Unmanned aerial vehicle return risk early warning method and system based on environmental perception

Abstract

The invention provides an unmanned aerial vehicle homeward voyage risk early warning method and system based on environmental perception, and is applied to the technical field of unmanned aerial vehicle autonomous control, and the method comprises the steps: collecting environmental data and airspace information, fusing the environmental data and airspace information to a three-dimensional grid map, calculating a risk probability value, marking a high-risk region, planning a safe homeward voyage trajectory, and generating a flight instruction. According to the scheme, the problem of poor adaptability to the dynamic environment in the prior art can be solved, and the return flight safety of the unmanned aerial vehicle in the complex environment is improved.

Description

Technical Field

[0001] This invention relates to the field of autonomous control technology for unmanned aerial vehicles (UAVs), and in particular to a method and system for early warning of UAV return-to-home risks based on environmental perception. Background Technology

[0002] With the rapid development of drone technology, drones are being used more and more widely in many fields, such as logistics and delivery, agricultural plant protection, and power line inspection. When performing these tasks, especially in complex environments such as cities and mountains, the drone's ability to return safely and autonomously is crucial for ensuring mission success, protecting equipment and assets, and even public safety.

[0003] The existing drone return-to-home technology has the following main shortcomings:

[0004] Static Environmental Awareness: Most drones rely on pre-loaded static electronic maps and GPS for return-to-home path planning. This approach cannot handle dynamically emerging threats during the return journey, such as suddenly established temporary construction zones, controlled airspace for large events, moving tall work vehicles like cranes, or even severe weather in localized areas like strong wind shear. This can cause drones to inadvertently enter unknown and dangerous areas during their return journey, leading to collisions or loss of control risks.

[0005] Single Risk Source Consideration: Traditional obstacle avoidance technologies primarily focus on avoiding static physical obstacles, neglecting non-physical risk sources. For example, in urban "canyons" filled with high-rise buildings, GPS signals can severely attenuate or even be interrupted, leading to a sharp decline in drone positioning accuracy. Simultaneously, complex electromagnetic environments can interfere with the communication link between the drone and the ground station. Existing return-to-home strategies rarely incorporate these "soft risks" into their path planning. If the drone enters a signal blind spot, it may fail due to positioning errors or loss of contact, resulting in mission failure.

[0006] Rigid decision-making mechanisms: Existing return-to-home strategies are usually preset and lack the ability to adaptively adjust based on the drone's own status and environmental changes. For example, regardless of whether the drone has sufficient power or is about to run out, it may adopt the same obstacle avoidance and detour strategy, which may prevent it from returning before running out of power when the battery is low. In addition, responses to sudden dangers usually rely on global path replanning, which is computationally intensive, slow in response, and difficult to deal with rapidly changing close-range threats during high-speed flight.

[0007] Algorithm and platform disconnect: Many advanced path planning algorithms perform well in simulation environments, but they fail to fully consider the reality of extremely limited onboard computing platform resources for UAVs, such as computing power, power consumption, and memory. Directly porting complex algorithms to resource-constrained embedded systems often results in excessive computational latency, failing to meet real-time obstacle avoidance requirements, and may even lead to system crashes due to exhaustion of computing resources.

[0008] In summary, existing technologies have significant shortcomings in addressing dynamic and multi-dimensional return-to-home risks. There is an urgent need for a new type of UAV return-to-home technology that can perceive and comprehensively assess multiple risks in real time, has adaptive decision-making capabilities, and is deeply integrated with the characteristics of the airborne platform, in order to improve its survivability and mission reliability in complex environments.

[0009] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0010] In view of this, the present invention provides a method and system for early warning of drone return-to-home risks based on environmental perception, which aims to solve the problem that drones cannot effectively cope with dynamic environmental threats and cannot guarantee flight safety when returning to home in complex environments in the prior art. By collecting environmental data in real time and generating dynamic risk maps, the method improves the safety of drones' autonomous return-to-home in complex environments.

[0011] This invention provides a method for early warning of drone return-to-home risk based on environmental perception, comprising:

[0012] The drone collects raw environmental data in real time, which characterizes the three-dimensional physical environment around it, using environmental sensors mounted on it, and receives dynamic airspace intelligence data through the communication module.

[0013] In the airborne processor, the raw environmental data and dynamic airspace intelligence data are processed in real time and integrated into a three-dimensional grid map. The risk probability value used to characterize the flight risk level is calculated for each grid in the three-dimensional grid map, thereby generating a dynamic risk probability map.

[0014] When the risk probability value of any grid in the dynamic risk probability map exceeds the risk threshold, the grid and the area within a predetermined range around it are marked as a high-risk avoidance zone.

[0015] Using a path planning algorithm, a safe return trajectory is searched and generated on a dynamic risk probability map, from the current location of the drone to the preset return point, while avoiding all high-risk avoidance zones;

[0016] Flight control commands are generated based on the safe return trajectory to control the drone to perform the return operation.

[0017] In some alternative embodiments, the raw environmental data includes at least three-dimensional point cloud data of terrain and buildings acquired by lidar.

[0018] In some alternative embodiments, dynamic airspace intelligence data includes at least geofencing information for temporary no-fly zones or real-time weather data.

[0019] In some optional embodiments, the step of calculating a risk probability value for each grid cell includes:

[0020] Based on a multi-factor risk assessment model, the risk probability value is calculated by combining the probability of static obstacles occupying the grid, the weight of temporary no-fly zones, and the impact factor of severe weather.

[0021] In some alternative embodiments, the probability of a static obstacle being occupied is calculated based on the point cloud density falling into the grid in the 3D point cloud data.

[0022] In some optional embodiments, the step of marking the grid and the area within a predetermined radius around it as a high-risk avoidance zone includes:

[0023] Centered on the grid with a risk probability value exceeding the risk threshold, a buffer expansion algorithm is applied to generate a buffer area covering the three-dimensional space surrounding the grid, and the entire buffer area is used as a high-risk avoidance zone.

[0024] In some optional embodiments, the method further includes, after generating the safe return trajectory:

[0025] While the drone is flying along the safe return trajectory, its real-time distance from the boundary of the high-risk avoidance zone is continuously monitored;

[0026] When the real-time distance is detected to be lower than the safety threshold, a local path replanning mechanism is triggered. Only the current path segment is recalculated to generate an alternative path segment that meets the safety distance, and then smoothly connects it with the original safe return trajectory.

[0027] In some alternative embodiments, the path planning algorithm is a heuristic search algorithm whose search cost function combines the physical length of the path and the risk probability value of the grids traversed along the path.

[0028] In some optional embodiments, the three-dimensional grid map is a multi-layer probabilistic map, with different layers storing static obstacle information, dynamic risk information, and communication signal quality information, respectively.

[0029] In some optional embodiments, the method further includes, before generating the dynamic risk probability map:

[0030] The risk threshold is dynamically adjusted based on the drone's real-time physical state parameters, which include at least the drone's remaining battery power or current flight speed.

[0031] In some alternative embodiments, the risk threshold is increased accordingly when the remaining battery power of the drone is below the battery power threshold; or, the risk threshold is decreased accordingly when the current flight speed of the drone is above the speed threshold.

