A semi-physical test system and method for robot operating system application verification

Abstract

A kind of semi-physical test system and method of robot operating system application verification, including GPS transmitting module, state monitoring module, formation similarity calculation module, autonomous path planning module and effect evaluation module.Wherein, GPS transmitting module is used to generate and emit GPS signal in indoor environment simulation;State monitoring module is used to monitor the measured stress of each unmanned aerial vehicle;Formation similarity calculation module is used for each unmanned aerial vehicle to autonomously calculate the expected target position of itself in formation;Autonomous path planning module is used for each unmanned aerial vehicle to autonomously calculate the expected control acceleration of itself, and determine expected stress;Effect evaluation module is used to exclude the hardware failure of each unmanned aerial vehicle, and the function and performance of robot operating system are evaluated.The application can reduce the frequency of large-scale unmanned aerial vehicle to field test, reduce test cost.

Description

Technical Field

[0001] This invention relates to the field of application verification testing of robot operating systems, and specifically to a semi-physical testing system and method for application verification of robot operating systems. Background Technology

[0002] The function of a robot operating system is to enable robots to receive, publish, and aggregate various sensing, control, status, and planning information. Existing robot operating systems face challenges such as: multi-domain heterogeneous resource management, where each type of robot can only perform fixed tasks in a specific environment, making it difficult to establish interconnectivity between robots and meet the needs of cross-domain collaboration; and the autonomous behavior control of complex robots, which is influenced by perception-planning-control loops, resulting in uncertain and complex robot behavior when facing complex environments and variable tasks. Therefore, the development of a swarm intelligence robot operating system is an inevitable trend.

[0003] A crucial aspect of developing a swarm intelligent robot operating system is conducting physical testing and application verification. A typical application is conducting field tests of collaborative control flight of hundreds or thousands of fixed-wing or multi-rotor drone swarms. Drones equipped with the robot operating system interconnect with each other, and the functionality and performance of the robot operating system are evaluated through the collaborative control effect of the drone swarm. However, conducting field tests of drone swarm collaborative control faces many difficulties, such as: (1) the transportation, deployment, and recovery costs of large-scale drones are very high; (2) the on-site debugging and preparation phase is very time-consuming, limiting the actual time available for testing and verifying the robot operating system; and (3) renting flight test sites is limited by objective conditions such as weather, airspace control, and transportation distance. A feasible solution to these difficulties is to simulate drone swarm flight in an indoor environment, conducting preliminary tests indoors to identify and eliminate as many drone hardware and software problems as possible, thus avoiding frequent transportation of equipment to the field for testing.

[0004] Currently, existing methods, devices, or systems for testing UAVs in indoor environments include: hardware-in-the-loop testing systems, flight control testing devices, power system test benches, and wind tunnel testing systems. For example, Patent 1 (authorization number CN103838152B) discloses a flight control testing device and control method; Patent 2 (application number CN105083588A) discloses a multi-rotor UAV performance testing platform and method, which includes a base, column, top frame, performance testing instruments, and computing processing center; Patent 3 (application number CN104990719A) discloses a UAV test bench system for testing UAV flight under high temperature, low temperature, and different wind speed conditions; Patent 4 (authorization number CN105784318B) discloses a low-speed wind tunnel model flight test system and method; Patent 5 (authorization number CN105632271B) discloses a low-speed wind tunnel model flight test ground simulation training system, which can realistically simulate the test scenario of UAV models flying unconstrained in six degrees of freedom in a low-speed wind tunnel.

[0005] While the above inventions can address software and hardware issues related to the dynamic response, thrust measurement, aging testing, and component performance testing of UAVs, providing comprehensive and scientific data support for UAV factory setup and routine debugging, and have wide-ranging, flexible, and effective applications—especially the static or dynamic wind tunnel testing, which can simulate the UAV's aerial flight environment and has been used to study the aerodynamics, flight mechanics, and flight control characteristics of aircraft—the following shortcomings exist when applying these inventions to the application verification testing and evaluation of robot operating systems:

[0006] (1) Existing testing systems or methods are generally applicable to single UAVs, which are expensive and not suitable for large-scale UAV testing, nor can they be directly used for testing related to the collaborative control of UAV swarms.

[0007] (2) Existing testing systems or methods are not applicable to robot operating systems and do not provide quantitative indicators to evaluate the functionality and performance of the operating system.

[0008] (3) Most existing testing systems or devices use separate instruments and equipment to measure and collect data. There is no overall solution or implementation case. The measurement data is inaccurate and incomplete, making it difficult to provide substantial data reference for the testing and application verification of fixed-wing or multi-rotor UAV clusters. Summary of the Invention

[0009] To address the aforementioned shortcomings in the existing technology, this invention provides a semi-physical testing system for verifying the application of a robot operating system. The testing system is set up indoors and includes: a GPS transmission module, a status monitoring module, a formation similarity calculation module, an autonomous path planning module, and an effect evaluation module.

[0010] The GPS transmitting module is used to: simulate and transmit GPS signals in an indoor environment; the indoor environment includes: time and space;

[0011] Each status monitoring module corresponds to one UAV under test. The status monitoring module is used to monitor the measured forces on each UAV under test.

[0012] The formation similarity calculation module is used for: each UAV to autonomously calculate the expected target position in the formation;

[0013] The autonomous path planning module is used for: each UAV to autonomously calculate the path from its fixed position to the desired target position using an autonomous path planning algorithm, and at the same time obtain the required desired control acceleration, and determine its own desired force based on the desired control acceleration;

[0014] The effect evaluation module is used to: use the minimum quantization error between the expected force and the measured force as a health indicator to eliminate UAV hardware faults; and use the minimum quantization error between the simulated standard force and the measured force as a performance evaluation indicator to evaluate the function and performance of the robot operating system.

[0015] The simulated standard force refers to the collective term for the thrust, lift, and torque that each UAV should exert, calculated by the effect evaluation module through simulation calculation of the path of each UAV flying from a fixed position to the desired target position.

[0016] Preferably, the status monitoring module is installed at the bottom of the support base and includes: a fixed support component, a sensor, and a data acquisition system;

[0017] The drone is fixed to the top of the fixed support component;

[0018] The sensor is disposed at the bottom of the fixed support component;

[0019] The data acquisition system is connected to the sensor;

[0020] The sensors are used to measure the lift, thrust, and torque of the drone;

[0021] The data acquisition system calculates the actual forces acting on each UAV based on the monitored lift, thrust, and torque.

[0022] Preferably, the sensor is a triaxial force and torque sensor.

[0023] Preferably, the formation similarity calculation module is specifically used for: each UAV to calculate its own expected target position in the formation based on the formation similarity index; wherein the formation similarity index adopts the Frobenius norm.

