Collaborative unmanned aerial vehicle cluster path planning and scheduling system

Abstract

The invention discloses a collaborative unmanned aerial vehicle cluster path planning and scheduling system, and particularly relates to the technical field of unmanned aerial vehicle intelligent control, and the system comprises a multi-mode sensing unit which is composed of a heterogeneous sensor array composed of LiDAR, binocular vision and millimeter wave radar, and an output dynamically updated three-dimensional Gaussian mixture map; the decision control unit is used for implementing double-layer optimization of mixed integer programming task allocation and artificial potential field path planning; the dynamic communication network adopts a hybrid networking protocol of TDMA backbone nodes and 802.11 ax terminal nodes; an energy management module; aiming at the insufficient environment perception and dynamic modeling capability in the prior art, the method achieves the effects that the centimeter-level positioning precision and the dynamic obstacle recognition rate are greater than 92%, the environment model is delayed and compressed to be within 200ms, the response speed is increased by 5 times by setting multi-modal sensor fusion, constructing a dynamically updated 3D Gaussian mixture map and combining an LSTM network to predict the obstacle trajectory in real time, and the dynamic obstacle recognition rate is greater than 92%. And the obstacle avoidance reliability in a complex scene is obviously enhanced.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle (UAV) intelligent control technology, and more specifically, to a collaborative UAV cluster path planning and scheduling system. Background Art

[0002] The collaborative UAV swarm path planning and scheduling system is of great significance. It improves mission execution efficiency and rationally allocates resources by optimizing path planning and efficient scheduling. It can achieve multi-machine collaboration to complete complex tasks, improve reliability, and enhance mission execution capabilities. It is widely used in military and civilian fields, expanding the application boundaries. It can also reduce flight energy consumption and equipment costs, avoid flight risks, and ensure personnel safety. It plays an irreplaceable core role in the modern UAV application system.

[0003] The existing technology has the following problems:

[0004] 1. Insufficient environmental perception and dynamic modeling capabilities

[0005] The lack of a multi-sensor collaborative fusion mechanism results in positioning errors greater than 3m, obstacle missed detection rates greater than 25% and environmental model update delays greater than 500ms in complex scenarios, making it difficult to support high-precision dynamic obstacle avoidance.

[0006] 2. Inefficient collaborative decision-making and task scheduling

[0007] Centralized architecture + static planning algorithm = single point failure risk. Task allocation for 50 nodes takes >30s, completion rate is <70%. Path replanning response delay increases collision risk by 3-5 times. Cluster expansion takes >30s.

[0008] 3. Inadequate communication and energy management

[0009] Conventional networking has a packet loss rate of >30% and a disconnection recovery time of >5s under interference. The extensive energy strategy causes energy consumption fluctuations of ±25%, the wireless charging scheduling error is >5m, and the battery life is limited.

[0010] 4. Bottlenecks in system robustness and verification efficiency

[0011] Traditional control architectures have weak anti-interference capabilities, low hardware redundancy, and physical verification cycles lasting several months. The recurrence rate of extreme scenarios is less than 20%, severely restricting the speed of technological iteration.

[0012] Therefore, to address the above problems, a collaborative UAV cluster path planning and scheduling system is proposed. Summary of the Invention

[0013] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a collaborative drone cluster path planning and scheduling system to solve the problems raised in the above-mentioned background technology.

[0014] To achieve the above objectives, the present invention provides the following technical solutions: a collaborative drone swarm path planning and scheduling system, comprising: a multimodal perception unit, which comprises a heterogeneous sensor array consisting of LiDAR, binocular vision, and millimeter-wave radar, and outputs a dynamically updated three-dimensional Gaussian mixture map; a decision control unit, which is used to implement a two-layer optimization of mixed integer programming task allocation and artificial potential field path planning, and output adaptive PID control instructions with six redundancies; a dynamic communication network, which adopts a hybrid networking protocol of TDMA backbone nodes and 802.11ax terminal nodes and integrates a lightweight Byzantine fault-tolerant consensus mechanism; and an energy management module, which can realize dynamic voltage and frequency regulation and glide path optimization based on atmospheric density.