[0032] In some optional embodiments, the method further includes:

[0033] During the drone's return journey, the quality parameters of its satellite navigation signal were continuously monitored;

[0034] When the quality parameters are below the reliability threshold, the drone's positioning data source is automatically switched to autonomous navigation positioning results based on lidar point cloud matching or visual odometry.

[0035] In some optional embodiments, the step of calculating the risk probability value for each grid cell further includes:

[0036] On a 3D grid map, a signal strength distribution layer is generated based on the measured wireless communication signal strength, and grids with signal strength below the communication threshold are included in the calculation of risk probability values.

[0037] In some optional embodiments, the method further includes:

[0038] Dynamic computing task allocation is performed based on the current load of the airborne processor and the bandwidth of the communication link with the ground station;

[0039] When the communication link bandwidth is higher than the bandwidth threshold, the steps of searching and generating the safe return trajectory are offloaded to the ground station server for execution, and the calculation results are received through the communication link.

[0040] In some alternative embodiments, the connection between the alternative path segment and the original safe return trajectory is achieved using a B-spline curve or a minimum hop trajectory generation algorithm to ensure curvature continuity at the connection point.

[0041] This invention provides an environmental perception-based UAV return-to-home risk warning system, comprising:

[0042] The data acquisition module is configured to collect raw environmental data characterizing the three-dimensional physical environment around the UAV in real time through environmental sensors mounted on the UAV, and to receive dynamic airspace intelligence data through the communication module.

[0043] The risk map generation module is connected to the data acquisition module and is configured to perform time-series synchronous processing on the received raw environmental data and dynamic airspace intelligence data, and integrate them into a three-dimensional raster map. It calculates the risk probability value for each raster in the three-dimensional raster map to characterize the flight risk level, thereby generating a dynamic risk probability map.

[0044] The path planning module is connected to the risk map generation module. It is configured to mark the grid and the area within a predetermined range around it as a high-risk avoidance zone when the risk probability value of any grid in the dynamic risk probability map exceeds the risk threshold. The path planning algorithm is used to search and generate a safe return trajectory on the dynamic risk probability map from the current position of the UAV to the preset return point, and avoid all high-risk avoidance zones.

[0045] The flight control command generation module is connected to the path planning module and is configured to generate flight control commands for controlling the UAV to perform return-to-home operations based on the generated safe return-to-home trajectory.

[0046] This invention provides a drone, including the aforementioned environmental perception-based drone return-to-home risk warning system.

[0047] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention.

[0048] The present invention provides a method and system for early warning of drone return-to-home risk based on environmental perception, which has the following beneficial effects:

[0049] This invention constructs a dynamic risk probability map by fusing real-time environmental data and dynamic airspace intelligence, achieving comprehensive perception of various risks in the UAV return-to-home environment. Risk thresholds are dynamically adjusted based on the UAV's own state, allowing the return-to-home strategy to adaptively balance safety and efficiency. Through path planning, the UAV can bypass high-risk areas, ensuring the safety of the return trajectory. Experiments show that this invention can effectively improve the success rate and safety of UAV return-to-home in complex dynamic environments. Attached Figure Description

[0050] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0051] Figure 1 This is a flowchart of an embodiment of the UAV return-to-home risk warning method based on environmental perception according to the present invention;

[0052] Figure 2 This is a schematic diagram of the structure of an environmental perception-based UAV return-to-home risk warning system according to an embodiment of the present invention. Detailed Implementation

[0053] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the invention will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0054] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0055] The flowchart shown in the attached diagram is merely an illustrative example and does not necessarily include all steps. For example, some steps may be broken down, while others may be combined or partially combined. Therefore, the actual execution order may change depending on the specific circumstances.

[0056] The return-to-home process for drones involves multi-sensor data fusion, environmental modeling, probabilistic risk assessment, and path planning. Multi-sensor data fusion, using methods such as Kalman filtering, spatiotemporally calibrates and complements heterogeneous data from lidar, visual sensors, and communication links, enhancing the accuracy and robustness of environmental perception. Environmental modeling typically employs data structures such as 3D grid maps or octrees to convert the fused environmental information into a computer-understandable representation, providing a foundation for subsequent path planning. Probabilistic risk assessment utilizes methods such as Bayesian networks to comprehensively consider multiple risk factors (such as obstacle occupancy, no-fly zones, and signal attenuation), quantifying environmental uncertainties and converting them into calculable risk probability values. Path planning, based on algorithms such as Algorithms and Restricted Response Timers (RRT), searches for the optimal path from the current location to the target point within the environmental model, considering both path length and risk probability to ensure the drone reaches its destination safely and efficiently. By fusing multi-source information to construct environmental cognition and quantifying uncertainties to aid decision-making, the process can guide drones to make reasonable return-to-home decisions in complex and dynamic environments.

[0057] like Figure 1 As shown in the figure, this embodiment of the invention provides a method for early warning of drone return-to-home risk based on environmental perception. The method includes the following steps:

[0058] Step S100: Data Acquisition. The UAV acquires raw data describing its surrounding environment through onboard sensor components and receives environmental intelligence information through a wireless communication module. These sensor components may include, but are not limited to, visual sensors, depth sensors, or lidar, for perceiving the three-dimensional physical environment around the UAV. The wireless communication module receives information from ground stations or other networks, such as flight restriction notices issued by airspace management agencies or weather warnings issued by meteorological departments.

[0059] Step S200: Dynamic Risk Probability Map Generation. In the UAV's onboard processor, preprocessing operations are performed on the raw data and environmental intelligence information collected in Step S100, such as timestamp synchronization and coordinate system transformation. The preprocessed data is fused into a three-dimensional spatial data structure, which characterizes the UAV's flight environment, such as a three-dimensional grid map. For each cell in this data structure, a risk probability value is calculated, reflecting the risk level the UAV might encounter while flying at that location. This step generates a dynamic risk probability map. In some implementations, this three-dimensional spatial data structure can also be an octree or a point cloud map, etc.

[0060] Step S300: High-Risk Avoidance Zone Marking. Evaluate the risk probability value of each cell in the dynamic risk probability map. If the risk probability value of a cell exceeds a preset risk threshold, the area is considered a high-risk area, requiring the UAV to avoid it during its return flight. Mark the high-risk cell and its surrounding area within a certain range as a high-risk avoidance zone. In some implementations, this predetermined range can be adaptively adjusted based on the UAV's flight speed and maneuverability.

[0061] Step S400: Safe Return Trajectory Generation. A safe return trajectory from the UAV's current location to a preset return point is searched on a dynamic risk probability map using a path planning algorithm. This algorithm takes the dynamic risk probability map as input, and the UAV's current location and the preset return point as the start and end points. The generated trajectory must avoid all high-risk avoidance zones marked in step S300. In some implementations, this path planning algorithm can be a graph search-based algorithm or an optimization-based algorithm.

