Urban low-altitude unmanned aerial vehicle multi-constraint path collaborative planning method and system

Abstract

The invention discloses an urban low-altitude unmanned aerial vehicle multi-constraint path collaborative planning method and system, and relates to the technical field of unmanned aerial vehicle control, and the method comprises the following steps: obtaining the three-dimensional grid environment data of a target region and the real-time operation state data of a local vehicle, and carrying out the fusion calculation of the multi-dimensional constraint feature data of each candidate node, the multi-dimensional constraint feature data at least comprises energy consumption constraint feature data and static obstacle avoidance constraint feature data. According to the invention, by introducing a dual cooperative discrimination mechanism of the conflict correlation degree representing the constraint conflict degree and the real-time power consumption rate, through the multi-dimensional cross verification, the comprehensive survival crisis faced by the unmanned aerial vehicle can be identified more accurately and in advance; the situation that the unmanned aerial vehicle is caught in a deadlock state of top-speed failure of electric quantity due to blind compromise in space risk avoiding is effectively avoided.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, specifically to a multi-constraint path collaborative planning method and system for urban low-altitude UAVs. Background Technology

[0002] With the booming development of the low-altitude economy, urban low-altitude drones are increasingly being used in logistics, security patrols, environmental monitoring, and emergency rescue. Unlike the open high-altitude environment, urban low-altitude airspace is characterized by dense buildings, complex weather conditions, and intricate air traffic networks. Therefore, generating a safe and efficient three-dimensional cooperative flight path for drones under multiple constraints has become a core technical challenge in the field of drone autonomous control and air traffic management.

[0003] Currently, existing UAV multi-constraint path planning is typically based on 3D grids or spatial node networks, employing heuristic search algorithms, multi-objective optimization algorithms, or swarm intelligence algorithms for trajectory searching. When dealing with multiple constraints such as obstacle avoidance, distance, and energy consumption, existing technologies generally adopt the method of constructing a comprehensive cost function, that is, assigning weight coefficients to each constraint feature and evaluating and selecting candidate nodes through linear weighted summation.

[0004] In complex urban environments, strong contradictions and conflicts often arise between various constraints. For example, to avoid large areas of dense buildings, drones must make significant detours, resulting in a sharp increase in flight distance. Existing technologies typically use fixed baseline weights or simple linear dynamic weights, lacking a quantitative assessment mechanism for the degree of such constraint contradictions. This leads algorithms to blindly compromise between multiple objectives when faced with extreme conflicts, ultimately planning an infeasible path that, while meeting avoidance requirements, consumes extremely high energy or even exceeds the drone's maximum endurance. Furthermore, most existing path planning algorithms, when considering energy constraints, often rely solely on static calculations based on the drone's absolute remaining battery power or the estimated energy consumption at the current node, severely neglecting the real-time rate of power consumption. When a drone encounters strong headwinds or performs emergency avoidance maneuvers, the rate of power consumption surges instantaneously. If the algorithm continues to perform smooth calculations using conventional weights at this time, the system's response to the energy crisis will be severely delayed, easily causing the drone to run out of power en route and crash, leading to serious safety accidents. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a multi-constraint path collaborative planning method and system for urban low-altitude unmanned aerial vehicles (UAVs).

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: A multi-constraint path collaborative planning method for urban low-altitude unmanned aerial vehicles (UAVs) includes the following steps: The system acquires three-dimensional grid environment data of the target area and real-time operating status data of the local machine, and integrates and calculates multi-dimensional constraint feature data of each candidate node. The multi-dimensional constraint feature data includes at least energy consumption constraint feature data and static obstacle avoidance constraint feature data. Numerical quantization processing is performed on multidimensional constraint feature data to generate benchmark weight coefficients for each constraint feature data; The difference between the baseline weight coefficient of the energy consumption constraint feature data and the baseline weight coefficient of other constraint feature data is calculated to generate a conflict correlation degree that characterizes the degree of constraint contradiction, and the real-time power consumption rate is extracted from the real-time operating status data of the local machine. Determine whether both the conflict correlation degree and the consumption rate are greater than the corresponding thresholds; If so, the baseline weight coefficients of the energy consumption constraint feature data are calculated exponentially, and the baseline weight coefficients of the remaining constraint feature data are calculated proportionally to generate the final weight coefficients of each constraint feature data. Otherwise, the baseline weight coefficients of each constraint feature data will be used as the final weight coefficients; The comprehensive cost of the candidate node is obtained by weighting and fusing the multidimensional constraint feature data with the final weight coefficients of each constraint feature data. In the 3D raster node network data, the candidate node with the lowest comprehensive cost value is extracted as the target node and connected to generate a cooperative path trajectory sequence and output it.

[0007] A multi-constraint path collaborative planning system for urban low-altitude unmanned aerial vehicles (UAVs) includes: The data acquisition module is used to acquire three-dimensional raster environment data of the target area and real-time running status data of the local machine, and to fuse and calculate the multi-dimensional constraint feature data of each candidate node. The multi-dimensional constraint feature data includes at least energy consumption constraint feature data and static obstacle avoidance constraint feature data. The data processing module is used to perform numerical quantization processing on multidimensional constraint feature data to generate the benchmark weight coefficients of each constraint feature data; calculate the difference between the benchmark weight coefficients of energy consumption constraint feature data and the benchmark weight coefficients of other constraint feature data to generate the conflict correlation degree that characterizes the degree of constraint contradiction; and extract the real-time power consumption rate from the real-time operating status data of the local machine. The data judgment and weight adjustment module is used to determine whether the conflict correlation degree and consumption rate are both greater than the corresponding thresholds; if so, the baseline weight coefficients of the energy consumption constraint feature data are calculated exponentially, and the baseline weight coefficients of the remaining constraint feature data are calculated proportionally to generate the final weight coefficients of each constraint feature data; otherwise, the baseline weight coefficients of each constraint feature data are used as the final weight coefficients. The data output module is used to perform weighted fusion calculation of multidimensional constraint feature data and the final weight coefficients of each constraint feature data to obtain the comprehensive cost value of candidate nodes; in the three-dimensional grid node network data, the candidate node with the smallest comprehensive cost value is extracted as the target node and connected to generate a collaborative path trajectory sequence and output it.

[0008] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention introduces a dual collaborative discrimination mechanism that combines conflict correlation degree (characterizing the degree of constraint contradiction) and real-time power consumption rate. This enables drones to not only perceive the rate of energy loss in real time, but also to accurately quantify the degree of antagonism between current spatial constraints (such as obstacle avoidance and detour due to dense building clusters) and energy consumption limitations. Through this multi-dimensional cross-verification, the comprehensive survival crisis faced by drones can be identified more accurately and proactively, effectively preventing drones from falling into a deadlock state of rapid power depletion due to blindly compromising with spatial avoidance. By introducing a nonlinear polarization weight adjustment strategy, the survivability of UAVs under extreme conditions is effectively improved. For extreme flight conditions with severe conflicts of multiple constraints and rapid power consumption, a condition-triggered strong intervention weight reconstruction mechanism is proposed. When both the conflict correlation and the consumption rate exceed the safety threshold, the algorithm completely breaks the smooth and linear compromise weight allocation method in traditional multi-objective programming. It decisively performs an exponential increase on the benchmark weight of energy consumption constraints, while performing a proportional decrease on the benchmark weights of other constraints. Through nonlinear polarization adaptive adjustment that sacrifices one component to save the other, it can give absolute dominance priority to energy consumption control in the moment of life and death, thereby forcibly intervening in the planning direction and generating an emergency shortcut path with power preservation and survival as the core in the shortest time, which greatly ensures the flight safety of UAVs. Attached Figure Description

[0009] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein: Figure 1 This is a flowchart of the present invention; Figure 2 This is a flowchart of the second-order correction of the weighting coefficients in this invention. Detailed Implementation

[0010] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0011] In traditional UAV path planning systems, fixed weight allocation and static evaluation rules cannot adapt to the complex and ever-changing environmental constraints of urban low-altitude environments and the dynamic fluctuations in UAV energy consumption. When a UAV encounters extreme conditions that cause nonlinear conflicts between various constraints during its flight cycle, the system cannot establish a dynamic correlation between spatial obstacle avoidance parameters and the UAV's energy consumption status, leading to a mismatch between the comprehensive cost criterion and real-time survivability. This static weighting mechanism reduces the reliability of multi-dimensional constraint feature fusion, causing the path planning algorithm to receive misleading weights in its cost values, ultimately affecting the safety and rationality of the cooperative path trajectory output.

[0012] For example, in urban low-altitude flight scenarios, when a drone encounters dense building clusters and sudden strong headwinds, its real-time power consumption rate rises sharply from an initial 1.2% per minute to 4.5%, and the trajectory deviation caused by obstacle avoidance increases by 40%. In this situation, traditional systems still use preset baseline weight coefficients to filter cross-dimensional constraint relevance, leading to key energy consumption degradation features being misclassified as routine parameters and failing to receive high priority. The candidate node cost value output by the multi-dimensional constraint fusion model loses the key data dimension representing the reduction of survival crisis. The system incorrectly identifies extremely energy-intensive detours as the optimal safe mode, and the generated trajectory sequence continuously recommends obstacle avoidance maneuvers that increase flight distance.

