A multi-agent evolution-based complex game field generation and equilibrium analysis method and system

Abstract

This invention provides a method and system for generating and analyzing complex game fields based on multi-agent evolution. It acquires environmental image data and flight status data of each drone in a pre-set drone swarm; constructs a dense environmental map based on the environmental image data and flight status data; calculates the relative distance and angle changes between drones to construct a high-precision relative pose state network; and applies the dense environmental map and the high-precision relative pose state network to map the positional relationships between pre-set dynamic obstacles, preset task objectives, and pre-set friendly drone swarms into potential energy field payoff functions to adjust the flight paths and task allocation strategies of each drone, generating a cooperative evolution game field to achieve equilibrium state analysis of the pre-set drone swarm. This invention enables the autonomous generation and equilibrium state analysis of cooperative game fields in unknown dynamic environments for drone swarms, improving the collaborative decision-making capability and task execution efficiency of swarm systems in complex environments.

Description

Technical Field

[0001] This invention relates to the field of game field generation and equilibrium analysis technology, and in particular to a method and system for generating and analyzing complex game fields based on multi-agent evolution. Background Technology

[0002] With the increasing demand for collaborative mission execution by UAV swarms in unknown dynamic environments, swarm systems must possess the ability to perceive environmental changes in real time and dynamically adjust formation structure and task allocation strategies. This requires UAV swarms not only to autonomously construct high-precision environmental spatial cognition models but also to achieve collaborative decision-making evolution among multiple agents based on local interactions and observation data, even in the absence of global prior information. This ensures the maintenance of overall system stability and mission efficiency under uncertain conditions such as dynamic obstacle interference, communication limitations, and mission target drift. Therefore, a comprehensive analytical framework that integrates environmental perception, relative pose deduction, dynamic role function recognition, and collaborative strategy optimization is urgently needed to support the self-organized operation and equilibrium state achievement of UAV swarms in highly uncertain environments.

[0003] The existing solution is a multi-agent collaborative decision-making method based on distributed reinforcement learning and local potential field coupling. Each UAV node independently runs a deep network to learn local policies, and combines a pre-set repulsive and gravitational potential field model to guide obstacle avoidance and approach behavior to nearby obstacles and target points. Each node exchanges action prediction and observation status within the communication range, and uses a consensus algorithm to align the local policy update direction. An experience playback mechanism is introduced to enhance the adaptability to dynamic changes in the environment. Existing solutions have some inherent flaws, including the fact that static modeling mechanisms relying on local potential fields cannot accurately reflect the spatiotemporal game relationship between dynamic obstacles and moving task objectives, resulting in potential field benefit assessment lagging behind environmental changes and causing path oscillations or task conflicts; independent learning strategies of each node lead to unclear role differentiation, which can easily result in strategy homogenization and decision redundancy in resource competition or task division scenarios, and cannot form differentiated behavior patterns based on differences in equipment capabilities and interaction history, thus affecting the overall evolutionary balance and task execution efficiency of the cluster; the use of relative spatial relationships is limited to simple distance threshold judgment, lacking a fine characterization of the evolution of geometric configurations among multiple machines, resulting in lagging formation structure adjustments and difficulty in supporting stable collaboration in highly dynamic environments. Summary of the Invention

[0004] This invention provides a method and system for generating and analyzing complex game fields based on multi-agent evolution. It addresses several issues in existing technologies, such as the inaccuracy of static modeling mechanisms relying on local potential fields in reflecting the spatiotemporal game relationship between dynamic obstacles and moving task objectives. This leads to potential field payoff assessment lagging behind environmental changes, causing path oscillations or task conflicts. Furthermore, the independent learning strategies of each node result in unclear role differentiation, leading to strategy homogenization and decision redundancy in resource competition or task division scenarios. This prevents the formation of differentiated behavior patterns based on device capabilities and interaction history, thus affecting the overall evolutionary equilibrium and task execution efficiency of the cluster. Finally, the utilization of relative spatial relationships is limited to simple distance threshold judgments, lacking a detailed characterization of the geometric configuration evolution among multiple machines, resulting in delayed formation structure adjustments and difficulty in supporting stable collaboration in highly dynamic environments.

[0005] In a first aspect, the present invention provides a method for generating and analyzing complex game fields based on multi-agent evolution, comprising: Acquire environmental image data and flight status data of each drone in a pre-set drone cluster; The environmental image data and flight status data are processed using a preset synchronous positioning and mapping technology to generate a dense environmental map. Spatial geometric relationship calculation is performed on the relative distances between the drones and the angular changes between the drones to construct a high-precision relative pose state network for the pre-set drone cluster. Based on the device performance parameters of each UAV and the environmental interaction history data of each UAV, the role evolution processing is driven to generate the role probability distribution data of each UAV. By applying the dense environmental map and the high-precision relative pose state network, the positional relationship between preset dynamic obstacles, preset mission objectives and preset friendly drone clusters is mapped into a potential energy field gain function. Based on the role probability distribution data and the potential energy field revenue function, the flight paths and task allocation strategies of each UAV are adjusted to generate a cooperative evolution game field of the pre-set UAV cluster in an unknown dynamic environment, so as to realize the equilibrium state analysis of the pre-set UAV cluster.

[0006] Optionally, environmental image data and flight status data of each drone in a pre-set drone cluster are acquired, including: The system utilizes the visual sensors carried by each drone in a pre-set drone cluster to acquire a pre-defined sequence of original images of the surrounding environment. Real-time flight attitude data of each UAV is collected using the inertial measurement unit carried by each UAV. Each image frame in the original image sequence and each measurement record in the real-time flight attitude data are respectively appended with a timestamp synchronized with each image frame and each measurement record to obtain multiple time-synchronized sensor data. The time-synchronized sensor data is packaged to generate an environmental status data package; The environmental status data packet is transmitted to the relay node of the preset drone cluster using a preset wireless transmission module to perform time synchronization verification on the environmental status data packet, and the verification results are integrated to generate a cluster environmental status dataset. The cluster environment status dataset is formatted to obtain environmental image data and flight status data sets from the processing results.

[0007] Optionally, the environmental image data and the flight status data are processed using a preset simultaneous localization and mapping (SLAM) technology to generate a dense environmental map, including: Multiple visual feature points are extracted from the environmental image data, and each visual feature point is associated with the location information corresponding to the acquisition time of each visual feature point in the flight status data to generate a target visual feature point set. Similarity matching is performed on each target visual feature point in the target visual feature point set to construct a feature point trajectory sequence; Based on the feature point trajectory sequence and the position information corresponding to the feature point trajectory sequence, calculate the pose change between adjacent acquisition times; The pose change is used to calculate the three-dimensional spatial coordinates of each of the target visual feature points to generate a sparse environment point cloud. The surface of the sparse environmental point cloud is reconstructed to generate the target environmental point cloud; The surface of the target environment point cloud and the geometric model of the pre-set dynamic obstacles are fused to generate a dense environment map.

[0008] Optionally, spatial geometric relationship calculations are performed on the relative distances between the drones and the angular changes between the drones to construct a high-precision relative pose state network for the pre-set drone cluster, including: The raw distance data and raw angle data between each UAV are measured using a pre-set sparse array microwave interferometric ranging system. Calculate the distance phase difference between each set of raw distance data adjacent to the measurement cycle and the angular phase difference between each set of raw angle data adjacent to the measurement cycle; Calculate the relative motion vector between each UAV based on the distance phase difference and the angle phase difference; The estimated relative pose of each of the UAVs is updated by applying the relative motion vectors to generate a dynamic relative pose table; The topological connection relationship of the pre-set UAV cluster is constructed based on the dynamic relative pose table; The topological connection relationship and the dynamic relative pose table are fused to generate a high-precision relative pose state network.