[0015] Preferably, the decision control unit includes: a task allocator for establishing an objective function: minΣ(αt_ij+βe_ij)+γmax(t_ij), where t_ij is time consumption and e_ij is energy consumption; a path planner, which adopts an artificial potential field method with a repulsive field weight w=K(1+e^(-d / σ)), where d is the obstacle distance; and a conflict resolution module, which predicts trajectory anomalies through an LSTM network and generates a three-dimensional collision avoidance corridor.

[0016] Preferably, the multimodal sensing unit includes: a meteorological compensation submodule capable of integrating wind speed, precipitation and electromagnetic interference data in real time; a data calibration component that uses extended Kalman filtering to achieve sensor spatiotemporal alignment with an alignment error of <2 cm.

[0017] Preferably, the dynamic communication network includes: a backbone node election algorithm based on link quality evaluation of QL=0.7SNR+0.3BER-1; an adaptive modulation unit supporting dynamic switching from QPSK to 256QAM with a switching delay of <10ms.

[0018] Preferably, the energy management module includes: a wireless charging scheduler capable of calculating the optimal service radius R_opt=√(P_txη / (πP_th)); a gliding strategy generator: calculating the optimal gliding angle according to the atmospheric density ρ(h):

[0019] θ_glide=arctan[(C_D / C_L)(1+2W / (ρ(h)S√(C_L2+C_D2)))].

[0020] A path planning method based on any one of the systems of claims 1 to 5, comprising:

[0021] (a) The improved Voronoi diagram is used to segment the 3D space, and the segmentation weight includes the energy consumption factor;

[0022] (b) Quantum genetic algorithm optimization path, the chromosome encoding contains polar coordinates represented by 3 qubits

[0023] (c) Dynamic conflict resolution: Apply time dilation Δt=kd / (v_max-v_curr) or spatial offset Δp=R(1-e^(-t / τ))n to the conflicting paths.

[0024] Preferably, the fitness function of step (b) is:

[0025] F = ω1Σt_i+ω2Σe_i+ω3(1 / (1+Σc_i))+ω4ΣΔθ2, where c_i is the collision risk coefficient and Δθ is the change in heading angle.

[0026] A computer-readable medium storing the method according to claims 6-7, characterized in that: it integrates a federated learning framework, adopts θ_global = Σ(θ_local_i + Laplace(0,b)), implements differential privacy protection, and has a built-in Gazebo / ROS2 digital twin interface that supports real-time simulation of 150 nodes.

[0027] Technical effects and advantages of the present invention:

[0028] 1. To address the inadequacies of existing technologies in environmental perception and dynamic modeling, the present invention integrates multimodal sensors, LiDAR, binocular vision, and millimeter-wave radar to construct a dynamically updated 3D Gaussian mixture map. Combined with an LSTM network, this system predicts obstacle trajectories in real time, achieving centimeter-level positioning accuracy and a dynamic obstacle recognition rate exceeding 92%. The environmental model delay is compressed to within 200ms, increasing response speed by five times and reducing the obstacle miss rate to below 5%, significantly enhancing obstacle avoidance reliability in complex scenarios (such as urban buildings and dense vegetation).

[0029] 2. To address the low efficiency of collaborative decision-making and task scheduling in existing technologies, this paper adopts a two-layer optimization architecture of mixed integer programming (MILP) and Q-learning dynamic priority weighting, combined with a quantum genetic algorithm, to achieve distributed task allocation and path planning, supporting replanning of 50-node clusters in seconds.

