Anti-riot capture net gun unmanned aerial vehicle with multi-angle capture capability

Abstract

The invention relates to an anti-riot capturing net gun unmanned aerial vehicle with multi-angle capturing capability. The unmanned aerial vehicle comprises a flight control unit, a sensing prediction unit, a strategy generation unit, an execution control unit and a vector emission unit. The flight control unit is used for autonomously controlling the flight attitude and trajectory of the unmanned aerial vehicle and outputting flight attitude parameters and trajectory data; the perception prediction unit combines the parameters with the multi-modal sensor data to identify the target object and predict the motion trail of the target object; the strategy generation unit generates an optimal interception window and direction and outputs an interception direction instruction and interception time sequence information; the execution control unit controls a target steering cabin in the vector transmitting unit based on the information, and transmits a transmitting control instruction in a specified time window; and the vector transmitting unit adjusts the direction of the cabin body according to the instruction and transmits a net gun to realize air capture of a high-dynamic target. The unmanned aerial vehicle has high flexibility and multi-angle quick response capability, and is suitable for target control and capture in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) technology, and in particular to a riot control and capture net gun UAV with multi-angle capture capabilities. Background Technology

[0002] In existing technologies, target control in riot control scenarios typically relies on launchers carried by ground personnel or close-range, low-speed strikes by drones equipped with net guns of a fixed orientation. To improve aerial capture efficiency, some systems have introduced visual recognition modules and basic flight path planning technologies, enabling drones to approach and capture targets autonomously to a certain extent.

[0003] However, existing technologies generally suffer from poor adaptability to target dynamics, fixed launch direction, and narrow response windows. Especially when facing rapidly changing, highly dynamic targets, fixed-orientation launchers struggle to achieve effective capture, resulting in low capture success rates and a high risk of accidental damage. Furthermore, multi-angle collaborative control and trajectory prediction often fail to form a closed-loop linkage, limiting the UAV's practical application capabilities in complex environments.

[0004] Therefore, there is an urgent need to provide a riot control and capture net gun drone with multi-angle capture capabilities. Summary of the Invention

[0005] This application provides a riot control net gun drone with multi-angle capture capability to improve the drone's ability to accurately capture moving targets from multiple angles in all space under high dynamic environments.

[0006] This application provides a riot control and capture net gun drone with multi-angle capture capability, including:

[0007] The flight control unit is used to control the attitude, position and trajectory of the UAV based on ground commands and autonomous navigation strategies, and outputs the current flight attitude parameters and flight trajectory data.

[0008] The perception and prediction unit is used to identify the target object and estimate its current three-dimensional position, velocity vector and future motion trajectory, as well as the environmental perception data collected by the multimodal sensor, based on the flight attitude parameters and flight trajectory data, thereby obtaining the target motion trajectory data of the target object.

[0009] The strategy generation unit is used to construct a multi-angle interception orientation matrix based on the target motion trajectory data and flight attitude parameters, calculate the optimal interception window and corresponding interception direction of the target object, and generate interception direction instructions and interception timing information.

[0010] The execution control unit is used to coordinate the various steering compartments of the vector launch unit, activate the target steering compartment according to the interception direction command, and issue launch control commands to the vector launch unit within the time window indicated by the interception timing information.

[0011] The vector launch unit is used to control the pneumatic launch assembly in the target turning cabin to complete the orientation adjustment and net gun launch according to the launch control command. The vector launch unit includes at least four turning cabins distributed on different spatial surfaces of the UAV shell, and each turning cabin is equipped with a rotatable and positionable pneumatic launch assembly.

[0012] The beneficial effects of this application mainly include: (1) By setting multiple steering cabins distributed on different spatial surfaces of the UAV shell, and equipping each cabin with a rotatable and positionable aerodynamic launch component, the UAV has the ability to launch in all directions of space. The launch direction can be dynamically adjusted according to the target position, which significantly improves the flexibility and effective coverage of aerial capture. (2) By using the perception prediction unit to fuse flight status and multimodal sensor data, the UAV can accurately identify the three-dimensional motion state of the target object and predict its trajectory, providing high-confidence motion information support for the formulation of subsequent capture strategies and improving the system's response accuracy in complex dynamic scenarios. (3) The strategy generation unit generates the optimal interception window and interception direction based on the predicted trajectory and flight attitude parameters, and outputs time-sensitive control commands, making the capture decision forward-looking and real-time, ensuring a high degree of matching between the launch timing and direction, thereby improving the interception success rate and reducing false launches. (4) By linking the execution control unit and the vector launch unit, the UAV can accurately select and launch the target steering cabin, effectively avoiding energy waste and misoperation of the entire aircraft, and has high execution efficiency and safety control capabilities, which is suitable for the rapid control needs of high-risk anti-riot scenarios. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of a riot control and capture net gun drone with multi-angle capture capability provided in the first embodiment of this application. Detailed Implementation

[0014] Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below.

[0015] The first embodiment of this application provides a riot control and capture net gun drone with multi-angle capture capabilities. Please refer to... Figure 1 This figure is a schematic diagram of the first embodiment of this application. The following is in conjunction with... Figure 1The first embodiment of this application provides a riot control and capture net gun drone with multi-angle capture capability.

[0016] The anti-riot capture net gun drone with multi-angle capture capability includes a flight control unit 101, a perception and prediction unit 102, a strategy generation unit 103, an execution control unit 104, and a vector launch unit 105.

[0017] The flight control unit 101 is used to control the attitude, position and trajectory of the UAV based on ground commands and autonomous navigation strategies, and output the current flight attitude parameters and flight trajectory data.

[0018] The flight control unit 101 is one of the core control components in the anti-riot capture net gun drone with multi-angle capture capability described in this invention. Its main function is to calculate and manage the flight status of the drone in real time, ensuring that the drone can perform precise maneuvers, hover stably or flexibly track target objects according to mission requirements, and provide high-precision flight status basic data for subsequent target identification and interception operations.

[0019] The flight control unit 101 integrates various high-performance sensors and embedded processors, including but not limited to a three-axis gyroscope, a three-axis accelerometer, a magnetometer, a barometric altimeter, a GNSS positioning module, and attitude calculation algorithms and flight control software stacks embedded in the processor. This unit can receive mission parameters from the ground command system in real time, including control information such as the flight area, target number, and prohibited area boundaries. Combined with an autonomous navigation strategy, it automatically performs closed-loop control of attitude angles (including pitch, roll, and yaw), spatial position (three-dimensional coordinates), and trajectory changes during flight.

[0020] During implementation, the flight control unit 101 first estimates the current flight attitude parameters of the UAV, namely pitch angle, roll angle, and yaw angle, in real time based on the acceleration and angular velocity data collected by the inertial measurement unit (IMU) and using fusion algorithms such as Kalman filter or extended Kalman filter. This data is then combined with the geographic location information and velocity vector provided by the GNSS module to form complete flight trajectory data. To ensure the stability of the attitude calculation, the flight control unit 101 is typically set with an attitude update frequency of no less than 100Hz and a trajectory update frequency of no less than 10Hz to meet the high dynamic response requirements of subsequent systems.

[0021] In addition, the flight control unit 101 also includes a flight mode switching management module for dynamically switching between various modes such as autocruise, hovering, path tracking, and obstacle avoidance. For example, during the initial takeoff phase, the system can use an automatic climb mode, and after approaching the target area, it can enter a hovering mode to wait for the perception and prediction unit 102 to lock onto the target. If the target moves rapidly, the flight control unit 101 switches to path tracking mode and adjusts the flight trajectory in real time according to the target's movement trajectory to ensure that the launch platform maintains a suitable pitch angle and stable speed. When potential collision risks are identified, it will also automatically switch to obstacle avoidance mode and adjust the flight path through sensor feedback.

[0022] The output of the flight control unit 101 includes two parts: flight attitude parameters and flight trajectory data. Flight attitude parameters refer to the current attitude state of the UAV in three-dimensional space, generally represented by Euler angles or quaternions, used to describe the UAV's current orientation. Flight trajectory data is the sequence of the UAV's past and present spatial positions, typically including latitude, longitude, altitude, horizontal speed, and heading angle. Both types of data are transmitted in real-time to the perception and prediction unit 102 via the communication interface of the flight control unit 101 for accurate calculation of relative position, angle, and velocity during target recognition.

[0023] In terms of software implementation, the flight control unit 101 can run on an embedded flight control platform based on a real-time operating system. Its control law design can employ model predictive control (MPC), PID control, or sliding mode control methods, optimized according to platform performance and mission complexity. For hardware, embedded main control chips with high-speed I / O, floating-point operation capabilities, and peripheral interface support can be selected, such as the STM32 series, PX4 flight control platform, or a customized FPGA+MCU hybrid architecture, to support stable and highly reliable flight mission execution.

[0024] In summary, the flight control unit 101 plays a crucial role in this invention by providing high-precision attitude and trajectory data, supporting perception and strategy calculation, and maintaining the stable flight attitude of the entire aircraft.

