Urban multi-source fusion risk field low-altitude multi-unmanned aircraft cooperative path planning method

Abstract

The application discloses a kind of urban multi-source fusion risk field's low-altitude multi-unmanned plane cooperative path planning method, its method includes: constructing multi-unmanned plane cooperative decision model;Cooperative decision rule module carries out joint value evaluation and correction processing to the original flight action of each unmanned plane of multi-unmanned plane system;Cooperative decision rule module has safety action projection mechanism module and safety action judgment rule inside, safety action projection mechanism module uses safety action judgment rule to judge safety constraint to the original flight action after correction processing, and modifies the multiple rounds of processing of safety constraint judgment, safety action projection mechanism module outputs executable safety action;Local re-planning module is based on executable safety action in the future time T1 within local safety trajectory of risk field planning based on executable safety action.This application realizes multi-unmanned plane cooperative path planning, formation switching, local risk response and action correction, and outputs executable path point sequence and control instruction.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) swarm control and path planning technology, and in particular to a low-altitude multi-UAV collaborative path planning method for urban multi-source fusion risk fields, which specifically integrates technologies such as multi-source information perception, risk field modeling, deep reinforcement learning and real-time motion planning. Background Technology

[0002] With the widespread application of drone technology in logistics delivery, urban inspection, emergency rescue, and other fields, traffic density in urban low-altitude airspace has increased dramatically. How to achieve safe and efficient coordinated flight of multiple drones in complex urban low-altitude environments characterized by dense buildings, variable airflow, dynamic obstacles (such as birds and other aircraft), and frequent temporary no-fly zones has become a core technological bottleneck restricting the development of this field.

[0003] Existing multi-UAV cooperative path planning methods mainly fall into three categories, all of which have limitations, as analyzed in detail below: The first category is methods based on graph search and geometric modeling, such as... , Methods such as artificial potential field method and mixed integer programming can obtain optimal solutions in static, regular environments. However, in three-dimensional urban low-altitude scenarios, when multiple strongly coupled constraints such as static buildings, dynamic no-fly zones, airspace boundaries, safe intervals between UAVs, and formation maintenance need to be considered simultaneously, the state space explodes exponentially, the computation time far exceeds real-time requirements, and it is difficult to respond to sudden environmental changes in an instant, resulting in poor scalability. The second category is heuristic optimization-based methods, such as particle swarm optimization and genetic algorithms. Although these methods reduce the difficulty of solving high-dimensional spaces through random search, they generally suffer from problems such as sensitivity to parameter selection, unstable iterative convergence, and susceptibility to getting trapped in local optima. Especially in scenarios requiring rapid and reliable actions, such as traversing narrow passages and avoiding sudden risks, their online adjustment capabilities are insufficient, and they cannot continuously output smooth, directly executable control commands, often leading to oscillations or stagnation of UAVs.

[0004] The third category is path planning methods based on deep reinforcement learning, which typically employ a framework of centralized training and distributed execution. While these methods possess some end-to-end learning capabilities, existing solutions still suffer from four key drawbacks: First, risk representation is fragmented, linearly superimposing static obstacles, dynamic threats, and boundary constraints through independent penalty terms, lacking a unified, physically meaningful joint risk field representation, leading to biases and omissions in the policy's perception of risk. Second, formation is singular and fixed, maintaining the same formation in different scenarios such as open areas, obstacle-dense areas, and narrow passages, failing to adapt to different situations. The adaptive adjustment of local environmental characteristics severely restricts traffic efficiency and safety; third, the continuous actions output by the strategy network (such as acceleration and angular velocity) are usually directly used as control commands, lacking a feasibility verification and correction mechanism for the dynamic limits and safety boundaries of the aircraft (such as minimum obstacle clearance), which can easily lead to dangerous actions that cannot be executed, such as exceeding limits and collisions; fourth, the lack of efficient local replanning and formation recovery mechanisms means that when the local environmental risk increases sharply, the system often needs to perform global replanning, resulting in a delayed response and making it difficult to recover after the formation breaks down, thus the overall system is not robust enough.

[0005] Therefore, existing technologies cannot simultaneously meet the requirements of safety, real-time performance, feasibility, and formation adaptability for multi-UAV collaborative flight in complex urban low-altitude environments. Summary of the Invention

[0006] The purpose of this invention is to solve the problems of dispersed risk representation, fixed formation, insufficient action safety, and lag in dynamic response in the prior art. It provides a low-altitude multi-UAV cooperative path planning method for urban multi-source fusion risk fields. Under the premise of satisfying flight dynamics constraints, obstacle avoidance constraints, no-fly zone constraints, boundary constraints, safety interval constraints between UAVs, and energy constraints, it can realize multi-UAV cooperative path planning, formation switching, local risk response, and action correction, and output executable path point sequences and control commands.

