Multi-unmanned aerial vehicle negotiation anti-collision method and system based on 5G unmanned aerial vehicle communication

Abstract

The invention discloses a multi-unmanned aerial vehicle negotiation anti-collision method and system based on 5G unmanned aerial vehicle communication, and belongs to the technical field of multi-unmanned aerial vehicle risk avoiding. The method comprises the following steps: collecting multi-source flight state data; after the multi-source flight state data are fused through Kalman filtering, an airspace situation map is constructed; a 5G network architecture is constructed, and real-time data interaction between the unmanned aerial vehicles is realized; a dynamic negotiation protocol is constructed, and fast exchange of obstacle avoidance proposals when collision risks are triggered by multiple machines is supported; and constructing a distributed collaborative obstacle avoidance algorithm, and generating a smooth obstacle avoidance path in real time. According to the invention, through a dynamic negotiation protocol, multiple machines are supported to rapidly exchange obstacle avoidance proposals when a collision risk is triggered; a global optimal strategy is generated through edge node intelligent arbitration; the priority mechanism gives consideration to task urgency and environment dynamic change, ensures that a high-value task unmanned aerial vehicle passes preferentially, balances fairness through distributed voting, avoids path stiffness caused by decision conflict or resource competition, and realizes efficient and reasonable collaborative decision.

Description

Technical Field

[0001] The present invention belongs to the technical field of multi-UAV risk avoidance, and specifically relates to a multi-UAV negotiation anti-collision method and system based on 5G UAV communication. Background Art

[0002] In recent years, with the technological advancements of 5G communications, edge computing, and swarm intelligence algorithms, UAV collaborative obstacle avoidance technology has gradually evolved from single-drone autonomy to multi-drone collaboration. By deploying onboard sensor networks and real-time communication systems, drones can obtain data such as the position, velocity, and heading of surrounding aircraft, and implement dynamic path planning using collaborative decision-making algorithms such as distributed model predictive control and multi-agent reinforcement learning. However, in multi-drone flight, traditional obstacle avoidance methods such as vision, radar, and GPS have limitations, as flight trajectories are complex and often exceed the perception range of a single drone. They are unable to effectively and timely detect and avoid collision risks between multiple drones. Existing technologies mostly rely on single sensing devices, such as vision or radar sensors. However, these methods are prone to misjudgment when handling collision prediction and path planning in complex environments and lack multi-drone collaboration and information sharing.

[0003] Existing technologies have achieved phased results in the field of multi-UAV collaborative obstacle avoidance, but they still face a fundamental conflict between collaborative mechanisms and real-time performance. Centralized solutions based on model predictive control (MPC) use a ground control station to globally optimize paths, which can reduce collision probability. However, these solutions rely on high-precision environmental models, consume a long computational time in dynamic airspace, and are unable to cope with sudden obstacles. Single-UAV obstacle avoidance algorithms driven by reinforcement learning utilize deep Q networks for end-to-end obstacle avoidance decision-making, resulting in high collision avoidance success rates but failing to account for multi-UAV scenarios. This can easily lead to "avoidance oscillation" in actual deployments. State-sharing systems based on broadcast communications can synchronize basic information, but lack a dynamic negotiation mechanism. In the event of multi-UAV conflict, they rely solely on one-way instructions or preset rules to adjust paths, leading to inefficiency and even conflicting decisions.

[0004] It can be seen that although some systems have attempted to achieve information sharing between drones through communication networks, most systems are limited to broadcasting flight status information and lack the ability to negotiate multiple drones and adjust paths in real time.

[0005] A Chinese patent discloses (application number: 202411680927.4) a method and device for automatic identification and obstacle avoidance of drone area boundaries. The invention method includes obtaining environmental perception data from drone sensors, classifying the environmental perception data into types, and obtaining corresponding image data, point cloud data, and continuous image frames; performing boundary recognition processing on the image data to obtain corresponding boundary recognition results, and performing obstacle feature extraction on the point cloud data to obtain corresponding obstacle features; performing long-range dependency feature prediction on continuous image frames through a preset Transformer model to obtain corresponding motion prediction information; performing obstacle avoidance path planning on the three-dimensional environment map based on dynamic boundary information to obtain the corresponding obstacle avoidance path; performing decision analysis on the obstacle avoidance path, dynamic boundary information, and preset mission objectives to generate automatic obstacle avoidance decisions.

[0006] A Chinese patent application (application number: 202411283767.X) discloses a method for distributed dynamic collaborative task allocation for fixed-wing UAV clusters, including the following steps: S1, obtaining status information of the environment, UAVs, and tasks; S2, constructing a task allocation mathematical model based on the task requirements and the aforementioned status information; S3, each UAV updates the sequence of tasks to be executed based on its own status and task requirements; S4, neighboring UAVs share task information via links and negotiate conflicting tasks; S5, obtaining emergency event information; S6, updating the aforementioned status information based on the emergency event information and task status; S7, looping through S3-S6 until a consistent task allocation result is obtained, and outputting the final task allocation result. By designing multiple evaluation indicators and constraints, a cluster task allocation mathematical model is constructed; this mathematical model is solved by combining interactive information and emergencies. This invention can reduce cluster communication requirements and improve the robustness of dynamic task allocation results while ensuring the effectiveness of the allocation results.

