Low-altitude aircraft autonomous operation obstacle avoidance method and system

Abstract

The invention provides an autonomous operation obstacle avoidance method and system for a low-altitude aircraft, and the method is executed by the low-altitude aircraft, and comprises the following steps: S1, data acquisition: collecting data through a multi-source heterogeneous sensor group carried by the low-altitude aircraft, airspace control information, real-time meteorological data and collaborative sensing data of other aircrafts sent by the ground sensing base station are received through the air-ground communication interface; preprocessing the collected original data, wherein the preprocessing comprises data format standardization, abnormal value elimination and time synchronization; and S2, data fusion processing: carrying out credibility evaluation on the preprocessed multi-source heterogeneous sensing data and collaborative sensing data. According to the invention, through three-dimensional credibility evaluation and an air-ground data compensation mechanism, the perception precision is greatly optimized, the perception data efficiency is improved in complex environments such as low illumination, strong electromagnetic interference, rainstorm and the like, and obstacle avoidance decision errors caused by data distortion of a single sensor are avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of aircraft operation, in particular to a low-altitude aircraft autonomous operation obstacle avoidance method and system. BACKGROUND

[0002] With the rapid development of low-altitude economy, unmanned aerial vehicles, electric vertical take-off and landing aircraft and other low-altitude aircraft are increasingly widely used in logistics distribution, emergency rescue, urban commuting and other scenarios. The low-altitude operation environment is complex and changeable, containing static obstacles (such as buildings, towers) and dynamic targets (such as other aircraft, birds), and is dynamically affected by factors such as weather conditions and airspace density, which puts high requirements on the real-time, reliability and safety of aircraft autonomous obstacle avoidance.

[0003] The existing low-altitude aircraft obstacle avoidance system mostly uses traditional Kalman filtering, Bayesian estimation and other algorithms to realize multi-sensor data fusion, and its core defect is that it does not distinguish between data superposition: it does not consider the environmental adaptability difference of different sensors (such as the error of photoelectric sensors increases under low light conditions, and the data distortion of radar sensors in electromagnetic interference environment), the cumulative effect of historical measurement errors, and the compensation effect of ground coordination data. For example, in heavy rain, the photoelectric sensor ranging error of a certain unmanned aerial vehicle increases from ±0.5m to ±3m, while the traditional fusion scheme still gives it the same weight as in sunny weather, resulting in a deviation of 2.8m in the fused data, which eventually leads to a collision with the building.

[0004] At the same time, the existing obstacle avoidance rule base is mostly a fixed rule set, and the rule priority is fixed and cannot be dynamically adjusted according to the running state. For example, the basic rule of the right aircraft avoiding the left aircraft when meeting horizontally is reasonable when the airspace density is very low (<1 aircraft per square kilometer), but in high-density airspace (>8 aircraft per square kilometer), the right aircraft may collide with the rear aircraft if it forces to avoid; for example, the scenario-based rule of increasing the avoidance distance in bad weather does not quantify the correlation between weather disturbance intensity and avoidance distance, resulting in insufficient rule adaptation accuracy. Secondly, the existing multi-agent cooperative obstacle avoidance mostly uses distributed game strategy, and each aircraft makes independent decision and informs the intention through communication, lacking global consistency constraint. For example, when two unmanned aerial vehicles meet, they both choose to avoid to the right, resulting in trajectory intersection and causing secondary conflict; in high-density airspace, the avoidance actions of multiple aircraft interfere with each other, forming an obstacle avoidance chain reaction, and finally leading to chaotic overall operation order. SUMMARY

[0005] The technical problem to be solved by the present application is to overcome the defects in the prior art, and the present application provides a low-altitude aircraft autonomous operation obstacle avoidance method and system.

[0006] To solve the above-mentioned existing low-altitude aircraft obstacle avoidance system is to adopt traditional Kalman filter, Bayesian estimation algorithm to realize multi-sensor data fusion, its core defect is that the data is superimposed without distinction: without considering the environmental adaptability of different sensors, the cumulative effect of historical measurement error, and the compensation effect of ground collaborative data.

[0007] At the same time, the existing obstacle avoidance rule base is mostly a fixed rule set, and the rule priority is fixed, which cannot be dynamically adjusted according to the running state. Secondly, the existing multi-agent collaborative obstacle avoidance mostly adopts distributed game strategy, each aircraft makes independent decision and then informs the intention through communication, lacking of global consistency constraint. The technical scheme adopted by the present application is: A low-altitude aircraft autonomous operation obstacle avoidance method and system, the method is executed by a low-altitude aircraft, comprising the following steps: S1, data acquisition: collecting data through a multi-source heterogeneous sensor group carried by itself, receiving airspace control information, real-time weather data and collaborative perception data of other aircrafts sent by a ground perception base station through an air-ground communication interface; preprocessing the collected raw data, including data format standardization, outlier rejection and time synchronization; S2, data fusion processing: performing credibility evaluation on the preprocessed multi-source heterogeneous perception data and collaborative perception data, calculating the perception credibility weight The credibility evaluation includes inherent weight coefficient assignment, error correction coefficient calculation and environment adaptation coefficient; When the credibility weight , The preset credibility threshold reduces the weight proportion of the data in weighted fusion, triggers the collaborative perception data compensation of the ground perception base station; all data are fused by a credibility weighted fusion model to generate unified perception state information; S3, rule adaptation processing: calling a preset dynamic avoidance rule base, the dynamic avoidance rule base is stored in a Redis distributed database; extracting rule applicability calculation parameters based on the unified perception state information; The rule applicability function Calculates the applicability of each avoidance rule; the preset applicability threshold Then execute the weight reduction processing; based on the priority reference value of the remaining rules and Final priority: , From high to low, the first 5 rules form an avoidance rule subset in the current scene; S4, safety interval calculation: according to the avoidance rule subset, target type and its own performance parameters, extracting the basic safety interval From a preset safety interval library; calculating the dynamic ; calculating the constraint condition parameters: minimum reaction interval T is the response time of the flight control system ; Minimum maneuvering safety interval Mass; through the constraint strip Differential safety interval Final safety interval is Updated once every 100ms; S5, collaborative decision processing: based on unified perception state information to detect potential operation conflict, conflict detection uses distance determination method: calculate the shortest straight line of itself and static obstacles The relative distance of the target In conflict; If there is a potential conflict, send the avoidance intention information of itself to other aircraft through the air-ground communication interface, and receive the avoidance intention information of other aircraft; determine the avoidance priority; build a multi-agent collaborative avoidance decision model with global consistency constraint; determine the avoidance action combination without secondary conflict and the leading avoidance aircraft and the subordinate avoidance aircraft, the avoidance action combination includes one or more of heading adjustment, height adjustment and speed adjustment; S6, execution and feedback: convert the avoidance action combination into flight control quantity; send the flight control quantity to the flight control system through the CAN bus to execute real-time collection of avoidance action execution state data and updated perception state information; Calculate the execution error When the error exceeds the preset threshold Difference correction; determine whether the current running state meets the safety interval threshold Then determine that the conflict is resolved, and the original flight trajectory is restored; if not, return to the rule adaptation processing step to the collaborative decision processing step to adjust the avoidance decision again until the conflict is resolved.

[0008] Preferably, the multi-source heterogeneous sensor group in step S1 comprises at least radar sensors, photoelectric sensors, positioning sensors, inertial measurement units and energy sensors, and the collected data includes self-running state data, surrounding static obstacle data and dynamic target original perception data.

[0009] Preferably, the inherent weight coefficient assignment, error correction coefficient calculation and environment adaptation coefficient calculation in step S2 are specifically as follows: , wherein: The inherent weight coefficient of the radar sensor is based on the type of sensor and the importance of data, and the number of radar sensors Bit sensor data Sensor Face collaborative data ; Error correction coefficient of the kth measurement, calculated based on historical measurement error statistics, formula is , which is the maximum allowable error of the kth measurement based on historical measurement error statistics ; Coefficient, calculated based on current environmental parameters, radar sensor data sensor data , positioning sensor unit , energy sensor ; The reliable weighted fusion model , which is the unified perception state information after fusion, and the data update frequency is 50 Hz.

[0010] Preferably, the dynamic avoidance rule library in S3 includes basic avoidance rules, scenario-based avoidance rules, and priority adaptation rules, and the rules are stored in a structured format of condition-action-priority benchmark value; The rule applicability calculation parameters include airspace operation density , weather disturbance strength , self-performance , and historical conflict frequency , and the extraction and normalization method of the rule applicability calculation parameters is as follows: Airspace operation density : calculated based on airspace control information sent by ground perception base station and dynamic target number perceived by itself, , normalized public ; Weather disturbance : calculated based on wind , , which ; Self-performance calculated, which is the maximum allowable error of the kth measurement ; Historical conflict frequency t: statistics of potential normalized ; The rule applicability function adopts weighted summation .

[0011] Preferably, the safety interval library in S4 is stored based on aircraft type and target type classification, specifically: UAV: static ; eVTOL: Static Obstacles ; Reference Airspace operation density Regression model Wherein: Is a preset correction coefficient; The obstacle avoidance priority described in step S5 is based on the mission type, aircraft mass, remaining : Mission type priority Emergency mission = 1.0, ordinary mission = 0.7, test flight mission = 0.5; Mass priority ; Energy priority 0]; The multi-agent collaborative obstacle avoidance decision-making model adopts a process of trajectory prediction - conflict detection - replanning: Trajectory prediction: Use the Extended Kalman Filter (EKF) algorithm to predict the State equation Measurement equation In State vector, State transition Is the observation matrix Gaussian white noise; Consistency constraint detection: Detect the conflict of the obstacle avoidance trajectory through the consistency constraint function , the Function; Trajectory replanning Mark the aircraft with a lower priority as the object to be replanned. Based on its remaining energy, maneuverability, and airspace idle resources, use the Rapidly-exploring Random Tree (RRT*) algorithm to replan the trajectory. The planning goal is to maximize the ] distance between the trajectory and the trajectories of other aircraft.

[0012] Preferably, the specific conversion relationship between the obstacle avoidance action combination described in step S6 and the flight control control quantity is: Heading adjustment action → Yaw angle instruction , Altitude adjustment action → Pitch angle , Speed adjustment action → Fuel .

