Multi-source perception autonomous avoidance method and device

Abstract

The application discloses a multi-source perception autonomous avoidance method and device, relates to the technical field of unmanned aerial vehicle autonomous avoidance, and is applied to a multi-source perception autonomous avoidance device; the method comprises the following steps: acquiring the operation identification information of cooperative unmanned aerial vehicles in a detection range in real time through a RemoteID module or an ADS-B module; acquiring the information of non-cooperative targets in the detection range in real time through an obstacle detection sensor; predicting the future trajectories of the cooperative unmanned aerial vehicles based on the operation identification information of the cooperative unmanned aerial vehicles; predicting the future trajectories of the non-cooperative targets based on the information of the non-cooperative targets; constructing a three-dimensional grid space with the current position of a target unmanned aerial vehicle as the center; generating an avoidance decision trajectory through repulsive force field projection and connected region search based on the mapping results of all the future trajectories in the three-dimensional grid space; and sending the avoidance decision trajectory to a flight control unit of the target unmanned aerial vehicle, so that the autonomous avoidance of a micro-light small unmanned aerial vehicle is realized.

Description

Technical Field

[0001] This invention relates to the field of building electrical systems, and more specifically to a multi-source sensing autonomous obstacle avoidance method and device. Background Technology

[0002] In recent years, with the accelerated development of the low-altitude economy, there is an urgent market demand for high-density drone flights in low-altitude application scenarios. However, currently, both domestically and internationally, drones, especially those in the low-altitude field, lack the relevant technical systems to ensure their safe operation under high-density flight conditions. Although the mandatory use of RemoteID is being actively promoted both domestically and internationally, providing basic technical conditions for drone regulation, small and lightweight drones in the low-altitude field still lack mature perception and avoidance (DAA) systems similar to those used in civil aircraft.

[0003] Existing technologies mostly employ neural network models for trajectory prediction and obstacle avoidance path decision-making. However, the lack of interpretability of neural networks introduces significant uncertainties when used for UAV flight safety control, thus affecting the safe operation of the system. Other solutions use artificial potential field methods for path planning, but these methods are highly complex, require high computing power from the onboard computing platform, and suffer from convergence difficulties under special circumstances.

[0004] For micro and lightweight drones, their onboard battery energy is limited and they are extremely sensitive to power consumption. Furthermore, their payload capacity and space are also severely constrained. The high computing power requirements of the existing technologies directly result in large power consumption and size of the computing devices, making them unsuitable for micro and lightweight drones. Therefore, how to achieve autonomous obstacle avoidance for micro and lightweight drones in low-altitude airspace is an urgent technical problem to be solved. Summary of the Invention

[0005] The technical problem to be solved by this invention is how to achieve autonomous obstacle avoidance for micro and light unmanned aerial vehicles in the low-altitude field. The purpose is to provide a multi-source perception autonomous obstacle avoidance method and device, which solves the above-mentioned problem.

[0006] This invention is achieved through the following technical solution:

[0007] In a first aspect, the present invention provides a multi-source sensing autonomous obstacle avoidance method, applied to a multi-source sensing autonomous obstacle avoidance device installed on a target UAV; the method includes:

[0008] The RemoteID module or ADS-B module can be used to obtain operational identification information of cooperative drones within the detection range;

[0009] Information on non-cooperative targets within the detection range is obtained through obstacle detection sensors; these non-cooperative targets include non-cooperative drones and other obstacles.

[0010] Based on the operational identification information of the cooperative drone, predict the future trajectory of the cooperative drone;

[0011] Based on the information about the non-cooperative target, predict the future trajectory of the non-cooperative target;

[0012] A three-dimensional grid space centered on the current position of the target UAV is constructed; based on the mapping result of the future trajectories of the cooperative UAV and the non-cooperative target in the three-dimensional grid space, an avoidance decision trajectory is generated through repulsive field projection and connected region search;

[0013] The avoidance decision trajectory is sent to the flight control unit of the target UAV so that the flight control unit can perform flight actions according to the avoidance decision trajectory.

[0014] Optionally, predicting the future trajectory of the cooperative drone based on its operational identification information includes:

[0015] If the operational identification information includes attitude information and position information, then the target motion equation is determined based on the attitude information, and based on the target motion equation, the attitude information, and the position information, the extended Kalman filter algorithm is used to predict the future trajectory of the cooperative UAV; the target motion equation is one of the following: uniform straight flight motion equation, turning motion equation, climbing motion equation, or descending motion equation;

[0016] If the operational identification information only includes location information, the unscented Kalman filter algorithm is used to predict the future trajectory of the cooperative UAV based on the location information.

[0017] Optionally, constructing a three-dimensional mesh space centered on the current position of the target UAV includes:

[0018] A square detection plane is defined with the centroid of the target UAV as the center and the preset maximum detectable range as the side length;

[0019] The square detection plane is divided into multiple grid units according to a preset safety distance;

[0020] Along the nose direction of the target UAV, the divided square detection plane is translated to the maximum detectable range according to the safe distance to form the three-dimensional grid space.

