Low-altitude large-scale model-driven intelligent examination and evaluation system for drone pilots

Abstract

This invention relates to the field of UAV testing technology and proposes a low-altitude large-scale model-driven intelligent UAV pilot testing and evaluation system, comprising: a low-altitude environment processing module for collecting data on complex low-altitude scenarios and constructing a 3D environment model adapted to low-altitude characteristics; a low-altitude large-scale model inference module, trained and optimized based on a low-altitude UAV pilot-specific dataset, receiving environmental features and test task instructions, and outputting pilot decision references and path planning schemes adapted to low-altitude airflow and obstacle distribution; and a test task generation module for generating differentiated low-altitude flight tasks by combining low-altitude control rules and test levels, adjusting task parameters in real time to adapt to environmental changes, and ensuring that tasks comply with low-altitude flight safety regulations. By constructing a 3D grid map and quantifying the low-altitude environmental cost field, the system accurately identifies low-altitude-specific obstacles and no-fly zones, dynamically adapts test tasks to low-altitude control rules, and effectively avoids low-altitude flight safety risks.

Description

Technical Field

[0001] This invention relates to the field of drone testing technology, and in particular to a low-altitude, large-model-driven intelligent testing and evaluation system for drone piloting. Background Technology

[0002] With the rapid development of the low-altitude economy, the application scenarios of drones are becoming increasingly widespread, leading to a surge in demand for skill assessments of drone pilots. Current drone pilot testing and evaluation systems largely rely on traditional GPS positioning, two-dimensional electronic maps, and preset flight path planning algorithms. Some systems have introduced basic visual obstacle avoidance modules, using data such as drone flight attitude and flight path deviation to complete basic assessment scoring. These systems can meet basic testing requirements in open airspace or standardized scenarios, and have also promoted the initial standardization of drone pilot assessments.

[0003] However, existing technologies have significant shortcomings when adapting to complex low-altitude scenarios: On the one hand, traditional path planning algorithms are not optimized for the characteristics of dense obstacles and frequent airflow disturbances in low-altitude environments. Relying solely on two-dimensional maps and fixed altitude constraints, they struggle to generate flight paths that balance safety and smoothness. Furthermore, their node selection efficiency is low, failing to meet the needs of dynamic obstacle avoidance and real-time path adjustment in low-altitude scenarios. On the other hand, existing evaluation systems are too simplistic, focusing primarily on basic indicators such as flight path deviation. They fail to emphasize the core weight of obstacle avoidance safety and do not incorporate the optimal decision-making scheme of a large low-altitude model to calculate operational deviation. Consequently, the evaluation results cannot accurately reflect the candidate's actual piloting ability in complex low-altitude environments. Additionally, the lack of a three-dimensional real-time monitoring mechanism for low-altitude no-fly zones and dynamic obstacles, along with a critical violation judgment mechanism, makes it difficult to guarantee the safety and compliance of the examination process. Summary of the Invention

[0004] In view of the problems existing in the prior art, the purpose of this invention is to provide a low-altitude, large-model-driven intelligent examination and evaluation system for unmanned aerial vehicle (UAV) piloting, so as to solve the problems mentioned in the background art.

[0005] To solve the above problems, the present invention adopts the following technical solution: a low-altitude large-model-driven intelligent examination and evaluation system for unmanned aerial vehicle (UAV) piloting, comprising:

[0006] The low-altitude environment processing module is used to collect data on complex low-altitude scenes, construct a 3D environment model adapted to low-altitude characteristics, identify low-altitude-specific obstacles and no-fly zones, and output an environmental feature dataset.

[0007] The low-altitude large model inference module is trained and optimized based on a dedicated dataset for low-altitude UAV piloting. It receives environmental features and examination task instructions, and outputs driving decision references and path planning schemes that are adapted to low-altitude airflow and obstacle distribution.

[0008] The exam task generation module is used to generate differentiated low-altitude flight tasks by combining low-altitude control rules and exam levels, and to adjust task parameters in real time to adapt to environmental changes, ensuring that the tasks comply with low-altitude flight safety regulations.

[0009] The UAV status real-time monitoring module is used to collect multi-dimensional status data of UAV position, attitude, speed, and energy consumption during low-altitude flight, simultaneously monitor the stability of the low-altitude communication link, and output status monitoring data stream;

[0010] The multi-dimensional intelligent assessment module, based on the inference results of a large model and the status data of the UAV, constructs an assessment system from the dimensions of low-altitude path compliance, obstacle avoidance accuracy, and operational stability, and outputs quantitative assessment results.

[0011] The assessment results feedback module is used to transform assessment results into visual reports, mark weaknesses in low-altitude flight, provide targeted improvement suggestions, and support the archiving and traceability of exam results.