[0024] Preferably, the path planning includes: Kinodynamic A based on Lévy heuristic search. * algorithm.

[0025] Preferably, the GPS transmitting module is installed indoors, specifically used to generate indoor GPS signals through data simulation and transmit them indoors; when the UAV uses GPS positioning, the UAV can autonomously calculate its own target position coordinates in the desired formation.

[0026] Preferably, the system further includes a GPS transponder; the GPS transponder consists of a receiver and a transmitter; the receiver of the GPS transponder is placed outdoors, and the transmitter is placed indoors; the GPS transponder is used to receive outdoor GPS signals and process the outdoor GPS signals before transmitting them indoors; the GPS signal processing includes, but is not limited to: static biasing of GPS signals and dynamic changes according to a defined pattern.

[0027] Preferably, the fixed support component includes, from bottom to top: a support base, a connecting rod, a spherical joint, and a bracket base plate; the connecting rod is placed vertically, and its top end is connected to the bracket base plate through the spherical joint; the UAV is fixed to the bracket base plate; the connecting rod is disposed on the support base; the support base needs to be securely fixed to the ground.

[0028] Preferably, the autonomous path planning module is further used to: determine the control commands of a single UAV based on its desired control acceleration at the fixed position.

[0029] Preferably, the calculation of the thrust, lift, and torque that each UAV should perform includes: the UAV executing autonomous flight control commands to obtain lift and thrust, and then applying one or more of the following reaction forces to the support base: lift, thrust, and torque.

[0030] Based on the same inventive concept, this invention also provides a semi-physical testing method for verifying robot operating system applications, comprising:

[0031] The following is executed using a hardware-in-the-loop testing system for verifying robot operating system applications provided by this invention:

[0032] Step 1:

[0033] Based on the desired target position, each UAV uses an autonomous path planning algorithm to calculate its desired control acceleration from the fixed position to the desired target position, and determines its desired force based on the desired control acceleration.

[0034] The measured forces experienced by each UAV in the monitoring cluster at a fixed position;

[0035] The minimum quantitative error between the expected force and the measured force is used as a health indicator to rule out hardware failures in the UAV.

[0036] Step 2:

[0037] The effect evaluation module calculates the path of each UAV from the fixed position to the desired target position through simulation, and calculates the thrust, lift and torque that each UAV should perform, which are collectively referred to as the simulation standard forces.

[0038] The measured forces experienced by each UAV in the monitoring cluster at a fixed position;

[0039] The minimum quantization error between the simulated standard force and the measured force is used as a performance evaluation index to evaluate the function and performance of the robot operating system.

[0040] Preferably, each UAV uses an autonomous path planning algorithm to calculate its desired control acceleration from a fixed position to the desired target position, including:

[0041] Based on the indoor map, the indoor space is mapped to a configuration space, and the drone is abstracted as a point and projected onto the configuration space;

[0042] Based on GPS signal positioning simulated indoors, the coordinates of the next desired target location are determined;

[0043] In order to reach the next desired target location, the robot operating system on board the drone calls the path planning algorithm to perform autonomous calculations, obtain the desired control acceleration at the fixed position, and generate control commands.

[0044] The indoor map is an accessible map or a map that includes static obstacles.

[0045] Preferably, the path planning includes:

[0046] Each UAV autonomously calculates its desired target position within the formation based on formation similarity indices; the path planning for flying from a fixed position to the desired target position employs a Kinodynamic A algorithm based on Lévy heuristic search. * algorithm.

[0047] Preferably, the Kinodynamic A based on Lévy heuristic search... * Algorithms, including:

[0048] The total path taken by each UAV from its fixed position to the desired target position in the formation; the total path is composed of a series of local paths connected end to end.

[0049] Preferably, the total path taken by each UAV from its fixed position to the desired target position in the formation includes:

[0050] The path search is calculated based on the Lévy heuristic search method, and multiple candidate acceleration solutions are obtained for each waypoint.

[0051] Based on the dynamic principles of UAVs, the coordinates of the UAV at each track point are calculated using candidate accelerations;

[0052] Based on the coordinates of each waypoint, the shortest feasible total path can be obtained by iteratively calculating the target loss function or by using a sampling search method.

[0053] Preferably, the candidate acceleration calculation formula is:

[0054]

[0055] In the formula, and These are the random acceleration sequences for the t-th and t+1-th iterations, respectively; α is the step scaling factor; and s is the step size. This is point-to-point multiplication; H(u) is the Heaviside transition function; ε is a random number drawn from a random distribution;

[0056] Among them, random acceleration sequence The calculation formula is:

[0057]

[0058] In the formula, Let be the candidate acceleration solution for a local path after t iterations; α = o(L / 10) or α = o(L / 100); L(s,λ) is the Levy random search path; typically, the variable s takes the value 1.5; λ is a variable;

[0059] The formula for calculating L(s,λ) is:

[0060]

[0061] In the formula, s0 is the minimum step size.

[0062] Preferably, the expression for the formation similarity index is:

[0063]

[0064] In the formula, For the current formation, use the symmetric normalized Laplace operator; For the desired formation, a symmetric normalized Laplace operator; ||·|| Fis the Frobenius operator; f is the formation similarity index value.

[0065] Preferably, the minimum quantization error between the expected force and the measured force is used as a health indicator to rule out hardware faults in the UAV, including:

[0066] The desired force vector of each UAV is used as the input vector, and the minimum quantization error is used as the output to train the self-organizing feature mapping model to obtain the first self-organizing feature mapping model; wherein, the minimum quantization error is the Euclidean distance between the input vector and the best matching unit;

[0067] By substituting the measured force vector into the first pre-trained self-organizing feature mapping model, the minimum quantization error curve is obtained.

[0068] When the minimum quantization error change curve shows a significant step or jump, it indicates that one or more drones in the cluster have experienced a hardware failure; otherwise, there are no drones in the cluster that have experienced a hardware failure.

[0069] Preferably, the minimum quantization error between the simulated standard force and the measured force is used as a performance evaluation index to evaluate the function and performance of the robot operating system, including:

[0070] Using the simulated standard force vectors of each UAV as input vectors and the minimum quantization error as output, the self-organizing feature mapping model is trained to obtain a second self-organizing feature mapping model; wherein, the minimum quantization error is the Euclidean distance between the input vector and the best matching unit;

[0071] The measured force vector is then fed into a pre-trained second self-organizing feature mapping model to obtain the minimum quantization error.

[0072] The performance of the robot operating system is evaluated by determining whether the minimum quantization error exceeds a threshold.