[0007] The patented method and device for automatic identification of drone area boundaries and obstacle avoidance have the following shortcomings: 1. Path planning relies on static boundary information of a three-dimensional environment map, resulting in a lag in response to sudden obstacles or dynamic targets;

[0008] 2. Obstacle avoidance decisions are based on a preset fixed rule base and lack the ability to globally optimize the real-time airspace heat map. This leads to problems such as redundant avoidance paths and increased energy consumption in scenarios with dense dynamic obstacles, making it difficult to meet the efficiency and robustness requirements of multi-aircraft collaboration in complex airspace.

[0009] The patent discloses a distributed dynamic collaborative task allocation method for fixed-wing UAV swarms, but the method has the following shortcomings: 1. Although the task allocation method reduces communication pressure through distributed negotiation, its periodic information synchronization mechanism leads to high decision-making delays, making it difficult to support millisecond-level obstacle avoidance negotiation requirements. In particular, when an emergency event is triggered, multiple rounds of negotiation processes need to be restarted, making it impossible to quickly generate a consistency strategy driven by real-time data.

[0010] 2. The mathematical model relies on centralized constraints, consumes a lot of local computing power, has poor compatibility with low-configuration drones, and is difficult to achieve lightweight deployment of large-scale clusters.

[0011] Therefore, how to solve the problems of high communication delay, low collaborative decision-making efficiency, and delayed obstacle avoidance response and insufficient global optimization in collaborative flight of multiple UAVs in dynamic environments is the technical problem that the present invention aims to solve. Summary of the Invention

[0012] The purpose of the present invention is to provide a multi-UAV negotiation and collision avoidance method and system based on 5G UAV communication to solve the problems raised in the above background technology.

[0013] The object of the present invention is achieved as follows: a multi-UAV negotiation and collision avoidance method based on 5G UAV communication, characterized in that the method comprises the following steps:

[0014] Step S1: Collect multi-source flight status data and capture the surrounding building outlines through binocular vision to generate a depth map;

[0015] Step S2: Multi-source flight status data is fused using Kalman filtering to construct an airspace situation map;

[0016] The airspace situation map contains the location, speed, and obstacle information of the UAV;

[0017] Step S3: Build a 5G network architecture to enable real-time data interaction between drones;

[0018] Step S4: Build a dynamic negotiation protocol to support multiple machines to quickly exchange obstacle avoidance proposals when collision risk is triggered;

[0019] Step S5: Build a distributed collaborative obstacle avoidance algorithm to generate a smooth obstacle avoidance path in real time.

[0020] Preferably, in step S3, a 5G network architecture is constructed to realize real-time data interaction between drones, specifically:

[0021] Step S3-1: Each drone is equipped with a 5G NR-U communication unit for data transmission;

[0022] The data transmission content includes: real-time flight status, path planning proposals, obstacle point cloud from millimeter-wave radar / visual perception, weather warning information and priority, and mission metadata of remaining battery power;

[0023] Step S3-2: Multiple drones share the same time-frequency resource block through sparse code multiple access (SCMA) technology, specifically:

[0024] Define the codebook matrix Where K is the number of users, N is the number of resource blocks, and the sparsity factor is d f ,satisfy:

[0025] ||C k ||0=d f ,

[0026] Step S3-3: Based on the real-time channel state information CSI, the high-priority UAV seizes the high signal-to-noise ratio subcarrier and adopts the message passing algorithm MPA demodulation:

[0027]

[0028] Among them, x k is the symbol bit variable of the kth UAV, p(y n |x) is the received signal y n Likelihood probability density function under known symbol bit x; μ j is the message value sent by adjacent node j to node k in the message passing algorithm MPA; x j is the symbol bit variable of neighbor node j;

[0029] Priority P i The rule for selecting subcarrier index m for the UAV is:

[0030]

[0031] Among them, h i,m is the channel response, σ 2 is the noise power.

[0032] Preferably, the dynamic negotiation protocol includes conflict detection, proposal interaction, arbitration mechanism and priority strategy, specifically:

[0033] Conflict detection: Based on the trajectory prediction model, define the collision risk index between the two aircraft;

[0034] The trajectory prediction model is based on the uniform acceleration motion model and Kalman filter prediction, fully considering the dynamic changes of the UAV;

[0035] The input parameters include the current position coordinates of the drone (x t ,y t ,z t ), the current velocity vector of the UAV (v x ,v y ,v z ), the current moment of the UAV's acceleration vector (a x ,a y ,a z) and the prediction step size τ, assuming that the acceleration remains unchanged within the short time prediction window, the future position prediction is given by the following formula:

[0036]

[0037] Where τ is the predicted collision time offset;

[0038] A one-dimensional Kalman filter is used to recursively update the position, velocity and acceleration, and output the trajectory sequence {(x t+τ ,y t+τ ,z t+τ )}, providing input for collision risk assessment and obstacle avoidance proposals;

[0039] Proposal interaction: The initiator sends an obstacle avoidance proposal, and the recipient responds with an acceptance or counter-proposal within a specified time.