[0012] Preferably, the low-altitude environment processing module constructs a three-dimensional environment model by including the following steps:

[0013] S11. Low-altitude scene data is collected synchronously through lidar, visual camera, and Beidou differential positioning module, including terrain elevation, building outline, dynamic obstacles and airflow disturbance information. The collection frequency is set to 10Hz to adapt to the low-altitude flight speed.

[0014] S12. Construct a low-altitude three-dimensional grid map based on the collected data. Set the grid cell side length to 0.5m and define different obstacle height thresholds through a piecewise function.

[0015] S13. Construct a low-altitude environment cost field, introduce a low-altitude risk weight coefficient, and calculate the low-altitude environment cost.

[0016] S14. Predict the trajectory of dynamic obstacles, output the position change trajectory within the next 3 seconds based on the Kalman filter algorithm, and update the status label of the dynamic area in the grid map.

[0017] S15. Integrate the constructed 3D raster map, cost field data, and obstacle trajectory prediction results to output a standardized low-altitude environment feature dataset.

[0018] Preferably, in step S13, the construction of the low-altitude environmental cost field includes the following steps:

[0019] S131. Classify low-altitude risk levels by dividing the scene into high-risk, medium-risk, and low-risk zones based on the height, density, and dynamic characteristics of obstacles.

[0020] S132. Calculate the low airflow disturbance factor based on wind speed sensor data;

[0021] S133. The cost of each cell in the raster map is superimposed to calculate the total cost of the cell, and the cost field is smoothed to eliminate abrupt changes in the raster boundary.

[0022] Preferably, the low-altitude large model inference module outputs driving decision references and path planning schemes including the following steps:

[0023] S21. Receive the standardized low-altitude environment feature dataset and preprocess the data, including outlier removal and low-altitude scene feature enhancement.

[0024] S22. Call the pre-trained low-altitude drone pilot model. This model is fine-tuned and optimized through massive low-altitude flight data and has the ability to plan low-altitude paths and make emergency obstacle avoidance decisions.

[0025] S23. Input the task instructions issued by the exam task generation module, including the starting point, ending point, necessary flight path and low-altitude no-fly zone. The model optimizes the low-altitude path based on the DCE-A* algorithm.

[0026] S24. Output the initial path planning scheme and driving operation reference instructions, including low-altitude hovering precision control parameters and obstacle avoidance steering angle. Receive UAV status monitoring data in real time, dynamically adjust the inference results, and when a sudden change in low-altitude airflow is detected, trigger the model's emergency decision-making mechanism to output speed adjustment and attitude calibration instructions.

[0027] Preferably, in step S23, the low-altitude path optimization includes the following steps:

[0028] S231. Adopt a direction cost enhancement strategy to give priority to scoring candidate nodes that face the target point;

[0029] S232. Introduce a Top-K node filtering mechanism. In each round of expansion, only the top K nodes with the highest directional scores are retained, redundant nodes are eliminated, and the altitude values ​​of path nodes are calibrated based on low-altitude constraints to ensure that the altitude of all nodes meets the low-altitude restrictions of the corresponding UAV type.

[0030] S233. Use a high-order Bézier curve to fit the initial path and optimize the path continuity;

[0031] S234. Calculate the total cost of the optimized path. If the cost is higher than the preset threshold, repeat steps S231-S234 until the path cost meets the requirements.

[0032] Preferably, the exam task generation module generates differentiated low-altitude flight tasks by including the following steps:

[0033] S31. Based on the type of drone and the examination level, retrieve the preset low-altitude mission template library, and combine it with the scene data output by the low-altitude environment perception and modeling module to select suitable mission elements, including low-altitude flight path, fixed-point hovering area, and dynamic obstacle avoidance mission.

[0034] S32. In accordance with the Civil Aviation Administration's low-altitude airspace control rules, verify the compliance of mission parameters to ensure that the mission area does not involve military restricted areas, densely populated areas, or other no-fly zones, and generate the initial test mission;

[0035] S33. Monitor changes in the low-altitude environment in real time. When a sudden obstacle or abnormal airflow is detected, dynamically adjust the mission route and difficulty coefficient. The adjustment range shall not exceed 20% of the initial mission. Convert the final mission command into a signal that the UAV can recognize and upload the signal synchronously.

[0036] Preferably, the multi-dimensional intelligent evaluation module outputs quantitative evaluation results including the following steps:

[0037] S41. Construct a low-altitude driving test evaluation index system, covering four core dimensions: route compliance, obstacle avoidance safety, operational stability, and energy consumption rationality. The weight of each dimension is set in combination with the characteristics of low-altitude scenarios, with obstacle avoidance safety accounting for no less than 30%.