[0025] Furthermore, the flight control unit is specifically used for:

[0026] Based on the attitude sensor data integrated in the flight control unit, a real-time three-dimensional quaternion attitude calculation module is constructed. By comparing the flight attitude parameters with the initial installation direction of each steering cabin in the UAV inertial coordinate system, the heading offset sensitivity map is calculated in real time to reflect the distribution of launch dead angle changes that may occur due to attitude deviation of each cabin under the current flight attitude.

[0027] The heading offset sensitive map is used as input and fused with the current predicted flight trajectory of the UAV for analysis. The trajectory control strategy output by the flight control unit is dynamically evaluated to determine whether it will continue to meet the angular reachability conditions of the target interception window in multiple time segments in the future. If it does not meet the conditions, micro-attitude adjustment is performed in advance, and the changes in directional availability of each steering cabin before and after the correction are fed back to the strategy generation unit in the form of a structured matrix.

[0028] During flight control, the spatial coupling relationship between the flight attitude parameters and the target motion trajectory data of the target object provided by the perception and prediction unit is continuously monitored. A cooperative trajectory strength factor is constructed. This cooperative trajectory strength factor is used to quantify the momentum compatibility between the UAV flight direction and the target object motion direction. The cooperative trajectory strength factor is provided as a dynamic priority value to the execution control unit to optimize the activation order of multiple steering cabins in the vector launch unit.

[0029] The flight attitude parameters, heading offset sensitivity map, angle reachability correction results, and cooperative trajectory intensity factor are periodically broadcast to the perception prediction unit, the strategy generation unit, and the execution control unit using a unified timestamp binding structure. This ensures that each functional module of the system can perform identification judgment, path generation, and launch scheduling decisions based on consistent and real-time flight control information in highly dynamic flight scenarios.

[0030] The flight control unit described in this embodiment is not merely a traditional flight control module that outputs attitude angles, speed, and acceleration, but an intelligent control core that actively participates in determining the feasibility of net gun firing, managing attitude corrections, optimizing spatial coupling, and coordinating cross-module data. It possesses technical features such as distributed multi-cabin configuration, dynamic collaborative interception, and high-precision directional control.

[0031] The flight control unit first relies on its integrated high-precision inertial measurement unit (IMU) to collect data from three-axis gyroscopes, three-axis accelerometers, and three-axis magnetometers, acquiring attitude changes in real time at a sampling frequency of 100 to 500 Hz. To avoid gimbal lock-up issues that occur with conventional Euler angle representations during large-angle movements, the flight control unit uses quaternions as the expression for flight attitude parameters. Quaternions are a rotational state representation method consisting of four real numbers, offering advantages such as high computational stability, absence of singularities, and ease of multi-axis attitude fusion.

[0032] During the calculation process, the flight control unit constructs a three-dimensional quaternion attitude calculation module, with the core algorithm being the extended Kalman filter. At each time step, this filter uses the current gyroscope integral value as the prediction input and the attitude direction obtained from the accelerometer and magnetometer as the observation input. It corrects the quaternion representation of the current flight attitude through a prediction-update-fusion process. The final quaternion error output by this module is controlled to be less than 0.5 degrees in the inertial coordinate system, thus ensuring that the attitude reference provided by the flight control unit is highly reliable during high-dynamic flight.

[0033] After obtaining high-precision flight attitude parameters, the flight control unit compares this data with the initial installation orientation of each steering pod in the UAV's inertial coordinate system. Each steering pod's spatial installation orientation is calibrated during the design and manufacturing phase and written into the spatial configuration table in the flight control unit during system initialization. The flight control unit maps the orientation of each pod from the airframe coordinate system to the current launch orientation in the inertial coordinate system through attitude transformation relationships and analyzes the impact of the current flight attitude on the launch capability of each pod.

[0034] Based on the above transformations, the flight control unit constructs a "heading offset sensitivity map," which is described in the form of a spatial vector field. This map shows the distribution of the offset between the original design launch direction and the current actual attitude direction of each steering module under the current flight attitude. For each module, the map records the changing trend of the angle between its launch vector and the target direction, and uses a deviation threshold to determine whether there is a launch blind spot, i.e., the original design direction has deviated from the capture area due to attitude changes. For example, if a module is designed to face forward and downward, but is turned to the opposite direction due to a large pitch angle of the UAV, its launch will be unusable. The data structure of this map is a multi-channel matrix, with each channel corresponding to one module, and its value being the change in angle between the launch direction and the target direction on a unit sphere.

[0035] The flight control unit then fuses the heading offset sensitivity map with the current predicted flight trajectory for analysis. Trajectory prediction is performed by the position estimation module within the flight control unit. This module uses the current position, velocity, and acceleration as inputs and employs motion equations to perform short-term forward predictions (typically 0.5 to 2 seconds into the future). For each future time point, the flight control unit applies attitude quaternion extensions to the predicted trajectory to determine if, during trajectory evolution, there are any time intervals where a specific steering module consistently fails to align with the target direction. If the analysis indicates that the current flight trajectory will cause a loss of angular accessibility for some modules within the expected interception time window, the flight control unit will intervene early to perform minor attitude adjustments. For example, if the target's direction of motion shifts to the left, the flight control unit can slightly adjust the current roll angle to the left by 5 to 10 degrees, thereby restoring the directional availability of the left rear module. This attitude adjustment is performed by the flight control unit executing control surface adjustment commands and does not change the overall navigation target; it only performs minor perturbation optimization based on the original trajectory.

[0036] To feed back the effects of this attitude optimization behavior to the strategy generation unit, the flight control unit constructs a structured matrix after completing the aforementioned actions. This matrix records the change in the angle between each steering module and the target direction on a unit sphere before and after adjustment. Each element in this matrix corresponds to the angular deviation of a specific module in a particular launch candidate direction. The strategy generation unit can adjust the intercept direction command and timing information based on this matrix to avoid hitting unavailable directions.

[0037] During flight control, the flight control unit continuously monitors the spatial coupling relationship between its own flight attitude parameters and the target trajectory data provided by the perception and prediction unit. To this end, the flight control unit constructs a "cooperative trajectory strength factor," which measures the consistency between the UAV's own velocity direction and the target object's motion direction in momentum space. The construction method is as follows: the flight control unit normalizes its own velocity vector and the target's motion direction vector, and calculates the angle between them. The smaller the angle, the more fully the flight direction aligns with the target direction, and the higher the probability of successful capture. Combining the UAV's current flight speed, the target speed, and the speed change trends of both, it assesses whether there is a tendency for consistency (i.e., whether they are pursuing in the same direction) and assigns a value between 0 and 1 as the coupling strength. The stronger the coupling, the higher the interception confidence. This factor is provided as a dynamic priority value to the execution control unit, used to select the module with the most consistent flight state and the highest launch success rate when multiple modules are simultaneously capable of launching.

[0038] To ensure time consistency across the entire system during dynamic, high-frequency operation, the flight control unit timestamps all the aforementioned critical data, including flight attitude parameters, heading offset sensitivity maps, angle reachability correction results, and cooperative trajectory strength factors. This binding is based on an internal real-time clock, with all data structures having millisecond-level absolute timestamps added during generation. These timestamps are then periodically broadcast in structured message format to the perception and prediction unit, policy generation unit, and execution control unit. The broadcast period can be set to 50ms or faster to ensure that the three units make judgments, control actions, and transmissions based on the same timing during mission execution, preventing command failures or acquisition deviations due to delays or asynchrony.

[0039] The perception and prediction unit 102 is used to identify the target object and estimate its current three-dimensional position, velocity vector and future motion trajectory based on the flight attitude parameters and flight trajectory data, as well as the environmental perception data collected by the multimodal sensor, thereby obtaining the target motion trajectory data of the target object.

[0040] The perception and prediction unit 102 is the core perception and analysis module in the anti-riot capture net gun UAV with multi-angle capture capability of the present invention. Its main function is to perform spatial identification, state estimation and motion trajectory prediction of the target object based on the flight attitude parameters and flight trajectory data provided by the flight control unit 101 and the environmental perception data collected by the multi-modal sensors, so as to provide reliable and real-time target motion trajectory data for the subsequent strategy generation unit 103.

[0041] The perception and prediction unit 102 includes at least one central processing unit, a data fusion module, an image processing and recognition module, a target tracking and prediction module, and several interfaces for acquiring multimodal perception data. The multimodal sensors may include visible light cameras, infrared thermal imagers, laser ranging modules, millimeter-wave radar, or structured light depth sensors, capable of acquiring complete environmental information under different lighting conditions, occlusion, and complex scenarios. These sensors are installed in different directions on the UAV platform and transmit data to the perception and prediction unit 102 via a high-speed bus, ensuring data latency is less than 50 milliseconds to support real-time analysis.

[0042] During implementation, the perception and prediction unit 102 first receives flight attitude parameters and flight trajectory data output by the flight control unit 101. This data is used to convert the raw images or point cloud data collected by each sensor into a unified inertial space coordinate system for geometric correction and pose calibration. Subsequently, the perception and prediction unit 102 performs feature extraction operations on multiple frames of images or sensor data sequences, including edge detection, temperature distribution extraction, and laser echo intensity analysis, to construct feature vectors describing suspicious moving targets in the scene. Using deep convolutional neural networks (such as YOLO) or traditional machine vision algorithms (such as HOG+SVM), preliminary identification and classification of the target object are completed, including contour recognition, heat source confirmation, and occlusion judgment.