[0030] 3. To address the inadequate communication and energy management performance of existing technologies, this invention designs a hybrid TDMA / 802.11ax networking protocol. The backbone nodes utilize lightweight Byzantine fault-tolerant consensus (disconnection recovery time <800ms). This protocol integrates dynamic voltage scaling (DVFS) with an atmospheric density-driven gliding strategy to optimize wireless charging service radius calculation. This reduces the communication packet loss rate from 30% to below 5%, narrows energy consumption fluctuations to ±8%, achieves a 28% overall energy saving, increases wireless charging station utilization to 85%, and extends battery life by 40%.

[0031] 4. To address the bottlenecks of system robustness and verification efficiency in existing technologies, this invention sets up six redundant actuators and adaptive PID control with a disturbance observer, and combines them with the Gazebo / ROS2 digital twin platform to realize 150-node extreme scenario simulation verification. It achieves an attitude control error of less than 0.5° under level 6 wind disturbance, and maintains a single-fault control efficiency of 80%. It also shortens the development cycle by 60%, reduces physical testing costs by 45%, and supports rapid iterative optimization of the federated learning framework. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 Schematic diagram of the workflow of the present invention. DETAILED DESCRIPTION

[0033] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0034] like Figure 1 As shown, a collaborative UAV cluster path planning and scheduling system is disclosed, which includes: a multimodal perception unit, which is composed of a heterogeneous sensor array consisting of LiDAR, binocular vision and millimeter wave radar, and outputs a dynamically updated three-dimensional Gaussian mixture map; a decision control unit, which is used to implement a two-level optimization of mixed integer programming task allocation and artificial potential field path planning, and outputs adaptive PID control instructions with six redundancy;

[0035] The dynamic communication network adopts a hybrid networking protocol of TDMA backbone nodes and 802.11ax terminal nodes, integrating a lightweight Byzantine fault-tolerant consensus mechanism; the energy management module can realize dynamic voltage and frequency adjustment and glide path optimization based on atmospheric density. Among them, LiDAR (10Hz scanning) provides centimeter-level point cloud, binocular vision (30fps) realizes texture recognition, millimeter-wave radar (77GHz) penetrates rain and fog, and data is fused through Kalman filtering (error <2cm). The upper-level MILP model (Gurobi solver) allocates tasks, and the lower-level potential field method (repulsive field coefficient σ=1.2) generates collision-free paths. The backbone node (TDMA time slot ≤5ms) ensures the transmission of key instructions, and the terminal node (802.11ax MU-MIMO) improves throughput. DVFS chips (such as TI The TPS65987D) dynamically adjusts voltage, and a gliding algorithm calculates the optimal pitch angle in real time based on barometer data. It integrates multimodal sensors and dynamic map updates to address blind spots in complex environmental perception. It offers full-scenario adaptability and achieves dual-layer real-time optimization of task allocation and path planning. A hybrid networking protocol ensures stable communication in high-interference environments, breaking through endurance bottlenecks through dynamic adjustment and gliding strategies.

[0036] As a preferred embodiment, the decision control unit includes: a task allocator for establishing an objective function: minΣ(αt_ij+βe_ij)+γmax(t_ij), where t_ij is time consumption and e_ij is energy consumption; a path planner, which adopts an artificial potential field method with a repulsive field weight w=K(1+e^(-d / σ)), where d is the obstacle distance; a conflict resolution module, which predicts trajectory anomalies through an LSTM network and generates a three-dimensional collision avoidance corridor. Further, mixed integer programming: constructing an objective function minΣ(0.6t_ij+0.3e_ij )+0.1max(t_ij), solved by branch and bound method, artificial potential field improvement: repulsion weight w=2.5*(1+e^(-d / 1.2)), when the obstacle distance d<5m, the weight increases exponentially, LSTM prediction network: input 10 frames of historical trajectory (100ms / frame), output the probability distribution map of the trajectory in the next 1s, MILP model combined with slack variables, reducing computational complexity by 30%, improving task allocation efficiency, adaptive repulsion field weight to avoid local optimal traps, dynamic obstacle avoidance capability, predictive conflict resolution, and can identify potential collision risks 300ms in advance.