[0021] Optionally, before determining a square detection plane centered on the centroid of the target UAV and with a preset maximum detectable range as its side length, the method further includes:

[0022] Determine the safe distance based on the type and maximum level flight speed of the target UAV;

[0023] Determine the theoretically maximum detection range achievable by the Remote ID module and the ADS-B module;

[0024] The maximum detectable range is determined based on the maximum detection distance and the safety distance; the formula is as follows:

[0025]

[0026] Where L is the maximum detectable range; The maximum detection range; This indicates a rounding down operation; d represents the safety distance.

[0027] Optionally, the mapping result of the future trajectories of the cooperative UAV and the non-cooperative target in the three-dimensional grid space is used to generate an avoidance decision trajectory through repulsive field projection and connected region search, including:

[0028] At each decision-making moment, the future trajectories of the cooperative drone, the non-cooperative target, and the target drone are mapped onto the three-dimensional grid space;

[0029] The set of grid cells occupied by the future trajectories of the cooperative UAV and the non-cooperative target in the three-dimensional grid space is defined as the first space, and the set of grid cells occupied by the future trajectory of the target UAV in the three-dimensional grid space is defined as the second space.

[0030] Based on the first space and the second space, calculate the projection diagram of the comprehensive repulsive force field;

[0031] In the comprehensive repulsive field projection diagram, search for the connected region with the lowest repulsive force value;

[0032] The center point of the connected region is calculated, and the center point is subjected to three-dimensional spatial trajectory fitting and kinematic smoothing optimization at multiple future decision moments to generate an avoidance decision trajectory.

[0033] Optionally, calculating the comprehensive repulsive field projection map based on the first space and the second space includes:

[0034] Along the nose direction of the target UAV, and at the specified safety interval, a series of corresponding two-dimensional grid planes are synchronously extracted from the first space and the second space; wherein, the nth two-dimensional grid plane includes the nth first subset extracted from the first space and the nth second subset extracted from the second space; n is a positive integer;

[0035] In the nth second subset, determine the central grid cell where the target UAV is currently located;

[0036] A decay coefficient matrix is ​​generated with the central grid cell as the center.

[0037] Calculate the Hadamard product of the nth first subset and the attenuation coefficient matrix to obtain the repulsive field projection of the nth two-dimensional grid plane;

[0038] The repulsive field projections of all two-dimensional grid planes are superimposed and normalized to obtain a comprehensive repulsive field projection map.

[0039] Optionally, the operation identification information further includes identity identification information; the step of searching for the connected region with the lowest repulsion value in the comprehensive repulsion field projection map includes:

[0040] The identification information is transmitted to the ground-based drone command and control platform via the LTE module;

[0041] Receive the verification result returned by the drone command and control platform, which indicates whether the cooperating drone is a peer drone;

[0042] If the verification result indicates that the cooperative drone is not a peer drone, then the connected region with the lowest repulsion value is selected in the integrated repulsion field projection diagram.

[0043] Optionally, after receiving a verification result returned by the drone command and control platform, indicating whether the cooperating drone is a peer drone, the method further includes:

[0044] If the verification result indicates that the cooperative UAV is a peer UAV, then in the integrated repulsive field projection diagram, a candidate connected region located to the right of the target UAV is determined;

[0045] Select the connected region with the lowest repulsion value from the candidate connected regions.

[0046] In a second aspect, the present invention provides a multi-source sensing autonomous obstacle avoidance device, characterized in that it is installed in a target drone; the device includes:

[0047] Both the Remote ID module and the ADS-B module are used to obtain operational identification information of cooperative UAVs within the detection range;

[0048] Obstacle detection sensors are used to acquire information about non-cooperative targets within the detection range;

[0049] The LTE module is used to communicate with the ground-based drone command and control platform;

[0050] A communication and expansion interface for communicating with the flight control unit of the target UAV;

[0051] Embedded computers are used to implement multi-source perception autonomous obstacle avoidance methods.

[0052] Optionally, the communication and expansion interface includes a UART interface and a CAN interface; the non-cooperative target detection sensor is one or more of a visual sensor, an infrared sensor, a lidar, and a millimeter-wave radar.

[0053] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0054] This application provides a multi-source perception autonomous obstacle avoidance method. It unifies the representation of target information detected at long distances by the Remote ID module and ADS-B module, as well as target information detected at close range by the obstacle detection sensor, in a three-dimensional grid space through trajectory prediction and grid mapping. This achieves unified fusion and efficient processing of multi-source heterogeneous information. This method abandons traditional complex multi-source data fusion algorithms, abstracting different types of targets (cooperative UAVs, non-cooperative UAVs, and other obstacles) as units in the grid space, greatly simplifying the data processing flow, reducing the computational dependence on fusion algorithms, and meeting the extremely stringent payload and power consumption requirements of micro-lightweight UAVs. Furthermore, by generating avoidance decision paths through repulsive field projection and connected region search, it enables autonomous obstacle avoidance for resource-constrained micro-lightweight UAVs. Attached Figure Description