[0040] Arbitration mechanism: If there is a conflict in proposals, the edge node calculates the global optimal solution based on the Nash equilibrium algorithm, giving priority to protecting the high-priority drone path;

[0041] Priority strategy: includes static priority and dynamic weight allocation. Static priority is pre-defined based on task urgency, while dynamic weight is dynamically adjusted based on remaining power and environmental risks.

[0042] Preferably, the Nash equilibrium algorithm is specifically:

[0043]

[0044] in, is the optimal strategy for the i-th UAV, s i is any feasible strategy for the i-th UAV, u i is the utility function of the i-th drone, which is used to evaluate the benefits under different strategies.

[0045] Preferably, step S4 supports multiple drones to quickly exchange obstacle avoidance proposals when a collision risk is triggered, and by calculating the relative motion relationship between drones and predicting the future collision time, active prediction of collision risk is achieved, specifically:

[0046] Predict collision time t based on relative position Δp and velocity Δv c , the collision risk value is calculated as:

[0047]

[0048] where e is a natural constant; λ is the risk slope factor; and τ is the predicted collision time offset, representing the threshold center time, which is used to balance the model's response to early and late collision times.

[0049] Preferably, in step S5, a distributed collaborative obstacle avoidance algorithm is constructed to generate a smooth obstacle avoidance path in real time, specifically:

[0050] Step S5-1: The drone uploads its perceived obstacle data in real time. The edge node generates a global airspace heat map and distributes it to all drones via the 5G network. The node feature update formula is:

[0051]

[0052] Among them, α ij is the attention weight, σ is the ReLU activation function; W (l) is the weight matrix of the lth layer, is the feature vector of node i in layer l, is the set of neighbor nodes;

[0053] Step S5-2: Each machine generates a local obstacle avoidance strategy based on the received global data through the graph neural network (GNN). The objective function is decomposed into local sub-problems:

[0054]

[0055] Among them, x i is the local obstacle avoidance strategy variable of the i-th UAV; ρ is the penalty factor, u i is the Lagrange multiplier corresponding to the i-th UAV; f i is the local objective function of the i-th UAV; z is the global consistency variable;

[0056] Step S5-3: Use the alternating direction multiplier method ADMM distributed optimization framework to iteratively reach global consensus.

[0057] Preferably, the alternating direction multiplier method ADMM outputs a composite adjustment instruction of heading, speed, and altitude, and the expression of the alternating direction multiplier method ADMM is:

[0058]

[0059] Among them, ||r k ||2 is the original residual, ||s k ||2 is the dual residual; ρ k is the penalty factor for the kth iteration;

[0060] When the original residual || r k ||2 is significantly larger than the dual residual ||s k ||2, indicating that the current solution deviates greatly from the original constraints and needs to be tightened to reduce the accelerated convergence; the dual residual ||s k ||2 is significantly larger than the original residual ||r k||2, indicating that the dual variable fluctuates greatly, and the penalty factor needs to be reduced to stabilize the optimization process; if the residuals of the two are relatively balanced, maintain the current ρ k value;

[0061] By dynamically adjusting ρ, the ADMM algorithm adaptively balances the feasibility of the original problem and the convergence of the dual variable, avoiding the tedious manual adjustment of parameters and improving the solution efficiency of multi-UAV path optimization.

[0062] A multi-UAV negotiation and collision avoidance system based on 5G UAV communication, characterized by comprising: a multi-UAV negotiation and collision avoidance method based on 5G UAV communication according to any one of claims 1 to 7;

[0063] The system includes a 5G communication module, a flight status receiving module, a collision risk assessment module, a negotiation module, a flight trajectory adjustment module, and a fusion interface module. The 5G communication module includes a 5G NR-U communication unit. Each drone is equipped with a 5G NR-U communication unit, which supports sparse code multiple access technology, allowing multiple drones to share the same time-frequency resource block and broadcast flight status including timing position, speed, heading, altitude, and mission priority.

[0064] The 5G communication module also includes a dynamic priority slot allocation algorithm, which allows high-priority drones to occupy low-interference channels first, ensuring millisecond-level transmission of critical emergency avoidance instructions;

[0065] Flight status receiving module: This module eliminates transmission delay and noise through the extended Kalman filter (EKF), constructs an airspace situation map in real time, and provides high-precision input for collision risk assessment.

[0066] Collision Risk Assessment Module: This module uses a trajectory prediction algorithm to calculate the probability of collision within a specified timeframe based on a dynamic model of relative speed and distance. This is then mapped to a risk value using a Sigmoid function. The trigger threshold is then set to balance false alarm rate with response speed.

[0067] Negotiation module: Integrates a distributed game solver to initiate multi-machine real-time negotiation via the 5G network when risks are detected;

[0068] Flight trajectory adjustment module: Generates smooth adjustment instructions based on the PID control algorithm, dynamically corrects the flight path, and supports compound adjustment of heading, speed, and altitude to ensure stable execution of obstacle avoidance actions and avoid the risk of loss of control caused by violent maneuvers;

[0069] Fusion interface module: used to bridge the gap between traditional sensing equipment and new 5G communication technology, achieving seamless integration of multi-level obstacle avoidance capabilities.

