A large-scale lightweight unmanned cluster test and simulation platform and working method

Abstract

The present invention discloses a large-scale lightweight unmanned cluster test simulation platform and a working method, including an operation module, a data module, a sensor module, a task file module, and a log file module. This platform can perform three-dimensional dynamic simulations of large-scale unmanned aircraft clusters. Due to the lightweight design of the platform and the flexibility of interfaces such as data and models, it can perform simulations of hundreds of unmanned aircraft with low-fidelity motion models or dozens of unmanned aircraft with high-fidelity motion models on ordinary computers. Moreover, the design of the platform includes parallel operation, facilitating its operation on a cluster computing system. The entire simulation process is presented in a visual form, and the simulation step size can be set by itself to achieve progress control of the simulation process.

Description

Technical Field

[0001] The present invention relates to a robot simulation platform, in particular to a large-scale lightweight unmanned cluster test simulation platform and working method. Background Art

[0002] Conducting large-scale unmanned aerial vehicle (UAV) cluster flight training in reality is a task with relatively high risks and high costs. However, with the continuous development of technology, large-scale UAV cluster confrontation in the future air combat field is inevitable. Therefore, in order to conduct training effectively and at low cost, many developers have developed various simulators and simulation platforms for simulation testing. There is a large gap between the existing simulation capabilities and the capabilities required for simulating a large number of aircraft. Many robot simulators are limited to 2D planes, such as the Stage robot simulator, etc., which greatly limits the simulation of aerial robots on the platform. Although there are some 3D robot simulators, when simulating a relatively large number of agents in these simulators, their performance may be greatly reduced. For example, the simulator Gazebo related to Stage, which is a dynamic three-dimensional robot simulator, is developed in close integration with the Robot Operating System (ROS), and it can be combined with SolidWorks to provide a closed kinematic chain for some CAD models with higher complexity. However, due to the relatively high cost of performing high-precision collision detection, the Gazebo simulator cannot be well extended when simulating hundreds of robot systems on a single computer.

[0003] Therefore, there is an urgent need to study a lightweight UAV cluster system simulation platform that can perform large-scale three-dimensional dynamic simulation. Summary of the Invention

[0004] In order to solve the above problems, the present invention discloses a large-scale lightweight unmanned cluster test simulation platform and working method, which can perform large-scale three-dimensional dynamic simulation of UAV clusters.

[0005] A large-scale lightweight unmanned cluster test simulation platform, characterized in that it includes the following modules: a task file module, used for editing task files, configuring initial conditions of the simulation, and configuring sensor plugins for the UAV group; a data module, used for storing model and script files; a sensor module, used for obtaining environmental actual situation information and the state information of the UAV group, and the actual situation information is interactively transmitted in the sensor modules of each UAV to generate various motion states of the UAVs; an operation module, specifying the task file to be used before the simulation and initiating a simulation task, retrieving the model and script files in the data module for simulation, calling the actual situation information obtained by the sensor module during the simulation, calculating and generating control instructions according to the desired state, and sending the control instructions to the sensor module to generate a new motion model; a log file module, used for recording the simulation data of the operation module.

[0006] Preferably, the task file is a specific task for simulation, including the task type and the parameters of the simulation; the task type includes an adversarial task, a game task, a collaborative task, or a playback task; the parameters of the simulation include the number of UAVs, the initial coordinates, and the map size.

[0007] Preferably, the model in the data module is a 3D aircraft model in the simulation; the script file is the underlying control logic of the UAV flight mode and is the basis for the UAV to change its flight attitude.

[0008] Preferably, the sensor module includes: an information acquisition plug-in, which acquires the current environmental situation information and the status information of the UAV swarm, converts them into sensor measurement values, and sends them to the controller plug-in; the controller plug-in, which acquires the information sent by the information acquisition plug-in, performs calculation and conversion, obtains the expected state at the next moment through neural network training, and converts the control instruction to send to the motion model plug-in; the motion model plug-in, which adjusts the pitch angle, yaw, and thrust according to the received execution instruction to generate a new state of the UAV swarm.