[0073] Preferably, the self-organizing feature mapping model employs the following learning function:

[0074] W i (t+1)=W i (t)-α(t)·h ci (t)·(F i (t)-W i (t))

[0075] In the formula, t+1 and t represent two adjacent operating cycles, and W i Let F be the weight vector of the i-th neuron in the organization feature mapping model. i (t) is the input vector during training, h ci(t) represents the method of acquiring neighboring neurons in cycle t; a(t) is the learning rate, which decreases as learning progresses.

[0076] Preferably, the Euclidean distance between the input vector and the best matching unit is calculated as follows:

[0077] E t =min||F(t)-BMU|| 2

[0078] In the formula, E t Let F(t) be the minimum quantization error in the t-th cycle, F(t) be the input vector in the t-th cycle, and BMU be the best matching unit.

[0079] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0080] 1. A semi-physical testing system and method for verifying the application of a robot operating system, comprising a GPS transmission module, a state monitoring module, a formation similarity calculation module, an autonomous path planning module, and an effect evaluation module. The GPS transmission module generates and transmits GPS signals in an indoor environment; the state monitoring module enables each drone to monitor its load through force sensors mounted on its support base; the formation similarity calculation module allows each drone to autonomously calculate its desired target position within the formation; the autonomous path planning module allows each drone to autonomously calculate the path from its fixed position to the desired target position, simultaneously obtaining the required desired control acceleration and determining the desired load based on the desired control acceleration; the effect evaluation module uses the minimum quantization error between the desired load and the measured load as a health indicator to eliminate drone hardware faults, and uses the minimum quantization error between the simulated standard load and the measured load as a performance evaluation indicator to evaluate the functionality and performance of the robot operating system.

[0081] 2. This invention enables the verification or evaluation of the functionality and performance of a robot operating system in an indoor environment. As a pre-test before field testing, it can promptly resolve potential software and hardware issues of various drones, reduce the frequency of transporting large-scale drones to the field, and effectively reduce the cost of field testing.

[0082] 3. An effect evaluation module was constructed using the technical solution provided by this invention. A single UAV was tested using Kinodynamic A algorithm based on Lévy heuristic search. * The algorithm employs path planning and formation similarity metrics, which can better evaluate the effectiveness of robot operating systems for swarm control. Kinodynamic A is based on Lévy heuristic search. *The algorithm enables efficient autonomous path planning for a single drone. The formation similarity index enables decentralized, self-organizing, and collaborative control of drone swarms. Attached Figure Description

[0083] Figure 1 This is a schematic diagram of the robot operating system application verification and testing system of the present invention;

[0084] Figure 2 This is a flowchart of the test method for determining whether the hardware of each UAV is faulty according to the present invention.

[0085] Figure 3 Flowchart of a semi-physical testing method for verifying the application of the robot operating system of the present invention;

[0086] Figure 4 This invention relates to a method for evaluating the functionality and performance of a robot operating system.

[0087] Figure 5 A schematic diagram of the system platform built for the application case of Example 3;

[0088] Figure 6 The health indicator change curve of the drone hardware is used as an application case in Example 3;

[0089] Figure 7 The curves showing the performance evaluation metrics of the robot operating system in Example 3 are shown.

[0090] The components include: 1-Safety net; 2-Industrial fan; 3-Support base; 4-The fixed-wing UAV under test, of which there are 4; 5-Effect evaluation module; 6-GPS transmission module; 7-Sensor data acquisition card; and 8-Three-dimensional force and torque sensor. Detailed Implementation

[0091] This invention provides a semi-physical testing system for verifying robot operating system applications. The hardware can be functionally categorized into four types:

[0092] (1) Test workstation. Its functions include running ground station software, performance evaluation module, and status monitoring module;

[0093] (2) Unmanned Aerial Vehicles (UAVs) and their fixed support structures. Software for autonomous operation formation similarity calculation and path planning for each UAV;

[0094] (3) Monitoring sensors and data acquisition system. Output data is received by the test workstation.

[0095] (4) GPS transmitter module. Generates indoor GPS signals.

[0096] The present invention provides a method for defining the functions and performance of a robot operating system, the process of which is as follows:

[0097] Step 1: Set up a "maintain formation" task for each drone to determine if there are any hardware malfunctions in each drone.

[0098] At this point, the input to the performance evaluation module is the expected and measured forces on each UAV, and the output is a health index used to determine whether any UAV has a hardware fault. The performance evaluation module runs on a test workstation.

[0099] All drones in the cluster publish their expected force data to the effect evaluation module. In the effect evaluation module, the expected force U... 1:t The rows of the vector store the runtime cycle, and the columns store the three-dimensional force data of each UAV. The three-axis forces of each UAV are grouped together and stored sequentially. The measured force F... t The vector has the same data structure as the expected force vector.

[0100] Given an operating period of 1:t, the expected force U of all drones in the cluster. 1:t The resulting vectors, also known as positive samples, are used as training data to train the parameters of the SOM. The expression for the learning function used here is:

[0101] W i (t+1)=W i (t)-α(t)·h ci (t)·(F i (t)-W i (t))

[0102] In the formula, t+1 and t represent two adjacent operating cycles, and W i Let F be the weight vector of the i-th neuron in the SOM. i (t) is the input vector, h ci (t) represents the method for acquiring neighboring neurons in cycle t.

[0103] During testing, the input to the SOM is the measured force vector, and the output is the minimum quantization error. The minimum quantization error is defined as the Euclidean distance between the input vector F(t) and the best-fit unit (BMU), and its calculation expression is:

[0104] E t =min||F(t)-BMU|| 2

[0105] Minimum quantization error E t It is used as a health indicator to determine if there is a hardware malfunction in the drone.

[0106] When the minimum quantization error change curve shows obvious step or jump changes, it indicates that a drone malfunction has been detected.

[0107] After ruling out any faults in the drone hardware, such as the control and execution systems, the functional and performance evaluation of the robot operating system can then be conducted.

[0108] Step 2: Set up a "formation change task" for each drone in the drone swarm. This task is used to evaluate the functionality and performance of the robot operating system.

[0109] The performance evaluation module (running on the test workstation) calculates the path of each UAV from the fixed position to the desired target position through simulation, and calculates the thrust, lift and torque that each UAV should perform, collectively referred to as the simulated standard forces.

[0110] The calculation method for minimum quantization error is the same as in step 1, the difference being the training data for the training model parameters; here, simulated standard forces are used. During testing, the input to the SOM is the measured force on each UAV, and the output is the functional and performance evaluation results of the robot operating system. For example, four levels can be defined: D1-high, D2-warning, D3-medium, and D4-low. The higher the value of D, the worse the performance of the robot operating system.

[0111] To better understand this invention, the following description, in conjunction with the accompanying drawings and examples, will further illustrate the invention.

