-time response capabilities. 3) Based on the pigeon flock topological consensus decision-making mechanism, a multi-homogeneous parallel optimization and topological consistency decision-making method is proposed to ensure the close collaboration and feasibility of the cluster in a highly dynamic environment. Attached Figure Description

[0024] Figure 1 This refers to the dynamic bifurcation and consensus behavior of pigeon flocks under predator invasion.

[0025] Figure 2 This outlines the overall process of a large-scale UAV mission planning method based on topological parallel optimization of pigeon flock bifurcation and consensus behavior.

[0026] Figure 3 The results are simulations of navigation for large-scale UAV swarms in static and dynamic environments.

[0027] Figure 4 The virtual progress speed and actual speed of a large-scale drone swarm in a static environment.

[0028] Figure 5 This is the thrust curve of a large-scale UAV motor in a static environment.

[0029] Figure 6 The virtual progress speed and actual speed of a large-scale drone swarm in a dynamic environment.

[0030] Figure 7 This is the thrust curve of a large-scale UAV motor in a dynamic environment. Detailed Implementation

[0031] The effectiveness of the proposed manned / unmanned aerial vehicle (UAV) swarm dynamic optimal homotopy trajectory cluster collaborative planning method based on pigeon-like intelligence is verified through a specific example. In this example, the number of UAVs is set to 100, with 200 static obstacles and 10 dynamic obstacles. The simulation environment for this example is configured as follows: Ubuntu 20.04, Intel Core i9 14900k processor (6.00GHz), 128GB RAM, NVIDIA RTX 4090 GPU, and MATLAB 2024a. The specific practical steps of this example are as follows: Step 1: Initialize the dynamic navigation mission scenario for a large-scale UAV swarm S11. Initialize the three-dimensional workspace and obstacle set for the UAV swarm dynamic navigation task. This invention first establishes a continuous three-dimensional workspace At the same time, establish a set containing all static obstacles. ,in This represents the total number of static obstacles. The set of dynamic obstacles is established as follows: ,in This represents the total number of dynamic obstacles. Therefore, at any given discrete time step... The collision-free space of a drone swarm is represented as a time-varying set. .

[0032] S12. Set the starting and target areas for the drone swarm movement. The starting and target boundary constraints of the UAV cluster in the mission space are determined according to Equation (1), namely the starting convex hull terminal and the target convex hull terminal.

[0033] S13. Establish an individual nonlinear kinematics and dynamics model for the UAV. This invention treats all individuals in a drone swarm as isomorphic nonlinear models. As shown in equation (2). For the quality of drones, Let the moment of inertia be denoted by equation (3). Further decoupling of the system input is then performed according to equation (3), where... The length of the drone's arm. This is the constant torque of the motor.

[0034] Step 2: Generate a cluster of channel-centered bifurcation hub trajectories based on pigeon flock topological bifurcation search. S21. Generate a set of bifurcation hub trajectories based on the centralized visibility probability route map of pigeon flock intelligent channels. The channel-centered visibility probabilistic route map algorithm proposed in this invention first traverses any pair of obstacles in the environment and extracts the shortest connecting line segments between them. Then, using the midpoint of this line segment as the center and the length of the line segment as the diameter, an isotropic detection sphere is constructed. If the detection sphere does not contain any other obstacles, the line segment is determined to be a valid channel and added to the set of valid channels.

[0035] To ensure the central trajectory closely approximates the central axis of the local free space, for each valid channel, its left and right vertices are identified, and the geometric center and half-width of that channel are calculated. Before graph expansion, these geometric center anchor points are pre-assigned and initialized to the node set and the guard node set.

[0036] Using the initialized node set, a visibility probability roadmap expansion is performed to construct a connected graph containing all homotopy classes. For each undirected edge in the graph, the cost function is defined as shown in equation (5). Indicates the length cost weight. The centripetal deviation penalty weight.