[0037] As a preferred embodiment, the multimodal perception unit includes: a meteorological compensation submodule, which can integrate wind speed, precipitation and electromagnetic interference data in real time; a data calibration component, which uses extended Kalman filtering to achieve sensor spatiotemporal alignment, with an alignment error of <2cm. Furthermore, wind field data (collected by ultrasonic anemometer) is used for point cloud coordinate correction. The spatiotemporal alignment module uses hardware synchronization signals (PPS pulses) to unify the sensor clock, with a delay jitter of <1ms. The meteorological anti-interference capability still maintains 85% perception accuracy in a level 6 wind / rainstorm environment, cross-platform compatibility, and supports seamless access to sensors from different manufacturers.

[0038] As a preferred embodiment, the dynamic communication network includes: a backbone node election algorithm based on link quality evaluation of QL=0.7SNR+0.3BER-1; an adaptive modulation unit that supports dynamic switching from QPSK to 256QAM with a switching delay of <10ms. Furthermore, a weighted voting mechanism based on link quality QL=0.7*SNR(dB)+0.3*(1 / BER) dynamically selects the modulation order according to channel state information (CSI), with a spectrum efficiency of 8bps / Hz, maintaining an effective rate of 5Mbps under 150dB electromagnetic noise, and a network reconstruction time of <1s when 20% of the nodes fail.

[0039] As a preferred embodiment, the energy management module includes: a wireless charging scheduler capable of calculating the optimal service radius R_opt = √(P_txη / (πP_th)); a gliding strategy generator: calculating the optimal gliding angle according to the atmospheric density ρ(h):

[0040] θ_glide = arctan[(C_D / C_L)(1+2W / (ρ(h)S√(C_L2+C_D2)))] Furthermore, the angle of attack is adjusted in real time based on the atmospheric density lookup table (US Standard Atmosphere 1976 model) to calculate the service radius, optimizing charging efficiency. This reduces the density of wireless charging stations by 40%, achieves a breakthrough in glide energy savings, and reduces energy consumption during level flight by 35%.

[0041] A path planning method based on any one of the systems of claims 1 to 5, comprising:

[0042] (a) The improved Voronoi diagram is used to segment the 3D space, and the segmentation weight includes the energy consumption factor;

[0043] (b) Quantum genetic algorithm optimization path, the chromosome encoding contains polar coordinates represented by 3 qubits

[0044] (c) Dynamic conflict resolution: Apply time dilation Δt = kd / (v_max - v_curr) or spatial offset Δp = R(1-e^(-t / τ))n to the conflicting paths; improve the Voronoi diagram, introduce energy weighting factors, improve the efficiency of three-dimensional space segmentation, and increase the speed of Voronoi diagram generation by 5 times. A quantum genetic algorithm is used, and the chromosome is encoded with 3 qubits (r∈[0,100m],θ∈[0,360°], ), making the 50-node path search time less than 0.8s.

[0045] As a preferred embodiment, the fitness function of step (b) is:

[0046] F = ω1Σt_i+ω2Σe_i+ω3(1 / (1+Σc_i))+ω4ΣΔθ2, where c_i is the collision risk coefficient and Δθ is the heading angle change. The collision risk calculation and dynamic weight adjustment balance multiple objectives, improving the search efficiency of the Pareto optimal solution for time, energy consumption, and safety by 60%. Physical constraints are aligned, and the heading angle change Δθ2 reduces ineffective maneuvers by 30%.

[0047] A computer-readable medium storing the method as described in claims 6-7 integrates a federated learning framework, adopts θ_global = Σ(θ_local_i + Laplace(0,b)), and implements differential privacy protection. It has a built-in Gazebo / ROS2 digital twin interface and supports real-time simulation of 150 nodes. In particular, the federated learning framework is adopted, Laplace noise is added during parameter aggregation, the privacy budget ε = 0.5, and millisecond-level synchronization between the Gazebo simulator and the physical controller is achieved through the DDS protocol. The differential privacy mechanism reduces the data leakage risk to <5%, and supports stress testing of a 150-node cluster in a digital twin environment.