[0013] If the above problems are not addressed, misidentification of high-cost, dangerous paths will lead to a misalignment between flight plans and the actual physical limits of the drone, exacerbating the risk of crashing due to battery depletion. A rigid weighting mechanism will hinder the system from capturing the critical state where spatial constraints transform into survival crises, delaying the optimal time for emergency flight path planning. Defects in the delayed processing of static parameters will also cause a phase discrepancy between environmental avoidance requirements and the drone's own energy reserves, reducing the reliability of comprehensive cost-benefit calculations and target node extraction, ultimately creating a negative feedback loop that affects the stable operation of the entire urban low-altitude collaborative network.

[0014] To address the aforementioned issues, this application first considers establishing a dynamic correlation mechanism between environmental constraint parameters and the drone's energy consumption status. Traditional systems use fixed weights to fuse multi-dimensional features, resulting in the suppression of key energy consumption deterioration signals and an inability to reflect the drone's survival crisis. To resolve this, this application attempts to dynamically couple conflict correlation with real-time power consumption rate, adjusting the internal criteria of the multi-dimensional constraint feature fusion model through dual threshold judgments. Further analysis reveals that relying solely on single remaining power data or static spatial mapping relationships is insufficient to capture nonlinear conflict fluctuations under extreme conditions. Therefore, a benchmark weight difference needs to be introduced to collaboratively calculate the conflict correlation, and dynamic weights are generated based on its concurrent state with the consumption rate. By designing a linkage mechanism of exponentially increasing calculation and proportionally decreasing calculation, the final weight coefficients of each constraint feature are adaptively adjusted according to the actual crisis state of the drone, thereby solving the problems of evaluation lag and constraint mismatch.

[0015] like Figure 1-2 As shown, a multi-constraint path collaborative planning method for urban low-altitude unmanned aerial vehicles (UAVs) includes the following steps: Acquire 3D raster environment data of the target area and real-time running status data of the local machine, and fuse and calculate the multi-dimensional constraint feature data of each candidate node. The multi-dimensional constraint feature data includes at least energy consumption constraint feature data and static obstacle avoidance constraint feature data. Numerical quantization processing is performed on multidimensional constraint feature data to generate benchmark weight coefficients for each constraint feature data; The difference between the baseline weight coefficient of the energy consumption constraint feature data and the baseline weight coefficient of other constraint feature data is calculated to generate a conflict correlation degree that characterizes the degree of constraint contradiction, and the real-time power consumption rate is extracted from the real-time operating status data of the local machine. Determine whether both the conflict correlation degree and the consumption rate are greater than the corresponding thresholds; If so, the baseline weight coefficients of the energy consumption constraint feature data are calculated exponentially, and the baseline weight coefficients of the remaining constraint feature data are calculated proportionally to generate the final weight coefficients of each constraint feature data. Otherwise, the baseline weight coefficients of each constraint feature data will be used as the final weight coefficients; The comprehensive cost of the candidate node is obtained by weighting and fusing the multidimensional constraint feature data with the final weight coefficients of each constraint feature data. In the 3D raster node network data, the candidate node with the lowest comprehensive cost value is extracted as the target node and connected to generate a cooperative path trajectory sequence and output it.

[0016] Among them: 3D raster environment data: refers to the spatial grid data structure formed by discretizing the target's 3D physical airspace according to a specific resolution, which includes the digital representation of static or dynamic environmental information such as buildings, no-fly zones and free airspace.

[0017] Real-time operational status data of the drone: refers to the physical dynamic state and system parameters of the drone performing the path planning task at the current moment, which usually includes parameters such as three-dimensional spatial coordinates, velocity vector, acceleration and airborne energy system status.

[0018] Candidate node: refers to a discrete spatial location point in a three-dimensional grid topology network that is spatially adjacent to the node where the UAV is currently located and satisfies the reachability range of the UAV's single-step kinematics and dynamics.

[0019] Multidimensional constraint feature data: refers to a set of quantitative indicators that characterize the various spatial and physical constraints that a drone must meet when it moves from the current node to a candidate node. These typically include evaluation parameters such as spatial obstacle avoidance distance, collision avoidance interval, and airspace control compliance.

[0020] Energy consumption constraint feature data: a specific subset of multidimensional constraint feature data, used to quantify the energy cost or power load index expected to be consumed by a UAV in performing a specific single-step spatial transfer.

[0021] Numerical quantization processing refers to the computational process of converting constrained feature data with different physical dimensions and scale differences into a unified dimensionless numerical range through mathematical mapping methods such as normalization or standardization.

[0022] Benchmark weight coefficient: refers to the initial weighting ratio assigned to each constraint feature by the system under normal environmental conditions, used to characterize the basic mathematical proportion of each constraint condition in the normal multi-objective evaluation system.

[0023] Conflict correlation degree: refers to a numerical index that quantifies the degree of divergence and antagonism of objective functions among different constraints. In this application, it specifically refers to the physical contradiction between space avoidance cost and energy cost, characterized by the difference between energy consumption benchmark weight and other constraint benchmark weights.

[0024] Consumption rate: refers to the first time derivative of the actual decrease gradient of the onboard power or energy reserve of the UAV per unit time or unit spatial displacement.

[0025] Exponentially increasing calculation: refers to the process in which a specific feedback variable is introduced as an exponential term in a mathematical model, so that the specific benchmark weight coefficient exhibits a non-linear, exponentially amplified mathematical amplification effect as the feedback variable increases.

[0026] Proportional decay calculation: refers to the mathematical process of reducing the numerical value of a set of variables (the baseline weights of non-core constraint features) by introducing a specific decay multiplier (between 0 and 1) at the same decay rate.

[0027] Final weight coefficient: refers to the weighting factor that is finally output and actually applied to the node cost evaluation function after logical judgment and dynamic adjustment mechanism (including maintaining the normal benchmark or triggering polarization processing).

[0028] Comprehensive cost value: refers to the single scalar output obtained through weighted fusion calculation, used to comprehensively and quantitatively evaluate the total system cost of transferring a drone to a specific candidate node. Its value serves as the direct mathematical basis for ranking the feasibility of nodes.

[0029] Target node: refers to the optimal spatial discrete point selected from all candidate nodes within the current single-step planning cycle based on the principle of minimizing comprehensive cost.

[0030] Collaborative path trajectory sequence: refers to the set of discrete spatial waypoints generated in a three-dimensional spatial network by extracting target nodes through continuous multi-step planning and connecting them with spatial topology in a temporal sequence, which can be tracked and executed by the underlying layer of the UAV flight control system.

[0031] Specifically: Step 1: Acquire 3D raster environment data of the target area and real-time operating status data of the local machine, and fuse and calculate the multi-dimensional constraint feature data of each candidate node. The multi-dimensional constraint feature data should include at least energy consumption constraint feature data and static obstacle avoidance constraint feature data. The system acquires 3D raster environmental data obtained by discretizing the target flight airspace at a specific resolution, including the distribution of buildings and no-fly zones. Simultaneously, it extracts the UAV's current real-time operational status data, including its current spatial coordinates, real-time overall weight, and real-time wind resistance. Using the UAV's current node as a baseline, it expands outwards to extract multiple physically adjacent and reachable discrete grid center points as candidate nodes.

[0032] For each candidate node, a fusion calculation is performed to extract multi-dimensional constraint feature data. Specifically, when calculating energy consumption constraint feature data, it does not rely solely on geometric distance. Instead, it combines the spatial geometric transfer distance of the UAV from the current node to the candidate node with the overall mass and environmental drag coefficient from the UAV's real-time status data, and substitutes this data into a pre-defined physical dynamics energy consumption model to calculate the expected energy cost. Simultaneously, it calculates in parallel the shortest spatial distance from the candidate node to the nearest obstacle bounding box, generating constraint feature data for other dimensions such as static obstacle avoidance.

[0033] Step 2: Perform numerical quantization on the multidimensional constraint feature data to generate the baseline weight coefficients for each constraint feature data: Because the constraint feature data of different dimensions have serious differences in units of measurement, such as distance units and energy units being different, numerical quantification is used to eliminate the influence of units of measurement. Cost-based constraints (the larger the value, the higher the cost, such as energy consumption) and benefit-based constraints (the smaller the value, the higher the risk, such as obstacle avoidance distance) are distinguished, and forward or inverse range transformation mapping algorithms are used respectively to uniformly map all the original feature data to a dimensionless standard cost range of zero to one.

[0034] The arithmetic mean of the standard cost values ​​of all candidate nodes in the current local space along a specific dimension is calculated to generate an index characterizing the expected severity of local constraints. The expected severity values ​​of each constraint are then globally normalized so that the sum of the weights for each constraint is constant at one, thus generating a baseline weight coefficient for each constraint feature data. This baseline weight coefficient objectively reflects the true pressure distribution of the current local environment.

[0035] Step 3: Calculate the difference between the baseline weight coefficient of the energy consumption constraint feature data and the baseline weight coefficients of other constraint feature data to generate a conflict correlation degree characterizing the degree of constraint contradiction, and extract the real-time power consumption rate from the local real-time operating status data: Extract the energy consumption baseline weight coefficients generated in step two, and perform absolute difference calculations with the baseline weight coefficients of other environmental constraints such as static obstacle avoidance and collision avoidance to find a set of absolute difference data. Extract the maximum value in this set of absolute difference data and define it as the conflict correlation degree. The larger this value is, the greater the energy consumption deviation cost required to meet the most dangerous environmental constraint.