[0062] Step S500: Flight control command generation. Based on the safe return trajectory generated in step S400, flight control commands capable of controlling the UAV's flight attitude and motion state are generated. These flight control commands are sent to the UAV's flight control system, driving the UAV to fly along the planned safe return trajectory and ultimately safely return to the preset return point. In some implementations, the flight control commands may include information such as desired position, desired speed, and desired attitude.

[0063] Through the above steps, this embodiment uses airborne sensors and environmental intelligence information for environmental perception, generates a dynamic risk probability map, and performs path planning based on this map, thereby effectively avoiding potential risks during the return journey. Step S100 provides environmental information input, step S200 fuses multi-source information into a unified risk assessment, step S300 determines the areas to be avoided, step S400 generates a global return route, and step S500 executes the return action. This series of steps works in concert, elevating the return process from simple GPS positioning and navigation to an intelligent decision-making process that comprehensively considers multiple risk factors. This solves the problems of static environmental perception, single risk assessment dimensions, and rigid decision-making mechanisms in existing technologies, enabling UAVs to safely and autonomously complete the return mission in dynamic and complex environments.

[0064] Through the above solution, this embodiment provides a method for autonomous return of unmanned aerial vehicles (UAVs) in complex and dynamic environments. Compared with existing technologies, the advantages of this embodiment are: it can perceive and avoid multiple risk factors in real time during the return process, significantly improving the survivability and mission success rate of UAVs in unknown environments.

[0065] In one specific implementation, the raw environmental data includes at least three-dimensional point cloud data of terrain and buildings collected by a lidar system. In practice, the UAV can be equipped with an HDL-32E lidar, which provides three-dimensional point cloud data within a 360° horizontal and +10° to -30° vertical range using a surround scan method. First, the data acquisition module receives the raw point cloud data stream output by the lidar at a frequency of 10Hz. Each point cloud data point contains three-dimensional coordinates (x, y, z) and reflection intensity information. Specifically, the module preprocesses the received raw point cloud data, including noise filtering and outlier removal, to improve the quality of the point cloud data. Then, the preprocessed point cloud data is transformed into a globally unified three-dimensional raster map coordinate system according to a set coordinate transformation relationship. During the transformation, real-time pose information provided by the UAV's onboard high-precision inertial navigation system (INS) is used to perform precise coordinate transformation and attitude correction on the point cloud data, ensuring accurate alignment between the point cloud data and the raster map. Next, the converted 3D point cloud data is stored in a 3D grid map with a resolution of 0.5 meters. Each grid cell records the number of point clouds falling within it, which is used for subsequent obstacle occupancy probability assessment. In other alternative implementations, an RGBD camera or a binocular vision system can be used instead of LiDAR to acquire depth information and construct point cloud data; or multiple LiDARs can be arranged in an array to obtain 3D environmental information with a wider range or higher precision.

[0066] Through the above solution, this embodiment can use LiDAR to acquire high-precision three-dimensional point cloud data, thereby providing reliable environmental perception information for UAVs, and supporting more accurate risk assessment and safer path planning.

[0067] In one specific implementation, the process of receiving and processing dynamic airspace intelligence data is as follows:

[0068] First, the communication module periodically retrieves the latest dynamic airspace intelligence data from a designated UTM system or meteorological service platform via a pre-defined API interface. Specifically, for geofencing information of temporary no-fly zones, the system uses common data formats such as JSON or GeoJSON. This data format includes the unique identifier of the no-fly zone, the effective time window (start and end time), and a list of polygon vertex coordinates defining the no-fly zone boundary. Each vertex coordinate is represented by longitude and latitude, for example, "coordinates": [[116.3971, 39.9075], [116.4005, 39.9075], [116.4005, 39.9109], [116.3971, 39.9109], [116.3971, 39.9075]]. For real-time meteorological data, the system receives meteorological forecast files conforming to international standards such as GRIB2 or NetCDF. These documents contain forecasts of meteorological elements such as wind speed, wind direction, and precipitation probability at different geographical locations and altitudes.

[0069] Then, the risk map generation module parses the received data. For no-fly zone data, this module converts the polygon vertex coordinates into a set of graticles in a unified 3D raster map coordinate system. All areas within these graticles are marked as no-fly zones. For meteorological data, this module extracts wind speed and direction information near the drone's flight altitude and calculates wind shear intensity. If the wind shear intensity exceeds a preset safety threshold, the graticles within that area are marked as areas affected by severe weather.

[0070] Next, the risk map generation module considers no-fly zones and severe weather impacts as risk factors when calculating the risk probability value for each grid cell. For example, if a grid cell is located within a no-fly zone, its risk probability value will be set to 1.0, indicating that entry is completely prohibited. If a grid cell is located within an area affected by severe weather, its risk probability value will be weighted and adjusted according to the wind shear intensity.

[0071] In other alternative implementations, the geofencing information for the no-fly zone can also be in other formats, such as Shapefile or KML, and the system will process it using the corresponding parsing library. Real-time meteorological data can also be obtained directly by subscribing to a meteorological service API, rather than receiving meteorological forecast files.

[0072] Through the above solution, this embodiment enables drones to perceive and avoid dynamically changing airspace risks in real time, such as temporary no-fly zones and severe weather conditions, thereby further improving their safety and reliability in complex environments.

[0073] In one specific implementation, a multi-factor risk assessment model is used for the step of calculating the risk probability value for each grid cell. The specific implementation of this model includes: First, determining the specific risk factors involved in the model. In this embodiment, in addition to the probability of static obstacle occupancy, the weight of temporary no-fly zones, and the impact factor of severe weather, a communication link quality factor is also introduced. Then, corresponding weight coefficients are set for each risk factor. In this embodiment, the weight coefficients are preset as follows: (Probability of static obstacle occupancy) (Corresponding to the weight of the temporary no-fly zone) (corresponding to severe weather impact factors), and (Corresponding to the communication link quality factor), and the sum of all weight coefficients is 1. Next, calculate the specific values ​​for each risk factor. Static obstacle occupancy probability. The calculation method is described in other embodiments. Temporary no-fly zone weight. The value is either 1 or 0 depending on whether the grid is located within a no-fly zone geofence. Severe Weather Impact Factor The wind speed level in real-time meteorological data is linearly mapped to the interval [0, 1]. Communication link quality factor. Mapping is performed based on the Received Signal Strength Indication (RSSI) value; for example, when the RSSI is higher than -80 dBm, When RSSI is below -95dBm, The intermediate values ​​are calculated using linear interpolation. Finally, each factor and its corresponding weight are substituted into the following multi-factor risk assessment model to calculate the final risk probability value of the grid cell. :

[0074]

[0075] in: This represents the final risk probability value of the raster cell. It is the weighting coefficient of the probability of static obstacles being occupied. It is the probability that a grid cell is occupied by a static obstacle. It is the weighting coefficient for the temporary no-fly zone. This is a risk factor that the grid is located within a temporary no-fly zone. It is the weighting coefficient of the adverse weather factors. It is a risk factor for the grid to encounter severe weather. It is the weighting coefficient of the communication link quality factor. It is a risk factor for the communication link quality of the grid.

[0076] In some alternative implementations, the multi-factor risk assessment model can employ a neural network model or a Bayesian network model to learn the nonlinear relationships between various risk factors based on historical data. In other implementations, the weighting coefficients can be dynamically adjusted according to the UAV's mission type or flight environment.

