Dynamic target-oriented end-to-end unmanned aerial vehicle motion planning and control method

Abstract

The invention relates to the technical field of unmanned aerial vehicles, in particular to an end-to-end unmanned aerial vehicle motion planning and control method for a dynamic target, and the method comprises the steps: obtaining real-time tracking data; determining a first candidate line; determining a second candidate line; determining an undetermined flight route; predicting and adjusting the model; and selecting an actual flight route. According to the invention, through fusion of real-time environment perception, candidate route screening and end-to-end prediction optimization, accurate tracking and obstacle avoidance of a dynamic target by an unmanned aerial vehicle are realized, a shielding ratio threshold and a learning rate can be adaptively adjusted, the autonomous planning capability of the unmanned aerial vehicle in a complex environment is improved, and the unmanned aerial vehicle can be used in a complex environment. According to the invention, the method can still keep stable tracking under the condition that the target is partially shielded, the obstacles are dense or the motion trail is unstable, and effectively solves the problems of low flight accuracy and low response speed caused by excessive dependence on a deep learning model and a training-based processing mode.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle technology, and in particular to an end-to-end unmanned aerial vehicle motion planning and control method for dynamic targets. Background Art

[0002] With the continuous development of science and technology, the application of drones in various fields has gradually increased, especially in flight missions in dynamic environments, such as monitoring, search and rescue, etc. Faced with complex environmental factors and dynamically changing targets, how to ensure the efficiency, safety and accuracy of drones in performing tasks has become an important issue that needs to be solved urgently. In order to achieve this goal, optimizing the motion planning and control strategy of drones to cope with the ever-changing environment and mission requirements has become the key to improving the application effect of drones.

[0003] The patent document with publication number CN118229780A discloses a UAV landing method and system based on end-to-end position estimation, the method comprising: S1, constructing a high-fidelity simulation environment for UAV landing according to actual conditions, acquiring virtual camera images in the simulation environment, and establishing a dataset of UAV landing images and their corresponding target relative poses; S2, training the images and their corresponding target pose datasets using a deep learning algorithm, and establishing a mapping relationship model between the images and their corresponding target relative pose data; S3, acquiring target pose data based on the mapping relationship model, and using it for UAV landing.

[0004] It can be seen that the drone landing method and system based on end-to-end position estimation have the following problems: the method relies on the training of high-precision image data and relative pose data sets, and needs to collect data in a simulation environment. The training-based processing method has delays in real-time applications, especially when landing a drone in a dynamic environment, and cannot guarantee a quick response; although the establishment of the virtual environment imitates the real environment as much as possible, due to the influence of factors such as environment, lighting, wind speed, etc., the gap between the virtual environment and the actual environment may lead to pose estimation errors, thereby affecting the landing accuracy of the drone; this method relies too much on deep learning algorithms to establish the mapping relationship between image and pose data, and deep learning models usually require a large amount of labeled data for training, and the training process will face the problems of overfitting or slow convergence, which reduces the stability and availability of the model. Summary of the invention

[0005] To this end, the present invention provides an end-to-end UAV motion planning and control method for dynamic targets, which uses real-time environmental perception and feedback control mechanism to overcome the problems of low flight accuracy and low response speed in the prior art due to over-reliance on deep learning models and training-based processing methods.

[0006] To achieve the above object, the present invention provides an end-to-end unmanned aerial vehicle motion planning and control method for dynamic targets, including:

[0007] Obtain the real-time occlusion ratio of the tracking target, the real-time density of the front obstacle, the real-time relative distance between the unmanned aerial vehicle and the front obstacle, and the real-time tracking distance between the unmanned aerial vehicle and the tracking target collected during the tracking process of the unmanned aerial vehicle based on a preset tracking angle;

[0008] Determine a number of first candidate routes according to the preset tracking angle, the real-time occlusion ratio, the real-time tracking distance, and a preset occlusion ratio threshold;

[0009] Determine a number of second candidate routes according to each of the first candidate routes, a preset flight step size, and the real-time relative distance;

[0010] Determine a number of first undetermined flight routes according to each of the second candidate routes, the preset flight step size, the real-time tracking distance, and the real-time density;

[0011] Respectively obtain the real-time position coordinates of the unmanned aerial vehicle, the tracking target, and the front obstacle in three-dimensional space;

[0012] Input all the real-time position coordinates into a preset end-to-end prediction model to obtain a second undetermined flight route;

[0013] Adjust the preset occlusion ratio threshold according to the deviation between each of the first undetermined flight routes and the second undetermined flight route to form an adjusted occlusion ratio threshold, or adjust the preset learning rate of the preset end-to-end prediction model to form an adjusted learning rate;

[0014] Select an actual flight route based on the adjusted occlusion ratio threshold, or select an actual flight route based on the adjusted learning rate;

[0015] Control the unmanned aerial vehicle to move along the actual flight route.

[0016] Further, determining a number of first candidate routes according to the preset tracking angle, the real-time occlusion ratio, the real-time tracking distance, and a preset occlusion ratio threshold includes:

[0017] When the real-time occlusion ratio is greater than the preset occlusion ratio threshold, calculate the change speed of the real-time tracking distance within a preset past historical window to form a tracking distance change rate;

[0018] Perform linear extrapolation according to the tracking distance change rate and a preset prediction duration to form a predicted relative position;

[0019] Determine a number of the first candidate routes according to the predicted relative position and the preset tracking angle.

[0020] Further, determining a plurality of the first candidate routes according to the predicted relative position and the preset tracking angle includes:

[0021] Calculating the predicted relative position and the relative displacement direction angle with the drone as the zero point;

[0022] Taking the relative displacement direction angle as a reference, selecting a preset number of candidate heading angles according to a preset angle offset;

[0023] Determining a plurality of first candidate angles according to each of the candidate heading angles and the preset tracking angle;

[0024] Calculating all relative positions within a preset planning duration when advancing along each of the first candidate angles to determine a plurality of the first candidate routes.

[0025] Further, determining a plurality of first candidate angles according to each of the candidate heading angles and the preset tracking angle includes:

[0026] Calculating the absolute value of the relative deviation between each of the candidate heading angles and the preset tracking angle to form a plurality of angle deviations;

[0027] When the angle deviation is less than a preset angle deviation threshold, determining the candidate heading angle as the first candidate angle to determine a plurality of the first candidate angles.