[0009] Optionally, role evolution processing is driven based on the device performance parameters of each UAV and the environmental interaction history data of each UAV to generate role probability distribution data for each UAV, including: The device performance parameters of each UAV are subjected to role capability quantization encoding to generate an initial role capability vector. The initial character capability vector and the dynamic character performance records in the environmental interaction history data of each UAV are fused together to generate a comprehensive character evaluation index. Calculate the fitness weight of each UAV in different role tasks based on the comprehensive role evaluation index; Based on the fitness weights described above, different preset strategy exploration factors are input into the preset current role strategy set to generate role strategy mutation schemes; The role selection preferences of each UAV are updated by applying the aforementioned role strategy mutation scheme, generating multiple updated role selection preferences; Based on the updated role selection preferences, the values ​​of each role selection preference are probabilistically normalized to generate role probability distribution data for each drone.

[0010] Optionally, by applying the dense environmental map and the high-precision relative pose state network, the positional relationships between preset dynamic obstacles, preset mission objectives, and preset friendly drone clusters are mapped into a potential energy field gain function, including: The contour information of preset dynamic obstacles and the real-time position coordinates of preset friendly drone clusters are extracted from the dense environmental map and the high-precision relative pose state network, respectively. The preset task target coordinate information is parsed to generate a spatial location description of the task target; Calculate the minimum distance vector between the contour information and the real-time position coordinates, as well as the relative azimuth vectors between each friendly drone in the preset friendly drone cluster; Spatial relationship encoding is performed on the minimum distance vector and the relative orientation vector to generate a relative spatial relationship description; The relative spatial relationship is described in terms of potential energy field parameters. The potential energy field parameters and the spatial location description of the mission target are combined to generate a potential energy field benefit function.

[0011] Optionally, the flight paths and task allocation strategies of each UAV are adjusted based on the role probability distribution data and the potential energy field reward function to generate a cooperative evolution game field of the pre-set UAV cluster in an unknown dynamic environment, thereby achieving equilibrium state analysis of the pre-set UAV cluster, including: The probability distribution data of the roles and the potential energy field benefit function are transformed and processed respectively to obtain the initial task allocation strategy and path planning benefit value of each of the drones. The initial task allocation strategy and the path planning benefit value are used to construct multi-objective optimization constraints; Based on the multi-objective optimization constraints, the initial task allocation strategy is executed to detect policy conflict, so as to identify the policy execution conflict characteristics between the UAVs. The conflict characteristics of each strategy are subjected to collaborative strategy optimization processing to generate a set of conflict resolution strategies; The conflict resolution strategy set is applied to update the flight path and task allocation strategy of each UAV, generating updated flight paths and updated task allocation strategies. The overall distribution state between the updated flight path and the updated task allocation strategy is evaluated for equilibrium convergence to generate a cooperative evolution game field and equilibrium state analysis results.

[0012] Secondly, the present invention provides a system for generating and analyzing complex game fields based on multi-agent evolution, comprising: The acquisition module is used to acquire environmental image data and flight status data of each drone in the pre-set drone cluster; The processing module is used to process the environmental image data and the flight status data using a preset synchronous positioning and mapping technology to generate a dense environmental map. The calculation module is used to perform spatial geometric relationship calculation on the relative distance between each UAV and the angular change between each UAV, so as to construct a high-precision relative pose state network of the pre-set UAV cluster. An evolution module is used to drive role evolution processing based on the equipment performance parameters of each UAV and the environmental interaction history data of each UAV, and generate role probability distribution data of each UAV. The mapping module is used to apply the dense environmental map and the high-precision relative pose state network to map the positional relationship between preset dynamic obstacles, preset task objectives and preset friendly drone clusters into a potential energy field gain function. The adjustment module is used to adjust the flight path and task allocation strategy of each UAV according to the role probability distribution data and the potential energy field revenue function, and generate a cooperative evolution game field of the preset UAV cluster in an unknown dynamic environment to realize the equilibrium state analysis of the preset UAV cluster.

[0013] Thirdly, the present invention provides a computing device, including a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a method for generating and analyzing complex game fields based on multi-agent evolution as described in any of the first aspects.

[0014] Fourthly, the present invention provides a computer storage medium storing computer program instructions thereon, wherein the computer program instructions, when executed by a processor, implement the method for generating and analyzing complex game fields based on multi-agent evolution as described in any one of the first aspects.

[0015] This invention generates a dense environmental map by comprehensively utilizing environmental images and flight status data, constructs a high-precision pose state network by combining geometric calculations of relative measurements among multiple UAVs, and drives role probability evolution based on device performance and historical interaction data. Finally, through potential energy field mapping and multi-objective strategy adjustment, it realizes the autonomous generation and equilibrium state analysis of the collaborative game field of UAV swarm in unknown dynamic environments, significantly improving the collaborative decision-making ability and task execution efficiency of swarm systems in complex environments.

[0016] Furthermore, by using a sparse array microwave interferometric ranging system to obtain raw distance and angle data, and calculating the relative motion vector through phase difference calculation, a high-precision relative pose state network for UAV swarms is constructed by dynamically updating the pose table and integrating it with the construction of topological connection relationships. This provides an accurate and reliable relative pose foundation for subsequent game field generation, effectively improving the swarm collaborative positioning accuracy and system stability.

[0017] These or other aspects of the invention will become more apparent from the following description of the embodiments. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1A flowchart of a method for generating and analyzing complex game fields based on multi-agent evolution provided in an embodiment of the present invention; Figure 2 A schematic diagram of the structure of a complex game field generation and equilibrium analysis system based on multi-agent evolution provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a computing device provided in an embodiment of the present invention. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0021] In some of the processes described in the specification, claims, and accompanying drawings of this invention, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Figure 1 A flowchart of a method for generating and analyzing complex game fields based on multi-agent evolution is provided in this embodiment of the invention, as shown below. Figure 1 As shown, the method includes: Existing UAV swarm collaboration technologies based on fixed rules or centralized control suffer from three key drawbacks: first, they cannot build reliable environmental models in real time by fusing multi-source heterogeneous perception data in unknown dynamic environments; second, the lack of high-precision relative pose measurement methods leads to the accumulation of swarm collaborative positioning errors; and third, static role allocation mechanisms are ill-suited to adapting to sudden obstacles and dynamic changes in mission objectives. To address these issues, this invention proposes the following approach: generating a dense environmental map by processing visual and inertial data using synchronous positioning and mapping technology; constructing a relative pose state network by calculating precise geometric relationships between UAVs using the principle of microwave interferometry; driving the dynamic evolution of role probabilities by combining equipment performance and historical interaction data; and finally generating an equilibrium strategy for obstacle avoidance and target collaboration through potential energy field mapping, forming a multi-agent autonomous collaborative solution with a closed-loop perception-decision-execution process. Based on this, this invention provides a method for generating and analyzing complex game fields based on multi-agent evolution, such as… Figure 1 ,include: Step 101: Acquire environmental image data and flight status data of each drone in the pre-set drone cluster.

[0024] In this step, environmental image data refers to the original image sequence of the surrounding environment captured by the visual sensors on the UAV; flight status data refers to the real-time flight parameters of the UAV recorded by the inertial measurement unit.