[0112] Example 1:

[0113] This invention provides a semi-physical testing system for verifying robot operating system applications. This testing system is installed indoors, such as... Figure 1 As shown, it includes: a GPS transmission module, a status monitoring module, a formation similarity calculation module, an autonomous path planning module, and a performance evaluation module;

[0114] The GPS transmitting module is used to: simulate and transmit GPS signals indoors; the indoor environment includes: time and space;

[0115] Each status monitoring module corresponds to one UAV under test. The status monitoring module is used to monitor the measured forces on each UAV under test.

[0116] The formation similarity calculation module is used for: each UAV to autonomously calculate the expected target position in the formation;

[0117] The autonomous path planning module is used for: each UAV to calculate the path from its fixed position to the desired target position using an autonomous path planning algorithm, while obtaining the required desired control acceleration, and determining the desired force based on the desired control acceleration;

[0118] The effect evaluation module is used to verify and evaluate the functionality and performance of the robot operating system.

[0119] The simulated standard force refers to the collective term for the thrust, lift, and torque that each UAV should perform, calculated by the effect evaluation module (running on the test workstation) through simulation calculation of the path of each UAV from its fixed position to the desired target position.

[0120] (1) The GPS transmitter module generates and transmits GPS signals within the indoor time and space environment. The UAV acquires indoor GPS and performs navigation and positioning. Although the UAV is not at the desired target location, it can still perform flight attitude maneuvers toward the desired target location.

[0121] To address the issue of no GPS signal indoors, this invention activates a GPS transmitter module indoors. There are two methods for generating GPS signals indoors:

[0122] 1) The first method is to deploy GPS transmitter modules indoors to simulate the generation of GPS signals and transmit them indoors;

[0123] 2) The second method is to use a GPS transponder to introduce outdoor GPS signals into the indoor environment as the base signal. A GPS transponder consists of a receiver and a transmitter; the receiver is placed outdoors to receive GPS signals, while the transmitter is placed indoors to transmit GPS signals. The GPS transponder supports one or more combinations of GPS, GLONASS, and BeiDou, enabling omnidirectional, multi-point latitude and longitude positioning.

[0124] The second method differs from existing GPS transponders in that it processes outdoor static GPS signals based on the obtained outdoor GPS signals. The processing method involves offsetting and superimposing dynamic signals.

[0125] The indoor GPS signals obtained by the above methods can be static or dynamically changing according to a defined pattern.

[0126] (2) The function of the status monitoring module is to enable each UAV to monitor the force state of the UAV through the force sensor installed on its support base.

[0127] The condition monitoring module includes fixed support components, sensors, and a data acquisition system;

[0128] The drone is fixed to the top of the fixed support component;

[0129] The sensor is located at the bottom of the fixed support component;

[0130] The sensor can be a triaxial force and torque sensor; the data acquisition system is connected to the sensor;

[0131] The sensors are used to measure the lift, thrust, and torque of the drone;

[0132] Based on the monitored lift, thrust, and torque of each tested UAV, the data acquisition system calculates the actual forces acting on each UAV.

[0133] The fixed support components, from bottom to top, include: a support base, a connecting rod, a spherical joint, and a bracket base plate, arranged vertically; the top end of the connecting rod is connected to the bracket base plate through the spherical joint; the UAV is fixed on the bracket base plate, and the connecting rod is disposed on the support base.

[0134] (3) In order to maintain the desired formation, each UAV calculates the desired target position in the formation based on the formation similarity index. Each UAV arrives at the desired target position in the formation from its fixed position, calculates the desired control acceleration using an autonomous path planning algorithm, and determines the desired force based on the desired control acceleration.

[0135] (4) The function of the effect evaluation module is to use the minimum quantitative error between the expected force and the measured force as a health indicator to eliminate hardware faults of the UAV, and to use the minimum quantitative error between the simulated standard force and the measured force as a performance evaluation indicator to evaluate the function and performance of the robot operating system.

[0136] This invention does not need to consider communication delay errors. The GPS transmission module generates indoor static or slowly varying GPS signals, so communication delay is not a concern. The status monitoring module uses a common three-axis force and torque data acquisition system; with current technology, the delay error is very small and does not affect data acquisition. The UAV achieves autonomous flight within the framework of the perception, planning, and control software system; with current technology, the communication delay is in the tens or hundreds of milliseconds, which does not affect the implementation of this invention.

[0137] Example 2:

[0138] Based on the same inventive concept, this invention also provides a semi-physical testing method for verifying robot operating system applications, comprising:

[0139] The following is executed using a hardware-in-the-loop testing system for verifying robot operating system applications provided by this invention:

[0140] Step 1: Assign each drone the task of "maintaining formation" to determine if there is a hardware malfunction in each drone.

[0141] Step 2: Set up a "formation change task" for each drone in the drone swarm. This task is used to evaluate the functionality and performance of the robot operating system.

[0142] Step 1, specifically as follows Figure 2 As shown, it includes:

[0143] S1-1. Based on the desired target position, each UAV uses an autonomous path planning algorithm to calculate its desired control acceleration from the fixed position to the desired target position, and determines the desired force based on the desired control acceleration.

[0144] S1-2, Measured forces on each UAV in the monitoring cluster at a fixed position;

[0145] S1-3. The minimum quantization error between the expected force and the measured force is used as a health indicator to eliminate hardware faults in the UAV.

[0146] Step 2, specifically as follows: Figure 3 As shown, it includes:

[0147] S2-1, Effect Evaluation Module (running on the test workstation) calculates the path of each UAV from its fixed position to the desired target position through simulation, and calculates the thrust, lift and torque that each UAV should perform, collectively referred to as the simulation standard force.

[0148] S2-2, Measured forces on each UAV in the monitoring cluster at a fixed position;

[0149] S2-3. The minimum quantization error between the simulated standard force and the measured force is used as a performance evaluation index to evaluate the function and performance of the robot operating system.

[0150] Specifically, step S1-1 includes:

[0151] Based on the indoor map, the indoor space is mapped to a configuration space, and the drone is abstracted as a point and projected onto the configuration space;

[0152] Based on GPS signal positioning simulated indoors, the coordinates of the next desired target location are determined;

[0153] In order to reach the next desired target location, each drone's onboard robot operating system calls a path planning algorithm to perform calculations, obtains the desired control acceleration at the fixed position, and generates control commands.

[0154] The indoor map is an accessible map, or a map that includes static obstacles.

[0155] The path planning algorithm in this embodiment includes:

[0156] Each UAV is based on Lévy's heuristic search for Kinodynamic A * The algorithm autonomously calculates the total path from the fixed position to the desired target position in the formation;

[0157] Here, the total path is composed of many local paths connected end to end, and its specific calculation process is as follows:

[0158] The path search is calculated based on the Lévy heuristic search method, and multiple candidate acceleration solutions are obtained for each waypoint.