[0038] S42. Receive real-time monitoring data from the UAV and extract key evaluation parameters, including low-altitude flight path deviation, minimum safe distance from obstacles, hovering altitude error, and attitude angle fluctuation amplitude.

[0039] S43. Decision reference scheme based on low-altitude large model reasoning, calculate the deviation between actual operation and optimal scheme;

[0040] S44. Quantify and score the indicators of each dimension using a 100-point system. Intrusion into the low-altitude no-fly zone and collision with obstacles will be directly judged as unqualified and trigger the test termination mechanism.

[0041] S45. Based on the combined scores and deviation calculation results of each dimension, output the final evaluation level, generate evaluation details simultaneously, and mark the weak links in low-altitude driving.

[0042] Preferably, in step S44, the quantitative scoring of each dimension indicator includes the following steps:

[0043] S441. Extract drone status data in real time, calculate the three-dimensional distance between the drone and the no-fly zone boundary, determine critical violations based on spatial distance, and simultaneously calculate the minimum distance between the drone and dynamic or static obstacles to determine obstacle collision.

[0044] S442. When the path and obstacle avoidance are both qualified, collect the attitude angle data of the UAV during the flight after the flight ends, calculate the comprehensive fluctuation amplitude, and at the same time calculate the relative deviation between the actual energy consumption and the optimal energy consumption.

[0045] S443. Calculate the score for each item according to the percentage-based quantitative formula, and combine the weights of each dimension to calculate the final comprehensive score, and output the final comprehensive score.

[0046] The low-altitude, large-model-driven intelligent examination and evaluation system for unmanned aerial vehicle (UAV) piloting provided by this invention has the following advantages:

[0047] 1. By enhancing the direction cost strategy, the efficiency and directionality of path planning are improved. Combined with the Top-K node screening mechanism to prune redundant nodes, and the path nodes are calibrated based on low-altitude constraints, the flight altitude is ensured to meet the requirements of low-altitude control. Finally, a continuous and smooth low-altitude flight path is fitted with a high-order Bézier curve. This not only improves the computational efficiency of path planning, but also ensures the safety, compliance and flight stability of the path in low-altitude scenarios, providing a more accurate and practical path solution for UAV examinations.

[0048] 2. By constructing a multi-dimensional assessment system adapted to low-altitude scenarios, highlighting the core weight of obstacle avoidance safety, and combining real-time UAV data with the optimal solution of the large model to calculate the operational deviation, the system ultimately outputs a precise assessment level and a detailed list of weak points. It also calculates the three-dimensional distance between the UAV and low-altitude no-fly zones and obstacles in real time, accurately determines critical violations and triggers the exam termination mechanism, and further quantifies the attitude angle fluctuation amplitude and energy consumption deviation. Combined with the dimensional weights, a comprehensive score is obtained. This not only strengthens the real-time control of low-altitude flight safety but also makes the quantitative assessment of the exam more accurate and rigorous, providing candidates with an objective basis for evaluating their abilities.

[0049] 3. By addressing the characteristics of dense obstacles and large airflow disturbances in low-altitude environments, and by constructing a 3D grid map and quantifying the cost field of the low-altitude environment, the system accurately identifies low-altitude-specific obstacles and no-fly zones, dynamically adapts to low-altitude control rules, and adjusts examination tasks to effectively avoid low-altitude flight safety risks. Furthermore, relying on a dedicated large-scale model fine-tuned from massive amounts of low-altitude data, it outputs driving decision-making and path planning schemes tailored to the low-altitude environment, capable of responding in real-time to sudden changes in low-altitude airflow. Combined with the inference results of the large-scale model, a multi-dimensional evaluation system is constructed, ensuring both the safety and compliance of flights in low-altitude scenarios and improving the scientific rigor and accuracy of UAV pilot examinations, providing reliable technical support for low-altitude UAV pilot training and testing. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 A schematic diagram of the overall system flow of the low-altitude large-model-driven unmanned aerial vehicle pilot intelligent examination and evaluation system provided in this application;

[0052] Figure 2 A schematic diagram of the low-altitude environment processing construction process for the low-altitude large-model-driven intelligent examination and evaluation system for unmanned aerial vehicle piloting provided in this application;

[0053] Figure 3 A schematic diagram of the output flow of the low-altitude large-model inference module of the intelligent examination and evaluation system for unmanned aerial vehicle piloting driven by the low-altitude large-model provided in this application;

[0054] Figure 4 A schematic diagram illustrating the process of generating differentiated low-altitude flight missions for the low-altitude large-model-driven intelligent examination and evaluation system for UAV piloting provided in this application.