[0025] In this embodiment of the invention, firstly, the visual sensors on each UAV capture the original image sequence of the surrounding environment to form environmental image data, and at the same time, the inertial measurement units on each UAV record the flight attitude, speed and position information to form flight status data. Secondly, the two types of data are packaged and transmitted to the cluster relay node after being stamped with a synchronization timestamp. Then, the data packets are time-synchronized and format-unified. Finally, after eliminating transmission redundancy, a structured set of environmental image data and flight status data is output.

[0026] Step 102: Process the environmental image data and the flight status data using a preset synchronous positioning and mapping technology to generate a dense environmental map.

[0027] In this step, the preset synchronous positioning and mapping technology refers to the algorithm framework that is set in advance to achieve self-localization and environmental modeling simultaneously; the dense environment map refers to the environment model containing continuous geometric information of the object surface obtained through 3D reconstruction.

[0028] In this embodiment of the invention, firstly, visual feature points are extracted from environmental image data using simultaneous localization and mapping (SLAM) technology and associated with the position information in the flight status data at the corresponding time. Secondly, similarity matching is performed on feature points at consecutive time moments to establish a feature point trajectory sequence. Then, the pose change between adjacent time moments is calculated based on the trajectory sequence and position information. Next, the sparse environmental point cloud is generated by calculating the three-dimensional spatial coordinates of the feature points based on the pose change. Finally, a dense environmental map containing the outline of obstacles is generated by reconstructing the surface of the point cloud and fusing it with a preset obstacle model.

[0029] Step 103: Perform spatial geometric relationship calculation on the relative distance between each UAV and the angular change between each UAV to construct a high-precision relative pose state network of the pre-set UAV cluster.

[0030] In this step, relative distance refers to the straight-line interval length between UAVs in three-dimensional space; angular change refers to the difference in relative orientation between UAVs within a continuous measurement period; spatial geometric relationship calculation refers to the mathematical process of calculating spatial relationships based on measurement data; and high-precision relative pose state network refers to the topology that describes the relative position and attitude relationships between all UAVs in the cluster.

[0031] In this embodiment of the invention, the original distance and angle data between UAVs are first measured using a sparse array microwave interferometric ranging system. Then, the distance phase difference and angle phase difference between adjacent measurement periods are calculated. Subsequently, the relative motion vector between each UAV is calculated based on the phase difference. Then, the estimated relative pose is updated based on the relative motion vector to generate a dynamic relative pose table. Then, the topological connection relationship of the UAV cluster is constructed based on the pose table. Finally, the topological relationship and the pose table are fused to generate a high-precision relative pose state network.

[0032] Step 104: Drive role evolution processing based on the device performance parameters of each UAV and the environmental interaction history data of each UAV to generate role probability distribution data for each UAV.

[0033] In this step, equipment performance parameters refer to the hardware performance indicators of the UAV; environmental interaction history data refers to the behavioral records accumulated by the UAV in past missions; role evolution processing refers to the dynamic process of adjusting role allocation based on capabilities and historical performance; and role probability distribution data refers to the probability set of each UAV undertaking different tasks.

[0034] In this embodiment of the invention, firstly, the device performance parameters of each UAV are quantified and encoded to generate an initial role capability vector. At the same time, dynamic role performance records are extracted from historical environmental interaction data. Secondly, the capability vector and performance records are weighted and fused to generate a comprehensive role evaluation index. Then, the fitness weight of each UAV on different role tasks is calculated based on the evaluation index. Next, the strategy exploration factor is introduced into the current role strategy set to generate a role strategy mutation scheme. Then, the mutation scheme is applied to iteratively update the role selection preference. Finally, the preference values ​​are probability normalized to generate role probability distribution data.

[0035] Step 105: Using the dense environmental map and the high-precision relative pose state network, the positional relationship between the preset dynamic obstacles, the preset mission objectives, and the preset friendly drone cluster is mapped into a potential energy field gain function.

[0036] In this step, the preset mission objective refers to the pre-defined mission objective information; the positional relationship refers to the spatial relative relationship between the UAV, obstacles, and the target; the mapping operation refers to the process of converting the spatial relationship into a mathematical model; and the potential energy field benefit function refers to the decision evaluation function generated based on the positional relationship.

[0037] In this embodiment of the invention, the contour information of preset dynamic obstacles is first extracted from a dense environmental map, and the real-time position coordinates of preset friendly UAVs are obtained from a high-precision relative pose state network. Then, the preset task target coordinate information is parsed to generate a spatial position description of the task target. Subsequently, the minimum distance vector between the obstacle contour and each UAV and the relative azimuth vector between friendly UAVs are calculated. Then, a relative spatial relationship description is generated based on the distance vector and azimuth vector. Then, potential energy field action parameters are generated based on the spatial relationship description. Finally, the action parameters and the target position description are combined to generate a potential energy field benefit function.

[0038] Step 106: Adjust the flight path and task allocation strategy of each UAV according to the role probability distribution data and the potential energy field revenue function to generate a cooperative evolution game field of the pre-set UAV cluster in an unknown dynamic environment to achieve equilibrium state analysis of the pre-set UAV cluster.

[0039] In this step, the flight path refers to the planned flight trajectory of the UAV in space; the task allocation strategy refers to the plan for the UAV to undertake specific tasks; the adjustment operation refers to the process of optimizing the path and strategy according to the payoff function; the unknown dynamic environment refers to the changing environment that cannot be fully perceived in advance; the co-evolutionary game field refers to the decision space formed by multiple UAVs in the process of interaction; and the equilibrium state analysis operation refers to the process of evaluating the stability of the game field.

[0040] In this embodiment of the invention, firstly, the role probability distribution data is converted into the initial task allocation strategy for each UAV, and the potential energy field revenue function is quantified into path planning revenue value. Secondly, multi-objective optimization constraints are constructed based on the task allocation strategy and path revenue value. Subsequently, the initial task allocation strategy is executed according to the constraints, and strategy conflict detection is performed to identify the conflict characteristics of the strategy execution. Then, the conflict characteristics are optimized collaboratively to generate a conflict resolution strategy set. The strategy set is then applied to update the flight path and task allocation strategy of each UAV. Finally, the overall distribution state of the updated path and strategy is evaluated for equilibrium convergence to generate a collaborative evolution game field and equilibrium state analysis results.

[0041] For example, firstly, image data and flight status data of mountainous terrain are collected using drone-mounted cameras and inertial measurement units. Secondly, simultaneous localization and mapping (SLAM) technology is used to process the data and generate a dense terrain map containing rocks and vegetation. Then, the relative positions within the cluster are measured using a microwave interferometric ranging system, and a high-precision pose network is constructed. Next, the probability distribution of reconnaissance and patrol roles is calculated based on the drone's endurance and historical obstacle avoidance data. Then, based on the terrain map and pose network, the positions of moving vehicle targets and friendly drones are mapped to potential energy field payoff functions. Finally, the patrol paths and task assignments of each drone are adjusted according to the role probabilities and payoff functions, ultimately forming a collaborative tracking game field and outputting an equilibrium state analysis report.

[0042] This invention constructs an accurate environmental model through multi-source sensing data fusion, establishes a cluster collaboration foundation using a high-precision relative positioning network, achieves adaptive task allocation by combining a dynamic role evolution mechanism, and finally generates an equilibrium decision scheme through potential energy field game optimization, effectively improving the collaborative combat capability, autonomous decision-making efficiency, and mission adaptability of UAV clusters in unknown dynamic environments.

[0043] To address the issue of insufficient spatiotemporal synchronization accuracy of multi-source heterogeneous sensor data, this step achieves high-precision synchronous acquisition of environmental perception data through timestamp appending and data packet verification. This invention provides a specific embodiment: Step 101, acquiring environmental image data and flight status data of each drone in a pre-set drone cluster, specifically includes the following steps: Step 111: Use the visual sensors carried by each drone in the pre-set drone cluster to obtain a preset sequence of original images of the surrounding environment.