[0028] Further, determining a plurality of second candidate routes according to each of the first candidate routes, a preset flight step, and the real-time relative distance includes:

[0029] Obtaining the relative position of advancing the preset flight step along the first candidate route to form a first expected position;

[0030] Calculating an expected relative distance according to the first expected position and the real-time relative distance;

[0031] When the expected relative distance is greater than a preset safety distance, determining the first candidate route as the second candidate route to determine a plurality of the second candidate routes.

[0032] Further, determining a plurality of first undetermined flight routes according to each of the second candidate routes, the preset flight step, the real-time tracking distance, and the real-time density includes:

[0033] Calculating the standard deviation of all the real-time tracking distances within a preset past historical duration to form a tracking distance fluctuation value;

[0034] Calculating the standard deviation of all the real-time densities within the preset past historical duration to form a density fluctuation value;

[0035] Determine a number of first undetermined flight routes according to each of the second candidate routes, the preset flight step length, the tracking distance fluctuation value, and the density fluctuation value.

[0036] Further, determining a number of first undetermined flight routes according to each of the second candidate routes, the preset flight step length, the tracking distance fluctuation value, and the density fluctuation value includes:

[0037] Calculate the correlation coefficient of the tracking distance fluctuation value and the density fluctuation value to form a consistency.

[0038] When the consistency is greater than a preset consistency threshold, obtain the relative position of advancing the preset flight step length along the second candidate route to form a second expected position.

[0039] Calculate the expected tracking distance according to the second expected position and the real-time tracking distance.

[0040] When the expected tracking distance is within a preset standard tracking distance range, determine the second candidate route as the first undetermined flight route to determine a number of first undetermined flight routes.

[0041] Further, adjusting the preset occlusion ratio threshold according to the deviation between each of the first undetermined flight routes and the second undetermined flight routes to form an adjusted occlusion ratio threshold, or adjusting the preset learning rate of the preset end-to-end prediction model to form an adjusted learning rate includes:

[0042] Calculate the position deviation of each preset unit time of each of the first undetermined flight routes and the second undetermined flight routes within the preset planning duration.

[0043] Calculate the direction deviation of each preset unit time of each of the first undetermined flight routes and the second undetermined flight routes within the preset planning duration.

[0044] Perform normalization processing on the position deviation to form a normalized position deviation, and perform normalization processing on the direction deviation to form a normalized direction deviation.

[0045] Perform weighted summation on the normalized position deviation, the normalized direction deviation, the preset position deviation weight, and the preset direction deviation weight to form a number of overall deviation values.

[0046] Adjust the preset occlusion ratio threshold according to all the overall deviation values, the normalized position deviation, and the normalized direction deviation to form an adjusted occlusion ratio threshold, or adjust the preset learning rate to form an adjusted learning rate.

[0047] Further, adjusting the preset occlusion ratio threshold according to all the overall deviation values, the normalized position deviation, and the normalized direction deviation to form an adjusted occlusion ratio threshold, or adjusting the preset learning rate to form an adjusted learning rate includes:

[0048] Select the minimum value among all the overall deviation values to form a comparison deviation value;

[0049] When the comparison deviation value is greater than a preset overall deviation threshold, calculate the sum of the normalized position deviation and the normalized direction deviation to form a total deviation;

[0050] Calculate the ratio of the normalized position deviation to the total deviation to form a position ratio;

[0051] When the position ratio is greater than a preset ratio threshold, reduce the preset occlusion ratio threshold according to the relative deviation between the comparison deviation value and the preset comparison deviation threshold and a preset adjustment coefficient to form the adjusted occlusion ratio threshold;

[0052] When the position ratio is less than or equal to the preset ratio threshold, increase the preset learning rate according to the relative deviation between the comparison deviation value and the preset comparison deviation threshold and the preset adjustment coefficient to form the adjusted learning rate.

[0053] Further, selecting an actual flight route based on the adjusted occlusion ratio threshold, or selecting an actual flight route based on the adjusted learning rate includes:

[0054] When determining a number of first candidate routes based on the adjusted occlusion ratio threshold, select the first candidate route corresponding to the comparison deviation value as the actual flight route;

[0055] When obtaining a second undetermined flight route based on the adjusted learning rate, select the second undetermined flight route as the actual flight route.

[0056] Compared with the prior art, the beneficial effects of the present invention are as follows. By integrating real-time environment perception, candidate route screening, and end-to-end prediction optimization, the present invention realizes precise tracking and obstacle avoidance of dynamic targets by drones, can adaptively adjust the occlusion ratio threshold and the learning rate, improves the autonomous planning ability of drones in complex environments, enables them to maintain stable tracking even when the target is partially occluded, the obstacles are dense, or the movement trajectory is unstable, optimizes the path selection of drones at the same time, improves the flight efficiency while ensuring safety, reduces unnecessary energy consumption and path adjustments, improves the intelligence level and robustness of task execution, and effectively solves the problems of low flight accuracy and low response speed caused by over-reliance on deep learning models and training-based processing methods.

[0057] Furthermore, when the target is occluded, the future position of the target is predicted through historical motion trends, enabling the UAV to maintain continuous and stable tracking in complex environments. By generating candidate routes based on the predicted relative positions, the forward-looking nature of path planning is enhanced, reducing trajectory deviations or failures caused by the temporary disappearance of the target, endowing the UAV with stronger autonomous tracking capabilities and environmental adaptability, and improving the intelligence level of flight and the mission success rate.

[0058] Furthermore, by calculating the first candidate route based on the predicted relative positions, the motion trends of the target can be fully considered, improving the accuracy of path planning for the UAV in complex environments. The use of the relative displacement direction angle and the candidate heading angle makes the path selection more flexible and capable of adapting to dynamic adjustments in different scenarios. At the same time, the introduction of a preset planning duration helps the UAV strike a balance between short-term planning and long-term tracking, enhancing flight stability and the continuity of target tracking.

[0059] Furthermore, by setting an angle deviation threshold to screen reasonable candidate heading angles, ineffective path calculations are effectively reduced, optimizing the calculation efficiency. At the same time, by restricting the angle deviation range, the flight trajectory of the UAV is ensured to be more stable, reducing the impact of large yaw on target tracking accuracy and enhancing the autonomous navigation ability in dynamic environments.

[0060] Furthermore, by calculating the expected relative distance and comparing it with the preset safety distance, the second candidate routes that meet the safety requirements can be effectively screened out, avoiding possible collisions or accidents during the flight of the UAV. This process enhances the safety and reliability of path planning and improves flight stability in dynamic environments.