[0077] Through the above approach, this embodiment can more comprehensively assess flight risks and take into account multiple risk factors, thereby improving the safety of the drone's return flight.

[0078] In one specific implementation, the static obstacle occupancy probability is calculated based on the point cloud density falling into the grid in the three-dimensional point cloud data.

[0079] The raw 3D point cloud data provided by the data acquisition module is transmitted to the risk map generation module. Specifically, this module traverses each grid cell in the 3D raster map and counts the number of LiDAR point clouds falling within that grid cell. The point cloud density of the grid cell is calculated based on the number of point clouds by dividing the number of point clouds by the grid cell volume. The calculated point cloud density is input into a preset function model, which converts the point cloud density into a static obstacle occupancy probability. In one example, the function model can use the Sigmoid function. By adjusting the parameters of the Sigmoid function, a non-linear mapping between point cloud density and static obstacle occupancy probability can be achieved. For example, when the point cloud density within a grid cell is less than a set density threshold, the static obstacle occupancy probability of that grid cell is close to zero; when the point cloud density is greater than the density threshold, the static obstacle occupancy probability is close to 1. In other optional implementations, the function model can also use a Gaussian function or other piecewise linear functions to adapt to the risk assessment accuracy requirements of different application scenarios.

[0080] Through the above solution, this embodiment can more accurately assess the probability that a grid is occupied by static obstacles based on the point cloud density of the lidar, providing the UAV with more accurate environmental perception information, thereby improving the reliability of path planning.

[0081] In one specific implementation, the step of marking the grid and its surrounding predetermined area as a high-risk avoidance zone includes: First, the system obtains a dynamic risk probability map output by the risk map generation module. Specifically, this map is stored in the memory of the onboard computer in the form of a three-dimensional grid array, with each grid corresponding to a floating-point value representing the risk probability value of the spatial area. Then, the path planning module traverses the grid array and identifies all grids whose risk probability values ​​exceed a preset risk threshold. Next, for each identified high-risk grid, the module executes a buffer expansion algorithm. This embodiment uses a three-dimensional voxel dilation algorithm, which starts from the center point of the high-risk grid and expands outward along the X, Y, and Z axes by a predetermined number of grid units. Specifically, the expansion number along each axis is determined by the buffer radius Rbuffer and the grid size. For example, assuming the buffer radius Rbuffer = 5 meters and the grid size is 2 meters, the expansion number along each axis is 2.5, rounded up to 3 grid units. During the expansion process, the algorithm marks all expanded grids as "high-risk avoidance zones". These marked grids will be added to a separate grid set for use by subsequent path planning algorithms. It is worth noting that in this embodiment, the buffer expansion algorithm calculates a three-dimensional cube. In some other alternative implementations, a spherical expansion algorithm can also be used to make the buffer spherical, providing a more uniform safety margin. Furthermore, the shape of the buffer can be adjusted according to the size and motion characteristics of the UAV; for example, an ellipsoidal buffer can be used for a long, narrow UAV.

[0082] Through the above solution, this embodiment can provide a safer and more reliable flight path for drone planning, avoiding accidents caused by getting too close to potentially risky areas.

[0083] In one specific implementation, after generating the safe return trajectory, the process further includes: First, continuously monitoring the real-time distance between the drone and the boundary of the high-risk avoidance zone. Specifically, the system maintains a high-efficiency spatial index structure (e.g., kd-tree or R-tree) to store boundary information for all high-risk avoidance zones. Then, by querying this index structure, the boundary of the high-risk avoidance zone closest to the drone's current position can be quickly found. Next, the shortest distance from the drone's current position to this boundary is calculated. This process is repeated to achieve continuous monitoring of the real-time distance.

[0084] Then, when the real-time distance is detected to be below a safety threshold, a local path replanning mechanism is triggered. This mechanism activates a lightweight path planner that searches for new feasible paths only within a small local area near the drone's current location. This local area can be a spherical region with a radius of several meters centered on the drone.

[0085] Next, an alternative path segment that meets the safety distance is recalculated only for the current path segment. This alternative path segment can be generated using various methods, such as the artificial potential field method or the dynamic window method. These methods can calculate a local path that avoids high-risk avoidance zones and matches the current motion state of the UAV in a relatively short time.

[0086] Finally, the alternative path segment is smoothly connected to the original safe return trajectory. During the connection process, it is necessary to ensure the continuity between the alternative path segment and the original trajectory at the connection point to avoid abrupt attitude changes of the drone during flight.

[0087] Through the above solution, this embodiment can achieve a rapid response to sudden risks, and minimize deviations from the original return trajectory while ensuring the safety of the drone, thereby improving the efficiency and reliability of the return.

[0088] In one specific implementation, the path planning module uses the A* algorithm, a heuristic search algorithm, whose cost function is designed to simultaneously consider the physical length of the path and the risk probability value of the grid cells traversed by the path. First, during the algorithm initialization phase, risk weight coefficients need to be set. This is used to adjust the relative importance of path length and risk probability in the cost function. In a specific example, it will be... Setting it to 2.0 indicates that the algorithm is relatively sensitive to risk. Then, during the iterative search process of the A* algorithm, for each neighboring grid node n to be evaluated, its actual cost g(n) from the starting point to the current node is calculated as follows:

[0089]

[0090] in, It is the actual cost from the starting point to the current node n. The parent node from the starting point to the current node n. The actual cost. From the parent node The physical distance to the current node n. It is a preset risk weighting coefficient. It is the risk probability value of the grid corresponding to the current node n.

[0091] Specifically, if the physical distance from the parent node to the current node is 2 meters, the risk probability value of the current node is 0.5, and If the value is 2.0, then the cost increment from the parent node to the current node is... The A* algorithm selects the grid with the smallest total cost (i.e., the actual cost g(n) plus the heuristic cost h(n) to the destination) as the node for the next expansion, until the search reaches the target point.

[0092] Through the above solution, this embodiment enables the UAV to more intelligently balance safety and efficiency when planning its return route, avoiding the selection of routes that are shorter but riskier, thereby improving the safety of the return journey in complex environments.

[0093] In one specific implementation, the 3D grid map is organized using a multi-layered data structure, with each layer independently storing different types of risk information. First, the base layer, or static obstacle layer, stores static environmental information obtained from the processing of 3D point cloud data acquired by LiDAR scanning, represented as voxel occupancy probabilities. Specifically, each grid cell corresponds to a static occupancy probability value, calculated using the method described in Example 4, reflecting the likelihood that the grid cell is occupied by a physical obstacle. Next, the dynamic risk layer, superimposed on the static obstacle layer, stores dynamically changing risk information. This layer can be further subdivided into multiple sub-layers, such as a temporary no-fly zone sub-layer for storing geofence information broadcast by the ground-based UTM system; and a meteorological risk sub-layer for storing gridded meteorological data obtained from meteorological services. Each grid cell in the temporary no-fly zone sub-layer is marked as impassable if it is located within a no-fly zone. Each grid cell in the meteorological risk sub-layer is assigned a different risk level based on meteorological parameters such as wind speed and direction. Finally, the communication quality layer stores wireless communication link quality information between the UAV and the ground station. This layer uses Received Signal Strength Indicator (RSSI) data recorded during the UAV's historical flight to generate a continuous signal strength distribution map using an interpolation algorithm. Each grid cell corresponds to a signal strength value, reflecting the communication quality at that location. In other optional implementations, the data in each layer of the multi-layer probabilistic map can use different resolutions. For example, the static obstacle layer uses a higher resolution to accurately describe environmental details, while the dynamic risk layer uses a lower resolution to reduce computational load. Furthermore, different strategies can be used to fuse data from different layers, such as weighted averaging or logical operations, to adapt to different application scenarios.