[0044] In this step, the preset surrounding environment refers to the specific area range that the drone needs to explore in advance; the original image sequence refers to the set of unprocessed image data continuously collected by the visual sensor.

[0045] In this embodiment of the invention, firstly, the visual sensors equipped on each drone in a pre-deployed drone cluster continuously acquire images of the preset surrounding environment to obtain an original image sequence arranged in chronological order. Secondly, each visual sensor is controlled to continuously capture environmental images according to a set sampling frequency to ensure that the image sequence contains complete spatial environment information.

[0046] Step 112: Collect real-time flight attitude data of each UAV using the inertial measurement unit carried by each UAV.

[0047] In this step, the inertial measurement unit refers to the combination of sensors used to measure the motion state of an object; real-time flight attitude data refers to the flight status information of the UAV collected in real time.

[0048] In this embodiment of the invention, the flight status is first monitored in real time by the inertial measurement unit carried by each UAV. Then, the three-dimensional acceleration, angular velocity and magnetometer data of the UAV are collected at a fixed sampling period. Subsequently, these measurement data are combined into real-time flight attitude data including attitude angle, motion speed and spatial position.

[0049] Step 113: Add a timestamp synchronized with each image frame in the original image sequence and each measurement record in the real-time flight attitude data to each image frame and each measurement record to obtain multiple time-synchronized sensor data.

[0050] In this step, an image frame refers to a single static image that makes up a video or image sequence; a measurement record refers to the set of data obtained by the inertial measurement unit each time it samples; a synchronized timestamp refers to a unified time stamp added to data from different sensors; and time-synchronized sensor data refers to a set of multi-source sensor data with the same time reference.

[0051] In this embodiment of the invention, firstly, the acquisition time information of each image frame in the original image sequence is extracted, and at the same time, the sampling time information of each measurement record in the real-time flight attitude data is extracted. Secondly, a precise time stamp is assigned to each image frame and each measurement record. Finally, it is ensured that all sensor data have a unified time reference, forming multiple time-synchronized sensor data.

[0052] Step 114: Package the time-synchronized sensor data to generate an environmental status data package.

[0053] In this step, the packetization operation refers to the process of combining multiple data units into a single data packet; the environmental status data packet refers to the transmission data unit containing environmental awareness information.

[0054] In this embodiment of the invention, the time-synchronized sensor data is first classified and organized according to its data source. Then, the visual sensor data and inertial measurement data are paired and combined in chronological order. Subsequently, the paired sensor data is packaged using a data encapsulation protocol to generate an environmental status data packet containing timestamps and data type information.

[0055] Step 115: Use a preset wireless transmission module to transmit the environmental status data packet to the relay node of the preset UAV cluster to perform time synchronization verification on the environmental status data packet, integrate the verification results, and generate a cluster environmental status dataset.

[0056] In this step, the preset wireless transmission module refers to the wireless communication device that has been set up in advance; the relay node refers to the specific UAV in the cluster that is responsible for data forwarding; the time synchronization verification operation refers to the process of checking the consistency of data time; the verification result refers to the output information of the time synchronization check; the integration operation refers to the process of merging multiple data units into a complete data set; and the cluster environment status dataset refers to the environmental perception data set of the entire UAV cluster.

[0057] In this embodiment of the invention, environmental status data packets are first sent to the drone cluster relay node through a preset wireless transmission module. Then, the received data packets are checked for timestamp consistency at the relay node. Subsequently, the data packets that pass the verification are reassembled. Finally, the reassembled data is integrated to form a complete cluster environmental status dataset.

[0058] Step 116: Perform format unification processing on the cluster environment status dataset to obtain environmental image data and flight status data sets from the processing results.

[0059] In this step, format standardization refers to the process of converting data in different formats into a standard format; the processing result refers to the data output obtained after format standardization.

[0060] In this embodiment of the invention, the cluster environment status dataset is first standardized in terms of data format, then sensor data from different sources are converted into a unified data format, then the validity of the converted data is verified, and finally, environmental image data and flight status data sets that meet the requirements are separated from the processing results.

[0061] This invention achieves high-quality acquisition and preprocessing of UAV swarm environmental perception data through multi-source sensor data acquisition, high-precision time synchronization, reliable data transmission, and unified format processing. This provides an accurate and reliable data foundation for subsequent environmental modeling and collaborative decision-making, significantly improving the swarm system's perception capability and data reliability in complex environments.

[0062] To improve the completeness and accuracy of obstacle modeling in unknown environments, this step achieves dense environmental map construction by fusing visual feature point trajectory tracking with dynamic obstacle models. This invention provides a specific embodiment: Step 102, using preset simultaneous localization and mapping (SLAM) technology to process the environmental image data and the flight state data to generate a dense environmental map, specifically includes the following steps: Step 201: Extract multiple visual feature points from the environmental image data, and associate each visual feature point with the location information corresponding to the acquisition time of each visual feature point in the flight status data to generate a target visual feature point set.

[0063] In this step, visual feature points refer to key location points in the image that have significant distinguishability; acquisition time refers to the specific time point when sensor data is acquired; location information refers to spatial location data that matches the acquisition time of feature points; association operation refers to the process of establishing correspondence between data; and target visual feature point set refers to the set of feature points after filtering and association processing.

[0064] In this embodiment of the invention, visual feature points with significant distinguishability are first detected from environmental image data. Then, a correspondence is established between each visual feature point and the spatial location information in the flight status data corresponding to the acquisition time. Subsequently, the feature points are bound to the location information through data association operations. Finally, a target visual feature point set containing spatial location attributes is generated.

[0065] Step 202: Perform similarity matching on each target visual feature point in the target visual feature point set to construct a feature point trajectory sequence.

[0066] In this step, the similarity matching operation refers to the process of comparing the similarity of feature points; the feature point trajectory sequence refers to the motion path of feature points in consecutive frames.

[0067] In this embodiment of the invention, feature descriptors are first calculated for each target visual feature point in the target visual feature point set, then feature similarity is compared between different image frames, then the feature point matching relationship is determined according to the similarity threshold, and finally the feature point trajectory sequence is constructed by connecting the matching between consecutive frames.

[0068] Step 203: Calculate the pose change between adjacent acquisition times based on the feature point trajectory sequence and the position information corresponding to the feature point trajectory sequence.

[0069] In this step, position information refers to the spatial position data corresponding to each point in the trajectory sequence; pose change refers to the relative motion change value between adjacent time points.

[0070] In this embodiment of the invention, the spatial displacement information of each feature point in the feature point trajectory sequence is first obtained, then the attitude data in the corresponding position information of the trajectory sequence is combined, and then the relative pose change between adjacent acquisition times is calculated by kinematic principles. Finally, the motion transformation parameters of the UAV in continuous time moments are obtained.

[0071] Step 204: Apply the pose change amount to calculate the three-dimensional spatial coordinates of each of the target visual feature points to generate a sparse environment point cloud.

[0072] In this step, the three-dimensional spatial coordinate calculation operation refers to the process of calculating the spatial position of the point cloud; the sparse environment point cloud refers to the preliminary reconstructed three-dimensional point set.

[0073] In this embodiment of the invention, firstly, the pose change amount is used to establish the coordinate system transformation relationship, then the three-dimensional spatial coordinates of each target visual feature point are calculated, then the spatial position of the feature point is calculated by the triangulation principle, and finally a sparse environmental point cloud describing the environmental structure is generated.

