Virtual-real combined unmanned autonomous algorithm verification method and system

Abstract

The invention discloses a virtuality and reality combined unmanned autonomous algorithm verification method and system. The method comprises the following steps: performing scenario configuration; unmanned autonomous algorithm development; an unmanned autonomous algorithm calculation result is processed, one part of the calculation result is sent to the unmanned aerial vehicle ground station to be executed by the entity node, and the other part of the calculation result is sent to the unmanned equipment in the simulation deduction platform to be executed by the virtual node; after task execution is finished, evaluation analysis is carried out, and the result is fed back to an unmanned autonomous algorithm for algorithm adjustment; the system comprises an unmanned equipment real flight platform and a simulation deduction platform. By reasonably configuring the number of the entity nodes and the number of the virtual nodes, balance between cost and precision can be achieved, the state of an equipment sensor can be simulated more truly, the correctness and feasibility of a cluster task planning algorithm can be better verified, closed-circuit dynamic tests are carried out on all development stages of an unmanned autonomous algorithm, and the development efficiency of the unmanned autonomous algorithm is improved. The problems of large workload and high cost of external field actual flight verification are effectively solved.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle autonomous algorithm verification, and in particular to a virtual-real combined unmanned aerial vehicle autonomous algorithm verification method and a system thereof. Background Art

[0002] At present, the verification of autonomous algorithms for unmanned equipment mainly includes two methods: field actual flight verification and simulation scene verification. Field actual flight verification is labor-intensive, consuming a lot of time, manpower and material resources, and the actual flight scene is single, making it difficult to build diversified scenes. Autonomous algorithms are difficult, and test failures can easily cause property losses. Simulation scene verification is difficult to support the needs of simulation and performance evaluation due to the serious gap in modeling and simulation resources for unmanned equipment. The algorithm verification single is not realistic and credible enough. There is an overall lack of effective algorithm verification and evaluation methods.

[0003] In response to the above problems, combined with the task requirements of verifying new technologies for unmanned equipment, a verification method is urgently needed to solve the difficulties in unmanned autonomous algorithm verification, the long conversion cycle, the high technical risks, and the insufficient feasibility assessment of innovative technology applications, so as to improve the efficiency and accuracy of unmanned autonomous algorithm verification. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a virtual-real combined unmanned autonomous algorithm verification method and system.

[0005] Technical solution: The virtual-real combined unmanned autonomous algorithm verification method of the present invention comprises the following steps: S1: Scenario configuration, determine the mission scenario of the unmanned equipment, generate the mission scenario using the scenario editing module, and configure the formation flight mission of the unmanned swarm equipment; S2: Unmanned autonomous algorithm development, mainly including the development of UAV autonomous navigation and positioning algorithms and autonomous identification algorithms. The developed algorithms are put into the simulation deduction platform; S3: According to the mission scenario, determine which unmanned equipment nodes need to use physical nodes for algorithm verification and which nodes need to use virtual nodes for algorithm verification, and map the scenarios designed in the scenario editing with the actual flight scenarios; S4: Processing of calculation results of unmanned autonomous algorithms. Part of the calculation results are sent to the UAV ground station and executed by the physical node, and the other part is sent to the unmanned equipment in the simulation platform and executed by the virtual node; S5: The entity node executes the result of the unmanned autonomous algorithm development. The unmanned autonomous algorithm development module sends the calculation results of the algorithm to the UAV ground station. The ground station sends the mission parameters and location information of each entity UAV to the entity equipment node. After the entity equipment node executes the task, it sends the flight control parameters back to the UAV ground station. After receiving it, the ground station returns the flight control parameters to the corresponding digital twin node in the simulation environment. S6: The virtual node executes the result of the unmanned autonomous algorithm development, and the digital twin node corresponding to the physical node drone shares the position during the deduction process to maintain the formation; S7: After the mission is completed, an evaluation and analysis is conducted and feedback is given to the unmanned autonomous algorithm for algorithm adjustment.

[0006] Furthermore, the UAV autonomous navigation and positioning algorithm of step S2 utilizes real-time positioning and mapping technology, relies on visual data from multiple machines, selects feature points and corrects the position of the inertial navigation system, and achieves positioning in navigation failure environments such as indoors; the UAV autonomous target recognition algorithm combines the characteristics of the scene, addresses the problem of a small number of target images, and adopts a method of first performing target detection and then using metric learning for target matching.

[0007] Furthermore, the result of the unmanned autonomous algorithm in step S4 is directly run in the physical node through the airborne board, and the positioning and identification results are forwarded back to the corresponding digital twin node through the UAV ground station using the virtual-reality mapping method.