[0036] Simultaneously, it communicates with the drone's battery management system to extract real-time absolute battery power data within a continuous sampling time window. By calculating the time-first derivative of the percentage battery depletion, it obtains the real-time battery consumption rate, characterizing the acceleration of battery loss. This quantifies the degree of conflict between environmental space constraints and the aircraft's physical limitations.

[0037] Step 4: Determine if both the conflict correlation degree and consumption rate are greater than the corresponding thresholds: The system pre-defines conflict correlation and energy consumption rate thresholds based on the drone's physical limits and normal weather conditions. The flight control computer compares the real-time calculated conflict correlation and energy consumption rate with their respective thresholds and performs a strict logical AND operation. Only when the spatial constraint limit and the temporal energy loss limit are simultaneously exceeded (i.e., both requiring significant detours and experiencing rapid abnormal power loss), is the drone deemed to be in an extreme survival crisis of multi-constraint deadlock.

[0038] Step 5: If yes, perform an exponentially increasing calculation on the baseline weight coefficients of the energy consumption constraint feature data, and perform a proportionally decreasing calculation on the baseline weight coefficients of the remaining constraint feature data to generate the final weight coefficients for each constraint feature data. If the result of step four is yes, it indicates that the drone is facing an extreme deadlock crisis, a matter of life and death. Immediately trigger the nonlinear polarization intervention mechanism: using the real-time power consumption rate as the exponential variable in the natural exponential calculation, the baseline weight coefficient of the energy consumption dimension is amplified geometrically, rapidly approaching its absolute dominance in weight proportion. To maintain the normalization and constancy of the total system weight, the remaining allocable weight space is calculated based on the amplified energy consumption weight, thus deriving a unified attenuation ratio coefficient. This attenuation ratio coefficient is then simultaneously multiplied and added to the baseline weight coefficients of all other spatial constraints, performing a proportional reduction. This mechanism instantly assigns the highest priority to power preservation and survival, forcibly breaking the multi-objective optimization deadlock and achieving adaptive polarization reconstruction.

[0039] Step Six: Otherwise, use the baseline weighting coefficients of each constraint feature data as the final weighting coefficients: If the result of step four is negative, meaning that the conflict correlation and consumption rate have not simultaneously exceeded the threshold, it indicates that the UAV is in normal flight condition. In this case, no aggressive intervention is performed; instead, the data pass-through bypass is directly activated. The baseline weight coefficients of various constraints derived from the objective distribution of the local environment are directly assigned as the final weight coefficients output downstream. This effectively avoids the neurotic jitter of the flight path caused by overly sensitive weight adjustments during normal flight, ensuring the smoothness of the regular cruise trajectory and the balance of multiple objectives.

[0040] Step 7: Perform a weighted fusion calculation on the multidimensional constraint feature data and the final weight coefficients of each constraint feature data to obtain the comprehensive cost of the candidate node: After determining the final weighting coefficients applicable to the current working conditions, a data dimensionality reduction evaluation is performed. All candidate nodes are traversed, and their dimensionless standard cost values ​​for each constraint dimension are extracted. These values ​​are then multiplied by the corresponding final weighting coefficients, and the products are linearly summed. Through this weighted inner product operation, the high-dimensional feature matrix is ​​compressed and aggregated into a scalar value reflecting the overall feasibility of a single candidate node, i.e., the comprehensive cost value.

[0041] Step 8: In the 3D raster node network data, extract the candidate node with the lowest comprehensive cost as the target node, connect them, generate a cooperative path trajectory sequence, and output it: An extreme value search is performed on the candidate node set in the current period. By comparing the comprehensive cost of all candidate nodes, the node with the smallest value is precisely extracted and confirmed as the optimal single-step target node under the current physical state and environmental pressure. At the data structure level, this target node is appended to the end of the path linked list to complete the topological connection, and steps one through seven are iterated repeatedly from this node as the new starting point. After continuous expansion in time and space dimensions, a complete, continuous, and ordered three-dimensional cooperative path trajectory sequence is finally constructed and packaged into a standard control command protocol, which is then output to the UAV's underlying flight control system to perform physical maneuvers.

[0042] Specific Implementation: Flight Line Polarization Reconstruction under Extreme Operating Conditions: Scenario Preset and Initial Data: Assume a drone is flying through a dense urban building complex, and two candidate nodes are extracted from the current local airspace: Candidate Node 1 (Safe Detour Point): Requires a large detour, obstacle avoidance is extremely safe, but energy consumption is extremely high.

[0043] Candidate Node 2 (High-Risk Straight Flight Point): It traverses a narrow passage between two tall buildings in a straight line, which is extremely energy-efficient, but the obstacle avoidance risk is extremely high.

[0044] After feature fusion and dimensionless quantization, the results of each node under the three constraints (energy consumption) are obtained. Static obstacle avoidance Airspace control Standard value on (numerical range) (A larger value indicates a higher cost / risk), and a baseline weighting coefficient adaptively generated based on the current dense building environment. As shown below: The total weight of the benchmark is 1.

[0045] Real-time calculation of constraint conflict correlation : ; Meanwhile, the drone encountered a sudden strong headwind due to the narrow channel effect, and the BMS system reported the current power consumption rate. .

[0046] Assuming a preset conflict threshold Consumption rate threshold .

[0047] at this time, and If the condition satisfies the double AND logic, it is determined that an extreme deadlock crisis has been entered, triggering polarization intervention.

[0048] Exponential polarization reconstruction of weights: The system cuts off normal pass-through and applies an exponentially increasing weight to energy consumption (assuming a gain constant). The final weighting coefficients of the energy consumption constraint feature data are obtained. : ; Then calculate the proportional decay coefficient for the remaining weights. : ; Obstacle avoidance and control weights are reduced proportionally: The final weighting coefficients of the static obstacle avoidance constraint feature data: ; The final weighting coefficients for airspace control constraint feature data: ; The final weights after reconstruction: .

[0049] Comparison of comprehensive cost-value calculation and decision reversal: Still using the benchmark weight The normal logic of forced flight: Overall cost of candidate node 1: ; Overall cost of candidate node 2: ; Traditional result: due to Traditional systems would select node 1. In the face of a rapidly depleting power headwind, the drone would continue to perform a wide-ranging, circling flight, and would most likely crash due to power exhaustion.

[0050] Use the final weights after polarization reconstruction : Overall cost of candidate node 1: ; Cost of candidate node 2: ; because The system decisively reversed its decision, selecting node 2 as the target node. Within milliseconds, it abandoned absolute obstacle avoidance margin and chose to take a risky straight shortcut to ensure that the drone could escape the high-risk wind zone as quickly as possible with minimal energy depletion.

[0051] The core innovation of this application lies in breaking through the limitations of traditional path planning algorithms that rely solely on the remaining power for static evaluation. It innovatively introduces a dual collaborative discrimination mechanism that represents the degree of constraint contradiction and the real-time power consumption rate. This enables the UAV to not only perceive the rate of its own energy loss in real time, but also to accurately quantify the degree of confrontation between current spatial constraints (such as obstacle avoidance due to dense building clusters) and energy consumption limitations. Through this multi-dimensional cross-validation, the comprehensive survival crisis faced by the UAV can be identified more accurately and proactively, effectively preventing the UAV from falling into a deadlock state of rapid power depletion due to blindly compromising with spatial avoidance. By introducing a nonlinear polarization weight adjustment strategy, the survivability of UAVs under extreme conditions is effectively improved. For extreme flight conditions with severe conflicts of multiple constraints and rapid power consumption, a condition-triggered strong intervention weight reconstruction mechanism is proposed. When both the conflict correlation and the consumption rate exceed the safety threshold, the algorithm completely breaks the smooth and linear compromise weight allocation method in traditional multi-objective programming. It decisively performs an exponential increase on the benchmark weight of energy consumption constraints, while performing a proportional decrease on the benchmark weights of other constraints. Through nonlinear polarization adaptive adjustment that sacrifices one component to save the other, it can give absolute dominance priority to energy consumption control in the moment of life and death, thereby forcibly intervening in the planning direction and generating an emergency shortcut path with power preservation and survival as the core in the shortest time, which greatly ensures the flight safety of UAVs.

[0052] By directly using the benchmark weight coefficients generated by numerical quantization for fusion calculation under normal flight conditions without triggering the dual-threshold crisis, the UAV can ensure a smooth and comprehensive response to various constraints such as obstacle avoidance and no-fly zones under normal conditions. However, once the environmental changes trigger the threshold, it instantly switches to an extreme power preservation mode. Finally, based on the comprehensive cost value after multi-dimensional fusion, the optimal candidate node is accurately extracted in the three-dimensional grid node network, ensuring that the most feasible and safest cooperative path trajectory sequence can be output under the current state, whether it is normal cruise or crisis escape.