[0074] Step 205: Reconstruct the surface of the sparse environmental point cloud to generate the target environmental point cloud.

[0075] In this step, "surface" refers to the geometric shape of the object's surface as expressed by the point cloud data; "reconstruction operation" refers to the process of building a continuous surface; and "target environment point cloud" refers to the point cloud data after surface reconstruction.

[0076] In this embodiment of the invention, the surface normal vector of the sparse environmental point cloud is first estimated, then the missing regions of the point cloud are filled by a surface fitting algorithm, followed by surface smoothing and topology construction, and finally a continuous and complete target environmental point cloud is generated.

[0077] Step 206: Fuse the surface of the target environment point cloud with the geometric model of the preset dynamic obstacles to generate a dense environment map.

[0078] In this step, "surface" refers to the outer surface geometry of the reconstructed point cloud; "geometric model" refers to the mathematical geometric representation of the obstacle; and "fusion operation" refers to the process of integrating different data.

[0079] In this embodiment of the invention, the surface geometric information of the target environment point cloud is first extracted, then registered and aligned with the geometric model of a preset dynamic obstacle, then the obstacle model is integrated into the environment point cloud through a model fusion algorithm, and finally a dense environment map containing dynamic obstacle information is generated.

[0080] This invention achieves accurate environmental feature localization by extracting visual features and associating them with spatial locations. It establishes accurate motion estimation by tracking feature point trajectories and calculating pose changes. It generates a complete environmental geometric model through three-dimensional coordinate calculation and surface reconstruction. Finally, it forms a dense map containing dynamic obstacle information by fusing obstacle models, which significantly improves the accuracy and completeness of environmental modeling.

[0081] To address the problem of accumulated relative pose measurement errors in UAV swarms, this step constructs a high-precision relative pose state network through microwave interferometric ranging and motion vector fusion. A specific embodiment of this invention is provided: Step 103 involves performing spatial geometric relationship calculations on the relative distances and angular changes between the UAVs to construct the high-precision relative pose state network of the pre-set UAV swarm. This specifically includes the following steps: Step 301: Measure the original distance data between each UAV and the original angle data between each UAV using a preset sparse array microwave interferometric ranging system.

[0082] In this step, the preset sparse array microwave interferometric ranging system refers to the ranging device based on the principle of microwave interferometry that is deployed in advance; the raw distance data refers to the straight-line distance between UAVs obtained by direct measurement; and the raw angle data refers to the azimuth angle between UAVs obtained by direct measurement.

[0083] In this embodiment of the invention, microwave signals are first transmitted to each UAV through a pre-deployed sparse array microwave interferometric ranging system and the echoes are received. Then, the propagation time difference and phase difference of the microwave signals between the UAVs are measured. Finally, the original distance data and original angle data between the UAVs are obtained.

[0084] Step 302: Calculate the distance phase difference between each set of raw distance data adjacent to the measurement cycle and the angular phase difference between each set of raw angle data adjacent to the measurement cycle.

[0085] In this step, the measurement period refers to the time interval between two consecutive measurements; the distance phase difference refers to the phase change of the distance data between adjacent measurement periods; and the angle phase difference refers to the phase change of the angle data between adjacent measurement periods.

[0086] In this embodiment of the invention, the original distance data sequence within a continuous measurement period is first obtained, the phase difference between the original distance data of adjacent measurement periods is calculated, the original angle data sequence within a continuous measurement period is obtained, and finally the phase difference between the original angle data of adjacent measurement periods is calculated.

[0087] Step 303: Calculate the relative motion vector between each UAV based on the distance phase difference and the angle phase difference.

[0088] In this step, the relative motion vector refers to the vector that describes the relative motion state between the UAVs.

[0089] In this embodiment of the invention, the relative radial velocity components between UAVs are first calculated based on the distance phase difference, and the relative tangential velocity components between UAVs are calculated based on the angular phase difference. Then, the radial velocity components and the tangential velocity components are vector-synthesized to obtain the three-dimensional relative motion vector between each UAV.

[0090] Step 304: Apply the relative motion vector to update the estimated relative pose of each UAV to generate a dynamic relative pose table.

[0091] In this step, the estimated relative pose refers to the relative position and attitude of the UAV predicted based on historical data; the update operation refers to the process of correcting and refreshing the data; and the dynamic relative pose table refers to the data table that records the real-time relative pose.

[0092] In this embodiment of the invention, firstly, the historical estimated relative pose data of each UAV is obtained; secondly, the relative motion vector is applied to perform kinematic update calculations on the historical pose; then, the pose estimation error is corrected based on the update results; and finally, a dynamic relative pose table containing the latest timestamp is generated.

[0093] Step 305: Construct the topological connection relationship of the preset UAV cluster based on the dynamic relative pose table.

[0094] In this step, topology refers to the relational network that describes the connection structure between UAVs.

[0095] In this embodiment of the invention, the spatial position distribution in the dynamic relative pose table is first analyzed, then the connection weights are calculated based on the relative distance and azimuth angle between the UAVs, then the topological connection strength is determined based on the connection weights, and finally the topological connection relationship describing the spatial structure of the UAV cluster is constructed.

[0096] Step 306: Fuse the topological connection relationship and the dynamic relative pose table to generate a high-precision relative pose state network.

[0097] In this step, the fusion operation refers to the process of integrating multiple types of data.

[0098] In this embodiment of the invention, the topological connection relationship is first converted into an adjacency matrix representation, the dynamic relative pose table is then converted into a pose data matrix, and the adjacency matrix and pose data matrix are integrated through a matrix fusion algorithm. Finally, a high-precision relative pose state network containing topological and pose information is generated.

[0099] This invention achieves high-precision relative measurement through a microwave interferometric ranging system, accurately obtains relative motion information by calculating phase difference, realizes real-time pose correction through motion vector update, and finally constructs a complete relative pose state network by fusing topological relationships and pose data, which significantly improves the accuracy and reliability of relative positioning of UAV swarms.

[0100] To address the issue that static role allocation strategies cannot adapt to dynamic environments, this step achieves dynamic evolution of role probabilities through capability quantization encoding and strategy exploration factor injection. This invention provides a specific embodiment: Step 104, based on the device performance parameters of each UAV and the environmental interaction history data of each UAV, drives role evolution processing to generate role probability distribution data for each UAV, specifically including the following steps: Step 401: Perform role capability quantization encoding on the equipment performance parameters of each UAV to generate an initial role capability vector.

[0101] In this step, the character ability quantization and encoding process refers to the process of converting performance parameters into numerical vectors; the initial character ability vector refers to the numerical vector representing the basic abilities.

[0102] In this embodiment of the invention, the equipment performance parameters of each UAV are first obtained, including indicators such as maximum flight speed, payload capacity and sensor accuracy. Then, a quantization encoding algorithm is used to convert these performance parameters into numerical representations. Subsequently, an initial role capability vector reflecting the basic capabilities of the UAV is generated according to a preset encoding rule.

[0103] Step 402: Merge the initial character capability vector and the dynamic character performance records in the environmental interaction history data of each UAV to generate a comprehensive character evaluation index.

[0104] In this step, dynamic role performance record refers to performance data in historical tasks; fusion operation refers to the process of integrating multiple data; and comprehensive role evaluation index refers to the evaluation value of comprehensive ability.

[0105] In this embodiment of the invention, the dynamic role performance records of the UAV in past missions are first extracted from the historical data of environmental interaction, including indicators such as mission completion rate and collaborative efficiency. Then, a weighted fusion algorithm is used to integrate the initial role capability vector with the dynamic role performance records. Finally, a comprehensive role evaluation index is calculated through a comprehensive evaluation model.