[0013] A low-altitude aircraft autonomous operation obstacle avoidance system, the system includes: Data acquisition module: Configured to collect its own operation state data, surrounding obstacles, and dynamic target data through a multi-source heterogeneous sensor group, receive data from the ground perception base station and other aircraft through the air-ground communication interface, and preprocess the data; Trusted Fusion Module: Configured to assess the trustworthiness of preprocessed multi-source data and calculate perceived trustworthiness weights. The data compensation mechanism generates unified perception state information through a trusted weighted fusion model; Rule management module: Configured to store a dynamic avoidance rule library, extract rule applicability calculation parameters, calculate the applicability of each rule, dynamically adjust rule priority, and form a subset of avoidance rules; Safety Interval Module: Configured to extract the basic safety interval, calculate the dynamic correction amount and constraint parameters, and determine the differentiated safety interval threshold S through multiple constraints; Collaborative Decision Module: Configured to detect potential operational conflicts, interact with obstacle avoidance intent information, determine obstacle avoidance priorities, predict the trajectory using the EKF algorithm, detect conflicts based on the consistency constraint function, replan the trajectory using the RRT* algorithm, and output the combination of obstacle avoidance actions; Execution feedback module: configured to convert obstacle avoidance action combinations into flight control inputs, send them to the flight control system via CAN bus, collect execution status data and updated perception data, perform error correction through PID controller, and trigger closed-loop feedback adjustment; Air-to-ground communication module: configured to use 5G-A communication protocol, operating frequency band 2.4GHz / 5.8GHz, transmission rate ≥1Gbps, transmission latency ≤100ms, supporting data interaction between multiple aircraft and communication with ground sensing base stations; Power management module: configured to provide stable power supply to each module, including main power supply and backup power supply. When the main power supply fails, the backup power supply maintains the operation of the core functions of the system for ≥30 seconds.

[0014] Preferably, the multi-source heterogeneous sensor group includes a radar sensor, a photoelectric sensor, a positioning sensor, an inertial measurement unit, and an energy sensor; The data acquisition module removes outliers. The specific steps are as follows: Calculate a certain type of data ; Calculation standard ; like It was identified as an outlier and replaced by the mean of the data from the previous time step and the data from the next time step. ; The Redis database of the rule management module adopts a master-slave replication architecture. The master node stores the complete rule base, and the slave nodes are used for rule retrieval and adaptation. The data synchronization latency is ≤10ms. The rule base supports remote updates. It receives rule update packets sent by ground base stations through the air-to-ground communication module and automatically replaces invalid rules or adds scenario-based rules. The consistency constraint detection unit of the collaborative decision-making module adopts a parallel computing architecture to detect trajectory conflicts of multiple aircraft simultaneously. The detection delay increases linearly with the number of aircraft. The CAN bus of the execution feedback module adopts the CANFD protocol, which supports flexible data rates. Control quantity transmission uses standard data frames, and status data transmission uses extended data frames. The air-to-ground communication module supports multi-link redundancy, including a 5G-A main link and a LoRa backup link. When the main link signal strength is lower than -85dBm, it automatically switches to the backup link with a switching delay of ≤50ms to ensure uninterrupted data transmission.

[0015] Preferably, the system further includes: Fault diagnosis module: Configured to monitor the working status of each module in real time. When a module fault is detected, a fault emergency mechanism is triggered. Sensor failure: Automatically switch to redundant sensors and trigger ground data compensation; Communication module failure: Activate backup communication link; if backup link still fails, adopt preset emergency obstacle avoidance rules. Processor overload: Shut down non-core functions and prioritize the core processes of obstacle avoidance decision-making.

[0016] A computer-readable storage medium storing a computer program that, when executed by a processor of a low-altitude aircraft, implements the autonomous obstacle avoidance method for low-altitude aircraft as described in any one of claims 1-6. The computer-readable storage medium includes one or more of flash memory, SD card, and solid-state drive.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention significantly optimizes sensing accuracy through a three-dimensional reliability assessment and air-to-ground data compensation mechanism. It enhances the efficiency of sensing data in complex environments such as low light, strong electromagnetic interference, and heavy rain, avoiding obstacle avoidance decision errors caused by data distortion from a single sensor. When the reliability of a core sensor falls below a threshold, it automatically triggers ground data compensation or weight reallocation to ensure uninterrupted sensing links, addressing the lack of emergency mechanisms for sensor failure in existing technologies. By dynamically adjusting sensor data weights through an environmental adaptation coefficient, the performance degradation of radar, photoelectric, and other sensors under different electromagnetic interference and lighting conditions is controlled within 10%, adapting to the complex and ever-changing low-altitude operating environment.

[0018] This invention improves scenario adaptation accuracy by quantifying rule applicability and dynamically adjusting priorities. It can dynamically adjust rule priorities based on airspace density, weather disturbances, and aircraft performance, automatically selecting the optimal subset of rules in high-density airspace or low-energy scenarios. By using an applicability threshold, it reduces or freezes invalid rules to prevent them from triggering dangerous obstacle avoidance actions, thus improving the rationality of obstacle avoidance actions. Through global consistency constraints and dynamic priority determination, it significantly reduces the secondary conflict rate, achieving a conflict resolution success rate of ≥98% in high-density airspace, and supporting large-scale collaborative operation of aircraft. Attached Figure Description

[0019] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein: Fig. 1 This is a flowchart illustrating the steps of the autonomous obstacle avoidance method of the present invention. Fig. 2 This is a schematic diagram of the overall architecture of the autonomous obstacle avoidance system of this invention. Fig. 3 This is a flowchart illustrating the collaborative workflow of the various modules of the autonomous obstacle avoidance system of this invention. Detailed Implementation

[0020] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0021] Specific embodiments of the present invention are described below with reference to the accompanying drawings.

[0022] Please see Figs. 1-3 This embodiment proposes an autonomous obstacle avoidance method for low-altitude aircraft, applicable to unmanned aerial vehicles (UAVs) or electric vertical takeoff and landing (eVTOL) aircraft. The method is executed by the low-altitude aircraft and includes the following steps: Data acquisition steps: Data is collected by the multi-source heterogeneous sensor group on its own, which includes at least radar sensors, photoelectric sensors, positioning sensors, inertial measurement units and energy sensors. The collected data includes its own operating status data (position, velocity, acceleration, attitude angle, remaining energy), surrounding static obstacle data (position, outline, height) and dynamic target raw perception data (position, velocity, direction of movement, type).

[0023] The system receives airspace control information (no-fly zone boundaries, flight paths, airspace density thresholds), real-time meteorological data (wind speed, wind direction, visibility, precipitation intensity, light intensity), and collaborative sensing data (position, speed, obstacle avoidance intention, mission priority) from ground-based sensing base stations via the air-to-ground communication interface.

[0024] The collected raw data undergoes preprocessing, including data format standardization (converting it to structured data in the WGS84 coordinate system) and outlier removal (collection). (Except for data that is outside the sensor's measurement range) and time synchronization (synchronizing multi-source data to the same time axis based on GPS timestamps, with a time synchronization error of ≤10ms).

[0025] Data fusion processing steps: The credibility of preprocessed multi-source heterogeneous sensing data and collaborative sensing data is evaluated, and the sensing credibility weights corresponding to each data are calculated. Credibility assessment includes assigning inherent weighting coefficients, calculating error correction coefficients, and calculating environmental adaptability coefficients. The specific formulas are as follows: ,in: The inherent weighting coefficients are set based on sensor type and data importance, and the number of radar sensors... Position sensor data sensor Surface collaborative data ; The error correction coefficient is calculated based on historical measurement error statistics, and the formula is as follows: ,That For statistics Data k-th measurement The maximum permissible error; The coefficient is calculated based on current environmental parameters and radar sensor data. Number of sensors Zhao Qiang (degree), positioning sensor Quantity unit (Unaffected by the environment), energy sensor ( .

[0026] When the credibility of any data When the preset credibility threshold is reached, the weight ratio of the data in the weighted fusion is reduced (to 50% of the original weight), or the ground sensing base station collaborative sensing data compensation is triggered (if the ground data credibility is ≥0.3).

[0027] All data is fused using a trusted weighted fusion model to generate unified perception state information. That The data update frequency is 50Hz to obtain the unified perception state information after fusion (including its own state, obstacle state, dynamic target state, and environmental state).

[0028] The specific calculation method for the environmental adaptability coefficient is as follows: Radar sensor data environment adaptability coefficient Magnetic interference intensity; Photoelectric sensor data environment adaptation system : middle , ; Positioning sensor data environment adaptation coefficient middle Satellite signal strength; Ground collaborative data environment adaptation The communication distance between the aircraft and the ground sensing base station.

[0029] Rule adaptation processing steps: The system calls the preset dynamic obstacle avoidance rule library, which is stored in a Redis distributed database. It includes basic obstacle avoidance rules (10 rules), scenario-based obstacle avoidance rules (25 rules), and priority adaptation rules (15 rules). The rules are stored in a structured format of condition-action-priority baseline value.

[0030] Based on the application parameters of the unified sensing state information extraction rules, including spatial operational density, the calculation parameters are as follows: (Unit: frames / square kilometer), strong meteorological disturbance (Normalized to [0,1]), inherent performance (Normalized to [0,1]) and frequent historical conflicts (Normalized to [0,1]), the extraction and normalization methods for each parameter are as follows: Airspace Operations Density : Calculation based on airspace control information sent by ground sensing base stations and the number of dynamic targets sensed by the system itself. Normalization / square kilometer is the maximum permissible density); meteorological disturbances Based on wind calculate, ,That ; Its own performance Calculate, That High-speed acceleration, after normalization ; Historical conflict frequency t: Statistics on potential conflicts within the past 10 minutes After normalization .

[0031] By rule applicability function The applicability of each avoidance rule is calculated, and the rule applicability function uses a weighted summation. .

[0032] Preset applicability threshold Then, a demotion process will be implemented (the priority benchmark value will be reduced to 30% of the original benchmark value) or a temporary freeze will be performed (the rule candidate set will be removed).

[0033] Priority baseline value based on residual rules and Final priority: (Rule priority baseline value) Sort the rules from highest to lowest, and select the top 5 rules to form a subset of avoidance rules for the current scenario.