[0061] Furthermore, by comprehensively analyzing the fluctuations in the tracking distance and environmental density, this method can effectively screen out relatively stable flight routes, avoiding flight path deviations caused by drastic changes in the target distance or environmental density, improving flight safety and continuity, and at the same time optimizing the route planning to endow the UAV with stronger adaptability in complex environments.

[0062] Furthermore, by calculating the correlation between the tracking distance fluctuation value and the density fluctuation value, the adaptability to dynamic environmental changes is improved, avoiding the influence of local abnormal data on flight path judgment. Combining the verification mechanism of the preset flight step size, real-time tracking distance, and standard tracking distance range enables the UAV to select a more stable and safe flight route, thereby optimizing path planning and improving flight stability and target tracking accuracy.

[0063] Furthermore, through the normalization and weighted calculation of the flight path deviation, the adaptive adjustment of the occlusion ratio threshold or learning rate is achieved, enabling the UAV to plan a more stable flight path in complex environments. Adjusting the occlusion ratio threshold helps optimize the obstacle avoidance strategy and improve navigation safety, while adjusting the learning rate can enhance the model's adaptability to environmental changes, accelerate the learning convergence speed, improve the prediction accuracy, and thus enhance the flight stability and mission execution efficiency of the UAV.

[0064] Furthermore, by adjusting the occlusion ratio threshold and learning rate, this process can dynamically optimize the parameters according to the deviation during flight, thereby enhancing the adaptability and stability of the system. When there is a large deviation in the path, by adjusting the occlusion ratio threshold, the influence of obstacle occlusion can be effectively avoided, improving flight safety; while in the case of small deviation, by adjusting the learning rate, the learning effect of the model can be better improved, ensuring that the system adjusts and corrects more precisely during flight. This flexible parameter adjustment strategy helps improve the execution effect of the flight mission and reduce potential risks caused by environmental changes or deviations.

[0065] Furthermore, by dynamically adjusting the occlusion ratio threshold and learning rate, the flight route is flexibly optimized under different flight conditions. By precisely controlling the selection of candidate routes, the errors and deviations during flight can be minimized to the greatest extent, ensuring flight safety and efficiency, and enhancing the success rate and stability of the flight mission. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 is a flowchart of the end-to-end UAV motion planning and control method for dynamic targets in this embodiment;

[0067] Figure 2 is a decision logic diagram for determining the first candidate angle in this embodiment;

[0068] Figure 3 is a decision logic diagram for determining the second candidate route in this embodiment;

[0069] Figure 4 is a decision logic diagram for the first undetermined flight route in this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0070] In order to make the objectives and advantages of the present invention clearer, the present invention will be further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0071] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present invention and do not limit the protection scope of the present invention.

[0072] Please refer to Figure 1 as shown, which is a flowchart of the end-to-end UAV motion planning and control method for dynamic targets in this embodiment;

[0073] This embodiment provides an end-to-end UAV motion planning and control method for dynamic targets, including:

[0074] Obtain the real-time occlusion ratio of the tracking target, the real-time density of the front obstacle, the real-time relative distance between the UAV and the front obstacle, and the real-time tracking distance between the UAV and the tracking target collected during the tracking process of the UAV based on a preset tracking angle;

[0075] Determine a number of first candidate routes according to the preset tracking angle, the real-time occlusion ratio, the real-time tracking distance, and a preset occlusion ratio threshold;

[0076] Determine a number of second candidate routes according to each of the first candidate routes, a preset flight step length, and the real-time relative distance;

[0077] Determine a number of first undetermined flight routes according to each of the second candidate routes, the preset flight step length, the real-time tracking distance, and the real-time density;

[0078] Respectively obtain the real-time position coordinates of the UAV, the tracking target, and the front obstacle in three-dimensional space;

[0079] Input all the real-time position coordinates into a preset end-to-end prediction model to obtain a second undetermined flight route;

[0080] Adjust the preset occlusion ratio threshold according to the deviation between each of the first undetermined flight routes and the second undetermined flight route to form an adjusted occlusion ratio threshold, or adjust the preset learning rate of the preset end-to-end prediction model to form an adjusted learning rate;

[0081] Select an actual flight route based on the adjusted occlusion ratio threshold, or select an actual flight route based on the adjusted learning rate;

[0082] Control the UAV to move along the actual flight route.

[0083] By equipping the drone with an RGB camera and a depth camera, the target detection algorithm (YOLO or Mask R-CNN) is used to identify and track the target, and the proportion of occluded pixels of the target is calculated through image segmentation to determine the real-time occlusion ratio. At the same time, the lidar or structured light depth camera scans the front environment, and the real-time density of the front obstacle and the real-time relative distance between the drone and the front obstacle are calculated through point cloud data analysis. The front obstacles usually refer to some stable structures, such as tall buildings or trees, etc. These objects usually remain unchanged during flight, so they can be regarded as fixed obstacles in flight planning to help the drone plan a safer and more efficient flight path and avoid collisions with them. In addition, by combining GPS / IMU sensors and visual SLAM technology, the three-dimensional spatial distance between the drone and the tracking target is calculated in real time to obtain the real-time tracking distance.

[0084] The pre-set end-to-end prediction model is a trained neural network model. It is trained and inferred in an end-to-end manner, directly generating prediction results from the input data without manual feature engineering or additional intermediate steps. It is trained by using the adaptive reinforcement learning training method for drone dynamic target tracking on the initial end-to-end motion planning neural network for drone tracking dynamic target points. The specific training method is as follows:

[0085] First, in the expert data collection stage, collect the flight state and target position data of the drone for dynamic target tracking in a complex obstacle environment. Then, in the same environment, in the guidance strategy initialization stage, based on the collected data, conduct reinforcement supervised learning training to enable the trained neural network to imitate the behavior of experts and possess preliminary motion planning capabilities. Then, transfer the neural network corresponding to the behavior decision module trained in the guidance strategy initialization stage to the adaptive reinforcement learning module;

[0086] Specifically, in the expert guidance stage, an expert strategy based on MINCO trajectory optimization is adopted to generate high-quality trajectory data for the drone to perform dynamic target tracking in a complex environment with dense obstacles, which is used to train the subsequent end-to-end neural network. Although this strategy has a high computational cost, in the simulation environment, the computational resources can be allocated by pausing the drone dynamics, thus obtaining smooth, safe and speed-optimal trajectory data.