[0070] Compared with the prior art, the present invention has the following improvements and advantages:

[0071] 1. Through a dynamic negotiation protocol, multiple drones can quickly exchange obstacle avoidance proposals when collision risks are triggered, and the global optimal strategy is generated through intelligent arbitration of edge nodes. The priority mechanism takes into account both mission urgency and dynamic environmental changes, ensuring that drones with high-value missions have priority passage. At the same time, distributed voting balances fairness, avoids path deadlocks caused by decision conflicts or resource competition, and achieves efficient and reasonable collaborative decision-making.

[0072] 2. Through distributed optimization algorithms and intelligent control technology, the results of multi-machine negotiation are integrated to generate a smooth obstacle avoidance path in real time. By aggregating neighboring aircraft status information and the global airspace situation, the heading, speed, and altitude of each drone are dynamically adjusted to ensure accurate and stable obstacle avoidance actions. This not only avoids the risk of loss of control caused by violent maneuvers, but also maintains the overall efficiency of the cluster mission, adapting to the collaborative needs of high-speed flight and beyond-visual-range scenarios.

[0073] 3. Real-time data interaction between drones is achieved through the 5G network architecture. Advanced non-orthogonal multiple access technology is used to allow multiple drones to share communication resources. Combined with a dynamic priority mechanism, it ensures the rapid transmission of key commands. High-priority drones can take priority in occupying high-quality channels and maintain millisecond-level response capabilities in dense airspace, making multi-drone status synchronization and obstacle avoidance command transmission seamless, significantly improving communication efficiency and reliability in complex scenarios.

[0074] 4. By integrating vision, radar, GPS / Beidou / RTK, IMU and other sensing devices, a multi-level obstacle avoidance capability is built. It is compatible with mainstream hardware protocols and data formats, and supports existing drones to access the system through low-cost upgrades. While retaining the original sensor performance, 5G enhances perception and collaborative decision-making, significantly improving environmental adaptability and comprehensive obstacle avoidance efficiency, providing a flexible and economical solution for large-scale deployment. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 Schematic diagram of the process of the present invention.

[0076] Figure 2 Schematic diagram of the structure of the multi-UAV negotiation collision avoidance system.

[0077] Figure 3 Schematic diagram of the real-time position and velocity vectors of multiple drones.

[0078] Figure 4 Schematic diagram of the drone's obstacle avoidance trajectory. DETAILED DESCRIPTION

[0079] The present invention is further summarized below with reference to the accompanying drawings.

[0080] like Figure 1 As shown, a multi-UAV negotiation and collision avoidance method based on 5G UAV communication includes the following steps:

[0081] Step S1: Collect multi-source flight status data and capture the surrounding building outlines through binocular vision to generate a depth map;

[0082] When a swarm of express delivery drones is performing a mission, each drone obtains centimeter-level positioning data in real time through the RTKGPS / Beidou module, combines it with millimeter-wave radar to detect obstacles within 200 meters, such as other drones and birds, and uses binocular vision to capture the contours of surrounding buildings to generate a depth map.

[0083] The depth map is generated by capturing the outline of surrounding buildings through binocular vision, specifically:

[0084] The drone's binocular camera collects synchronized image pairs from the left and right perspectives, and uses the parallax information of binocular vision to perform depth estimation;

[0085] First, corresponding pixel pairs in the left and right images are identified through pixel block matching to generate a disparity map. Then, based on the baseline length B of the binocular camera and the camera focal length f, the disparity value d is converted to a depth value Z using the following triangulation relationship:

[0086]

[0087] Here, disparity d represents the horizontal positional offset between the left and right camera imaging planes for the same object point in the image. Finally, applying the above formula to each pixel in the image yields a complete 3D depth map, reflecting the building outlines, the spatial distribution of obstacles, and their relative depth relationships. The depth map generated through this process provides real-time visualization of the 3D spatial form of surrounding buildings, assisting with subsequent airspace situational awareness and obstacle avoidance path planning, and providing highly accurate and dynamic environmental input data for multi-UAV collaborative flight.

[0088] Step S2: Multi-source flight status data is fused using Kalman filtering to construct an airspace situation map;

[0089] The airspace situation map contains the drone's position, speed, and obstacle information. When drone A flies over a residential area, its radar detects drone B 50 meters ahead. The visual sensor simultaneously identifies the gaps between buildings, and the IMU provides a real-time velocity vector. This data is fused through a Kalman filter to construct an airspace situation map containing position, speed, and obstacle information, providing precise input for subsequent collaborative obstacle avoidance.

[0090] Drone A detects that the relative distance to Drone B has shortened to 40 meters (preset threshold: 50 meters). Based on relative speed (e.g., A is 5m / s eastbound and B is 4m / s southbound), the system predicts that the two will intersect in 6 seconds. The collision probability is calculated to be 65% (exceeding the 30% threshold), immediately triggering the negotiation process. Drone A, with high priority, sends a "turn right 15°" proposal, and Drone B, with low priority, responds with a counter-proposal of "slow down to 3m / s." The edge node arbitrates based on task priority (assuming A is 9 and B is 5) and ultimately selects A's right turn solution.