[0048] Detailed explanation of the workflow of the present invention:

[0049] After the system is started, it first collects environmental data in real time through a multimodal sensor array, LiDAR / vision / millimeter wave radar, and performs spatiotemporal alignment through an extended Kalman filter to construct a dynamically updated 3D Gaussian mixture map. At the same time, meteorological sensors compensate for the impact of wind field / precipitation on perception; after the decision layer receives the task instruction, the upper-level mixed integer programming engine (MILP) combined with Q-learning dynamic priority weights completes the 50-node task allocation and generates the spatiotemporal constraint matrix within 50ms, and the lower-level improved artificial potential field method plans the initial path based on the repulsive field model (σ=1.2); the path planner uses a quantum genetic algorithm to encode 3-qubit polar coordinates and optimize the three-dimensional voxel space (5m×5m×3m grid), predicts the obstacle trajectory in the next 1s through the LSTM network, and implements time dilation (Δt=kd / Δv) or spatial offset (Δp=R(1-e^(-t / τ)) on the conflicting path in real time; the control layer uses six redundant The system uses a high-speed actuator and an adaptive PID (Lyapunov exponential feedback) with a disturbance observer, outputting control commands at a 10Hz frequency, while dynamic voltage scaling (DVFS) optimizes energy consumption based on flight status. The communication network uses TDMA backbone nodes with a time slot of ≤5ms, and is hybrid-networked with 802.11ax terminal nodes. A lightweight Byzantine fault-tolerant protocol (an improved PBFT) is used to achieve a 95% data integrity rate under 20% node failures. The energy management module calculates the optimal glide angle θ_glide based on the atmospheric density model (USSA-76) and dynamically schedules the service radius of wireless charging stations. Full-process data is synchronized to the Gazebo / ROS2 digital twin platform for 150-node extreme scenario verification. The federated learning framework, with differential privacy ε=0.5, continuously optimizes parameters to form a closed-loop workflow of "perception → decision → execution → verification → iteration", achieving a 200ms dynamic obstacle avoidance response, an 89% task completion rate, and 28% overall energy savings.

[0050] Example 1: Forest fire prevention work

[0051] 1. Hardware Configuration

[0052] Drone platform: The quadrotor maintains a 350mm wheelbase and features a carbon fiber frame (treated with a high-temperature-resistant coating). The sensor module includes a thermal imaging camera (FLIR Tau2, 640×512 resolution, temperature range -40°C to 550°C); a gas sensor (MQ-2 smoke sensor + CO detection module); and a Livox Mid-40 LiDAR (which penetrates smoke and generates real-time 3D terrain maps).

[0053] Mission payload: Fire extinguishing bomb bay (500g payload, capable of carrying four chemical fire extinguishing bombs); emergency material delivery device (maximum delivery weight 300g); and communication module, which consists of a satellite communication relay (Beidou short message module, supporting communication in areas without network access) and an enhanced image transmission system. The enhanced image transmission system uses a 2.4GHz + 5.8GHz dual-band transmission range of 5km (in complex terrain).

[0054] Ground facilities: Mobile command station, which integrates the QGC ground station and fire analysis system; portable charging station, which can support fast battery replacement (battery replacement can be completed in 5 minutes).

[0055] 2. Software parameter configuration

[0056] Path planning: Fire spread model integration: Predict fire paths based on meteorological data (wind speed, humidity) 3. Typical workflow;

[0057] Emergency obstacle avoidance parameters: conflict_zones = predict_fire_spread(wind_speed,humidity)

[0058] safe_altitude = max(50m,2×flame_height) #Safe flight altitude calculation

[0059] Quantum genetic algorithm weight adjustment: ω1 = 0.5 (fire urgency), ω2 = 0.3 (endurance), ω3 = 0.2 (terrain complexity);

[0060] Communication protocol: includes anti-interference mode, which can automatically switch to TDMA backbone node to dominate communication in smoke environment; emergency channel, which reserves 400MHz frequency band for Beidou short message transmission (packet loss rate <5%).