[0034] The rules in the dynamic avoidance rule base include, but are not limited to: Basic obstacle avoidance rules: Rule 1: When flying in the same direction, the following aircraft Rule 2: When meeting laterally, the aircraft on the right shall give way to the aircraft on the left. ; Rule 3: When flying head-on, both parties should move to the right to give way, with the right of first refusal giving way. ; Contextualized obstacle avoidance rules: Rule 11: Empty Prioritize using height adjustment to avoid collisions, based on priority benchmark values. ; Rule 12: Weather Disturbance Large safety margin increased to 1.5 times, priority baseline value. ; Rule 13: When the static obstacle is a tall structure (height > 100m), the avoidance distance shall not be less than 1.2 times the height of the obstacle, and priority shall be given to avoiding it. Priority adaptation rules: Rule 31: Residual Energy ; Rule 32: Task Priority ; Rule 33: Maneuvering Acceleration .

[0035] Steps for calculating the safety margin: Extract the basic safety margin from the preset safety margin library according to the avoidance rule subset, target type (static obstacle / dynamic target), and its own performance parameters The safety margin library is stored classified by aircraft type and target type, specifically: Drone: static ; eVTOL: static obstacle .

[0036] Calculate the dynamic parameter airspace operation density regression model Where: The maximum speed of the aircraft, no ; is the preset correction coefficient.

[0037] Calculate the constraint condition parameters: The minimum reaction time where T is the response time of the flight control system ; The minimum maneuvering safety margin mass (unit: kg).

[0038] Through the constraint conditions Differentiated safety margin The final safety margin is Updated every 100 ms.

[0039] Steps for collaborative decision-making processing: Detect potential operation conflicts based on the unified perception status information. The conflict detection uses the distance determination method: calculate the shortest straight line between itself and the static obstacle The relative distance to the target is in conflict.

[0040] If there is a potential conflict, send its own obstacle avoidance intention information (intended obstacle avoidance direction, adjustment amplitude, estimated completion time) to other aircraft through the air-ground communication interface, and receive the obstacle avoidance intention information of other aircraft. The communication data format uses JSON format, and the fields include: aircraft ID, current position, speed, obstacle avoidance type, priority.[[ID=​​​​​​​Emergency missions (rescue, firefighting) = 1.0, routine missions (logistics, inspection) = 0.7, test flight missions = 0.5; Quality First (Maximum mass of aircraft-class vehicles). Energy .

[0042] A multi-agent collaborative obstacle avoidance decision-making model with global consistency constraints is constructed, employing a trajectory prediction-conflict detection-replanning process: Trajectory prediction: The Extended Kalman Filter (EKF) algorithm is used to predict the trajectory within the next 5 seconds. Equations of state Equation middle State vector, State transition Observation matrix White noise (variance respectively) , ); Consistency constraint detection: through consistency constraint function Obstacle avoidance trajectory conflict, its (Returns 1 if the intersection of the trajectories is not empty, otherwise returns 0); Trajectory Lower-priority aircraft are marked as re-planning targets. Based on their remaining energy, maneuverability, and available airspace resources (obtained through ground sensing base stations), a Fast Search Random Tree (RRT*) algorithm is used to re-plan their trajectories. The planning objective is to find the best match between their trajectory and that of other aircraft. .

[0043] Determine the combination of obstacle avoidance maneuvers that will not cause secondary conflicts, as well as the leading obstacle avoidance aircraft (highest priority) and subordinate obstacle avoidance aircraft (lower priority than the leading one). The combination of obstacle avoidance maneuvers includes one or more of the following: heading adjustment, altitude adjustment (climb / descent), and speed adjustment.

[0044] Trajectory The specific steps of the RRT* algorithm used are as follows: Initialization: Setting the planned spatial boundaries (based on airspace control information for flightability) ; Random sampling: randomly within the planned space The sampling probability is: 90% for the vacant portion of the flyable area and 10% for the edge portion; Nearest neighbor search: in the generated trajectory tree ; Node expansion Collision detection: judgment Does the line segment conflict with obstacles or the trajectory of other aircraft? If so, Return to the data fusion processing steps; Parent node reselection: Calculate all nearest neighbors Choose the lowest cost option. The parent node; Path optimization: If the nearest neighbor node has a lower cost, then update the parent node; Termination condition: When the iteration number is reached... The iteration terminates when the distance is less than 10m. Optimal obstacle avoidance trajectory.

[0045] Execution and feedback steps: The obstacle avoidance maneuver combination is converted into flight control input, and the specific conversion relationship is as follows.

[0046] Heading adjustment maneuver → Yaw angle indication .

[0047] Altitude adjustment action → Pitch angle ; Speed ​​adjustment action → oil .

[0048] The flight control inputs are sent to the flight control system via the CAN bus. The CAN bus transmission rate is 1Mbps and the transmission delay is ≤20ms.

[0049] The system collects real-time obstacle avoidance action execution status data (actual yaw angle, pitch angle, throttle value) and updated perception status information at a frequency of 100Hz.

[0050] Calculation execution error When the error exceeds the preset threshold Differential correction (adjusting the control quantity through a PID controller).

[0051] Determine whether the current operating status meets the safety interval threshold. If the conditions are met, the conflict is resolved and the original flight path is restored; otherwise, the process returns to the rule adaptation processing step and then to the collaborative decision-making processing step to readjust the obstacle avoidance decision until the conflict is resolved.

[0052] The PID controller parameters used for error correction are as follows: Yaw angle P ; Pitch angle ; accelerator ; PID controller output error.

[0053] The dynamic selection strategy for obstacle avoidance maneuver combinations is as follows: When airspace Prioritize heading adjustments (60%), supplemented by speed adjustments (40%). Use a combination of heading adjustment (30%) + altitude adjustment (40%) + speed adjustment (30%); First, select the height adjustment action (accounting for 70%), and then supplement it with the speed adjustment (30%). When the remaining energy Climb maneuvers (altitude adjustments are only allowed for descent), prioritize speed adjustments (60%) + heading adjustments (40%). When meteorological disturbances Reduce the course adjustment range (≤15°), and prioritize speed adjustment (50%) + altitude adjustment (50%).

[0054] A low-altitude autonomous obstacle avoidance system for unmanned aerial vehicles (UAVs) or electric vertical takeoff and landing (EVL) aircraft, comprising: Data acquisition module: Configured to collect its own operational status data, surrounding obstacle and dynamic target data through a multi-source heterogeneous sensor group, and receive data from ground sensing base stations and other aircraft through an air-to-ground communication interface, and preprocess the data (format standardization, outlier removal, time synchronization); the multi-source heterogeneous sensor group includes a radar sensor (model: HJ-200, detection range 0.1-5km, ranging accuracy ±0.5m, sampling rate 50Hz), an optoelectronic sensor (model: GD-300, field of view 60°, image resolution 1920×1080, sampling rate 30Hz), a positioning sensor (GPS+BeiDou dual-mode, positioning accuracy ±0.3m, update rate 10Hz), an inertial measurement unit (model: IMU-100, sampling rate 100Hz, acceleration measurement range ±16g, angular velocity measurement range ±2000° / s), and an energy sensor (model: NL-50, measurement range 0-10000J, accuracy ±1%, sampling rate 10Hz); Outlier removal in the data acquisition module The specific steps are as follows: Calculate a certain type of data ; Calculation standard ; like It was identified as an outlier and replaced by the mean of the data from the previous time step and the data from the next time step. .

[0055] Trusted Fusion Module: Configured to assess the trustworthiness of preprocessed multi-source data and calculate perceived trustworthiness weights. The data compensation mechanism generates unified perception state information through a trusted weighted fusion model; the module is implemented using an FPGA chip (model: Xilinx Zynq UltraScale+MPSoC), the fusion algorithm is processed in parallel, and the data processing latency is ≤30ms.

[0056] Rule management module: Configured to store a dynamic avoidance rule library, extract rule applicability calculation parameters, calculate the applicability of each rule, dynamically adjust rule priorities, and form a subset of avoidance rules; the module uses an ARM Cortex-A76 processor (2.8GHz), and the rule retrieval and adaptation latency is ≤50ms; The Redis database of the rule management module adopts a master-slave replication architecture. The master node stores the complete rule base, and the slave nodes are used for rule retrieval and adaptation. The data synchronization latency is ≤10ms. The rule base supports remote updates by receiving rule update packets sent by ground base stations through the air-to-ground communication module, automatically replacing invalid rules or adding scenario-based rules.

[0057] Safety Interval Module: Configured to extract the basic safety interval, calculate the dynamic correction amount and constraint parameters, and determine the differentiated safety interval threshold S through multiple constraints; the module's safety interval calculation delay is ≤20ms.

[0058] Collaborative Decision Module: Configured to detect potential operational conflicts, interact with obstacle avoidance intent information, determine obstacle avoidance priorities, predict trajectories using the EKF algorithm, detect conflicts based on the consistency constraint function, replan the trajectory using the RRT* algorithm, and output obstacle avoidance action combinations; the module uses a GPU (model: NVIDIA Jetson AGX Orin) to accelerate trajectory prediction and replanning, with a decision latency ≤100ms; The consistency constraint detection unit of the collaborative decision-making module adopts a parallel computing architecture to detect trajectory conflicts of multiple aircraft (≤20) at the same time. The detection latency increases linearly with the number of aircraft. For each additional aircraft, the detection latency increases by ≤5ms.

[0059] Execution feedback module: configured to convert obstacle avoidance action combinations into flight control inputs, send them to the flight control system via CAN bus, collect execution status data and updated perception data, perform error correction through a PID controller, and trigger closed-loop feedback adjustment; the module's execution control delay is ≤50ms; The CAN bus of the execution feedback module adopts the CANFD protocol, which supports flexible data rates. Control signal transmission uses standard data frames (ID range: 0x000-0x7FF), and status data transmission uses extended data frames (ID range: 0x800-0x1FFF). The data frame format includes: frame start, arbitration field, control field, data field (8 bytes), CRC field, ACK field, and frame end.

[0060] Air-to-ground communication module: configured to use 5G-A communication protocol, operating frequency band 2.4GHz / 5.8GHz, transmission rate ≥1Gbps, transmission latency ≤100ms, supports data interaction between multiple aircraft (obstacle avoidance intentions, trajectory data) and communication with ground sensing base stations (airspace control, meteorological data). The air-to-ground communication module supports multi-link redundancy, including a 5G-A main link and a LoRa backup link (communication distance ≤10km, transmission rate ≥100kbps). When the main link signal strength is lower than -85dBm, it automatically switches to the backup link with a switching delay of ≤50ms to ensure uninterrupted data transmission.