[0008] Furthermore, the virtual node in step S4 adopts a virtual-reality mapping method based on digital twins, uses a machine learning algorithm, and models the entity through the collected historical data model, while simplifying the motion and control model to quickly simulate large-scale drone cluster nodes.

[0009] Furthermore, the evaluation and analysis after the execution of the task in step S7 mainly includes the evaluation of the unmanned autonomous positioning and recognition algorithm. Among them, the unmanned autonomous positioning assessment is to calculate the Euclidean distance between the position of the algorithm and the positioning of the drone itself RMSE : Assume that n The positioning result, i The actual position of the second positioning is ( ), the estimated position is ( ), , The evaluation of the drone autonomous identification algorithm is to calculate the F1 value of the target. Precision = TP / (TP+FP), Recall = TP / (TP+FN), , Among them, TP represents the number of correctly identified targets, and FP represents the number of incorrectly identified targets.

[0010] The virtual-real unmanned autonomous algorithm verification system described in the present invention includes an unmanned equipment real flight platform and a simulation and deduction platform; the unmanned equipment real flight platform includes physical equipment nodes and networking positioning system hardware entities, and the simulation and deduction platform includes digital simulation virtual nodes, virtual network simulators, simulation control systems, unmanned autonomous algorithm development modules, three-dimensional scene visualization modules and performance analysis and evaluation modules; the unmanned equipment real flight platform and the simulation platform communicate through a UAV ground station, the interior of the real flight platform communicates through a UWB networking hardware entity, and the digital twin virtual nodes communicate with each other through virtual network nodes.

[0011] Furthermore, the hardware entity of the networked positioning system includes a UWB indoor positioning system and an unmanned equipment self-networking data link module. The UWB indoor positioning system includes multiple UWB indoor positioning base stations and airborne UWB positioning tags on the unmanned equipment. The positioning of the unmanned equipment is realized by relying on the UWB indoor positioning system. Multiple UWB indoor positioning base stations are deployed to construct the flight positioning coordinate space of the unmanned aerial vehicle, and the reference coordinate system is established. The UWB positioning tags on the unmanned equipment communicate with the UWB base stations fixedly installed in the indoor space to realize the positioning of the unmanned aerial vehicle itself.

[0012] Furthermore, the three-dimensional scene visualization module receives the position data of the drone and displays the flight status of the drone in the air in three dimensions, which is used for large-scale drone cluster simulation three-dimensional display, calling up vision scenes at different angles and formulating drone vision.

[0013] Furthermore, the effectiveness analysis and evaluation module evaluates the effectiveness of the adversarial action by collecting simulation data, calculating the evaluation indicators and indicator system, and displaying the results, thereby providing support for the continuous iteration of the UAV autonomous algorithm.

[0014] Furthermore, the digital simulation virtual node includes a single system that simulates a real UAV equipment, takes the power unit control quantity as input, and takes the UAV status and sensor information as output, providing multi-rotor and fixed-wing UAV single models.

[0015] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: the present invention adopts a simulation architecture that combines virtual and real. First, a variety of scenarios are constructed through a simulation environment to improve the robustness of the unmanned autonomous target positioning and recognition algorithm; second, based on the assumed scale, some actual flight equipment is selected to verify the algorithm, and the results are synchronously updated with the virtual nodes to improve the realism and credibility of the algorithm verification; third, a balance between cost and accuracy is achieved by reasonably configuring the number of physical nodes and virtual nodes, and closed-circuit dynamic tests can be carried out on each development stage of the unmanned autonomous algorithm, effectively solving the problem of large workload and high cost of field actual flight verification. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of the present invention; Figure 2 It is a system framework diagram of the present invention. DETAILED DESCRIPTION

[0017] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0018] like Figure 1 As shown in the figure, the unmanned autonomous algorithm verification method combining virtual and real is described, which mainly includes the following steps: S1: Taking the target reconnaissance of a swarm of ten unmanned aircraft as an example, a hypothetical scenario of unmanned swarm reconnaissance is set up, including the tasks editing of unmanned aircraft, simulation model parameter design, command and control behavior modeling, program management and force deployment, etc. The scenario editing module is used to generate combat scenarios and configure the tasks of different unmanned aircraft.

[0019] S2: Develop unmanned autonomous algorithms based on mission requirements, including autonomous positioning algorithms, target recognition algorithms, and other autonomous algorithm development for drones. The model can be trained and tuned according to the constraints of the interface through the development platform of the simulation environment.

[0020] S3: Set the number of virtual and real drones. For example, a cluster of ten drones can be set up with four real drones and six virtual drones. Part of the results of the autonomous algorithm formed in step 2 are sent to real drones, and the other part is sent to virtual drones.

