An unmanned aerial vehicle safety obstacle avoidance method and system based on near-surface airflow sensing

Abstract

This invention discloses a method and system for safe obstacle avoidance of unmanned aerial vehicles (UAVs) based on near-surface airflow perception. The method collects the UAV's linear velocity, attitude quaternions, body angular velocity, and rotor speeds to construct a continuous state time window. It uses a temporal convolutional network to extract the temporal characteristics of rotor-induced airflow caused by disturbances on obstacle surfaces, and outputs the estimated obstacle distance through a multilayer perceptron. A nominal translational acceleration is generated based on a reference trajectory. Based on the estimated distance, a safety set and exponential control obstacle function constraints are established. A quadratic programming problem that minimizes the nominal input correction is solved to obtain the safe translational acceleration. The safe translational acceleration is then converted into thrust, desired attitude, and rotor commands. This invention can utilize the UAV's existing flight state to achieve near-surface obstacle perception, reducing reliance on cameras or lidar. It balances trajectory tracking and obstacle avoidance through safety constraints, making it suitable for transparent, weakly textured, or perception-degraded environments.

Description

Technical Field

[0001] This invention relates to the fields of autonomous control, aerodynamic perception and safety control of unmanned aerial vehicles (UAVs), and specifically to a method and system for safe obstacle avoidance of UAVs based on near-surface airflow perception. Background Technology

[0002] Quadrotor drones, characterized by their compact structure, high maneuverability, and ability to fly in confined spaces, have been used for tasks such as inspection, search and rescue, warehousing, and indoor exploration. In environments near walls, with transparent obstacles, or where lighting changes significantly, the aircraft needs to promptly determine the distance to obstacles and make safe corrections to control commands. Existing obstacle avoidance solutions typically rely on cameras, lidar, ultrasonic sensors, or depth sensors. Cameras are susceptible to low light, strong light, smoke, and transparent or low-texture surfaces; lidar and depth sensors increase weight, power consumption, and cost, and may miss detections on transparent surfaces. For small drones, these external sensing devices further reduce payload and endurance margins.

[0003] The rotor generates lift while simultaneously creating induced airflow. When a UAV approaches a wall, enclosure, or other obstacle, the flow field, pressure distribution, and effective thrust of this induced airflow change with distance. The rotor speed, attitude, and velocity adjustments made by the flight controller to maintain attitude and trajectory incorporate obstacle distance information. Existing technologies treat these near-surface aerodynamic effects primarily as disturbances to be suppressed, lacking a method to integrate them as active sensing information with a controller possessing verifiable safety constraints. Therefore, a safe obstacle avoidance technology is needed that does not rely on conventional external sensing sensors, can estimate the distance to near-surface obstacles using the UAV's existing flight state, and can correct the nominal control input in real time when estimation errors exist.

[0004] Chinese patent application CN201910310817.1 proposes an obstacle avoidance device and method for quadcopter UAVs based on infrared light ranging. The method installs four sets of infrared light transmitting and receiving devices, ultrasonic ranging modules, and servo motor-plane mirror scanning devices on the UAV. The UAV changes its flight direction according to the measured distance to the obstacle and a preset threshold. However, there are two problems: (1) The method requires additional configuration of multiple sets of infrared sensors, ultrasonic modules, and servo motor scanning mechanisms, which increases the structural complexity, weight, and energy consumption of the UAV. This is not suitable for small quadcopter UAVs with limited load and endurance. (2) Infrared ranging depends on the effective reflection of infrared light by the surface of the obstacle. Its measurement results are easily affected by strong ambient light, transparent or mirror obstacles, surface reflectivity, and changes in incident angle. This may result in ranging errors or missed detections, making it difficult to adapt to complex perception degradation environments.