[0055] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0056] Figure 1 This is a schematic diagram of the structure of a multi-source sensing autonomous obstacle avoidance device provided in an embodiment of this application;

[0057] Figure 2 This is a schematic diagram of an application scenario provided by an embodiment of this application;

[0058] Figure 3 A flowchart illustrating a multi-source sensing autonomous obstacle avoidance method provided in an embodiment of this application;

[0059] Figure 4 A schematic diagram of a three-dimensional mesh trajectory projection provided in an embodiment of this application;

[0060] Figure 5 This is a schematic diagram of the integrated repulsive field projection provided in the embodiments of this application. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0062] To achieve autonomous obstacle avoidance for small, lightweight unmanned aerial vehicles (UAVs) in low-altitude environments, this application provides a multi-source sensing autonomous obstacle avoidance method, applied to a multi-source sensing autonomous obstacle avoidance device, which can be installed on the target UAV. Please refer to... Figure 1 This is a schematic diagram of the structure of a multi-source sensing autonomous obstacle avoidance device provided in an embodiment of this application. The multi-source sensing autonomous obstacle avoidance device includes an obstacle detection sensor, a RemoteID module, an ADS-B module, an LTE module, an RTK / GNSS positioning module, an embedded computer, and communication and expansion interfaces.

[0063] Both the Remote ID module and the ADS-B module are used to obtain operational identification information of cooperative UAVs within the detection range.

[0064] Obstacle detection sensors are used to acquire information about non-cooperative targets within the detection range;

[0065] The LTE module is used to communicate with the ground-based drone command and control platform;

[0066] The communication and expansion interface is used to communicate with the flight control unit of the target UAV;

[0067] The RTK / GNSS positioning module is used to obtain its own location information;

[0068] The embedded computer includes a method for implementing the multi-source sensing autonomous obstacle avoidance method provided in the embodiments of this application.

[0069] In one possible embodiment, the communication and expansion interfaces include a UART interface, a CAN interface, and an ETH interface. The obstacle detection sensor is one or more of a visual sensor, an infrared sensor, a LiDAR, and a millimeter-wave radar. The RemoteID module supports 2.4G or 5.8G Wi-Fi and Bluetooth broadcast transmission and reception. The embedded computer includes an embedded CPU processor, memory, and storage.

[0070] Please refer to Figure 2 This is a schematic diagram illustrating an application scenario provided by an embodiment of this application. The application scenario includes a target drone, a base station, and a drone command and control platform. The target drone is equipped with a multi-source perception autonomous obstacle avoidance device. This device connects to the target drone's flight control unit via a UART interface, establishes a connection with the base station via an LTE module, and then communicates with the drone command and control platform via the Internet.

[0071] The above has introduced multi-source perception autonomous obstacle avoidance devices and application scenarios. Below, based on… Figure 1 The multi-source sensing autonomous obstacle avoidance device shown and Figure 2 The application scenarios shown are for Figure 3 The multi-source sensing autonomous obstacle avoidance method shown in the figure will be introduced.

[0072] S301. Obtain operational identification information of cooperative UAVs within the detection range through the RemoteID module or ADS-B module.

[0073] In practical implementation, the multi-source perception autonomous obstacle avoidance device integrates a Remote ID module and an ADS-B module, both mandated by national standards, for collaborative perception. The Remote ID module can listen to Wi-Fi or Bluetooth broadcast signals in the 2.4GHz or 5.8GHz bands, analyze the broadcast payloads that conform to preset protocols, and thus obtain the operational identification information of cooperative drones within the detection range. Simultaneously, the ADS-B module can receive the ADS-B Out signals broadcast by cooperative drones within the detection range and extract their operational identification information from these signals.

[0074] The cooperating drones are other drones equipped with and having their Remote ID or ADS-B modules activated. Operational identification information includes at least location information, specifically latitude, longitude, and altitude. More preferably, the operational identification information further includes attitude information, identification information, and size information, which are described below.

[0075] (1) Location information: used to indicate the precise location of the UAV in the three-dimensional airspace, specifically including longitude, latitude and altitude three-dimensional coordinate data.

[0076] (2) Attitude information: used to describe the flight attitude of the UAV in the air, specifically including heading angle, pitch angle and acceleration.

[0077] (3) Identification information: used to uniquely identify the drone, such as SN serial number or aircraft registration number.

[0078] (4) Size information: used to describe the physical outline dimensions of the UAV, specifically including length, width and height.

[0079] S302. Obtain information about non-cooperative targets within the detection range through obstacle detection sensors.

[0080] In practical implementation, the multi-source perception autonomous obstacle avoidance device integrates one or more obstacle detection sensors, which can acquire in real time the historical location and size information of non-cooperative targets within the detection range. Non-cooperative targets include non-cooperative drones and other obstacles. Non-cooperative drones are other drones that are not equipped with or have not activated the Remote ID module and ADS-B module. Other obstacles include dynamic obstacles (such as birds, balloons, etc.) and static obstacles (such as buildings, power lines, trees, etc.).