[0007] The objective of this invention is achieved through the following technical solution: A method for low-altitude multi-UAV cooperative path planning in urban multi-source fusion risk fields, the method comprising: S1. Construct a multi-UAV collaborative decision-making model. The multi-UAV collaborative decision-making model includes a low-altitude three-dimensional collaborative task scenario, a multi-UAV monitoring system, a formation switching rule module, and a local replanning module. The low-altitude three-dimensional collaborative task scenario is internally set with a bounded flyable domain and a risk field created by the fusion of multi-source scene data. The multi-UAV monitoring system stores a task set containing several flight stages and a collaborative decision-making rule module preset according to the flight stage. The multi-source scene data includes local observation data and external observation data of the multi-UAV system, no-fly restriction information, and obstacle data. S2. The multi-UAV monitoring system monitors and obtains the original flight actions of each UAV in the multi-UAV system at the current time. The formation switching rule module has preset formation switching rules and adjusts the UAVs in the multi-UAV system to adopt formation. The collaborative decision-making rule module performs joint value assessment and correction processing on the original flight actions of each UAV in the multi-UAV system. S3, the collaborative decision-making rule module contains a safety action projection mechanism module and a safety action judgment rule. The safety action projection mechanism module uses the safety action judgment rule to make safety constraint judgments on the original flight action after modification, as well as multiple rounds of modification processing and safety constraint judgment. The safety action projection mechanism module outputs executable safety actions. S4. The local replanning module plans a local safe trajectory within a future time T1 in the risk field based on executable safe actions.

[0008] To better achieve the present invention, the present invention also includes the following methods: S5. The multi-UAV collaborative decision-making model maps the executable safety actions and local safety trajectories of each UAV in the multi-UAV system into a set of sequential control commands for each UAV within a future time T1 and sends them to the UAV's onboard controller or / and the ground station control terminal.

[0009] Preferably, in method S1, the local observation data of the multi-UAV system is UAV aerial data obtained through UAV onboard sensors, including position, speed, yaw angle, pitch angle, relative position of neighboring UAVs, flight phase, and mission target guidance information; the external observation data of the multi-UAV system is dynamic and static obstacle data obtained through UAV visual sensors; the no-fly zone restriction information is no-fly zone data and temporary airspace restriction zone data obtained through external airspace interfaces; and the obstacle data is a set of obstacle data within the bounded flyable domain of the low-altitude three-dimensional collaborative mission scenario.

[0010] Furthermore, in method S1, the risk field is constructed based on the comprehensive risk of the risk function, and the expression of the risk function in the risk field is as follows: ,in For the UAV in position vector ,time Comprehensive risks Static risk is constructed based on the distance between the drone and static obstacles. For dynamic risk based on the distance between the drone and the center of the dynamic obstacle, To assess boundary proximity risk based on the shortest distance between the drone and the boundary of the flyable domain, The risk of a safety interval is constructed based on the deviation between the actual spacing and the preset safety interval. The energy cost risk is constructed based on a combination of speed, acceleration, and remaining range. This is the risk weighting coefficient.

[0011] Preferably, the multi-UAV system includes one lead UAV and at least two follower UAVs; the flight phases of the mission division include takeoff and convergence phase, cruise search phase, narrow passage crossing phase, local avoidance phase, formation restoration phase, and terminal approach phase.

[0012] Preferably, in method S3, the formation switching rules in the formation switching rule module include the following: When the local passage width is higher than the first width threshold and the obstacle density is lower than the first density threshold, a horizontal formation is used; when the local passage width is lower than the first width threshold or the overall risk in the horizontal formation flight risk field increases rapidly, a column formation is used; when the obstacle density is greater than the first density threshold or when coordinated obstacle avoidance is required, a diamond formation or a column formation is used. The formation switching rule module obtains the local channel width and obstacle density in front of the overall flight path through the UAV vision sensor of the multi-UAV system to determine the formation switching rule. If the switching conditions are met for several consecutive time periods, the formation switching process is performed.