[0043] After completing target recognition, the perception and prediction unit 102 initiates the target tracking and prediction module. Utilizing temporal feature modeling algorithms, such as Kalman filtering, extended Kalman filtering, or nonlinear time series prediction methods based on LSTM networks, it dynamically models the spatial position, velocity vector, and acceleration of the target object across consecutive frames. The perception and prediction unit 102 represents the target object's three-dimensional position at the current moment as a point (x, y, z) in a Cartesian coordinate system, and its velocity vector as vx, vy, vz. It further calculates the predicted position set over several future time steps, forming complete target motion trajectory data. This prediction process comprehensively considers the UAV's own flight trajectory and sensor perception delays to ensure the prediction results are both engineering-feasible and real-time.

[0044] It is worth emphasizing that, in handling complex scenes with partial occlusion, rapid turning, or multi-target interference, the perception and prediction unit 102 can also introduce an occlusion-robust tracking strategy. Through dynamic feature weighting and multi-frame similarity discrimination mechanisms, it effectively maintains target continuity and suppresses misjudgment. The target motion trajectory data output by this unit not only includes the target's position information in the future time period, but also the confidence score, prediction error upper bound, and confidence ellipsoid estimation results corresponding to the current frame, which are used to assist the policy generation unit 103 in priority judgment and risk assessment.

[0045] To ensure the real-time performance and reliability of the overall system, the perception and prediction unit 102 can run on a high-performance edge computing platform, such as RK3588 or a custom FPGA acceleration board, and adopts lightweight neural network model compression and tensor quantization technology to ensure millisecond-level image processing and prediction calculation on the power-constrained UAV platform.

[0046] Through the above structure and process, the perception and prediction unit 102 can not only accurately identify and track target objects moving at high speed or along complex paths, but also predict their future trajectory based on the current flight status and environmental data, thereby providing the strategy generation unit 103 with complete, continuous target motion trajectory data with time dimension, ensuring the accurate closed-loop response capability of the entire capture chain.

[0047] Furthermore, the perception and prediction unit is specifically used for:

[0048] During the target object recognition process, in cases where there is occlusion or image degradation in the environmental perception data collected by the multimodal sensor, the contour features and thermal intensity features associated with the target object are extracted from the infrared thermal imaging image and visible light image frame sequences, respectively, and an inter-frame velocity vector estimation sequence is constructed.

[0049] The velocity vector estimation sequence and infrared thermal intensity features are input into a confidence weighted model to generate a confidence weight matrix. Based on this confidence weight matrix, multiple candidate trajectory segments formed by the contour features are weighted and fused to output a continuous frame-level position estimation sequence.

[0050] A prediction error ellipsoid model is constructed based on the continuous frame-level position estimation sequence. The prediction error ellipsoid model is used to quantify the trajectory estimation uncertainty of the current target object and output the ellipsoid principal axis direction and error covariance matrix.

[0051] The shape of the tracking region and the sampling resolution range of the next frame image are dynamically adjusted according to the ellipsoid principal axis direction and the error covariance matrix, and the correction window is input into the subsequent image processing flow to optimize the target recognition robustness of the multimodal sensor.

[0052] When the environmental perception data is continuously missing or the confidence weight matrix is ​​lower than the preset confidence threshold, the trajectory compensation mechanism is invoked. The continuous frame-level position estimation sequence is used as input, and time series modeling is performed through a long short-term memory neural network model or a Kalman filter model. The target motion trajectory data of the target object is output as the final trajectory estimation result.

[0053] This implementation primarily addresses situations in complex dynamic environments where the target object may be occluded, or experience image degradation, blurring, or lighting interference. It ensures that the perception and prediction unit can stably and continuously output the target object's motion trajectory data for subsequent strategy generation, thereby enabling intelligent decision-making regarding the net gun's firing direction and timing. The core technical approach of this implementation is: fusing multimodal sensing information, constructing a robust trajectory estimation mechanism based on confidence levels, and employing a time-series compensation model to restore trajectory continuity under occlusion or anomaly conditions.

[0054] First, during target object recognition, the system needs to simultaneously receive environmental perception data from multimodal sensors. Here, multimodal refers to data that simultaneously includes visible light image data and infrared thermal imaging image data. Visible light image frame sequences are continuous frame images acquired by ordinary image sensors or cameras, possessing good spatial texture resolution capabilities; while infrared thermal imaging images are thermal radiation image data acquired in real time by an infrared thermal imaging module, offering advantages such as the ability to penetrate smoke and maintain visibility in low light conditions. The system inputs these two types of image frames into the feature extraction module respectively.

[0055] During the feature extraction stage, the system performs feature separation processing on the target object in each frame of the image. In the visible light image frame sequence, geometric edge features related to the target contour are extracted first, such as the object's outline, closed regions, and edge segments with significant changes in directional gradient. Semantic edge regions extracted using Canny, Sobel, or U-Net models are commonly employed. In infrared images, the system extracts thermal intensity regions, i.e., closed hotspot regions with temperatures higher than the background. These regions typically represent humans, animals, vehicle engines, or other targets with significant thermal radiation. The system performs region binarization, region connectivity determination, and maximum response localization operations on the target regions in the thermal image to obtain the thermal intensity center point and hotspot contour.

[0056] Next, the system needs to model the positional changes of the same target object between adjacent frames in the image sequence, thereby constructing an inter-frame velocity vector estimation sequence. This sequence is constructed based on calculating the motion vector of the target's center point frame by frame in temporal order. If the target center coordinates are extracted at times t and t+1, the system calculates its spatial displacement and divides it by the time interval to obtain the velocity estimate for that time period. To improve robustness, the system can use optical flow methods (such as Lucas-Kanade optical flow) or keypoint matching methods (such as ORB or SIFT feature matching) to determine the target displacement across frames. If target detection fails in a frame, the system will use the predicted velocity vector from the previous frame for linear extrapolation.

[0057] To further enhance prediction reliability in occlusion scenarios, the system inputs the extracted inter-frame velocity vector estimation sequence and infrared thermal intensity features into a confidence-weighted model. This model evaluates the confidence of trajectory candidates in each frame and weights and fuses multiple trajectory segments using a weight matrix. The confidence-weighted model can employ a logistic regression-based or shallow neural network structure, where input features include velocity vector stability (velocity change amplitude), thermal intensity distribution consistency (the degree of overlap between the current hotspot and the historical average hotspot), and contour matching similarity (such as IoU overlap rate). The model output is a confidence value between 0 and 1, representing the confidence level of the trajectory segment in the current frame. All trajectory candidates are weighted and fused based on their corresponding confidence values ​​to form a continuous frame-level position estimation sequence, representing the predicted trajectory of the target object from time t to frame t+N.

[0058] To further address the uncertainty in the predicted trajectory, the system constructs a prediction error ellipsoid model for the aforementioned frame-level position estimation sequence. The error ellipsoid measures the spatial distribution boundary of trajectory points, especially in cases of occlusion or image degradation. The principal axis direction and magnitude of the ellipsoid reflect the main direction and intensity of the prediction uncertainty. The construction method is as follows: First, the system calculates the covariance of the position estimation points within N frames, obtaining a covariance matrix in three-dimensional space, where the diagonal elements represent the variances in the X, Y, and Z directions, respectively. Then, the covariance matrix is ​​decomposed eigenvalues ​​to obtain eigenvectors and corresponding eigenvalues. The eigenvectors represent the principal axis directions of the ellipsoid, and the eigenvalues ​​are the squares of the lengths of each axis of the ellipsoid. This error ellipsoid model can intuitively represent the direction and magnitude of the spatial uncertainty of the current prediction result.

[0059] The output of the prediction error ellipsoid model is used to dynamically adjust the shape of the tracking region and the sampling resolution range of the next frame image. The tracking region is adjusted as follows: the principal axis of the error ellipsoid is used as the principal direction of the prediction window for the next frame. The axis length of the error ellipsoid in this direction is multiplied by the confidence weighting factor to obtain the major axis dimension of the search window. Simultaneously, the minor axis dimension is set according to the lateral axis length, thus constructing a tilted, asymmetric ellipsoidal search region. Furthermore, the system estimates the target blur level based on the overall volume of the ellipsoid. If the volume is large, the sampling resolution is automatically reduced to improve tracking speed; if the volume is small, the resolution is increased to improve recognition accuracy. Finally, this search window will be used in subsequent image processing steps to optimize the recognition accuracy of the target object in multimodal images.

[0060] When the system fails to extract effective features from the environmental perception data for multiple consecutive frames, or when the overall average value of the confidence weight matrix output by the confidence weighted model is lower than the set confidence threshold, it indicates that the current visual information quality is insufficient to support reliable tracking. The system will then invoke a trajectory compensation mechanism to prevent trajectory interruption. This compensation mechanism is based on time-series modeling methods, extracting historical trajectory data from the aforementioned consecutive frame-level position estimation sequences to construct a prediction model for generating future trajectories.