[0087] The steps of the expert guidance stage are as follows:

[0088] Initialization: Set the target trajectory, point cloud map and data set;

[0089] State acquisition: Obtain the current drone state (position, speed, acceleration, etc.);

[0090] Path planning: Use the JPS algorithm to generate a collision-free path;

[0091] Flight corridor construction: Construct a safe flight area around the initial path;

[0092] Trajectory optimization: Generate a smooth and kinematically feasible trajectory through MINCO optimization;

[0093] Data recording: Record the state sequence and update the dataset;

[0094] Target change handling: Re-plan when the target position changes, and repeat the above steps until the target is reached.

[0095] Initialization training of the policy: Train the initial control policy using expert data, and the steps are as follows:

[0096] State space design: Include the current UAV state and future trajectory information in the expert data;

[0097] Action space: The size of the rotor throttle output by the behavior decision module;

[0098] Reward function: Include rewards for position, speed, quaternion, and angular velocity errors;

[0099] Algorithm: Perform reinforcement learning training based on the PPO algorithm, and optimize the network parameters until the model converges;

[0100] Adaptive training of reinforcement learning: Further optimize the control policy through autonomous exploration based on the expert policy, and the steps are as follows:

[0101] Policy transfer: Transfer the network parameters obtained from the initialization training to the adaptive module;

[0102] State space adjustment: Only include on-board sensor information (including but not limited to radar, position, speed, and acceleration);

[0103] Reward function optimization: Include rewards for speed alignment, heading alignment, energy efficiency, attitude stability, action smoothness, and safety distance;

[0104] Training process: Interact with the environment to generate trajectory data, calculate the total loss function, and update the network parameters until the model converges.

[0105] Through the above three stages, train the initial end-to-end motion planning neural network for the UAV to track dynamic target points from expert guidance to autonomous learning, gradually improve the dynamic target tracking ability in complex environments, and form a preset end-to-end prediction model.

[0106] The preset tracking angle refers to the line-of-sight angle of the drone relative to the tracking target, which usually depends on the mission requirements (such as side tracking or frontal tracking) and the target movement pattern, and is usually set between 30° and 90°. In this embodiment, it is set to 45°, ensuring that the drone can keep the target in sight during tracking and avoid obstacles at the same time.

[0107] The preset occlusion ratio threshold refers to the maximum ratio of the target that the drone allows to be occluded, which depends on the robustness of the visual algorithm and the mission requirements, and is usually set between 5% and 15%. In this embodiment, it is set to 10%, ensuring that the target will not be occluded for a long time and avoiding flight instability caused by frequent route adjustments.

[0108] The preset flight step refers to the minimum displacement of each flight adjustment of the drone, which depends on the control accuracy of the drone and the environmental complexity, and is usually set between 0.5m and 5m. In this embodiment, it is set to 1m, ensuring smooth tracking and reducing the path planning calculation amount.

[0109] The preset learning rate refers to the step size for updating the parameters of the end-to-end prediction model, which depends on the model convergence speed and stability, and is usually set between 0.0001 and 0.01. In this embodiment, it is set to 0.001, ensuring efficient convergence during training and avoiding model instability caused by oscillations at the same time.

[0110] By obtaining real-time environmental data such as the occlusion ratio, obstacle density, and relative distance when the drone tracks the target, gradually screening candidate flight routes, and optimizing them in combination with the preset end-to-end prediction model. By comparing the deviations of different routes, dynamically adjusting the occlusion ratio threshold or the model learning rate, so as to select the optimal actual flight route and control the drone to move along this route, realizing precise tracking and obstacle avoidance of dynamic targets.

[0111] By integrating real-time environmental perception, candidate route screening, and end-to-end prediction optimization, the drone realizes precise tracking and obstacle avoidance of dynamic targets, can adaptively adjust the occlusion ratio threshold and the learning rate, improves the autonomous planning ability of the drone in complex environments, enables it to maintain stable tracking when the target is partially occluded, the obstacles are dense or the movement trajectory is unstable, and at the same time optimizes the path selection of the drone, improves the flight efficiency while ensuring safety, reduces unnecessary energy consumption and path adjustments, improves the intelligence and robustness of task execution, and effectively solves the problems of low flight accuracy and low response speed caused by over-reliance on deep learning models and training-based processing methods.

[0112] Specifically, determining a number of first candidate routes according to the preset tracking angle, the real-time occlusion ratio, the real-time tracking distance, and the preset occlusion ratio threshold includes:

[0113] When the real-time occlusion ratio is greater than the preset occlusion ratio threshold, calculate the change speed of the real-time tracking distance within a preset historical window in the past to form a tracking distance change rate;

[0114] Perform linear extrapolation based on the tracking distance change rate and a preset prediction duration to form a predicted relative position;

[0115] Determine a number of the first candidate routes according to the predicted relative position and the preset tracking angle.

[0116] Linear extrapolation is a method for predicting future values based on the changing trend of historical data. Its basic principle is to use the change rate within a past period of time to estimate the value at a future time point. In this embodiment, linear extrapolation calculates the change rate of the tracking distance within a preset historical window in the past and multiplies it by the preset prediction duration to calculate the future relative position of the target. It is applicable to the situation where the target movement is relatively stable, can provide a reasonable prediction of the movement trend in a short time, and improve the tracking stability of the UAV for dynamic targets.

[0117] The preset historical window refers to the past time period used to calculate the tracking distance change rate, which depends on the flight speed of the UAV, the target movement characteristics, and the dynamic changes of the environment. It is usually set between 1 s and 5 s. In this embodiment, it is set to 3 s, which can balance real-time performance and prediction accuracy, and avoid unstable prediction caused by too short a window or lag effect caused by too long a window.

[0118] First, when the occlusion ratio of the target exceeds the preset threshold, calculate the change rate of the tracking distance of the UAV within a past period of time, and then predict the future relative position of the target through linear extrapolation. Then, in combination with the preset tracking angle, generate multiple first candidate routes according to the predicted relative position, providing a preliminary alternative plan for the motion planning of the UAV.

[0119] Predict the future position of the target through the historical motion trend when the target is occluded, enabling the UAV to maintain continuous and stable tracking in a complex environment. By generating candidate routes based on the predicted relative position, the forward-looking nature of path planning is improved, and the trajectory deviation or failure caused by the temporary disappearance of the target is reduced, enabling the UAV to have stronger autonomous tracking ability and environmental adaptability, and improving the intelligence level and mission success rate of flight.