[0009] The present invention also discloses a working method for a large-scale lightweight unmanned cluster test simulation platform, including the following steps: Step 1, edit a task file in the task file module and read the task file through the operation module. Step 2, when it is determined according to the task file that the task type is an adversarial task, a game task, or a collaborative task, call the model and script file in the data module to configure the environment and then transfer to Step 3; when it is determined that the task type is a playback task, call the playback function configuration and then transfer to Step 5. Step 3, call each module of the platform for cluster simulation. Step 4, output the cluster simulation situation in Step 3, evaluate the simulation situation, input the score situation, and input them into the log file module for storage for future call. Step 5, read the data stored in the log file module for simulation playback.

[0010] Preferably, the working process of the operation module in Step 1 includes: Step 1.1, construct the underlying execution logic of the platform, read the settings of each plug-in, set various logical judgments in the UAV simulation process, and integrate the lightweight neural network tiny-dnn for flight combat learning. Step 1.2, execute the corresponding simulation according to the settings in the task file.

[0011] Preferably, step 3 is specifically as follows: Step 3.1, obtain information through an information acquisition plug-in, where the information includes environmental actual situation information and the status information of the UAV swarm. Step 3.2, transmit the information obtained in step 3.1 into the plug-ins of the sensor module, and circulate and transmit it among the various sensor plug-ins. By changing the thrust, pitch, roll, and yaw of the UAVs, the status of the UAVs is changed to generate new status information of the UAV swarm. Step 3.3, feedback the new status information of the UAV swarm generated in step 3.2 to the information acquisition plug-in in step 3.1, and perform calculations and status changes again according to the new information, so that the UAVs can change their status in real time.

[0012] Preferably, the environmental actual situation information m in step 3.1 includes the relative position, speed, relative heading, and distance of n surrounding UAVs, denoted as m t,i , where the subscripts t and i represent the i-th UAV at time t, i = 1...n; the status information of the UAV swarm includes the number of real-time friendly and enemy aircraft, flight status, and position.

[0013] Preferably, step 3.2 is specifically as follows: Step 3.2.1, input the information obtained in step 3.1 into a custom plug-in, and calculate the expected status for the next step according to the algorithms and functions defined by the custom plug-in. Step 3.2.2, generate control information according to the expected status and input it into the controller plug-in. The controller plug-in inputs the local situation information m of the i-th UAV at time t t,i into the neural network, trains to obtain a feasible flight status information a for the next moment t+1,i , and decodes the status information and converts it into a control instruction of the PID controller and sends it to the motion model plug-in. Step 3.2.3, the motion model plug-in changes the thrust, pitch, etc. of the UAV according to the control instruction to change the status, so that the UAV changes from status a t,i to the new status a t+1,i . Step 3.2.4, the UAVs perform simulation and confrontation through the interaction plug-in, and return the actual situation information to step 3.2.1 for recalculation, so that the UAVs can change their status in real time to cope with the change of the environmental situation.

[0014] Preferably, step 4 is specifically as follows: The UAV cluster outputs a simulation situation once after completing a simulation and stores and records it in the log file module. At the same time, input the simulation situation into the index plug-in to evaluate the simulation situation. The index plug-in analyzes the flight status and interaction situation of the UAVs in the cluster simulation according to the set reward function, calculates the score situation, and stores and records it in the log file module.

[0015] Beneficial effects:

[0016] (1) The large-scale lightweight unmanned cluster test simulation platform provided by the present invention is a 3D simulation environment. The simulation platform consumes low resources, memory and CPU, and has a low computational algorithm complexity and high computational efficiency, achieving its lightweight design. It can perform aircraft cluster simulations of dozens to hundreds of aircraft with high-fidelity or low-fidelity motion models on ordinary computers.

[0017] (2) The data, model and other interfaces of the simulation platform provided by the present invention are flexible. The custom plug-in module and data module can be designed by researchers according to research needs, with rich functions, increasing the scalability and practicability.