[0159] Based on the dynamic principles of UAVs, the coordinates of the UAV at each waypoint are calculated using candidate accelerations;

[0160] Based on the coordinates of each candidate waypoint, the shortest feasible total path can be obtained by iterative calculation using the objective loss function or by sampling search.

[0161] The algorithm used in this invention will be described in detail below:

[0162] (I) To improve the efficiency of UAV path planning, this invention is based on Kinodynamic A, inspired by Lévy search. * Algorithm. This method differs from existing A algorithms. * Compared to path planning algorithms, it has a shorter path search time, making it suitable for fast-flying fixed-wing drones.

[0163] Kinodynamic A based on Lévy heuristic search * Algorithm, abbreviated as LIK-A * This algorithm applies Lévy random search to Kinodynamic A. * The algorithm's path search process. The implementation steps are as follows:

[0164] Step 1): Map the known flight space of the drone formation to a configuration space C.

[0165] Step 2): Randomly generate initial candidate acceleration solutions. n is the number of waypoints. The acceleration in three directions can be decomposed into: Adopting the classic A * The algorithm uses the same minimum loss function to optimize the solution. And reserve it for the next iteration.

[0166] Step 3): Update the solution using formulas (1) and (2) and calculate the loss function value. If the updated solution is better than the previous iteration solution, then use it as the current solution.

[0167] The mathematical expression for solving the acceleration solution is:

[0168]

[0169] In the formula, and These are the random acceleration sequences for the t-th and t+1-th iterations, respectively; α is the step scaling factor; and s is the step size. This is point-to-point multiplication; H(u) is the Heaviside transition function; ε is a random number drawn from a random distribution;

[0170]

[0171]

[0172] In the formula, Let be a candidate acceleration solution for a local path in the t-th iteration; α = o(L / 10) or α = o(L / 100); L(s,λ) is the Levy random search path; typically, the variable s takes the value of 1.5; λ is a variable; s0 is the minimum step size.

[0173] Step 4): Generate normally distributed random numbers r∈(0,1), and set the threshold p. a = [0,1] for comparison. If r > p a If so, some candidate solutions will be updated; otherwise, the original solutions will be retained.

[0174] Step 5): Calculate the current loss function value and compare it with the loss function value of the solution from the previous step. If the loss function value is smaller, the solution with the smaller loss function value is taken as the current global optimal solution.

[0175] Step 6): Determine if the termination condition is met. If If the convergence criterion is met, the algorithm stops. Otherwise, return to step 2) to perform the calculation.

[0176] (ii) Formation similarity evaluation in the effect evaluation module.

[0177] The formation similarity index enables self-organized and collaborative control of the drone swarm, with each drone autonomously calculating its desired target position within the formation. To maintain formation flight and successfully avoid obstacles, each drone in the formation utilizes an onboard computer-based robotic operating system to schedule and run the formation control algorithm for adaptive control.

[0178] An undirected graph G = (ν, ε) is used to represent the formation shape of N drones, where ν = {1, 2, ..., N} is a set of vertices. It is an edge set. Each edge in graph G is associated, with non-negative numbers as weights. In an undirected graph G, vertex i represents the position vector of the i-th drone. The edge e of the undirected graph... ij This represents the geometric distance between drones i and j.

[0179] For an undirected graph matrix, the Laplacian operator contains information about the graph structure. The adjacency matrix A and membership matrix D of an undirected graph G have the following Laplacian matrix:

[0180] L = DA (4)

[0181] The symmetric normalized Laplacian matrix of an undirected graph G is defined as:

[0182]

[0183] In the formula, I∈R N×N It is an identity matrix.

[0184] The Frobenius norm is used to characterize the similarity index of UAV formations, and the expression is:

[0185]

[0186] In the formula, For the current formation, use the symmetric normalized Laplace operator; For the desired formation, a symmetric normalized Laplace operator; ||·|| F is the Frobenius operator; f is the formation similarity index value.

[0187] (III) Evaluation of the functions and performance of the robot operating system.

[0188] The hardware of the semi-physical test system for verifying robot operating system applications can be functionally categorized into four types:

[0189] (1) Test workstation. Its functions include running ground station software, performance evaluation module, and status monitoring module;

[0190] (2) Unmanned Aerial Vehicles (UAVs) and their fixed support structures. This includes modules for calculating the similarity of autonomous UAV formations and for autonomous path planning.

[0191] (3) Monitoring sensors and data acquisition system. Output data is received by the test workstation.

[0192] (4) GPS transmitter module. Generates indoor GPS signals.

[0193] like Figure 4 The following is a method for evaluating the functionality and performance of a robot operating system.

[0194] In steps S1-3, the minimum quantization error between the expected force and the measured force is used as a health indicator to eliminate hardware faults in the UAV, including:

[0195] Assign a "maintain formation" task to each drone in the drone swarm to determine if there is a hardware malfunction in each drone.

[0196] At this point, the input to the effect evaluation module software is the expected force and the measured force of each UAV, and the output is a health index, which is used to troubleshoot hardware faults in each UAV.

[0197] All drones in the cluster publish their expected force data to the effect evaluation module. In the effect evaluation module, the expected force U... 1:t The rows of the vector store the runtime cycle, and the columns store the three-dimensional force data of each UAV. The three-axis forces of each UAV are grouped together and stored sequentially. The measured force F... t The vector has the same data structure as the expected force vector.

[0198] Given an operating period of 1:t, the expected force U of all drones in the cluster. 1:t The constituent vectors are used as positive samples to train the SOM parameters. The expression for the learning function used here is:

[0199] W i (t+1)=W i (t)-α(t)·h ci (t)·(F i (t)-W i (t)) (7)

[0200] In the formula, t+1 and t represent two adjacent operating cycles, and W i Let F be the weight vector of the i-th neuron in the SOM. i (t) is the input vector, h ci (t) represents the method for acquiring neighboring neurons in cycle t.

[0201] During testing, the input to the SOM is the measured force vector, and the output is the minimum quantization error. The minimum quantization error is defined as the Euclidean distance between the input vector F(t) and the best-fit unit (BMU), and its calculation expression is:

[0202] E t =min||F(t)-BMU|| 2 (8)

[0203] Minimum quantization error E t It is used as a health indicator to determine whether there are any faults in the hardware of each drone.

[0204] When the minimum quantization error change curve shows obvious step or jump changes, it indicates that a drone malfunction has been detected.

[0205] Assuming the drone control and execution systems are fault-free, the functionality and performance of the robot operating system will be evaluated.

[0206] In steps S2-3: the minimum quantization error between the simulated standard force and the measured force is used as a performance evaluation index to evaluate the function and performance of the robot operating system, including:

[0207] Set up a "formation change task" for each drone in the drone swarm. This task is used to evaluate the functionality and performance of the robot operating system.

