Unmanned aerial vehicle water surface risk perception and active cooperative planning method and system

Abstract

The present application relates to the technical field of unmanned aerial vehicle control, and more particularly to a method and system for risk perception and active cooperation planning of unmanned aerial vehicles on water surface, comprising: acquiring distance perception, flight state and local environment map information to perform effectiveness screening and generate candidate risk events; performing local risk analysis in combination with the aforementioned information and historical observation features to output risk certainty evaluation results; comparing and determining conditions to generate hierarchical disposal instructions, triggering local conservative risk constraints for low certainty, and triggering local range review and update for high certainty; extracting instruction processing results to convert into unified risk areas; synchronously applying the unified risk areas to the exploration target selection and trajectory avoidance planning process to execute high-risk target suppression and generate avoidance trajectories; and outputting mode switching and flight execution results based on the avoidance trajectories to finally form an active avoidance control closed loop.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, and in particular to a method and system for UAV surface risk perception and proactive collaborative planning. Background Technology

[0002] When UAVs perform autonomous exploration missions in unknown environments, the flight system needs to continuously acquire environmental boundary information to avoid physical obstacles. Water surfaces, as unique physical risk targets, possess strong specular reflection and signal absorption characteristics, resulting in highly discontinuous echo data acquired by onboard sensing equipment in terms of time sequence. Traditional UAV obstacle avoidance technologies mostly employ a loosely coupled, single-line control architecture for perception and planning. Due to the lack of dynamic hierarchical processing strategies for observational uncertainties and rigid no-entry boundaries shared across modules, the underlying controller is highly susceptible to misjudging the absolute safety of the space below due to missed single-frame signals in dynamic water surface environments. This can lead to serious problems such as UAVs excessively intruding into high-risk water areas under exploration commands, resulting in irreversible crashes. Summary of the Invention

[0003] To overcome the above shortcomings, this invention provides a method and system for UAV water surface risk perception and active collaborative planning, aiming to improve the problem that traditional UAV obstacle avoidance mostly adopts loosely coupled single-line logic, which causes UAVs to crash due to excessive intrusion into water areas due to unstable water surface observation and lack of shared collision avoidance boundaries.

[0004] In a first aspect, the present invention provides the following technical solution: a method for unmanned aerial vehicle (UAV) surface risk perception and active collaborative planning, comprising the following steps: S1. Acquire the distance perception information, flight status information and local environment map information of the UAV, perform validity screening on the distance perception information, and generate candidate risk events when the screening is deemed valid; S2. Combining the candidate risk events, the flight status information, the local environment map information, and the extracted historical observation features, perform local risk analysis and output the risk certainty assessment result; S3. Compare the risk certainty assessment result with the preset classification judgment conditions to generate a classification handling instruction. If the risk certainty assessment result falls into the low certainty range, a local conservative risk constraint instruction is triggered. If the risk certainty assessment result falls into the high certainty range, a local scope review and update instruction is triggered. S4. Extract the processing results corresponding to the local conservative risk constraint instruction or the local range review and update instruction, and convert the processing results into a unified risk area; S5. Simultaneously apply the unified risk area to the UAV's target selection process and trajectory avoidance planning process, so as to perform high-risk target suppression in the target selection process and generate an avoidance trajectory in the trajectory avoidance planning process; S6. Based on the avoidance trajectory output mode switching information and flight execution results, an active avoidance control closed loop is formed.

[0005] By adopting the above technical solutions, risks are classified and transformed into a unified risk area, thereby simultaneously constraining target selection and trajectory planning to form an avoidance closed loop. This improves the problem that traditional UAV obstacle avoidance mostly adopts loosely coupled single-line logic, which causes UAVs to crash due to excessive intrusion into water areas due to unstable water surface observation and lack of shared collision avoidance boundaries.

[0006] Secondly, this invention provides the following technical solution: an unmanned aerial vehicle (UAV) surface risk perception and active collaborative planning system, comprising the following modules: The perception acquisition and filtering module is used to acquire the distance perception information, flight status information and local environment map information of the UAV, perform validity filtering on the distance perception information, and generate candidate risk events when the filtering is deemed valid; The local risk analysis module is used to combine the candidate risk events, the flight status information, the local environment map information, and the extracted historical observation features to perform local risk analysis and output the risk certainty assessment results. The graded instruction generation module is used to compare the risk certainty assessment results with preset graded judgment conditions and generate graded disposal instructions. The risk area conversion module is used to extract the processing results corresponding to the graded disposal instructions and convert the processing results into a unified risk area; The collaborative planning constraint module is used to simultaneously apply the unified risk area to the UAV's target selection process and trajectory avoidance planning process; The avoidance execution output module is used to form an active avoidance control closed loop based on the avoidance trajectory output mode switching information and flight execution results generated during the trajectory avoidance planning process.

[0007] The present invention has the following beneficial effects: 1. In this invention, by classifying and handling risks and transforming them into a unified risk area, the target selection and trajectory planning are simultaneously constrained to form an avoidance closed loop. This improves the problem that traditional UAV obstacle avoidance mostly adopts loosely coupled single-line logic, which causes UAVs to crash due to excessive intrusion into water areas because of unstable water surface observation and lack of shared collision avoidance boundaries.

[0008] 2. In this invention, the effective point cloud is filtered by calculating the dynamic incident angle of the sensor beam angle and pitch angle, thereby blocking the multipath interference caused by the change of body attitude. This improves the problem that most traditional sensing modules use indiscriminate data reception, and the obstacle avoidance system frequently triggers errors due to signal drift caused by water surface reflection when the body is tilted.

[0009] 3. In this invention, by performing a logical AND operation between the Boolean truncation mask of the logical false value and the exploration gain, the problem that traditional task planning mostly adopts a soft cost decay mechanism is improved. Since conventional penalties are difficult to offset the traction of exploration gains, this causes UAVs to easily intrude into dangerous blind spots in water.