[0091] Using drones as nodes and collision risk as edge weights, GNN aggregates the states of neighboring drones. Drone A's GNN model analyzes its interactions with drones B, C, and D, generating a local strategy of "turn right 10° and climb 3 meters." Combined with ADMM distributed optimization, each drone iterates within 20ms to reach global consensus, avoiding path oscillations caused by conflicting individual drone policies (e.g., two drones simultaneously turning left, causing a secondary collision).

[0092] The edge node analyzes network-wide data and detects a temporary cluster of obstacles in a certain area due to construction. It then issues a "redirection zone marker" to all drones. Drones respond in real time based on a lightweight model. Drone A adjusts its heading based on the edge command and uses a PID controller to smoothly adjust the rudder deflection angle (for example, from 15° to 10°) to ensure stable obstacle avoidance. Drone B, with its battery level below 20%, triggers a dynamic priority increase, prioritizing straight flight.

[0093] Step S3: Build a 5G network architecture to enable real-time data interaction between drones, specifically:

[0094] Step S3-1: Each drone is equipped with a 5G NR-U communication unit for data transmission;

[0095] The data transmission content includes: real-time flight status, path planning proposals, obstacle point cloud from millimeter-wave radar / visual perception, weather warning information and priority, and mission metadata of remaining battery power;

[0096] Step S3-2: Multiple drones share the same time-frequency resource block through sparse code multiple access (SCMA) technology, specifically:

[0097] Define the codebook matrix Where K is the number of users, N is the number of resource blocks, and the sparsity factor is d f ,satisfy:

[0098] ||C k ||0=d f ,

[0099] Step S3-3: Based on the real-time channel state information CSI, the high-priority UAV seizes the high signal-to-noise ratio subcarrier and adopts the message passing algorithm MPA demodulation:

[0100]

[0101] Among them, x k is the symbol bit variable of the kth UAV, p(y n |x) is the received signal y n Likelihood probability density function under known symbol bit x; μ j is the message value sent by adjacent node j to node k in the message passing algorithm MPA; x j is the symbol bit variable of neighbor node j;

[0102] Priority P i The rule for selecting subcarrier index m for the UAV is:

[0103]

[0104] Among them, h i,m is the channel response, σ 2 is the noise power.

[0105] Step S4: Build a dynamic negotiation protocol to support multiple machines to quickly exchange obstacle avoidance proposals when collision risk is triggered. Specifically:

[0106] The dynamic negotiation protocol includes conflict detection, proposal interaction, arbitration mechanism, and priority strategy. Conflict detection: Based on the trajectory prediction model, the collision risk index between the two aircraft is defined;

[0107] The trajectory prediction model is based on the uniform acceleration motion model and Kalman filter prediction, fully considering the dynamic changes of the UAV;

[0108] The input parameters include the current position coordinates of the drone (x t ,y t ,z t ), the current velocity vector of the UAV (v x ,v y ,v z ), the current moment of the UAV's acceleration vector (a x ,a y ,a z ) and the prediction step size τ, assuming that the acceleration remains unchanged within the short time prediction window, the future position prediction is given by the following formula:

[0109]

[0110] Where τ is the predicted collision time offset;

[0111] A one-dimensional Kalman filter is used to recursively update the position, velocity and acceleration, and output the trajectory sequence {(x t+τ ,y t+τ ,z t+τ )}, providing input for collision risk assessment and obstacle avoidance proposals;

[0112] Proposal interaction: The initiator sends an obstacle avoidance proposal, and the recipient responds with an acceptance or counter-proposal within a specified time;

[0113] Arbitration mechanism: If there is a conflict in proposals, the edge node calculates the global optimal solution based on the Nash equilibrium algorithm, giving priority to protecting the high-priority drone path;

[0114] The Nash equilibrium algorithm is specifically:

[0115]

[0116] in, is the optimal strategy for the i-th UAV, s i is any feasible strategy for the i-th UAV, u i is the utility function of the i-th drone, which is used to evaluate the benefits under different strategies;

[0117] Priority strategy: includes static priority and dynamic weight allocation. Static priority is pre-defined based on task urgency, while dynamic weight is dynamically adjusted based on remaining power and environmental risks.

[0118] Supports multiple drones to quickly exchange obstacle avoidance proposals when collision risk is triggered. By calculating the relative motion relationship between drones, the future collision time is predicted, and the collision risk is proactively predicted. Specifically:

[0119] Predict collision time t based on relative position Δp and velocity Δv c , the collision risk value is calculated as:

[0120]

[0121] where e is a natural constant; λ is the risk slope factor; and τ is the predicted collision time offset, representing the threshold center time, which is used to balance the model's response to early and late collision times.