[0055] Figure 5 This is a schematic diagram illustrating the output quantitative evaluation result flow of the low-altitude large model-driven intelligent examination and evaluation system for unmanned aerial vehicle piloting provided in this application. Detailed Implementation

[0056] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0057] like Figures 1-5 As shown, this embodiment proposes a low-altitude large-model-driven intelligent examination and evaluation system for unmanned aerial vehicle (UAV) piloting, including:

[0058] The low-altitude environment processing module is used to collect data on complex low-altitude scenes, construct a 3D environment model adapted to low-altitude characteristics, identify low-altitude-specific obstacles and no-fly zones, and output an environmental feature dataset.

[0059] The low-altitude large model inference module is trained and optimized based on a dedicated dataset for low-altitude UAV piloting. It receives environmental features and examination task instructions, and outputs driving decision references and path planning schemes that are adapted to low-altitude airflow and obstacle distribution.

[0060] The exam task generation module is used to generate differentiated low-altitude flight tasks by combining low-altitude control rules and exam levels, and to adjust task parameters in real time to adapt to environmental changes, ensuring that the tasks comply with low-altitude flight safety regulations.

[0061] The UAV status real-time monitoring module is used to collect multi-dimensional status data of UAV position, attitude, speed, and energy consumption during low-altitude flight, simultaneously monitor the stability of the low-altitude communication link, and output status monitoring data stream;

[0062] The multi-dimensional intelligent assessment module, based on the inference results of a large model and the status data of the UAV, constructs an assessment system from the dimensions of low-altitude path compliance, obstacle avoidance accuracy, and operational stability, and outputs quantitative assessment results.

[0063] The assessment results feedback module is used to transform assessment results into visual reports, mark weaknesses in low-altitude flight, provide targeted improvement suggestions, and support the archiving and traceability of exam results.

[0064] In this embodiment, the low-altitude environment processing module constructs a three-dimensional environment model through the following steps:

[0065] S11. Low-altitude scene data is collected synchronously through lidar, visual camera, and Beidou differential positioning module, including terrain elevation, building outline, dynamic obstacles and airflow disturbance information. The collection frequency is set to 10Hz to adapt to the low-altitude flight speed.

[0066] S12. Construct a low-altitude 3D raster map based on the collected data. The raster cell side length is set to 0.5m. Define different obstacle height thresholds using a piecewise function. The formula is: ,in, This refers to the actual height of the low-altitude obstacle. For different obstacle height thresholds, For grid coordinates;

[0067] Specifically, in low-altitude scenarios, obstacles are small in size, densely distributed, and vary greatly in height. A uniform height threshold can easily lead to recognition errors. By calculating the grid height threshold, obstacle areas and free space can be accurately marked, avoiding the risk of misjudgment. This result is the underlying spatial cognition foundation for subsequently constructing the low-altitude environment cost field, triggering dynamic obstacle avoidance commands, quantifying obstacle avoidance compliance in examinations, and supporting the generation of accurate path planning for low-altitude adaptive large models. It provides core support for the safety and accuracy of the entire system.

[0068] S13. Construct a low-altitude environment cost field, introduce a low-altitude risk weight coefficient, and calculate the low-altitude environment cost using the following formula: In the formula, For the cost of the low-altitude environment, This is the low-altitude risk weighting coefficient. The distance from the drone to the low-altitude obstacle. This is the distance cost coefficient. Value for low-level airflow disturbance;

[0069] Specifically, by introducing a low-altitude risk weighting coefficient, the distance risk between the UAV and obstacles and the low-altitude airflow disturbance risk are quantified into a unified environmental cost, constructing a cost field that fits the characteristics of low-altitude flight. This result not only provides a risk-oriented path planning basis for subsequent low-altitude adaptation large models, allowing the planned path to actively avoid high-risk areas, but also provides a quantitative risk benchmark for examination and evaluation. It is a core link in ensuring low-altitude flight safety and improving the accuracy of system decision-making and evaluation.

[0070] S14. Predict the trajectory of dynamic obstacles, output the position change trajectory within the next 3 seconds based on the Kalman filter algorithm, and update the status label of the dynamic area in the grid map.

[0071] S15. Integrate the constructed 3D raster map, cost field data, and obstacle trajectory prediction results to output a standardized low-altitude environment feature dataset.

[0072] In this embodiment, step S13, the construction of the low-altitude environmental cost field includes the following steps:

[0073] S131. Classify low-altitude risk levels by dividing the scene into high-risk, medium-risk, and low-risk zones based on the height, density, and dynamic characteristics of obstacles.