[0061] Power management module: configured to provide stable power supply to each module, including main power supply (lithium battery pack, voltage 24V, capacity 20Ah) and backup power supply (supercapacitor, capacity 500F). When the main power supply fails, the backup power supply can maintain the operation of the core functions of the system for ≥30 seconds.

[0062] Fault diagnosis module: Configured to monitor the operating status of each module in real time (power supply voltage, operating temperature, data transmission status). When a module fault is detected (such as no data output from the sensor, communication module disconnection, processor overload), a fault emergency mechanism is triggered. Sensor failure: Automatically switch to redundant sensors (e.g., if the positioning sensor fails, enable the inertial measurement unit to assist in positioning) and trigger ground data compensation; Communication module failure: Activate the backup communication link. If the backup link still fails, adopt the preset emergency obstacle avoidance rules (maintain current altitude, reduce speed to 30% of maximum speed, and fly straight along the original route). Processor overload: Shut down non-core functions (such as historical data statistics) and prioritize the core process of obstacle avoidance decision-making.

[0063] The technical solution of this invention is based on the core logic of data-driven, dynamic adaptation, global constraints, and closed-loop feedback, constructing a five-layer full-link architecture of perception, rules, intervals, decision-making, and execution. Each layer achieves data interoperability and collaborative optimization through standardized interfaces. Data acquisition layer: Through multi-source heterogeneous sensors and air-to-ground communication interfaces, three-dimensional data acquisition is achieved, including local data from the aircraft, collaborative data from ground base stations, and interactive data from other aircraft, providing comprehensive data support for subsequent processing; Trusted Fusion Layer: Performs trustworthiness assessment and weighted fusion of multi-source data, triggers data compensation mechanism, outputs unified and trustworthy perception status information, and solves the problem of data distortion; Rule adaptation layer: Based on perceived state information, dynamically adjust the priority of avoidance rules, freeze invalid rules, form a subset of scenario-based rules, and solve the problem of rule rigidity; Safety Interval Layer: Combining the basic interval, dynamic correction amount and multiple constraints, calculate the differentiated safety interval threshold to solve the interval ambiguity problem; Collaborative decision-making layer: Based on conflict detection, intent interaction, trajectory prediction and consistency constraints, it outputs a combination of obstacle avoidance actions without secondary conflicts to solve the collaborative conflict problem; Execution feedback layer: Converts decision commands into flight control quantities, monitors execution status in real time, triggers closed-loop corrections, and ensures accurate execution of obstacle avoidance actions.

[0064] This invention constructs a three-dimensional credibility assessment system and an air-to-ground data compensation mechanism by moving from indiscriminate fusion to trusted weighted fusion plus data compensation. The specific implementation logic is as follows: Three-dimensional credibility assessment: Calculating data credibility weights from three dimensions: inherent weights, error correction, and environmental adaptation. The limitations of traditional single-dimensional weighting. For example, the environmental adaptation coefficient of photoelectric sensors under low light conditions (L=500 lux)... Credibility weight At a threshold of 0.1, no compensation is needed; if the light intensity further decreases to L=300 lux, Historical errors due to deterioration of light but Ground data compensation was implemented, and the weight of ground collaborative data was increased from 0.05 to 0.2 to ensure fusion accuracy.

[0065] Weighted fusion model optimization: A normalized weighted summation formula is adopted to ensure that the sum of the weights is 1, avoiding data superposition distortion. At the same time, the fusion model supports sensor redundancy configuration. When a sensor fails, its weight is automatically reassigned to other trusted sensors. For example, when the positioning sensor fails, its inherent weight of 0.2 is assigned to the radar sensor (0.1) and the inertial measurement unit (0.1), maintaining the stability of the fusion model.

[0066] This invention transforms a static rule set into an evolvable rule system, quantifies rule applicability, and dynamically adjusts priority, achieving runtime self-adaptation of the rule base. Rule applicability function calibration: The weight system of the applicability function was determined through a large number of simulation experiments (1000 sets of data from different scenarios). Ensure that the applicability accurately reflects the effectiveness of the rule in the current scenario. For example, in high-density air... Historical Conflicts In this scenario, the applicability of the priority height adjustment rule is considered. Priority Ultimate Excellence The rules are ranked lower to ensure that rules that are more suitable for high-density scenarios are selected first.

[0067] Rule lifecycle management: The rule base supports full lifecycle management from activation to demotion to freezing to updating. Rules are updated remotely via ground base stations to adapt to obstacle avoidance requirements in new scenarios (such as urban canyons and low-altitude sea areas). For example, a new rule that prioritizes avoidance along building edges in an urban canyon scenario is sent to the rule management module via the communication module, automatically integrated into the rule base, and participates in applicability calculations.

[0068] This invention constructs a quantitative model consisting of a basic interval, dynamic correction, and multiple constraints, moving from empirical intervals to multi-factor coupled constraint intervals, ensuring the rationality and dynamic adaptability of the safety interval. Physics with multiple constraints The device has enough time to respond to obstacle avoidance commands. Obstacles and maneuvers, both and the base The lower limit of the safety interval, dynamic Changes in environment and operating status; For example, a certain eVTOL operates at a speed of 100 m / s, with 50% remaining energy, a wind speed of 8 m / s, and an airspace density of 6 aircraft per square kilometer. , , Dynamic correction amount Final safe interval Ensure obstacle avoidance safety in high-speed, medium-energy, medium-weather disturbance, and high-density scenarios.

[0069] Self-calibration of dynamic correction coefficients: This involves using machine learning algorithms (such as linear regression) to self-calibrate dynamic correction coefficients. Self-calibration is implemented, optimizing coefficient values ​​based on historical obstacle avoidance data to improve the accuracy of dynamic corrections. For example, through training with 1000 sets of historical data, without... The value was adjusted to 0.28, making the dynamic correction amount more suitable for the operating characteristics of the drone.

[0070] This invention moves from distributed game theory to consistency-consistent collaboration, using global consistency constraints and dynamic priority determination to achieve multi-agent collaboration without secondary conflicts. The trajectory prediction algorithm generates the future 5-second trajectories for each aircraft. Set intersection operations are then used to determine if the trajectories overlap. If overlap exists... First, adjust the trajectories of lower-priority aircraft to ensure no global conflicts. For example, in a coordinated delivery operation with three drones, trajectory prediction reveals that the trajectories of drone 2 and drone 3 overlap after 3 seconds. If the priority of UAV 2 (normal mission, 2kg mass, 60% energy) is lower than that of UAV 3 (normal mission, 3kg mass, 70% energy), then UAV 2 will undergo trajectory replanning, with its heading adjusted by 15° to ensure that the trajectories do not overlap.

[0071] The system considers three dimensions—mission type, mass, and energy—to avoid unfairness caused by judging based on a single dimension (e.g., judging solely by mass might lead to small emergency mission aircraft avoiding large, general mission aircraft). For example, the obstacle avoidance priority of a small rescue drone (mass 1kg, mission priority 1.0, energy 50%). Priority of a logistics drone (10kg, mission priority 0.7, energy 80%); Large logistics drones still need to give way to small rescue drones. Therefore, after priority determination, a logic is added to force emergency mission aircraft to be prioritized to ensure that emergency missions are not affected.

[0072] The core modules of this invention achieve autonomous obstacle avoidance through the following collaborative workflow: Data Acquisition and Preprocessing: The data acquisition module collects data from multiple sources and performs standardization, outlier removal, and time synchronization, outputting structured data to the trusted fusion module; Trusted Fusion and State Awareness: The trusted fusion module calculates the data trustworthiness weights, triggers the compensation mechanism, and generates unified state awareness information through weighted fusion, which is then synchronized to the rule management module, security interval module, and collaborative decision-making module. Rule adaptation and interval calculation: The rule management module generates a subset of rules based on the perceived state information, the safety interval module calculates the differentiated safety interval threshold, and the results of both are output to the collaborative decision-making module; Collaborative decision-making and action generation: The collaborative decision-making module detects conflicts, interacts with obstacle avoidance intentions, and outputs a combination of obstacle avoidance actions to the execution feedback module based on rule subsets, safety interval thresholds, priorities, and consistency constraints. Execution feedback and closed-loop adjustment: The execution feedback module converts the action combination into flight control inputs and monitors the execution status in real time. If the obstacle avoidance effect does not meet the safety interval, it returns to the rule adaptation or collaborative decision-making steps for readjustment until the conflict is resolved.

[0073] Hardware system implementation details 1. Hardware configuration of the data acquisition module Radar sensor: HJ-200 millimeter-wave radar is selected, with an operating frequency of 24GHz, a detection range of 0.1-5km, a ranging accuracy of ±0.5m, an angle measurement accuracy of ±1°, a sampling rate of 50Hz, and support for target velocity measurement (range -50-50m / s, accuracy ±0.1m / s). It has anti-rain and fog interference capabilities (performance degradation ≤10% when rainfall ≤50mm / h). Photoelectric sensor: The GD-300 high-definition photoelectric sensor is selected, which includes a CMOS image sensor and an optical lens. The image resolution is 1920×1080, the frame rate is 30fps, the field of view is 60°, the focal length is 10mm, and it supports autofocus and exposure adjustment. The low light sensitivity is ≤0.1 lux (black and white mode). It has target contour extraction and classification functions (can distinguish aircraft, birds and buildings). Positioning sensor: Uses BDS / GPS dual-mode positioning module, supports SBAS differential positioning, positioning accuracy ±0.3m (horizontal) and ±0.5m (vertical), update rate 10Hz, cold start time ≤30s, hot start time ≤1s, and has anti-occlusion capability (can still maintain positioning accuracy ±1m when occlusion time ≤10s). Inertial Measurement Unit (IMU): IMU-100 MEMSIMU is selected, with a sampling rate of 100Hz, an acceleration measurement range of ±16g, a measurement accuracy of ±0.01g, an angular velocity measurement range of ±2000° / s, a measurement accuracy of ±0.1° / s, and a temperature compensation function (operating temperature -40℃ to 85℃). Energy sensor: NL-50 lithium battery energy sensor is selected, with a measurement range of 0-10000J, accuracy of ±1%, and sampling rate of 10Hz. It calculates the remaining energy by detecting battery voltage and current, and supports overcharge and over-discharge warnings. Air-to-ground communication modules: 5G-A communication module (model: 5G-8000) and LoRa module (model: LoRa-915) are selected. The 5G-A module operates at a frequency of 2.4GHz / 5.8GHz, with a transmission rate of 1Gbps, latency ≤100ms, and communication distance ≤10km; the LoRa module operates at a frequency of 915MHz, with a transmission rate of 100kbps, latency ≤200ms, and communication distance ≤10km, and supports multi-link redundancy switching. Data preprocessing unit: The STM32H743VIT6 microcontroller with a main frequency of 480MHz is selected. It integrates an FPU floating-point unit and is responsible for data format standardization (converting to JSON format), outlier removal (3σ criterion), and time synchronization (GPS timestamp alignment). The preprocessing latency is ≤10ms.