[0122] In step S5, a distributed collaborative obstacle avoidance algorithm is constructed to generate a smooth obstacle avoidance path in real time. Specifically:

[0123] Step S5-1: The drone uploads its perceived obstacle data in real time. The edge node generates a global airspace heat map and distributes it to all drones via the 5G network. The node feature update formula is:

[0124]

[0125] Among them, α ij is the attention weight, σ is the ReLU activation function; W (l) is the weight matrix of the lth layer, is the feature vector of node i in layer l, is the set of neighbor nodes;

[0126] Step S5-2: Each machine generates a local obstacle avoidance strategy based on the received global data through the graph neural network (GNN). The objective function is decomposed into local sub-problems:

[0127]

[0128] Among them, x i is the local obstacle avoidance strategy variable of the i-th UAV; ρ is the penalty factor, u i is the Lagrange multiplier corresponding to the i-th UAV; f i is the local objective function of the i-th UAV; z is the global consistency variable;

[0129] Step S5-3: Use the alternating direction multiplier method ADMM distributed optimization framework to iteratively reach global consensus.

[0130] The alternating direction multiplier method ADMM outputs a composite adjustment instruction of heading, speed, and altitude. The expression of the alternating direction multiplier method ADMM is:

[0131]

[0132] Among them, ||r k ||2 is the original residual, ||s k ||2 is the dual residual; ρ k is the penalty factor for the kth iteration;

[0133] When the original residual || r k ||2 is significantly larger than the dual residual ||s k ||2, indicating that the current solution deviates greatly from the original constraints and needs to be tightened to reduce the accelerated convergence; the dual residual ||s k ||2 is significantly larger than the original residual ||r k ||2, indicating that the dual variable fluctuates greatly, and the penalty factor needs to be reduced to stabilize the optimization process; if the residuals of the two are relatively balanced, maintain the current ρ k value;

[0134] By dynamically adjusting ρ, the ADMM algorithm adaptively balances the feasibility of the original problem and the convergence of the dual variable, avoiding the tedious manual adjustment of parameters and improving the solution efficiency of multi-UAV path optimization.

[0135] like Figure 2 As shown, a multi-UAV negotiation and collision avoidance system based on 5G UAV communication is shown. The system includes a 5G communication module, a flight status receiving module, a collision risk assessment module, a negotiation module, a flight trajectory adjustment module, and a fusion interface module. The 5G communication module includes a 5G NR-U communication unit. Each UAV is equipped with a 5G NR-U communication unit, which supports sparse code multiple access technology, allowing multiple UAVs to share the same time-frequency resource block and broadcast flight status of timing position, speed, heading, altitude, and mission priority.

[0136] The 5G communication module also includes a dynamic priority slot allocation algorithm, which allows high-priority drones to occupy low-interference channels first, ensuring millisecond-level transmission of critical emergency avoidance instructions;

[0137] Flight status receiving module: This module eliminates transmission delay and noise through the extended Kalman filter (EKF), constructs an airspace situation map in real time, and provides high-precision input for collision risk assessment.

[0138] Collision Risk Assessment Module: This module uses a trajectory prediction algorithm to calculate the probability of collision within a specified timeframe based on a dynamic model of relative speed and distance. This is then mapped to a risk value using a Sigmoid function. The trigger threshold is then set to balance false alarm rate with response speed.

[0139] Negotiation module: Integrates a distributed game solver to initiate multi-machine real-time negotiation via the 5G network when risks are detected;

[0140] Flight trajectory adjustment module: Generates smooth adjustment instructions based on the PID control algorithm, dynamically corrects the flight path, and supports compound adjustment of heading, speed, and altitude to ensure stable execution of obstacle avoidance actions and avoid the risk of loss of control caused by violent maneuvers;

[0141] Fusion interface module: used to bridge the gap between traditional sensing equipment and new 5G communication technology, achieving seamless integration of multi-level obstacle avoidance capabilities.

[0142] In order to better illustrate the advantages of the technical solution of the present invention, the following simulation test is disclosed:

[0143] In the simulation test, 10 logistics drones were simulated performing delivery tasks in a 1km×1km virtual airspace. Three cross-route points were preset and five dynamic obstacles (simulating flying birds) were randomly injected. Data collection included drone status (position, speed, and priority were recorded every 0.1 seconds), communication logs (recording the delay and packet loss of each signaling interaction), and obstacle avoidance actions (recording the heading adjustment angle, speed change, and execution timestamp).

[0144] Key steps description:

[0145] Data collection: Update the drone's position every 0.1 seconds to simulate sensor data;

[0146] Communication broadcast: broadcast status via 5G network, simulating a 5% packet loss rate;

[0147] Collision detection: trigger risk when the distance is less than 50 meters, and record the number of collisions;

[0148] Negotiation and execution: High-priority proposals are executed first, and end-to-end delays are recorded;

[0149] Energy consumption statistics: Calculate additional energy consumption based on the square of speed change.