[0106] Step 403: Calculate the fitness weight of each UAV in different role tasks based on the comprehensive role evaluation index.

[0107] In this step, role task refers to the specific task type that needs to be performed; fitness weight refers to the quantitative value of the degree of suitability.

[0108] In this embodiment of the invention, the adaptability of each UAV to different role tasks is first analyzed based on the comprehensive role evaluation index. Then, the fitness calculation model is used to calculate the fitness weights of tracking, cover and reconnaissance tasks respectively. Finally, a weight set containing multi-task fitness values ​​is generated.

[0109] Step 404: Input different preset strategy exploration factors into the preset current role strategy set according to the fitness weights, and generate a role strategy mutation scheme.

[0110] In this step, the preset strategy exploration factor refers to the parameter that controls the exploration intensity; the preset current role strategy set refers to the set of strategies set in advance; and the role strategy mutation scheme refers to the scheme that includes new strategies.

[0111] In this embodiment of the invention, the intensity parameter of strategy exploration is first determined according to each fitness weight, then the preset strategy exploration factor is input into the preset current role strategy set, and then a role strategy mutation scheme containing new strategy combinations is generated through a strategy mutation algorithm.

[0112] Step 405: Apply the aforementioned role strategy mutation scheme to update the role selection preferences of each of the drones, generating multiple updated role selection preferences.

[0113] In this step, "role selection preference" refers to the selection bias data; "update operation" refers to the process of modifying the data; and "updated role selection preference" refers to the modified bias data.

[0114] In this embodiment of the invention, the current role selection preference data of each UAV is first obtained, then the preference data is adjusted by applying a role strategy mutation scheme, and then multiple updated role selection preferences are generated by a preference update algorithm.

[0115] Step 406: Perform probability normalization processing on the values ​​of each of the updated role selection preferences according to the updated preferences, and generate role probability distribution data for each of the drones.

[0116] In this step, probability normalization refers to the process of converting the data into a probability distribution.

[0117] In this embodiment of the invention, firstly, all updated character selection preference values ​​are collected; secondly, a probability normalization algorithm is used to convert these values ​​into a probability distribution; then, the sum of the probabilities of each character is ensured to be 1; and finally, character probability distribution data for each drone is generated.

[0118] This invention converts device performance into capability vectors through quantization and encoding, performs fusion evaluation by combining historical performance data, explores factor generation mutation schemes based on fitness weight input strategy, and finally generates role probability distributions through preference updates and probability normalization, thereby realizing dynamic adaptive allocation of UAV roles and significantly improving the task adaptability and collaborative efficiency of the cluster system.

[0119] To address the issue that traditional path planning cannot simultaneously address multi-objective optimization, this step generates a multi-objective reward function through spatial relationship encoding and potential energy field parameter transformation. This invention provides a specific embodiment: Step 105, applying the dense environmental map and the high-precision relative pose state network, maps the positional relationships between preset dynamic obstacles, preset task objectives, and preset friendly drone clusters into a potential energy field reward function, specifically including the following steps: Step 501: Extract the contour information of preset dynamic obstacles and the real-time position coordinates of preset friendly UAV clusters from the dense environmental map and the high-precision relative pose state network, respectively.

[0120] In this step, contour information refers to the obstacle's shape feature data; real-time position coordinates refer to the spatial position data at the current moment.

[0121] In this embodiment of the invention, the contour information of the preset dynamic obstacles, including boundary features and shape parameters, is first extracted from the dense environmental map. Then, the real-time position coordinates of the preset friendly UAV cluster, including three-dimensional spatial coordinates and attitude information, are obtained from the high-precision relative pose state network. Finally, the contour information and position coordinates are sorted and classified.

[0122] Step 502: Parse the preset task target coordinate information to generate a spatial location description of the task target.

[0123] In this step, the preset task target coordinate information refers to the pre-defined target location data; the parsing operation refers to the action of analyzing and processing the data; and the task target spatial location description refers to the characteristic description of the target location.

[0124] In this embodiment of the invention, the preset task target coordinate information, including the latitude, longitude and elevation data of the target point, is first read. Then, the coordinate information is parsed and processed, including coordinate transformation and format standardization. Subsequently, a task target spatial location description containing target spatial attributes and priority information is generated.

[0125] Step 503: Calculate the minimum distance vector between the contour information and the real-time position coordinates, as well as the relative azimuth vector between each friendly drone in the preset friendly drone cluster.

[0126] In this step, the minimum distance vector refers to the direction of the shortest distance; the relative azimuth vector refers to the angle of the relative direction.

[0127] In this embodiment of the invention, firstly, the Euclidean distance from each boundary point in the contour information to the real-time position coordinates is calculated and the minimum value is taken to generate the minimum distance vector. Secondly, the relative azimuth angles between each drone in the preset friendly drone cluster are calculated to generate the relative azimuth vector. Finally, a set of vectors describing the spatial relationship is obtained.

[0128] Step 504: Encode the spatial relationship between the minimum distance vector and the relative orientation vector to generate a relative spatial relationship description.

[0129] In this step, spatial relation encoding refers to the digital processing of spatial relations; relative spatial relation description refers to the characteristic representation of spatial relations.

[0130] In this embodiment of the invention, the minimum distance vector is first normalized, the relative orientation vector is then oriented and encoded, the processed vectors are then fused using a spatial relationship model, and finally a relative spatial relationship description containing distance and orientation information is generated.

[0131] Step 505: Convert the relative spatial relationship description into potential energy field parameters.

[0132] In this step, the conversion operation refers to the data format transformation process; the potential field action parameter refers to the control parameter of the potential field.

[0133] In this embodiment of the invention, a mapping rule from relative spatial relationship description to potential energy field parameters is first established. Then, the repulsive force parameter is calculated based on the distance relationship. Subsequently, the attractive force parameter is calculated based on the orientation relationship. Finally, all kinds of parameters are integrated into potential energy field action parameters.

[0134] Step 506: Combine the potential energy field parameters and the spatial location description of the mission target to generate a potential energy field benefit function.

[0135] In this step, the combination operation refers to the data integration process, including multi-source data integration methods such as parameter weighting, matrix operations, and function construction.

[0136] In this embodiment of the invention, the potential energy field parameters are first weighted, then the weighted parameters are combined with the spatial location description of the task target in a matrix, a multi-objective benefit function is constructed through a function generation algorithm, and finally the potential energy field benefit function is generated.

[0137] This invention extracts the contours of environmental obstacles and the real-time location information of UAVs, analyzes the spatial attributes of the mission target, calculates accurate spatial relationship vectors, generates potential energy field parameters through encoding and conversion, and finally combines them to form a multi-objective benefit function. This achieves intelligent perception of environmental spatial relationships and accurate construction of potential energy field models, significantly improving the obstacle avoidance capability and collaborative operation efficiency of UAV swarms.

[0138] To address the strategy conflict problem in multi-UAV collaborative decision-making, this step analyzes the equilibrium state of the game field through conflict detection and collaborative optimization. This invention provides a specific embodiment: Step 106, adjusting the flight paths and task allocation strategies of each UAV based on the role probability distribution data and the potential energy field payoff function, generates a collaborative evolution game field of the pre-set UAV cluster in an unknown dynamic environment to achieve equilibrium state analysis of the pre-set UAV cluster. This specifically includes the following steps: Step 601: Transform the probability distribution data of the roles and the potential energy field benefit function respectively to obtain the initial task allocation strategy and path planning benefit value of each UAV.