[0074] 2. Hardware Implementation of Trusted Fusion Module Core chip: Xilinx Zynq UltraScale+ MPSoC FPGA (model: XCZU9EG-2FFVB1156I) is selected, which includes a 4-core ARM Cortex-A53 processor and FPGA logic units (200k logic units), and supports parallel processing of algorithms; Storage unit: 2GB DDR4 SDRAM (2400MHz) for data caching, 128MB QSPI Flash for storing fusion algorithm firmware; Interface circuit: Communicates with the data acquisition module via SPI interface (100Mbps) to receive preprocessed multi-source data; interacts with the unified perception status information fused by the rule management module, safety interval module, and collaborative decision-making module via AXI4 bus (1Gbps); Power supply circuit: A DC-DC power module (input 24V, output 3.3V / 1.8V / 1.0V) is used, with a power supply ripple of ≤50mV to ensure stable chip operation.

[0075] 3. Hardware Implementation of the Rule Management Module Processor: ARM Cortex-A76 processor (model: RK3588), 2.8GHz clock speed, quad-core architecture, integrated NEON acceleration unit, supports fast operation of rule retrieval and adaptation algorithms; Storage unit: Configured with 4GB LPDDR4X memory (4266MHz speed), 64GB eMMC flash memory for storing dynamic avoidance rule base (Redis database), supporting local caching and remote updates of the rule base; Interface circuit: Receives unified perception status information from the trusted fusion module via Ethernet interface (Gigabit Ethernet), and sends rule subsets to the security interval module and collaborative decision-making module via UART interface (115200bps). Heat dissipation design: A passive heatsink + thermal pad heat dissipation solution is adopted, with an operating temperature range of -20℃ to 60℃, ensuring that the processor temperature is ≤85℃ when running at full load.

[0076] 4. Hardware Implementation of Safety Interval Module Processor: STM32H750VBT6 microcontroller with a main frequency of 480MHz, integrated FPU floating-point unit, responsible for safety interval calculation, calculation latency ≤20ms; Storage unit: 512KBSRAM is configured for caching calculation parameters, and 16MB Flash is configured for storing the safety interval library and calculation algorithm; Interface circuit: It receives sensing status information with the trusted fusion module through the I2C interface (1MHz rate), receives a subset of rules with the rule management module through the SPI interface (50Mbps rate), and sends the security interval threshold with the collaborative decision-making module through the UART interface.

[0077] 5. Hardware Implementation of the Collaborative Decision-Making Module Core processor: NVIDIA Jetson AGX Orin GPU (model: AGX Orin 64GB) is selected, which includes a 12-core ARM Cortex-A78AE processor and a GPU core (2048 CUDA cores), with a computing power of 200 TOPS, and supports acceleration of trajectory prediction and replanning algorithms; Storage unit: 64GB LPDDR5 memory (6400MHz), 1TB NVMe SSD for storing trajectory data and algorithm models; Interface circuit: Communicates with other modules through PCIe 4.0 interface (16Gbps), receives rule subsets, safety interval thresholds, and perception status information, and sends obstacle avoidance action combinations to the execution feedback module; Power supply circuit: It adopts a wide voltage input power module (input 12V-24V, output 5V / 3.3V) with a maximum output current of 20A to meet the high power consumption requirements of the GPU (typical power consumption of 30W).

[0078] 6. Hardware implementation of the execution feedback module Flight control interface unit: The STM32F407VET6 microcontroller is selected, with a main frequency of 168MHz. It integrates a CAN controller, supports the CANFD protocol, and communicates with the flight control system through the CAN bus (speed 1Mbps) to send flight control inputs and receive execution status data. PID control unit: Employs a dedicated PID controller chip (model: ADN8830), supporting hardware configuration and dynamic adjustment of proportional, integral, and derivative coefficients, with a control accuracy ≤0.1%; Status acquisition unit: Acquires flight control execution status data (yaw angle, pitch angle, throttle value) through ADC interface (12-bit precision, 100Hz sampling rate), and receives updated perception status information with the trusted fusion module through SPI interface; Fault detection unit: integrates voltage detection, current detection, and temperature detection circuits. When the execution error exceeds the threshold or a hardware failure occurs, an alarm signal is triggered.

[0079] 7. Hardware Implementation of Power Management Module Main power supply: Lithium battery pack (24V / 20Ah), energy density 200Wh / kg, cycle life ≥1000 times, supports fast charging (2C charging rate). Backup power supply: Supercapacitors (500F / 2.7V) are selected, configured in series 4 and parallel 3, with a rated voltage of 10.8V, a capacity of 500F, and support for high current discharge (100A), which can maintain the operation of the core functions of the system for ≥30 seconds; Power management chip: The TIBQ76952 battery management chip is selected, which supports battery voltage, current and temperature monitoring, and overcharge, over-discharge, overcurrent and over-temperature protection; the TITPS65987DDH power path management chip is selected to achieve seamless switching between main power and backup power, with a switching latency of ≤1ms. Power supply distribution: The DC-DC power supply module provides the appropriate voltage for each module: 3.3V for data acquisition module, 3.3V / 1.8V / 1.0V for trusted fusion module, 5V / 3.3V for rule management module, 12V / 5V / 3.3V for collaborative decision-making module, and 3.3V for execution feedback module.

[0080] Software system implementation details 1. Data Acquisition Module Software Implementation Sensor drivers: The driver interfaces for each sensor are written in C language to realize data acquisition, configuration parameter settings (such as radar detection range, photoelectric sensor frame rate), and fault detection functions; the radar sensor driver is based on the SPI protocol, the photoelectric sensor driver is based on the I2C protocol, the positioning sensor driver is based on the UART protocol, the IMU driver is based on the SPI protocol, and the energy sensor driver is based on the I2C protocol. Data preprocessing software: written in C++, including a data format standardization module (converts raw sensor data into JSON format, with fields including sensor ID, data type, value, and timestamp), an outlier removal module (3σ criterion), and a time synchronization module (aligns multi-source data based on GPS timestamps, with a synchronization error ≤10ms). Communication software: Based on the TCP / IP protocol, a communication program for the 5G-A module and the LoRa module is written to realize data interaction with ground sensing base stations and other aircraft. CRC32 check is used to ensure data transmission integrity, and a retransmission mechanism (timeout retransmission time 500ms) ensures that data is not lost.

[0081] Trusted Fusion Module Software Implementation Credibility weight calculation software: Written in C language and running on an ARM core of FPGA, it implements the assignment of inherent weight coefficients, calculation of error correction coefficients, calculation of environment adaptation coefficients, and outputs the credibility weight of each data point. The error correction coefficient is calculated using a sliding window algorithm (window size 1000), and the environment adaptation coefficient is calculated based on a preset piecewise function. Weighted fusion algorithm software: FPGA logic is written in VerilogHDL to implement a trusted weighted fusion model. Calculation, data width 32-bit floating point type, calculation precision ≤ 0.001; Data compensation software: written in C language, when a certain data... According to the compensation logic, the weight of this data is reduced or ground collaborative data compensation is enabled, and the compensation algorithm latency is ≤10ms. Software flow: Initialization → Receive multi-source data → Calculate credibility weights → Determine if compensation is needed → Yes (reduce weights / ground compensation) → Weighted fusion → Output unified perception status information; No → Weighted fusion → Output unified perception status information.

[0082] Rule management module software implementation Redis database configuration: Redis version 6.2.6 is used, configured with master-slave replication architecture. The master node stores the complete rule base (45 rules), and the slave nodes are used for rule retrieval and adaptation. Data synchronization adopts RDB persistence method (synchronization period of 10 seconds). Rule applicability calculation software: Written in Python, running on an ARM Cortex-A76 processor, extracting parameters from unified perception state information. Substitute applicability function after processing Calculate the applicability of each rule; Rule priority adjustment software: Written in C++, it calculates the final priority based on rule applicability and priority baseline values. Then select the first 5 rules to form a rule subset, and adjust the latency to ≤30ms; Rule update software: The rule update interface is written based on the HTTP protocol. It receives rule update packets (JSON format) sent by ground base stations, automatically replaces invalid rules or adds new rules, and the update process does not affect the current obstacle avoidance decision.

[0083] Safety Interval Module Software Implementation Safety Interval Storage: Basic safety intervals are stored using an SQLite database, categorized and indexed by aircraft type and target type, with a query latency of ≤5ms; Dynamic correction calculation software: written in C language, based on linear regression model Correction amount Remote configuration of ground base stations; Constraint calculation software: using C language ,pass Define the lower limit of the safety interval, and the final safety interval. ; Software flow: Initialization → Receive sensing state information and rule subset → Extract basic security space Calculate → Determine the safety interval To the collaborative decision-making module.