[0094] Through the above scheme, this embodiment can effectively separate and manage different types of risk information, which is convenient for independent updates and maintenance. It can also flexibly select and combine different risk layers according to different task requirements, thereby improving the flexibility and adaptability of the UAV return-to-home risk warning system.

[0095] In one specific implementation, before generating the dynamic risk probability map, the system first acquires the UAV's current real-time physical state parameters. Specifically, the UAV's flight control system sends telemetry data, including the remaining battery percentage and three-axis velocity vectors, to the onboard processor at a frequency of 10Hz. This data is then fed into an adaptive risk threshold adjustment module. This module internally maintains a risk threshold adjustment strategy related to the UAV's physical state. In one example, this strategy can be represented as a two-dimensional lookup table, where the horizontal axis represents the remaining battery range (e.g., 100%-75%, 75%-50%, 50%-25%, below 25%), and the vertical axis represents the flight speed range (e.g., below 5 m / s, 5-10 m / s, above 10 m / s). Each cell in the table stores a corresponding risk threshold adjustment amount. Based on the current remaining battery and flight speed, the module retrieves the corresponding adjustment amount from the lookup table and adds it to a preset base risk threshold to obtain the final dynamic risk threshold used for subsequent high-risk avoidance zone delineation. In some other alternative implementations, the adjustment strategy can also be implemented using a parameterized mathematical model, for example, by using a polynomial function or neural network to fit the relationship between the risk threshold and physical state parameters such as remaining battery power and flight speed.

[0096] Through the above scheme, this embodiment enables the drone's return-to-home strategy to be more flexible, allowing it to adjust its risk tolerance according to its own status, focusing more on flight efficiency when the battery is sufficient, and more on flight safety when the battery is insufficient.

[0097] In one specific implementation, the flight control system sends the UAV's current real-time physical state parameters, including remaining battery percentage and flight speed in all three axes, to the risk map generation module at a frequency of 20Hz. Then, an adaptive threshold adjuster is deployed within the module. This adjuster contains two independent linear adjustment functions:

[0098] Battery adjustment function: When the received remaining battery percentage is... When the battery level falls below a preset baseline (e.g., 30%), the function calculates the adjusted risk threshold based on the following formula. :

[0099]

[0100] in, It is a risk threshold adjusted based on the amount of electricity. This is the basic risk threshold (e.g., 0.8). It is the power sensitivity coefficient (e.g., 0.01). This is the baseline for battery level (e.g., 30%). This represents the current remaining battery percentage. This formula implies that the lower the battery level, the higher the risk threshold, and the more likely the drone is to choose a shorter, but potentially riskier, path.

[0101] Speed ​​adjustment function: When the received flight speed When the speed exceeds a preset baseline (e.g., 8 m / s), the function calculates the speed-adjusted risk threshold according to the following formula. :

[0102]

[0103] in, It is a risk threshold adjusted based on speed. It is the basic risk threshold. : is the speed sensitivity coefficient (e.g., 0.005). : This is the speed baseline (e.g., 8 m / s). : represents the current flight speed. This formula implies that the higher the speed, the lower the risk threshold, and the more likely the drone is to choose a more conservative but safer path.

[0104] The adjuster takes into account the results of both functions. For example, it can choose the value that makes a more significant adjustment to the base threshold (i.e., more conservative or more aggressive) as the final output risk threshold. This is used for subsequent high-risk avoidance zone delineation and route planning.

[0105] In other alternative implementations, the aforementioned power and speed adjustment functions can also be non-linear, such as using a sigmoid function or a piecewise linear function, to achieve more refined threshold adjustments. The baseline for power and speed can be adjusted according to the specific model of the UAV and mission requirements. Through the above scheme, this embodiment enables the UAV's return-to-home strategy to intelligently adjust its risk tolerance based on its own state, maximizing mission success rate and resource utilization efficiency while ensuring a safety baseline.

[0106] In one specific implementation, the method further includes adding a satellite navigation signal quality monitor to the data acquisition module. This monitor can be an independent software thread that receives raw NMEA data output from a satellite navigation receiver such as GPS or BeiDou from the UAV flight control system at a frequency of 5Hz.

[0107] Specifically, the monitor parses fields in the NMEA data, including the positioning solution status (such as fixed solution, floating-point solution, no solution), number of visible satellites, position precision factor (PDOP), horizontal position precision (HDOP), and vertical position precision (VDOP).

[0108] The navigation signal quality assessment value is calculated according to the following rules: If the positioning solution is unsolvable, the quality assessment value is 0; if the number of visible satellites is less than 4, or the PDOP value is greater than 2.5, or the HDOP or VDOP value is greater than 2.0, the quality assessment value is linearly mapped according to the degree of deviation, for example, when the PDOP is 5, the quality assessment value is 0.5. Otherwise, the quality assessment value is 1.

[0109] The system sets a reliability threshold, for example, 0.7. If the navigation signal quality assessment value is lower than this threshold, the satellite navigation signal is considered unreliable. Furthermore, to avoid misjudgments caused by momentary signal fluctuations, the positioning data source will only be switched if the navigation signal quality assessment value remains below 0.7 for more than 2 seconds.

[0110] Specifically, the UAV flight control system is configured to support dynamic switching between multiple positioning data sources. In addition to satellite navigation systems, the autonomous navigation positioning results output by an onboard LiDAR-based SLAM (Simultaneous Localization and Mapping) algorithm can also be selected. The SLAM algorithm uses point cloud data generated by the LiDAR scanning the surrounding environment and estimates the UAV's pose through feature extraction, matching, and optimization steps. In other optional implementations, autonomous navigation positioning results based on visual odometry can also be used. Visual odometry uses image sequences captured by an onboard camera and calculates the UAV's pose changes through image processing and motion estimation methods.

[0111] Then, when the satellite navigation signal quality monitor determines that satellite navigation is unreliable, it sends a positioning source switching command to the flight control system. The flight control system then switches the primary positioning data source from GPS to the autonomous navigation positioning results output by the SLAM algorithm or visual odometry. During the switching of positioning data sources, the flight control system performs a smooth transition, for example, by using Kalman filtering to fuse multiple positioning data sources, to ensure the continuity and smoothness of the UAV's positioning.