[0018] (3) When the simulation platform provided by the present invention runs the simulation, the scheduling logic between modules is clear, and the functions of the components of the simulated unmanned aircraft are clear, which is convenient for editing and calling. The information flow in the simulation process interacts repeatedly, enabling the unmanned aircraft to continuously take countermeasures according to real-time state changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the overall framework of the unmanned aircraft cluster system simulation platform according to an embodiment of the present invention;

[0020] Figure 2 It is a simulation flow chart of the simulation platform according to an embodiment of the present invention;

[0021] Figure 3 It is the components of the simulated unmanned aircraft according to an embodiment of the present invention;

[0022] Figure 4 It is an information interaction diagram during the simulation process according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0024] The present invention discloses that it can perform three-dimensional dynamic simulation, and through the lightweight design of the platform and the flexibility of data, model and other interfaces, the platform can perform aircraft cluster simulations of hundreds of aircraft with low-fidelity motion models or dozens of aircraft with high-fidelity motion models on ordinary computers. And the design of the platform includes running in parallel, that is, calculating multiple processes simultaneously at the same time, and each process has a CPU for operation, so as to run on a cluster computing system.

[0025] Such asFigure 1 As shown in the figure, it is a schematic diagram of a large-scale lightweight unmanned cluster test simulation platform, including an operation module, a data module, a sensor module, a mission file module, and a log file module.

[0026] Before the simulation starts, edit the mission file in the mission file module, including determining the autonomous control logic such as the model, flight mode, and control algorithm of the unmanned aerial vehicle (UAV), as well as configuring the initial conditions for various simulations. Then, configure the sensor plugins for the UAVs in the sensor module, and configure the initial values, algorithms, and functions of each sensor plugin. The operation module reads the mission file in the mission file module, calls the script files and models in the data module, and starts the simulation. When the simulation starts, the information acquisition plugin in the sensor module obtains the initial actual situation information of the environmental situation, and the processed information starts to circulate in the sensor module. Each plugin outputs according to its own different functions, which is reflected as different flight behaviors of the UAVs in the simulation. When the simulation ends, the data of the simulation is output and stored by the log module, and the results can be directly retrieved for subsequent analysis.

[0027] Specifically,

[0028] The mission file module is used to edit the mission file, configure the initial conditions of the simulation, and configure the sensor plugins for the UAV swarm. When preparing to conduct a simulation, it is necessary to first specify the mission file, that is, specify the specific mission of the simulation: including cluster tasks such as confrontation, game, and cooperation, or the playback function, and set the start, end time, or end conditions of the simulation in the mission file, set the time step of the simulation, the size of the simulation environment, the coordinate origin of the simulation scene, and the initial position coordinates and states of the UAV swarm. Configure the initial conditions of the simulation: set whether to open the visualization window and whether to automatically start the simulation until it ends; set the UAV model used in the simulation, the number of UAVs; set the data to be recorded during the simulation and the file location for storing the data; set the plugins used by the UAV swarm during the simulation process and set the number of threads of the simulation.

[0029] The data module is used to store the model and the script files for UAV flight control. The model includes the 3D model of the UAV and the digital terrain model. The 3D models of UAVs provided by the simulation platform include fixed-wing aircraft and quadcopter models. Researchers can also import their own designed 3D aircraft models and add custom attack weapon models to them. The digital terrain model can set different scenarios according to simulation requirements, such as mountains, islands, plains, etc. The script files for UAV flight control are also saved in the data module for easy subsequent calling.

[0030] The sensor module is used to obtain the real-time situation information of the environment. This real-time situation information is interactively transmitted among the sensor modules of each unmanned aerial vehicle (UAV) to generate the motion models of each UAV. When the platform conducts simulation, most of the behaviors of the UAVs are controlled and decided by the sensor module. Different parts in the sensor module have different functions. During the simulation process, the state information of the UAVs and the environment is transmitted among the various plugins. These plugins, together with the UAV models and script files in the data module, jointly form a simulated UAV. The plugins in the sensor module include:

[0031] The information acquisition plugin acquires the real-time situation information of the environment and the state information of the UAV swarm and sends them to the custom plugin. The state information of the UAV swarm includes the number of friendly and enemy aircraft that change in real time, flight states, etc.