[0150] The experimental simulation diagram simulates the delivery mission of 10 drones in a 1km×1km virtual airspace, visualizing the starting and final positions of the drones, flight speed and direction, and obstacle distribution. The “Drone Flight Path and Obstacle Distribution” diagram visualizes 10 drones in a complex scenario with randomly injected dynamic obstacles and multiple drones crossing routes, such as Figure 3 As shown, the final positions of the drones are evenly distributed and maintain a safe distance from each other, with no clustering or dense intersections. This demonstrates that multiple drones can achieve a good spatial distribution through dynamic obstacle avoidance negotiation in a high-density environment, avoiding potential collision risks. Furthermore, in areas with dense obstacles, the flight paths do not bend sharply, with no "wraparounds" or excessive detours, and the transitions are smooth. This demonstrates that the distributed negotiation and dynamic optimization algorithm can achieve efficient flight path planning while ensuring obstacle avoidance safety, and demonstrates the system's path optimization and dynamic smooth adjustment capabilities in a multi-obstacle environment. The velocity vector directions in the "UAV Final Position and Velocity Direction" plot are mostly diverse, indicating that each drone maintains autonomy and path diversity at the end, avoiding path blockage or flight bottlenecks caused by repeated paths. Furthermore, the similar lengths of the velocity arrows indicate that the overall flight speed of the drone group is consistent, meeting the goals of efficient collaboration and smooth obstacle avoidance.

[0151] The simulation results are shown in Table 1:

[0152] Table 1

[0153] Number of collisions Average number of obstacle avoidances Obstacle avoidance success rate Maximum communication delay Average energy consumption 23 2.3 100％ 48.6ms 40

[0154] like Figure 4As shown, 10 drones triggered a total of 23 collision risk detections in a complex scenario. Through a real-time negotiation mechanism and dynamic priority arbitration among multiple drones, all collision risks were effectively resolved, achieving a 100% obstacle avoidance success rate. This demonstrates that the system can ensure efficient and robust obstacle avoidance among multiple drones in high-density, complex and dynamic environments. The maximum communication delay statistical value was 48.6ms, demonstrating that the 5G network can maintain low latency in complex environments. The average energy consumption statistical value shows that during multiple dynamic negotiations and track corrections, the system can balance obstacle avoidance safety and energy consumption control, achieving good overall flight efficiency.

[0155] The foregoing is merely an embodiment of the present invention and is not intended to limit the present invention. It will be apparent to those skilled in the art that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are intended to be included within the scope of the claims of the present invention.

Claims

1. A multi-UAV negotiation and collision avoidance method based on 5G UAV communication, characterized by: The method comprises the following steps: Step S1: Collect multi-source flight status data and capture the surrounding building outlines through binocular vision to generate a depth map; Step S2: Multi-source flight status data is fused using Kalman filtering to construct an airspace situation map; The airspace situation map contains the location, speed, and obstacle information of the UAV; Step S3: Build a 5G network architecture to enable real-time data interaction between drones; Step S4: Build a dynamic negotiation protocol to support multiple machines to quickly exchange obstacle avoidance proposals when collision risk is triggered; Step S5: Build a distributed collaborative obstacle avoidance algorithm to generate a smooth obstacle avoidance path in real time.

2. The multi-UAV negotiation and collision avoidance method based on 5G UAV communication according to claim 1 is characterized in that: In step S3, a 5G network architecture is constructed to realize real-time data interaction between drones, specifically: Step S3-1: Each drone is equipped with a 5G NR-U communication unit for data transmission; The data transmission content includes: real-time flight status, path planning proposals, obstacle point cloud from millimeter-wave radar / visual perception, weather warning information and priority, and mission metadata of remaining battery power; Step S3-2: Multiple drones share the same time-frequency resource block through sparse code multiple access (SCMA) technology, specifically: Define the codebook matrix Where K is the number of users, N is the number of resource blocks, and the sparsity factor is d f ,satisfy: Step S3-3: Based on the real-time channel state information CSI, the high-priority UAV seizes the high signal-to-noise ratio subcarrier and adopts the message passing algorithm MPA demodulation: Among them, x k is the symbol bit variable of the kth UAV, p(y n |x) is the received signal y n Likelihood probability density function under known symbol bit x; μ j is the message value sent by adjacent node j to node k in the message passing algorithm MPA; x j is the symbol bit variable of neighbor node j; Priority P i The rule for selecting subcarrier index m for the UAV is: Among them, h i,m is the channel response, σ 2 is the noise power.

3. The multi-UAV negotiation and collision avoidance method based on 5G UAV communication according to claim 1 is characterized in that: The dynamic negotiation protocol includes conflict detection, proposal interaction, arbitration mechanism, and priority strategy, specifically: Conflict detection: Based on the trajectory prediction model, define the collision risk index between the two aircraft; The trajectory prediction model is based on the uniform acceleration motion model and Kalman filter prediction, fully considering the dynamic changes of the UAV; The input parameters include the current position coordinates of the drone (x t ,y t ,z t ), the current velocity vector of the UAV (v x ,v y ,v z ), the current moment of the UAV's acceleration vector (a x ,a y ,a z ) and the prediction step size τ, assuming that the acceleration remains unchanged within the short time prediction window, the future position prediction is given by the following formula: Where τ is the predicted collision time offset; A one-dimensional Kalman filter is used to recursively update the position, velocity and acceleration, and output the trajectory sequence {(x t+τ ,y t+τ ,z t+τ )}, providing input for collision risk assessment and obstacle avoidance proposals; Proposal interaction: The initiator sends an obstacle avoidance proposal, and the recipient responds with an acceptance or counter-proposal within a specified time; Arbitration mechanism: If there is a conflict in proposals, the edge node calculates the global optimal solution based on the Nash equilibrium algorithm, giving priority to protecting the high-priority drone path; Priority strategy: includes static priority and dynamic weight allocation. Static priority is pre-defined based on task urgency, while dynamic weight is dynamically adjusted based on remaining power and environmental risks.