[0208] The performance evaluation module (running on the test workstation) simulates the flight path of each UAV from the fixed position to the desired target position, and calculates the thrust, lift, and torque that each UAV should exert, collectively referred to as the simulated standard forces. The method for calculating the minimum quantization error is the same as in steps S1-3, the difference being that the positive samples used to train the SOM parameters are different; here, the simulated standard forces are used.

[0209] The input to the SOM (System of Objectives) in the performance evaluation module is the measured force experienced by each UAV, and the output is the result of evaluating the functionality and performance of the robot operating system using performance evaluation indicators. Four levels are defined here: D1 - High, D2 - Warning, D3 - Medium, and D4 - Low. The higher the value of D, the worse the performance of the robot operating system.

[0210] Example 3:

[0211] This embodiment uses specific examples to introduce the hardware-in-the-loop testing system and method for verifying the application of the robot operating system provided by the invention.

[0212] This case study will not discuss or describe code errors that are easily identifiable by the naked eye, such as drone hardware malfunctions or robot operating system software interface errors.

[0213] (I) Construction of an indoor semi-physical testing system platform.

[0214] like Figure 5 As shown, a test system was built in an indoor rectangular area, with the area's length, width, and height set to 8m, 7m, and 2.4m, respectively. The test system consists of a GPS transmission module, a status monitoring module, a formation similarity calculation module, an autonomous path planning module, and a performance evaluation module.

[0215] This case study involves four fixed-wing unmanned aerial vehicles (UAVs) adapted to a robot operating system, designated UAV_1, UAV_2, UAV_3, and UAV_4. The main technical specifications of these UAVs are: wingspan 1.95m, fuselage length 1.287m, empty weight (without battery) less than 1kg, maximum takeoff weight 5kg, flight speed 8-20m / s, and idle operating time up to 80 minutes. Each UAV's main components include: motors, ESCs, propellers, batteries, remote controller, PIXHAWK flight controller, and onboard sensors. Onboard sensors include an accelerometer / gyroscope, magnetometer, barometer, and airspeed differential pressure sensor. The remote controller is 2.4GHz, uses 8-channel software, and has a remote control range greater than 500m. Each UAV is equipped with a P900 communication link for communication between them. Each UAV can package its own attitude information into MAVLink protocol data and transmit it to other UAVs via the P900. Each drone acquires the pose information of other drones and calculates its desired target position within the formation using formation similarity metrics. Although each drone aims to fly to its desired target position, it is limited to performing only the actions required for flight due to its fixed position.

[0216] This solution involves the modification, upgrade, and software adaptation of all fixed-wing UAVs. Key measures include: modifying the internal structure of the UAVs; redesigning and adjusting the overall center of gravity; redesigning the internal wiring layout; mounting the onboard NVIDIA Jetson TX2 computer, custom communication board, and vision sensors on a machined wooden base; and developing software for the UAVs, including communication protocol adaptation, data acquisition, and data distribution. Specific measures are as follows:

[0217] (1) In terms of structure, the reserved installation positions for key components are upgraded, such as the onboard computer NVIDIA Jetson TX2, industrial-grade communication board, communication converter, branch voltage reduction module, landing gear, wiring layout, etc.

[0218] (2) In terms of hardware, it is equipped with components such as an airborne computer, an industrial-grade communication board, a communication adapter, a branch voltage reduction module, and a signal conditioning module;

[0219] (3) In terms of software, establish a communication adaptation protocol toolchain between the flight controller and the airborne computer, and install communication data protocol conversion software and its related dependent toolchain.

[0220] Fixed-wing drones require a frontal windward position to achieve adequate buoyancy for takeoff, cruising, and landing. Otherwise, a sudden drop in buoyancy when wind speed equals drone speed can lead to stall and crash. However, indoor spaces are generally limited and do not meet the flight space requirements for fixed-wing drones. To solve this problem, the drone is securely fixed to a support base, and an industrial fan generates wind. When the generated wind speed is sufficiently high, the drone gains buoyancy when facing the wind, enabling it to assume the desired flight path and perform maneuvers.

[0221] The indoor semi-physical testing system platform also includes a power supply system, which provides external power to the various components within the indoor testing system. Each module can also be powered independently.

[0222] The drone uses a support base structure with three-axis force and torque sensors installed at the bottom. The main function of these sensors is to collect the three-axis forces acting on the drone and build a data acquisition system.

[0223] The purpose of the industrial fan is to provide a positive windward surface for the fixed-wing drone, enabling it to gain buoyancy. The selected industrial fan has an output wind speed of 30 m / s, which meets the minimum flight speed requirement of the fixed-wing drone.

[0224] To address the lack of GPS signal indoors, GPS transmitter modules are installed or deployed indoors. A ground station computer simulates and generates GPS signals, which are then transmitted indoors via the transmitter modules. Indoor GPS signals are static.

[0225] (II) Test Task

[0226] The drone swarm is configured for diamond formation flight. Each drone can package its position and velocity into MAVLINK data via a communication link and send it to other drones through the flight controller's TELEM2 interface. The other drones then parse the formation information upon receiving it.

[0227] Each UAV can autonomously calculate its desired target position in the formation based on the formation similarity index, and use an autonomous path planning algorithm to calculate the path from its fixed position to the desired target position. At the same time, it can obtain the required desired control acceleration. Each UAV determines the desired force based on the desired control acceleration. The minimum quantization error between the desired force and the measured force is used as a health index to eliminate UAV hardware failures.

[0228] The performance evaluation module (running on the test workstation) simulates the path of each UAV from its fixed position to the desired target position, and calculates the thrust, lift and torque that each UAV should perform, collectively referred to as the simulated standard forces. The minimum quantization error between the simulated standard forces and the measured forces is used as a performance evaluation index to evaluate the function and performance of the robot operating system.

[0229] (III) Test Procedure

[0230] 1. Preparation stage.

[0231] (1) Set up the test system and power on all equipment to a stable standby state;

[0232] (2) Power on the UAVs in sequence according to their numbers UAV_1 to UAV_4. The interval between powering on each UAV should be greater than 30 seconds. Wait for all UAVs to enter a stable state.

[0233] (3) After each UAV enters GPS stable state, the ground station (QGroundControl software is used in this case) will announce the simulated absolute altitude of each UAV via voice. When the minimum quantization error of the absolute altitude of all UAVs does not exceed 3 meters, it is a normal state. The smaller the minimum quantization error of the initial altitude of all UAVs, the better.

[0234] 2. Testing phase.