[0005] Chinese patent application CN202011506198.2 proposes a method for tracking the safe trajectory of a UAV based on predictive control and barrier functions. This method uses model predictive control to generate a reference control quantity and constructs a control barrier function based on the distance between the UAV and the obstacle obtained by the airborne sensor. The reference control quantity is corrected by quadratic programming. However, there are two problems: (1) This method assumes that the airborne sensor can directly and continuously obtain obstacle information and distance. It does not involve using rotor speed, inertial measurement information and flight state time-series changes to estimate the obstacle distance. Therefore, it is still limited by external perception degradation factors such as weak light, smoke, transparent or low texture surfaces; (2) This method uses a first-order control barrier function constraint based on the position safety domain. However, the obstacle distance usually has a high relative order with respect to the quadcopter thrust or translation acceleration input. This patent does not specifically deal with the relative order problem and is difficult to directly use for close-range obstacle avoidance control with limited reaction time. Summary of the Invention

[0006] This invention aims to address the problems of missed detection, increased load, and insufficient safety constraints in existing small UAVs that rely on cameras and lidar for near-range obstacle avoidance in transparent, weakly textured, or perception-degraded environments. It provides a technical solution that transforms the near-surface effect of rotor-induced airflow into a distance observation and uses this to construct safety control constraints.

[0007] The technical solution of the present invention is as follows: This invention provides a method for safe obstacle avoidance of unmanned aerial vehicles (UAVs) based on near-surface airflow perception, comprising the following steps: Step S1: Collect continuous flight status data of the UAV during flight, and construct a fixed-length status data sequence according to the time sequence of the continuous flight status data; Step S2: Input the state data sequence into the pre-trained temporal distance estimation model, extract the temporal features of rotor-induced airflow caused by the near-face effect of obstacles, and output the estimated distance of the UAV relative to the obstacles; Step S3: Calculate the nominal translational acceleration based on the reference trajectory generated by the trajectory generation module and the current position of the UAV, and construct a safety set and exponential control obstacle function constraints based on the estimated distance; Step S4: Under the constraints of the exponential control barrier function, solve the quadratic programming problem with the objective of minimizing the correction amount of the nominal translational acceleration to obtain the safe translational acceleration; Step S5: Convert the safe translational acceleration into total thrust command and desired attitude command to control the UAV to move away from the obstacle.

[0008] This invention also provides a UAV safety obstacle avoidance system based on near-plane airflow perception, corresponding to the above method, the system comprising: The status acquisition module is used to collect continuous flight status data of the UAV during flight, and the continuous flight status data is constructed into a fixed-length status data sequence according to the time sequence; The distance estimation module is used to input the state data sequence into a pre-trained temporal distance estimation model, extract the temporal features of rotor-induced airflow caused by the near-face effect of obstacles, and output the estimated distance of the UAV relative to the obstacles. The nominal control module is used to calculate the nominal translational acceleration based on the reference trajectory generated by the trajectory generation module and the current position of the UAV, and to construct a safety set and exponential control obstacle function constraints based on the estimated distance; The safety filtering module is used to solve a quadratic programming problem under the constraints of the exponential control barrier function, with the objective of minimizing the correction amount of the nominal translational acceleration, to obtain the safety translational acceleration; The flight execution module is used to convert the safe translational acceleration into total thrust commands and desired attitude commands to control the UAV to move away from obstacles.

[0009] The advantages of this invention compared to the prior art are as follows: 1) By using the aircraft's existing inertia, attitude and rotor speed data to sense obstacles, the reliance on external sensing devices such as cameras and lidar can be reduced, which is especially suitable for transparent or weakly textured obstacles. 2) Temporal convolutional networks utilize historical information from multiple flight states simultaneously, enabling them to extract stable near-surface features from noisy, time-delayed, and coupled aerodynamic responses, thereby improving the real-time performance and robustness of distance estimation. 3) The estimated distance is directly embedded into the exponential control barrier function, and the nominal input is minimized through quadratic programming, so that trajectory tracking performance and safety can be balanced. 4) The perception and control form a closed loop, and the system can be updated in a rolling manner according to the control cycle. It can respond to both static transparent obstacles and dynamic obstacles approaching the UAV. 5) The technical solution can be deployed in existing flight control systems or airborne computers through software, requiring minimal hardware modifications and exhibiting good engineering adaptability. Attached Figure Description