[0037] Next, a depth-first search algorithm is used to extract all feasible paths from the connected graph. The total cost of each path in the set is calculated, and the paths are sorted in ascending order of cost. Then, the sorted path set is traversed, and the 3D paths are analyzed using the 2D H-Signature method. Homotopy comparison is performed on the projections on the horizontal plane. Only the optimal path belonging to different homotopy classes is retained, and the second-best homotopy redundant paths are eliminated until a set number of bifurcation paths are extracted. Finally, the set of bifurcation central trajectories is output.

[0038] S22. Generate a set of linear transition boundary trajectories based on the pigeon flock hierarchical interaction mechanism. According to equation (6), the linear transition boundary trajectory of the central trajectory cluster is constructed based on the generated bifurcation central trajectory.

[0039] S23. Central trajectory cluster relocalization based on real-time reconstruction behavior of pigeon flock formation The proposed central trajectory cluster relocation method first extracts the intersection point of the boundary trajectory and the straight line through which the central trajectory passes, and uses it as the reference point to be evaluated. The Euclidean distance between this reference point and the nearest obstacle boundary point is calculated. The edge reference safety point is calculated according to equation (7), thus obtaining the centripetal compression coefficient shown in equation (6). Finally, the relocated safety reference point and the reconstructed boundary trajectory representation are obtained according to equations (9) and (10), thus obtaining a safe relocation linear transfer boundary trajectory set. Then, an infinite number of feasible and non-intersecting internal curves are generated according to equation (11), thus forming a relocation internal curve set, and outputting a relocation central trajectory cluster containing the reconstructed boundary trajectory set and the internal curve set, providing a safe and feasible optimized local topology space for UAV swarms.

[0040] Step 3: Parallel optimization of model prediction profile control based on pigeon flock coordination and progress consistency mechanism S31. Local optimization of cluster individuals based on standard model predictive contour control First, construct the model predictive control objective function as shown in equation (12), where, Indicates the prediction time domain, These are the penalty weights for contour error and advancement progress, respectively.

[0041] S32. Group model prediction profile control based on pigeon flock coordination progress consistency mechanism This invention introduces a collaborative progress consistency penalty term, as shown in equation (17), into the cost function of each individual, where, This represents the consistency penalty coefficient.

[0042] S33. Parallel solution of multihomopy optimal control problem based on pigeon bifurcation driving. Combining the decoupled location error and consistency penalty in steps S31 and S32, for each individual Construct the prediction time domain as shown in equation (18). The complete local objective function within. Where, the parameters... , and The penalty weights represent the contour error, hysteresis error, and propulsion speed. Positive definite matrix. Used to penalize changes in control input between adjacent time steps. Local planner The trajectory optimization process is represented as the optimal control problem shown in equation (19).

[0043] Step 4: Make a decision on the optimal central trajectory cluster based on pigeon flock bifurcation and consensus behavior. During dynamic navigation, the index of the optimal central trajectory cluster is determined by equation (22). The weights are set to... .

[0044] Step 5: Output the current drone swarm trajectory tracking status in the mission scenario. In static and dynamic mission scenarios of UAV swarm dynamic navigation based on pigeon flock bifurcation and consensus behavior, the system outputs the position, virtual progress speed, actual speed, and motor thrust of each individual UAV in the swarm in real time. The simulation ends when all UAVs have reached the designated area.

Claims

1. A large-scale UAV mission planning method based on pigeon flock bifurcation consensus topology optimization, characterized in that: The steps are as follows: Step 1: Initialize the dynamic navigation mission scenario for a large-scale UAV swarm, including: initializing the three-dimensional workspace and obstacle set for the dynamic navigation mission of the UAV swarm; setting the starting and target areas for the movement of the UAV swarm; and establishing the nonlinear kinematics and dynamics models of individual UAVs. Step 2: Generate a cluster of channel-centered bifurcation hub trajectories based on pigeon flock topological bifurcation search, including: Generate a set of bifurcation hub trajectories based on the probability route map of the centralized visibility of pigeon flock intelligent channels; generate a set of linear transition boundary trajectories based on the hierarchical interaction mechanism of pigeon flock; and relocate the hub trajectory cluster based on the real-time reconstruction behavior of pigeon flock formation. Step 3: Parallel optimization of model prediction profile control based on the pigeon flock cooperative progress consistency mechanism, including: local optimization of individual clusters based on standard model prediction profile control; group model prediction profile control based on the pigeon flock cooperative progress consistency mechanism; and parallel solution of multi-homopy optimal control problems driven by pigeon flock bifurcation. Step 4: Make a decision on the optimal central trajectory cluster based on pigeon flock bifurcation and consensus behavior; Step 5: Output the current drone cluster trajectory tracking status in the mission scenario.