[0081] S303. Based on the operational identification information of the cooperative drones, predict the future trajectory of the cooperative drones.

[0082] In the specific implementation process, there may be differences in the data completeness between the operation identification information obtained from the Remote ID module and the operation identification information obtained from the ADS-B module. Therefore, different trajectory prediction methods are adopted, which will be explained in detail below.

[0083] In the first scenario, the operational identification information includes attitude information and position information.

[0084] The target's motion equation is determined based on the attitude information, and the future trajectory of the cooperative UAV is predicted based on the target's motion equation, attitude information, and position information using an extended Kalman filter (EKF).

[0085] In this case, the received attitude information (heading angle, pitch angle, and acceleration) is first analyzed to determine the current flight mode of the cooperative UAV, thereby adaptively selecting a matching target motion equation, such as a uniform straight flight motion equation, a turning motion equation, a climbing motion equation, or a descending motion equation.

[0086] When the heading angle of the cooperative UAV remains stable and its acceleration is close to zero, it is determined that the cooperative UAV is in uniform straight flight mode, and the uniform straight flight motion equation is selected. When the heading angle of the cooperative UAV changes continuously and regularly, and a corresponding centripetal acceleration is detected, it is determined that the cooperative UAV is in turning mode, and the turning motion equation is selected. When the pitch angle of the cooperative UAV is continuously positive and accompanied by continuous positive acceleration in the altitude channel, it is determined that the cooperative UAV is in climb mode, and the climb motion equation is selected. When the pitch angle of the cooperative UAV is continuously negative and accompanied by continuous negative acceleration in the altitude channel, it is determined that the cooperative UAV is in descent mode, and the descent motion equation is selected.

[0087] Furthermore, the target motion equation is used as the state transition equation of the EKF filter, and combined with the real-time acquired attitude and position information, the future trajectory of the cooperative UAV is predicted, thereby outputting the operating trajectory of the cooperative UAV within a certain period of time (e.g., 50 seconds).

[0088] In the second scenario, the operation identification information only includes location information.

[0089] Based on location information, the Unscented Kalman Filter (UKF) algorithm is used to predict the future trajectory of the cooperative drone.

[0090] In this case, since the attitude information of the cooperative drone cannot be obtained directly, the unscented Kalman filter algorithm is used to predict the future trajectory of the cooperative drone directly based on the position information, namely a series of historical and real-time position observation data, so as to output the running trajectory of the cooperative drone in the future (e.g., 50 seconds).

[0091] In this embodiment, when information is complete, specific flight modes are identified by parsing attitude information, and the corresponding target motion equation is selected to drive the extended Kalman filter. This ensures that the prediction model closely matches the target's actual physical motion, significantly improving prediction accuracy during critical maneuvers such as turning and climbing. When information is limited, an unscented Kalman filter is used. It does not require precise physical equations and robustly captures and extrapolates the target's motion trend directly from historical position data, effectively avoiding a precipitous drop in prediction performance due to missing information. By designing differentiated trajectory prediction strategies, an optimal balance between prediction accuracy and resource efficiency is achieved, providing highly reliable trajectory prediction assurance for the safe and autonomous avoidance of unmanned aerial vehicles (UAVs).

[0092] S304. Based on information about non-cooperative targets, predict the future trajectory of non-cooperative targets.

[0093] In the specific implementation process, for non-cooperative targets (including non-cooperative drones and other obstacles) detected by obstacle detection sensors, since such targets do not provide their own attitude information, the unscented Kalman filter algorithm is uniformly adopted based on historical position information to predict the future trajectory of non-cooperative targets, and finally outputs the running trajectory of non-cooperative targets in the future (e.g., 50 seconds).

[0094] S305. Construct a three-dimensional mesh space centered on the current position of the target UAV.

[0095] In one possible embodiment, a square detection plane is defined with the centroid of the target UAV as the center and a preset maximum detectable range as the side length; the square detection plane is divided into multiple grid cells according to a preset safety distance; along the nose direction of the target UAV, the divided square detection plane is translated to the maximum detectable range according to the safety distance to form a three-dimensional grid space.

[0096] In the specific implementation process, firstly, an L×L square detection plane S is defined directly in front of the target UAV's nose, where L is the preset maximum detectable range. The detection plane S is perpendicular to the target UAV, and the target UAV is located at the center of the detection plane S. Secondly, the detection plane S is divided into m×m square grid cells of equal area according to a safety interval d, where L=m×d. Then, using the above-mentioned completed gridded detection plane S as the starting section, it is gradually translated along the nose direction at a safety interval d until the maximum detectable range L is reached. Each translation replicates an identical grid plane, ultimately resulting in an L×L×L cubic three-dimensional grid space.

[0097] In this embodiment, by constructing a three-dimensional mesh space centered on the centroid of the target UAV and with the nose direction as the main axis, there is no need to perform indiscriminate and costly modeling of the global environment. Limited computing resources are dynamically and centrally applied to the assessment of key airspace ahead of the target UAV's future flight path. This effectively avoids the waste of computing resources in irrelevant directions (such as the sides or rear), ensuring that when the system encounters potential conflicts, it can reserve the highest perception accuracy and the fastest decision response speed for the most urgent and dangerous areas.