[0032] The custom plugin is used to set the corresponding algorithms. According to the design logic and algorithms, it can calculate the motion state to be achieved in the next stage, and via the controller plugin, determine how the UAV should adjust the pitch angle, yaw, thrust, etc. to achieve the desired heading, and generate execution commands for the motion model plugin according to the required state. Researchers can set the corresponding functions according to their research needs. Simple functions include calculating the minimum distance between each UAV in the simulation, setting the UAV to attack the enemy aircraft with the shortest distance, etc. Complex functions include integrating classic swarm algorithms and training the offensive and defensive strategies of the UAV swarm through reinforcement learning.

[0033] The controller plugin, based on the preprocessed environmental situation information, inputs the local situation information m of the i-th UAV at time t t,i into the neural network, trains to obtain a feasible flight state information a at the next moment t+1,i , and decodes the state information, converting it into the input of the PID controller, thereby controlling the thrust, roll, pitch, and yaw of the UAV to achieve the desired state.

[0034] The motion model plugin is directly associated with the controller plugin. Based on the current state of the UAV and the control output of the controller plugin, it enables the UAV to move from state a t,i to the new state a t+1,i in the global coordinate system.

[0035] The interaction plugin defines the constraint conditions and physical interactions among the UAVs in the simulation. It includes collisions between UAVs and collisions between UAVs and enemy bases or the ground.

[0036] The index plugin analyzes the flight states and interaction situations of the UAVs in the swarm simulation according to the set reward function, calculates the results of the simulation after information processing, and finally stores the data in the log file module according to the category.

[0037] The running module is essential for simulating the confrontation between UAV swarms using this platform. All other modules need to be associated with this module and read and run by the running module. If you want to perform a confrontation task between two teams of UAVs, you should select the corresponding function in the running module and specify the corresponding task file, model, etc. for it. If you want to replay the simulation process, you should select the replay function in the running module and specify the recording file path stored in the log file module to perform the simulation replay.

[0038] The log file module. After the simulation ends, if you need to analyze the simulation results, you can look for the corresponding recording file in the log file module. If you want to perform a simulation replay, select the corresponding function in the running module and then specify the replay file to perform a visual replay; if you want to view the score situation of both UAV swarms, you can also directly open the score recording file to read the score situation. The log file module will automatically classify and store the data during the simulation, facilitating subsequent research calls.

[0039] The present invention also discloses a working method of a large-scale lightweight unmanned cluster test simulation platform, as Figure 2 shown, specifically including:

[0040] Step 1, edit the task file in the task file module and read the task file through the running module.

[0041] Step 2, after judging the task type, call the corresponding function according to different tasks and start the simulation according to the following steps.

[0042] When judging that the task type is an adversarial task, a game task, or a collaborative task according to the task file, call the data module to perform the initial configuration of the simulation environment and then transfer to Step 3; when judging that the task type is a replay task, call the replay function configuration and then transfer to Step 5.

[0043] Step 3, call each module of the platform to perform cluster simulation.

[0044] Step 3.1, when the specified task is adversarial, game, or collaborative, call the data module to perform the initial configuration of the simulation environment according to the task file, and the initial configuration content is shown in Table 1.

[0045] Table 1 Initial configuration of the simulation environment

[0046] simulation scene terrain aircraft model flight data record storage file start time dimension mountain aircraft type limit height flight trajectory file type end time number of aircraft plain motion model limit speed use plug-in storage location end condition coordinate origin sea surface weapon module flight mode interaction algorithm

[0047] Step 3.2, after the initial environment and data configuration are completed, the information acquisition plugin transmits the initial configuration information to the sensor module. Each plugin calculates based on this information. The custom plugin can calculate the motion state to be achieved in the next stage according to the design logic and algorithm, and the controller plugin determines how the UAV should adjust the pitch angle, yaw, thrust, etc. to achieve the required heading, and generates an execution command for the motion model plugin according to the required state. Finally, the motion model plugin generates a new state for the UAV in the global coordinate according to the current state of the UAV and the control output of the controller plugin.