2. The method according to claim 1, characterized in that: The specific process of step one is as follows: S11. Output the current drone swarm trajectory tracking status in the mission scenario. First, establish a continuous three-dimensional workspace. ,in The space is real number space; at the same time, a set containing all static obstacles is established. ,in U Indicates the total number of static obstacles; dynamic The obstacle set is established as ,in V This represents the total number of dynamic obstacles; therefore, at any given discrete time step... The collision-free space of a drone swarm is represented as a time-varying set. ; S12. Set the starting and target areas for the drone swarm movement. S13. Establish an individual nonlinear kinematics and dynamics model for the UAV. Treating all individuals in a drone swarm as isomorphic nonlinear quadrotor models ,in Here is the state vector of the quadcopter UAV. For location, It is a linear velocity vector. To describe the unit quaternion for rotation from the body coordinate system to the inertial coordinate system, The angular velocity in the body coordinate system. Input for the system.

3. The method according to claim 1, characterized in that: The specific process of step two is as follows: S21. Generate a set of bifurcation hub trajectories based on the centralized visibility probability route map of pigeon flock intelligent channels. First, iterate through any pairs of obstacles in the environment. Extract the shortest connecting line segment between them. ,in and Next, using the midpoint of the line segment... Centered on, Construct an isotropic detection sphere with diameter as the reference value. If the probe sphere does not contain any other obstacles, then the condition is satisfied. If the line segment is found to be a valid channel, it is added to the set of valid channels. In order to make the central trajectory as close as possible to the central axis of the local free space, for each effective channel... Identify left and right vertices and Calculate the geometric center of the channel and half width Before performing graph expansion, anchor the geometric center point. Pre-allocate and initialize to the node set and the set of guard nodes In the middle; using the initialized node set, perform visibility probability roadmap expansion to construct a connected graph containing all homotopy classes. ,in For a graph, the set of undirected edges is given. Each undirected edge in Assign a pigeon-centered communication cost function that simultaneously penalizes the total path length and centripetal deviation: (1) in, Indicates the length cost weight. The weight for the centripetal deviation penalty. Indicates the edge of the line With corresponding channel The intersection points; then, a depth-first search algorithm is used to search the connected graph. Extract all feasible path sets ; Calculate the total cost of each path in the set. The paths are then sorted in ascending order of cost; subsequently, the sorted path set is traversed, and the 3D paths are analyzed using the two-dimensional H-Signature method. Homotopy comparisons are performed on the projections on the horizontal plane. Only the optimal paths belonging to different homotopy classes are retained, while suboptimal homotopy redundant paths are eliminated. This process continues until a predetermined number of bifurcation paths are extracted, ultimately outputting a set of intelligent bifurcation central trajectories for the pigeon flock. , which contains a finite number of lines starting from the origin To the finish line bifurcation center trajectory ; S22. Generate a set of linear transition boundary trajectories based on the pigeon flock hierarchical interaction mechanism. First, based on the intelligent bifurcation center trajectory of the pigeon flock generated in step S21 Construct linear transition boundary trajectories; define the set of linear transition boundary trajectories as follows: For any point on the terminal boundary of the initial convex hull any point on the boundary of the target convex hull terminal By performing deformable linear transfer on the central trajectory, connection point pairs are generated. linear transition boundary trajectory ; S23. Central trajectory cluster relocalization based on real-time reconstruction behavior of pigeon flock formation First, extract the boundary trajectory. The straight line through which the central trajectory passes The intersection point will serve as a reference point for evaluation. ; Calculate the distance between the reference point and the nearest obstacle boundary point. Euclidean distance If the distance is less than the preset safe collision distance for obstacles ,Right now This triggers the centripetal compression mechanism; the edge reference safe point is calculated as follows: (2) Thus, the centripetal compression coefficient is obtained as: (3) If the distance is less than the preset obstacle safe collision distance ,Right now Then let Finally, the reference point is moved along the trajectory pointing to the central pivot. Compress the direction to obtain a safe reference point after relocation: (4) After obtaining the safe reference point for relocation, based on the ordered set of reference parameters For each parameter range Perform deformable linear transitions to reconstruct the continuous boundary trajectory; the reconstructed continuous boundary trajectory is represented as: (5) This yields a set of safe relocation linear transfer boundary trajectories. Then, a convex combination method is adopted, utilizing the convex combination coefficients. By linearly weighting the linear transition boundary trajectories, an infinite number of feasible and non-intersecting internal curves are generated. ,Right now: (6) This constitutes the set of internal curves for relocation. The output includes a repositioning hub trajectory cluster containing the reconstructed boundary trajectory set and the internal curve set. This provides a safe and feasible optimized local topology space for drone swarms.