[0021] S4: The physical drone executes the unmanned autonomous algorithm result. The physical drone's self-organizing network terminal, onboard board and flight control are introduced in a physical way. After the physical drone performs the task, the flight control parameters are sent back to the drone ground station. The ground station sends the flight control parameters to the simulation platform, and the platform calculates the position of the physical drone and then feeds it back to the corresponding digital twin node of the simulation environment through the ground station.

[0022] S5: The virtual drone executes the results of the unmanned autonomous algorithm, shares its position with the physical drone during the simulation, and jointly executes the results of the algorithm.

[0023] S6: After the task is completed, the unmanned autonomous algorithm is evaluated and analyzed, and feedback is given to the unmanned autonomous algorithm. Steps S2 to S6 are repeated until the evaluation algorithm meets the task requirements.

[0024] like Figure 2 As shown, the virtual-real combined unmanned autonomous algorithm verification system described in the present invention includes two parts: an unmanned equipment actual flight platform and a simulation and deduction platform.

[0025] (1) Unmanned equipment flight platform: The unmanned equipment flight platform includes physical equipment nodes and networking positioning system hardware entities.

[0026] The physical equipment nodes include four drones and four unmanned vehicles. The wheelbase of the drone is ≤310mm, and the flight time of the drone is not less than 13 minutes; the unmanned vehicle has a load capacity of 5kg and a flight time of 4.5 hours.

[0027] The hardware entity of the network positioning system includes the UWB indoor positioning system and the unmanned equipment self-networking data link module.

[0028] Among them, the UWB indoor positioning system is mainly achieved by deploying four UWB indoor positioning base stations, and the drone positioning is mainly achieved by relying on the UWB indoor positioning system. Four UWB indoor positioning base stations are deployed to build the flight positioning coordinate space of the unmanned aerial vehicle and complete the establishment of the reference coordinate system. At the same time, the reference coordinate system also supports mutual conversion with the local coordinate system. At the same time, the UWB positioning tag is also carried on the UWB aircraft, which can realize the positioning of the unmanned aerial vehicle itself by communicating with the UWB base station fixedly installed in the indoor space.

[0029] Among them, the unmanned equipment self-organizing network data link module mainly deploys the ground end of the self-organizing network at the ground station of the UAV, and deploys the airborne end of the self-organizing network on the UAV. The whole relies on the indoor self-organizing network data link module to realize data link transmission and self-organizing network guarantee in outdoor environment.

[0030] (2) Simulation platform: The simulation platform includes six parts: digital simulation virtual nodes, virtual network simulator, simulation control system, unmanned autonomous algorithm development, three-dimensional real-scene visualization, and performance analysis and evaluation.

[0031] Among them, 3D visual scene visualization supports the deployment of reconnaissance and early warning forces, the deployment of attack forces, and the monitoring of battlefield situations. It supports receiving the position and other data of drones, and displays the flight status of drones in the air in 3D, meeting the needs of 3D display and scene simulation of large-scale drone clusters. It can call up scenes of different angles and formulate drone fields of view. Effectiveness analysis and evaluation supports the evaluation of the effectiveness of confrontation actions. By collecting simulation data, calculating evaluation indicators and indicator systems, and displaying results, it provides support for the research of drone intelligent confrontation actions and operations.

[0032] Among them, the digital simulation virtual model includes simulating the real UAV equipment single system, taking the power unit control quantity as input and the UAV status and sensor information as output, providing multiple types of UAV single models such as multi-rotor and fixed-wing. On the basis of highly realistic single UAV model and multi-machine cluster simulation, a complex and high-precision nonlinear model is used for each UAV, and multiple simulation control closed loops are topologically connected to obtain a highly realistic complex cluster simulation model.

Claims

1. A virtual and real combined unmanned autonomous algorithm verification method, characterized in that: The steps include: S1: Scenario configuration, determine the mission scenario of the unmanned equipment, generate the mission scenario using the scenario editing module, and configure the formation flight mission of the unmanned swarm equipment; S2: Unmanned autonomous algorithm development, mainly including the development of UAV autonomous navigation and positioning algorithms and autonomous identification algorithms. The developed algorithms are put into the simulation deduction platform; S3: According to the mission scenario, determine which unmanned equipment nodes need to use physical nodes for algorithm verification and which nodes need to use virtual nodes for algorithm verification, and map the scenarios designed in the scenario editing with the actual flight scenarios; S4: Processing of calculation results of unmanned autonomous algorithms. Part of the calculation results are sent to the UAV ground station and executed by the physical node, and the other part is sent to the unmanned equipment in the simulation platform and executed by the virtual node; S5: The entity node executes the result of the unmanned autonomous algorithm development. The unmanned autonomous algorithm development module sends the calculation results of the algorithm to the UAV ground station. The ground station sends the mission parameters and location information of each entity UAV to the entity equipment node. After the entity equipment node executes the task, it sends the flight control parameters back to the UAV ground station. After receiving it, the ground station returns the flight control parameters to the corresponding digital twin node in the simulation environment. S6: The virtual node executes the result of the unmanned autonomous algorithm development, and the digital twin node corresponding to the physical node drone shares the position during the deduction process to maintain the formation; S7: After the mission is completed, an evaluation and analysis is conducted and feedback is given to the unmanned autonomous algorithm for algorithm adjustment.