[0074] S132. Calculate the low-airflow disturbance factor based on wind speed sensor data, using the following formula: ,in, This refers to the instantaneous wind speed at low altitudes.

[0075] S133. The cost of each cell in the raster map is summed to obtain the total cost of the cell, using the following formula: ,in, The total cost of a single cell in a raster map. The unit risk cost is determined by the risk level: 5.0 for high-risk areas, 2.5 for medium-risk areas, and 1.0 for low-risk areas. The cost field is also smoothed to eliminate abrupt changes in grid boundaries.

[0076] Specifically, by classifying low-altitude risk levels to accurately quantify the regional risks of obstacles, and combining wind speed calculations to capture the unique airflow effects of low-altitude environments, a smooth grid cost field is finally generated. This provides a risk-oriented path planning basis that is more in line with the complex environment of low-altitude environments for subsequent large-scale low-altitude adaptation models, and also provides a quantitative risk benchmark for examination and evaluation, effectively improving the system's decision-making accuracy and safety in low-altitude scenarios.

[0077] In this embodiment, the low-altitude large model inference module outputs driving decision references and path planning schemes, including the following steps:

[0078] S21. Receive the standardized low-altitude environment feature dataset and preprocess the data, including outlier removal and low-altitude scene feature enhancement.

[0079] S22. Call the pre-trained low-altitude drone pilot model. This model is fine-tuned and optimized through massive low-altitude flight data and has the ability to plan low-altitude paths and make emergency obstacle avoidance decisions.

[0080] S23. Input the task instructions issued by the exam task generation module, including the start point, end point, mandatory flight path, and low-altitude no-fly zone. The model optimizes the low-altitude path based on the DCE-A* algorithm, with the following formula: In the formula, The shortest path length. This is the path length weighting coefficient. This is the total path length. This is the environmental cost weighting coefficient, with a value range of [0.4, 0.5]. This is a highly volatile weighting coefficient, with a value range of [0.1, 0.2]. This represents the fluctuation value of low-altitude flight altitude.

[0081] S24. Output the initial path planning scheme and driving operation reference instructions, including low-altitude hovering precision control parameters and obstacle avoidance steering angle. Receive UAV status monitoring data in real time, dynamically adjust the inference results, and when a sudden change in low-altitude airflow is detected, trigger the model's emergency decision-making mechanism to output speed adjustment and attitude calibration instructions.

[0082] Specifically, by enhancing the characteristics of low-altitude scenarios through preprocessing, a dedicated large model fine-tuned by massive amounts of low-altitude data is invoked. Combined with the DCE-A* algorithm, an optimal path is generated that takes into account path length, low-altitude environmental costs, and altitude fluctuations. Pilot operation commands are then output, and the system can dynamically respond to environmental changes such as sudden changes in low-altitude airflow. This provides accurate decision-making references and path planning solutions for drone examinations, ensuring the safety of low-altitude flight and improving the scientific nature and dynamic adaptability of the examination assessment.

[0083] In this embodiment, step S23, low-altitude path optimization includes the following steps:

[0084] S231. Employ a direction cost enhancement strategy to prioritize scoring candidate nodes that face the target point, using the following formula: In the formula, Score for direction. The angle between the node's expansion direction and the target direction;

[0085] S232. Introduce a Top-K node filtering mechanism. In each round of expansion, only the top K nodes with the highest directional scores are retained, redundant nodes are eliminated, and the altitude values ​​of path nodes are calibrated based on low-altitude constraints to ensure that the altitude of all nodes meets the low-altitude restrictions of the corresponding UAV type.

[0086] S233. A higher-order Bézier curve is used to fit the initial path to optimize path continuity. The fitting formula is: In the formula, for Coordinates of points on the Bézier curve at time , Let be the order of the Bézier curve. For the first Coordinates of the path control points for The Bessel basis function of order 1 is given by the following formula: In the formula, The combination number, i.e., the binomial coefficient, is given by the formula: , These are the parameters of the curve;

[0087] S234. Calculate the total cost of the optimized path. If the cost is higher than the preset threshold, repeat steps S231-S234 until the path cost meets the requirements.

[0088] Specifically, the efficiency and directionality of path planning are improved by using a direction cost enhancement strategy. Redundant nodes are pruned by a Top-K node screening mechanism, and path nodes are calibrated based on low-altitude constraints to ensure that the flight altitude meets the requirements of low-altitude control. Finally, a continuous and smooth low-altitude flight path is fitted using a high-order Bézier curve. This not only improves the computational efficiency of path planning, but also ensures the safety, compliance and flight stability of the path in low-altitude scenarios, providing a more accurate and practical path solution for drone testing.