4. The method according to claim 1, characterized in that: The specific process of step three is as follows: S31. Local optimization of cluster individuals based on standard model predictive contour control S32. Group model prediction profile control based on pigeon flock coordination progress consistency mechanism A penalty term for consistency in collaborative progress is introduced into the cost function for each individual: (7) in, This represents the consistency penalty coefficient; The introduction of this technology forces the cluster to maintain a synchronized formation topology in a highly dynamic environment, thereby maximizing tracking progress while ensuring safe tracking of the preset internal curve within the central trajectory cluster. S33. Parallel solution of multihomopy optimal control problem based on pigeon bifurcation driving. Combining the contour error, hysteresis error, and consistency penalty obtained in steps S31 and S32, for each individual In the prediction time domain The complete local objective function within is expressed as: (8) The penalty matrix is ​​defined as follows: ,parameter , and Penalty weights representing contour error, hysteresis error, and propulsion speed; positive definite matrix. Used to penalize changes in control input between adjacent time steps; By all The summation of the local costs of each agent, the total traversal cost of the central trajectory cluster is defined as follows: Subsequently, the local planner The trajectory optimization process can be represented as the following optimal control problem: (9) In the optimal control problem, the traversal cost objective function expresses the overall planning objective; the agent dynamics, initial conditions, and state / input constraints of the UAV are strictly limited by the corresponding equations and inclusion relations; the set of feasible states... This constitutes the central trajectory cluster constraint of topological homotopy and the dynamic obstacle avoidance constraint between agents in the optimal control problem.

5. The method according to claim 1, characterized in that: The specific process of step four is as follows: During dynamic navigation, the drone swarm only selects candidate relocation center trajectory clusters. Select and execute the unique optimal central trajectory cluster Since all local planners in each homotopy class minimize local cost functions with identical structures, the traversal cost can be directly obtained by solving the problem. To assess the quality of each central trajectory cluster; Furthermore, a consensus decision weight was introduced. The index of the optimal central trajectory cluster is determined by the following formula: (10)。 6. The method according to claim 5, characterized in that: The specific weight allocation rules are as follows: If the current central trajectory cluster was selected for execution in the previous iteration cycle, priority compensation is given, and the weight is set to... ; Otherwise, it will be considered a candidate central trajectory cluster, and the weight will be set to... .

7. The method according to claim 6, characterized in that: For priority compensation coefficient The physical meaning of the extreme case is as follows: when When the decision-making strategy degenerates into a strict minimum-cost decision, it is entirely inclined to explore new and better topological paths; when At this point, the local planner will maintain the central trajectory cluster of the same homotopy class as the previous iteration cycle and abandon the exploration of new topologies.

8. The method according to claim 1, characterized in that: The specific process of step five is as follows: In static and dynamic mission scenarios of UAV swarm dynamic navigation based on pigeon flock bifurcation and consensus behavior, the position, virtual progress speed and actual speed, and motor thrust of each individual UAV in the swarm are output in real time; the simulation ends when all UAVs have reached the designated area.

Images