[0010] The accompanying drawings, which are included to provide a further understanding of the embodiments of this specification and form part of the embodiments of this specification, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating the UAV safety obstacle avoidance method based on near-surface airflow perception according to the present invention. Figure 2 This is a schematic diagram of a transparent static obstacle test scenario and the results of distance estimation and safety control in one embodiment. Figure 3 This is a schematic diagram illustrating the results of a dynamic obstacle approach avoidance test, distance estimation, and safety control in one embodiment. Figure 4 This is a schematic diagram of a UAV safety obstacle avoidance system based on near-surface airflow perception, provided as an embodiment of this specification. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0012] like Figure 1 As shown in the figure, this specification provides an embodiment of a UAV safety obstacle avoidance method based on near-surface airflow perception. The method may specifically include the following steps: Step S1: Collect continuous flight status data of the UAV during flight. The continuous flight status data includes at least linear velocity, attitude quaternion, body angular velocity and rotational speed of each rotor. The continuous flight status data is constructed into a fixed-length state data sequence in chronological order. Specifically, in the first Construct a single-time state vector at each sampling time. , formula (1).

[0013] in, The linear velocity in the inertial coordinate system. For attitude quaternions, The angular velocity in the body coordinate system. For the first The rotational speed of each rotor.

[0014] Recently The state vectors at each time point are arranged in chronological order to obtain the state data sequence. , formula (2).

[0015] in, At least linear velocity, attitude quaternion, airframe angular velocity, and rotor speeds. This represents the length of the sliding time window.

[0016] Each data source is first resampled according to a unified timestamp. Missing short-term data can be imputed using either preserve-the-data or interpolation methods, and then standardized according to the training set statistics. The sliding time window moves forward one step at each new sampling time, thereby continuously generating distance estimates.

[0017] Step S2: Input the state data sequence into the pre-trained temporal distance estimation model, extract the temporal features of rotor-induced airflow caused by the near-face effect of obstacles, and output the estimated distance of the UAV relative to the obstacles; Specifically, the temporal distance estimation model in step S2 includes a temporal convolutional network and a multilayer perceptron connected in sequence. The temporal convolutional network includes at least two causal convolutional temporal blocks, which employ dilated convolutions to expand the temporal receptive field and include at least one of nonlinear activation, batch normalization, dropout layers, and residual connections. The multilayer perceptron maps the high-dimensional temporal features output by the temporal convolutional network to the estimated distance.

[0018] The time-series distance estimation model can be expressed as: ,in These are the network parameters obtained through training. Temporal convolutional networks use causal convolution to ensure that the first... The output at each time step depends only on current and historical data; multiple time blocks use an increasing expansion rate to cover a longer historical range with fewer network layers. Specifically, each time block may include one-dimensional convolution, ReLU activation, batch normalization, dropout layers, and residual connections.

[0019] The high-dimensional features output by the temporal convolutional network are flattened or pooled and then input into a multilayer perceptron. The multilayer perceptron uses several fully connected layers to regress the features into obstacle distances.

[0020] The time-series distance estimation model is trained under supervision using flight data labeled with the actual distances to obstacles. The training objective is to minimize the mean square error between the estimated distance and the actual distance. The flight data includes the continuous flight status of the UAV under different obstacle distances, flight speeds, and flight directions.

[0021] During supervised training, the actual distance is obtained synchronously. As labels, the network parameters are updated by minimizing the mean squared error: Formula (3) In an alternative embodiment, the state sequence length The value is set to 20, the sampling interval is 10 ms, corresponding to 200 ms of historical information; the temporal convolutional network is set with four hidden layers, each with 16 channels, and ReLU, batch normalization and dropout layer with a dropout rate of 10% are set after each hidden layer; the hidden layer widths of the multilayer perceptron are 64, 32 and 32 respectively.