[0013] Preferably, in method S2, the multi-UAV system includes one lead UAV and at least two follower UAVs. For the lead UAV, a multi-objective function is constructed, including the mission target direction, the overall risk distribution ahead, obstacle avoidance constraints, flight boundary constraints, and mission advancement efficiency. The multi-objective function constraints aim to avoid areas with high overall risk and advance towards the mission target direction to generate the lead UAV's original flight maneuvers. The follower UAVs generate their original flight maneuvers by following the lead UAV and constructing distance constraints and current formation constraints with the lead UAV or adjacent follower UAVs. The collaborative decision-making rule module constructs joint value assessment rules, which include: the current action combination enables the UAV swarm to stably approach the mission target; the current action combination can avoid high-risk areas; the current action combination can maintain a safe distance; the current action combination can maintain the target formation and energy consumption is within a preset allowable range. The joint value assessment rules are used to perform a joint value assessment on the original flight maneuvers of each UAV in the multi-UAV system. If the joint value assessment passes, the original flight maneuver combination of each UAV is adopted. If the joint value assessment fails, the collaborative decision-making rule module performs correction processing until the joint value assessment passes.

[0014] Preferably, the collaborative decision-making rule module internally constructs a key flight segment priority invocation mechanism module. The key flight segments include at least the following: segments where the UAV approaches a building, a dynamic no-fly zone, or a flight boundary; segments where the safety distance between UAVs is insufficient; segments where the following UAVs significantly deviate from their formation position; segments where the UAV swarm fails to effectively approach the mission target; and segments where flight maneuvers lead to abnormally high energy consumption or a rapid increase in local risks. The key flight segment priority invocation mechanism module uses historical replays to obtain the correction processing corresponding to the key flight segments as correction processing experience samples. The collaborative decision-making rule module uses the correction processing experience samples for learning and training and promotes rapid correction processing until it passes the joint value assessment.

[0015] Preferably, in method S3, the original flight maneuver after modification is judged by flight prediction. The safety constraint judgment of the safety action judgment rule includes the speed exceeding the limit, acceleration exceeding the limit, excessive yaw rate, insufficient obstacle clearance, insufficient safety distance between UAVs, approaching the flight boundary, or insufficient remaining energy that the original flight maneuver prediction will cause. If the original flight maneuver is modified to obtain a set of executable actions, and the modification is adjusted under common constraint conditions including speed constraint, acceleration constraint, yaw rate constraint, pitch angle constraint, obstacle clearance constraint, safety distance between UAVs constraint, flight boundary constraint, and remaining energy constraint, then the safety action projection mechanism module selects the executable action that is closest to the original flight maneuver from the set of executable actions as the executable safety action.

[0016] Preferably, the local replanning module is configured with the following replanning mechanism, which initiates local replanning processing when any of the following triggering conditions are met: a. The peak value of the local risk field exceeds the preset threshold; b. Collisions may occur within the predicted time window T2; c. The actual formation deviation during flight continuously exceeds the permissible threshold; d. The efficiency of target advancement remains below the threshold.

[0017] Compared with the prior art, the present invention has the following advantages and beneficial effects: This invention integrates and creates a risk field in low-altitude three-dimensional collaborative mission scenarios, incorporating multi-source scene data such as static, dynamic, boundary, interval, and energy data. It can match the spatiotemporal continuous safety constraints of various collaborative flights of multiple UAVs within a bounded flyable domain and wind conditions. It can naturally weigh different types of risks and make truly globally optimal decisions, avoiding the problem of neglecting some risks due to partial risk components, and fundamentally solving the problem of fragmented risk perception. Under the premise of satisfying flight dynamics constraints, obstacle avoidance constraints, no-fly zone constraints, boundary constraints, safety interval constraints between UAVs, and energy constraints, this invention can realize multi-UAV collaborative path planning, formation switching, local risk response, and action correction, and output executable path point sequences and control commands. Attached Figure Description

[0018] Figure 1 This is a flowchart of the low-altitude multi-UAV cooperative path planning method of the present invention; Figure 2 This is a schematic diagram illustrating the principle of the low-altitude multi-UAV cooperative path planning method in the embodiment. Figure 3 This is a schematic diagram illustrating the formation switching of multiple drones at different flight phases, as exemplified in the embodiment. Figure 4 This is a schematic diagram illustrating the principle of local replanning in the example. Detailed Implementation