[0139] In this step, conversion processing refers to the operation of data format conversion, including data conversion methods such as format conversion, quantization processing, and mapping transformation; initial task allocation strategy refers to the preliminary task allocation plan; path planning benefit value refers to the quantified value of the benefits of path planning.

[0140] In this embodiment of the invention, the role probability distribution data is first processed by strategy mapping to convert it into a specific task allocation scheme. At the same time, the potential energy field benefit function is quantified to convert it into a path planning benefit value. Finally, the initial task allocation strategy and path planning benefit value of each UAV are obtained.

[0141] Step 602: Apply the initial task allocation strategy and the path planning benefit value to construct multi-objective optimization constraints.

[0142] In this step, the multi-objective optimization constraint refers to the restriction conditions of multi-objective optimization.

[0143] In this embodiment of the invention, firstly, task allocation constraints are formulated based on the initial task allocation strategy, and secondly, path optimization constraints are formulated based on the path planning benefit value. Then, the two types of constraints are integrated and processed to finally construct the multi-objective optimization constraints.

[0144] Step 603: Detect policy conflicts in the initial task allocation strategy based on the multi-objective optimization constraints to identify policy execution conflict characteristics between the UAVs.

[0145] In this step, the strategy conflict detection operation refers to the process of identifying strategy conflicts; the strategy execution conflict characteristics refer to the specific manifestations of strategy conflicts.

[0146] In this embodiment of the invention, firstly, conflict analysis is performed on the initial task allocation strategy based on multi-objective optimization constraints; secondly, the task overlap area and resource competition relationship between UAVs are detected; then, the conflict characteristics that may occur during the strategy execution process are identified; and finally, a detailed description of the conflict characteristics of strategy execution is obtained.

[0147] Step 604: Perform collaborative strategy optimization processing on the conflict characteristics of each strategy to generate a conflict resolution strategy set.

[0148] In this step, collaborative strategy optimization refers to the collaborative optimization process; conflict resolution strategy set refers to the set of conflict resolution solutions.

[0149] In this embodiment of the invention, the characteristics of policy execution conflict are first analyzed through collaborative optimization, then conflict resolution schemes are formulated, including task reallocation and resource coordination strategies, and then an optimized set of strategies is generated through a collaborative algorithm, finally forming a complete set of conflict resolution strategies.

[0150] Step 605: Apply the conflict resolution strategy set to update the flight path and task allocation strategy of each UAV, and generate the updated flight path and updated task allocation strategy.

[0151] In this step, the update operation refers to the data refresh process, including data refresh methods such as parameter adjustment, strategy modification, and status update; the updated flight path refers to the optimized flight trajectory; and the updated task allocation strategy refers to the optimized task plan.

[0152] In this embodiment of the invention, firstly, a set of conflict resolution strategies is applied to optimize and adjust the flight paths of each UAV. Secondly, the task allocation strategy is replanned and reassigned. Subsequently, updated flight paths and updated task allocation strategies are generated, and finally, the coordinated update of all UAV strategies is completed.

[0153] Step 606: Perform an equilibrium convergence determination on the overall distribution state between the updated flight path and the updated task allocation strategy to generate a cooperative evolution game field and equilibrium state analysis results.

[0154] In this step, the overall distribution state refers to the overall distribution of the system; the equilibrium convergence determination operation refers to the process of determining the equilibrium state; and the equilibrium state analysis result refers to the result data of the equilibrium analysis.

[0155] In this embodiment of the invention, firstly, an overall coordination analysis is performed on the updated flight path and task allocation strategy; secondly, the convergence degree and equilibrium characteristics of the system state are evaluated; then, a co-evolutionary game field and equilibrium state analysis results are generated through an equilibrium determination algorithm; and finally, the equilibrium state analysis of the entire system is completed.

[0156] This invention generates initial strategies and reward values ​​by converting role probabilities and potential energy field gains, performs conflict detection and collaborative optimization by constructing multi-objective constraints, and finally achieves collaborative decision-making for UAV swarms through strategy updates and equilibrium determination, significantly improving the collaborative efficiency and task execution capability of multi-agent systems in complex environments.

[0157] Figure 2 This invention provides a schematic diagram of the structure of a complex game field generation and equilibrium analysis system based on multi-agent evolution, as shown in the embodiment of the invention. Figure 2 As shown, the system includes: The acquisition module 21 is used to acquire environmental image data and flight status data of each drone in the preset drone cluster; Processing module 22 is used to process the environmental image data and the flight status data using a preset synchronous positioning and map building technology to generate a dense environmental map; The calculation module 23 is used to perform spatial geometric relationship calculation on the relative distance between each UAV and the angular change between each UAV, so as to construct a high-precision relative pose state network of the pre-set UAV cluster. Evolution module 24 is used to drive role evolution processing based on the equipment performance parameters of each UAV and the environmental interaction history data of each UAV, and generate role probability distribution data of each UAV. Mapping module 25 is used to apply the dense environmental map and the high-precision relative pose state network to map the positional relationship between preset dynamic obstacles, preset task objectives and preset friendly drone clusters into a potential energy field gain function. The adjustment module 26 is used to adjust the flight path and task allocation strategy of each UAV according to the role probability distribution data and the potential energy field revenue function, and generate a cooperative evolution game field of the preset UAV cluster in an unknown dynamic environment to realize the equilibrium state analysis of the preset UAV cluster.

[0158] Figure 2 The aforementioned system for generating and analyzing complex game fields based on multi-agent evolution can execute... Figure 1The implementation principle and technical effects of the complex game field generation and equilibrium analysis method based on multi-agent evolution described in the illustrated embodiment will not be repeated here. The specific methods by which each module and unit performs operations in the complex game field generation and equilibrium analysis system based on multi-agent evolution described in the above embodiments have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0159] In one possible design, Figure 2 The illustrated embodiment of a complex game field generation and equilibrium analysis system based on multi-agent evolution can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0160] The processing component 32 is used to: acquire environmental image data and flight status data of each drone in a pre-set drone cluster; process the environmental image data and flight status data using a pre-set synchronous positioning and mapping technology to generate a dense environmental map; perform spatial geometric relationship calculation on the relative distances and angular changes between the drones to construct a high-precision relative pose state network of the pre-set drone cluster; drive role evolution processing based on the device performance parameters and environmental interaction history data of each drone to generate role probability distribution data of each drone; apply the dense environmental map and the high-precision relative pose state network to map the positional relationships between pre-set dynamic obstacles, pre-set task objectives, and pre-set friendly drone clusters into a potential energy field benefit function; adjust the flight paths and task allocation strategies of each drone based on the role probability distribution data and the potential energy field benefit function to generate a cooperative evolution game field of the pre-set drone cluster in an unknown dynamic environment to achieve equilibrium state analysis of the pre-set drone cluster.

[0161] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0162] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0163] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0164] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0165] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0166] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0167] This invention also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown is a method for generating and analyzing complex game fields based on multi-agent evolution.