[0022] It should be noted that the specific values ​​mentioned above are for illustrative purposes only and can be adjusted according to the actual application, taking into account the size of the drone, control frequency, rotor response speed, and computing resources.

[0023] Training data should cover scenarios including free flight, gradually approaching an obstacle, flying parallel to an obstacle, and moving away from an obstacle, and include various speeds, attitudes, and obstacle directions. Actual distances can be obtained during the training phase using a motion capture system, laser rangefinder, or other external measuring devices; after training is complete, this external measuring device is not required during the online operation phase.

[0024] Step S3: Calculate the nominal translational acceleration based on the reference trajectory generated by the trajectory generation module and the current position of the UAV, and construct a safety set and exponential control obstacle function constraints based on the estimated distance; In step S3, the nominal translational acceleration is calculated based on the UAV's position tracking error, velocity tracking error, reference trajectory acceleration, and gravity compensation term.

[0025] Specifically, based on the desired location Expected speed and expected acceleration Calculate position error and speed error The nominal translational acceleration can be expressed as: Formula (4) in, and It is a positive definite gain matrix.

[0026] Let the preset safety distance be To estimate distance Construct a safety function and define the security set. .

[0027] For a distance constraint of relative order 2, the exponential control barrier function constraint condition is constructed as follows: Formula (5) in, , The positive parameter is chosen to ensure the stability of the corresponding characteristic polynomial, and its value is such that the corresponding characteristic polynomial is stable. and These represent the first and second derivatives of the security function with respect to time, respectively.

[0028] Step S4: Under the constraints of the exponential control barrier function, solve the quadratic programming problem with the objective of minimizing the correction amount of the nominal translational acceleration to obtain the safe translational acceleration; In the embodiments of this specification, the quadratic programming problem is expressed as: under the condition of satisfying the exponential control barrier function constraint, such that... Minimum, of which The nominal translational acceleration, The value is the safe translational acceleration to be solved; the quadratic programming problem is solved in a rolling manner according to a preset control cycle.

[0029] Specifically, the exponential control obstacle function constraint, combined with the UAV translational dynamics, is transformed into a function relating the unknown safe translational acceleration. The affine inequality is used to solve the following quadratic programming problem in each control cycle: Formula (6) The objective function aims to make the safe translational acceleration as close as possible to the nominal translational acceleration. Therefore, when the distance between the UAV and the obstacle is sufficient, the safety filter does not change or only slightly alters the trajectory control commands. When the estimated distance approaches the safety boundary, the constraint is activated, and the safety filter generates acceleration components away from the obstacle. To handle estimation errors, the safety distance can be increased with a confidence margin, or adaptively adjusted based on the upper bound of the estimation model's error on the validation set. .

[0030] Step S5: Convert the safe translational acceleration into total thrust command and desired attitude command to control the UAV to move away from the obstacle.

[0031] In step S5, the total thrust command and the desired attitude command are calculated based on the safe translational acceleration and the preset yaw angle, and the total thrust command and the desired attitude command are converted into the rotational speed commands of each rotor through the attitude controller, the angular velocity controller and the control distribution module.

[0032] Specifically, safe translational acceleration Used to calculate the required total thrust direction and magnitude, and to generate the desired attitude. The attitude controller and angular velocity controller generate control torque based on the desired attitude, desired angular velocity, and current state. The control distribution module then maps the total thrust and control torque into rotor speed commands.

[0033] To verify the effectiveness of the method proposed in this patent, the following transparent static and dynamic obstacle avoidance tests were conducted.

[0034] like Figure 2 As shown, multiple transparent acrylic boxes are arranged near a reference trajectory. A drone approaches each box sequentially along a predetermined trajectory, estimating the distance using its onboard flight status and correcting the nominal input with a safety filter. In the experiment, without using cameras or lidar to directly identify the transparent boxes, the drone can generate a safety control component to move away from obstacles when approaching them.