[0019] The present invention will be further described in detail below with reference to embodiments: Example like Figure 1 , Figure 2 As shown, a low-altitude multi-UAV cooperative path planning method for urban multi-source fusion risk fields includes the following steps: S1. Construct a multi-UAV collaborative decision-making model. This model includes a low-altitude 3D collaborative mission scenario, a multi-UAV monitoring system, a formation switching rule module, and a local replanning module. The low-altitude 3D collaborative mission scenario internally sets up a bounded flyable domain and a risk field created by fusing multi-source scenario data (the wind field is a unified 3D space constructed by mapping the fused multi-source scenario data to the final unified multi-UAV collaborative decision-making model, and the wind field covers the entire bounded flyable domain, providing consistent situational awareness for all UAVs). It receives multi-source scenario data from airborne sensors, neighboring UAV communication links, ground station mission systems, and external airspace interfaces, and establishes a fused risk field accordingly. In this embodiment, the local observation data of the multi-UAV system is UAV aerial data obtained through UAV airborne sensors. Airborne sensors include at least one or more of the following: visual sensors, lidar, millimeter-wave radar, inertial measurement units, and satellite positioning modules. UAV aerial data includes position, speed, yaw angle, pitch angle, relative position of neighboring UAVs, flight phase, and mission target guidance information. The external observation data of the multi-UAV system is dynamic and static obstacle data obtained through UAV visual sensors. No-fly zone information is dynamically updated information such as no-fly zone data and temporary airspace restriction zone data obtained through external airspace interfaces. Obstacle data is a set of obstacle data within the bounded flyable domain of the low-altitude three-dimensional collaborative mission scenario.

[0020] This embodiment selects a 3D urban low-altitude environment of 3000m×3000m×300m to construct a low-altitude 3D collaborative mission scenario. Within the scenario, 35 static building obstacles, 8 dynamic no-fly zones, and several inspection target points are set. The multi-UAV system consists of 1 lead UAV and 3 follower UAVs. The dynamic parameters of each UAV can be set according to Table 1.

[0021] In some embodiments, the risk field is constructed based on the comprehensive risk of the risk function, and the expression of the risk function in the risk field is as follows: ,in For the UAV in position vector ,time Comprehensive risks Based on the distance between the drone and static obstacles (such as buildings) Static risks in construction , This is the gain coefficient. To prevent extremely small positive numbers with a denominator of zero, the static risk increases dramatically when the drone approaches the surface of any static obstacle (such as a building). Based on the distance between the drone and the center of the dynamic obstacle The dynamic risk can be constructed using the same formula as the static risk. The dynamic risk increases as the drone approaches any dynamic obstacle. Based on the shortest distance between the drone and the boundary of the flyable domain The boundary proximity risk is constructed using the static risk formula. As the UAV approaches the boundary of the flyable domain, the boundary proximity risk increases, as shown in the following expression: , These are preset parameters. To prevent extremely small positive numbers with a denominator of zero. To mitigate the risk of a safety distance based on the deviation between the actual spacing and a preset safety interval among drones in a multi-UAV system, this embodiment uses the actual spacing between the i-th and j-th drones in the formation. With preset safety interval The deviation can be constructed using a Gaussian function or a quadratic penalty function. In this embodiment, the safety interval risk value is taken as follows: A risk penalty will be incurred when the actual distance is less than the safe distance. This embodiment uses speed as the energy cost risk, which is constructed based on a combination of speed, acceleration, and remaining range. acceleration and remaining range It can be obtained using the following formula: , , , These are preset parameters. Preset, non-negative risk weighting coefficients (used to adjust the relative importance of various risks according to task requirements) satisfy... The example values ​​in this embodiment are as follows: , , , , .

[0022] The multi-UAV monitoring system stores a set of tasks comprising several flight phases and a pre-defined collaborative decision-making rule module based on the flight phases. Preferably, the flight phases of the task division include a takeoff and convergence phase, a cruise search phase, a narrow passage crossing phase, a local avoidance phase, a formation restoration phase, and a terminal approach phase. The multi-source scene data in this embodiment includes local observation data and external observation data of the multi-UAV system, no-fly zone restriction information, and obstacle data. Preferably, the multi-UAV system includes one lead UAV and at least two follower UAVs.

[0023] S2. The multi-UAV monitoring system monitors and obtains the original flight actions of each UAV in the multi-UAV system at the current time. The formation switching rule module has preset formation switching rules and adjusts the UAVs in the multi-UAV system to adopt a formation. The collaborative decision-making rule module performs joint value assessment and correction processing on the original flight actions of each UAV in the multi-UAV system.

[0024] In some embodiments, a multi-UAV system includes one lead UAV and at least two follower UAVs. For the lead UAV, a multi-objective function is constructed, incorporating the mission objective direction, the overall forward risk distribution, obstacle avoidance constraints, flight boundary constraints, and mission propulsion efficiency. This multi-objective function constraint generates the lead UAV's initial flight maneuvers based on data such as the location, flight speed, attitude angles, relative positions of adjacent UAVs, local risk fields, target direction, and flight phase, aiming to avoid areas with high overall risk and propel the UAV towards the mission objective direction. The initial flight maneuvers include at least longitudinal acceleration, yaw rate, and pitch rate, used to control the UAV's speed changes, horizontal turning, and altitude adjustments. The follower UAVs generate their initial flight maneuvers by following the lead UAV and constructing distance constraints and current formation constraints (enabling the follower UAVs to maintain target formation during obstacle avoidance and gradually recover to target formation positions after deviation) with respect to the lead UAV.