[0112] Next, the path planning module generates a safe return trajectory based on the new positioning data source. Since SLAM algorithms or visual odometry are based on local environmental features, their positioning accuracy is typically higher than GPS positioning accuracy in environments with poor satellite navigation signals, thus improving the accuracy and reliability of path planning. In other alternative implementations, multiple autonomous navigation positioning algorithms can be enabled simultaneously and weighted and fused according to their respective positioning accuracy and reliability to further improve the robustness of positioning.

[0113] Through the above solution, this embodiment can automatically switch to autonomous navigation and positioning mode in environments with poor satellite navigation signal quality, thereby ensuring the positioning accuracy and reliability of the UAV during its return journey. Compared with existing technologies, the advantages of this embodiment are: improved adaptability of the UAV to autonomous return in complex environments, reduced risk of return failure due to positioning errors, and expanded application scenarios for the UAV.

[0114] In one specific implementation, the step of calculating the risk probability value for each grid cell further includes: the system first measures the wireless communication signal strength between the UAV and the ground station or base station in real time using an onboard wireless communication module (e.g., a Wi-Fi chip supporting the 802.11ax standard or a 5G NR module). Specifically, the signal strength can be an indicator such as Received Signal Strength Indication (RSSI), Received Signal Quality (RSRQ), or Signal-to-Noise Ratio (SNR). The measurement frequency can be set to 1Hz to ensure timely reflection of signal changes. Then, the measured signal strength values ​​and their corresponding UAV geographical locations (provided by GPS or visual odometry) are recorded to form a set of geographically labeled signal strength sample points. Next, the risk map generation module uses these sample points and employs a spatial interpolation algorithm (e.g., inverse distance weighted interpolation, Kriging interpolation, or radial basis function interpolation) to generate a continuous signal strength distribution layer on a three-dimensional grid map. The signal strength value of each grid cell represents the predicted wireless communication quality at that location. For grids with signal strength below a preset communication threshold (e.g., -90dBm, which can be adjusted based on the actual communication environment and the performance of the UAV's communication equipment), their risk probability value is set to a non-zero value, such as 0.3. This value is weighted and summed with other risk factors in Example 3 (probability of static obstacle occupancy, weight of temporary no-fly zones, and adverse weather impact factor) to jointly determine the final risk probability value. Through this scheme, the path planning algorithm tends to avoid areas with weak wireless communication signals when planning the return trajectory, reducing the risk of loss of control due to communication interruption. In some other optional implementations, in addition to measuring the downlink signal strength, the uplink signal strength can also be measured simultaneously to comprehensively evaluate the quality of two-way communication. Alternatively, the signal strength distribution layer can use different color codes to represent the signal strength level, rather than just a binary risk probability value.

[0115] Through the above solution, this embodiment can incorporate wireless communication quality into the return-to-home risk assessment system, enabling the UAV to select a path with good signal in complex electromagnetic environments, thereby reducing the risk of communication interruption and enhancing the reliability of the return-to-home process.

[0116] In one specific implementation, before the drone performs its return mission, the onboard mission allocation decision-maker first reads the CPU and GPU utilization of the onboard processor (e.g., NVIDIA Jetson Orin NX) and assesses the current 5G communication link bandwidth with the ground station. Specifically, processor utilization is obtained by reading performance counters provided by the operating system, and link bandwidth is estimated by monitoring the transmit and receive rates of the network interface. A processor utilization threshold of 80% and a bandwidth threshold of 50 Mbps are set.

[0117] First, the task allocation decision-maker determines whether the processor utilization exceeds 80%. If it does, it indicates that onboard computing resources are strained. Next, it determines whether the communication link bandwidth is higher than 50Mbps. If it is, it indicates that there is a sufficient high-speed data channel between the UAV and the ground station. If both of these conditions are met, the task allocation decision-maker will trigger the task offloading process.

[0118] Specifically, the task unloading process is as follows: The airborne data acquisition module compresses the raw 3D point cloud data acquired by the LiDAR using fast lossless compression algorithms such as LZ4 or Zstd to reduce data transmission volume. Then, the compressed point cloud data, the UAV's current pose information, and the coordinates of the return target point are sent to the ground station server via a 5G communication link. After receiving the data, the ground station server decompresses the point cloud data and executes the risk map generation and global path planning steps of Example 1. After completing the path planning, the ground station compresses the waypoint sequence of the global path (e.g., selecting a waypoint every 5 meters) and a portion of the raster data of the risk map (for local obstacle avoidance) and sends it back to the UAV. After receiving the waypoint sequence from the ground station, the airborne path planning module performs local path adjustment and trajectory smoothing to generate the final flight control commands. If the link bandwidth is lower than the bandwidth threshold, all computational tasks are still executed locally on the UAV.

[0119] In other alternative implementations, task offloading decisions can be based not only on processor load and link bandwidth, but also on other factors such as the UAV's remaining battery power, the available computing resources of the ground station server, and the complexity of the return path. Furthermore, the task allocation decision-maker can employ different task allocation strategies; for example, offloading only the computationally intensive and risky map generation step to the ground station, while retaining the path planning step for execution on the airborne platform. The communication protocol between the ground station and the UAV can use more reliable protocols such as QUIC or WebTransport.

[0120] Through the above solution, this embodiment can make full use of the ground station's computing resources, and even when airborne computing resources are limited, it can still ensure the real-time performance and safety of the UAV's return flight, while reducing the power consumption of the airborne platform.

[0121] In one specific implementation, during the smooth transition, the intersection point of the original safe return trajectory and the alternative path segment generated by local replanning is first determined. Then, based on these two intersection points, a B-spline curve is constructed. Specifically, the order of the B-spline curve is set to 3 to ensure good smoothness. The number of control points is adaptively adjusted according to the distance to the intersection points, generally set to 4 to 6, with the first and last control points coinciding with the intersection points to ensure the curve passes through these points.

[0122] Next, by optimizing the algorithm (e.g., gradient descent), the positions of the remaining control points are adjusted so that the tangent direction of the B-spline curve at the intersection point is as consistent as possible with the original safe return trajectory and the alternative path segment, in order to minimize curvature abrupt changes. At the same time, the distribution range of the control points is constrained to avoid the generated curve being too curved or exceeding the maximum flight capability of the UAV.

[0123] In some alternative implementations, a minimum hop trajectory generation algorithm can be used instead of a B-spline curve. This algorithm directly generates a smooth trajectory that satisfies the continuity constraints of position, velocity, and acceleration by minimizing the integral of the trajectory's hop (the derivative of acceleration). Specifically, the position, velocity, and acceleration of the original safe return trajectory and the alternative path segment at their intersection point can be used as boundary conditions to construct an optimization problem. Solving this problem yields a fifth-order or higher Bernstein polynomial curve connecting the two intersection points. By adjusting the coefficients of the Bernstein polynomial, fine control over the trajectory shape can be achieved.

[0124] Through the above scheme, this embodiment can ensure a smooth transition between the alternative path and the original return trajectory, avoiding instability of the UAV attitude due to sudden trajectory changes, thereby improving flight safety and control accuracy.

[0125] This invention provides an environmental perception-based UAV return-to-home risk warning system, such as... Figure 2 As shown, the system includes a data acquisition module M100, a risk map generation module M200, a path planning module M300, and a flight control command generation module M400.