[0010] 4. In this invention, by incorporating the repulsive potential field penalty term into the trajectory optimization cost function for iterative solution, a smooth and continuous anti-collision physical repulsive force is established, thereby improving the problem that traditional flight control obstacle avoidance mostly adopts a rule-based hard triggering mechanism. Since rigid blocking logic is prone to causing low-level level jumps on dynamic water surfaces, it can cause severe oscillations in the UAV's heading. Attached Figure Description

[0011] Figure 1 This is a flowchart of the UAV surface risk perception and active collaborative planning method proposed in this invention. Figure 2 This is a flowchart illustrating the underlying calculation process for effectiveness screening and local risk analysis of the UAV surface risk perception and proactive collaborative planning method proposed in this invention. Figure 3 This is a flowchart illustrating the risk assessment, grading decision-making, and regional reshaping process of the UAV surface risk perception and proactive collaborative planning method proposed in this invention. Figure 4 This is a flowchart of the cross-domain dual-path collaborative constraint and physical drive control of the UAV surface risk perception and active collaborative planning method proposed in this invention. Figure 5 This is an architecture diagram of the UAV surface risk perception and active collaborative planning system proposed in this invention. Detailed Implementation

[0012] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Example 1: In a first embodiment of the present invention, the present invention provides a method for unmanned aerial vehicle (UAV) surface risk perception and active cooperative planning, such as... Figures 1-2As shown, it includes the following steps: S1. Acquire distance perception information, flight status information, and local environment map information of the UAV; perform validity screening on the distance perception information; and generate candidate risk events when the screening is deemed valid. Furthermore, in S1, validity screening is performed on the distance-aware information, and candidate risk events are generated when the screening is deemed valid, including: Extract the sensor vertical beam angle corresponding to the distance perception information, and extract the current pitch angle and current relative altitude corresponding to the flight status information; The sensor's vertical beam angle is summed with the current pitch angle to output dynamic incident angle evaluation parameters. Extract the preset critical incident angle threshold and the maximum effective detection height constant; Determine whether the dynamic incident angle evaluation parameter is less than the critical incident angle threshold, and simultaneously determine whether the current relative height is less than the maximum effective detection height constant; When the dynamic incident angle evaluation parameter is determined to be less than the critical incident angle threshold and the current relative height is less than the maximum effective detection height constant, the distance sensing information is marked as effective feature point cloud data. Perform density-based spatial clustering on the effective feature point cloud data, extract the boundary coordinates of the clustered connected regions, and use the boundary coordinates of the clustered connected regions as candidate risk events.

[0014] Specifically, step S1 is executed to construct the front-end data input channel and the underlying false alarm prevention mechanism of the entire active avoidance system. The system's underlying communication interface is responsible for acquiring three heterogeneous data sources from the UAV's onboard hardware: distance perception information, flight status information, and local environment map information. Distance perception information consists of a 3D point cloud generated by the onboard millimeter-wave radar or lidar within the current sampling period. Flight status information is calculated and streamed in real time by the inertial measurement unit built into the flight control system. Local environment map information is stored in the memory space of the onboard navigation computer.

[0015] The system receives and caches the three types of data mentioned above, performs a validity screening operation on the distance sensing information, and uses physical space constraint laws to eliminate invalid multipath interference signals caused by water surface reflection. When the screening status is determined to be valid, the system extracts abnormal echo points to generate candidate risk events, and writes the generated candidate risk events into shared video memory for use by the subsequent local risk analysis module.

[0016] The specific execution process for performing validity screening and generating candidate risk events based on distance-aware information consists of two stages: attitude verification and spatial clustering.

[0017] During the attitude verification phase, the system reads the hardware configuration file to extract the sensor vertical beam angle corresponding to the distance perception information. Simultaneously, the system parses the flight status information data frames to extract the UAV's current pitch angle and current relative altitude. The underlying computing unit sums the sensor vertical beam angle and the current pitch angle, outputting dynamic incident angle evaluation parameters. The mathematical model of this computational logic is expressed as follows: ; in This represents the dynamic incident angle evaluation parameter. This indicates the sensor's vertical beam angle. Indicates the current pitch angle.

[0018] The arithmetic unit retrieves the preset critical incident angle threshold and the maximum effective detection height constant from the system's non-volatile memory. The system establishes a Boolean logic comparator to determine whether the dynamic incident angle evaluation parameter is less than the critical incident angle threshold, and simultaneously determines whether the current relative height is less than the maximum effective detection height constant. The physical constraint inequalities for the above determination are as follows: ; ; in This represents the critical incident angle threshold, which is obtained through calibration testing before the equipment leaves the factory, based on the physical dielectric constant of the water surface and the electromagnetic wave reflectivity attenuation model. This indicates the drone's current relative altitude. This represents the maximum effective detection altitude constant of the radar equipment.

[0019] When the logic comparator determines that the dynamic incident angle evaluation parameter is less than the critical incident angle threshold and the current relative height is less than the maximum effective detection height constant, it confirms that the current UAV attitude has not caused the reflection cross-section of the sensing beam to completely deviate from the receiving antenna, and the relative height meets the receiver's signal-to-noise ratio (SNR) minimum requirement. In response to this determination, the system marks the distance sensing information of the current frame as valid feature point cloud data. If either condition is not met, the system generates a data discard command, blocking the transmission of the current frame data to subsequent modules to prevent artifact signals from being mixed into the risk analysis stream due to drastic attitude changes.

[0020] After acquiring valid feature point cloud data, the processing flow enters the spatial clustering stage. The system calls the onboard computing power module to perform density-based spatial clustering operations on the valid feature point cloud data. Density clustering, based on the set neighborhood search radius and the minimum inclusion limit of core points, merges the spatially distributed valid feature point clouds into independent clustered connected regions. The system extracts the peripheral three-dimensional extreme points of the clustered connected regions, generating the boundary coordinates of the clustered connected regions. The system directly instantiates the generated boundary coordinates of the clustered connected regions into candidate risk event data packets. The system outputs the memory pointer address of the candidate risk event data packets to the bus, completing the complete data flow loop from underlying perception input to risk target screening.