[0098] In one possible embodiment, a safe distance is determined based on the type and maximum level flight speed of the target UAV; the theoretically achievable maximum detection range of the Remote ID module and ADS-B module is determined; and the maximum detectable range is determined based on the maximum detection range and the safe distance; the formula is as follows:

[0099]

[0100] Where L is the maximum detectable range; Maximum detection range; This indicates a round-down operation; d is the safety margin.

[0101] In the specific implementation process, before environmental modeling and obstacle avoidance decisions, the safety distance and maximum detectable range can be obtained from pre-configured parameters. The safety distance, as the basic unit size in the three-dimensional mesh space, is determined comprehensively based on the type of the target UAV and its maximum level flight speed. Target UAV types include multi-rotor, vertical takeoff and landing (VTOL) aircraft, and fixed-wing aircraft. The principle for determining it is: the worse the UAV's maneuverability and the faster its maximum level flight speed, the larger the safety buffer space required to cope with emergencies; therefore, the larger the safety distance value.

[0102] The maximum detectable range defines the spatial boundary for the system's threat perception and decision-making. Its base value is the theoretically maximum detection distance achievable by the RemoteID and ADS-B modules. To achieve standardized spatial discretization, the maximum detectable range L is rounded to the largest integer multiple of the safety distance d, i.e., L = m × d. , This represents the theoretically maximum detection range achievable by the Remote ID module or ADS-B module. This indicates a round-down operation.

[0103] In this application, different parameters are determined according to the performance configuration of different models, avoiding the universality problem caused by fixed parameters. At the same time, by quantizing the maximum detectable range into integer multiples of the safety spacing, a perfect mathematical foundation is laid for the subsequent construction of a three-dimensional mesh space composed of m×m×m standard mesh units.

[0104] S306. Based on the mapping results of the future trajectories of cooperative drones and non-cooperative targets in a three-dimensional grid space, avoidance decision trajectories are generated through repulsive field projection and connected region search.

[0105] In one possible embodiment, S306 is specifically implemented as follows:

[0106] At each decision-making moment, the future trajectories of the cooperative drone, the non-cooperative target, and the target drone are mapped onto a 3D grid space. The set of grid cells occupied by the future trajectories of the cooperative drone and the non-cooperative target in the 3D grid space is defined as the first space, and the set of grid cells occupied by the future trajectory of the target drone in the 3D grid space is defined as the second space. Based on the first and second spaces, a comprehensive repulsive field projection map is calculated. In the comprehensive repulsive field projection map, the connected region with the lowest repulsive force value is searched. The center point of the connected region is calculated, and the center point of the connected region is subjected to 3D spatial trajectory fitting and kinematic smoothing optimization for multiple future decision-making moments to generate an avoidance decision trajectory.

[0107] In the specific implementation process, firstly, at each decision time t of the target UAV's operation, the future trajectories of the cooperative UAV, the non-cooperative target, and the target UAV itself are mapped to a three-dimensional grid space based on their respective size information, and the one or more grid cells occupied by each are determined.

[0108] Secondly, treating a single grid cell as a matrix element, the grid cells occupied by the future trajectories of all cooperative UAVs and non-cooperative targets in the three-dimensional grid space are defined as the first space V1. This first space V1 mathematically represents the distribution of all external threats in the environment. The set of grid cells occupied by the future trajectory of the target UAV in the three-dimensional grid space is defined as the second space V2. This second space V2 mathematically represents the planned occupancy of the target UAV at the present and future moments.

[0109] Then, based on the first and second spaces, a comprehensive repulsive field projection map is calculated, and the connected regions with the lowest repulsive values ​​are searched. The comprehensive repulsive field projection map is used to comprehensively indicate the global dynamic risk distribution in the airspace ahead of the target UAV. The repulsive value of each grid cell in the comprehensive repulsive field projection map quantitatively indicates the comprehensive collision risk level faced by the planned path passing through that location; the lower the repulsive value, the safer the direction. A connected region refers to a set of spatially connected areas in the comprehensive repulsive field projection map, composed of grid cells whose repulsive values ​​are all below a preset safety threshold.

[0110] Next, the geometric center point of the connected region is calculated, and the center points extracted from multiple consecutive decision moments in the future are combined to form a discrete point sequence in three-dimensional space. Then, a cubic spline curve or polynomial fitting method is used to fit the discrete point sequence to generate an initial geometric path. Finally, an optimization problem is constructed with the control point coordinates and / or time parameters of the initial geometric path as optimization variables, motion smoothness and / or trajectory execution time as optimization objectives, and the dynamic limits of the target UAV as constraints. By solving this optimization problem, a time-parameterized trajectory, i.e., the avoidance decision trajectory, is obtained. The constraints include velocity constraints, acceleration constraints, curvature constraints, and boundary state constraints. The avoidance decision trajectory defines the position, velocity, and acceleration of the target UAV on the time axis.