[0048] Specifically, in Step 3.2.1, the information obtained in Step 3.1 is input into the custom plugin, and the expected state for the next step is calculated according to the algorithm and function defined by the custom plugin.

[0049] Step 3.2.2, control information is generated according to this expected state and input into the controller plugin. The controller plugin inputs the local situation information m of the i-th UAV at time t t,i into the neural network, and trains to obtain a feasible flight state information a at the next moment t+1,i , and decodes the state information and converts it into a control instruction of the PID controller and sends it to the motion model plugin.

[0050] Step 3.2.3, the motion model plugin changes the thrust, pitch, etc. of the UAV according to the control instruction to change the state, so that the UAV changes from state a t,i to the new state a t+1,i .

[0051] Step 3.2.4, the UAVs perform simulation and confrontation through the interaction plugin. The state of the UAV swarm is constantly changing, including position, speed, and heading information. The state information after each interaction is returned to Step 3.2.1 for information feedback and recalculation, so that the UAVs reach a better state and position.

[0052] During the simulation process, the acquisition of information is continuously carried out, enabling the UAVs to perform real-time state changes to cope with changes in the environmental situation.

[0053] Step 3.3, the state information of the UAV swarm changed through Step 3.2 is fed back to the UAV cluster simulation, and multiple constraint conditions are established in combination with the local situation information of the UAVs. The local situation information of the UAVs is the information within a certain range obtained by each UAV, and the environmental situation information obtained by the entire cluster is obtained by integrating this information. When calculating the next state, it will be affected by the positions and states of the surrounding friendly UAVs, such as maintaining a certain distance to avoid collisions, etc., so that the state of the UAVs is continuously calculated, controlled, and changed according to the acquired information.

[0054] The UAV swarm simulation module obtains real-time environmental situation information and the status information of the UAV swarm, and returns to step 3.1 to form feedback control.

[0055] Step 4: The UAV swarm simulation module outputs the simulation situation and stores it in the log file module. At the same time, the simulation situation is input into the index plugin to evaluate the simulation. The index plugin analyzes the flight status and interaction of the UAVs in the swarm simulation according to the set reward function, and calculates the score and stores it in the log file module. Different reward functions can be set in the index plugin for evaluation, and the reward function is as shown in Equation (1). Set the corresponding reward type according to different research types.

[0056] S = a1 * (number of enemy UAVs destroyed) + a2 * (enemy positions destroyed)

[0057] - a3 * (number of our own UAVs destroyed) - a4 * (remaining number of enemies) (1)

[0058] In the formula, S represents the total score, and the coefficient a represents the score value, which can be customized.

[0059] The simulation situation and score data of the swarm simulation are stored by the log file module, and the output file type is set in the task file, including txt file, csv file, bin file and chart. The simulation platform can store different scenario modes in different types of files.

[0060] Step 5: Read the data stored in the log file module for simulation playback. When the specified task type is simulation playback, the simulation function configuration is called, and only the stored data needs to be read for playback. If you want to analyze and study the simulation results, you can open the corresponding file in the folder where the record file is stored.

[0061] As Figure 3 shown, it is the component of the UAV. When the simulation platform conducts UAV swarm simulation, it is necessary to configure the UAV. The aircraft model refers to the 3D model of the UAV in the simulation. This model can be designed and made by the researcher and then imported. Some classic UAV models are also provided in the platform and can be directly used. The script file sets the underlying control logic of the UAV flight mode and is the basis for the UAV to change its flight attitude.

[0062] The other components are the information acquisition plugin, the controller plugin, the motion model plugin, the interaction plugin, the index plugin, and the custom plugin. These plugins are designed and configured in the sensor module.

[0063] The information acquisition plugin obtains the changes in the real-time simulation status information, including the remaining amounts of friendly and enemy UAVs. Each UAV obtains the status information of the nearest (the more the number, the greater the computational load, which will affect the simulation speed) enemy aircraft around it and performs preprocessing. Another plugin controls the UAVs to change their status based on this information, obtains the position coordinates of key enemy areas so that the UAVs can calculate the optimal attack route; obtains the ground and obstacle information in the simulation environment.