[0025] To avoid a decline in overall coordination caused by each UAV acting independently based on its own local information, the system combines the state information, flight stage, target formation, fusion risk field, and original flight actions of each UAV after generating its initial flight maneuvers, forming multi-UAV joint state-action information. Based on this joint state-action information, the system performs a unified evaluation of the overall effect of the current action combination. The overall effect evaluation includes at least the following: mission advancement effect evaluation, obstacle avoidance effect evaluation, no-fly zone avoidance effect evaluation, boundary maintenance effect evaluation, safe interval between UAVs evaluation, formation stability evaluation, and energy consumption evaluation. Through these evaluations, this embodiment determines whether the current combination of UAV actions is conducive to the safe, stable, and coordinated completion of the current stage of the task by the multi-UAV system. The collaborative decision-making rule module constructs joint value evaluation rules, which include: the current action combination enables the UAV swarm to stably approach the mission target; the current action combination avoids high-risk areas; the current action combination maintains a safe interval; and the current action combination maintains the target formation while energy consumption remains within a preset allowable range. The joint value assessment rules are used to evaluate the original flight maneuvers of each UAV in the multi-UAV system. If the joint value assessment passes, the combination of the original flight maneuvers of each UAV is adopted. If the joint value assessment fails, the collaborative decision-making rule module performs corrections until the joint value assessment passes. In this way, each UAV can quickly generate flight maneuvers based on local observation information during execution, while the system can make unified judgments on the joint behavior of multiple UAVs at the overall level. This balances the real-time performance of the individual UAVs' dispersed actions in the formation with the consistency of multi-UAV collaboration, and provides a basis for subsequent key flight segment identification, safety action correction, and local replanning.

[0026] S3, the collaborative decision-making rule module contains a safety action projection mechanism module and a safety action judgment rule. The safety action projection mechanism module uses the safety action judgment rule to perform safety constraint judgment on the original flight action after modification, as well as multiple rounds of modification processing and safety constraint judgment. The safety action projection mechanism module outputs the executable safety actions corresponding to each UAV in the multi-UAV system.

[0027] In some embodiments, such as Figure 3 As shown, the formation switching rule module stores the local channel widths monitored by each drone. Obstacle density Target direction deviation and flight phase markers The formation switching rules in the formation switching rules module include the following: When the local passage width is higher than the first width threshold and the obstacle density is lower than the first density threshold, a horizontal formation is used; when the local passage width is lower than the first width threshold or the overall risk in the horizontal formation flight risk field increases rapidly, a column formation is used; when the obstacle density is greater than the first density threshold or when cooperative obstacle avoidance is required, a diamond formation or a column formation is used; ...

[0028] The formation switching rule module acquires local channel width and obstacle density in front of the multi-UAV system through UAV vision sensors to determine formation switching rules. If the switching conditions are met for several consecutive time periods, formation switching is performed (formation switching is executed when the switching conditions are met for multiple consecutive sampling periods; during the switching process, the relative displacement change rate of the following UAVs per unit time is limited to reduce the possibility of the formation breaking apart). An example is shown below: The following switching logic can be set: When... and When, use a horizontal line; when When in the middle section or needing to avoid obstacles on both sides, a diamond formation should be used; when or At this time, a column formation is adopted. Formation switching adopts a hysteresis determination method, and template switching is performed when the switching conditions are met for three consecutive sampling periods. During switching, the lead UAV maintains the target's advancement direction basically unchanged, while the follower UAVs gradually adjust according to the target's relative pose sequence.

[0029] In some embodiments, the collaborative decision-making rule module also includes a key flight segment priority recall mechanism module. During the execution of urban low-altitude missions by multiple UAVs, the system continuously records the flight status, original flight maneuvers, risk changes, formation changes, mission progress, and corresponding feedback results of each UAV at different times, forming a mission execution record. Since different flight processes have different reference values ​​for subsequent collaborative control parameter adjustments, the system does not recall all mission execution records equally, but prioritizes extracting key flight segments that have a significant impact on safe flight, formation maintenance, and mission progress. Preferably, key flight segments include at least the following: segments where UAVs approach buildings, dynamic no-fly zones, or flight boundaries; segments where the safety distance between UAVs is insufficient; segments where following UAVs significantly deviate from their formation positions; segments where the UAV swarm fails to effectively approach the mission target; and segments where flight maneuvers lead to abnormally high energy consumption or rapidly increasing local risks. The key flight segment priority recall mechanism module uses historical playback to obtain the correction processing corresponding to key flight segments as correction processing experience samples. The collaborative decision-making rule module uses these correction processing experience samples for learning and training, promoting rapid correction processing until it passes the joint value assessment.