[0111] Considering that drones, as physical entities, are constrained by their dynamic limitations (such as maximum acceleration, maximum roll angle, and maximum pitch rate), they cannot change direction instantaneously. Unsmoothed paths force drones to perform aggressive maneuvers such as sharp turns, rapid acceleration, or sudden deceleration, which not only consumes excessive energy and causes onboard load fluctuations but may also lead to dangerous stall or loss of control. Therefore, in this embodiment, a continuous three-dimensional spatial curve with smooth curvature is generated through trajectory fitting and kinematic smoothing—the avoidance decision trajectory. The smoothing optimization process uses the drone's kinematic limits as constraints, ensuring that all flight state variables (velocity, acceleration, etc.) remain within the safety envelope. This provides the drone with a stable and controllable avoidance maneuver, significantly improving flight quality and safety.

[0112] In one possible embodiment, the comprehensive repulsive field projection map is calculated based on the first space and the second space, including:

[0113] Along the nose direction of the target UAV, a series of corresponding two-dimensional grid planes are synchronously intercepted from the first space and the second space at safe intervals; wherein, the nth two-dimensional grid plane contains the nth first subset intercepted from the first space and the nth second subset intercepted from the second space; n is a positive integer; in the nth second subset, the central grid cell where the target UAV is currently located is determined; with the central grid cell as the center, an attenuation coefficient matrix is ​​generated; the Hadamard product of the nth first subset and the attenuation coefficient matrix is ​​calculated to obtain the repulsive field projection of the nth two-dimensional grid plane; the repulsive field projections of all two-dimensional grid planes are superimposed and normalized to obtain a comprehensive repulsive field projection map.

[0114] In the specific implementation process, firstly, along the nose direction of the target UAV, a two-dimensional grid plane SV1n is intercepted from the first space V1 at a safe distance d, and simultaneously a two-dimensional grid plane SV2n is intercepted from the second space V2 (n is a multiple of d). Secondly, in SV2n, the central grid cell where the target UAV is currently located is determined, and an attenuation coefficient matrix Rn is generated with the central grid cell as the center. The coefficient of the grid cell where the center is located in this matrix is ​​set to the maximum value (1). The coefficients of other grid cells in the matrix diffuse and attenuate outwards according to the Manhattan distance or Euclidean distance between them and the center, according to the preset attenuation parameters, and their minimum coefficient is not lower than the preset threshold (e.g., 0.4). Then, the Hadamard product of SV1n and Rn is calculated to obtain the repulsive field projection of the nth two-dimensional grid plane, such as Figure 4 As shown.

[0115] Extrapolating to the farthest point of the three-dimensional mesh space using the above method yields m repulsive field projections. These m projections are then summed and normalized to obtain a comprehensive repulsive field projection map of all cooperative and non-cooperative targets relative to the target drone, as shown below. Figure 5 As shown.

[0116] In this embodiment, by sequentially extracting two-dimensional planes along the flight direction and calculating the repulsive field projections separately, the complex three-dimensional obstacle avoidance problem is successfully decomposed into a series of more manageable two-dimensional problems. Finally, through superposition and normalization, the threat information in the three-dimensional airspace ahead is efficiently aggregated into a two-dimensional comprehensive repulsive field projection map, greatly reducing computational complexity while ensuring the integrity of decision-making information.

[0117] In one possible embodiment, the step of searching for the connected region with the lowest repulsion value in the comprehensive repulsion field projection map includes:

[0118] The identification information of the cooperating drone is sent to the ground drone command and control platform via the LTE module; the verification result returned by the drone command and control platform is received, indicating whether the cooperating drone is a peer drone; if the verification result indicates that the cooperating drone is not a peer drone, the connected region with the lowest repulsion value is selected in the integrated repulsion field projection map.

[0119] In the specific implementation process, the multi-source autonomous avoidance device establishes a stable data link with the ground-based UAV command and control platform through the LTE module, encapsulates the identification information of the cooperating UAV into a data packet of a specific protocol format, transmits it to the UAV command and control platform through the base station, and waits for and receives the verification result returned by the UAV command and control platform.

[0120] The drone command and control platform can pre-store a data table containing the identification information of all drones equipped with multi-source autonomous obstacle avoidance (MAO) devices. It matches the received identification information against this data table. If a match is successful, a verification result indicating that the cooperating drone is a peer drone is generated; if the match fails, a verification result indicating that the cooperating drone is not a peer drone is generated. A peer drone is defined as a drone also equipped with MAO devices. This verification result is then forwarded via a base station to the LTE module of the target drone's MAO device.

[0121] Furthermore, if the cooperating drone is not a peer drone or the multi-source autonomous avoidance device does not receive the verification result returned by the drone command and control platform within the preset time, then in the integrated repulsion field projection map, the sum or average value of the repulsion values ​​of all grid cells in each connected region is calculated, and the connected region with the lowest global repulsion value is selected.