[0035] In a comparative embodiment, a multilayer perceptron without temporal convolution feature extraction combined with an artificial potential field method was used as the benchmark method. Under three box-like conditions, the maximum distance estimation error of this embodiment reached approximately 0.133 m, 0.162 m, and 0.108 m, respectively, corresponding to root mean square errors of approximately 0.048 m, 0.051 m, and 0.035 m; compared to the benchmark method, the average distance estimation error was reduced. This result demonstrates that temporal features based on historical states are helpful in identifying near-surface aerodynamic responses.

[0036] As maximum flight speed increases, the response time allowed for the safety filter decreases. Success rates at high speeds can be improved by expanding the training range, increasing the control frequency, increasing the safety distance, and introducing estimation uncertainty margins.

[0037] like Figure 3 As shown, the drone is hovered at a height of approximately 0.8 m, and a transparent enclosure mounted on a mobile platform is gradually brought closer to the drone. When the estimated distance decreases to near the safety boundary, the safety constraint is activated, and the drone retreats in a direction away from the enclosure; after the obstacle stops approaching, the drone maintains the new safe position.

[0038] In one experimental example, with a preset safe distance of 0.15 m, the UAV retreated a cumulative distance of approximately 0.7 m, and the root mean square error of the distance estimation was approximately 0.026 m. The exponential control obstacle function value remained non-negative during the experiment. The above values ​​are used to illustrate the feasibility of the invention. Specific systems can be configured with safe distances and control parameters according to the aircraft size, rotor diameter, obstacle shape, and mission requirements.

[0039] Furthermore, the temporal distance estimation model is not limited to the number of layers and channels mentioned above. It can also employ gated temporal convolution, lightweight recurrent neural networks, temporal attention networks, or combinations thereof, as long as it can extract near-face airflow features from a time series containing rotor speed and flight status. The distance output can be a scalar minimum distance, a distance vector in the body coordinate system, distances in multiple directions, or a distance distribution.

[0040] For multi-directional obstacle perception, feature branches can be constructed for each rotor or airframe direction, and distance estimates in the forward, backward, left, right or up and down directions can be output. Based on this, the safety filter constructs multiple control obstacle function constraints and solves them jointly in the same quadratic programming.

[0041] Furthermore, assume that the quaternions for the drone's position, velocity, attitude, and angular velocity are respectively... , , and The quality is The inertia matrix is The rigid body dynamics of a quadcopter drone can be expressed as: Formula (7) in, The rotation matrix is ​​obtained from the attitude quaternions. As the thrust shaft of the aircraft, To control the torque, Given gravitational acceleration, the reference thrust generated by a single rotor in free flight is denoted as . .

[0042] When the distance between the rotor and the surface of the obstacle is The characteristic radius of the rotor is When, the dimensionless spacing is The effective thrust in the near-surface state can be written as ,in This is the near-surface correction function that varies with the spacing.

[0043] During closed-loop flight, changes in the near-surface correction function cause the flight control system to continuously compensate for rotor speed, attitude, and velocity. A single state variable at a single moment is unlikely to uniquely correspond to obstacle distance, while the coupled changes of multiple state variables over a period of time contain dynamic characteristics that can be used to estimate distance. Therefore, this invention does not require explicit identification. The mapping is learned from continuous flight data using a temporal network.

[0044] In another application embodiment, a quadcopter platform weighing approximately 1.65 kg is used. The flight controller runs open-source flight control software and obtains rotor speed feedback through a two-way digital electronic speed controller (ESC) protocol. The inertial measurement unit outputs acceleration and angular velocity at 200 Hz, rotor speed is read at 100 Hz, and the onboard computer runs a distance estimation network and safety filter at 100 Hz. Position and velocity can be provided by motion capture, visual inertial odometry, satellite navigation, or other positioning modules.