[0030] To determine the importance of each flight segment, the system evaluates it from four aspects: action judgment deviation, safety risk level, formation deviation, and mission stagnation. Action judgment deviation reflects the difference between the system's original flight actions and the actual feedback results. Safety risk level reflects whether there are situations such as insufficient obstacle clearance, approaching a dynamic no-fly zone, approaching a flight boundary, or insufficient safe spacing between UAVs within the segment. Formation deviation reflects the deviation of each UAV from the target formation position. Mission stagnation reflects whether the UAV swarm experiences a decrease in target approach speed or a significant slowdown in mission progress. Based on these evaluation results, the system assigns a priority to each flight segment. Flight segments with large action judgment deviations, high safety risks, significant formation deviations, or mission stagnation are assigned higher priority; flight segments with stable flight status, low risk levels, good formation maintenance, and normal mission progress are assigned lower priority. The priority can be expressed as:

[0031] in, Indicates the priority of flight segments. This indicates the deviation in action judgment (in this embodiment, it is preferably the average value of the timing difference error within the segment). This indicates the degree of security risk (in this embodiment, it is preferably the quantification of the average security risk within the segment). This indicates the degree of formation deviation (in this embodiment, the average value of formation deviation within a segment is preferred, and segments with severely dispersed formations have high priority). This indicates the degree of stagnation in task progress (in this embodiment, it is preferably the average value of the distance approaching the target point per unit time within the segment). , , , This represents the corresponding weight coefficient. This represents a small constant used to avoid zero priority. The degree of safety risk can be characterized by the minimum clearance gap, the degree of formation deviation can be characterized by the combined deviation of adjacent aircraft spacing and azimuth, and the degree of mission stagnation can be characterized by insufficient target distance descent at adjacent moments. Through priority design, the system can prioritize high-risk, low-progress, and easily unstable flight segments during subsequent adjustments to collaborative control parameters. During these adjustments, the system prioritizes high-priority flight segments, adjusting control parameters and action generation rules that are prone to increased risk, formation breakup, mission stagnation, or abnormal energy consumption. This key flight segment priority mechanism avoids the problem of a large number of smooth flight processes masking critical risk processes, enabling the system to have more stable collaborative control effects in scenarios such as narrow passage crossings, approaching dynamic no-fly zones, local obstacle avoidance, insufficient UAV spacing, formation recovery, and target advancement obstruction.

[0032] In some embodiments, a flight prediction judgment is performed on the original flight maneuver after modification (the original flight maneuver includes longitudinal acceleration, yaw rate, and pitch rate; since the original flight maneuver may lead to speed exceeding limits, acceleration exceeding limits, excessive yaw rate, insufficient obstacle clearance, insufficient safety distance between UAVs, approaching the flight boundary, or insufficient remaining energy, flight prediction judgment is required). The safety constraint judgment of the safety action judgment rule includes the speed exceeding limits, acceleration exceeding limits, excessive yaw rate, insufficient obstacle clearance, insufficient safety distance between UAVs, approaching the flight boundary, or insufficient remaining energy that the original flight maneuver prediction may lead to. If the original flight maneuver is modified to obtain a set of executable actions, and the modification is adjusted under common constraints including speed constraints, acceleration constraints, yaw rate constraints, pitch rate constraints, obstacle clearance constraints, safety distance constraints between UAVs, flight boundary constraints, and remaining energy constraints, then the safety action projection mechanism module selects the executable action closest to the original flight maneuver from the set of executable actions as the executable safety action. The safety action projection mechanism module contains a safety action projection function. Safety action projection can be implemented through a quadratic programming solver, a hierarchical constraint pruning method, or a sequential projection method. The quadratic programming solver is used to directly obtain the executable action with the minimum distance from the original flight action under all constraints. The hierarchical constraint pruning method is used to modify the original action layer by layer according to the order of dynamic constraints, safety interval constraints, obstacle clearance constraints, boundary constraints, and energy constraints. The sequential projection method is used to project the original action sequentially into the executable range corresponding to different constraints until an executable action that satisfies the constraints is obtained.

[0033] S4. The local replanning module plans a local safe trajectory within a future time T1 in the risk field based on executable safety actions. In some embodiments, the local replanning module sets the following replanning mechanism: when any of the following triggering conditions are met, the local replanning process (regenerating local track segments and formation instructions) is initiated: a. The peak value of the local risk field is higher than the preset threshold.

[0034] b. Collisions may occur within the predicted time window T2.