[0021] S2. Combining candidate risk events, flight status information, local environmental map information, and extracted historical observation features, perform local risk analysis and output risk certainty assessment results; Furthermore, in S2, by combining candidate risk events, flight status information, local environmental map information, and extracted historical observation features, a local risk analysis is performed, outputting a risk certainty assessment result, including: Extract a set of local 3D point clouds within the spatial coordinate range corresponding to the candidate risk event based on local environmental map information; The principal component analysis logic is invoked to extract the normal vectors of feature points within the local 3D point cloud set, the spatial dispersion of the extracted normal vectors is calculated, and the spatial dispersion is used as the surface fluctuation feature value. The current flight speed vector of the UAV is extracted based on the flight status information. The absolute value of the speed is extracted by performing L2 calculation on the current flight speed vector and used as the physical kinetic energy feature value. Extract prior probability parameters of spatiotemporal distribution for the current spatial region from historical observation features; Obtain the preset first weighted number, second weighted number, and third weighted number; The first weighted value is multiplied by the surface wave characteristic value, the second weighted value is multiplied by the physical kinetic energy characteristic value, and the third weighted value is multiplied by the spatiotemporal distribution prior probability parameter. Perform a logical summation on the results of all product operations, and output the summation value as the risk certainty assessment result.

[0022] Specifically, the system executes step S2, initiating the multi-dimensional feature extraction and joint settlement process for multi-source input data. The local risk analysis module receives candidate risk events output from the previous step via the system's underlying data bus. Simultaneously, the local risk analysis module retrieves flight status information, local environment map information, and extracted historical observation features from system memory and the historical database.

[0023] The underlying parsing program reads the three-dimensional spatial coordinate range encapsulated in the candidate risk event data packet. The processing unit uses external local environment map information as a spatial comparison benchmark and performs Boolean intersection operations on the coordinate regions. The processing unit extracts a local three-dimensional point cloud set within the aforementioned spatial coordinate range. This point cloud set maps all discrete electromagnetic reflection feature points within the water surface area to be evaluated.

[0024] The system invokes the hardware matrix operation unit to perform principal component analysis on the local 3D point cloud set. The matrix operation unit constructs the covariance matrix of all feature points within the set and performs eigenvalue decomposition to extract the local normal vectors corresponding to each feature point. The system aggregates all extracted normal vectors and performs statistical spatial dispersion calculation. The calculated spatial dispersion value is directly defined as the surface fluctuation characteristic value by the system's underlying registers. The mathematical expression model of this calculation process is as follows: ; in This represents the surface wave characteristic value extracted and calculated. The larger the value, the stronger the physical interference of multipath scattering generated by the water surface ripples on the detection signal. This represents the total number of feature points contained within a local 3D point cloud set. Indicates the first Local normal vectors of feature points. It represents the spatial average vector of the normal vectors of all feature points within a local 3D point cloud set.

[0025] The system initiates a parallel dynamic state analysis thread within the same clock cycle as the surface fluctuation extraction. The processing unit parses the received flight state information and extracts the current flight velocity vector representing the instantaneous spatial motion state of the UAV. The computing module performs a mathematical norm 2 operation on the current flight velocity vector to extract the absolute velocity value in scalar form, and sets the extracted absolute velocity value as the physical kinetic energy feature value. The mapping equation for this feature extraction dimension is as follows: ; in This represents the set physical kinetic energy characteristic value. This represents the lateral velocity component of the current flight velocity vector in the body coordinate system. This represents the longitudinal velocity component of the current flight velocity vector in the body coordinate system. This represents the vertical velocity component of the current flight velocity vector in the body coordinate system.

[0026] The processing unit executes the historical data retrieval process. Based on the physical geographic coordinates of the current UAV, the system extracts the spatiotemporal distribution prior probability parameters for the current spatial region from the retrieved historical observation features. This parameter records the temporal cumulative amount of water surface features encountered by the UAV within the target coordinate grid in historical flight batches.

[0027] The system constructs a multi-dimensional feature weighted fusion processor. The fusion processor loads preset first, second, and third weighted values ​​from the system's read-only memory. The computation control unit performs a multiplication operation between the first weighted value and the surface fluctuation feature value. The computation control unit performs a multiplication operation between the second weighted value and the physical kinetic energy feature value. The computation control unit performs a multiplication operation between the third weighted value and the spatiotemporal distribution prior probability parameter. The underlying accumulator inputs the results of the above three independent multiplication operations into the logical summation calculation module and outputs a unique summation value. This calculated value is designated by the system as the risk certainty assessment result. Its underlying weighted logic formula is constructed as follows: ; in This represents the risk certainty assessment result output by the fusion calculation. This indicates the preset first weight number. This represents the surface fluctuation feature value extracted in the above steps. This represents the preset second weight number. This represents the physical kinetic energy characteristic value extracted in the above steps. This indicates the pre-defined third weight number. This represents the extracted spatiotemporal distribution prior probability parameters.

[0028] After the calculation is completed, the system will push the packaged risk certainty assessment results to the input port of the subsequent hierarchical instruction generation module through the internal transmission bus, establishing a complete processing data flow from heterogeneous feature extraction, weighted fusion to single quantitative index output.

[0029] like Figure 3 As shown, S3 compares the risk certainty assessment result with the preset grading judgment conditions and generates a grading disposal instruction. If the risk certainty assessment result falls into the low certainty range, a local conservative risk constraint instruction is triggered. If the risk certainty assessment result falls into the high certainty range, a local scope review and update instruction is triggered. Furthermore, in S3, if the risk certainty assessment result falls into the low certainty range, a local conservative risk constraint instruction is triggered, including: Obtain the preset low-risk and high-risk judgment thresholds; Determine whether the risk certainty assessment result is greater than or equal to the low-risk assessment threshold, and simultaneously determine whether the risk certainty assessment result is less than the high-risk assessment threshold; When the risk certainty assessment result is greater than or equal to the low risk assessment threshold and less than the high risk assessment threshold, the risk certainty assessment result falls into the low certainty range. In response to a judgment state that falls into a low-determinism range, a local conservative risk constraint instruction is generated; The underlying execution rules of the local conservative risk constraint command are set to lock the maximum downward vertical flight speed limit of the UAV, while maintaining the current horizontal exploration heading control logic of the UAV unchanged.