[0089] In this embodiment, the exam task generation module generates differentiated low-altitude flight tasks by including the following steps:

[0090] S31. Based on the type of drone and the examination level, retrieve the preset low-altitude mission template library, and combine it with the scene data output by the low-altitude environment perception and modeling module to select suitable mission elements, including low-altitude flight path, fixed-point hovering area, and dynamic obstacle avoidance mission.

[0091] S32. In accordance with the Civil Aviation Administration's low-altitude airspace control rules, verify the compliance of mission parameters to ensure that the mission area does not involve military restricted areas, densely populated areas, or other no-fly zones, and generate the initial test mission;

[0092] S33. Monitor changes in the low-altitude environment in real time. When a sudden obstacle or abnormal airflow is detected, dynamically adjust the mission route and difficulty coefficient. The adjustment range shall not exceed 20% of the initial mission. Convert the final mission command into a signal that the UAV can recognize and upload the signal synchronously.

[0093] Specifically, by generating suitable low-altitude flight mission elements based on drone type and exam level, and verifying mission compliance in accordance with the Civil Aviation Administration's low-altitude control rules to avoid no-fly risks, the system can dynamically respond to environmental changes such as sudden low-altitude obstacles and abnormal airflow, generating differentiated missions that meet both exam requirements and low-altitude safety regulations. This ensures the accuracy and compliance of exam missions and enhances the system's dynamic adaptability in complex low-altitude scenarios, providing safe and scientific mission support for drone pilot exams.

[0094] In this embodiment, the multi-dimensional intelligent evaluation module outputs quantitative evaluation results through the following steps:

[0095] S41. Construct a low-altitude driving test evaluation index system, covering four core dimensions: route compliance, obstacle avoidance safety, operational stability, and energy consumption rationality. The weight of each dimension is set in combination with the characteristics of low-altitude scenarios, with obstacle avoidance safety accounting for no less than 30%.

[0096] S42. Receive real-time monitoring data from the UAV and extract key evaluation parameters, including low-altitude flight path deviation, minimum safe distance from obstacles, hovering altitude error, and attitude angle fluctuation amplitude.

[0097] S43. Based on the decision reference scheme of low-altitude large model inference, calculate the deviation between the actual operation and the optimal scheme, using the following formula: ,in, This represents the deviation value of the optimal solution. These are the actual operating parameters. For optimal operating parameters, These are the parameter weighting coefficients;

[0098] S44. Quantify and score the indicators of each dimension using a 100-point system. Intrusion into the low-altitude no-fly zone and collision with obstacles will be directly judged as unqualified and trigger the test termination mechanism.

[0099] S45. Based on the combined scores and deviation calculation results of each dimension, output the final evaluation level, generate evaluation details simultaneously, and mark the weak links in low-altitude driving.

[0100] Specifically, by constructing a multi-dimensional evaluation system adapted to low-altitude scenarios, highlighting the core weight of obstacle avoidance safety, and combining real-time drone data with the optimal solution of a large model to calculate the operational deviation, the system quantifies the score on a percentage basis and directly judges serious violations such as intrusion into no-fly zones as unqualified. Finally, it outputs a precise evaluation level and a detailed list of weak points, which not only ensures the scientific and rigorous nature of drone pilot test evaluation, but also strengthens the precise control of low-altitude flight safety.

[0101] In this embodiment, step S44, which involves quantifying and scoring the indicators of each dimension, includes the following steps:

[0102] S441. Extract drone status data in real time, calculate the three-dimensional distance between the drone and the no-fly zone boundary, and determine the critical violation behavior based on the spatial distance. The formula is as follows: ,in, Provide the real-time coordinates of the drone. The coordinates of the nearest point to the no-fly zone boundary. For the three-dimensional distance between the drone and the no-fly zone boundary, when If the condition is not met, it is directly judged as unqualified. At the same time, the minimum distance between the drone and dynamic or static obstacles is calculated to determine obstacle collision. The formula is as follows: ,in, The three-dimensional coordinates of the obstacle. The distance between the drone and the obstacle, when If a collision violation is triggered, the exam will be terminated immediately. This refers to the distance between the distance sensor on the drone and the drone's wingspan.

[0103] S442. When both the path and obstacle avoidance are satisfactory, collect the attitude angle data of the UAV during the flight operation after the flight ends, and calculate the comprehensive fluctuation amplitude using the following formula: ,in, This represents the deviation value for operational stability. This represents the total number of data collections. For the first The actual attitude angle of the drone at each time step. For the first The optimal attitude angle output by the low-altitude large model at each time step is calculated, and the relative deviation between the actual energy consumption and the optimal energy consumption is calculated simultaneously. The formula is as follows: , This represents the deviation value for the rationality of energy consumption. The actual energy consumption for the drone to complete the exam. The optimal energy consumption calculated for the low-altitude large model based on low-altitude scenarios;

[0104] S443. Calculate the score for each item according to the percentage-based quantitative formula, and combine the weights of each dimension to calculate the final comprehensive score. The formula is as follows: , For the final overall score, For the weights of each dimension, The system calculates scores for each dimension and outputs the final composite score.