[0045] In the offline training example, the collected data is divided into training and validation / test sets at 75% and 25% respectively, and the Adam optimizer and learning rate are used. The model is trained with a batch size of 1024. During online runtime, only forward inference is retained, and distance estimation results are updated over a sliding time window. The nominal control gain and safety parameters can be adjusted according to platform dynamics, for example... , , , The above parameters do not constitute a limitation on the scope of protection of this invention.

[0046] To avoid abrupt control changes due to abnormal network output, at least one of the following can be set at the distance estimation output: physical range clipping, low-pass filtering, rate of change limiting, and confidence level assessment. When the estimated confidence level is below a threshold, the safety margin can be increased, the flight speed reduced, or the system can switch to hover / backward mode.

[0047] In addition to quadcopters, the present invention can be extended to multi-rotor aircraft with multiple rotors and capable of obtaining rotor speed or motor current feedback. The flight status may also include at least one of motor current, ESC duty cycle, airframe acceleration, thrust command, and barometric altitude to enhance the observability of the aerodynamic status.

[0048] Based on the same inventive concept, embodiments of this specification also provide a drone safety obstacle avoidance system based on near-surface airflow perception, such as... Figure 4 The diagram shown is a structural schematic of a UAV safety obstacle avoidance system based on near-surface airflow perception, provided in an embodiment of this specification.

[0049] In the embodiments described in this specification, the UAV safety obstacle avoidance system based on near-surface airflow perception may specifically include: The status acquisition module 401 is used to acquire continuous flight status data of the UAV during flight, and the continuous flight status data is constructed into a fixed-length status data sequence according to the time sequence. The distance estimation module 402 is used to input the state data sequence into a pre-trained temporal distance estimation model, extract the temporal features of rotor-induced airflow caused by the near-face effect of obstacles, and output the estimated distance of the UAV relative to the obstacles. The nominal control module 403 is used to calculate the nominal translational acceleration based on the reference trajectory generated by the trajectory generation module and the current position of the UAV, and to construct a safety set and exponential control obstacle function constraints based on the estimated distance; The safety filtering module 404 is used to solve a quadratic programming problem under the constraints of the exponential control barrier function, with the objective of minimizing the correction amount of the nominal translational acceleration, to obtain the safety translational acceleration; The flight execution module 405 is used to convert the safe translational acceleration into total thrust command and desired attitude command to control the UAV to move away from the obstacle.

[0050] Specifically, the state acquisition module includes an inertial measurement unit, a flight controller, a speed information acquisition interface, and a rotor speed information acquisition interface; the system, without relying on cameras or lidar to directly measure obstacles, utilizes the changes in flight state caused by the disturbance of the obstacle surface by the rotor-induced airflow to complete distance estimation and achieve obstacle avoidance.

[0051] The specific application examples of the UAV safety obstacle avoidance system based on near-surface airflow perception are the same as the corresponding method examples described above, and will not be repeated here.

[0052] The above description is merely a preferred embodiment of the present invention. For those skilled in the art, any equivalent substitutions, parameter adjustments, and module combinations made without departing from the concept of the present invention should fall within the protection scope of the present invention.

Claims

1. A method for safe obstacle avoidance of unmanned aerial vehicles (UAVs) based on near-surface airflow perception, characterized in that, Includes the following steps: Step S1: Collect continuous flight status data of the UAV during flight, and construct a fixed-length status data sequence according to the time sequence of the continuous flight status data; Step S2: Input the state data sequence into the pre-trained temporal distance estimation model, extract the temporal features of rotor-induced airflow caused by the near-face effect of obstacles, and output the estimated distance of the UAV relative to the obstacles; Step S3: Calculate the nominal translational acceleration based on the reference trajectory generated by the trajectory generation module and the current position of the UAV, and construct a safety set and exponential control obstacle function constraints based on the estimated distance; Step S4: Under the constraints of the exponential control barrier function, solve the quadratic programming problem with the objective of minimizing the correction amount of the nominal translational acceleration to obtain the safe translational acceleration; Step S5: Convert the safe translational acceleration into total thrust command and desired attitude command to control the UAV to move away from the obstacle.