[0122] In this embodiment, when it is confirmed that the other party is not a peer drone, the device unconditionally adopts the most conservative global avoidance strategy, that is, it selects the channel with the lowest global repulsion value for avoidance, granting the highest security privileges to the other party, and strictly adhering to airspace rules. When communication with the ground platform is interrupted or the response times out, it is assumed that the other party is not a peer drone and adopts the most conservative global avoidance strategy, effectively preventing the device from falling into a decision deadlock state of "wait-stagnation" due to unreliable external communication links, and ensuring that the device has the ability to autonomously degrade operation and maintain basic safety under any abnormal circumstances.

[0123] In one possible embodiment, after receiving a verification result returned by the drone command and control platform indicating whether the cooperating drone is a peer drone, the method further includes:

[0124] If the verification results indicate that the cooperating drone is a peer drone, then in the integrated repulsive field projection map, determine the candidate connected region located to the right of the target drone; select the connected region with the lowest repulsive force value from the candidate connected regions.

[0125] In the specific implementation process, a body coordinate system is established with the target UAV's center of mass as the origin and the nose direction as the X-axis. From all connected regions of the comprehensive repulsive field projection map, candidate connected regions located in the positive direction of the Y-axis of the body coordinate system (i.e., the right side) are selected. Within the set of candidate connected regions obtained after selection, the sum or average value of the repulsive force of all grid cells in each connected region is calculated, and the connected region with the lowest global repulsive force value in the set is selected.

[0126] In this embodiment, by first screening the right-hand region, the avoidance behavior strictly adheres to the right-hand rule, enabling both drones to generate deterministic and predictable behavioral expectations, fundamentally avoiding collision risks caused by inconsistent decision-making logic. Under the premise of satisfying the rules, by secondarily selecting the connected region with the lowest repulsion value within the right-hand region, it is ensured that the target drone can still choose the path with the lowest risk and highest safety while complying with the rule framework.

[0127] S307. The avoidance decision trajectory is sent to the flight control unit of the target UAV so that the flight control unit can execute flight actions according to the avoidance decision trajectory.

[0128] In practical implementation, after generating the avoidance decision trajectory, the multi-source sensing autonomous avoidance device transmits the avoidance motion decision trajectory to the target UAV's flight control unit every second via a preset communication protocol through a UART or CAN interface. Upon receiving the avoidance decision trajectory, the UAV's flight control unit calculates it into specific low-level control commands (such as attitude angles and throttle), driving the UAV's actuators (such as motors and control surfaces) to precisely execute the actual avoidance flight maneuvers.

[0129] In one possible embodiment, the multi-source perception autonomous obstacle avoidance device can also upload the obstacle avoidance decision trajectory to a ground-based drone command and control platform via an LTE module.

[0130] In one possible embodiment, the multi-source perception autonomous obstacle avoidance device also includes a system monitoring module, which monitors the overall hardware and software operation status of the multi-source perception autonomous obstacle avoidance device in real time and records necessary logs including path decision, obstacle information and obstacle avoidance decision process.

[0131] In summary, this application provides a multi-source perception autonomous obstacle avoidance method. Through a Remote ID module, an ADS-B module, and multiple obstacle detection sensors, it simultaneously achieves comprehensive perception of various threats, including static obstacles, moving obstacles, cooperative drones, and non-cooperative drones. By employing 3D mesh spatial modeling and repulsive field projection technology, the complex 3D path planning problem is transformed into an efficient 2D image search problem. Combined with an optimized trajectory prediction algorithm, the system can reliably complete obstacle avoidance decision-making and planning in real time under limited embedded computing power, effectively avoiding both moving and static obstacles. Furthermore, this method does not rely on the other aircraft's obstacle avoidance system and can safely coexist with various aircraft, whether equipped with or without such equipment, in the same airspace. It provides an efficient and feasible technical solution for current low-altitude safe operations and offers strong technical support for the commercially desired high-density drone operations.

[0132] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0133] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0134] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-source sensing autonomous obstacle avoidance method, characterized in that, The method is applied to a multi-source sensing autonomous obstacle avoidance device installed on a target drone; the method includes: The RemoteID module or ADS-B module can be used to obtain operational identification information of cooperative drones within the detection range; Information on non-cooperative targets within the detection range is obtained through obstacle detection sensors; these non-cooperative targets include non-cooperative drones and other obstacles. Based on the operational identification information of the cooperative drone, predict the future trajectory of the cooperative drone; Based on the information about the non-cooperative target, predict the future trajectory of the non-cooperative target; A three-dimensional grid space centered on the current position of the target UAV is constructed; based on the mapping result of the future trajectories of the cooperative UAV and the non-cooperative target in the three-dimensional grid space, an avoidance decision trajectory is generated through repulsive field projection and connected region search; The avoidance decision trajectory is sent to the flight control unit of the target UAV, so that the flight control unit can perform flight actions according to the avoidance decision trajectory; The mapping result of the future trajectories of the cooperative UAV and the non-cooperative target in the three-dimensional grid space is used to generate an avoidance decision trajectory through repulsive field projection and connected region search, including: At each decision-making moment, the future trajectories of the cooperative drone, the non-cooperative target, and the target drone are mapped onto the three-dimensional grid space; The set of grid cells occupied by the future trajectories of the cooperative UAV and the non-cooperative target in the three-dimensional grid space is defined as the first space, and the set of grid cells occupied by the future trajectory of the target UAV in the three-dimensional grid space is defined as the second space. Based on the first space and the second space, calculate the projection diagram of the comprehensive repulsive force field; In the comprehensive repulsive field projection diagram, search for the connected region with the lowest repulsive force value; The center point of the connected region is calculated, and the center point is subjected to three-dimensional spatial trajectory fitting and kinematic smoothing optimization at multiple future decision moments to generate an avoidance decision trajectory.