4. The multi-UAV negotiation and collision avoidance method based on 5G UAV communication according to claim 3 is characterized in that: The Nash equilibrium algorithm is specifically: in, is the optimal strategy for the i-th UAV, s i is any feasible strategy for the i-th UAV, u i is the utility function of the i-th drone, which is used to evaluate the benefits under different strategies.

5. The multi-UAV negotiation and collision avoidance method based on 5G UAV communication according to claim 1 is characterized in that: Step S4 supports the rapid exchange of obstacle avoidance proposals among multiple drones when a collision risk is triggered. By calculating the relative motion relationship between drones and predicting the future collision time, active prediction of collision risk is achieved. Specifically, Predict collision time t based on relative position Δp and velocity Δv c , the collision risk value is calculated as: where e is a natural constant; λ is the risk slope factor; and τ is the predicted collision time offset, representing the threshold center time, which is used to balance the model's response to early and late collision times.

6. The multi-UAV negotiation and collision avoidance method based on 5G UAV communication according to claim 1, characterized in that: In step S5, a distributed collaborative obstacle avoidance algorithm is constructed to generate a smooth obstacle avoidance path in real time, specifically: Step S5-1: The drone uploads its perceived obstacle data in real time. The edge node generates a global airspace heat map and distributes it to all drones via the 5G network. The node feature update formula is: Among them, α ij is the attention weight, σ is the ReLU activation function; W (l) is the weight matrix of the lth layer, is the feature vector of node i in layer l, is the set of neighbor nodes; Step S5-2: Each machine generates a local obstacle avoidance strategy based on the received global data through the graph neural network (GNN). The objective function is decomposed into local sub-problems: Among them, x i is the local obstacle avoidance strategy variable of the i-th UAV; ρ is the penalty factor, u i is the Lagrange multiplier corresponding to the i-th UAV; f i is the local objective function of the i-th UAV; z is the global consistency variable; Step S5-3: Use the alternating direction multiplier method ADMM distributed optimization framework to iteratively reach global consensus.

7. The multi-UAV negotiation and collision avoidance method based on 5G UAV communication according to claim 6, characterized in that: The alternating direction multiplier method ADMM outputs a composite adjustment instruction of heading, speed, and altitude. The expression of the alternating direction multiplier method ADMM is: Among them, ||r k ||2 is the original residual, ||s k ||2 is the dual residual; ρ k is the penalty factor for the kth iteration; When the original residual || r k ||2 is significantly larger than the dual residual ||s k ||2, indicating that the current solution deviates greatly from the original constraints and needs to be tightened to reduce the accelerated convergence; the dual residual ||s k ||2 is significantly larger than the original residual ||r k ||2, indicating that the dual variable fluctuates greatly, and the penalty factor needs to be reduced to stabilize the optimization process; if the residuals of the two are relatively balanced, maintain the current ρ k value; By dynamically adjusting ρ, the ADMM algorithm adaptively balances the feasibility of the original problem and the convergence of the dual variable, avoiding the tedious manual adjustment of parameters and improving the solution efficiency of multi-UAV path optimization.

8. A multi-UAV negotiation and collision avoidance system based on 5G UAV communication, characterized by: A multi-UAV negotiation and collision avoidance method based on 5G UAV communication comprising any one of claims 1 to 7; The system includes a 5G communication module, a flight status receiving module, a collision risk assessment module, a negotiation module, a flight trajectory adjustment module, and a fusion interface module. The 5G communication module includes a 5G NR-U communication unit. Each drone is equipped with a 5G NR-U communication unit, which supports sparse code multiple access technology, allowing multiple drones to share the same time-frequency resource block and broadcast flight status including timing position, speed, heading, altitude, and mission priority. The 5G communication module also includes a dynamic priority slot allocation algorithm, which allows high-priority drones to occupy low-interference channels first, ensuring millisecond-level transmission of critical emergency avoidance instructions; Flight status receiving module: This module eliminates transmission delay and noise through the extended Kalman filter (EKF), constructs an airspace situation map in real time, and provides high-precision input for collision risk assessment. Collision Risk Assessment Module: This module uses a trajectory prediction algorithm to calculate the probability of collision within a specified timeframe based on a dynamic model of relative speed and distance. This is then mapped to a risk value using a Sigmoid function. The trigger threshold is then set to balance false alarm rate with response speed. Negotiation module: Integrates a distributed game solver to initiate multi-machine real-time negotiation via the 5G network when risks are detected; Flight trajectory adjustment module: Generates smooth adjustment instructions based on the PID control algorithm, dynamically corrects the flight path, and supports compound adjustment of heading, speed, and altitude to ensure stable execution of obstacle avoidance actions and avoid the risk of loss of control caused by violent maneuvers; Fusion interface module: used to bridge traditional sensing equipment with new 5G communication technology, achieving seamless connection of multi-level obstacle avoidance capabilities.

Images