[0105] Specifically, by calculating the three-dimensional distance between the drone and the low-altitude no-fly zone and obstacles in real time, the system accurately determines the critical violation and triggers the exam termination mechanism. Then, it quantifies the attitude angle fluctuation amplitude and energy consumption deviation, and combines the dimensional weights to obtain a comprehensive score. This not only strengthens the real-time control of low-altitude flight safety, but also makes the quantitative assessment of the exam more accurate and rigorous, providing candidates with an objective basis for evaluating their abilities.

[0106] The above embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Although the invention has been described in detail with reference to the embodiments, those skilled in the art should understand that various combinations, modifications, or equivalent substitutions of the technical solutions of the invention do not depart from the spirit and scope of the invention and should be covered within the scope of the claims of the invention.

Claims

1. A low-altitude, large-model-driven intelligent examination and evaluation system for unmanned aerial vehicle (UAV) piloting, characterized in that: include: The low-altitude environment processing module is used to collect data on complex low-altitude scenes, construct a 3D environment model adapted to low-altitude characteristics, identify low-altitude-specific obstacles and no-fly zones, and output an environmental feature dataset. The low-altitude large model inference module is trained and optimized based on a dedicated dataset for low-altitude UAV piloting. It receives environmental features and examination task instructions, and outputs driving decision references and path planning schemes that are adapted to low-altitude airflow and obstacle distribution. The exam task generation module is used to generate differentiated low-altitude flight tasks by combining low-altitude control rules and exam levels, and to adjust task parameters in real time to adapt to environmental changes, ensuring that the tasks comply with low-altitude flight safety regulations. The real-time drone status monitoring module is used to collect multi-dimensional status data of the drone's position, attitude, speed, and energy consumption during low-altitude flight, simultaneously monitor the stability of the low-altitude communication link, and output status monitoring data streams. The multi-dimensional intelligent assessment module, based on the inference results of a large model and the status data of the UAV, constructs an assessment system from the dimensions of low-altitude path compliance, obstacle avoidance accuracy, and operational stability, and outputs quantitative assessment results. The assessment results feedback module is used to transform assessment results into visual reports, mark weaknesses in low-altitude flight, provide targeted improvement suggestions, and support the archiving and traceability of exam results.

2. The low-altitude large-model driven unmanned aerial vehicle (UAV) pilot intelligent examination and evaluation system according to claim 1, characterized in that, The low-altitude environment processing module constructs a three-dimensional environment model through the following steps: S11. Low-altitude scene data is collected synchronously through lidar, visual camera, and Beidou differential positioning module, including terrain elevation, building outline, dynamic obstacles and airflow disturbance information. The collection frequency is set to 10Hz to adapt to the low-altitude flight speed. S12. Construct a low-altitude three-dimensional grid map based on the collected data. Set the grid cell side length to 0.5m and define different obstacle height thresholds through a piecewise function. S13. Construct a low-altitude environment cost field, introduce a low-altitude risk weight coefficient, and calculate the low-altitude environment cost. S14. Predict the trajectory of dynamic obstacles, output the position change trajectory within the next 3 seconds based on the Kalman filter algorithm, and update the status label of the dynamic area in the grid map. S15. Integrate the constructed 3D raster map, cost field data, and obstacle trajectory prediction results to output a standardized low-altitude environment feature dataset.

3. The low-altitude large-model driven unmanned aerial vehicle (UAV) pilot intelligent examination and evaluation system according to claim 2, characterized in that, In step S13, the construction of the low-altitude environmental cost field includes the following steps: S131. Classify low-altitude risk levels by dividing the scene into high-risk, medium-risk, and low-risk zones based on the height, density, and dynamic characteristics of obstacles. S132. Calculate the low airflow disturbance factor based on wind speed sensor data; S133. The cost of each cell in the raster map is superimposed to calculate the total cost of the cell, and the cost field is smoothed to eliminate abrupt changes in the raster boundary.