[0084] Implementation of collaborative decision-making module software Conflict detection software: written in C++, it calculates the distance between itself and obstacles and dynamic targets based on the distance determination method. When the distance is less than the safety interval threshold, it is determined to be a potential conflict. The detection delay is ≤10ms. Obstacle avoidance intention interaction software: A data interaction program based on the UDP protocol is written to send its own obstacle avoidance intention information (JSON format) and receive obstacle avoidance intention information from other aircraft, with an interaction cycle of 200ms; Obstacle avoidance priority determination software: The priority determination algorithm is written in C++ and is based on a three-dimensional evaluation model of mission type, aircraft mass, and remaining energy to realize the dynamic calculation of obstacle avoidance priority; Trajectory prediction software (EKF algorithm implementation): The Extended Kalman Filter (EKF) algorithm is implemented using C++ language combined with the Eigen linear algebra library to predict the flight trajectory of an aircraft within the next 5 seconds; Consistency Constraint Detection Software: A parallel conflict detection algorithm is written in C++, which realizes multi-aircraft trajectory conflict detection based on trajectory point intersection operation; Trajectory replanning software (RRT* algorithm implementation): The Fast Search Random Tree (RRT*) algorithm is implemented using C++ language in conjunction with PCL (PointCloudLibrary) for replanning conflict trajectories; Obstacle avoidance action combination generation software: The action selection algorithm is written in C++ and dynamically selects the optimal obstacle avoidance action combination based on rule subsets, safety interval thresholds and aircraft status. Implementation of the feedback module software Flight control input conversion software: The control input conversion algorithm is written in C language to convert obstacle avoidance action combinations into control commands that the flight control system can recognize; Execution status acquisition software: The status acquisition program is written in C language and acquires flight control execution status data through the ADC interface and CAN bus; PID control algorithm software: A discretized PID controller is implemented using C language to perform error correction; Closed-loop feedback logic software: The closed-loop feedback program is written in C++ to realize real-time monitoring and decision adjustment of obstacle avoidance effect; Fault diagnosis module hardware and software implementation Hardware configuration Fault detection sensors: configured with a voltage sensor (model: INA219, measuring range 0-36V, accuracy ±0.5%), a temperature sensor (model: DS18B20, measuring range -55℃-125℃, accuracy ±0.5℃), and a communication status detection circuit; Core processor: The STM32L476RGT6 low-power microcontroller with a main frequency of 80MHz is selected, which is responsible for module status monitoring and fault diagnosis. Interface circuit: Communicates with voltage and temperature sensors via I2C interface, exchanges status data with other modules via UART interface, and outputs fault alarm signals via GPIO interface; Alarm device: Integrated buzzer and LED indicator. The buzzer alarm frequency is 2Hz, and the LED indicator is solid red (fault) / solid green (normal).

[0085] Software Implementation A fault diagnosis program was written in C language to implement status monitoring, fault diagnosis, and emergency handling for each module, as detailed below: Module status monitoring: Power supply status monitoring: Real-time acquisition of power supply voltage for each module (3.3V for data acquisition module, 3.3V / 1.8V for trusted fusion module, etc.). When the voltage is lower than 90% or higher than 110% of the rated value, it is determined to be a power supply fault. Operating temperature monitoring: Collects the operating temperature of the processor, FPGA, and GPU. When the temperature exceeds 85℃ (processor) or 95℃ (FPGA / GPU), it is determined to be a temperature fault. Data transmission status monitoring: Monitor the frame rate of data interaction between modules. If there is no data or the data loss rate is >10% for three consecutive cycles (e.g., the data output frame rate of the trusted fusion module is <40Hz), it is determined to be a communication failure. Sensor status monitoring: Monitor the validity of sensor data output (e.g., if a radar sensor has no target data for 10 consecutive frames), it is determined to be a sensor malfunction; Fault severity classification: Minor faults: single non-core sensor failure (such as energy sensor), slight temperature exceedance (<90℃); Moderate faults: single-link failure of communication link (such as 5G-A link disconnection), and decrease in the accuracy of core sensors; Critical faults: main power supply failure, core processor overload (CPU utilization > 95% for 10 seconds), multiple modules failing simultaneously; Emergency response mechanism: Minor faults: Activate sensor redundancy (e.g., estimate remaining energy by battery voltage when the energy sensor fails), reduce power consumption of non-core functions, and continuously monitor fault status; Moderate fault: Switch to backup communication link (switch to LoRa when 5G-A is disconnected), adjust algorithm parameters (such as reducing processor frequency when temperature exceeds the limit), and send fault alarm to ground base station; Serious Failure: Activate backup power and execute emergency avoidance maneuvers (maintain current altitude, reduce speed to 30%v0, and fly straight along the original route), triggering the emergency landing procedure (if the failure cannot be recovered); Fault recording and reporting: The fault type, occurrence time, and handling result are stored in the local Flash (16MB capacity) and reported to the ground base station in real time through the air-to-ground communication module, which facilitates subsequent fault analysis.

[0086] Example 1: Obstacle Avoidance Scenario for Mountain Emergency Rescue Drones Scene parameters Aircraft: Hexacopter rescue drone, weight 8kg, maximum speed 80m / s, rated maximum maneuvering acceleration 3g, total energy 15000J, mission type: emergency rescue (transporting first aid medicines to people trapped in mountainous areas). Operating environment: mountainous terrain (altitude 500-1000m), airspace density 2 aircraft / km², wind speed 6m / s (gusts 8m / s), visibility 6km, light intensity 800lux (evening), electromagnetic interference intensity 60dBμV / m; Obstacles: Static obstacles (mountain rocks, power lines, rocks 50m high, power lines 80m high), dynamic targets (bird flocks, 5-8 birds, flight speed 10-15m / s, flight altitude 700-750m). Task requirements: obstacle avoidance delay ≤200ms, collision risk ≤1%, ensuring fast and safe delivery of medicines.

[0087] Execution process Data acquisition and preprocessing: Sensor data collection: Radar sensors detected mountain rocks (distance 400m, coordinates (X1,Y1,Z1=650m)), power transmission lines (distance 300m, coordinates (X2,Y2,Z2=800m)), and a flock of birds (distance 250m, coordinates (X3,Y3,Z3=720m)); the positioning sensor obtained its own coordinates (X0,Y0,Z0=700m) and velocity of 70m / s; the energy sensor collected the remaining energy of 12000J. Ground data reception: Ground sensing base stations transmit meteorological data (wind speed 6m / s, visibility 6km) and airspace control information (no no-fly zone). Preprocessing: Remove one frame of abnormal data (3m deviation) from the radar sensor, and synchronize multi-source data based on GPS timestamps with a synchronization error of 8ms.

[0088] Trusted fusion processing: Credibility weight calculation: radar transmission Magnetic interference E=60dBμV / m photoelectric sensing (800 lux of light) Visibility: 6km Positioning sensor Sex measurement form Quantity sensing ; Weighted fusion After integration, the precise location and movement status of rocks, power transmission lines, and flocks of birds were clearly identified, with a perception error of 0.8m.

[0089] Rule adaptation handling: Parameter extraction: empty Weather disturbance performance parameters ; Applicability Calculation: Priority Passage Rules for Emergency Mission Aircraft Priority Dynamic target priority height adjustment avoidance rules Priority Take the first 5 rules to form a subset.

[0090] Safety interval calculation: Basic safety interval: dynamic Dynamic correction amount ; Constraint parameters ; Safety Interval Threshold (Because bird flocks are dynamic targets and have sufficient remaining energy, the safety distance is mainly determined by energy constraints.)

[0091] Collaborative decision processing: Conflict detection: If the relative distance between oneself and the flock of birds is 250m < 1252.06m, it is considered a potential conflict; Priority determination: Its own mission priority is 1.0, which is forced to be at the top. Other targets have no cooperating aircraft and do not require interaction intent. Trajectory Prediction: EKF predicts trajectories for the next 5 seconds and finds similarities to bird flock trajectories after 3 seconds. ; Trajectory replanning: The trajectory is replanned using the RRT* algorithm, and the altitude is adjusted to 780m (higher than the flight altitude of the bird flock, which is 720-750m). The heading remains unchanged. The minimum distance between the replanned trajectory and the bird flock is 1300m ≥ 1.2 × 1252.06 ≈ 1502.47m. Action combination: Select altitude adjustment (80m climb) single action, action parameter: pitch angle .

[0092] Implementation and Feedback: Control input conversion: Pitch angle command 12.9°, throttle value maintained at 0.8 (maintaining speed 70m / s); Execution monitoring: When the flight control system executes the climb maneuver, the attitude monitoring unit collects the actual pitch angle value of 12.7°, with an error of 0.2°≤0.5°, requiring no calibration; Closed-loop feedback: After 3 seconds, the plane climbed to 780m and was 1350m away from the flock of birds (≥1502.47m), thus resolving the conflict. The plane continued to fly, avoiding power lines and mountain rocks, and successfully delivered the medicine. The total obstacle avoidance time was 180ms.

[0093] Implementation effect Perception accuracy: The unified perception state information error is 0.8m, which meets the perception requirements of complex mountainous terrain; Obstacle avoidance effectiveness: Successfully avoided static obstacles and dynamic flocks of birds, with no risk of collision; Real-time performance: The end-to-end latency is 180ms, meeting the rapid response requirements of emergency rescue. Energy consumption: Energy consumption for climbing 80m = 8 × 9.8 × 80 = 6272J, with 5728J of energy remaining, which still meets the return trip requirements.

[0094] Example 2: eVTOL Urban Commuting Obstacle Avoidance Scenario Scene parameters Aircraft: eVTOL urban commuter aircraft, mass 800kg, maximum speed 120m / s, rated maximum maneuvering acceleration 2g, total energy 500000J, mission type: ordinary commuting (carrying 4 passengers from city area A to area B, range 20km). Operating environment: low altitude in urban areas (flight altitude 300-500m), airspace density 9 aircraft / square kilometer (high density), wind speed 5m / s, visibility 10km, light intensity 15000lux (sunny day), electromagnetic interference intensity 40dBμV / m; Obstacles: Static obstacles (skyscraper, height 450m, coordinates (X1,Y1,Z1=450m)), dynamic targets (3 other eVTOLs, numbered 2, 3, and 4, speed 100-110m / s, flight altitude 350-400m, mission type is ordinary commuting). Task requirements: Smooth obstacle avoidance trajectory, high passenger comfort (acceleration change rate ≤ 3m / s³), secondary collision rate ≤ 2%, and on-time arrival rate ≥ 95%.

[0095] Execution process Data acquisition and preprocessing: Sensor data acquisition: Radar sensors detected a skyscraper (distance 800m, coordinates (X1,Y1,Z1=450m)), eVTOL2 (distance 600m, coordinates (X2,Y2,Z2=380m), velocity 105m / s), eVTOL3 (distance 700m, coordinates (X3,Y3,Z3=360m), velocity 100m / s), and eVTOL4 (distance 500m, coordinates (X4,Y4,Z4=390m), velocity 110m / s); its own coordinates (X0,Y0,Z0=370m), velocity 110m / s; the energy sensor acquired 400,000J of remaining energy. Ground data reception: Ground sensing base stations transmit airspace density of 9 aircraft / square kilometer, meteorological data (wind speed 5 m / s), and airspace control information (recommended airway X0±500m). Preprocessing: No anomalies were found in the multi-source data, and the time synchronization error was 5ms.