[0064] The interaction plugin defines the constraints and physical interactions between UAVs in the simulation. When the distance between two or more friendly and enemy UAVs is zero, a collision occurs; when the end point of the attack trajectory of a UAV intersects with an enemy UAV, it is considered a hit.

[0065] After the interaction, the state of the UAV swarm has changed. The metric plugin calculates based on the changed state information of the swarm, inputs it into the reward function, and outputs real-time score data.

[0066] As Figure 4 shown, it is a schematic diagram of information interaction during the simulation process. Before the simulation, the operation module reads the task information or log file in the task file, and then initiates the simulation task to each UAV swarm. The information acquisition plugin obtains the models and script files stored in the data module. Each UAV swarm automatically obtains the live status information through the information acquisition plugin, reads the simulation environment configuration, and the information flow is shared and transmitted among friendly UAVs, and quickly obtains the status and position of enemy UAVs, enabling the entire swarm to quickly understand the current overall situation and conduct adversarial games according to the set task type. The sensor module calculates based on the real-time information of the situation and controls the UAVs to take countermeasures and reach a new state. The real-time status information is obtained again by each UAV swarm, recalculated, and the UAVs are controlled to reach a new state. The data information generated during the entire simulation process is transmitted and stored in the log file module.

[0067] As Figure 2 shown in the set environment of the simulation experiment and the specific adversarial situation process, the embodiment provides a basic attack plan. In this embodiment, it is set that in the confrontation, the red side defends a fixed base with 100 quadrotor UAVs, and the blue side consists of 100 fixed-wing UAVs to form an attacking swarm. The set range is a square mountainous area with a length of 50 kilometers and a width of 50 kilometers. The simulation process is that the blue side UAVs destroy the red side UAVs and the base after taking off.

[0068] The situation information in the simulation environment includes two parts of entity information, namely the red side and the blue side, as shown in Table 2 specifically.

[0069] Table 2 Specific configuration information of part of the simulation environment

[0070] category information scene mountain Red side 100 quadrotor UAVs, fixed base and coordinates Blue side 100 fixed-wing UAVs and initial coordinates input and output dimensions of neural network 102；3 controller PID, LQR

[0071] The embodiments perform simulations according to the workflow steps, record the scores once for each simulation, and as the number of iterative simulations increases, the offensive behavior of the blue fixed-wing aircraft becomes more strategic, and the obtained scores gradually increase.

[0072] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A large-scale lightweight unmanned cluster test and simulation platform, characterized in that: It includes the following modules: A task file module, which is used to edit task files, configure the initial conditions of the simulation, and configure sensor plugins for the UAV swarm; A data module, which is used to store model and script files; A sensor module, which includes an information acquisition plugin, a controller plugin, and a motion model plugin. This module is used to obtain the actual situation information of the environment and the status information of the UAV swarm. The actual situation information is interactively transmitted in the sensor modules of each UAV to generate various motion states of the UAV. Specifically, The information acquisition plugin obtains the current actual situation information of the environment and the status information of the UAV swarm, converts them into sensor measurement values, and sends them to the controller plugin; The controller plugin obtains the information sent by the information acquisition plugin, performs calculation and conversion, obtains the expected state at the next moment through neural network training, and converts the control instruction and sends it to the motion model plugin; The motion model plugin adjusts the pitch angle, yaw, and thrust according to the received execution instruction to generate a new state of the UAV swarm; An operation module, which specifies the task file to be used before the simulation and initiates the simulation task, retrieves the model and script files in the data module for simulation, calls the actual situation information obtained by the sensor module during the simulation, calculates and generates control instructions according to the expected state, and sends the control instructions to the sensor module to make it generate a new motion model and update the status information of the UAV swarm; A log file module, which is used to record the simulation data of the operation module; Each module runs in parallel, that is, multiple processes are calculated simultaneously at the same time, and each process has a CPU for operation.