[0035] c. The actual formation deviation during flight continuously exceeds the permissible threshold.

[0036] d. The efficiency of target advancement remains below the threshold.

[0037] In some embodiments, such as Figure 4 As shown, local replanning only applies to the future. Local trajectories within a given walking distance are refreshed, while previously executed trajectories and stable distant tracks are retained. Local replanning generates local track segments using candidate track sampling based on the current risk field, local cost minimization search, or policy network rolling prediction. An example is shown below: In the future... Within a local window of varying distance, the risk field and obstacle predictions are updated, refreshing only the track segments within that local window, and simultaneously re-selecting the formation based on the latest channel conditions. Under a set of exemplary simulation conditions, the performance of the standard MADDPG, the fixed formation with no motion projection scheme, and the technical solution of this invention were compared, and the results are shown in Table 2.

[0038]

[0039] S5. The multi-UAV collaborative decision-making model maps the executable safety actions and local safety trajectories of each UAV in the multi-UAV system into a set of sequential control commands for each UAV within a future time T1, and sends them to the UAV's onboard controller and / or the ground station control terminal. The set of sequential control commands includes the three-dimensional waypoint sequence, velocity command, yaw rate command, pitch rate command, and formation switching command for each UAV. The set of sequential control commands is converted into trajectory tracking control quantities via the onboard flight control interface, or sent to the execution terminal of each UAV via the ground station mission management interface.

[0040] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A low-altitude multi-UAV cooperative path planning method for urban multi-source fusion risk fields, characterized in that: The methods include: S1. Construct a multi-UAV collaborative decision-making model. The multi-UAV collaborative decision-making model includes a low-altitude three-dimensional collaborative task scenario, a multi-UAV monitoring system, a formation switching rule module, and a local replanning module. The low-altitude three-dimensional collaborative task scenario is internally set with a bounded flyable domain and a risk field created by the fusion of multi-source scene data. The multi-UAV monitoring system stores a task set containing several flight stages and a collaborative decision-making rule module preset according to the flight stage. The multi-source scene data includes local observation data and external observation data of the multi-UAV system, no-fly restriction information, and obstacle data. S2. The multi-UAV monitoring system monitors and obtains the original flight actions of each UAV in the multi-UAV system at the current time. The formation switching rule module has preset formation switching rules and adjusts the UAVs in the multi-UAV system to adopt formation. The collaborative decision-making rule module performs joint value assessment and correction processing on the original flight actions of each UAV in the multi-UAV system. S3, the collaborative decision-making rule module contains a safety action projection mechanism module and a safety action judgment rule. The safety action projection mechanism module uses the safety action judgment rule to make safety constraint judgments on the original flight action after modification, as well as multiple rounds of modification processing and safety constraint judgment. The safety action projection mechanism module outputs executable safety actions. S4. The local replanning module plans a local safe trajectory within a future time T1 in the risk field based on executable safe actions.

2. The low-altitude multi-UAV cooperative path planning method for urban multi-source fusion risk fields according to claim 1, characterized in that: It also includes the following methods: S5. The multi-UAV collaborative decision-making model maps the executable safety actions and local safety trajectories of each UAV in the multi-UAV system into a set of sequential control commands for each UAV within a future time T1 and sends them to the UAV's onboard controller or / and the ground station control terminal.

3. The low-altitude multi-UAV cooperative path planning method for urban multi-source fusion risk fields according to claim 1, characterized in that: In method S1, the local observation data of the multi-UAV system is UAV aerial data obtained through UAV onboard sensors. The UAV aerial data includes position, speed, yaw angle, pitch angle, relative position of neighboring UAVs, flight phase, and mission target guidance information. The external observation data of the multi-UAV system is dynamic and static obstacle data obtained through UAV visual sensors. The no-fly zone restriction information is no-fly zone data and temporary airspace restriction zone data obtained through external airspace interfaces. The obstacle data is a set of obstacle data within the bounded flyable domain of the low-altitude three-dimensional collaborative mission scenario.

4. The low-altitude multi-UAV cooperative path planning method for urban multi-source fusion risk fields according to claim 1, characterized in that: In method S1, the risk field is constructed based on the comprehensive risk of the risk function, and the expression of the risk function in the risk field is as follows: ,in For the UAV in position vector ,time Comprehensive risks Static risk is constructed based on the distance between the drone and static obstacles. For dynamic risk based on the distance between the drone and the center of the dynamic obstacle, To assess boundary proximity risk based on the shortest distance between the drone and the boundary of the flyable domain, The risk of a safety interval is constructed based on the deviation between the actual spacing and the preset safety interval. The energy cost risk is constructed based on a combination of speed, acceleration, and remaining range. This is the risk weighting coefficient.