2. The UAV obstacle avoidance method based on near-surface airflow perception according to claim 1, characterized in that, The state data sequence in step S1 is represented as follows: ,in At least linear velocity, attitude quaternion, airframe angular velocity, and rotor speeds. The length of the sliding time window; flight statuses from different sources are synchronized and normalized according to a unified timestamp.

3. The UAV safety obstacle avoidance method based on near-surface airflow perception according to claim 1, characterized in that, The temporal distance estimation model in step S2 includes a temporal convolutional network and a multilayer perceptron connected in sequence; the temporal convolutional network includes at least two causal convolutional temporal blocks, which employ dilated convolution to expand the temporal receptive field and include at least one of nonlinear activation, batch normalization, dropout layers, and residual connections; the multilayer perceptron maps the high-dimensional temporal features output by the temporal convolutional network to the estimated distance.

4. The UAV obstacle avoidance method based on near-surface airflow perception according to claim 3, characterized in that, The time-series distance estimation model is trained under supervision using flight data labeled with the actual distances to obstacles. The training objective is to minimize the mean square error between the estimated distance and the actual distance. The flight data includes the continuous flight status of the UAV under different obstacle distances, flight speeds, and flight directions.

5. The method according to claim 1, characterized in that, The security set is defined as Among them, the security function , For the estimated distance, The preset safety distance; the exponential control barrier function constraint is as follows: ,in and Positive parameters that ensure the stability of the corresponding characteristic polynomial. and These represent the first and second derivatives of the security function with respect to time, respectively.

6. The method according to claim 1, characterized in that, The quadratic programming problem is expressed as: under the condition of satisfying the exponential control barrier function constraint, such that... Minimum, of which The nominal translational acceleration, The safety translational acceleration to be solved; The quadratic programming problem is solved on a rolling basis according to a preset control cycle.

7. The method according to claim 1, characterized in that, In step S3, the nominal translational acceleration is calculated based on the UAV's position tracking error, velocity tracking error, reference trajectory acceleration, and gravity compensation term. In step S5, the total thrust command and the desired attitude command are calculated based on the safe translational acceleration and the preset yaw angle, and the total thrust command and the desired attitude command are converted into the rotational speed commands of each rotor through the attitude controller, the angular velocity controller and the control distribution module.

8. A UAV safety obstacle avoidance system based on near-surface airflow perception, characterized in that, include: The status acquisition module is used to collect continuous flight status data of the UAV during flight, and the continuous flight status data is constructed into a fixed-length status data sequence according to the time sequence; The distance estimation module is used to input the state data sequence into a pre-trained temporal distance estimation model, extract the temporal features of rotor-induced airflow caused by the near-face effect of obstacles, and output the estimated distance of the UAV relative to the obstacles. The nominal control module is used to calculate the nominal translational acceleration based on the reference trajectory generated by the trajectory generation module and the current position of the UAV, and to construct a safety set and exponential control obstacle function constraints based on the estimated distance; The safety filtering module is used to solve a quadratic programming problem under the constraints of the exponential control barrier function, with the objective of minimizing the correction amount of the nominal translational acceleration, to obtain the safety translational acceleration; The flight execution module is used to convert the safe translational acceleration into total thrust commands and desired attitude commands to control the UAV to move away from obstacles.

9. The UAV safety obstacle avoidance system based on near-surface airflow perception according to claim 8, characterized in that, The state acquisition module includes an inertial measurement unit, a flight controller, a speed information acquisition interface, and a rotor speed information acquisition interface. The system can estimate distance and achieve obstacle avoidance by utilizing the changes in flight state caused by the disturbance of the obstacle surface by the rotor-induced airflow without relying on cameras or lidar to directly measure obstacles.

Images