[0235] (1) Three minutes after the last drone is powered on, the status data of all drones can be seen on the ground station software. Now it is necessary to check the data status of the drones on the ground station to ensure that all drones are in normal status.

[0236] (2) Power-driven unlocking and locking test. Use the remote controller to perform unlocking and locking tests on each drone one by one. All drones should be able to unlock and lock normally;

[0237] (3) Test the control surface attitude feedback and attitude control of each UAV one by one to ensure that the control surface response is normal under manual and stabilized conditions, and confirm that all UAVs can switch modes normally.

[0238] 3. Task settings.

[0239] (1) Each drone is equipped with a remote controller for the highest priority protection. The remote controller's flight mode can be set to mission.

[0240] (2) Check the flight path and safe return-to-home altitude for all UAVs in mission mode. Note that all UAVs should allow sufficient altitude difference when in mission mode. Altitude is a virtual value received by the UAV's GPS. The flight path for UAV_1 is set to 55 meters, and both the return-to-home altitude and flight altitude for UAV_1 are set to 55 meters. The mission waypoint altitude for UAV_2 is set to 60 meters, and both the return-to-home altitude and loiter altitude are set to 60 meters. The mission waypoint altitude for UAV_3 is set to 65 meters, and both the return-to-home altitude and loiter altitude are set to 65 meters. The mission waypoint altitude for UAV_4 is set to 70 meters, and both the return-to-home altitude and loiter altitude are set to 70 meters.

[0241] 4. During the formation flight phase, conduct a functional and performance evaluation of the robot's operating system.

[0242] After UAVs 2 through 4 are added to the mission in sequence, they enter the formation flight phase after the takeoff phase. The formation is determined by the FORM_TYPE parameter of UAV 1 and is set to a diamond formation. Each UAV can autonomously calculate its desired target position within the formation and use an autonomous path planning algorithm to calculate the path from its fixed position to the desired target position. Simultaneously, it obtains the required desired control acceleration and determines the desired force based on this acceleration. The performance evaluation module uses the minimum quantization error between the desired force and the measured force as a performance evaluation index to evaluate the functionality and performance of the robot operating system.

[0243] (1) Check whether there is a hardware failure in each drone.

[0244] Each UAV is assigned the task of "maintaining formation". The performance evaluation module takes as input the expected and measured forces on each UAV and outputs the result excluding hardware malfunctions. The SOM takes as input the measured force vector and outputs the minimum quantization error.

[0245] When the minimum quantization error change curve shows obvious step or jump changes, it indicates that one or more drones in the cluster have malfunctioned.

[0246] Figure 6 To determine whether there is a fault in the drone hardware, the health indicator change curve uses the minimum quantization error as the health indicator. If there is no large step change and the change range is very small, it indicates that the drone hardware, such as the flight control and actuators, is not faulty.

[0247] Assuming the drone control and execution systems are fault-free, the functionality and performance of the robot operating system will be evaluated.

[0248] (2) Evaluate the functional integrity of the robot operating system.

[0249] If the software is incomplete, there will be code error logs on the onboard computer interface of each drone, which will be highlighted and easily identified by the developers.

[0250] (3) Evaluate the performance level of the robot operating system.

[0251] Each UAV is assigned a "formation change task". The performance evaluation module (running on the test workstation) calculates the thrust, lift, and torque that each UAV should generate by flying from its fixed position to the desired target position. These are collectively referred to as simulated standard forces. The method for calculating the minimum quantization error is the same as in step 1, except that the positive samples used to train the SOM parameters are different; simulated standard forces are used here. The input to the SOM in the performance evaluation module is the measured force of each UAV, and the output is the result of evaluating the function and performance of the robot operating system using performance evaluation indicators. In this case, four levels are defined as: D1 - High, D2 - Warning, D3 - Medium, and D4 - Low. The higher the level value, the worse the performance of the robot operating system. Here, the values ​​are D1 = 0.25, D2 = 0.5, D3 = 0.75, and D4 = 1.

[0252] The performance evaluation module displays the real-time operating status of the robot's operating system. Assuming the drone is in good condition, it determines whether the operating system software performance meets the standards. Minimum quantization error, as a performance evaluation metric, is used to assess errors in the robot operating system's application layer algorithms, such as calculation errors and unstable communication between drones—errors that are not easily detected by the naked eye. A larger minimum quantization error value indicates a software error in the robot's operating system, requiring code debugging and repair until the minimum quantization error between the expected and measured forces is within the allowable range before the application test can be passed.

[0253] The output of the path planning algorithm was artificially modified to cause the measured forces on the UAV to deviate from the simulated standard forces, such as... Figure 7 The performance evaluation index change curve of the robot operating system is shown. The value of the performance evaluation index indicates that the performance level of the robot operating system is D3. At this time, it does not meet the requirements of the field test and the robot operating system program needs to be further debugged and optimized.

[0254] 5: Mission completion phase.

[0255] After the formation mission is completed, UAV_4 will exit the formation first, followed by UAV_3 and UAV_2, and UAV_1 will be the last to stop.

Claims

1. A semi-physical testing system for verifying robot operating system applications, characterized in that, The testing system is set up indoors and includes: a GPS transmission module, a status monitoring module, a formation similarity calculation module, an autonomous path planning module, and a performance evaluation module; The GPS transmitting module is used to: simulate and transmit GPS signals in an indoor environment; the indoor environment includes: time and space; Each status monitoring module corresponds to one UAV under test. The status monitoring module is used to monitor the measured forces on each UAV under test. The formation similarity calculation module is used for: each UAV to autonomously calculate the expected target position in the formation; The autonomous path planning module is used for: each UAV to autonomously calculate the path from its fixed position to the desired target position using an autonomous path planning algorithm, while obtaining the required desired control acceleration, and determining the desired force based on the desired control acceleration; The effect evaluation module is used to: use the minimum quantization error between the expected force and the measured force as a health indicator to eliminate UAV hardware faults; and use the minimum quantization error between the simulated standard force and the measured force as a performance evaluation indicator to evaluate the function and performance of the robot operating system. The simulated standard force refers to the collective term for the thrust, lift, and torque that each UAV should perform, calculated by the effect evaluation module through simulation calculation of the path of each UAV flying from the fixed position to the desired target position. The minimum quantization error between the expected force and the measured force is used as a health indicator to eliminate UAV hardware faults, including: The desired force vector of each UAV is used as the input vector, and the minimum quantization error is used as the output to train the self-organizing feature mapping model, thus obtaining the first self-organizing feature mapping model; wherein, the minimum quantization error is the Euclidean distance between the input vector and the best matching unit; By substituting the measured force vector into the first pre-trained self-organizing feature mapping model, the minimum quantization error curve is obtained. When the minimum quantization error change curve shows obvious step or jump changes, it indicates that one or more drones in the cluster have experienced a hardware failure; otherwise, there are no drones in the cluster that have experienced a hardware failure. The minimum quantization error between the simulated standard force and the measured force is used as a performance evaluation index to evaluate the function and performance of the robot operating system, including: Using the simulated standard force vectors of each UAV as input vectors and the minimum quantization error as output, the self-organizing feature mapping model is trained to obtain a second self-organizing feature mapping model; wherein, the minimum quantization error is the Euclidean distance between the input vector and the best matching unit; The measured force vector is then fed into a pre-trained second self-organizing feature mapping model to obtain the minimum quantization error. The performance of the robot operating system is evaluated by determining whether the minimum quantization error exceeds a set threshold range.