[0030] In S3, if the risk certainty assessment result falls into the high certainty range, a local scope review and update instruction is triggered, including: Obtain the preset high-risk judgment threshold; Determine whether the risk certainty assessment result is greater than or equal to the high-risk determination threshold; When the risk certainty assessment result is greater than or equal to the high-risk assessment threshold, the risk certainty assessment result is determined to fall into the high certainty range. In response to a judgment state that falls within a high-certainty range, a local range verification and update instruction is generated; The underlying execution rules for the local scope review and update instructions are set to extract the three-dimensional spatial physical coordinate range corresponding to the candidate risk event, configure the three-dimensional spatial physical coordinate range as a rigid no-fly state, and perform an outward mathematical morphological expansion operation on the three-dimensional spatial physical coordinate range configured as a rigid no-fly state.

[0031] Specifically, the classification instruction generation module receives the risk certainty assessment results pushed out by the local risk analysis module via the system's underlying data bus. The logic comparator within this module acts as the execution entity, responsible for performing a hard-wired numerical comparison between the received risk certainty assessment results and the system's preset classification judgment conditions. The underlying control flow is differentiated into two independent instruction triggering links based on the level state of the numerical comparison.

[0032] The arithmetic unit retrieves the pre-programmed low-risk and high-risk thresholds from the read-only address segment of the system memory chip. The logic comparator determines whether the risk certainty assessment result is greater than or equal to the low-risk threshold. Simultaneously within the same computation cycle, the logic comparator determines whether the risk certainty assessment result is less than the high-risk threshold. The mathematical inequality corresponding to the above logical comparison process is as follows: ; in This indicates the low-risk threshold obtained from the system memory chip. This indicates the risk certainty assessment result based on the preceding data stream input. This indicates the high-risk threshold obtained from the system memory chip.

[0033] When the comparator confirms that the risk certainty assessment result meets the above discrimination inequality conditions, the system's underlying state machine marks the spatial attributes of the current data as falling into the low certainty interval. In response to the judgment state of falling into the low certainty interval, the controller compiles and generates a locally conservative risk constraint instruction in system memory. The processor forcibly overrides the underlying execution rules of this locally conservative risk constraint instruction to the flight control solution node. Specifically, the underlying execution rules restrict the Z-axis downward motion degree of freedom of the control system. The system locks the maximum downward vertical flight speed limit of the UAV in the current control cycle to prevent the aircraft from diving into suspected water areas. The system maintains the UAV's current horizontal exploration heading control logic unchanged, ensuring that the aircraft's exploration actions in the two-dimensional horizontal plane are not interrupted.

[0034] For the other decision link in the control flow, the logic comparator independently determines whether the risk certainty assessment result is greater than or equal to the high-risk determination threshold. The discrimination inequality for this logic branch is constructed as follows: ; in This indicates the risk certainty assessment result of the preceding input. This indicates the preset high-risk threshold.

[0035] When the logic comparator determines that the risk certainty assessment result meets the inequality condition, the system state machine determines that the risk certainty assessment result falls into the high certainty interval. After receiving the state trigger signal indicating that the result has fallen into the high certainty interval, the controller assembles and generates local range verification and update instructions. The computing unit loads the underlying execution rules of the local range verification and update instructions and extracts the three-dimensional spatial physical coordinate range corresponding to the candidate risk events based on the address mapping table.

[0036] The memory management unit configures the block attributes corresponding to the 3D physical coordinate range to a rigid no-fly state. The computing hardware performs an outward mathematical morphological expansion operation on the 3D physical coordinate range configured as rigidly no-fly, expanding the original observation risk boundary outward along the normal direction. The underlying mathematical formula for the mathematical morphological expansion operation is as follows: ; in This represents the expanded three-dimensional space coordinate set output after performing a mathematical morphological dilation operation. This represents the initial three-dimensional physical coordinate range of the area that is configured to be in a rigid no-fly state. This represents the Minkowski addition operator in the field of mathematical morphology operations. This represents the kernel parameters of the three-dimensional structure element stored in the register.

[0037] After the instructions are compiled, the hierarchical instruction generation module encapsulates the local conservative risk constraint instructions or local scope review and update instructions into data packets. The bus driver pushes the data packets containing these instructions into the system transmission queue, outputting them as the final hierarchical disposal instructions to the subsequent risk area conversion module. The instruction generation and distribution process thus forms a physical closed loop.

[0038] like Figure 3 As shown, S4 extracts the processing results corresponding to the local conservative risk constraint instruction or the local scope review and update instruction, and transforms the processing results into a unified risk area; Furthermore, in S4, the processing results corresponding to local conservative risk constraint instructions or local scope review and update instructions are extracted, and the processing results are transformed into a unified risk area, including: Monitor the currently active command trigger type; When the local conservative risk constraint command is detected to be active, a one-dimensional temporary restricted area with a height lower limit is generated within the current feedforward planning field of view of the UAV, and the one-dimensional temporary restricted area is set as a unified risk area. When the local area review and update command is detected to be active, the basic 3D bounding box corresponding to the candidate risk event is extracted, and all feature vertices on the surface of the basic 3D bounding box are extracted. Calculate the direction of the external normal vector corresponding to all feature vertices; All feature vertices are translated outward along the direction of the external normal vector. The boundary expansion distance parameter generated by parameter optimization is used to construct a set of polygon bounding boxes after translation and reshaping. The set of polygon bounding boxes is then set as a unified risk region.

[0039] Specifically, the risk area conversion module receives graded processing instructions pushed into the transmission queue by the front-end graded instruction generation module via the system's high-speed data bus. The module's built-in status register monitors the currently active instruction trigger type in real time. The underlying control logic triggers the corresponding hardware interrupt based on the flag bit read from the status register, directing the data flow to two independent physical space mapping links.

[0040] When the status register detects that the local conservative risk constraint instruction is active, the spatial computing unit extracts the current feedforward planning field of view of the UAV. Within this current feedforward planning field of view, the geometry rendering program generates a one-dimensional temporary restricted area patch with a lower height limit indicator. This process restricts the motion constraints in three-dimensional space to an absolute height truncation along a single vertical axis. The mathematical model of the generated one-dimensional temporary restricted area patch is expressed as follows: ; in This represents the set of spatial coordinates of the generated one-dimensional temporary restricted area. It represents any coordinate point in three-dimensional physical space. This represents the projection vector of the coordinate point onto the horizontal two-dimensional plane. This represents the set of horizontal boundaries of the current feedforward planning field of view. This represents the vertical height component of the coordinate point. This indicates the coordinates of the rigid safety baseline threshold with a height lower limit indicator.