[0126] The data acquisition module M100 is configured to collect raw environmental data characterizing the three-dimensional physical environment surrounding the UAV in real time using various environmental sensors mounted on the UAV, and to receive dynamic airspace intelligence data via a communication module. In one implementation, the data acquisition module M100 includes, but is not limited to, various environmental sensors such as lidar, cameras, and ultrasonic sensors, used to acquire three-dimensional information, image information, and distance information of the environment. The communication module can be a 4G / 5G communication module used to receive dynamic airspace intelligence data from ground stations or other data sources, such as temporary no-fly zone information and meteorological information. In other optional implementations, the data acquisition module M100 can employ other types of sensors, such as millimeter-wave radar and infrared sensors, to adapt to different environmental and mission requirements. The communication module can also employ other communication methods such as satellite communication and Wi-Fi.

[0127] The risk map generation module M200, connected to the data acquisition module M100, is configured to synchronously process received raw environmental data and dynamic airspace intelligence data in a time series and fuse them into a 3D raster map. It calculates a risk probability value for each raster in the 3D raster map, representing the flight risk level, thereby generating a dynamic risk probability map. In one implementation, the risk map generation module M200 uses methods such as Kalman filtering or particle filtering to fuse data from different sources. The 3D raster map can employ an octree structure or other spatial indexing structures to improve storage and retrieval efficiency. The calculation of risk probability values ​​can employ various methods, such as probabilistic inference based on Bayesian networks or support vector machine classification. In other optional implementations, the risk map generation module M200 can employ other types of data fusion algorithms, such as extended Kalman filtering or information filtering, to adapt to different sensor noise characteristics and data correlation levels. The 3D raster map can also employ other types of spatial data structures, such as KD trees or quadtrees. The calculation of risk probability values ​​can employ other types of machine learning algorithms, such as neural networks or decision trees.

[0128] The path planning module M300 is connected to the risk map generation module M200 and is configured to mark the grid and its surrounding area within a predetermined range as a high-risk avoidance zone when the risk probability value of any grid in the dynamic risk probability map exceeds a risk threshold. It then uses a path planning algorithm to search and generate a safe return trajectory on the dynamic risk probability map from the UAV's current position to a preset return point, avoiding all high-risk avoidance zones. In one implementation, the path planning module M300 uses an expansion algorithm or other region growth algorithm to mark the area around the risk grid as a high-risk avoidance zone. The path planning algorithm can be the A* algorithm, Dijkstra's A* algorithm, RRT algorithm, etc. In other optional implementations, the path planning module M300 can employ different region growth algorithms, such as erosion algorithms, opening algorithms, etc., to adapt to different risk avoidance requirements. The path planning algorithm can also be other types of search algorithms, such as genetic algorithms, particle swarm optimization, etc.

[0129] The flight control command generation module M400 is connected to the path planning module M300 and is configured to generate flight control commands for controlling the UAV to perform a return-to-home operation based on the generated safe return-to-home trajectory. In one implementation, the flight control command generation module M400 converts the safe return-to-home trajectory into a series of waypoints and generates corresponding speed and attitude control commands based on the UAV's dynamic model. The flight control commands can be encapsulated using the MAVLink protocol or other UAV control protocols. In other alternative implementations, the flight control command generation module M400 can employ other types of trajectory planning algorithms, such as model predictive control and optimal control. The flight control commands can employ other types of UAV control protocols, such as DJI SDK and ROS.

[0130] In the aforementioned system, the data acquisition module M100 is responsible for acquiring environmental information in real time, the risk map generation module M200 is responsible for converting the environmental information into a risk assessment, the path planning module M300 is responsible for generating a safe trajectory based on the risk assessment, and the flight control command generation module M400 is responsible for converting the safe trajectory into commands executable by the UAV. These modules cooperate and work together to enable the UAV to perceive and avoid risks in complex environments, ultimately returning safely to the preset return point. This solves the problem in existing technologies where UAVs cannot safely return in dynamic and complex environments, improving the survivability and mission reliability of UAVs in unknown environments.

[0131] Through the above solution, this embodiment enables the UAV to perceive the risks of the surrounding environment in real time and autonomously plan a safe return trajectory, thereby improving the return safety of the UAV in complex and dynamic environments.

[0132] In one specific implementation, the risk map generation module M200 includes a multi-factor fusion unit, which consists of three parallel sub-modules: a static occupancy probability calculation module, a no-fly zone risk assessment module, and a meteorological risk assessment module. The static occupancy probability calculation module receives LiDAR point cloud data from the data acquisition module M100, converts the point cloud data into 3D raster occupancy information using voxelization, and uses a ray casting algorithm to eliminate shadow areas caused by LiDAR viewpoint obstruction, improving the accuracy of the occupancy probability. This module also includes a dynamic threshold adjuster that dynamically adjusts the occupancy probability threshold based on the UAV's attitude angle, reducing misjudgments caused by UAV tilt. The no-fly zone risk assessment module receives geofence information published by the UTM system from the data acquisition module M100, uses a spatial inclusion algorithm to determine whether a raster is located within a no-fly zone, and assigns different risk weights based on the type of no-fly zone (e.g., temporary no-fly zone, permanent no-fly zone). The meteorological risk assessment module receives meteorological data, including wind speed, wind direction, temperature, and humidity, from the meteorological service interface of the data acquisition module M100. It then establishes a meteorological risk model to calculate the risk level under different meteorological conditions. For example, when the wind speed exceeds a preset threshold, the risk probability value of the corresponding area is increased. The multi-factor fusion unit weights and sums the outputs of the three sub-modules according to preset weighting coefficients to obtain the comprehensive risk probability value for each grid cell, and stores it in a 3D grid map.

[0133] In other alternative implementations, the static occupancy probability calculation module can employ different point cloud processing algorithms, such as using a deep learning-based point cloud segmentation algorithm to segment point cloud data into different objects, thereby improving the accuracy of the occupancy probability; the no-fly zone risk assessment module can employ different spatial inclusion algorithms, such as using an octree-based spatial indexing algorithm, to improve the efficiency of no-fly zone judgment; and the meteorological risk assessment module can employ different meteorological risk models, such as using a neural network-based meteorological risk prediction model, to improve the accuracy of meteorological risk prediction.

[0134] Through the above-described solution, this embodiment can more accurately assess the risks of the UAV flight environment, providing a more reliable basis for subsequent path planning. Compared with existing technologies, the advantages of this embodiment are: it can comprehensively consider multiple risk factors, improving the accuracy and comprehensiveness of risk assessment.