5. The low-altitude multi-UAV cooperative path planning method for urban multi-source fusion risk fields according to claim 1, characterized in that: The multi-UAV system includes one lead UAV and at least two follower UAVs; the flight phases of the mission are divided into takeoff and convergence phase, cruise search phase, narrow passage crossing phase, local avoidance phase, formation restoration phase, and terminal approach phase.

6. The low-altitude multi-UAV cooperative path planning method for urban multi-source fusion risk fields according to claim 1, characterized in that: In method S3, the formation switching rules in the formation switching rule module include the following: When the local passage width is higher than the first width threshold and the obstacle density is lower than the first density threshold, a horizontal formation is used; when the local passage width is lower than the first width threshold or the overall risk in the horizontal formation flight risk field increases rapidly, a column formation is used; when the obstacle density is greater than the first density threshold or when coordinated obstacle avoidance is required, a diamond formation or a column formation is used. The formation switching rule module obtains the local channel width and obstacle density in front of the overall flight path through the UAV vision sensor of the multi-UAV system to determine the formation switching rule. If the switching conditions are met for several consecutive time periods, the formation switching process is performed.

7. A low-altitude multi-UAV cooperative path planning method for urban multi-source fusion risk fields according to claim 1 or 6, characterized in that: In method S2, the multi-UAV system includes one lead UAV and at least two follower UAVs. For the lead UAV, a multi-objective function is constructed, which includes the mission objective direction, the overall risk distribution ahead, obstacle no-fly avoidance constraints, flight boundary constraints, and mission propulsion efficiency. The multi-objective function constraints are used to avoid areas with high overall risk and propel the UAV toward the mission objective direction to generate the original flight maneuvers of the lead UAV. The follower UAVs generate their original flight maneuvers by following the lead UAV and constructing distance constraints and current formation constraints with the lead UAV or adjacent follower UAVs. The collaborative decision-making rule module constructs joint value assessment rules, which include: the current action combination enables the UAV swarm to stably approach the mission target; the current action combination can avoid high-risk areas; the current action combination can maintain a safe distance; and the current action combination can maintain target formation while energy consumption is within a preset allowable range. The joint value assessment rules are used to perform a joint value assessment on the original flight actions of each UAV in the multi-UAV system. If the joint value assessment passes, the original flight action combination of each UAV is adopted; if the joint value assessment fails, the collaborative decision-making rule module performs correction processing until the joint value assessment passes.

8. The low-altitude multi-UAV cooperative path planning method for urban multi-source fusion risk fields according to claim 7, characterized in that: The collaborative decision-making rule module internally constructs a key flight segment priority invocation mechanism module. The key flight segments include at least the following: segments where the UAV approaches a building, a dynamic no-fly zone, or a flight boundary; segments where the safety distance between UAVs is insufficient; segments where the following UAVs significantly deviate from their formation position; segments where the UAV swarm fails to effectively approach the mission target; and segments where flight maneuvers lead to abnormally high energy consumption or a rapid increase in local risks. The key flight segment priority invocation mechanism module uses historical replays to obtain the correction processing corresponding to the key flight segments as correction processing experience samples. The collaborative decision-making rule module uses the correction processing experience samples for learning and training and promotes rapid correction processing until it passes the joint value assessment.

9. The low-altitude multi-UAV cooperative path planning method for urban multi-source fusion risk fields according to claim 1, characterized in that: In method S3, the original flight maneuver after modification is evaluated for flight prediction. The safety constraint judgment of the safety action judgment rule includes the speed exceeding the limit, acceleration exceeding the limit, excessive yaw rate, insufficient obstacle clearance, insufficient safety distance between UAVs, approaching the flight boundary, or insufficient remaining energy that the original flight maneuver prediction may cause. If the original flight maneuver is modified to obtain a set of executable actions, and the modification is adjusted under common constraints including speed constraint, acceleration constraint, yaw rate constraint, pitch angle constraint, obstacle clearance constraint, safety distance between UAVs constraint, flight boundary constraint, and remaining energy constraint, then the safety action projection mechanism module selects the executable action that is closest to the original flight maneuver from the set of executable actions as the executable safety action.

10. The low-altitude multi-UAV cooperative path planning method for urban multi-source fusion risk fields according to claim 1, characterized in that: The local replanning module is configured with the following replanning mechanism, which initiates local replanning processing when any of the following triggering conditions are met: a. The peak value of the local risk field exceeds the preset threshold; b. Collisions may occur within the predicted time window T2; c. The actual formation deviation during flight continuously exceeds the permissible threshold; d. The efficiency of target advancement remains below the threshold.

Images