[0041] The risk region transformation module sets the generated one-dimensional temporary restricted area as a unified risk region and locks its physical memory address to prevent external processes from overwriting it.

[0042] On the other side of the instruction mapping chain, when the status register detects that a local range review and update instruction is active, the memory read pointer extracts the underlying 3D bounding box bound to the candidate risk event. The vertex parsing logic traverses the mesh topology of the underlying 3D bounding box and extracts all feature vertices on the surface of the underlying 3D bounding box.

[0043] The vector solver calculates the direction of the external normal vector for all feature vertices based on the normals of the spatial triangular facets formed by adjacent feature vertices. The coordinate transformation unit reads the boundary expansion distance parameter output by the independent optimization module after performing extremum solving. The processor allocates floating-point computing resources to translate all feature vertices outward along the calculated direction of the external normal vector to the boundary expansion distance parameter. The underlying affine transformation formula for the expansion translation reshaping process is constructed as follows: ; in Indicates the number after outward expansion and translation. The three-dimensional coordinate vectors of each feature vertex in a unified world coordinate system. This represents the first element extracted from the surface of the base 3D bounding box before translation. The initial three-dimensional coordinate vector of each feature vertex. The parameter represents the boundary extension distance generated by parameter optimization, and is a one-dimensional scalar constant. The solution obtained is the first... The directions of the external normal vectors corresponding to each feature vertex are preprocessed into unit direction vectors with a magnitude of one.

[0044] The mesh reconstruction algorithm reconnects all feature vertices that have undergone translation transformation according to the initial index relationship, constructing a set of polygon bounding boxes after translation and reshaping. The system sets the generated set of polygon bounding boxes as a unified risk region.

[0045] The risk area transformation module performs data serialization on the defined unified risk areas. The data distribution interface pushes the serialized unified risk area data packets to the collaborative planning constraint module. The data flow transmission node triggered by the instruction action to reshape the physical boundary completes the logical loop.

[0046] like Figure 4 As shown, S5 applies a unified risk area simultaneously to the UAV's target selection process and trajectory avoidance planning process, in order to suppress high-risk targets during the target selection process and generate avoidance trajectories during the trajectory avoidance planning process. Furthermore, in S5, high-risk target suppression is implemented during the target selection process, including: Extract the set of candidate target points generated by the UAV within the current control cycle; For any candidate target point within the candidate target point set, extract the unknown space information gain value and flight distance penalty value corresponding to the exploration action; The difference between the unknown space information gain value and the flight distance penalty value is calculated to output the initial exploration gain value. Read the 3D spatial projection coordinates of any candidate target point; Determine whether the three-dimensional spatial projection coordinates fall within the boundary of the unified risk area; When it is determined that the 3D spatial projection coordinates fall inside the boundary of a unified risk area, a Boolean truncation mask configured as a logical false value is generated. Perform a logical AND operation between the initial exploration gain value and the Boolean truncation mask, and output the final exploration gain that blocks the abandoned state to limit the allocation of exploration scheduling resources to arbitrary candidate target points.

[0047] In S5, the avoidance trajectory is generated during the trajectory avoidance planning process, including: Extract the discrete trajectory point sequence from the bottom layer and traverse any trajectory control point within the discrete trajectory point sequence; Calculate the nearest Euclidean distance from any trajectory control point to the external physical boundary of the unified risk area; Extract the preset water surface safety warning distance constant; Compare the magnitudes of the nearest Euclidean distance and the water surface safety warning distance constant; When the nearest Euclidean distance is determined to be less than the water surface safety warning distance constant, the spatial gradient direction vector corresponding to the nearest Euclidean distance is calculated. A repulsive potential penalty term is generated by combining the spatial gradient direction vector; The repulsive potential field penalty term is used as an additional cost input parameter and is incorporated into the underlying trajectory optimization cost function. Nonlinear iterative solution calculation is performed, and the set of heading points that deviate from the unified risk region is output as the avoidance trajectory.

[0048] Specifically, the collaborative planning constraint module receives unified risk area data packaged and distributed by the preceding process through its internal high-speed cache. The system task allocation scheduler generates an initial set of candidate target points within the current control cycle and pushes this set into the memory address space of the exploration target selection process.

[0049] The target selection process initiates a traversal command, sequentially reading any candidate target point within the candidate target point set. The data evaluation node extracts the unknown space information gain value corresponding to the UAV's exploration action towards that point, and simultaneously extracts the flight distance penalty cost required for the device to reach that coordinate. Here, the unknown space information gain value is the business characteristic state value defined in this embodiment, and its physical meaning refers to the expected three-dimensional volume of the unknown physical environment area that the system expects to newly explore after the UAV's onboard sensors cover the candidate target point. The flight distance penalty cost value is the kinematic cost parameter set by the system, and its physical meaning refers to the combined cost of the estimated absolute length of the spatial path consumed by the UAV flying from the current spatial coordinates to the candidate target point and the power consumption. The arithmetic logic unit performs a scalar subtraction operation on the unknown space information gain value and the flight distance penalty cost value, and outputs the initial exploration gain value. The physical model for this basic gain settlement is expressed as follows: ; in This represents the initial exploration reward value output by the arithmetic logic unit. This represents the gain value of the extracted unknown spatial information. This represents the penalty value for the extracted flight distance.

[0050] The spatial verification component reads the 3D spatial projection coordinates of any candidate target point currently being traversed. The collision detection engine executes a boundary cross-test algorithm to determine whether the 3D spatial projection coordinates fall within the boundary of a unified risk area. When the collision detection engine determines that the 3D spatial projection coordinates clearly fall within the boundary of a unified risk area, the underlying control register immediately generates a Boolean truncation mask whose state is forcibly configured as a logical false value. The processor instruction set performs a hardware-level logical AND operation on the initial exploration gain value and the Boolean truncation mask. The system output is reduced to the final exploration gain that blocks the abandoned state. The scheduler uses this final exploration gain to limit the system to allocate underlying flight exploration scheduling resources to the currently dangerous candidate target point, completing the direct physical stripping of high-risk wading targets at the macro-mission planning node.