[0061] The compensation mechanism supports two models: a nonlinear trajectory prediction method based on a Long Short-Term Memory (LSTM) neural network model and a linear state estimation method based on Kalman filtering. In LSTM modeling, the system uses time points and position vectors from the position sequence as input to construct time-series features. A three-layer gated neural network is then used to learn the latent state and model long-term dependencies. Training can utilize known occluded segments from the target dataset as labeled data. In Kalman filtering modeling, the system uses standard state-space modeling, defining the target state as a joint vector of position and velocity. State update and prediction equations are established using the state transition matrix and observation matrix, and a Bayesian filtering process is used to continuously output state estimates.

[0062] Regardless of the model used, the final output is the predicted position of the target object at consecutive future moments. This result is regarded as the target motion trajectory data of the target object and is passed to subsequent modules as one of the inputs required by the strategy generation unit. This ensures that the entire capture system still has trajectory continuity and robustness in scenarios with occlusion interference, image distortion, or visual perception failure.

[0063] The strategy generation unit 103 is used to construct a multi-angle interception orientation matrix based on the target motion trajectory data and flight attitude parameters, calculate the optimal interception window and corresponding interception direction of the target object, and generate interception direction instructions and interception timing information.

[0064] The strategy generation unit 103 is a key decision-making module in the anti-riot capture net gun UAV with multi-angle capture capability described in this invention. It is mainly used to fuse the target motion trajectory data provided by the perception and prediction unit 102 with the flight attitude parameters provided by the flight control unit 101, establish a multi-angle interception orientation matrix, and determine the optimal interception window of the target object and its corresponding interception direction accordingly. Finally, it generates interception direction instructions and interception timing information, providing accurate and executable decision-making basis for subsequent launch control actions.

[0065] In actual operation, the strategy generation unit 103 first receives the target motion trajectory data output by the perception and prediction unit 102. This trajectory data typically includes the target's three-dimensional predicted position sequence at multiple future moments, current velocity vector, heading change trend, prediction confidence level, and its error range. Simultaneously, the strategy generation unit 103 receives flight attitude parameters output by the flight control unit 101, including pitch angle, roll angle, and yaw angle, to determine in real time whether the UAV's current spatial orientation and launch platform attitude have the capability to respond to a certain direction capture.

[0066] The strategy generation unit 103 has a built-in spatial geometric analysis module for constructing a multi-angle interception azimuth matrix. This module calculates the set of directions in which the UAV can effectively launch within a specific time window based on the current flight attitude parameters and the spatial distribution parameters of the multiple steering pods carried by the UAV. Based on this, the system matches the target trajectory provided by the perception and prediction unit 102 with this set of directions. Based on the probability that the target will enter the geometric firing arc of a specific launch direction at a future moment, the system dynamically evaluates the feasibility of all candidate directions and the expected benefits of successful capture.

[0067] The construction process of the multi-angle interception azimuth matrix is ​​based on the UAV coordinate system and combines the spatial distribution parameters of the steering pods on the airframe structure. The launch direction of each pod is modeled as a unit direction vector, and all direction vectors form a direction set. Each vector in the direction set represents a specific spatial sector that the steering pod can cover after rotational adjustment. In the specific implementation, the system first normalizes these direction vectors and represents them as a combination of azimuth and pitch angles in polar coordinates, for example, recorded as a two-dimensional array θ. i , Subsequently, the strategy generation unit 103 combines the current flight attitude parameters provided by the flight control unit 101 and uses Euler angle transformation or quaternion transformation to map the launch vectors of each cabin to the inertial space coordinate system, forming a global orientation model facing the ground target.

[0068] The matrix is ​​ultimately presented as a sparse direction coverage table, where each row represents a possible launch direction, and each column records its corresponding module number, direction vector, degree of adaptation to the current attitude (e.g., required rotation angle), timestamp of the predicted hit window, and angle with the target's current trajectory, among other multi-dimensional indicators. Based on this, the system establishes an evaluation function that comprehensively considers factors such as angle deviation, rotation time, and target approach velocity, assigning a score to each candidate option in the matrix. By sorting and filtering this matrix, a sequence of candidate launch directions that satisfy the optimal acquisition timing and direction constraints can be determined, and the optimal interception direction corresponding to the target interception window can be ultimately selected. This matrix not only provides a structured decision-making basis for subsequent launch control but also enables quantitative scheduling support for multi-module collaborative acquisition capabilities.

[0069] To accurately pinpoint the optimal interception timing, the strategy generation unit 103 introduces a time-related window analysis mechanism. This mechanism combines the target's velocity change trend and flight attitude response delay to determine at what point in time and in which spatial direction the target will achieve launch reachability from the current moment. Following the principles of minimizing advance response, maximizing success rate, and minimizing flight adjustments, the system selects a set of optimal interception windows and corresponding interception directions from all feasible directions. Here, the "interception window" refers not only to a specific time point but also to the allowable deviation time interval, typically defined in milliseconds, to ensure fault tolerance in launch control.

[0070] Based on the above calculations, the strategy generation unit 103 generates interception direction commands and interception timing information. The interception direction commands are used to explicitly specify the spatial azimuth number of the target turning hull or the corresponding launch direction vector, while the interception timing information is used to define the specific launch time point or time range. This information will be output to the execution control unit 104 in the form of structured data, which will then complete the subsequent launch action scheduling.

[0071] The strategy generation unit 103 can be implemented on an embedded multi-threaded processing platform, employing a modular structure to divide functional sub-modules such as sensing data reception, direction matrix calculation, trajectory prediction and alignment, optimization algorithm execution, and control parameter output. The optimization algorithm can utilize a cost function-based evaluation mechanism, which simultaneously considers multiple factors such as target proximity, launch direction coverage, prediction error range, flight adjustment load, and control response delay, to numerically optimize the globally optimal launch scheme. This unit can achieve efficient processing via CPU or high-parallel computation using GPU or FPGA resources, meeting the real-time response requirements under complex conditions.

[0072] In summary, the strategy generation unit 103 combines the dynamic target behavior modeling results with its own flight state to construct a multi-angle decision space oriented towards spatial structure constraints, and extracts the optimal launch direction and time window with the highest probability of strike from it, so that the subsequent execution control links are highly targeted and efficient.

[0073] Furthermore, the strategy generation unit is specifically used for:

[0074] Based on the spatial distribution parameters of each steering cabin in the vector launch unit on the UAV shell, an initial direction vector for each steering cabin is constructed.

[0075] Based on the flight attitude parameters output by the flight control unit, an attitude transformation operation is performed on the initial direction vector. Quaternion rotation or Euler angle transformation is used to map the initial direction vector to the inertial space coordinate system to form a multi-angle interception orientation matrix.

[0076] Assign a directional reachability score and a cabin adjustment cost estimate to each directional vector in the multi-angle interception orientation matrix. The directional reachability score is used to reflect the probability that the target object enters the spatial region of that direction in the target motion trajectory data of the target object output by the perception and prediction unit. The cabin adjustment cost estimate is used to characterize the adjustment time or energy consumption required for the steering cabin to rotate from the current attitude to that direction.

[0077] Based on the target motion trajectory data of the target object, a set of candidate interception directions with an angle less than a preset threshold between the target object and the direction vector is identified. The set of candidate interception directions is then sorted according to the direction reachability score and the estimated cost of cabin adjustment. Finally, the optimal interception window and the corresponding interception direction are determined, and interception direction instructions and interception timing information are generated.

[0078] First, the strategy generation unit receives structural parameters from the vector launch unit. Each steering module has a fixed spatial position and initial orientation in the body coordinate system. Let the launch direction of the t-th steering module be... for:

[0079]

[0080] in, Let represent the initial direction vector of the t-th compartment. This vector is a unit vector and represents its standard launch direction in the body coordinate system. For example, the forward-facing compartment could be .

[0081] Where x0 represents the component of the cabin orientation along the X-axis of the body coordinate system, which usually represents the direction of the drone's nose (i.e., forward); y0 represents the component of the cabin orientation along the Y-axis of the body coordinate system, which usually represents the horizontal direction of the drone's fuselage (i.e., left / right direction); and z0 represents the component of the cabin orientation along the Z-axis of the body coordinate system, which usually represents the vertical direction of the drone (i.e., up / down direction).

[0082] To adapt to the current flight attitude of the drone, each initial direction vector needs to be... Transforming to the inertial space coordinate system, we obtain the mapped direction:

[0083]

[0084] in, It is the orientation vector after attitude correction, representing the actual launch direction of the t-th cabin in the inertial coordinate system under the current flight attitude. It is the key data for subsequent trajectory matching and angle determination.

[0085] R body→world It is a 3×3 direction cosine matrix, i.e., attitude transformation matrix, used to rotate vectors in the body coordinate system to the inertial space coordinate system;

[0086] The flight attitude of a UAV is output in real time by the flight control unit. Common attitude representation methods include Euler angles (pitch, roll, yaw) and quaternions q = [q0, q1, q2, q3]. Quaternions have the advantages of no singularity and numerical stability, and are widely used in modern flight control systems.