2. The virtual-real combined unmanned autonomous algorithm verification method according to claim 1 is characterized in that: The UAV autonomous navigation and positioning algorithm of step S2 utilizes real-time positioning and mapping technology, relies on visual data from multiple machines, selects feature points and corrects the position of the inertial navigation system, and obtains positioning in an indoor navigation failure environment; the UAV autonomous target recognition algorithm combines the characteristics of the scene and, in order to address the problem of a small number of target images, adopts a method of first performing target detection and then using metric learning for target matching.

3. The virtual-real combined unmanned autonomous algorithm verification method according to claim 1 is characterized in that: The result of the unmanned autonomous algorithm in step S4 is directly run in the physical node through the airborne board, and the positioning and identification results are forwarded back to the corresponding digital twin node through the drone ground station using the virtual-real mapping method.

4. The virtual-real combined unmanned autonomous algorithm verification method according to claim 1 is characterized in that: The virtual node in step S4 adopts a virtual-reality mapping method based on digital twins, uses a machine learning algorithm, and models the entity through the collected historical data model, while simplifying the motion and control model to quickly simulate large-scale drone cluster nodes.

5. The virtual-real combined unmanned autonomous algorithm verification method according to claim 1 is characterized in that: After the task execution in step S7 is completed, the evaluation and analysis mainly includes the evaluation of the unmanned autonomous positioning and recognition algorithm. Among them, the unmanned autonomous positioning assessment is to calculate the Euclidean distance between the position of the algorithm and the positioning of the drone itself RMSE : Assume that n The positioning result, i The actual position of the second positioning is ( ), the estimated position is ( ), ， The evaluation of the drone autonomous identification algorithm is to calculate the F1 value of the target. Precision = TP / (TP+FP), Recall = TP / (TP+FN), ， Among them, TP represents the number of correctly identified targets, and FP represents the number of incorrectly identified targets.

6. A virtual and real combined unmanned autonomous algorithm verification system, characterized in that: It includes an unmanned equipment actual flight platform and a simulation and deduction platform; the unmanned equipment actual flight platform includes physical equipment nodes and networking positioning system hardware entities, and the simulation and deduction platform includes digital simulation virtual nodes, virtual network simulators, simulation control systems, unmanned autonomous algorithm development modules, three-dimensional scene visualization modules and performance analysis and evaluation modules; the unmanned equipment actual flight platform and the simulation platform communicate through the UAV ground station, the actual flight platform communicates internally through UWB networking hardware entities, and the digital twin virtual nodes communicate with each other through virtual network nodes.

7. The virtual-real combined unmanned autonomous algorithm verification system according to claim 6 is characterized in that: The hardware entities of the networked positioning system include a UWB indoor positioning system and an unmanned equipment self-networking data link module. The UWB indoor positioning system includes multiple UWB indoor positioning base stations and airborne UWB positioning tags on unmanned equipment. The positioning of unmanned equipment is achieved by relying on the UWB indoor positioning system. Multiple UWB indoor positioning base stations are deployed to build a flight positioning coordinate space for the unmanned aerial vehicle, complete the establishment of a reference coordinate system, and the UWB positioning tags on the unmanned equipment communicate with UWB base stations fixedly installed in the indoor space to achieve the positioning of the unmanned aerial vehicle itself.

8. The virtual-real combined unmanned autonomous algorithm verification system according to claim 6 is characterized in that: The three-dimensional scene visualization module receives the position data of the drone and displays the flight status of the drone in the air in three dimensions. It is used for large-scale drone cluster simulation three-dimensional display, calling up vision scenes at different angles and formulating the drone vision.

9. The virtual-real combined unmanned autonomous algorithm verification system according to claim 6 is characterized in that: The effectiveness analysis and evaluation module evaluates the effectiveness of the adversarial action by collecting simulation data, calculating the evaluation index and index system, and displaying the results, thereby providing support for the continuous iteration of the UAV autonomous algorithm.

10. The virtual-real combined unmanned autonomous algorithm verification system according to claim 6, characterized in that: The digital simulation virtual node includes a single system that simulates a real UAV equipment, takes the power unit control quantity as input, and takes the state and sensor information of the UAV as output, and provides single models of multiple types of UAVs such as multi-rotor and fixed-wing.

Images