[0096] Trusted fusion processing: Credibility weight calculation: radar transmission Electrical sensors Sensing Sex measurement unit sensor Ground collaborative data ; Weighted fusion: After fusion, the precise trajectory of each dynamic target is clearly defined, with a perception error of 0.5m.

[0097] Rule adaptation handling: Parameter extraction: spatial domain elephant disturbance ; Applicability Calculation: Priority Altitude Adjustment Avoidance Rules for High-Density Airspace First (Ranked first); Ordinary commuter aircraft should give way based on quality priority. Priority (Ranked second) Rule subset: Select the first 5 rules, with the core rules being priority height adjustment and quality priority avoidance.

[0098] Safety interval calculation: Basic safety interval: dynamic target ; Dynamic correction amount , ; constraint parameters ; Safety Interval Threshold .

[0099] Collaborative decision processing: Conflict detection: If the relative distance between itself and eVTOL4 is 500m < 1210.5m, it is determined to be a potential conflict; Intent interaction: Send its own obstacle avoidance intent (intention to climb and adjust altitude) through the 5G-A communication module, and receive obstacle avoidance intents from eVTOL2, 3, and 4 (all of which are to maintain altitude and adjust heading). Priority determination: itself: ; eVTOL4: Mass 750kg, Remaining Energy 350,000J ; Its priority is higher than eVTOL4, making it the dominant obstacle avoidance aircraft; Trajectory prediction: EKF predicts that its own trajectory will overlap with that of eVTOL4 after 4 seconds (C=1). Track replanning: Maintaining its own heading, climbing to an altitude of 420m, eVTOL4 adjusts its heading by 10°. After replanning, the minimum distance between the two tracks is 1350m ≥ 1.2 × 1210.5 ≈ 1452.6m; Maneuver combination: Select the combination of altitude adjustment (50m climb) + speed maintenance, pitch angle The rate of change of acceleration is 2.5 m / s³ (meets comfort requirements).

[0100] Implementation and Feedback: Control input conversion: Pitch angle command 6.5°, throttle value 0.85 (maintain speed 110m / s); Execution monitoring: The flight control system executed the climb maneuver with an attitude error of 0.3°, requiring no correction. Closed-loop feedback: After 4 seconds, the plane climbed to 420m, and the distance to eVTOL4 was 1380m ≥ 1452.6m, thus resolving the conflict; the plane continued to fly along the recommended flight path, avoiding skyscrapers, and finally arrived at the destination on time, without any secondary conflicts throughout the entire process.

[0101] Implementation effect Collaborative obstacle avoidance effect: 0% secondary conflict rate among multiple entities, meeting the needs of high-density airspace collaboration; Passenger comfort: Acceleration change rate ≤ 3m / s³, no obvious bumps; Energy consumption: Energy consumption for climbing 50m = 800 × 9.8 × 50 = 392000J, with 8000J of energy remaining (enough for landing). Real-time performance: The end-to-end latency is 190ms, which meets the requirements for obstacle avoidance during high-speed operation.

[0102] Example 3: Obstacle Avoidance Scenario for Power Line Inspection Drones Scene parameters Aircraft: Multi-rotor power line inspection drone, weight 3kg, maximum speed 50m / s, rated maximum maneuvering acceleration 2.5g, total energy 5000J, mission type: power line inspection (inspection of high-voltage transmission lines and towers). Operating environment: Suburban power transmission line corridor (flight altitude 50-100m), airspace density 1 aircraft / square kilometer, wind speed 3m / s, visibility 7km, light intensity 3000 lux (cloudy day), electromagnetic interference intensity 80dBμV / m (electromagnetic radiation from power transmission lines). Obstacles: Static obstacles (transmission towers, height 80m, spacing 200m, coordinates (X1,Y1,Z1=80m), (X2,Y2,Z2=80m)), dynamic targets (kite, distance 150m, coordinates (X3,Y3,Z3=70m), no fixed direction of movement, speed 5-8m / s); Task requirements: The obstacle avoidance process should not affect the inspection accuracy (distance from the power transmission line ≥ 10m), have low energy consumption (remaining energy ≥ 30%), and be flexible in obstacle avoidance actions.

[0103] Execution process Data acquisition and preprocessing: Sensor data acquisition: Radar sensors detected a power transmission tower (distance 200m, coordinates (X1, Y1, Z1=80m)) and a kite (distance 150m, coordinates (X3, Y3, Z3=70m)); positioning sensors acquired their own coordinates (X0, Y0, Z0=75m) and speed 40m / s; photoelectric sensors acquired images of the power transmission line (for inspection). Ground data reception: Ground sensing base stations send power transmission line corridor control information (flight range X0±50m) and meteorological data (wind speed 3m / s); Preprocessing: Due to electromagnetic interference, the radar sensor exhibits a data deviation of 2 frames (2m). Then discard it and replace it with the average of the preceding and following frames.

[0104] Trusted fusion processing: Credibility weight calculation: radar transmission (electricity photoelectric sensor (Illumination 3000 lux, Positioning sensor Inertial measurement Measurement sensor ; Weighted fusion: After fusion, the real-time position of the kite is clearly determined, with a sensing error of 0.6m.

[0105] Rule adaptation handling: Parameter extraction: spatial domain strong disturbance Its own performance parameters ; Applicability Calculation: Low-Altitude Static Obstacle Priority Course Adjustment Avoidance Rules ,excellent Dynamic targets (without fixed trajectories) prioritize speed and heading adjustment rules. Priority (Ranked first).

[0106] Safety interval calculation: Basic safety interval: dynamic target (kite) ; Dynamic correction amount: ; constraint parameters ; Safety Interval Threshold .

[0107] Collaborative decision processing: Conflict detection: If the relative distance between the user and the kite is 150m < 757.6m, it is considered a potential conflict. Maneuver combination: Select speed adjustment (decrease to 30m / s) + heading adjustment (yaw 15° to the right) combination maneuver, maneuver parameters: throttle value 0.6 (30 / 50), yaw angle 15°; Track prediction: The adjusted EKF prediction track does not overlap with the kite's trajectory. .

[0108] Implementation and Feedback: Actions performed: The flight control system performs speed and heading adjustments, with an execution error of 0.5° (heading) and 0.03° (throttle). Closed-loop feedback: After 3 seconds, the distance to the kite is 800m ≥ 757.6m, the conflict is resolved; the original speed of 40m / s is restored, the inspection continues, and the tower inspection task is completed.

[0109] Implementation effect Inspection accuracy: During obstacle avoidance, the distance between the obstacle and the power transmission line is maintained at 15m, which does not affect the image acquisition accuracy; Energy consumption: Obstacle avoidance energy consumption = speed adjustment energy consumption + heading adjustment energy consumption ≈ 500J, with remaining energy of 4500J (90%), which meets the needs of subsequent inspections; Flexibility: Flexible movement combinations, adaptable to dynamic scenarios where kites have no fixed trajectory, with a 100% obstacle avoidance success rate.

[0110] Comparative test To verify the technical effectiveness of the present invention, a comparative experiment was designed: the present invention (experimental group) and existing mainstream obstacle avoidance technologies (control group, using traditional Kalman filter fusion + fixed rule base + empirical safety interval + distributed game collaboration) were used as research objects, and the core performance indicators were tested under the same simulation environment and real-world scenario.

[0111] Test environment Simulation environment: Prescan + MATLAB / Simulink, simulating multiple scenarios with airspace density of 0-10 aircraft / km², wind speed of 0-20m / s, and obstacle density of 0-5 objects / km²; Real-world scenarios: suburban areas (low-density airspace), city centers (high-density airspace), and mountainous areas (complex terrain), with 100 sets of samples tested in each scenario. Test metrics Perception accuracy: The average error (m) of uniformly perceived state information; Rule adaptation accuracy: the degree of matching between the rule subset and the scenario (%); Reasonableness of safety distance: The rate (%) of conformity between the actual distance after obstacle avoidance and the safety distance threshold; Collaborative obstacle avoidance success rate: the success rate (%) of multi-subject obstacle avoidance without secondary conflict; End-to-end latency (ms); Collision risk (%) Energy efficiency optimization rate: the percentage reduction in energy consumption compared to the control group (%) Experimental Results and Analysis The simulation test data is shown in the table below: The field test data are shown in the table below: Results Analysis Perception performance: Through trusted weighted fusion and data compensation, the present invention reduces the perception error by 75% compared with the prior art. It can still maintain high-precision perception in complex terrain and electromagnetic interference environments, and solves the problem of data distortion in the prior art. Rule adaptation: The dynamic rule evolution mechanism improves the adaptation accuracy by 22.8%. In high-density airspace and severe weather scenarios, it can quickly adjust rule priorities and avoid interference from invalid rules. Safety distance: The multi-constraint quantification model improves the safety distance compliance rate by 57.5%, ensuring obstacle avoidance safety under different speed, energy, and environmental conditions, and reducing collision risk by 85.5%; Collaborative obstacle avoidance: Consistency constraints and priority determination improve the success rate of collaboration by 15.1%, and there are no secondary conflicts in high-density airspace (10 aircraft / square kilometer), meeting the needs of large-scale operation; Real-time performance and energy consumption: End-to-end latency is reduced by 26% and energy consumption is optimized by more than 15%, balancing real-time obstacle avoidance and endurance requirements, making it suitable for long-range scenarios.

[0112] This invention addresses the core shortcomings of existing low-altitude aircraft obstacle avoidance technologies, such as unreliable perception, rigid rules, ambiguous spacing, and collaborative conflicts. It constructs a full-link autonomous obstacle avoidance system encompassing perception, rules, spacing, decision-making, and execution. Through four core innovative modules—reliable perception fusion, dynamic rule evolution, constrained safety spacing, and global collaborative decision-making—it achieves high-precision, high-reliability, and low-energy-consumption autonomous obstacle avoidance in complex environments. It can be widely applied to various low-altitude aircraft application scenarios, including drone logistics delivery, eVTOL urban commuting, emergency rescue, and power line inspection, providing core technological support for the large-scale and safe development of the low-altitude economy. It possesses significant engineering application value and market prospects.