[0120] Specifically, determining a number of the first candidate routes according to the predicted relative position and the preset tracking angle includes:

[0121] Calculate the predicted relative position and the relative displacement direction angle with the UAV as the zero point;

[0122] Based on the relative displacement direction angle, select a preset number of candidate heading angles according to a preset angle offset.

[0123] Determine a number of first candidate angles according to each of the candidate heading angles and the preset tracking angle;

[0124] Calculate all relative positions during a preset planning duration when advancing along each of the first candidate angles to determine a number of the first candidate routes.

[0125] The preset angle offset refers to the angle offset range relative to the relative displacement direction angle, which depends on the maneuverability of the UAV, the environmental complexity, and the flexibility of the target movement. It is usually set between 5° and 30°. In this embodiment, it is set to 15°. This can ensure a sufficient heading adjustment range while avoiding a decrease in flight efficiency caused by excessive offset, and improve the path optimization ability of the UAV.

[0126] The preset candidate quantity refers to the number of candidate heading angles selected based on the relative displacement direction angle, which depends on the size of the angle offset range and the complexity of the path optimization calculation. It is usually set between 5 and 15. In this embodiment, it is set to 9, which ensures the diversity of path planning, improves the adaptability to complex environments, and at the same time controls the calculation amount within a reasonable range.

[0127] The preset planning duration refers to the time length for predicting the flight trajectory of the UAV under the preset candidate angles, which depends on the flight speed of the UAV, the obstacle avoidance requirements, and the target movement speed. It is usually set between 0.5 seconds and 3 seconds. In this embodiment, it is set to 2 seconds, which can ensure that the UAV has sufficient maneuvering adjustment space in a dynamic environment and will not be too long to cause excessive prediction errors, and improves the feasibility of the flight path.

[0128] First, calculate the predicted relative position and determine the relative displacement direction angle of this position relative to the current position of the UAV. Then, based on this direction angle, select multiple candidate heading angles according to the preset angle offset. Next, determine multiple first candidate angles according to the relationship between these candidate heading angles and the preset tracking angle. Finally, simulate the movement trajectory of the UAV within the preset planning duration along each first candidate angle and calculate the corresponding relative positions, thereby generating multiple first candidate routes to provide alternative paths for the movement planning of the UAV.

[0129] By calculating the first candidate routes based on the predicted relative position, it is possible to fully consider the target movement trend and improve the path planning accuracy of the UAV in complex environments. The method of using the relative displacement direction angle and candidate heading angles makes the path selection more flexible and can adapt to dynamic adjustments in different scenarios. At the same time, the introduction of the preset planning duration helps the UAV to achieve a balance between short-term planning and long-term tracking, and improves flight stability and the continuity of target tracking.

[0130] Please continue to refer to Figure 2As shown, it is the determination logic diagram for determining the first candidate angle in this embodiment;

[0131] Determining a number of first candidate angles based on each of the candidate heading angles and the preset tracking angle includes:

[0132] Calculating the absolute value of the relative deviation between each of the candidate heading angles and the preset tracking angle to form a number of angle deviations;

[0133] When the angle deviation is less than the preset angle deviation threshold, determining the candidate heading angle as the first candidate angle to determine a number of the first candidate angles.

[0134] The preset angle deviation threshold is a parameter used to screen the first candidate angles, representing the maximum allowable deviation range between the candidate heading angle and the preset tracking angle, which depends on the movement flexibility of the drone, the environmental complexity, and the dynamic characteristics of the tracking target. It is usually set between 5° and 20°. In this embodiment, it is set to 10°, which can not only ensure that the drone has sufficient maneuverability in a complex environment but also maintain a relatively stable tracking path, avoiding control errors caused by unnecessary angle offsets.

[0135] First, calculate the absolute value of the relative deviation between the candidate heading angle and the preset tracking angle to obtain a number of angle deviations. Subsequently, compare these angle deviations with the preset angle deviation threshold, screen out the candidate heading angles that meet the deviation requirements as the first candidate angles, and finally determine a number of first candidate angles that meet the tracking requirements, providing a basis for path optimization.

[0136] By setting the angle deviation threshold to screen reasonable candidate heading angles, the calculation of invalid paths is effectively reduced, and the calculation efficiency is optimized. At the same time, by restricting the angle deviation range, the flight trajectory of the drone is ensured to be more stable, the influence of large yaw on the target tracking accuracy is reduced, and the autonomous navigation ability in a dynamic environment is improved.

[0137] Please continue to refer to Figure 3 As shown, it is the determination logic diagram for determining the second candidate route in this embodiment;

[0138] Determining a number of second candidate routes based on each of the first candidate routes, the preset flight step, and the real-time relative distance includes:

[0139] Obtaining the relative position of advancing the preset flight step along the first candidate route to form a first expected position;

[0140] Calculating the expected relative distance based on the first expected position and the real-time relative distance;

[0141] When the expected relative distance is greater than the preset safety distance, determine the first candidate route as the second candidate route to determine a number of the second candidate routes;

[0142] Among them, the expected relative distance is obtained by calculating the Euclidean distance between the first expected position and the coordinates of the front obstacle through the Pythagorean theorem, and the coordinates of the front obstacle are calculated through the preset tracking angle and the real-time relative distance.

[0143] The preset safety distance refers to the minimum safety distance that the drone maintains from obstacles or targets when planning the flight path, which depends on the flight speed of the drone, the sensor accuracy, environmental conditions (such as wind speed), and the response time of the obstacle avoidance algorithm. It is usually set between 1.5 meters and 5 meters. In this embodiment, it is set to 3 meters to balance safety and path optimization in a complex environment and reduce unnecessary obstacle avoidance adjustments.

[0144] First, according to each first candidate route, the preset flight step size, and the real-time relative distance, obtain the relative position of advancing a preset flight step size along the first candidate route to form the first expected position. Then, calculate the expected relative distance based on the first expected position and the real-time relative distance. Finally, if the expected relative distance is greater than the preset safety distance, determine that the first candidate route is the second candidate route, and further determine a number of the second candidate routes.

[0145] By calculating the expected relative distance and comparing it with the preset safety distance, the second candidate routes that meet the safety requirements can be effectively screened out, avoiding possible collisions or accidents during the flight of the drone. This process enhances the safety and reliability of path planning and improves the flight stability in a dynamic environment.

[0146] Specifically, determining a number of first undetermined flight routes according to each of the second candidate routes, the preset flight step size, the real-time tracking distance, and the real-time density includes:

[0147] Calculate the standard deviation of all the real-time tracking distances within the past preset historical duration to form a tracking distance fluctuation value;

[0148] Calculate the standard deviation of all the real-time densities within the past preset historical duration to form a density fluctuation value;

[0149] Determine a number of first undetermined flight routes according to each of the second candidate routes, the preset flight step size, the tracking distance fluctuation value, and the density fluctuation value.