2. The multi-source sensing autonomous obstacle avoidance method according to claim 1, characterized in that, The prediction of the future trajectory of the cooperative drone based on its operational identification information includes: If the operational identification information includes attitude information and position information, then the target motion equation is determined based on the attitude information, and based on the target motion equation, the attitude information, and the position information, the extended Kalman filter algorithm is used to predict the future trajectory of the cooperative UAV; the target motion equation is one of the following: uniform straight flight motion equation, turning motion equation, climbing motion equation, or descending motion equation; If the operational identification information only includes location information, the unscented Kalman filter algorithm is used to predict the future trajectory of the cooperative UAV based on the location information.

3. The multi-source sensing autonomous obstacle avoidance method according to claim 1, characterized in that, The construction of the three-dimensional mesh space centered on the current position of the target UAV includes: A square detection plane is defined with the centroid of the target UAV as the center and the preset maximum detectable range as the side length; The square detection plane is divided into multiple grid units according to a preset safety distance; Along the nose direction of the target UAV, the divided square detection plane is translated to the maximum detectable range according to the safe distance to form the three-dimensional grid space.

4. The multi-source sensing autonomous obstacle avoidance method according to claim 3, characterized in that, Before defining a square detection plane centered on the centroid of the target UAV and with a preset maximum detectable range as its side length, the method further includes: Determine the safe distance based on the type and maximum level flight speed of the target UAV; Determine the theoretically maximum detection range achievable by the Remote ID module and the ADS-B module; The maximum detectable range is determined based on the maximum detection distance and the safety distance; the formula is as follows: ； Where L is the maximum detectable range; The maximum detection range; This indicates a rounding down operation; d represents the safety distance.

5. The multi-source sensing autonomous obstacle avoidance method according to claim 3, characterized in that, The calculation of the comprehensive repulsive field projection map based on the first space and the second space includes: Along the nose direction of the target UAV, and at the specified safety interval, a series of corresponding two-dimensional grid planes are synchronously extracted from the first space and the second space; wherein, the nth two-dimensional grid plane includes the nth first subset extracted from the first space and the nth second subset extracted from the second space; n is a positive integer; In the nth second subset, determine the central grid cell where the target UAV is currently located; A decay coefficient matrix is ​​generated with the central grid cell as the center. Calculate the Hadamard product of the nth first subset and the attenuation coefficient matrix to obtain the repulsive field projection of the nth two-dimensional grid plane; The repulsive field projections of all two-dimensional grid planes are superimposed and normalized to obtain a comprehensive repulsive field projection map.

6. The multi-source sensing autonomous obstacle avoidance method according to claim 1, characterized in that, The operational identification information also includes identity identification information; the step of searching for the connected region with the lowest repulsion value in the comprehensive repulsion field projection map includes: The identification information is transmitted to the ground-based drone command and control platform via the LTE module; Receive the verification result returned by the drone command and control platform, which indicates whether the cooperating drone is a peer drone; If the verification result indicates that the cooperative drone is not a peer drone, then the connected region with the lowest repulsion value is selected in the integrated repulsion field projection diagram.

7. The multi-source sensing autonomous obstacle avoidance method according to claim 6, characterized in that, After receiving a verification result returned by the drone command and control platform, indicating whether the cooperating drone is a peer drone, the method further includes: If the verification result indicates that the cooperative UAV is a peer UAV, then in the integrated repulsive field projection diagram, a candidate connected region located to the right of the target UAV is determined; Select the connected region with the lowest repulsion value from the candidate connected regions.

8. A multi-source sensing autonomous obstacle avoidance device, characterized in that, Installed in the target drone; the device includes: Both the Remote ID module and the ADS-B module are used to obtain operational identification information of cooperative UAVs within the detection range; Obstacle detection sensors are used to acquire information about non-cooperative targets within the detection range; The LTE module is used to communicate with the ground-based drone command and control platform; A communication and expansion interface for communicating with the flight control unit of the target UAV; An embedded computer for implementing the multi-source sensing autonomous obstacle avoidance method as described in any one of claims 1 to 7.

9. The multi-source sensing autonomous obstacle avoidance device according to claim 8, characterized in that, The communication and expansion interfaces include a UART interface and a CAN interface; the obstacle detection sensor is one or more of a visual sensor, an infrared sensor, a lidar, and a millimeter-wave radar.

Images