4. The low-altitude large-model driven unmanned aerial vehicle (UAV) pilot intelligent examination and evaluation system according to claim 3, characterized in that, The low-altitude large-model inference module outputs driving decision references and path planning schemes, including the following steps: S21. Receive the standardized low-altitude environment feature dataset and preprocess the data, including outlier removal and low-altitude scene feature enhancement. S22. Call the pre-trained low-altitude drone pilot model. This model is fine-tuned and optimized through massive low-altitude flight data and has the ability to plan low-altitude paths and make emergency obstacle avoidance decisions. S23. Input the task instructions issued by the exam task generation module, including the starting point, ending point, necessary flight path and low-altitude no-fly zone. The model optimizes the low-altitude path based on the DCE-A* algorithm. S24. Output the initial path planning scheme and driving operation reference instructions, including low-altitude hovering precision control parameters and obstacle avoidance steering angle. Receive UAV status monitoring data in real time, dynamically adjust the inference results, and when a sudden change in low-altitude airflow is detected, trigger the model's emergency decision-making mechanism to output speed adjustment and attitude calibration instructions.

5. The low-altitude large-model driven unmanned aerial vehicle (UAV) pilot intelligent examination and evaluation system according to claim 4, characterized in that, In step S23, the low-altitude path optimization includes the following steps: S231. Adopt a direction cost enhancement strategy to give priority to scoring candidate nodes that face the target point; S232. Introduce a Top-K node filtering mechanism. In each round of expansion, only the top K nodes with the highest directional scores are retained, redundant nodes are eliminated, and the altitude values ​​of path nodes are calibrated based on low-altitude constraints to ensure that the altitude of all nodes meets the low-altitude restrictions of the corresponding UAV type. S233. Use a high-order Bézier curve to fit the initial path and optimize the path continuity; S234. Calculate the total cost of the optimized path. If the cost is higher than the preset threshold, repeat steps S231-S234 until the path cost meets the requirements.

6. The low-altitude large-model driven unmanned aerial vehicle (UAV) pilot intelligent examination and evaluation system according to claim 5, characterized in that, The exam task generation module generates differentiated low-altitude flight tasks by including the following steps: S31. Based on the type of drone and the examination level, retrieve the preset low-altitude mission template library, and combine it with the scene data output by the low-altitude environment perception and modeling module to select suitable mission elements, including low-altitude flight path, fixed-point hovering area, and dynamic obstacle avoidance mission. S32. In accordance with the Civil Aviation Administration's low-altitude airspace control rules, verify the compliance of mission parameters to ensure that the mission area does not involve military restricted areas, densely populated areas, or other no-fly zones, and generate the initial test mission; S33. Monitor changes in the low-altitude environment in real time. When a sudden obstacle or abnormal airflow is detected, dynamically adjust the mission route and difficulty coefficient. The adjustment range shall not exceed 20% of the initial mission. Convert the final mission command into a signal that the UAV can recognize and upload the signal synchronously.

7. The low-altitude large-model driven unmanned aerial vehicle pilot intelligent examination and evaluation system according to claim 6, characterized in that, The multi-dimensional intelligent evaluation module outputs quantitative evaluation results including the following steps: S41. Construct a low-altitude driving test evaluation index system, covering four core dimensions: route compliance, obstacle avoidance safety, operational stability, and energy consumption rationality. The weight of each dimension is set in combination with the characteristics of low-altitude scenarios, with obstacle avoidance safety accounting for no less than 30%. S42. Receive real-time monitoring data from the UAV and extract key evaluation parameters, including low-altitude flight path deviation, minimum safe distance from obstacles, hovering altitude error, and attitude angle fluctuation amplitude. S43. Decision reference scheme based on low-altitude large model reasoning, calculate the deviation between actual operation and optimal scheme; S44. Quantify and score the indicators of each dimension using a 100-point system. Intrusion into the low-altitude no-fly zone and collision with obstacles will be directly judged as unqualified and trigger the test termination mechanism. S45. Based on the combined scores and deviation calculation results of each dimension, output the final evaluation level, generate evaluation details simultaneously, and mark the weak links in low-altitude driving.

8. The low-altitude large-model driven unmanned aerial vehicle (UAV) pilot intelligent examination and evaluation system according to claim 7, characterized in that, In step S44, the quantitative scoring of each dimension indicator includes the following steps: S441. Extract drone status data in real time, calculate the three-dimensional distance between the drone and the no-fly zone boundary, determine critical violations based on spatial distance, and simultaneously calculate the minimum distance between the drone and dynamic or static obstacles to determine obstacle collision. S442. When the path and obstacle avoidance are both qualified, collect the attitude angle data of the UAV during the flight after the flight ends, calculate the comprehensive fluctuation amplitude, and at the same time calculate the relative deviation between the actual energy consumption and the optimal energy consumption. S443. Calculate the score for each item according to the percentage-based quantitative formula, and combine the weights of each dimension to calculate the final comprehensive score, and output the final comprehensive score.

Images