2. The large-scale lightweight unmanned cluster test simulation platform according to claim 1, wherein: The task file is the specific task of the simulation, including the task type and the parameters of the simulation; the task type includes an adversarial task, a game task, a collaborative task, or a playback task; the parameters of the simulation include the number of UAVs, the initial coordinates, and the map size.

3. The large-scale lightweight unmanned cluster test simulation platform according to claim 2, characterized in that: The model in the data module is a 3D aircraft model in the simulation; the script file is the underlying control logic of the UAV flight mode and is the basis for the UAV to be able to change its flight attitude.

4. The working method of the large-scale lightweight unmanned cluster test simulation platform according to claim 1, characterized in that: It includes the following steps: Step 1, edit the task file in the task file module and read the task file through the operation module; Step 2, when it is determined according to the task file that the task type is an adversarial task, a game task, or a collaborative task, call the model and script files in the data module to configure the environment and then go to Step 3; when it is determined that the task type is a playback task, call the playback function configuration and then go to Step 5; Step 3, call each module of the platform for cluster simulation. Specifically, it includes: Step 3.1, obtain information through the information acquisition plugin, and the information includes the actual situation information of the environment and the status information of the UAV swarm; Step 3.2, input the information obtained in Step 3.1 into the plugins of the sensor module, and circulate and transmit it between the sensor plugins. By changing the thrust, pitch, roll, and yaw of the UAV, the state of the UAV is changed to generate new status information of the UAV swarm; Step 3.3: Feed the status information of the new UAV swarm generated in Step 3.2 back to the information acquisition plugin in Step 3.1, and recalculate and change the status based on the new status information, enabling the UAVs to change their status in real time; Step 4: Output the cluster simulation situation in Step 3, evaluate the simulation situation, input the score situation, and store them together in the log file module for future invocation; Step 5: Read the data stored in the log file module for simulation playback.

5. The working method of the large-scale lightweight unmanned cluster test simulation platform according to claim 4, characterized in that: The working process of the operation module in Step 1 includes: Step 1.1: Build the underlying execution logic of the platform, read the settings of each plugin, set various logical judgments during the UAV simulation, and integrate the lightweight neural network tiny-dnn for flight combat learning; Step 1.2: Execute the corresponding simulation according to the settings in the task file.

6. The working method of the large-scale lightweight unmanned cluster test simulation platform according to claim 5, characterized in that: The environmental situation information m in step 3.1 includes the relative positions, speeds, relative headings, and distances of n surrounding drones, denoted as m t,i , where the subscripts t and i represent the i-th drone at time t, i = 1...n; the status information of the drone swarm includes the real-time number of friendly and enemy aircraft, flight status, and position.

7. According to the working method of the large-scale lightweight unmanned cluster test simulation platform described in claim 6, Step 3.2 is specifically: Step 3.2.1: Input the information obtained in Step 3.1 into the custom plugin, and calculate the expected status for the next step according to the algorithms and functions defined by the custom plugin; Step 3.2.2, generate control information according to the expected state and input it into the controller plugin. The controller plugin inputs the local situation information m of the i-th drone at time t t,i into the neural network, and trains to obtain a feasible flight state information a at the next moment t+1,i , and decode this state information, convert it into a control instruction of the PID controller and send it to the motion model plugin; Step 3.2.3, the motion model plugin changes the thrust, pitch, roll, and yaw of the UAV according to the control instruction to change the state, so that the UAV changes from state a t,i to reach the new state a t+1,i ; Step 3.2.4: The UAVs perform simulation confrontation through the interaction plugin, and return the live status information to Step 3.2.1 for recalculation, enabling the UAVs to change their status in real time to cope with the changes in the environmental situation.

8. The working method of the large-scale lightweight unmanned cluster test simulation platform according to claim 6 or 7, characterized in that: Step 4 is specifically: After the UAV cluster completes a simulation, output the simulation situation once and store the record in the log file module. At the same time, input the simulation situation into the index plugin to evaluate the simulation. The index plugin analyzes the flight status and interaction situation of the UAVs in the cluster simulation according to the set reward function, and calculates the score situation and stores the record in the log file module.

Images