[0053] Existing technologies often treat environmental geometric data (such as distance to obstacles) and onboard operational data (such as the drone's weight and wind conditions) separately during pathfinding. This disconnect between geometric planning and physical execution capabilities leads to the system frequently planning theoretically shortest, obstacle-free paths that are actually beyond the drone's current load capacity or energy budget, resulting in unworkable paths. To address this, we propose acquiring 3D raster environmental data of the target area and real-time onboard operational data, and then fusing and calculating multi-dimensional constraint feature data for each candidate node. This multi-dimensional constraint feature data includes at least energy consumption constraint feature data and static obstacle avoidance constraint feature data, as detailed below: Step 1: Using the onboard flight control computer, on the one hand, read the pre-loaded or real-time constructed 3D grid environment data of the target area (to obtain obstacle distribution and no-fly zone coordinates); on the other hand, synchronously extract the real-time operating status data of the aircraft (current coordinates) through onboard sensors (such as IMU inertial measurement unit, BMS battery management system, barometer, etc.). Current quality Current wind resistance status ).in: Current UAV path planning mapping technologies typically employ a uniform global resolution grid. This leads to several issues: If the global resolution is too high (mesh too small), while obstacle avoidance accuracy is extremely high, millions of search nodes are generated. This results in excessively long computation times for subsequent candidate node expansion and cost fusion, severely straining the limited onboard computing power and failing to meet the real-time response requirements of high-speed UAV flight. Conversely, if the global resolution is too low (mesh too large), although computation is fast, building edges are severely coarsened (i.e., a small gap is completely filled by the large grid), causing the UAV to fail to find any passable gaps in narrow urban canyons, directly leading to path planning failure or false collisions. Therefore, this paper proposes acquiring 3D raster environmental data of the target area, specifically including: Extract building density attribute data for the target area; Determine whether the building density attribute data is greater than the preset density threshold; If so, high-resolution three-dimensional raster environment data is generated using the first spatial spacing parameter; Otherwise, a low-resolution 3D raster environment data is generated using a second spatial spacing parameter that is greater than the first spatial spacing parameter.

[0054] Before the drone enters the target area or receives a flight mission, the system obtains the terrain overview of the target area by calling up the city's GIS elevation map or the preliminary scan point cloud of the airborne lidar. The total area of ​​the target area is then extracted. and the total area covered by all the buildings therein. Calculate their ratio to generate building density attribute data. : .

[0055] The system compares the real-time calculated density attribute data with a preset density threshold. If the density attribute data is greater than the preset density threshold, it indicates that the drone is about to fly into a densely populated area such as a CBD center with many high-rise buildings. In this extremely confined environment, to ensure the drone can accurately pass between two closely spaced buildings without colliding, the system must employ extremely high map precision. Therefore, the system uses a very small first spatial spacing parameter to divide the airspace into extremely fine "high-resolution 3D raster environmental data".

[0056] If the density attribute data is less than or equal to the preset density threshold, it indicates that the drone is in an open area with sparse buildings, such as a river surface, a large park, or the suburbs. At this time, the airspace is wide and the risk of collision is extremely low. The system decisively calls a larger second spatial spacing parameter to divide the airspace into coarser-grained "low-resolution three-dimensional raster environment data".

[0057] By extracting building density attribute data and comparing thresholds, the spatial spacing parameters are dynamically switched based on the degree of environmental congestion, achieving adaptive adjustment of the underlying data structure resolution. In complex and dense areas requiring precise operations (such as building navigation), a high-precision grid is provided to ensure absolute safety, collision avoidance, and gap penetration capabilities; while in open and safe areas, the grid precision is proactively reduced. This strategy geometrically compresses the total number of nodes, effectively offloading the memory and CPU load from the onboard computer. The sharp reduction in underlying environmental nodes directly leads to a significant reduction in the array length required for all subsequent calculations (such as calculating multidimensional features, conflict correlation, weighted fusion, etc.). This reduces the computation cycle of the entire planning algorithm from seconds to milliseconds, giving urban low-altitude UAVs truly agile real-time collaborative planning capabilities during high-speed flight.

[0058] Step 2: Based on the current node of the drone Based on the maximum single-step maneuverability and grid resolution of the UAV, the system expands outward in three-dimensional space to determine a set of adjacent and physically reachable discrete grid center points, forming a candidate node set. .

[0059] Step 3: For each candidate node in the set A fusion calculation must be performed. First, calculate the spatial geometric transfer distance: ; Subsequently, without directly stating the distance Equivalent to energy consumption, but rather compared to the machine's state (mass). Environmental resistance The data is then fused into a pre-defined UAV physical dynamics energy consumption model. Candidate nodes are then calculated. Energy consumption constraint characteristic data The specific formula is as follows: ; Step 4: While calculating energy consumption, simultaneously calculate the constraint features of other dimensions. Specifically: Acquire three-dimensional raster environmental data of the target area and real-time operating status data of the local machine, and also simultaneously acquire dynamic airspace control data and spatiotemporal trajectory prediction data of other UAVs in the same airspace; In addition to energy consumption constraint data and static obstacle avoidance constraint data, the multidimensional constraint feature data also includes collision avoidance interval constraint data and airspace control constraint data.

[0060] The specific steps for generating static obstacle avoidance constraint feature data, collision avoidance interval constraint feature data, and airspace control constraint feature data include: The entity features in the 3D raster environment data are abstracted into spatial bounding box data. The shortest distance from the candidate node to the outer edge of the spatial bounding box data is calculated to generate static obstacle avoidance constraint feature data. Extract the timestamp of the predicted arrival at the candidate node from the local machine, align it with the spatiotemporal trajectory prediction data of other UAVs, calculate the relative spatial distance under the same time section, and generate anti-collision interval constraint feature data. Calculate the distance from candidate nodes to the no-fly zone boundary in the dynamic airspace control data, and generate airspace control constraint feature data.

[0061] When preprocessing 3D raster environment data, the system does not directly use the original entity features of buildings composed of tens of thousands of complex polygons. Instead, it abstracts these features into regular spatial bounding box data (e.g., using the AABB axis-aligned bounding box algorithm). For candidate nodes, the system calculates their Euclidean shortest distances to the outer surfaces of all spatial bounding boxes, extracts the minimum values, and generates static obstacle avoidance constraint feature data. .

[0062] Based on the current speed of the drone, the precise timestamp required for the drone to physically reach the candidate node is derived. Subsequently, the system iterates through the spatiotemporal trajectory prediction data of other drones acquired simultaneously, forcibly extracting their predicted coordinates at the same time point. The spatial relative distances between the candidate node and all other drones at that moment are calculated, and the minimum values ​​are extracted to generate anti-collision interval constraint feature data. .

[0063] The system parses dynamic airspace control data, obtains the boundary set of temporary no-fly zones or altitude-restricted zones, calculates the shortest distance of candidate nodes to all no-fly zone boundaries, and generates airspace control constraint feature data. .

[0064] Finally, the system will calculate the above... , , The energy consumption constraint characteristic data calculated in the aforementioned steps They are packaged together to form the complete multidimensional constraint feature data of the candidate node.

[0065] In the aforementioned technology, three-dimensional grid environment data and real-time operating status data of the local machine are acquired simultaneously, and fusion calculation is forced to be performed in the evaluation stage of each candidate node. That is, a mathematical model that couples spatial geometric parameters with the physical dynamic parameters of the machine body is established, thereby extracting multi-dimensional constraint feature data including energy consumption dimension. This breaks the limitation of traditional algorithms that only expand nodes based on spatial geometric reachability. By integrating the real-time status of the machine body mass, drag and other factors into energy consumption calculation, the evaluation indicators of each candidate node output are ensured to not only meet the spatial obstacle avoidance safety requirements, but also fully comply with the current physical dynamic limits and energy budget of UAVs.

[0066] Existing technologies typically employ static, fixed weights assigned through human experience. This static mechanism suffers from two major drawbacks: first, it cannot eliminate dimensional barriers between heterogeneous data such as spatial distance and energy joules, leading to easy divergence in underlying numerical calculations or being swallowed up by large numerical features; second, static weights cannot adapt to the complex and ever-changing environment of urban low-altitude airspace. For example, maintaining extremely high obstacle avoidance weights in open areas would result in wasted computing power and flight path redundancy. Therefore, this paper proposes to perform numerical quantization processing on multi-dimensional constraint feature data to generate corresponding baseline weight coefficients for each constraint feature data, as detailed below: Step 1: Based on the differences in the physical properties of each constraint, call different quantization mapping functions to convert the multidimensional constraint feature data. Unified mapping to dimensionless standard value ,in, This serves as the traversal index for the candidate nodes. , The total number of candidate nodes extracted in the current local space. Traversal index of constrained feature dimensions , Due to energy consumption constraints, This is a static obstacle avoidance constraint feature. For collision avoidance spacing constraints, This refers to the characteristics of airspace control constraints.

[0067] For cost constraints (such as energy consumption constraints) (The larger the value, the higher the cost), positive range quantization is used for calculation: ; For benefit-based constraints (such as obstacle avoidance distance constraints) (The smaller the value, the higher the risk, and the greater the cost should be), inverse range quantization is used for calculation: ; All heterogeneous physical data are uniformly converted into a standard mathematical expression that "the larger the value, the higher the threat / cost".

[0068] Step 2: To ensure the weights reflect the true characteristics of the current environment, calculate the... Item constraints in the current all The arithmetic mean of the dimensionless standard cost values ​​at each candidate node is used to generate the expected severity of the local constraint. : ; If an area is densely packed with buildings, causing all candidate nodes to incur obstacle avoidance costs... If all are too high, then the calculated The value will increase significantly, objectively reflecting the physical fact that the current local airspace is under tremendous obstacle avoidance pressure.