2. The system as described in claim 1, characterized in that, The status monitoring module is installed at the bottom of the support base and includes: a fixed support component, a sensor, and a data acquisition system; The drone is fixed to the top of the fixed support component; The sensor is disposed at the bottom of the fixed support component; The data acquisition system is connected to the sensor; The sensors are used to measure the lift, thrust, and torque of the drone; The data acquisition system calculates the actual forces acting on each UAV based on the monitored lift, thrust, and torque. The sensor is a triaxial force and torque sensor.

3. The system as described in claim 1, characterized in that, The formation similarity calculation module is specifically used for: each UAV to calculate its expected target position in the formation based on the formation similarity index; wherein, the formation similarity index adopts the Frobenius norm.

4. The system as described in claim 1, characterized in that, The path planning includes: the Kinodynamic A* algorithm based on Lévy heuristic search; The GPS transmitter module is installed indoors and is specifically used to generate indoor GPS signals through data simulation and transmit them indoors; when the UAV uses GPS positioning, the UAV can autonomously calculate the target position coordinates in the desired formation. The system also includes a GPS transponder; the GPS transponder consists of a receiver and a transmitter; the receiver of the GPS transponder is placed outdoors, and the transmitter is placed indoors; the GPS transponder is used to receive outdoor GPS signals and process the outdoor GPS signals before transmitting them indoors; the GPS signal processing includes, but is not limited to: static biasing of GPS signals and dynamic changes according to a defined pattern.

5. The system as described in claim 2, characterized in that, The fixed support component, from bottom to top, includes: a support base, a connecting rod, a spherical joint, and a bracket base plate; the connecting rod is placed vertically, and its top end is connected to the bracket base plate through the spherical joint; the drone is fixed to the bracket base plate; the connecting rod is mounted on the support base; the support base needs to be securely fixed to the ground.

6. The system as described in claim 1, characterized in that, The autonomous path planning module is also used to: determine autonomous flight control commands for a single UAV based on its desired control acceleration at a fixed position; The UAV executes autonomous flight control commands to obtain lift and thrust, and then applies one or more of the following reaction forces to the support base: lift, thrust, and torque.

7. A semi-physical testing method for verifying robot operating system applications, characterized in that, include: Perform the following using the hardware-in-the-loop test system for robot operating system application verification provided by any one of claims 1 to 6: Step 1: Based on the desired target position, each UAV uses an autonomous path planning algorithm to calculate its desired control acceleration from the fixed position to the desired target position, and determines the desired force based on the desired control acceleration; The measured forces experienced by each UAV in the monitoring cluster at a fixed position; The minimum quantitative error between the expected force and the measured force is used as a health indicator to rule out hardware failures in the UAV. Step 2: The effect evaluation module calculates the path of each UAV from the fixed position to the desired target position through simulation, and calculates the thrust, lift and torque that each UAV should perform, which are collectively referred to as the simulation standard forces. The measured forces experienced by each UAV in the monitoring cluster at a fixed position; The minimum quantization error between the simulated standard force and the measured force is used as a performance evaluation index to evaluate the function and performance of the robot operating system.

8. The method as described in claim 7, characterized in that, Each UAV uses an autonomous path planning algorithm to calculate its desired control acceleration from a fixed position to the desired target position, including: Based on the indoor map, the indoor space is mapped to a configuration space, and the drone is abstracted as a point and projected onto the configuration space; Based on GPS signal positioning simulated indoors, the coordinates of the next desired target location are determined; In order to reach the next desired target location, the robot operating system on the drone calls the path planning algorithm to perform calculations, obtains the desired control acceleration at the fixed position, and generates control commands. The indoor map is an accessible map, or a map that includes static obstacles.

9. The method as described in claim 7, characterized in that, The path planning includes: Each UAV autonomously calculates its desired target position within the formation based on formation similarity indices; the path planning for flying from a fixed position to the desired target position employs a Kinodynamic A algorithm based on Lévy heuristic search. * algorithm; The Kinodynamic A based on Lévy heuristic search * Algorithms, including: The total path taken by each UAV from its fixed position to the desired target position in the formation; the total path is composed of a series of local paths connected end to end; The total path taken by each UAV from its fixed position to the desired target position in the formation includes: Map the known flight space of the drone formation to a configuration space C; Randomly generate initial candidate acceleration solutions. n is the number of waypoints. These are accelerations in three directions; The path search is calculated based on the Lévy heuristic search method, and multiple candidate acceleration solutions are obtained for each waypoint. Based on the dynamic principles of UAVs, the coordinates of the UAV at each waypoint are calculated using candidate accelerations; Based on the coordinates of each waypoint, the shortest feasible total path can be obtained by iteratively calculating the target loss function or by using a sampling search method. The formula for calculating the candidate acceleration is: In the formula, and These are the random acceleration sequences for the t-th and t+1-th iterations, respectively. s is the step scaling factor; s is the step size; This is point-to-point multiplication; Here is the Heaviside transition function; A random number drawn from a random distribution; Among them, random acceleration sequence The calculation formula is: In the formula, This is a candidate acceleration solution for a local path after t iterations; or ; For Levy's random search path; typically, variables The value is 1.5; As a variable; in, The formula for calculation is: In the formula, Minimum step size; The expression for the formation similarity index is: In the formula, For the current formation, use the symmetric normalized Laplace operator; For the desired formation, a symmetric normalized Laplace operator; It is the Frobenius norm; This represents the formation similarity index value.

10. The method as described in claim 7, characterized in that, The self-organizing feature mapping model uses the following learning function: In the formula, t +1 and t These are two adjacent operating cycles. W i For the self-organizing feature mapping model, the first i The weight vector of each neuron. F i ( t () is the input vector during training. h ci ( t) for t The method of acquiring neighboring neurons during runtime; This is the learning rate, which decreases as learning progresses; The Euclidean distance between the input vector and the best matching unit is calculated as follows: In the formula, E t The minimum quantization error at the t-th running cycle. F ( t ) is the input vector at the t-th running cycle, and BMU is the best matching unit.

Images