[0135] In one specific implementation, the path planning module M300 includes a distance monitoring unit, a local planning trigger, and a local path generator. The distance monitoring unit continuously receives the UAV's position coordinates reported by the UAV flight control system, while simultaneously accessing the latest high-risk avoidance zone data maintained by the risk map generation module M200. The distance monitoring unit calculates the minimum distance between the UAV's current position and the boundaries of all high-risk avoidance zones and compares this distance value with a preset safety threshold. The safety threshold can be set to a fixed value, such as 3 meters, or dynamically adjusted according to the UAV's flight speed. When the minimum distance is less than the safety threshold, the distance monitoring unit sends a trigger signal to the local planning trigger. Upon receiving the trigger signal, the local planning trigger immediately interrupts the currently executing global path tracking task and activates the local path generator. The local path generator extracts a 20-meter segment from the current global safe return trajectory, starting from the UAV's current position, as the local planning range. The local path generator uses a Dynamic Window (DWA) algorithm to search for new alternative path segments within the local planning range. The DWA* algorithm uses the drone's current speed as its initial speed, simulates various speed combinations of the drone within a certain time window (e.g., 2 seconds), and evaluates whether the trajectory under each speed combination will collide with a high-risk avoidance zone. The DWA* algorithm selects an optimal trajectory as an alternative path segment that both avoids the high-risk avoidance zone and allows the drone to return to the global safe return trajectory as quickly as possible. The alternative path segment and the original global safe return trajectory are smoothly connected using a fifth-order polynomial curve to ensure continuity at the connection point.

[0136] Through the above solution, this embodiment can quickly respond to sudden risks that arise during the flight of a drone and are relatively close to the target, avoid collision risks, and improve the safety of drone flight.

[0137] This invention provides a drone that includes an environmental perception-based drone return-to-home risk warning system as described in any of the above embodiments.

[0138] Through the above-described solution, this embodiment enables the UAV to autonomously and safely return to its home location in complex environments. Compared with existing technologies, the advantages of this embodiment are: it can comprehensively perceive and avoid both static and dynamic physical risks, while also considering non-physical risk factors (such as communication signal quality); it has a high degree of decision-making adaptability; it can dynamically adjust its return-to-home strategy according to the UAV's own state and environmental changes; it has excellent real-time response performance; and it can respond promptly to sudden threats.

[0139] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for early warning of drone return-to-home risk based on environmental perception, characterized in that, include: The drone collects raw environmental data in real time, which characterizes the three-dimensional physical environment around it, using environmental sensors mounted on it, and receives dynamic airspace intelligence data through the communication module. In the airborne processor, the raw environmental data and the dynamic airspace intelligence data are processed in a time-series synchronization and fused into a three-dimensional grid map. A risk probability value for each grid in the three-dimensional grid map is calculated to characterize the flight risk level, thereby generating a dynamic risk probability map. When the risk probability value of any grid in the dynamic risk probability map exceeds the risk threshold, the grid and the area within a predetermined range around it are marked as a high-risk avoidance zone. Using a path planning algorithm, a safe return trajectory is searched and generated on the dynamic risk probability map, from the current position of the UAV to the preset return point, and avoiding all the high-risk avoidance zones. Flight control commands are generated based on the safe return trajectory to control the UAV to perform the return operation.

2. The method according to claim 1, characterized in that, The raw environmental data includes at least three-dimensional point cloud data of terrain and buildings collected by lidar, and the dynamic airspace intelligence data includes at least geofence information of temporary no-fly zones or real-time meteorological data. The step of calculating the risk probability value for each of the grid cells includes: Based on the multi-factor risk assessment model, the risk probability value is calculated by combining the static obstacle occupancy probability corresponding to the grid, the weight of the temporary no-fly zone, and the adverse weather impact factor. The static obstacle occupancy probability is calculated based on the point cloud density falling into the grid in the three-dimensional point cloud data.

3. The method according to claim 1, characterized in that, The step of marking the grid and the area within a predetermined radius around it as the high-risk avoidance zone includes: Centered on the grid where the risk probability value exceeds the risk threshold, a buffer expansion algorithm is applied to generate a buffer area covering the three-dimensional space surrounding the grid, and the entire buffer area is used as the high-risk avoidance zone.

4. The method according to claim 1, characterized in that, The method further includes the following after generating the safe return trajectory: While the drone is flying along the safe return trajectory, its real-time distance to the boundary of the high-risk avoidance zone is continuously monitored; When the real-time distance is detected to be lower than the safety threshold, a local path replanning mechanism is triggered. Only the current path segment is recalculated to generate an alternative path segment that meets the safety distance, and then smoothly connects it with the original safe return trajectory. The connection between the alternative path segment and the original safe return trajectory is achieved using a B-spline curve or minimum hop trajectory generation algorithm to ensure curvature continuity at the connection point.

5. The method according to claim 1, characterized in that, The path planning algorithm is a heuristic search algorithm, and its search cost function combines the physical length of the path and the risk probability value of the grids passed through on the path. The three-dimensional raster map is a multi-layer probabilistic map, with different layers storing static obstacle information, dynamic risk information, and communication signal quality information, respectively.

6. The method according to claim 1, characterized in that, The method further includes the following steps before generating the dynamic risk probability map: The risk threshold is dynamically adjusted based on the real-time physical state parameters of the drone itself, wherein the real-time physical state parameters include at least the drone's remaining battery power or current flight speed. When the remaining battery power of the drone is lower than the battery power threshold, the risk threshold is increased accordingly; or, when the current flight speed of the drone is higher than the speed threshold, the risk threshold is decreased accordingly.

7. The method according to claim 1, characterized in that, The method further includes: During the return journey of the UAV, the quality parameters of its satellite navigation signal are continuously monitored; When the quality parameter is lower than the reliability threshold, the UAV's positioning data source is automatically switched to the autonomous navigation positioning result based on lidar point cloud matching or visual odometry.

8. The method according to claim 1, characterized in that, The step of calculating the risk probability value for each of the grid cells further includes: On the three-dimensional grid map, a signal strength distribution layer is generated based on the measured wireless communication signal strength, and the grids with signal strength below the communication threshold are included in the calculation of the risk probability value.

9. The method according to claim 1, characterized in that, The method further includes: Dynamic computing task allocation is performed based on the current load of the airborne processor and the bandwidth of the communication link with the ground station; When the bandwidth of the communication link is higher than the bandwidth threshold, the steps of searching and generating the safe return trajectory are offloaded to the ground station server for execution, and the calculation results are received through the communication link.

10. A UAV return-to-home risk early warning system based on environmental perception, characterized in that, include: The data acquisition module is configured to collect raw environmental data characterizing the three-dimensional physical environment around the UAV in real time through environmental sensors mounted on the UAV, and to receive dynamic airspace intelligence data through the communication module. The risk map generation module is connected to the data acquisition module and is configured to perform time-series synchronization processing on the received raw environmental data and dynamic airspace intelligence data, and merge them into a three-dimensional raster map. It calculates a risk probability value for each raster in the three-dimensional raster map to characterize the flight risk level, thereby generating a dynamic risk probability map. The path planning module is connected to the risk map generation module and is configured to mark the grid and the area within a predetermined range around it as a high-risk avoidance zone when the risk probability value of any grid in the dynamic risk probability map exceeds the risk threshold. The path planning algorithm is used to search and generate a safe return trajectory on the dynamic risk probability map from the current position of the UAV to the preset return point and bypass all the high-risk avoidance zones. A flight control command generation module is connected to the path planning module and is configured to generate flight control commands for controlling the UAV to perform a return-to-home operation based on the generated safe return-to-home trajectory.

Images