[0087] If represented by quaternions as q = [q0, q1, q2, q3], then the attitude transformation matrix is:

[0088]

[0089] Where q0 is the scalar part of the quaternion, usually output by the inertial fusion algorithm of the flight control system (such as extended Kalman filter); q1, q2, and q3 are the vector parts of the quaternion, corresponding to the rotation information of the airframe around the three axes x, y, and z; each matrix element is derived from the standard quaternion rotation transformation, the principle of which is to rotate and map vectors in space through unit quaternions.

[0090] This transformation matrix is ​​used to rotate the direction vectors in the body coordinate system to the inertial coordinate system, thereby obtaining the set of multi-angle interception directions after attitude correction.

[0091]

[0092] in, It is the orientation vector after attitude correction, representing the actual launch direction of the t-th cabin in the inertial coordinate system under the current flight attitude; n is the number of turning cabins;

[0093] Next, the policy generation unit performs reachability scoring on each direction vector based on the target motion trajectory data of the target object output by the perception and prediction unit. Let the future position of the target object at time t be:

[0094]

[0095] in, This represents the three-dimensional position of the target object at a future time point t; the components x(t), y(t), and z(t) represent the east (X), north (Y), and vertical (Z) coordinates of the target object in the world coordinate system, respectively; this position is obtained by combining a perception prediction unit (such as multi-frame image sequences, LiDAR, infrared tracking data) with a time series prediction model (such as Kalman filtering, LSTM neural network);

[0096] The drone's current location is Then the angle α between the direction vector and the target direction t (t) is:

[0097]

[0098] in, This indicates the drone's current position coordinates, in units of 1 and 2. Consistency is expressed as follows:

[0099]

[0100] The coordinates are measured by the flight control unit using a combination of the inertial navigation system (INS) and the global positioning system (GPS); they can be set as a reference point in the world coordinate system (e.g., the takeoff point as the origin) and are continuously updated in the flight control system. x0 represents the spatial position coordinates of the UAV along the X-axis of the world coordinate system at the current moment. y0 represents the spatial position coordinates of the UAV along the Y-axis of the world coordinate system at the current moment. z represents the spatial position coordinates of the UAV along the Z-axis of the world coordinate system at the current moment.

[0101] The dot product is used to calculate the included angle.

[0102] Because the direction vector is a unit vector;

[0103] This refers to the spatial distance between the drone and the target object.

[0104] Based on the above angle calculation results, the directional reachability scoring function is defined as follows:

[0105]

[0106] Where: w1 and w2 are weight parameters set empirically, with recommended values ​​of 1.0 and 0.5 respectively;

[0107] σ is a distance scaling factor that controls the tolerance of exponential decay;

[0108] For ground-based crowd interception missions, a σ value of 8–12 meters is recommended.

[0109] For high-speed vehicles or wide-area surveillance, a σ = 20–30 meters is recommended.

[0110] The specific value of σ can be obtained through scene simulation parameter tuning or optimization in the training set.

[0111] In the above formula, the first term reflects the degree of directional matching, and the second term reflects the degree of proximity to the target.

[0112] To obtain the total accessibility score at multiple time points, the accessibility score function is integrated or discretely summed in the time domain:

[0113]

[0114] in, Represents a certain emission direction vector The cumulative directional reachability score within the time window [t0, t0+Δt] is used to comprehensively evaluate the overall probability of target capture in that direction; t0 represents the current moment, or the start time when the strategy generation unit begins evaluating the predicted trajectory; it is provided by the system clock or the flight control main control module; Δt is the length of the evaluation time window, i.e., the total duration of forward prediction from the current moment; K is the total number of samples, calculated according to the following formula:

[0115]

[0116] Where δt is the sampling interval, the recommended value is 0.05 seconds.

[0117] Simultaneously, the estimated cost of hull adjustment is calculated for each direction. Let the current hull orientation be... The target direction is The angle between the two is:

[0118]

[0119] in, It is a launch direction vector; β is the actual orientation unit vector of the t-th turning module at the current moment; t The spatial angle between the current direction and the target direction, usually expressed in radians (rad) or degrees (°), represents the minimum rotation angle required for the steering pod to change from its current attitude to the target launch direction.

[0120] Estimated cost of cabin adjustment The calculation can be based on rotation time and energy consumption, as shown in the following formula:

[0121]

[0122] in:

[0123] ω max The maximum angular velocity of rotation of the hull is expressed in rad / s.

[0124] k is the energy consumption factor required per unit angle of rotation (e.g., in J / rad);

[0125] The first term represents the required rotation time, and the second term is the energy consumption estimate.

[0126] After obtaining the above two metrics, the strategy generation unit further constructs a comprehensive scoring function for interception direction ranking:

[0127]

[0128] in:

[0129] α and β are the overall scoring weights; α controls the degree of preference for target proximity and direction matching; β controls the degree of penalty for cabin adjustment costs (time / energy consumption), which can be set through simulation training, field testing or engineering experience; among them, the recommended value for α is 1.0 and the recommended value for β is 0.5.

[0130] A higher score indicates a better direction.

[0131] Finally, the direction vector with the highest score is selected from all scoring candidates. And generate accordingly:

[0132] 1. Interception Direction Command: Identify the selected hull number τ opt and its corresponding direction vector

[0133] 2. Interception timing information: Calculate the most likely time point t when the target object enters the radiating field in this direction. fire And set the tolerance range Δ, and finally output the time window:

[0134] T fire =[t fire -Δ,tfire +Δ]

[0135] Among them, T fire Indicates the launch time window interval; t fire This indicates the optimal launch time, which is the time when the system predicts the target is most likely to enter the effective capture range in the current direction;

[0136] Δ is a time tolerance parameter used to define a safe time range that allows for a certain error or buffer in the transmission control command; the recommended value is 0.2 seconds.

[0137] This information will be received by the execution control unit, which will then issue a launch control command to the vector launch unit to complete the acquisition operation in the correct direction and at the right time.

[0138] The following is a reference implementation code for the strategy generation unit:

[0139]

[0140]

[0141]

[0142] The execution control unit 104 is used to coordinate the various steering compartments of the vector launch unit, activate the target steering compartment according to the interception direction command, and issue launch control commands to the vector launch unit within the time window indicated by the interception timing information.

[0143] In the anti-riot capture net gun UAV with multi-angle capture capabilities of the present invention, the execution control unit 104 plays a key bridging role between strategy generation and actual execution, and is responsible for converting the interception direction command and interception timing information output by the strategy generation unit 103 into specific launch control actions. This unit not only needs to accurately understand the intent of the capture strategy, but also must coordinate the working state of multiple steering cabins in the vector launch unit 105 under strict time control to ensure that the launch component in the selected direction performs the net gun launch operation within the appropriate time window.

[0144] In the specific implementation process, the execution control unit 104 first receives the interception direction command and interception timing information output by the strategy generation unit 103 via a data bus or dedicated interface. The interception direction command generally includes the identification information of the target turning hull, which may be represented in the form of a number, coordinate vector, or direction angle, while the interception timing information clearly indicates when the launch operation should be executed, usually expressed in the form of an absolute timestamp or relative time delay. The execution control unit 104 uses these two pieces of information as input conditions to enter the scheduling process.

[0145] In the scheduling process, the execution control unit 104 first calls the system's status query interface to retrieve the status of all current steering cabins, including the cabin's rotational alignment status, the pre-charge status of the aerodynamic launch assembly, whether the power supply is stable, and whether it is in a occupied or conflicting state. If the target steering cabin has not yet completed directional alignment or has not yet entered the ready-to-launch state, the execution control unit 104 will send a pre-activation signal to the steering cabin to initiate the directional adjustment action and launch preparation process, while continuously monitoring its feedback status until preparation is complete.

[0146] Before the time window set in the interception timing information approaches, the execution control unit 104 performs a synchronization operation to ensure that its local time is synchronized with the system clock or GPS time at the millisecond level, thus avoiding launch delays or advances due to time deviations. After confirming that all preparations are complete, when the system time is aligned with the time window specified in the interception timing information, the execution control unit 104 immediately issues a launch control command to the vector launch unit 105 to which the target turning hull belongs. This command includes specific start signals, allowable thresholds, duration, and optional compensation parameters to ensure that the aerodynamic launch assembly can complete an accurate launch in the predetermined orientation.

[0147] The execution control unit 104 also has a conflict management mechanism. If system interference, directional error, cabin abnormality, or external obstruction is detected during the launch preparation phase or launch execution, it will immediately interrupt the launch process and report the launch failure status to the strategy generation unit 103, so that the unit can regenerate alternative interception direction commands and timing information. This feedback mechanism is implemented through asynchronous communication and does not affect the real-time performance of the main flight control process.