[0168] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0169] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0170] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0171] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating and analyzing complex game fields based on multi-agent evolution, characterized in that, include: Acquire environmental image data and flight status data of each drone in a pre-set drone cluster; The environmental image data and flight status data are processed using a preset synchronous positioning and mapping technology to generate a dense environmental map. Spatial geometric relationship calculation is performed on the relative distances between the drones and the angular changes between the drones to construct a high-precision relative pose state network for the pre-set drone cluster. Based on the device performance parameters and environmental interaction history data of each UAV, a role evolution process is driven to generate role probability distribution data for each UAV. The role evolution process refers to the dynamic process of adjusting role allocation based on capabilities and historical performance, including: quantifying and encoding the device performance parameters of each UAV to generate an initial role capability vector; whereby the role capability quantification and encoding process refers to the process of converting performance parameters into numerical vectors; and fusing the initial role capability vector with the dynamic role performance records in the environmental interaction history data of each UAV to generate a comprehensive role evaluation index; whereby the dynamic role performance records refer to the tables in historical tasks. The current data includes the dynamic role performance record, which includes task completion rate and collaboration efficiency. The fitness weight of each UAV in different role tasks is calculated based on the comprehensive role evaluation index. Different preset strategy exploration factors are input into a preset current role strategy set according to the fitness weights to generate a role strategy mutation scheme. Here, the strategy exploration factor refers to the parameter controlling the exploration intensity. The role selection preferences of each UAV are updated using the role strategy mutation scheme, generating multiple updated role selection preferences. The values ​​of each updated role selection preference are then subjected to probability normalization processing to generate role probability distribution data for each UAV. By applying the dense environmental map and the high-precision relative pose state network, the positional relationship between preset dynamic obstacles, preset mission objectives and preset friendly drone clusters is mapped into a potential energy field gain function. Based on the role probability distribution data and the potential energy field revenue function, the flight paths and task allocation strategies of each UAV are adjusted to generate a cooperative evolution game field of the pre-set UAV cluster in an unknown dynamic environment, so as to realize the equilibrium state analysis of the pre-set UAV cluster.

2. The method according to claim 1, characterized in that, Acquire environmental image data and flight status data of each drone in a pre-configured drone cluster, including: The system utilizes the visual sensors carried by each drone in a pre-set drone cluster to acquire a pre-defined sequence of original images of the surrounding environment. Real-time flight attitude data of each UAV is collected using the inertial measurement unit carried by each UAV. Each image frame in the original image sequence and each measurement record in the real-time flight attitude data are respectively appended with a timestamp synchronized with each image frame and each measurement record to obtain multiple time-synchronized sensor data. The time-synchronized sensor data is packaged to generate an environmental status data package; The environmental status data packet is transmitted to the relay node of the preset drone cluster using a preset wireless transmission module to perform time synchronization verification on the environmental status data packet. The verification results are then integrated to generate a cluster environmental status dataset. The cluster environment status dataset is formatted to obtain environmental image data and flight status data sets from the processing results.

3. The method according to claim 1, characterized in that, The environmental image data and flight status data are processed using a preset simultaneous localization and mapping (SLAM) technology to generate a dense environmental map, including: Multiple visual feature points are extracted from the environmental image data, and each visual feature point is associated with the location information corresponding to the acquisition time of each visual feature point in the flight status data to generate a target visual feature point set. Similarity matching is performed on each target visual feature point in the target visual feature point set to construct a feature point trajectory sequence; Based on the feature point trajectory sequence and the position information corresponding to the feature point trajectory sequence, calculate the pose change between adjacent acquisition times; The pose change is used to calculate the three-dimensional spatial coordinates of each of the target visual feature points to generate a sparse environment point cloud. The surface of the sparse environmental point cloud is reconstructed to generate the target environmental point cloud; The surface of the target environment point cloud and the geometric model of the pre-set dynamic obstacles are fused to generate a dense environment map.

4. The method according to claim 1, characterized in that, Spatial geometric relationship calculations are performed on the relative distances and angular changes between the drones to construct a high-precision relative pose state network for the pre-set drone cluster, including: The raw distance data and raw angle data between each UAV are measured using a pre-set sparse array microwave interferometric ranging system. Calculate the distance phase difference between each set of raw distance data adjacent to the measurement cycle and the angular phase difference between each set of raw angle data adjacent to the measurement cycle; Calculate the relative motion vector between each UAV based on the distance phase difference and the angle phase difference; The estimated relative pose of each UAV is updated using the relative motion vector to generate a dynamic relative pose table; wherein, the estimated relative pose refers to the relative position and attitude of the UAV predicted based on historical data. The topological connection relationship of the pre-set UAV cluster is constructed based on the dynamic relative pose table; The topological connection relationship and the dynamic relative pose table are fused to generate a high-precision relative pose state network.

5. The method according to claim 1, characterized in that, Using the dense environmental map and the high-precision relative pose state network, the positional relationships between preset dynamic obstacles, preset mission objectives, and preset friendly drone clusters are mapped into a potential energy field gain function, including: The contour information of preset dynamic obstacles and the real-time position coordinates of preset friendly drone clusters are extracted from the dense environmental map and the high-precision relative pose state network, respectively. The preset task target coordinate information is parsed to generate a spatial location description of the task target; Calculate the minimum distance vector between the contour information and the real-time position coordinates, as well as the relative azimuth vectors between each friendly drone in the preset friendly drone cluster; Spatial relationship encoding is performed on the minimum distance vector and the relative orientation vector to generate a relative spatial relationship description; The relative spatial relationship is described in terms of potential energy field parameters. The potential energy field parameters and the spatial location description of the mission target are combined to generate a potential energy field benefit function.

6. The method according to claim 1, characterized in that, Based on the role probability distribution data and the potential energy field reward function, the flight paths and task allocation strategies of each UAV are adjusted to generate a cooperative evolution game field of the pre-set UAV cluster in an unknown dynamic environment, thereby realizing the equilibrium state analysis of the pre-set UAV cluster, including: The probability distribution data of the roles and the potential energy field benefit function are transformed and processed respectively to obtain the initial task allocation strategy and path planning benefit value of each of the drones. The initial task allocation strategy and the path planning benefit value are used to construct multi-objective optimization constraints; Based on the multi-objective optimization constraints, the initial task allocation strategy is executed to detect policy conflict, so as to identify the policy execution conflict characteristics between the UAVs. The conflict characteristics of each strategy are subjected to collaborative strategy optimization processing to generate a set of conflict resolution strategies; The conflict resolution strategy set is applied to update the flight path and task allocation strategy of each UAV, generating updated flight paths and updated task allocation strategies. The overall distribution state between the updated flight path and the updated task allocation strategy is evaluated for equilibrium convergence to generate a cooperative evolution game field and equilibrium state analysis results.

7. A system for generating and analyzing complex game fields based on multi-agent evolution, used in the method for generating and analyzing complex game fields based on multi-agent evolution as described in any one of claims 1 to 6, characterized in that, include: The acquisition module is used to acquire environmental image data and flight status data of each drone in the pre-set drone cluster; The processing module is used to process the environmental image data and the flight status data using a preset synchronous positioning and mapping technology to generate a dense environmental map. The calculation module is used to perform spatial geometric relationship calculation on the relative distance between each UAV and the angular change between each UAV, so as to construct a high-precision relative pose state network of the pre-set UAV cluster. An evolution module is used to drive role evolution processing based on the equipment performance parameters of each UAV and the environmental interaction history data of each UAV, and generate role probability distribution data of each UAV. The mapping module is used to apply the dense environmental map and the high-precision relative pose state network to map the positional relationship between preset dynamic obstacles, preset task objectives and preset friendly drone clusters into a potential energy field gain function. The adjustment module is used to adjust the flight path and task allocation strategy of each UAV according to the role probability distribution data and the potential energy field revenue function, and generate a cooperative evolution game field of the preset UAV cluster in an unknown dynamic environment to realize the equilibrium state analysis of the preset UAV cluster.

8. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the method for generating and analyzing complex game fields based on multi-agent evolution as described in any one of claims 1 to 6.

9. A computer storage medium, characterized in that, The system contains a computer program that, when executed by a computer, implements a method for generating and analyzing complex game fields based on multi-agent evolution as described in any one of claims 1 to 6.

Images