[0150] The preset historical duration refers to the data time window used to calculate the tracking distance fluctuation value and the density fluctuation value, that is, the statistical range of the tracking distance and density data within a certain period in the past, which depends on the degree of environmental dynamic changes, the flight speed of the drone, and the system's requirement for real-time performance. It is usually set between 5 seconds and 30 seconds. In this embodiment, it is set to 15 seconds to balance data stability and response speed.

[0151] First, calculate the standard deviation of all real-time tracking distances within the preset historical duration in the past to obtain the tracking distance fluctuation value, so as to reflect the change of the target distance. Then, calculate the standard deviation of all real-time densities within the same historical duration to form the density fluctuation value, so as to measure the change trend of the environmental density. Finally, based on the second candidate route, the preset flight step, the tracking distance fluctuation value, and the density fluctuation value, screen out the first undetermined flight route that meets the requirements of stability and safety.

[0152] By comprehensively analyzing the fluctuations of the tracking distance and the environmental density, this method can effectively screen out relatively stable flight routes, avoid flight path deviations caused by drastic changes in the target distance or environmental density, improve the safety and continuity of flight, and at the same time optimize the route planning, enabling the drone to have stronger adaptability in complex environments.

[0153] Please continue to refer to Figure 4 as shown, which is the determination logic diagram of the first undetermined flight route in this embodiment;

[0154] Determining a number of first undetermined flight routes according to each of the second candidate routes, the preset flight step, the tracking distance fluctuation value, and the density fluctuation value includes:

[0155] Calculate the correlation coefficient of the tracking distance fluctuation value and the density fluctuation value to form a consistency;

[0156] When the consistency is greater than the preset consistency threshold, obtain the relative position of advancing the preset flight step along the second candidate route to form a second expected position;

[0157] Calculate the expected tracking distance according to the second expected position and the real-time tracking distance;

[0158] When the expected tracking distance is within the preset standard tracking distance range, determine the second candidate route as the first undetermined flight route to determine a number of first undetermined flight routes. Among them, the expected tracking distance is obtained by calculating the Euclidean distance between the second expected position and the coordinates of the tracking target through the Pythagorean theorem, and the coordinates of the tracking target are calculated through the preset tracking angle and the real-time tracking distance.

[0159] The preset consistency threshold is a parameter used to determine whether there is a strong correlation between the tracking distance fluctuation value and the density fluctuation value. It depends on the environmental stability requirements, the sensitivity of the drone to external disturbances, and the complexity of the target area. It is usually set between 0.6 and 0.9. In this embodiment, it is set to 0.75 to allow a certain degree of environmental change adaptability while ensuring path stability.

[0160] The preset standard tracking distance range refers to the ideal tracking distance interval between the drone and the target. It depends on the specific requirements of the flight mission, the moving speed of the target, the complexity of the flight environment, and the dynamic control ability of the drone. It is usually set between 1 meter and 50 meters. In this embodiment, it is set to 10 meters to 30 meters, ensuring that in most scenarios, the tracking target always remains within a reasonable distance, which can not only ensure the real-time tracking effect but also avoid interference from being too far or too close, guaranteeing flight stability and control accuracy.

[0161] First, calculate the correlation coefficient of the tracking distance fluctuation value and the density fluctuation value over a past period of time to obtain the consistency. When the consistency is greater than the preset consistency threshold, move forward a preset flight step along the second candidate route to obtain the corresponding second expected position. Then, calculate the expected tracking distance based on this second expected position and the real-time tracking distance, and compare it with the preset standard tracking distance range. If the expected tracking distance is within this range, determine the second candidate route as the first pending flight route, thereby screening out the flight path that meets the standards.

[0162] By calculating the correlation between the tracking distance fluctuation value and the density fluctuation value, the adaptability to environmental dynamic changes is improved, and the influence of local abnormal data on flight path judgment is avoided. Combining the verification mechanism of the preset flight step, real-time tracking distance, and standard tracking distance range enables the drone to select a more stable and safe flight route, thereby optimizing the path planning and improving flight stability and target tracking accuracy.

[0163] Specifically, adjusting the preset occlusion ratio threshold according to the deviation between each of the first pending flight routes and the second pending flight routes to form an adjusted occlusion ratio threshold, or adjusting the preset learning rate of the preset end-to-end prediction model to form an adjusted learning rate includes:

[0164] Calculate the position deviation of each preset unit time of each of the first pending flight routes and the second pending flight routes within the preset planning duration;

[0165] Calculate the direction deviation of each preset unit time of each of the first pending flight routes and the second pending flight routes within the preset planning duration;

[0166] Normalize the position deviation to form a normalized position deviation, and normalize the direction deviation to form a normalized direction deviation;

[0167] Perform a weighted sum of the normalized position deviation, the normalized direction deviation, the preset position deviation weight, and the preset direction deviation weight to form a number of overall deviation values;

[0168] Adjust the preset occlusion ratio threshold according to all the overall deviation values, the normalized position deviation, and the normalized direction deviation to form an adjusted occlusion ratio threshold, or adjust the preset learning rate to form an adjusted learning rate.

[0169] Normalization refers to converting data with different dimensions and ranges to a unified standard scale or range for comparison and calculation. In this scenario, normalization usually scales data such as position deviation and direction deviation by their maximum values so that their values fall within a standard interval (between 0 and 1). This can eliminate the dimensional differences between different data and ensure that their contributions to the overall deviation calculation are equal when performing a weighted sum.

[0170] The preset position deviation weight is a relative importance coefficient used to measure the position deviation in the overall deviation calculation, and the preset direction deviation weight is a relative importance coefficient used to measure the direction deviation in the overall deviation calculation. Both depend on the UAV mission requirements, environmental complexity, flight stability requirements, and sensitivity to position accuracy and direction adjustment. Usually, the sum of the preset position deviation weight and the preset direction deviation weight is 1. The position deviation weight is generally set between 0.4 and 0.7, and the direction deviation weight is set between 0.3 and 0.6. In this embodiment, the preset position deviation weight is set to 0.6, and the preset direction deviation weight is set to 0.4. A higher preset position deviation weight can ensure that the UAV optimizes position accuracy first during flight to avoid position drift, while an appropriate allocation of the direction deviation weight helps maintain heading stability and improve the accuracy and smoothness of the overall flight trajectory.