[0148] From a hardware perspective, the execution control unit 104 can be deployed in the UAV's main control chip or an independent embedded microcontroller, possessing independent clock management, cabin communication interface, and fault tolerance capabilities. In high-concurrency scenarios, this unit can simultaneously manage cabin switching requests from multiple directions, and schedule launch logic in parallel through multi-threading or event-driven methods to maintain system response speed and command execution accuracy.

[0149] Through the above process and mechanism, the execution control unit 104 ensures the accurate implementation of the capture strategy, enabling the target turning cabin specified in the vector launch unit 105 to launch the net gun at a predetermined time, achieving a high success rate in intercepting high-speed, non-linear moving targets.

[0150] Furthermore, the execution control unit is specifically used for:

[0151] Based on the interception timing information generated by the strategy generation unit, the system time synchronization module is invoked to perform a high-precision launch time alignment operation. The system time synchronization module includes a global positioning system synchronization unit based on satellite navigation timing or a highly stable local clock with an integrated temperature-compensated crystal oscillator, which is used to align the execution time of the launch control command to the standard reference time under a unified system time base.

[0152] The center point of the time window and the allowable time offset tolerance field contained in the corresponding interception timing information are embedded in each generated launch control command. The execution control unit dynamically adjusts the aerodynamic trigger advance of the corresponding cabin according to the allowable time offset tolerance field to compensate for the cabin response delay and the rotational inertia of the servo motor, so as to achieve quasi-synchronous launch of each steering cabin in the vector launch unit.

[0153] When the launch control command planned launch time of multiple vector launch units corresponding to the turning cabins falls within the same interception timing information time window, the execution control unit performs task preemption control according to the cabin priority strategy, selects the two turning cabins with the highest priority within the maximum allowed number of parallel cabins to perform launch preparation, and suspends the delay readjustment logic for the remaining cabins to be launched.

[0154] In this embodiment, to ensure that the anti-riot capture net gun UAV with multi-angle capture capabilities can achieve high-precision, low-latency launch control operations in multi-target tracking and high-concurrency launch scenarios, a launch window precision synchronization mechanism centered around the "execution control unit" is proposed. This mechanism aims to solve the following key problems: first, whether the launch responses of different steering cabins can be accurately aligned within the same interception timing information time window; second, how to compensate for minor errors caused by the trigger delay of the cabin's aerodynamic launch components or the inertia of attitude rotation response; and third, how to coordinate task resource conflicts, improve scheduling efficiency, and prevent misfires when the system supports the parallel preparation of multiple cabins.

[0155] Specifically, after receiving the interception direction command and interception timing information generated by the strategy generation unit, the execution control unit first calls the system time synchronization module to unify the current system's time base and plans the standard trigger time point for each transmission control command accordingly. This time synchronization module has two implementation methods: one is to use a global positioning system synchronization unit based on satellite navigation timing, which locks the system's main control board time to the globally unified Coordinated Universal Time (UTC) through a 1PPS signal (one pulse per second) from the GPS system, and achieves millisecond-level or even better alignment accuracy through continuous time offset correction. The other method is to integrate a temperature-compensated crystal oscillator (TCXO) to form a highly stable local clock. The TCXO chip, through compensation of the temperature drift curve, can control the local crystal oscillator frequency error within ±2ppm. Combined with GPS time calibration at startup, it can maintain high-precision local time operation for a long time. The system can select different synchronization methods according to the actual usage scenario, prioritizing GPS timing in open urban environments and switching to TCXO self-stabilization mode in severely obstructed or indoor scenarios.

[0156] After time synchronization is complete, the execution control unit embeds two key time parameters into the command data packet when generating each launch control command. The first parameter is the center point of the time window defined in the intercept timing information. This value is determined by the strategy generation unit based on target trajectory prediction and system reachability score analysis, representing the optimal launch time when the target object is most likely to enter the current launch range of the hull. The required accuracy is typically in the millisecond range. The second parameter is the allowed time offset tolerance field, which defines a time tolerance range symmetrically extended to the left and right of the window center point. This is used to adjust for small time errors caused by factors such as hull attitude fine-tuning, control response jitter, and aerodynamic trigger delays. It is generally set to ±100 to ±300 milliseconds.

[0157] The execution control unit dynamically calculates the lead time in the launch control command for each cabin based on the two time parameters mentioned above and the structural response characteristics of each steering cabin. This lead time refers to the specific time value that the launch control command should arrive at a particular steering cabin in the vector launch unit relative to the center point of the time window. The calculation method includes two aspects: first, the response time of the cabin's pneumatic valves is statistically derived from historical cabin aerodynamic trigger data, typically ranging from 20 to 80 milliseconds; second, the rotational inertia lag time is estimated based on the current attitude change angle of the steering cabin. This time can be estimated by real-time feedback from sensors on the cabin's current angle change rate (provided by a gyroscope or encoder) and combined with the servo motor's maximum speed and angular acceleration model. For example, if a cabin needs to rotate 45 degrees from its current position, its servo motor's maximum angular velocity is 180 degrees per second, and the rotational inertia lag is expected to be 0.25 seconds, then the lead time must not be less than this value to ensure that the final launch action is completed within the time window.

[0158] This lead time will be aligned with the system's unified time base, and the launch control command issuance time will be set in real time, so that each steering module can complete launch preparation and execute launch in a logically quasi-synchronous manner, meeting the accuracy requirements for time window control.

[0159] Furthermore, in multi-target dynamic tracking missions, there may be multiple vector launch units whose turning cabins correspond to launch control commands, and their planned launch times all fall within the same interception timing information time window. In such cases, to prevent conflicts in aerodynamic resources or system energy consumption load, the execution control unit will perform task preemption control according to a preset cabin priority strategy. The cabin priority setting can refer to multiple dimensions, such as whether the cabin's current position is aligned with the target direction, the cabin's historical hit rate statistics, aerodynamic energy storage status, the current battery load ratio, and the system's task level assessment of the target's importance, ultimately forming a unified cabin task priority table.

[0160] When multiple modules compete for the same launch window resource, the execution control unit will prioritize selecting the two modules with the highest scores for launch preparation, and the remaining modules will be suspended. For suspended modules, the system will automatically execute delay readjustment logic after the launch time window ends, that is, request the strategy generation unit to reassess the target trajectory and system response capability. If the launch conditions are still met, the module will be rescheduled for the next launch scheduling cycle; otherwise, the module will not be activated again to avoid wasting resources or repeated launches.

[0161] Through the coordinated implementation of the aforementioned time synchronization mechanism, dynamic advance compensation strategy, and cabin launch scheduling preemption logic, the execution control unit in this invention can ensure precise triggering within a millisecond-level time window under multi-cabin collaboration or high-frequency scheduling conditions. Compared to existing triggering methods based on local clock polling or non-feedback execution, the solution provided by this invention significantly improves the effectiveness and stability of complex target acquisition, and is particularly suitable for scenarios such as dynamically moving targets, sudden group behavior control, or precise interception of high-value targets.

[0162] The entire control process can be deployed on an embedded processor with a high-precision clock source and concurrent thread scheduling capabilities, such as a processing unit using the ARM Cortex-A series architecture, combined with an RTOS operating system for thread priority scheduling and timestamp recording, and the control latency can be controlled within tens of milliseconds.

[0163] Vector launch unit 105 is used to control the pneumatic launch assembly in the target turning cabin to complete orientation adjustment and net gun launch according to the launch control command. The vector launch unit includes at least four turning cabins distributed on different spatial surfaces of the UAV shell, and each turning cabin is equipped with a rotatable and positionable pneumatic launch assembly.

[0164] The vector launch unit 105 is a key actuator in the anti-riot capture net gun UAV of this invention, which has multi-angle capture capabilities. Its main function is to drive the pneumatic launch assembly inside the target turning cabin to complete the orientation adjustment according to the launch control command issued by the execution control unit 104, and to achieve precise launch of the net gun within a specified time window. This unit directly determines the capture coverage range and actual interception effect of the UAV against high-speed targets and targets with changing attitudes, and is the final point of impact for the tactical output of the entire system.

[0165] The vector launch unit 105 consists of multiple steering pods arranged on multiple spatial surfaces. There are at least four of these pods, which are installed on different spatial surfaces of the UAV shell, preferably in the front, rear, left, and right directions, but can also be extended to the top and bottom directions according to actual mission needs. Each steering pod has an optimized structural layout to ensure that it has an independent rotation channel and launch outlet without interfering with flight aerodynamic stability, thus ensuring that it has controllable attitude adjustment capability at any time.

[0166] Inside each steering chamber is a rotatable pneumatic launch assembly. This assembly typically includes a pneumatic energy storage device, a launch tube, a rotary servo structure, a pneumatic nozzle assembly, a gun release limit device, and a mechanical restraint mechanism. The launch tube houses the foldable net gun projectile. The pneumatic energy storage device stores compressed gas or compresses air in real time via a miniature high-pressure pump when not triggered. The rotary servo structure usually consists of a set of high-precision brushless servo motors and encoders, achieving a positioning accuracy within 0.1 degrees. Feedback control ensures that the launch tube always points towards the interception direction calculated by the strategy generation unit 103. The overall angular range of the launch assembly is at least ±90 degrees of rotation coverage and has rapid angle adjustment capability. The angular velocity is preferably above 60 degrees per second to adapt to directional changes caused by high-speed target movement.