[0051] In the underlying kinematics calculation phase, the trajectory avoidance planning process extracts the underlying discrete trajectory point sequence from the navigation controller's feedforward queue. The system's calculation thread then traverses any trajectory control point within this discrete trajectory point sequence.

[0052] The distance calculation unit invokes a spatial proximity search algorithm to calculate the nearest Euclidean distance from any trajectory control point currently traversed to the external physical boundary of the unified risk area. The system extracts a preset water surface safety warning distance constant from the flight control safety configuration file. The hardware comparator performs a numerical determination on the relationship between the nearest Euclidean distance and the water surface safety warning distance constant.

[0053] When the hardware comparator determines that the nearest Euclidean distance is less than the water surface safety warning distance constant, the calculus kernel calculates the spatial gradient direction vector corresponding to this nearest Euclidean distance using three-dimensional differentiation. The system then combines the direction of the spatial gradient direction vector to generate a repulsive potential field penalty term with physical dimensions. The specific repulsive potential field penalty mapping formula is established as follows: ; in This represents the parameter of the repulsive potential field penalty term generated by the calculus kernel. This represents the repulsive gain constant coefficient built into the memory. This represents the calculated nearest Euclidean distance. This represents the extracted water surface safety warning distance constant. This represents the spatial gradient direction vector output by the derivative calculation.

[0054] The trajectory optimization solution module uses the repulsive potential field penalty term as an additional cost input parameter, directly integrating it into the underlying trajectory optimization cost function. The numerical computation engine calls a sequential quadratic programming algorithm to perform nonlinear iterative calculations on the reconstructed target cost function. After the iterative gradient converges, the system outputs a set of heading points that deviate from the unified risk region in the three-dimensional physical space. The bus controller uses this safe set of heading points as the final avoidance trajectory and pushes it to the next-level physical action control interface.

[0055] like Figure 4 As shown, S6, based on the avoidance trajectory output mode switching information and flight execution results, forms an active avoidance control closed loop; Furthermore, in S6, based on the avoidance trajectory output mode switching information and flight execution results, an active avoidance control closed loop is formed, including: Read the feedforward yaw angle deviation data and feedforward pitch angle deviation data corresponding to the avoidance trajectory; Based on the feedforward yaw angle deviation data mapping, pulse width modulation duty cycle command parameters are generated for the yaw servo. Based on the feedforward elevation angle deviation data mapping, command parameters are added to the base speed of the power motor to increase the speed. The pulse width modulation duty cycle command parameters and the basic speed increase command parameters are packaged and combined to generate flight execution results; Configure the status flag of the mode switching information to actively avoid forced takeover mode; The status flag of the active avoidance of forced takeover mode and the flight execution result are transmitted synchronously to the underlying flight control execution hardware unit to trigger the physical servo and power motor drive.

[0056] Specifically, the avoidance execution output module extracts the avoidance trajectory, which is sent to the underlying cache by the collaborative planning constraint module, from the system's internal data sharing queue. The controller parses the spatial heading point sequence within the avoidance trajectory and inputs it, along with the current actual flight attitude data output by the airborne inertial navigation system, into the underlying difference comparator. The difference comparator performs a subtraction operation, and the system reads the feedforward yaw angle deviation data and feedforward pitch angle deviation data corresponding to the avoidance trajectory based on the calculation result.

[0057] The underlying hardware abstraction layer retrieves the pre-configured yaw response function from the control memory. The computing unit performs linear interpolation and amplification operations based on the feedforward yaw angle deviation data, mapping and generating pulse width modulation duty cycle command parameters for the yaw servo. The mathematical mapping model for this hardware pulse control is established as follows: ; in This represents the pulse width modulation duty cycle command parameter for the yaw servo output. This represents the baseline duty cycle constant used by the yaw servo of the drone to maintain a mechanically neutral position. This represents the proportional gain coefficient for yaw angle control, which is embedded in the flight control system. This indicates the feedforward yaw angle deviation data read by the difference comparator.

[0058] The dynamic calculation node extracts feedforward elevation angle deviation data within the same control clock cycle. The processor calls the dynamic response physical model and generates command parameters for increasing the base speed of the motor based on the feedforward elevation angle deviation data. The dynamic compensation mapping equation is constructed as follows: ; in This indicates that the generated command parameters are increased based on the base speed of the motor. This represents the force response conversion coefficient that relates the physical elevation angle to the motor output speed. This indicates the read feedforward elevation angle deviation data. This represents the physical compensation constant used to overcome the weight of the UAV and the attenuation of the downwash airflow on the wading surface.

[0059] The main control chip's memory management unit implements a completely new communication data frame structure. The program logic performs bit-segment concatenation and cyclic redundancy check (CRC) code appending operations, combining the pulse width modulation duty cycle command parameters and the basic rotational speed increase command parameters into a data package. After the data frame is encapsulated, a unique flight execution result is generated.

[0060] The system task scheduler sends a high-priority hardware interrupt request to the top-level state machine. Upon receiving this request, the state machine forcibly intervenes in the system's global state register, overwriting the status flag of the mode switching information to actively avoid forced takeover mode. This write operation completely severs the scheduling channel for flight control authority of the regular exploration mission threads at the system's underlying level.

[0061] The data distribution gateway extracts the status flag bit written to the active avoidance forced takeover mode and binds it as a frame header instruction with the aforementioned generated flight execution result. The communication interface module synchronously transmits the above combined data stream to the underlying flight control execution hardware unit via the airborne controller local area network bus. The digital-to-analog converter of the flight control execution hardware unit interprets the digital signal into physical drive levels, directly triggering the mechanical deflection of the physical servo and the speed jump drive of the power motor. The UAV system thus completely completes the end-to-end active avoidance control closed loop from the front-end water surface risk perception input, multi-level data processing flow to the end-end physical avoidance action output.