[0061] 3. Forest fire prevention workflow

[0062] Phase 1. Fire reconnaissance phase

[0063] Multi-drone coordinated launch: Three drones form a formation. The master drone is equipped with a thermal imaging camera (FLIR Tau2) to scan the surface temperature (accuracy ±2°C). Slave drone 1 uses LiDAR to build a 3D map of the fire scene (updated at 1Hz). Slave drone 2 uses an MQ-2 sensor to monitor the CO concentration in real time (threshold >50ppm triggers an alarm).

[0064] Data fusion and labeling: The onboard computer (Jetson Nano) integrates heat source distribution, terrain data, and gas concentrations, uses the ORB-SLAM algorithm to mark the fire boundary (error <3m), and predicts the fire spread path (emergency mode is activated when wind speed >8m / s).

[0065] Real-time backhaul: Fire scene coordinates and temperature gradient maps are synchronized to the ground command station through 5.8GHz image transmission (5km range) and Beidou short message dual channels.

[0066] Phase 2. Firefighting mission execution phase

[0067] Dynamic task allocation: The MILP model uses minΣ(0.4t_ij+0.3e_ij+0.3d_fire) as the objective function to allocate fire bomb throwing (covering a radius of 8m) and beacon delivery tasks. The quantum genetic algorithm optimizes the flight path (avoiding updraft areas).

[0068] Anti-disturbance control: Under heat wave disturbances, the adaptive PID with sliding mode observer (overshoot <3%) adjusts the motor speed in real time, and combined with DVFS, dynamically adjusts the voltage (energy consumption is reduced by 15%).

[0069] Firefighting operation: After arriving at the target point, the fire extinguishing magazine solenoid valve is triggered (response time 0.2s), the positioning beacon (UWB accuracy ±0.1m) is released simultaneously, and the firefighting effect is verified by LiDAR.

[0070] Phase 3. Emergency return and maintenance

[0071] Return trigger: When the battery voltage is less than 14.8V or the fuselage temperature is greater than 60℃ (for 10 seconds), it will automatically switch to TDMA backbone communication mode and return to home at a safe altitude (≥50m).

[0072] Quick maintenance: After landing, use compressed air to cool the motor (cool down to 40°C in 2 minutes), clean the thermal imaging lens with anhydrous ethanol (to prevent smoke and dust adhesion), and check the carbon brush wear (limit value <3mm).

[0073] Data review: A federated learning framework (differential privacy ε = 0.5) analyzes task data, optimizes the fire prediction model, and iteratively updates the cluster system.

[0074] Implementation Results: The forest fire prevention workflow begins with a three-aircraft formation coordinated reconnaissance: the master aircraft uses thermal imaging to accurately scan the fire scene's temperature distribution, slave aircraft 1 utilizes LiDAR to build a 3D terrain model, and slave aircraft 2 monitors CO2 concentrations in real time. This data is fused via Jetson Nano to mark the fire boundary and predict the spread path. During the firefighting phase, the MILP model dynamically allocates tasks, a quantum genetic algorithm optimizes the bombing path, and adaptive PID control is combined to mitigate heat wave disturbances, accurately dropping fire bombs (with an accuracy of ±5m) and navigation beacons. Low battery voltage or high temperature triggers an emergency return, ensuring stable communication via dual TDMA / Beidou channels. Upon return, the motor is rapidly cooled and the sensor is cleaned. Federated learning analyzes the task data and optimizes the algorithm. This entire process reduces fire response time to 2.5 minutes and improves firefighting efficiency by 67%.