[0167] When the vector launch unit 105 receives the launch control command from the execution control unit 104, the system's internal decoding module first parses the target turning chamber identifier, launch orientation data, and launch trigger timing information in the command. Subsequently, the rotating servo structure inside the target turning chamber begins to perform positioning actions, ensuring that the launch axis of the pneumatic launch assembly is strictly aligned with the designated launch direction. At the same time, the pneumatic energy storage device enters a pre-charge state, ready to release high-pressure gas to propel the net gun projectile at the commanded time.

[0168] Within the time window specified by the interception timing information, the vector launch unit 105 controls the opening of the spray valve assembly, releasing high-pressure gas instantaneously. This propels the net gun projectile at high speed along the calibrated launch direction. The net gun unfolds during flight, forming a coverage area to surround and entangle the target, thus completing the capture action. To ensure launch safety and accuracy, the vector launch unit 105 continuously monitors the positioning status of the steering chamber, the air pressure level inside the chamber, the projectile loading status, and the valve response status throughout the launch process. If any parameter is abnormal, the launch is aborted and feedback is sent to the execution control unit 104.

[0169] All steering pods are connected to the main control circuit of the vector launch unit 105 via a high-speed bus. The main control circuit has concurrent scheduling capabilities, enabling the simultaneous activation of two or more pods when necessary, achieving multi-directional coverage interception. Each pod's aerodynamic launch component has independent operating capabilities, independent of the operating status of other pods, avoiding the "dead zone" problem existing in traditional fixed unidirectional launch methods, and significantly improving the UAV's spatial adaptability and response speed in complex dynamic scenarios.

[0170] The vector launch unit 105's structural design balances lightweight design with strength requirements. It can utilize a carbon fiber composite shell and an aluminum alloy or engineering plastic launch component body to ensure structural stability while meeting flight weight limits. The control system can run on an FPGA+MCU heterogeneous architecture, ensuring parallel real-time execution and redundancy verification capabilities for multi-cabin control tasks.

[0171] In summary, the vector launch unit 105, through its multi-directionally distributed steering cabin, precision rotatable positioning structure, and high-pressure pneumatic launch mechanism, can rapidly adjust its launch direction according to external control commands and accurately complete the launch action, thus constructing an active acquisition capability that covers the entire space, has a fast response, and high precision.

[0172] Furthermore, the vector launch unit includes multiple steering cabins distributed on different spatial surfaces of the UAV shell. The multiple steering cabins are installed according to the principle of symmetry of octahedral spatial structure, so that each steering cabin corresponds to an octahedral vertex direction in the UAV inertial coordinate system, thereby forming a uniform launch coverage layout in three-dimensional spherical space. The layout parameters are initialized and written into the control unit of each steering cabin as spatial configuration parameters.

[0173] Each steering hull has an internal structure equipped with a rotatable pneumatic launch assembly. The pneumatic launch assembly supports a pitch range of at least ±90 degrees and a yaw range of at least ±90 degrees in its mechanical structure, and has a dynamic attitude response speed of not less than 60 degrees per second in the electronic control system. This dynamic attitude response speed is fed back in real time through an internal encoder and used as the attitude closed-loop control input.

[0174] Each steering cabin includes a local attitude conflict detection unit. After receiving the launch control command, the local attitude conflict detection unit is used to determine whether its target launch direction has a mechanical interference risk with other steering cabins based on its current orientation and built-in spatial configuration parameters. If the determination result is that there is a risk, it returns a conflict status code to the external control interface and refuses to execute the launch preparation action.

[0175] Provided that the local attitude conflict detection unit returns to a non-conflict state, the steering pod encodes its attitude response speed, current orientation, and remaining rotation as state information and provides it to the execution control unit to assist it in selecting the optimal launch path and scheduling the target steering pod, thereby realizing the dynamic redundancy management and launch direction coverage balance of the vector launch unit under multi-pod conditions.

[0176] The core design of the vector launch unit involved in this embodiment lies in achieving a multi-directional intelligent capture structure with full-space coverage capabilities through optimized spatial layout of the multi-steering cabin, enhanced independent control capabilities, and a local conflict detection mechanism. This ensures good attitude coordination and execution safety during actual launch control. This vector launch unit provides a feasible structural foundation for riot control and capture net gun UAVs with multi-angle capture capabilities.

[0177] First, the vector launch unit comprises multiple steering pods. These pods are not simply installed at equal intervals in a planar manner, but are designed in a three-dimensional layout strictly according to the "octahedral spatial structure symmetry principle." The so-called octahedral spatial structure refers to selecting the directions of eight vertices of a cube in three-dimensional coordinate space as the installation orientation of the steering pods, so that the entire UAV shell forms eight equidistant direction vectors on the spherical surface. Each direction vector points from the center of the sphere (i.e., the center of the UAV) to a vertex of the cube, thus ensuring that these steering pods form a basically isostatic distribution structure in space.

[0178] In engineering implementation, this octahedral distribution model needs to be accurately simulated using 3D modeling software, such as SolidWorks or UG NX, to define the mounting normal direction, position offset, connection structure, and cable guidance path of each steering cabin on the UAV's outer shell surface. During manufacturing, high-precision injection molding, laser orientation positioning, and snap-fit ​​alignment are used to ensure that each cabin maintains the same attitude orientation as the design model after installation and maintains good spatial orientation distribution stability under different flight attitudes. The layout parameters include the initial orientation vector, coordinate position, mounting reference plane normal vector, and polar coordinate encoding relative to the UAV center for each cabin. These parameters are written into the non-volatile memory inside the steering cabin control unit by the initialization module during system startup and serve as the basic data for cabin control, attitude verification, and spatial conflict judgment.

[0179] Secondly, the core of the steering hull is its internal rotatable and positionable pneumatic launch assembly. This assembly not only launches the capture net gun but also possesses attitude self-adaptation capabilities. Mechanically, each pneumatic launch assembly should include at least two degrees of freedom rotational axes: one for controlling the pitch angle and the other for controlling the yaw angle, thus supporting dual-axis attitude adjustment within a ±90-degree range. To achieve this capability, a servo motor structure or a small gimbal structure can be used as the drive core, coupled with high-strength shock-absorbing limiters and electric stop devices to ensure attitude stability and repeatability during high-speed rotation. This device should also include at least one position encoder, preferably a magnetoelectric encoder or a photoelectric Hall encoder, for real-time acquisition of the current hull attitude angle. The electronic control system samples this angle value at a high frequency (preferably greater than 100Hz) and uses it in the control loop to ensure that the target angle in the rotational attitude closed-loop control system can be tracked in real time.

[0180] Regarding response speed, the hull rotation mechanism must meet a dynamic attitude response speed of no less than 60 degrees per second. This indicator is determined by a combination of the drive motor's rated torque, reduction ratio, and load inertia. For example, if a servo motor with an unloaded response speed of 120 degrees per second is used, maintaining a speed of no less than 60 degrees per second through load adjustment will meet the requirement. The electronic control system needs to establish a PID control loop by combining the rotation angle error with the current speed, or use a model predictive control (MPC) algorithm to adjust the attitude path, thereby ensuring that the hull can complete attitude alignment within 200 to 300 milliseconds after the system issues a launch direction command.

[0181] To achieve attitude safety management, each steering pod also integrates a local attitude conflict detection unit. This detection unit is not used for global path planning, but rather serves as a safety pre-judgment module within the pod. Its core function is to determine, upon receiving launch control commands from the execution control unit, whether mechanical interference with other steering pods will occur during launch, based on its current orientation and built-in spatial configuration parameters. This module is typically implemented as embedded code logic residing in the firmware of the pod control unit. Upon receiving launch control commands, this module retrieves its locally recorded initial orientation vector and target launch orientation vector, compares them with the orientation distribution recorded by other pods, and determines whether mechanical attitude interference exists by calculating criteria such as the included angle, the degree of projection overlap, and whether the rotation trajectory intervals intersect.

[0182] If a mechanical interference risk is detected, the system will immediately return a conflict status code to the external control interface. This status is typically represented by a digital flag or a dedicated protocol data frame, prohibiting the current module from continuing the launch preparation process. This judgment and return action does not rely on active intervention from the execution control unit; it is a local self-protection mechanism of the module, capable of releasing risk information in advance before the control center makes scheduling decisions, thus preventing the two modules from overlapping in the same direction and causing rotational conflicts.

[0183] If the detection result is "no conflict," the capsule will enter the launch preparation process and output its status data in real time after encoding. This status information includes: the current pitch and yaw attitude angles (collected by the encoder), the remaining rotation (calculated from the angle difference between the target direction and the current attitude), and the current attitude response speed (obtained from the real-time speed feedback in the controller). These parameters are packaged into structured data in a unified format and periodically reported to the execution control unit for it to assess whether the capsule is currently feasible for launch.