[0062] Example 2: In applications where UAVs perform low-altitude autonomous search and rescue and mapping exploration missions in unknown wetland environments such as mountainous reservoirs or flood-stricken areas, the system faces highly specific water-related safety challenges. Due to the specular reflection and low texture of the water surface, the water surface echoes acquired by airborne sensing equipment are extremely unstable, resulting in strong flickering and discontinuity in the temporal sequence of observation data. This scenario presents a serious technical problem: traditional loosely coupled obstacle avoidance systems cannot cope with this low-determinism risk. On the one hand, single-frame missed detections in the lower-level sensing layer can easily lead to misjudgments of safety below, inducing the UAV to dive towards the water surface and crash. On the other hand, occasional false alarms from water wave reflections can trigger rigid global avoidance, causing frequent sudden braking or heading lock-up. The system lacks a dynamic, graded handling mechanism for water-related risks and a unified, shared physical collision avoidance boundary, ultimately leading to the stagnation of exploration missions and an inability to balance absolute flight safety and autonomous detection efficiency over complex water areas. To solve these problems, this invention provides a UAV water surface risk perception and active collaborative planning system, such as... Figure 5 As shown. The specific implementation process of this system is as follows: The provided UAV surface risk perception and active collaborative planning system is deployed within the airborne computing motherboard. The front-end perception acquisition and filtering module acquires distance perception information, flight status information, and local environmental map information via a bus. Based on spatial geometric constraints, it eliminates invalid signals and pushes candidate risk events into the lower-level queue. The local risk analysis module extracts historical observation features, performs multi-dimensional feature fusion calculations with candidate risk events, and quantitatively outputs the risk certainty assessment results. The hierarchical instruction generation module intercepts this result based on hard-line comparison logic, generating local conservative risk constraint instructions in low-certainty intervals and local range verification and update instructions in high-certainty intervals. The risk region transformation module parses the instruction data structure and reshapes a unified risk region with clearly defined physical boundaries. The collaborative planning constraint module maps this risk region to the navigation calculation node, simultaneously executing a high-risk target mask suppression algorithm and collision avoidance trajectory generation operations. The avoidance execution output module reads the calculated avoidance trajectory, sends mode switching information and entity flight execution results to the hardware interface, directly driving the underlying servo motor response, completing the end-to-end active obstacle avoidance hardware closed loop.

[0063] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for unmanned aerial vehicle (UAV) surface risk perception and proactive collaborative planning, characterized in that, Includes the following steps: S1. Acquire the distance perception information, flight status information and local environment map information of the UAV, perform validity screening on the distance perception information, and generate candidate risk events when the screening is deemed valid; S2. Combining the candidate risk events, the flight status information, the local environment map information, and the extracted historical observation features, perform local risk analysis and output the risk certainty assessment result; S3. Compare the risk certainty assessment result with the preset classification judgment conditions to generate a classification handling instruction. If the risk certainty assessment result falls into the low certainty range, a local conservative risk constraint instruction is triggered. If the risk certainty assessment result falls into the high certainty range, a local scope review and update instruction is triggered. S4. Extract the processing results corresponding to the local conservative risk constraint instruction or the local range review and update instruction, and convert the processing results into a unified risk area; S5. Simultaneously apply the unified risk area to the UAV's target selection process and trajectory avoidance planning process, so as to perform high-risk target suppression in the target selection process and generate an avoidance trajectory in the trajectory avoidance planning process; S6. Based on the avoidance trajectory output mode switching information and flight execution results, an active avoidance control closed loop is formed.

2. The UAV surface risk perception and active collaborative planning method according to claim 1, characterized in that, In S1, the step of performing validity screening on the distance-aware information and generating candidate risk events when the screening is deemed valid includes: Extract the sensor vertical beam angle corresponding to the distance perception information, and extract the current pitch angle and current relative altitude corresponding to the flight status information; The sensor's vertical beam angle and the current pitch angle are summed to output a dynamic incident angle evaluation parameter. Extract the preset critical incident angle threshold and the maximum effective detection height constant; Determine whether the dynamic incident angle evaluation parameter is less than the critical incident angle threshold, and simultaneously determine whether the current relative height is less than the maximum effective detection height constant; When it is determined that the dynamic incident angle evaluation parameter is less than the critical incident angle threshold and the current relative height is less than the maximum effective detection height constant, the distance perception information is marked as effective feature point cloud data; Perform density-based spatial clustering on the effective feature point cloud data, extract the boundary coordinates of the clustered connected regions, and use the boundary coordinates of the clustered connected regions as the candidate risk events.

3. The UAV surface risk perception and active cooperative planning method according to claim 1, characterized in that, In S2, the process of combining the candidate risk events, the flight status information, the local environment map information, and the extracted historical observation features to perform local risk analysis and output a risk certainty assessment result includes: Based on the local environment map information, extract a set of local 3D point clouds within the spatial coordinate range corresponding to the candidate risk event; The principal component analysis logic is invoked to extract the normal vectors of feature points inside the local 3D point cloud set, the spatial dispersion of the extracted normal vectors is calculated, and the spatial dispersion is used as the surface fluctuation feature value. Based on the flight status information, the current flight speed vector of the UAV is extracted, the L2 norm is calculated on the current flight speed vector to extract the absolute value of the speed, and the absolute value of the speed is used as the physical kinetic energy feature value; Extract the spatiotemporal distribution prior probability parameters for the current spatial region from the historical observation features; Obtain the preset first weighted number, second weighted number, and third weighted number; The first weighted value is multiplied by the surface wave characteristic value, the second weighted value is multiplied by the physical kinetic energy characteristic value, and the third weighted value is multiplied by the spatiotemporal distribution prior probability parameter. Perform a logical summation on the results of all product operations, and output the summation value as the risk certainty assessment result.

4. The UAV surface risk perception and active cooperative planning method according to claim 1, characterized in that, In S3, the step of triggering a local conservative risk constraint instruction if the risk certainty assessment result falls into the low certainty range includes: Obtain the preset low-risk and high-risk judgment thresholds; Determine whether the risk certainty assessment result is greater than or equal to the low-risk determination threshold, and simultaneously determine whether the risk certainty assessment result is less than the high-risk determination threshold; When the risk certainty assessment result is determined to be greater than or equal to the low-risk determination threshold and less than the high-risk determination threshold, the risk certainty assessment result is determined to fall into the low certainty interval. In response to the determination state falling into the low-determinism range, the local conservative risk constraint instruction is generated; The underlying execution rule of the local conservative risk constraint instruction is set to lock the maximum downward vertical flight speed limit of the UAV and maintain the current horizontal exploration heading control logic of the UAV unchanged.