[0171] First, calculate the position deviation and direction deviation of all the first undetermined flight routes and the second undetermined flight routes within each preset unit time during the preset planning duration. Then, normalize these deviation data to obtain the normalized position deviation and the normalized direction deviation. Next, combine the preset position deviation weight and direction deviation weight, and perform a weighted sum on the normalized data to calculate the overall deviation value. Finally, dynamically adjust the preset occlusion ratio threshold or the learning rate of the preset end-to-end prediction model according to all the overall deviation values, the normalized position deviation, and the normalized direction deviation to adapt to the actual flight situation and optimize path selection and model training.

[0172] Through the normalization and weighted calculation of the flight path deviation, the adaptive adjustment of the occlusion ratio threshold or the learning rate is realized, enabling the UAV to more stably plan the flight path in complex environments. Adjusting the occlusion ratio threshold helps optimize the obstacle avoidance strategy and improve navigation safety, while adjusting the learning rate can enhance the model's adaptability to environmental changes, accelerate the learning convergence speed, improve the prediction accuracy, and thus enhance the flight stability and mission execution efficiency of the UAV.

[0173] Specifically, adjusting the preset occlusion ratio threshold according to all the overall deviation values, the normalized position deviation, and the normalized direction deviation to form an adjusted occlusion ratio threshold, or adjusting the preset learning rate to form an adjusted learning rate includes:

[0174] Select the minimum value among all the overall deviation values to form a comparison deviation value;

[0175] When the comparison deviation value is greater than the preset overall deviation threshold, calculate the sum of the normalized position deviation and the normalized direction deviation to form a total deviation;

[0176] Calculate the ratio of the normalized position deviation to the total deviation to form a position ratio;

[0177] When the position ratio is greater than the preset ratio threshold, reduce the preset occlusion ratio threshold according to the relative deviation between the comparison deviation value and the preset comparison deviation threshold and a preset adjustment coefficient to form the adjusted occlusion ratio threshold. The relative deviation between the comparison deviation value and the preset comparison deviation threshold is positively correlated with the adjusted occlusion ratio threshold;

[0178] When the position ratio is less than or equal to the preset ratio threshold, increase the preset learning rate according to the relative deviation between the comparison deviation value and the preset comparison deviation threshold and a preset adjustment coefficient to form the adjusted learning rate. The relative deviation between the comparison deviation value and the preset comparison deviation threshold is positively correlated with the adjusted learning rate.

[0179] The preset ratio threshold is a parameter used to measure whether the position ratio exceeds a certain standard value, which depends on the stability requirements of the system and the error tolerance that may occur during the flight. It is usually set between 0 and 1. In this embodiment, it is set to 0.7, which can effectively judge whether adjustment is needed during the flight, thus ensuring the accuracy and stability of the flight path and avoiding fluctuations caused by excessive adjustment.

[0180] The preset adjustment coefficient is a coefficient used to adjust the learning rate or other parameters, depending on the requirements of the flight mission, the error tolerance, and the sensitivity requirements of the system. It is usually set between 0 and 1 and is set to 0.1 in this embodiment. This helps to balance the response speed and stability during flight, avoid excessive or insufficient learning during the adjustment process, and ensure that the system can be optimized flexibly and smoothly according to real-time feedback.

[0181] By selecting the minimum value among the overall deviation values as the comparison deviation value, it is determined whether it is greater than the preset overall deviation threshold. If it exceeds the threshold, the sum of the position deviation and the direction deviation is calculated, and by comparing the position ratio with the preset ratio threshold, the occlusion ratio threshold or the learning rate is adjusted. When the position ratio is greater than the threshold, the occlusion ratio threshold is decreased; when the position ratio is less than or equal to the threshold, the learning rate is increased. Finally, the flight path and control strategy are optimized through these adjustments.

[0182] By adjusting the occlusion ratio threshold and the learning rate, this process can dynamically optimize the parameters according to the deviations during flight, thereby enhancing the adaptability and stability of the system. When there are large deviations in the path, by adjusting the occlusion ratio threshold, the influence of obstacle occlusion can be effectively avoided, improving flight safety; while in the case of small deviations, by adjusting the learning rate, the learning effect of the model can be better improved, ensuring that the system can be adjusted and corrected more accurately during flight. This flexible parameter adjustment strategy helps to improve the execution effect of the flight mission and reduce potential risks caused by environmental changes or deviations.

[0183] Specifically, selecting the actual flight route based on the adjusted occlusion ratio threshold or based on the adjusted learning rate includes:

[0184] When determining a number of first candidate routes based on the adjusted occlusion ratio threshold, select the first candidate route corresponding to the comparison deviation value as the actual flight route;

[0185] When obtaining the second pending flight route based on the adjusted learning rate, select the second pending flight route as the actual flight route.

[0186] When selecting the actual flight route, first based on the adjusted occlusion ratio threshold, if this threshold affects the selection of candidate routes, then through the comparison deviation value, select the corresponding first candidate route as the actual flight route; if the selection is based on the adjusted learning rate, then by selecting the second pending flight route, the actual flight route is further determined. These steps ensure the real-time optimization of the flight path and dynamically adjust the route selection according to the system feedback.

[0187] By dynamically adjusting the occlusion ratio threshold and the learning rate, the flight route can be flexibly optimized under different flight conditions. Through precise control of the selection of candidate routes, the errors and deviations during flight can be minimized to ensure flight safety and efficiency, and improve the success rate and stability of flight missions.

[0188] So far, the technical solution of the present invention has been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. An end-to-end UAV motion planning and control method for dynamic targets, characterized in that, Including: Obtaining the real-time occlusion ratio of the tracking target collected during the tracking of the UAV based on a preset tracking angle, the real-time density of the front obstacle, the real-time relative distance between the UAV and the front obstacle, and the real-time tracking distance between the UAV and the tracking target; Determining a number of first candidate routes according to the preset tracking angle, the real-time occlusion ratio, the real-time tracking distance, and a preset occlusion ratio threshold; Determining a number of second candidate routes according to each of the first candidate routes, a preset flight step length, and the real-time relative distance; Determining a number of first undetermined flight routes according to each of the second candidate routes, the preset flight step length, the real-time tracking distance, and the real-time density; Respectively obtaining the real-time position coordinates of the UAV, the tracking target, and the front obstacle in three-dimensional space; Inputting all the real-time position coordinates into a preset end-to-end prediction model to obtain a second undetermined flight route; Adjusting the preset occlusion ratio threshold according to the deviation between each of the first undetermined flight routes and the second undetermined flight route to form an adjusted occlusion ratio threshold, or adjusting the preset learning rate of the preset end-to-end prediction model to form an adjusted learning rate; Selecting an actual flight route based on the adjusted occlusion ratio threshold, or selecting an actual flight route based on the adjusted learning rate; Controlling the UAV to move along the actual flight route.