[0184] This status information will also play a crucial role in multi-module scheduling. For example, when multiple module candidates meet the launch requirements in the same direction, the execution control unit can prioritize the module with the shortest rotation path, fastest speed, and fewest conflicts based on the status information, thereby achieving dynamic load balancing among redundant modules. Furthermore, in long-term missions, by analyzing the rotational load and response stability provided by each module, module maintenance early warning and usage frequency optimization scheduling can also be implemented.

[0185] Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of this application. Therefore, the scope of protection of this application should be determined by the scope defined in the claims of this application.

Claims

1. A riot control and capture net gun drone with multi-angle capture capability, characterized in that, include: The flight control unit is used to control the attitude, position and trajectory of the UAV based on ground commands and autonomous navigation strategies, and outputs the current flight attitude parameters and flight trajectory data. The perception and prediction unit is used to identify the target object and estimate its current three-dimensional position, velocity vector and future motion trajectory, as well as the environmental perception data collected by the multimodal sensor, based on the flight attitude parameters and flight trajectory data, thereby obtaining the target motion trajectory data of the target object. The strategy generation unit is used to construct a multi-angle interception orientation matrix based on the target motion trajectory data and flight attitude parameters, calculate the optimal interception window and corresponding interception direction of the target object, and generate interception direction instructions and interception timing information. The execution control unit is used to coordinate the various steering compartments of the vector launch unit, activate the target steering compartment according to the interception direction command, and issue launch control commands to the vector launch unit within the time window indicated by the interception timing information. The vector launch unit is used to control the pneumatic launch assembly in the target turning cabin to complete the orientation adjustment and net gun launch according to the launch control command. The vector launch unit includes at least four turning cabins distributed on different spatial surfaces of the UAV shell, and each turning cabin is equipped with a rotatable and positionable pneumatic launch assembly.

2. The anti-riot capture net gun drone with multi-angle capture capability according to claim 1, characterized in that, The strategy generation unit is specifically used for: Based on the spatial distribution parameters of each steering cabin in the vector launch unit on the UAV shell, an initial direction vector for each steering cabin is constructed. Based on the flight attitude parameters output by the flight control unit, an attitude transformation operation is performed on the initial direction vector. Quaternion rotation or Euler angle transformation is used to map the initial direction vector to the inertial space coordinate system to form a multi-angle interception orientation matrix. Assign a directional reachability score and a cabin adjustment cost estimate to each directional vector in the multi-angle interception orientation matrix. The directional reachability score is used to reflect the probability that the target object enters the spatial region of that direction in the target motion trajectory data of the target object output by the perception and prediction unit. The cabin adjustment cost estimate is used to characterize the adjustment time or energy consumption required for the steering cabin to rotate from the current attitude to that direction. Based on the target motion trajectory data of the target object, a set of candidate interception directions with an angle less than a preset threshold between the target object and the direction vector is identified. The set of candidate interception directions is then sorted according to the direction reachability score and the estimated cost of cabin adjustment. Finally, the optimal interception window and the corresponding interception direction are determined, and interception direction instructions and interception timing information are generated.

3. The anti-riot capture net gun drone with multi-angle capture capability according to claim 1, characterized in that, The perception and prediction unit is specifically used for: During the target object recognition process, in cases where there is occlusion or image degradation in the environmental perception data collected by the multimodal sensor, the contour features and thermal intensity features associated with the target object are extracted from the infrared thermal imaging image and visible light image frame sequences, respectively, and an inter-frame velocity vector estimation sequence is constructed. The velocity vector estimation sequence and infrared thermal intensity features are input into a confidence weighted model to generate a confidence weight matrix. Based on this confidence weight matrix, multiple candidate trajectory segments formed by the contour features are weighted and fused to output a continuous frame-level position estimation sequence. A prediction error ellipsoid model is constructed based on the continuous frame-level position estimation sequence. The prediction error ellipsoid model is used to quantify the trajectory estimation uncertainty of the current target object and output the ellipsoid principal axis direction and error covariance matrix. The shape of the tracking region and the sampling resolution range of the next frame image are dynamically adjusted according to the ellipsoid principal axis direction and the error covariance matrix, and the correction window is input into the subsequent image processing flow to optimize the target recognition robustness of the multimodal sensor. When the environmental perception data is continuously missing or the confidence weight matrix is ​​lower than the preset confidence threshold, the trajectory compensation mechanism is invoked. The continuous frame-level position estimation sequence is used as input, and time series modeling is performed through a long short-term memory neural network model or a Kalman filter model. The target motion trajectory data of the target object is output as the final trajectory estimation result.

4. The anti-riot capture net gun drone with multi-angle capture capability according to claim 1, characterized in that, The execution control unit is specifically used for: Based on the interception timing information generated by the strategy generation unit, the system time synchronization module is invoked to perform a high-precision launch time alignment operation. The system time synchronization module includes a global positioning system synchronization unit based on satellite navigation timing or a highly stable local clock with an integrated temperature-compensated crystal oscillator, which is used to align the execution time of the launch control command to the standard reference time under a unified system time base. The center point of the time window and the allowable time offset tolerance field contained in the corresponding interception timing information are embedded in each generated launch control command. The execution control unit dynamically adjusts the aerodynamic trigger advance of the corresponding cabin according to the allowable time offset tolerance field to compensate for the cabin response delay and the rotational inertia of the servo motor, so as to achieve quasi-synchronous launch of each steering cabin in the vector launch unit. When the launch control command planned launch time of multiple vector launch units corresponding to the turning cabins falls within the same interception timing information time window, the execution control unit performs task preemption control according to the cabin priority strategy, selects the two turning cabins with the highest priority within the maximum allowed number of parallel cabins to perform launch preparation, and suspends the delay readjustment logic for the remaining cabins to be launched.

5. The anti-riot capture net gun drone with multi-angle capture capability according to claim 1, characterized in that, The vector launch unit includes multiple steering cabins distributed on different spatial surfaces of the UAV shell. The multiple steering cabins are installed according to the principle of symmetry of octahedral spatial structure, so that each steering cabin corresponds to an octahedral vertex direction in the UAV inertial coordinate system, thereby forming a uniform launch coverage layout in three-dimensional spherical space. The layout parameters are initialized and written into the control unit of each steering cabin as spatial configuration parameters. Each steering hull has an internal structure equipped with a rotatable pneumatic launch assembly. The pneumatic launch assembly supports a pitch range of at least ±90 degrees and a yaw range of at least ±90 degrees in its mechanical structure, and has a dynamic attitude response speed of not less than 60 degrees per second in the electronic control system. This dynamic attitude response speed is fed back in real time through an internal encoder and used as the attitude closed-loop control input. Each steering cabin includes a local attitude conflict detection unit. After receiving the launch control command, the local attitude conflict detection unit is used to determine whether its target launch direction has a mechanical interference risk with other steering cabins based on its current orientation and built-in spatial configuration parameters. If the determination result is that there is a risk, it returns a conflict status code to the external control interface and refuses to execute the launch preparation action. Provided that the local attitude conflict detection unit returns to a non-conflict state, the steering pod encodes its attitude response speed, current orientation, and remaining rotation as state information and provides it to the execution control unit to assist it in selecting the optimal launch path and scheduling the target steering pod, thereby realizing the dynamic redundancy management and launch direction coverage balance of the vector launch unit under multi-pod conditions.

6. The anti-riot capture net gun drone with multi-angle capture capability according to claim 1, characterized in that, The flight control unit is specifically used for: Based on the attitude sensor data integrated in the flight control unit, a real-time three-dimensional quaternion attitude calculation module is constructed. By comparing the flight attitude parameters with the initial installation direction of each steering cabin in the UAV inertial coordinate system, the heading offset sensitivity map is calculated in real time to reflect the distribution of launch dead angle changes that may occur due to attitude deviation of each cabin under the current flight attitude. The heading offset sensitive map is used as input and fused with the current predicted flight trajectory of the UAV for analysis. The trajectory control strategy output by the flight control unit is dynamically evaluated to determine whether it will continue to meet the angular reachability conditions of the target interception window in multiple time segments in the future. If it does not meet the conditions, micro-attitude adjustment is performed in advance, and the changes in directional availability of each steering cabin before and after the correction are fed back to the strategy generation unit in the form of a structured matrix. During flight control, the spatial coupling relationship between the flight attitude parameters and the target motion trajectory data of the target object provided by the perception and prediction unit is continuously monitored. A cooperative trajectory strength factor is constructed. This cooperative trajectory strength factor is used to quantify the momentum compatibility between the UAV flight direction and the target object motion direction. The cooperative trajectory strength factor is provided as a dynamic priority value to the execution control unit to optimize the activation order of multiple steering cabins in the vector launch unit. The flight attitude parameters, heading offset sensitivity map, angle reachability correction results, and cooperative trajectory intensity factor are periodically broadcast to the perception prediction unit, the strategy generation unit, and the execution control unit using a unified timestamp binding structure. This ensures that each functional module of the system can perform identification judgment, path generation, and launch scheduling decisions based on consistent and real-time flight control information in highly dynamic flight scenarios.

Images