5. The UAV surface risk perception and active collaborative planning method according to claim 1, characterized in that, In S3, if the risk certainty assessment result falls into the high certainty range, triggering a local scope review and update instruction includes: Obtain the preset high-risk judgment threshold; Determine whether the risk certainty assessment result is greater than or equal to the high-risk determination threshold; When the risk certainty assessment result is determined to be greater than or equal to the high-risk determination threshold, the risk certainty assessment result is determined to fall into the high certainty interval. In response to the determination state that falls within the high certainty interval, the local range verification and update instruction is generated; The underlying execution rule of the local range review and update instruction is set to extract the three-dimensional spatial physical coordinate range corresponding to the candidate risk event, configure the three-dimensional spatial physical coordinate range as a rigid no-fly state, and perform an outward mathematical morphological expansion operation on the three-dimensional spatial physical coordinate range configured as a rigid no-fly state.

6. The UAV surface risk perception and active cooperative planning method according to claim 1, characterized in that, In S4, the step of extracting the processing result corresponding to the local conservative risk constraint instruction or the local range review and update instruction, and converting the processing result into a unified risk region, includes: Monitor the currently active command trigger type; When the local conservative risk constraint command is detected to be active, a one-dimensional temporary restricted area with a height lower limit is generated within the current feedforward planning field of view of the UAV, and the one-dimensional temporary restricted area is set as the unified risk area. When the local range verification and update instruction is detected to be active, the basic 3D bounding box corresponding to the candidate risk event is extracted, and all feature vertices on the surface of the basic 3D bounding box are extracted. Calculate the direction of the external normal vector corresponding to all the feature vertices; All feature vertices are translated outward along the direction of the external normal vector, and the boundary expansion distance parameter generated by parameter optimization is used to construct a set of polygon bounding boxes after translation and reshaping. The set of polygon bounding boxes is then set as the unified risk region.

7. The UAV surface risk perception and active cooperative planning method according to claim 1, characterized in that, In S5, the high-risk target suppression during the target selection process includes: Extract the set of candidate target points generated by the UAV during the current control cycle; For any candidate target point within the set of candidate target points, extract the unknown space information gain value and flight distance penalty value corresponding to the exploration action; The difference between the unknown space information gain value and the flight distance penalty value is calculated to output the initial exploration gain value. Read the three-dimensional spatial projection coordinates of any candidate target point; Determine whether the three-dimensional spatial projection coordinates fall within the boundary of the unified risk area; When it is determined that the three-dimensional spatial projection coordinates fall inside the boundary of the unified risk area, a Boolean truncation mask configured as a logical false value is generated. Perform a logical AND operation between the initial exploration gain value and the Boolean truncation mask, and output the final exploration gain that blocks the abandoned state to limit the allocation of exploration scheduling resources to the arbitrary candidate target point.

8. The UAV surface risk perception and active cooperative planning method according to claim 1, characterized in that, In S5, generating the avoidance trajectory during the trajectory avoidance planning process includes: Extract the discrete trajectory point sequence from the bottom layer, and traverse any trajectory control point within the discrete trajectory point sequence; Calculate the nearest Euclidean distance from the arbitrary trajectory control point to the external physical boundary of the unified risk area; Extract the preset water surface safety warning distance constant; Compare the magnitudes of the nearest Euclidean distance and the water surface safety warning distance constant; When it is determined that the nearest Euclidean distance is less than the water surface safety warning distance constant, the spatial gradient direction vector corresponding to the nearest Euclidean distance is calculated; The repulsive potential penalty term is generated by combining the spatial gradient direction vector; The repulsive potential field penalty term is used as an additional cost input parameter and fed into the underlying trajectory optimization cost function. Nonlinear iterative solution calculation is performed, and the set of heading points that deviate from the unified risk region is output as the avoidance trajectory.

9. The UAV surface risk perception and active cooperative planning method according to claim 1, characterized in that, In S6, the process of forming an active avoidance control closed loop based on the avoidance trajectory output mode switching information and flight execution results includes: Read the feedforward yaw angle deviation data and feedforward pitch angle deviation data corresponding to the avoidance trajectory; Based on the aforementioned feedforward yaw angle deviation data mapping, pulse width modulation duty cycle command parameters for the yaw servo are generated. Based on the aforementioned feedforward elevation angle deviation data, a command parameter for increasing the base speed of the power motor is generated. The pulse width modulation duty cycle command parameters and the basic speed increase command parameters are packaged and combined to generate the flight execution result. Configure the status flag of the mode switching information to actively avoid forced takeover mode; The status flag of the active avoidance forced takeover mode and the flight execution result are synchronously transmitted to the underlying flight control execution hardware unit to trigger the physical servo and the power motor drive.

10. A UAV surface risk perception and active collaborative planning system, characterized in that, The method for UAV surface risk perception and active cooperative planning according to any one of claims 1-9 includes the following modules: The perception acquisition and filtering module is used to acquire the distance perception information, flight status information and local environment map information of the UAV, perform validity filtering on the distance perception information, and generate candidate risk events when the filtering is deemed valid. The local risk analysis module is used to combine the candidate risk events, the flight status information, the local environment map information, and the extracted historical observation features to perform local risk analysis and output the risk certainty assessment result. The graded instruction generation module is used to compare the risk certainty assessment results with preset graded judgment conditions and generate graded disposal instructions. The risk area conversion module is used to extract the processing results corresponding to the graded disposal instructions and convert the processing results into a unified risk area; The collaborative planning constraint module is used to simultaneously apply the unified risk area to the UAV's target selection process and trajectory avoidance planning process; The avoidance execution output module is used to form an active avoidance control closed loop based on the avoidance trajectory output mode switching information and flight execution results generated during the trajectory avoidance planning process.

Images