[0069] Step 3: Calculate the expected severity of each constraint. Perform global normalization to generate baseline weight coefficients for each constraint. To ensure that the sum of the system weights is always 1 (i.e. The calculation formula is as follows: ; in, This is a traversal set of all constraint feature dimensions. It completes the transformation of the original multidimensional feature data into baseline weight coefficients that reflect the objective pressure of the local environment. The adaptive quantization closed loop.

[0070] The aforementioned technology introduces a bidirectional range mapping algorithm based on physical extrema to separate cost-type and benefit-type data by removing physical units, completing numerical quantification, and then calculating the expected severity distribution within the candidate space. This allows the environmental data itself to participate in determining the weight allocation, automatically generating objective benchmark weight coefficients. This forcibly pulls spatial and dynamic data with completely different physical scales back to their original state. The unified comparison benchmark ensures the mathematical stability and robustness of subsequent weighted calculations, and makes the allocation of weights entirely determined by the local real environment in which the UAV is currently located. This provides an objective, accurate and highly dynamic comparison benchmark for subsequent conflict correlation calculations and exponential polarization adjustments in response to extreme working conditions.

[0071] Traditional UAV path planning systems suffer from two serious flaws: First, when evaluating system safety, they rely solely on the absolute values ​​or independent weights of each constraint, failing to perceive the internal logical conflicts and contradictions between different constraints (such as power preservation and obstacle avoidance); second, when assessing energy consumption safety, they rely only on static thresholds such as whether the remaining power is below 20%, completely ignoring the dynamic power loss trends caused by sudden environmental changes (such as gusts of wind shear), resulting in a severe lag in the system's response to extreme crises. To address this, a method is proposed that calculates the difference between the baseline weight coefficients of energy consumption constraint characteristic data and the baseline weight coefficients of other constraint characteristic data to generate a conflict correlation degree characterizing the degree of constraint contradiction. Furthermore, the real-time power consumption rate is extracted from the real-time operating status data of the UAV, as detailed below: Step 1: Calculate the difference between the baseline weight coefficient of the energy consumption constraint feature data and the baseline weight coefficients of other constraint feature data to generate a conflict correlation degree that characterizes the degree of constraint contradiction. Specifically, this includes: Extract the first benchmark weight coefficient corresponding to the energy consumption constraint feature data, and the second, third, and fourth benchmark weight coefficients corresponding to the static obstacle avoidance constraint feature data, collision avoidance interval constraint feature data, and airspace control constraint feature data, respectively; Calculate the absolute differences between the first benchmark weight coefficient and the second, third, and fourth benchmark weight coefficients, respectively. Extract the largest value from the absolute difference data and use it as the degree of conflict correlation.

[0072] Extract the energy consumption baseline weighting coefficients generated in the previous calculation step. and the baseline weighting coefficients for all other environmental constraints. Traverse all Calculate the absolute difference between the energy consumption benchmark weight and the weights of other benchmarks. The calculation formula is as follows: ; Step 2: After obtaining a set of absolute difference data, the system applies a maximum value extraction algorithm to find the largest value among all differences, and defines it as the conflict correlation degree within the current planning period. The calculation formula is as follows: ; The physical significance of selecting the maximum absolute difference lies in the fact that as long as there is any extremely dangerous environmental constraint in the local airspace (such as an extremely close obstacle causing...) (Extremely high), while the energy consumption constraint is in a normal state at this time ( (lower), the huge difference between the two will make The value increased sharply.

[0073] Step 3: While calculating the conflict correlation, the system communicates asynchronously or synchronously with the UAV's Battery Management System (BMS) to obtain power data within continuous sampling periods. Let the current time be... The previous sampling time was (in Then the real-time power consumption rate The formula for calculating (i.e., the first derivative of the decrease in battery power) is: ; For a moment The real-time absolute battery percentage or energy value fed back by the drone's battery management system; This rate It's not a static remaining battery level, but rather a measure of how quickly the battery is depleted, especially when the drone encounters strong headwinds or performs dramatic altitude maneuvers. This will result in a momentary jump in value.

[0074] In the aforementioned technology, a derived mathematical index, conflict correlation, is constructed by calculating the maximum absolute difference between the energy consumption baseline weight and other spatial constraint baseline weights. Simultaneously, a time dimension is introduced, and the dynamic energy consumption rate is obtained by calculating the first derivative of the energy consumption within a continuous time window. This precisely visualizes the invisible and intangible physical contradiction during flight (the need for significant detours versus energy conservation) into a mathematical variable that the computer can use for threshold comparison—the conflict correlation—compensating for the blind spots of traditional algorithms in multi-objective game perception. By extracting the energy consumption rate, the system does not need to wait until the battery is depleted to trigger an alarm. When the drone encounters severe turbulence causing abnormal acceleration in power loss, the high conflict correlation can be used to predict the impending multi-constraint deadlock crash crisis in advance, providing an extremely sensitive precondition for the next step of implementing aggressive polarization intervention strategies (exponential adjustments).

[0075] In autonomous decision-making systems for unmanned aerial vehicles (UAVs), any intervention triggering mechanism based on a single threshold or dimension has fatal flaws: if the triggering condition is too sensitive (e.g., changing the global flight path based solely on a sudden, rapid power loss), it can cause the UAV to exhibit neurotic flight path jitter, severely affecting the continuity of flight missions; if the triggering condition is too simplistic, it can easily generate false alarms, leading to wasted computing power and unnecessary avoidance maneuvers. Therefore, a method is proposed that determines whether both the conflict correlation degree and the consumption rate are greater than their corresponding thresholds, as detailed below: Step 1: The flight control system will preload preset conflict correlation thresholds at the underlying level. and preset consumption rate threshold These two thresholds are not arbitrarily set, but are pre-calibrated through offline simulation based on the drone's model parameters and routine meteorological data (such as average wind resistance) of the target area. When performing this judgment step, these two baseline limits are first retrieved from memory.

[0076] Step 2: Calculate in real time and Each value is compared with its corresponding threshold, and a strict logical AND operation is performed to generate a judgment label. The calculation formula is as follows: ; Scenario A (only) This indicates a significant discrepancy between the space obstacle avoidance requirements and energy consumption requirements at this point (e.g., a large no-fly zone ahead requires extensive detours), but the power consumption rate is... Still within the normal threshold ( (For example, if the drone is facing a tailwind, or if it has an extremely high battery level and is discharging slowly). At this point, the system determines that it has not entered an extreme crisis. ).

[0077] Case B (only) This indicates that the battery is rapidly depleting (e.g., encountering a sudden strong headwind or performing a violent maneuver), but the conflict correlation is low. Not high ( (For example, if you are in an open area, you don't need to fly around; you can go in a straight line.) At this point, you only need to accelerate in a straight line to pass through the wind zone; this is not considered an extreme crisis. ).

[0078] Case C (all values ​​are greater than the threshold): that is and This indicates that the drone is facing a predicament where it must take a significant detour while simultaneously losing power at an abnormally high rate. This concurrent state of "having to take a significant detour while experiencing rapid power loss" is identified by the system as an "extreme survival crisis of multiple constraint deadlocks." ).

[0079] Step 3: According to This directly determines the direction of the downstream computing link: if it is If so, the branch "Otherwise, use the baseline weight coefficients of each constraint feature data as the final weight coefficients" is triggered; if it is... If so, the polarization intervention branch is triggered, which is to "perform an exponentially increasing calculation on the baseline weight coefficients of the energy consumption constraint feature data and a proportionally decreasing calculation on the baseline weight coefficients of the remaining constraint feature data to generate the final weight coefficients of each constraint feature data".

[0080] The aforementioned technology employs dual concurrent logic to jointly cross-validate multi-source heterogeneous parameters (structural conflict correlation and dynamic consumption rate). This requires that both the spatial dimension's contradiction limit and the temporal dimension's energy loss limit be simultaneously overcome before a positive judgment can be output. Through a logical AND mechanism, false positives caused by simple sensor noise, instantaneous gust wind shear, or single local obstacles are effectively filtered out, significantly improving the system's robustness in complex urban scenarios. Furthermore, only in the event of a genuine multiple deadlock crisis is a subsequent drastic weight reconstruction triggered. This ensures that the UAV maintains a smooth, multi-target (avoidance, altitude restriction) optimal flight path for most of its normal flight time, while ensuring that intervention commands are not missed or delayed in the rare, extremely critical situations.

[0081] When drones face extremely harsh environments (such as severe headwinds and being surrounded by tall buildings), traditional linear programming algorithms or fixed-weight mechanisms can get stuck in a deadlock: the system wants to avoid the buildings while also conserving power. However, due to the limited adjustment range of linear weights, the drone will choose a slightly longer and more energy-intensive compromise path. Under conditions of extremely rapid power consumption, this slow compromise can directly lead to the drone crashing due to battery depletion before completing obstacle avoidance. To address this, we propose an exponentially increasing calculation of the baseline weight coefficients for energy-constrained feature data and a proportionally decreasing calculation for the baseline weight coefficients of other constraint feature data, generating the final weight coefficients for each constraint feature data, as follows: The baseline weighting coefficients of the energy consumption constraint characteristic data are calculated exponentially, specifically including: Extract the first derivative of the real-time power consumption rate over a continuous sampling period to generate power change factor data that characterizes the acceleration of power loss. Use the sudden change in electricity volume factor data as the exponential term variable in the natural index calculation formula; The baseline weight coefficients of the energy consumption constraint characteristic data are multiplied by the calculation result of the natural exponential formula to generate the final weight coefficients, which are exponentially amplified.