[0075] Finally, a few points should be explained: First, in the description of this application, it should be noted that, unless otherwise specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense, and may refer to mechanical or electrical connections, internal communication between two components, or direct connection. "Up," "down," "left," and "right" are only used to indicate relative positional relationships. When the absolute positions of the objects being described change, the relative positional relationships may also change.

[0076] Secondly: The drawings of the embodiments disclosed in the present invention only involve structures related to the embodiments disclosed in the present invention. Other structures may refer to conventional designs. The same embodiment and different embodiments of the present invention may be combined with each other without conflict.

[0077] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A collaborative UAV cluster path planning and scheduling system, characterized by: It includes: a multimodal perception unit, which is composed of a heterogeneous sensor array consisting of LiDAR, binocular vision and millimeter-wave radar, and outputs a dynamically updated three-dimensional Gaussian mixture map; a decision control unit, which is used to implement a two-level optimization of mixed integer programming task allocation and artificial potential field path planning, and outputs adaptive PID control instructions with six redundancies; The dynamic communication network adopts a hybrid networking protocol of TDMA backbone nodes and 802.11ax terminal nodes, integrating a lightweight Byzantine fault-tolerant consensus mechanism; the energy management module can achieve dynamic voltage and frequency regulation and glide path optimization based on atmospheric density.

2. A collaborative UAV cluster path planning and scheduling system as claimed in claim 1, characterized in that: The decision control unit includes: a task allocator for establishing the objective function: minΣ(αt_ij+βe_ij)+γmax(t_ij), where t_ij is the time consumption and e_ij is the energy consumption; a path planner, which adopts the artificial potential field method with repulsive field weight w=K(1+e^(-d / σ)), where d is the obstacle distance; and a conflict resolution module, which predicts trajectory anomalies through an LSTM network and generates a three-dimensional collision avoidance corridor.

3. A collaborative UAV cluster path planning and scheduling system as claimed in claim 1, characterized in that: The multimodal sensing unit includes: a meteorological compensation submodule that can fuse wind speed, precipitation and electromagnetic interference data in real time; and a data calibration component that uses an extended Kalman filter to achieve temporal and spatial alignment of sensors with an alignment error of less than 2 cm.

4. A collaborative UAV cluster path planning and scheduling system as claimed in claim 1, characterized in that: The dynamic communication network includes: a backbone node election algorithm based on link quality evaluation of QL=0.7SNR+0.3BER-1; Adaptive modulation unit supports dynamic switching from QPSK to 256QAM with a switching delay of <10ms.

5. The collaborative UAV cluster path planning and scheduling system according to claim 1, characterized in that: The energy management module includes: a wireless charging scheduler that can calculate the optimal service radius R_opt = √(P_txη / (πP_th)); a gliding strategy generator that calculates the optimal gliding angle according to the atmospheric density ρ(h): θ_glide=arctan[(C_D / C_L)(1+2W / (ρ(h)S√(C_L2+C_D2)))].

6. A path planning method based on any system of claims 1-5, characterized in that: Include: (a) The improved Voronoi diagram is used to segment the 3D space, and the segmentation weight includes the energy consumption factor; (b) Quantum genetic algorithm optimization path, the chromosome encoding contains polar coordinates represented by 3 qubits (c) Dynamic conflict resolution: Apply time dilation Δt=kd / (v_max-v_curr) or spatial offset Δp=R(1-e^(-t / τ))n to the conflicting paths.

7. The path planning method according to claim 6, wherein: The fitness function of step (b) is: F = ω1Σt_i+ω2Σe_i+ω3(1 / (1+Σc_i))+ω4ΣΔθ2, where c_i is the collision risk coefficient and Δθ is the heading angle change.

8. A computer-readable medium storing the method according to claim 6-7, characterized in that: The integrated federated learning framework adopts θ_global=Σ(θ_local_i+Laplace(0,b)) to achieve differential privacy protection. It has a built-in Gazebo / ROS2 digital twin interface and supports real-time simulation of 150 nodes.

Images