2. The end-to-end UAV motion planning and control method for dynamic targets according to claim 1, characterized in that Determining a number of first candidate routes according to the preset tracking angle, the real-time occlusion ratio, the real-time tracking distance, and the preset occlusion ratio threshold includes: When the real-time occlusion ratio is greater than the preset occlusion ratio threshold, calculating the change speed of the real-time tracking distance within a preset historical window in the past to form a tracking distance change rate; Performing linear extrapolation according to the tracking distance change rate and a preset prediction duration to form a predicted relative position; Determining a number of the first candidate routes according to the predicted relative position and the preset tracking angle.

3. The end-to-end UAV motion planning and control method for dynamic targets according to claim 2, wherein Determining a number of the first candidate routes according to the predicted relative position and the preset tracking angle includes: Calculating the predicted relative position and the relative displacement direction angle with the UAV as the zero point; Taking the relative displacement direction angle as a reference, and selecting a preset number of candidate heading angles according to a preset angle offset; Determining a number of first candidate angles according to each of the candidate heading angles and the preset tracking angle; Calculating all relative positions within a preset planning duration when advancing along each of the first candidate angles to determine a number of the first candidate routes.

4. The end-to-end UAV motion planning and control method for dynamic targets according to claim 3, wherein, Determining a number of first candidate angles according to each of the candidate heading angles and the preset tracking angle includes: Calculating the absolute values of the relative deviations between each of the candidate heading angles and the preset tracking angle to form a number of angle deviations; When the angle deviation is less than a preset angle deviation threshold, determining the candidate heading angle as the first candidate angle to determine a number of the first candidate angles.

5. The end-to-end UAV motion planning and control method for dynamic targets according to claim 4, characterized in that Determining a number of second candidate routes according to each of the first candidate routes, a preset flight step length, and the real-time relative distance includes: Obtaining the relative position of advancing the preset flight step length along the first candidate route to form a first expected position; Calculate an expected relative distance based on the first expected position and the real-time relative distance; When the expected relative distance is greater than a preset safety distance, determine the first candidate route as the second candidate route to determine a plurality of the second candidate routes.

6. The end-to-end UAV motion planning and control method for dynamic targets according to claim 5, characterized in that, Determining a plurality of first pending flight routes according to each of the second candidate routes, the preset flight step, the real-time tracking distance, and the real-time density includes: Calculate the standard deviation of all the real-time tracking distances within a preset past historical duration to form a tracking distance fluctuation value; Calculate the standard deviation of all the real-time densities within the preset past historical duration to form a density fluctuation value; Determine a plurality of first pending flight routes according to each of the second candidate routes, the preset flight step, the tracking distance fluctuation value, and the density fluctuation value.

7. The end-to-end UAV motion planning and control method for dynamic targets according to claim 6, characterized in that Determining a plurality of first pending flight routes according to each of the second candidate routes, the preset flight step, the tracking distance fluctuation value, and the density fluctuation value includes: Calculate the correlation coefficient of the tracking distance fluctuation value and the density fluctuation value to form a consistency; When the consistency is greater than a preset consistency threshold, obtain a relative position of advancing the preset flight step along the second candidate route to form a second expected position; Calculate an expected tracking distance according to the second expected position and the real-time tracking distance; When the expected tracking distance is within a preset standard tracking distance range, determine the second candidate route as the first pending flight route to determine a plurality of first pending flight routes.

8. The end-to-end UAV motion planning and control method for dynamic targets according to claim 7, characterized in that Adjust the preset occlusion ratio threshold according to the deviation between each of the first pending flight routes and the second pending flight routes to form an adjusted occlusion ratio threshold, or adjust the preset learning rate of the preset end-to-end prediction model to form an adjusted learning rate, including: Calculate the position deviation of each preset unit duration of each of the first pending flight routes and the second pending flight routes within the preset planning duration; Calculate the direction deviation of each preset unit duration of each of the first pending flight routes and the second pending flight routes within the preset planning duration; Perform normalization processing on the position deviation to form a normalized position deviation, and perform normalization processing on the direction deviation to form a normalized direction deviation; Perform weighted summation on the normalized position deviation, the normalized direction deviation, a preset position deviation weight, and a preset direction deviation weight to form a plurality of overall deviation values; Adjust the preset occlusion ratio threshold according to all the overall deviation values, the normalized position deviation, and the normalized direction deviation to form an adjusted occlusion ratio threshold, or adjust the preset learning rate to form an adjusted learning rate.

9. The end-to-end UAV motion planning and control method for dynamic targets according to claim 8, characterized in that, Adjust the preset occlusion ratio threshold according to all the overall deviation values, the normalized position deviation, and the normalized direction deviation to form an adjusted occlusion ratio threshold, or adjust the preset learning rate to form an adjusted learning rate, including: Select the minimum value among all the overall deviation values to form a comparison deviation value; When the comparison deviation value is greater than a preset overall deviation threshold, calculate the sum of the normalized position deviation and the normalized direction deviation to form a total deviation; Calculate the ratio of the normalized position deviation to the total deviation to form a position ratio; When the position ratio is greater than a preset ratio threshold, reduce the preset occlusion ratio threshold according to the relative deviation between the comparison deviation value and the preset comparison deviation threshold and a preset adjustment coefficient to form the adjusted occlusion ratio threshold; When the position ratio is less than or equal to the preset ratio threshold, increase the preset learning rate according to the relative deviation between the comparison deviation value and the preset comparison deviation threshold and the preset adjustment coefficient to form the adjusted learning rate.

10. The end-to-end UAV motion planning and control method for dynamic targets according to claim 9, characterized in that, Selecting an actual flight route based on the adjusted occlusion ratio threshold or selecting an actual flight route based on the adjusted learning rate includes: When determining a number of first candidate routes based on the adjusted occlusion ratio threshold, select the first candidate route corresponding to the comparison deviation value as the actual flight route; When obtaining a second undetermined flight route based on the adjusted learning rate, select the second undetermined flight route as the actual flight route.

Images