[0082] Step 1: Upon determining that a crisis has been triggered, the flight control computer retrieves historical rate data cached in memory. It then extracts the current rate of data consumption. The consumption rate of the previous cycle The difference method is used to calculate its value in the continuous sampling period. The first time derivative within the time interval. The formula is as follows: Power fluctuation factor data: ; If the drone is simply flying at a constant speed against the wind, although it drains the battery quickly ( (Large), but because of its stable speed, , calculated It will approach 0; only when the drone suddenly encounters a gust of wind or extremely violent maneuvering, causing a sudden increase in the rate of power loss ( Much larger Only then will a very large positive change factor be generated. The obtained power surge factor data As a core variable, it is fed into the position of the exponent term of the natural exponential function to calculate a nonlinear amplifying multiplier. .

[0083] Extract the energy consumption benchmark weighting coefficients calculated based on the objective environment in the early stage. Multiply it directly by the amplification multiplier obtained in step two. The calculation formula is as follows: ; in: is the base of the natural logarithm, a constant of 2.71828; This is a preset exponential sensitivity adjustment constant (a dimensionless positive number used to adapt to the battery discharge characteristics of different models).

[0084] Step 2: To ensure that the final sum of the weight coefficients of all feature data is 1 at any given time (i.e., normalization constraint: The system must dynamically calculate the remaining allocable space for the remaining weights based on the final weights after the rapid expansion of energy consumption, and then derive the proportional attenuation coefficient. : ; because After exponential amplification, it will inevitably be much larger than... Therefore, the calculated It must be an extremely small decimal (i.e., a strong decay).

[0085] Step 3: Calculate the attenuation coefficient It operates synchronously on all non-energy-constrained dimensions. For all ,implement: By multiplying by the same The weight values ​​of other constraints (such as obstacle avoidance and collision avoidance) are significantly compressed, but because the compression is proportional, their relative importance order is preserved, avoiding logical confusion.

[0086] In the aforementioned technology, a natural exponential operation based on the consumption rate is introduced as an amplifier to nonlinearly polarize and elevate the weights of energy consumption constraints. Simultaneously, a proportional decay operation based on the summation conservation principle is used to globally compress the weights of other spatial constraints. This exponential increase allows the system to assign overwhelming priority to energy consumption constraints within milliseconds, enabling nonlinear polarization decisions in crisis situations. Through proportional decay calculations, while drastically amplifying a certain dimension, the underlying mathematical logic of weight normalization is strictly adhered to. The remaining decayed constraints are not directly cleared to zero (direct clearing would lead to collisions), but rather proportionally relegated to a secondary role. This maintains the most basic spatial collision avoidance baseline while ensuring absolute power supply, thus guaranteeing the mathematical stability of the multi-dimensional feature fusion system.

[0087] If the path planning system is overly sensitive to any minute environmental fluctuations or energy consumption changes and performs drastic weight adjustments (e.g., frequently exponentially increasing or decreasing a constraint), it will lead to extremely serious negative consequences: the drone will veer sharply into a straight line when the power is slightly depleted, and then suddenly climb sharply when it gets close to a building, resulting in a jagged flight path, which greatly reduces the mechanical lifespan of the aircraft and the stability of passengers / cargo. Complex nonlinear polarization calculations are energy-intensive instructions in airborne microcomputers, and frequent calls to them at unnecessary times will strain the core computing resources of the flight control system. Therefore, it is proposed that the baseline weight coefficients of each constraint characteristic data be used as the final weight coefficients, as follows: When the dual logical thresholds are not simultaneously breached, a direct data bypass branch is established, which incorporates the baseline weight coefficients that already include local environment adaptive characteristics. Direct pass-through assignment as the final weight coefficient Without attaching any extreme intervention algorithms.

[0088] When the underlying flight control computer receives the signal output by the previous determination module, At this point, the system instruction stream automatically merges into the current "otherwise" branch. Under this branch, the system bypasses all complex nonlinear polarization adjustment modules, addressing all constraint feature dimensions. It directly performs numerical pass-through assignment operations in the same dimension. For any ,implement: ; It should be noted that here It is not a fixed constant set by humans in the traditional sense, but a value that has been adaptively quantized and normalized in the previous steps of this application based on the expected severity of constraints in the current local three-dimensional grid.

[0089] Therefore, direct as This means the system recognizes the distribution of natural environmental pressure within the current local airspace. For example, if the current area is a high-density building zone, the obstacle avoidance baseline weights calculated in previous steps will be considered. The drone will naturally ascend gradually, and the system will directly adopt the weight of this ascent to guide the drone to perform smooth, energy-efficient conventional obstacle avoidance and maneuvering.

[0090] In the aforementioned technologies, by strictly adhering to a balance strategy calculated based on the real local environment during the regular mission cycle, which accounts for most of the total flight time, the system avoids both excessive power conservation and overly sensitive obstacle avoidance. This allows for the planning of a smooth flight path that best suits the dynamics of the UAV. By replacing complex polarization calculations with simplified assignment operations, the system significantly reduces computational latency and processor power consumption during normal flight. This enables high-frequency path collaborative planning iterations to be realized under limited onboard computing power, providing the system with an absolutely stable and reliable fallback / default safety baseline.

[0091] Because the actuators (motors, servos) in the execution logic of the UAV's underlying flight control computer cannot directly read and process high-dimensional conflict data, multi-objective optimization problems, without mathematical dimensionality reduction, will lead to decision paralysis in the path search algorithm during the node evaluation stage, making it impossible to select a unique and definite subsequent spatial node. Therefore, this paper proposes to perform a weighted fusion calculation of multi-dimensional constraint feature data and the final weight coefficients of each constraint feature data to obtain the comprehensive cost value of candidate nodes, as follows: Step 1: Traverse all candidate nodes in the current space For each specific candidate node, extract its standard cost array across all constraint dimensions. And extract the final weight coefficient array currently issued. .

[0092] Perform a dot product (inner product) operation on these two arrays, that is, multiply each feature value by its corresponding final weight and sum them all. The calculation formula is as follows: ; in: : No. The final comprehensive value of each candidate node; : No. The final weight coefficients of the feature dimensions of the constraint item; : No. The candidate node in the th Dimensionless standard substitution value under the constraint of term.

[0093] Step 2: Due to And the total weight The comprehensive cost of each candidate node is calculated using the weighted fusion formula described above. It will inevitably be strictly constrained to Within the closed interval.

[0094] If the current flight is under normal conditions, Relatively balanced This reflects the smoothing of the overall cost between obstacle avoidance and energy consumption in conventional space at this node.

[0095] If an extreme crisis (exponential polarization intervention) is triggered, then The energy consumption weight approaches 1.0, while other weights approach 0. At this point, the formula effectively degenerates into... This means that, at this extreme moment, the overall cost of a candidate node is almost entirely determined by its energy consumption cost, and the environmental obstacle avoidance requirements are forcibly downgraded by mathematical algorithms, thereby implementing a top-level intervention strategy of absolute power supply at the underlying computing level.

[0096] In the aforementioned technology, a linear weighted fusion calculation model is employed to perform a mathematical inner product summation between the dimensionless multidimensional constraint feature data and the final weight coefficients output by the crisis judgment mechanism, aggregating the multidimensional feature matrix into a single scalar value. This highly condenses and compresses the extremely complex environmental game, physical dynamic constraints, and intervention strategies under sudden crises into a single scalar index that directly reflects feasibility, directly eliminating the decision-making ambiguity caused by multidimensional constraints. Through this simple linear multiplication-addition operation, the computational cost of complex state machine branching and nonlinear exponential operations is isolated upstream. In actual high-frequency node traversal searches, the algorithm only needs to compare the size of a one-dimensional scalar, greatly improving the execution efficiency of airborne microprocessors in real-time path iterative calculations within a three-dimensional grid network.

[0097] Since the preceding steps (which involve complex polarization intervention and weighted fusion) can only output a series of abstract, discrete scalar values ​​(cost values), the underlying flight control system of the UAV cannot understand these multidimensional games and mathematical weights. The flight control system can only receive explicit control commands such as "where to go (spatial coordinates)" and "in what order (time series)". How to smoothly transform high-dimensional mathematical decisions into a sequence of actions executable by physical entities is the core problem that this step needs to solve. To this end, we propose extracting the candidate node with the minimum comprehensive cost value from the 3D grid node network data as the target node for connection, generating a cooperative path trajectory sequence and outputting it, as follows: Step 1: In the 3D raster node network, for the currently expanded... Sort the candidate nodes in ascending order by their combined cost value or directly perform a minimum value operation to extract the results. The node with the smallest value has its state data set. Assign the value to the target node of the current step. The calculation is as follows: ,in ; The selection of the minimum value here means that, after comprehensively considering obstacle avoidance, airspace control, and extreme energy consumption crisis intervention strategies, the node... This represents the safest and most economical space transfer decision that drones can make under the current physical conditions and environmental pressures.