[0113] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method for autonomous obstacle avoidance of low-altitude aircraft, characterized in that, The method is performed by a low-altitude aircraft and includes the following steps: S1. Data Acquisition: It collects data through its own multi-source heterogeneous sensor group and receives airspace control information, real-time meteorological data and collaborative sensing data from other aircraft sent by ground sensing base stations through the air-to-ground communication interface. The collected raw data is preprocessed, including data format standardization, outlier removal, and time synchronization. S2. Data Fusion Processing: The credibility of the preprocessed multi-source heterogeneous sensing data and collaborative sensing data is assessed, and the sensing credibility weights corresponding to each data point are calculated. Credibility assessment includes assigning inherent weighting coefficients, calculating error correction coefficients, and environmental adaptability coefficients; When the credibility of any data , A preset confidence threshold is set to reduce the weight of this data in the weighted fusion, triggering compensation for collaborative sensing data from ground sensing base stations. All data are fused using a trusted weighted fusion model to generate unified perception state information; S3. Rule Adaptation Processing: Call the preset dynamic avoidance rule library, which is stored in a Redis distributed database; extract rule applicability calculation parameters based on unified awareness status information; By rule applicability function Calculate the applicability of each avoidance rule; preset the applicability threshold. Then, a weight reduction process will be implemented; based on the priority baseline value of the remaining rules and... Final priority: , Sort the rules from highest to lowest, and select the top 5 rules to form a subset of avoidance rules for the current scenario; S4. Safety Interval Calculation: Based on the avoidance rule subset, target type, and its own performance parameters, the basic safety interval is extracted from the preset safety interval library. ; Calculate dynamics ; Calculate constraint parameters: minimum reaction chamber When T is the flight control system response time ; Minimum maneuver safety space Quality; through constraints Differentiated security room The final safe interval is Updated every 100ms; S5. Collaborative Decision Processing: Detects potential operational conflicts based on unified perception state information. Conflict detection uses a distance-based method: calculating the shortest straight line between itself and static obstacles. Relative distance of the target In conflict; If a potential conflict exists, the system sends its own obstacle avoidance intention information to other aircraft through the air-to-ground communication interface and receives obstacle avoidance intention information from other aircraft; determines the obstacle avoidance priority; constructs a multi-agent collaborative obstacle avoidance decision model with global consistency constraints; and determines the combination of obstacle avoidance actions without secondary conflicts, as well as the leading obstacle avoidance aircraft and subordinate obstacle avoidance aircraft. The combination of obstacle avoidance actions includes one or more of the following: heading adjustment, altitude adjustment, and speed adjustment. S6. Execution and Feedback: Convert obstacle avoidance maneuvers into flight control inputs; send the flight control inputs to the flight control system via the CAN bus to collect real-time obstacle avoidance maneuver execution status data and updated perception status information; Calculation execution error When the error exceeds the preset threshold Difference correction; determining whether the current operating status meets the safety interval threshold. If the conditions are met, the conflict is resolved and the original flight path is restored; otherwise, the process returns to the rule adaptation processing step and then to the collaborative decision-making processing step to readjust the obstacle avoidance decision until the conflict is resolved.

2. The method for autonomous obstacle avoidance of low-altitude aircraft according to claim 1, characterized in that, The multi-source heterogeneous sensor group mentioned in step S1 includes at least a radar sensor, a photoelectric sensor, a positioning sensor, an inertial measurement unit, and an energy sensor. The collected data includes its own operating status data, surrounding static obstacle data, and dynamic target raw perception data.

3. The method for autonomous obstacle avoidance of low-altitude aircraft according to claim 1, characterized in that, The specific formulas for assigning inherent weight coefficients, calculating error correction coefficients, and calculating environmental adaptation coefficients in step S2 are as follows: ,in: The inherent weighting coefficients are set based on sensor type and data importance, and the number of radar sensors... Position sensor data sensor Surface collaborative data ; The error correction coefficient is calculated based on historical measurement error statistics, and the formula is as follows: ,That For statistics Data k-th measurement The maximum permissible error; The coefficient is calculated based on current environmental parameters and radar sensor data. Number of sensors Positioning sensor Quantity unit Energy sensor ; The trusted weighted fusion model That The data update frequency is 50Hz to obtain the unified perception state information after fusion.

4. The method for autonomous obstacle avoidance of low-altitude aircraft according to claim 1, characterized in that, The dynamic avoidance rule base mentioned in step S3 includes basic avoidance rules, scenario-based avoidance rules, and priority adaptation rules. The rules are stored in a structured format of condition-action-priority baseline value. The rule applicability calculation parameters include airspace operation density. Strong meteorological disturbances Its own performance and frequent historical conflicts The extraction and normalization methods for the rule applicability calculation parameters are as follows: Airspace Operations Density : Calculation based on airspace control information sent by ground sensing base stations and the number of dynamic targets sensed by the system itself. Normalization ; meteorological disturbances Based on wind calculate, ,That ; Its own performance Calculate, That High-speed acceleration, after normalization ; Historical conflict frequency t: Statistics on potential conflicts within the past 10 minutes After normalization ; The rule applicability function uses a weighted summation. .

5. The method for autonomous obstacle avoidance of low-altitude aircraft according to claim 1, characterized in that, The safety interval database mentioned in step S4 is stored based on aircraft type and target type, specifically as follows: Drones: Static ; eVTOL: Static Barrier ; Reference Airspace operation secret Regression model Wherein: This is the preset correction factor; The obstacle avoidance priority described in step S5 is based on mission type, aircraft mass, and remaining... : Task type optimization Emergency missions = 1.0, regular missions = 0.7, test flight missions = 0.5; Quality First ; Energy ; The multi-agent collaborative obstacle avoidance decision-making model adopts a process of trajectory prediction-conflict detection-replanning: Trajectory prediction: The Extended Kalman Filter (EKF) algorithm is used to predict the trajectory within the next 5 seconds. Equations of state Equation middle State vector, State transition Observation matrix White noise; Consistency constraint detection: through consistency constraint function Obstacle avoidance trajectory conflict, its number; Trajectory Repetition Lower-priority aircraft are marked as re-planning targets. Based on their remaining energy, maneuverability, and available airspace resources, the Fast Search Random Tree (RRT*) algorithm is used to re-plan their trajectories. The planning objective is to find the best match between their trajectory and that of other aircraft. .

6. The method for autonomous obstacle avoidance of low-altitude aircraft according to claim 1, characterized in that, The specific conversion relationship between the obstacle avoidance action combination described in step S6 and the flight control quantity is as follows: Heading adjustment maneuver → Yaw angle indication Altitude adjustment action → Pitch angle Speed ​​adjustment action → oil .

7. A low-altitude aircraft autonomous obstacle avoidance system, characterized in that, The system includes: Data acquisition module: configured to collect its own operational status data, surrounding obstacles and dynamic target data through a multi-source heterogeneous sensor group, receive data from ground sensing base stations and other aircraft through the air-to-ground communication interface, and preprocess the data; Trusted Fusion Module: Configured to assess the trustworthiness of preprocessed multi-source data and calculate perceived trustworthiness weights. The data compensation mechanism generates unified perception state information through a trusted weighted fusion model; Rule management module: Configured to store a dynamic avoidance rule library, extract rule applicability calculation parameters, calculate the applicability of each rule, dynamically adjust rule priority, and form a subset of avoidance rules; Safety Interval Module: Configured to extract the basic safety interval, calculate the dynamic correction amount and constraint parameters, and determine the differentiated safety interval threshold S through multiple constraints; Collaborative Decision Module: Configured to detect potential operational conflicts, interact with obstacle avoidance intent information, determine obstacle avoidance priorities, predict the trajectory using the EKF algorithm, detect conflicts based on the consistency constraint function, replan the trajectory using the RRT* algorithm, and output the combination of obstacle avoidance actions; Execution feedback module: configured to convert obstacle avoidance action combinations into flight control inputs, send them to the flight control system via CAN bus, collect execution status data and updated perception data, perform error correction through PID controller, and trigger closed-loop feedback adjustment; Air-to-ground communication module: configured to use 5G-A communication protocol, operating frequency band 2.4GHz / 5.8GHz, transmission rate ≥1Gbps, transmission latency ≤100ms, supporting data interaction between multiple aircraft and communication with ground sensing base stations; Power management module: configured to provide stable power supply to each module, including main power supply and backup power supply. When the main power supply fails, the backup power supply maintains the operation of the core functions of the system for ≥30 seconds.

8. The low-altitude aircraft autonomous obstacle avoidance system according to claim 7, characterized in that, The multi-source heterogeneous sensor group includes a radar sensor, a photoelectric sensor, a positioning sensor, an inertial measurement unit, and an energy sensor; The data acquisition module removes outliers. The specific steps are as follows: Calculate a certain type of data ; Calculation standard ; like It was identified as an outlier and replaced by the mean of the data from the previous time step and the data from the next time step. ; The Redis database of the rule management module adopts a master-slave replication architecture. The master node stores the complete rule base, and the slave nodes are used for rule retrieval and adaptation. The data synchronization latency is ≤10ms. The rule base supports remote updates. It receives rule update packets sent by ground base stations through the air-to-ground communication module and automatically replaces invalid rules or adds scenario-based rules. The consistency constraint detection unit of the collaborative decision-making module adopts a parallel computing architecture to detect trajectory conflicts of multiple aircraft simultaneously. The detection delay increases linearly with the number of aircraft. The CAN bus of the execution feedback module adopts the CANFD protocol, which supports flexible data rates. Control quantity transmission uses standard data frames, and status data transmission uses extended data frames. The air-to-ground communication module supports multi-link redundancy, including a 5G-A main link and a LoRa backup link. When the main link signal strength is lower than -85dBm, it automatically switches to the backup link with a switching delay of ≤50ms to ensure uninterrupted data transmission.

9. The low-altitude aircraft autonomous obstacle avoidance system according to claim 7, characterized in that, The system also includes: Fault diagnosis module: Configured to monitor the working status of each module in real time. When a module fault is detected, a fault emergency mechanism is triggered. Sensor failure: Automatically switch to redundant sensors and trigger ground data compensation; Communication module failure: Activate backup communication link; if backup link still fails, adopt preset emergency obstacle avoidance rules. Processor overload: Shut down non-core functions and prioritize the core processes of obstacle avoidance decision-making.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by the processor of the low-altitude aircraft, implements the autonomous obstacle avoidance method for low-altitude aircraft as described in any one of claims 1-6. The computer-readable storage medium includes one or more of flash memory, SD card, and solid-state drive.

Images