[0098] Step 2: Extract the target node Then, it is pushed into the set of collaborative path trajectory sequences. At the tail end, complete the topology connection. Then, update the current node status of the drone to... Starting from this point, steps one through seven are triggered iteratively to expand outwards to new candidate nodes and recalculate the cost value. This iterative process of "computation-extraction-connection-recomputation" continues until the latest target node reaches or is sufficiently close to the final target node. .

[0099] .

[0100] Step 3: When the sequence After generation is complete (or a local sequence satisfying a certain time look-ahead window has been generated), the discrete set of nodes is packaged into a waypoint command protocol format recognizable by the flight control system. Subsequently, it is output to the attitude calculation module and power control module of the UAV through the data bus, guiding the UAV to fly according to the predetermined trajectory sequence.

[0101] In the aforementioned technology, a local optimum extraction algorithm (extracting the node with the minimum overall cost) is employed at the bottom layer of the three-dimensional grid topology network. Combined with the dynamic stacking connection operation of the data linked list, a complete and ordered set of collaborative path trajectory sequences is gradually constructed through iterative cycles in both time and space dimensions, and then output through a standard interface. All the technical advantages accumulated in the preceding steps (such as the absolute power preservation strategy in crisis situations and the smooth obstacle avoidance strategy in normal situations) are thoroughly solidified into a clear, continuous, and unambiguous three-dimensional spatial flight path, directly driving the UAV to avoid risks. This achieves a precise leap from mathematical decision-making to physical execution. Since each node in the sequence is the optimal solution selected layer by layer under an extremely rigorous dual-threshold crisis judgment and dynamic weight reconstruction mechanism, the entire trajectory sequence spliced ​​from these nodes can maintain the comprehensive balance of multiple objectives in normal situations and possess advanced dynamic robustness for power preservation and escape in extreme local conflict deadlocks, ensuring the global safety and local optima of the trajectory sequence.

[0102] Due to the classic local optimum trap in complex global path planning, traditional planning algorithms often only consider the absolute energy required for the next step (e.g., 500 joules), ignoring the implications of those 500 joules for the drone at that moment. If the drone has just taken off and is fully charged, consuming 500 joules is insignificant; however, if the drone is at the end of its mission and has only 5% battery remaining, those 500 joules could be fatal. Traditional methods lack a dynamic assessment of the relative relationship between individual actions and global assets, making them highly susceptible to short-sighted and reckless decisions under low-battery conditions, leading to crashes. Therefore, this paper proposes: After extracting the candidate node with the minimum comprehensive cost and before making the connection, the following steps are also included: The candidate node with the lowest comprehensive cost value is selected as the current pre-selected node. Obtain the local energy consumption prediction data corresponding to the currently selected node, and extract the global remaining power data from the real-time operating status data of the local machine; Calculate the proportion of local energy consumption prediction data to the global remaining power data, and generate single-step energy consumption deviation feedback data; Determine whether the single-step energy consumption deviation feedback data is greater than the preset single-step resource tolerance threshold; If so, a second adjustment to the final weight coefficient will be triggered, and the target node will be redefined; Otherwise, the currently selected node will be used as the target node.

[0103] Triggering a secondary adjustment to the final weight coefficients and redetermining the target node specifically includes: Using the single-step energy consumption deviation feedback data as a compensation factor, the final weight coefficients of each constraint characteristic data are secondarily corrected to obtain the final corrected weight coefficients of each constraint characteristic data. Based on the final corrected weight coefficients of each constraint feature data, the multidimensional constraint feature data is recalculated by weighted fusion to obtain the corrected comprehensive cost value of each candidate node. The node with the minimum modified comprehensive value is extracted again as the new target node.

[0104] Using single-step energy consumption deviation feedback data as a compensation factor, a secondary correction calculation is performed on the final weight coefficients of various constraint characteristic data, specifically including: The weighted compensation increment data is obtained by multiplying the single-step energy consumption deviation feedback data with the preset compensation gain coefficient. The weighted incremental data is superimposed onto the final weight coefficient of the energy consumption constraint feature data to generate the final corrected weight coefficient of the energy consumption constraint feature data. Then, a proportional attenuation calculation is performed on the final weight coefficients of the remaining constraint feature data to generate the final corrected weight coefficients of each constraint feature data.

[0105] Among them: Current pre-selected nodes: refer to candidate nodes that stand out after the initial weighted fusion calculation (with the smallest comprehensive cost), but have not yet undergone global energy security verification and are in a pending state.

[0106] Local energy consumption prediction data: refers to the absolute physical power (e.g., how many milliampere-hours or joules) that the drone is expected to consume in a single step motion from its current actual location to the current pre-selected node.

[0107] Global remaining battery power data: refers to the total remaining battery power actually available in the drone's battery system at the current moment.

[0108] Single-step energy consumption deviation feedback data: a dimensionless proportional value.

[0109] Single-step resource tolerance threshold: The system's preset safety red line, representing the maximum proportion of power consumption that the system can tolerate within a single planning step (for example, it is set to consume a maximum of 2% of the current remaining power in a single step).

[0110] Compensation gain coefficient: A proportional adjustment parameter in a control system, used to amplify or reduce the proportional error of energy consumption and convert it into a compensation amount applicable to the weighting domain.

[0111] Weighted compensation incremental data: After gain calculation, it is specifically used to be superimposed on the original energy consumption weight to further increase the specific value of energy consumption priority.

[0112] Final corrected weight coefficient: After two rounds of feedback adjustment, the weight factor is finally used for the second (reincarnation-style) node evaluation weighting calculation.

[0113] Step 1: The system does not directly... (The candidate node with the lowest overall value extracted, i.e., the currently selected node) is confirmed as the final target, and the drone flies to it. Expected power consumption And the drone's current total remaining battery power. .

[0114] The consumption ratio is then calculated, and single-step energy consumption deviation feedback data is generated. : ; System judgment Is it greater than the single-step resource tolerance threshold? : Otherwise, branch (safe): If This indicates that although this maneuver might involve a detour, its power consumption is entirely manageable relative to the current total power. At this point, the system directly accepts the maneuver. As the final output target node, it enters the next planning cycle.

[0115] Then the branch (dangerous): If This indicates that although the node scored the highest in the initial assessment, it would consume a very high proportion of the remaining power in this step, posing a serious risk of local overdraft, and thus a secondary correction mechanism must be triggered.

[0116] Step 2: After triggering the secondary correction, the system uses this excess deviation as a compensation factor. With compensation gain coefficient Multiply to calculate the additional weight compensation increment data that needs to be added. : ; This increment is forcibly added to the original final energy consumption weight to generate the final corrected weight coefficient. : ; The final weighting coefficients of the calculated energy consumption constraint characteristic data are obtained through... The function ensures that the weight limit does not overflow 1.0.

[0117] Step 3: To maintain the overall weighting system normalization (summing to 1), since the energy consumption weight is forcibly increased again, the weights of other spatial constraints must yield. The system calculates the secondary attenuation coefficient. : ; The final weights of the remaining constraints are then proportionally decayed: .

[0118] Step 4: At this point, the system has a set of final correction weighting coefficients that are more sensitive to energy consumption. The system uses this new set of weights to return to the fusion calculation step, and recalculates the weighted inner product of all candidate nodes in the current space to obtain the corrected comprehensive cost of each candidate node.

[0119] Finally, under the new evaluation system, the node with the minimum adjusted comprehensive cost value is re-examined and selected as the new target node. Since energy consumption has a very high weight at this point, the previously selected high-energy-consuming node is no longer considered a target node. The value of such nodes will inevitably skyrocket, leading to their elimination. The system will automatically select a safer node with lower energy consumption and greater conservatism.

[0120] In the aforementioned technology, a post-safety check gate based on the energy consumption ratio (single-step / global) is inserted before the final output node. If the tolerance threshold is reached, closed-loop feedback control logic is introduced to convert the deviation ratio into a gain multiplier, and to perform mandatory secondary weighting compensation and recalculation on the energy consumption weight. When the UAV has sufficient power, the denominator (global remaining power) is large, the feedback data is extremely small, and the system dares to perform large-scale maneuvers and obstacle avoidance; when the power is depleted, even a small maneuver will instantly amplify its consumption ratio and break through the threshold, forcing the system to become extremely conservative and energy-sparing, achieving the coexistence of planning strategy and current energy reserves. The secondary correction mechanism is equivalent to adding a "final audit program" to the UAV's decision-making system. Even if the primary algorithm gives a dangerous suggestion with high energy consumption due to certain extreme terrain, this post-feedback mechanism can accurately intercept it, and through secondary weight polarization, forcibly find an absolutely power-saving node among the remaining alternative nodes, effectively improving the reliability of long-endurance and complex mission end-of-flight operations.

[0121] A multi-constraint path collaborative planning system for urban low-altitude unmanned aerial vehicles (UAVs) includes: The data acquisition module is used to acquire three-dimensional raster environment data of the target area and real-time running status data of the local machine, and to fuse and calculate the multi-dimensional constraint feature data of each candidate node. The multi-dimensional constraint feature data includes at least energy consumption constraint feature data and static obstacle avoidance constraint feature data. The data processing module is used to perform numerical quantization processing on multidimensional constraint feature data to generate the benchmark weight coefficients of each constraint feature data; calculate the difference between the benchmark weight coefficients of energy consumption constraint feature data and the benchmark weight coefficients of other constraint feature data to generate the conflict correlation degree that characterizes the degree of constraint contradiction; and extract the real-time power consumption rate from the real-time operating status data of the local machine. The data judgment and weight adjustment module is used to determine whether the conflict correlation degree and consumption rate are both greater than the corresponding thresholds; if so, the baseline weight coefficients of the energy consumption constraint feature data are calculated exponentially, and the baseline weight coefficients of the remaining constraint feature data are calculated proportionally to generate the final weight coefficients of each constraint feature data; otherwise, the baseline weight coefficients of each constraint feature data are used as the final weight coefficients. The data output module is used to perform weighted fusion calculation of multidimensional constraint feature data and the final weight coefficients of each constraint feature data to obtain the comprehensive cost value of candidate nodes; in the three-dimensional grid node network data, the candidate node with the smallest comprehensive cost value is extracted as the target node and connected to generate a collaborative path trajectory sequence and output it.

[0122] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A multi-constraint path collaborative planning method for urban low-altitude unmanned aerial vehicles (UAVs), characterized in that, Includes the following steps: The system acquires three-dimensional grid environment data of the target area and real-time operating status data of the local machine, and integrates and calculates multi-dimensional constraint feature data of each candidate node. The multi-dimensional constraint feature data includes at least energy consumption constraint feature data and static obstacle avoidance constraint feature data. Numerical quantization processing is performed on multidimensional constraint feature data to generate benchmark weight coefficients for each constraint feature data; The difference between the baseline weight coefficient of the energy consumption constraint feature data and the baseline weight coefficient of other constraint feature data is calculated to generate a conflict correlation degree that characterizes the degree of constraint contradiction, and the real-time power consumption rate is extracted from the real-time operating status data of the local machine. Determine whether both the conflict correlation degree and the consumption rate are greater than the corresponding thresholds; If so, the baseline weight coefficients of the energy consumption constraint feature data are calculated exponentially, and the baseline weight coefficients of the remaining constraint feature data are calculated proportionally to generate the final weight coefficients of each constraint feature data. Otherwise, the baseline weight coefficients of each constraint feature data will be used as the final weight coefficients; The comprehensive cost of the candidate node is obtained by weighting and fusing the multidimensional constraint feature data with the final weight coefficients of each constraint feature data. In the 3D raster node network data, the candidate node with the lowest comprehensive cost value is extracted as the target node and connected to generate a cooperative path trajectory sequence and output it.

2. The urban low-altitude UAV multi-constraint path collaborative planning method according to claim 1, characterized in that: After extracting the candidate node with the minimum comprehensive cost and before making the connection, the following steps are also included: The candidate node with the lowest comprehensive cost value is selected as the current pre-selected node. Obtain the local energy consumption prediction data corresponding to the currently selected node, and extract the global remaining power data from the real-time operating status data of the local machine; Calculate the proportion of local energy consumption prediction data to the global remaining power data, and generate single-step energy consumption deviation feedback data; Determine whether the single-step energy consumption deviation feedback data is greater than the preset single-step resource tolerance threshold; If so, a second adjustment to the final weight coefficient will be triggered, and the target node will be redefined; Otherwise, the currently selected node will be used as the target node.

3. The urban low-altitude UAV multi-constraint path collaborative planning method according to claim 2, characterized in that: Triggering a secondary adjustment to the final weight coefficients and redetermining the target node specifically includes: Using the single-step energy consumption deviation feedback data as a compensation factor, the final weight coefficients of each constraint characteristic data are secondarily corrected to obtain the final corrected weight coefficients of each constraint characteristic data. Based on the final corrected weight coefficients of each constraint feature data, the multidimensional constraint feature data is recalculated by weighted fusion to obtain the corrected comprehensive cost value of each candidate node. The node with the minimum modified comprehensive value is extracted again as the new target node.

4. The urban low-altitude UAV multi-constraint path collaborative planning method according to claim 3, characterized in that: Using single-step energy consumption deviation feedback data as a compensation factor, a secondary correction calculation is performed on the final weight coefficients of various constraint characteristic data, specifically including: The weighted compensation increment data is obtained by multiplying the single-step energy consumption deviation feedback data with the preset compensation gain coefficient. The weighted incremental data is superimposed onto the final weight coefficient of the energy consumption constraint feature data to generate the final corrected weight coefficient of the energy consumption constraint feature data. Then, a proportional attenuation calculation is performed on the final weight coefficients of the remaining constraint feature data to generate the final corrected weight coefficients of each constraint feature data.

5. The urban low-altitude UAV multi-constraint path collaborative planning method according to claim 1, characterized in that: Acquire three-dimensional raster environmental data of the target area and real-time operating status data of the local machine, and also simultaneously acquire dynamic airspace control data and spatiotemporal trajectory prediction data of other UAVs in the same airspace; In addition to energy consumption constraint data and static obstacle avoidance constraint data, the multidimensional constraint feature data also includes collision avoidance interval constraint data and airspace control constraint data.

6. The urban low-altitude UAV multi-constraint path collaborative planning method according to claim 5, characterized in that: The specific steps for generating static obstacle avoidance constraint feature data, collision avoidance interval constraint feature data, and airspace control constraint feature data include: The entity features in the 3D raster environment data are abstracted into spatial bounding box data. The shortest distance from the candidate node to the outer edge of the spatial bounding box data is calculated to generate static obstacle avoidance constraint feature data. Extract the timestamp of the predicted arrival at the candidate node from the local machine, align it with the spatiotemporal trajectory prediction data of other UAVs, calculate the relative spatial distance under the same time section, and generate anti-collision interval constraint feature data. Calculate the distance from candidate nodes to the no-fly zone boundary in the dynamic airspace control data to generate airspace control constraint feature data.

7. The urban low-altitude UAV multi-constraint path collaborative planning method according to claim 6, characterized in that: The difference between the baseline weight coefficient of energy consumption constraint feature data and the baseline weight coefficient of other constraint feature data is calculated to generate a conflict correlation degree that characterizes the degree of constraint contradiction. Specifically, this includes: Extract the first benchmark weight coefficient corresponding to the energy consumption constraint feature data, and the second, third, and fourth benchmark weight coefficients corresponding to the static obstacle avoidance constraint feature data, collision avoidance interval constraint feature data, and airspace control constraint feature data, respectively; Calculate the absolute differences between the first benchmark weight coefficient and the second, third, and fourth benchmark weight coefficients, respectively. Extract the largest value from the absolute difference data and use it as the degree of conflict correlation.

8. The urban low-altitude UAV multi-constraint path collaborative planning method according to claim 1, characterized in that: The baseline weighting coefficients of the energy consumption constraint characteristic data are calculated exponentially, specifically including: Extract the first derivative of the real-time power consumption rate over a continuous sampling period to generate power change factor data that characterizes the acceleration of power loss. The data on the sudden change in electricity volume is used as the exponential term variable in the natural index calculation formula. The baseline weight coefficients of the energy consumption constraint characteristic data are multiplied by the calculation result of the natural exponential formula to generate the final weight coefficients, which are exponentially amplified.

9. The urban low-altitude UAV multi-constraint path collaborative planning method according to claim 1, characterized in that: Acquire 3D raster environment data of the target area, specifically including: Extract building density attribute data for the target area; Determine whether the building density attribute data is greater than the preset density threshold; If so, high-resolution three-dimensional raster environment data is generated using the first spatial spacing parameter; Otherwise, a second spatial spacing parameter greater than the first spatial spacing parameter is used to generate low-resolution 3D raster environment data.

10. A multi-constraint path collaborative planning system for urban low-altitude unmanned aerial vehicles (UAVs), characterized in that, include: The data acquisition module is used to acquire three-dimensional raster environment data of the target area and real-time running status data of the local machine, and to fuse and calculate the multi-dimensional constraint feature data of each candidate node. The multi-dimensional constraint feature data includes at least energy consumption constraint feature data and static obstacle avoidance constraint feature data. The data processing module is used to perform numerical quantization processing on multidimensional constraint feature data to generate the benchmark weight coefficients of each constraint feature data; calculate the difference between the benchmark weight coefficients of energy consumption constraint feature data and the benchmark weight coefficients of other constraint feature data to generate the conflict correlation degree that characterizes the degree of constraint contradiction; and extract the real-time power consumption rate from the real-time operating status data of the local machine. The data judgment and weight adjustment module is used to determine whether the conflict correlation degree and consumption rate are both greater than the corresponding thresholds; if so, the baseline weight coefficients of the energy consumption constraint feature data are calculated exponentially, and the baseline weight coefficients of the remaining constraint feature data are calculated proportionally to generate the final weight coefficients of each constraint feature data; otherwise, the baseline weight coefficients of each constraint feature data are used as the final weight coefficients. The data output module is used to perform weighted fusion calculation of multidimensional constraint feature data and the final weight coefficients of each constraint feature data to obtain the comprehensive cost value of candidate nodes; in the three-dimensional grid node network data, the candidate node with the smallest comprehensive cost value is extracted as the target node and connected to generate a collaborative path trajectory sequence and output it.

Images