Unmanned aerial vehicle landing control method and system based on visual tactile technology

Abstract

The invention provides an unmanned aerial vehicle landing control method and system based on a visual tactile technology, and relates to the technical field of unmanned aerial vehicle intelligent control. Outputting a visual feature set including a slope angle and a material type and a tactile feature set including a stiffness coefficient, a friction coefficient and a frequency spectrum entropy value; performing multi-level calculation by combining the slope angle, the material type and the spectrum entropy value to obtain a safety threshold value; generating a tactile deviation through searching based on the material type, and generating a risk alarm instruction and a risk category when the tactile deviation exceeds a safety threshold; executing hierarchical control according to the risk alarm instruction and the risk category; after hierarchical control, the change rate of the rigidity coefficient and the change rate of the friction coefficient in the tactile feature set are monitored in real time, successful landing is judged when preset conditions are met, and full-process safety control over landing of the unmanned aerial vehicle under the complex terrain is achieved through deep fusion of visual tactile multi-mode sensing and dynamic risk assessment.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle (UAV) intelligent control technology, and in particular to a UAV landing control method and system based on visual-tactile technology. Background Art

[0002] Traditional drone landing control relies primarily on single-mode vision or lidar technology, which presents significant limitations in complex terrain (such as sandy areas, gravel slopes, and vegetated areas). Visual sensors are susceptible to interference from lighting and occlusion, and cannot directly perceive the physical properties of the terrain. Pure mechanical feedback control lacks the ability to predict terrain material, leading to slippage, rollovers, and other accidents during landing due to insufficient friction or sudden changes in stiffness. Existing multimodal perception fusion technologies often focus on data-level splicing, lacking quantitative modeling of physical properties (such as stiffness and friction coefficient). Furthermore, safety thresholds are often statically set, making them incapable of adapting to dynamic terrain changes.

[0003] Therefore, it is necessary to provide a UAV landing control method and system based on visual-tactile technology to solve the above technical problems. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a drone landing control method and system based on visual-tactile technology. Through multimodal perception, feature fusion and adaptive control strategy, it solves the problems of dynamic instability and slip risk during drone landing in complex terrain. It is suitable for scenarios such as drone logistics distribution, emergency rescue, and terrain survey.

[0005] The present invention provides a method for controlling the landing of a UAV based on visual-tactile technology, the method comprising the following steps: The visual data stream and tactile data stream are collected by the onboard visual sensor and the onboard tactile sensor array respectively, and are processed synchronously in time and space; Feature extraction is performed on the synchronously processed visual data stream and tactile data stream respectively, and a visual feature set including slope angle and material type and a tactile feature set including stiffness coefficient, friction coefficient and spectral entropy value are output; Perform multi-level calculations based on the slope angle, material type, and spectrum entropy value to obtain a safety threshold; Based on the material type, generating a tactile deviation by querying a preset terrain physical database, and generating a risk warning instruction and a risk category when the tactile deviation exceeds the safety threshold; executing hierarchical control according to the risk warning instruction and the risk category, wherein the hierarchical control includes at least one of vertical speed adaptation, dynamic adjustment of admittance parameters, and safe altitude increase; After the hierarchical control, the change rate of the stiffness coefficient and the change rate of the friction coefficient in the tactile feature set are monitored in real time, and a successful landing is determined when preset conditions are met.

[0006] Preferably, the feature extraction of the visual feature set includes: Performing terrain analysis on the spatiotemporally synchronized visual data stream, wherein the terrain analysis includes: A random sampling consensus algorithm is used to fit the ground plane of the landing area, the angle between the normal vector of the ground plane and the vertical direction is calculated, and the angle is converted using an inverse cosine function to obtain the slope angle; Use lightweight convolutional neural networks to classify the materials of the collected RGB images and output the material type; The slope angle and material type are combined into a visual feature set.

[0007] Preferably, the feature extraction of the tactile feature set includes: The stiffness coefficient is calculated by linearly fitting the force-displacement curve of the contact process using the least squares method; The dynamic ratio of the tangential force to the normal force is calculated within a preset sliding window, and the dynamic ratio within the preset window is weighted averaged using a time exponential to obtain the friction coefficient; Band-limited spectrum analysis is performed on the acceleration signal, and the Shannon entropy is calculated based on the spectrum energy distribution obtained by the analysis as the spectrum entropy value.

[0008] Preferably, the step of performing multi-level calculations based on the slope angle, material type, and spectrum entropy value to obtain a safety threshold comprises: Based on the slope angle and material type, query a predefined safety reference database to obtain a safety reference value corresponding to the slope angle and material type; Based on the spectrum entropy value, calculating the adjustment factor by a preset piecewise linear function; The safety reference value and the adjustment factor are linearly combined to calculate a safety threshold.

[0009] Preferably, the step of generating a tactile deviation based on the material type by querying a preset terrain physical database, and generating a risk warning instruction and a risk category when the tactile deviation exceeds the safety threshold, includes: Querying a preset terrain physical database to obtain a standard stiffness coefficient range and a standard friction coefficient range corresponding to the material type; Calculating a stiffness deviation rate and a friction deviation rate based on the standard stiffness coefficient range and the standard friction coefficient range, and taking the larger value of the stiffness deviation rate and the friction deviation rate as the tactile deviation; Risk grading is performed based on the tactile deviation and the safety threshold to generate risk warning instructions and risk categories, wherein the risk warning instructions include a first-level warning and a second-level warning, and the risk categories include dynamic instability and slip risks corresponding to the first-level warning and the second-level warning, respectively.

[0010] Preferably, the execution of the hierarchical control includes: Select a control combination method according to the risk category: when the risk category is dynamic instability, simultaneously perform safety altitude increase, dynamic adjustment of admittance parameters and vertical speed adaptation; When the risk category is slip risk, dynamic adjustment of admittance parameters and vertical speed adaptation are performed simultaneously; The adjustment range is set based on the risk warning instruction level, wherein the safety height improvement includes: In the first level warning, the safety altitude threshold is increased by the first offset; In the second level warning, the safety height threshold is increased by a second offset, wherein the first offset is a preset multiple of the second offset; The dynamic adjustment of the admittance parameters includes: For dynamic instability: increase the admittance damping coefficient to a first percentage range of the reference value, and reduce the inertia coefficient to a second percentage range of the reference value; To address the risk of slippage: increase the admittance damping coefficient to the third percentage range of the baseline value; The vertical speed adaptation includes: Under the first level warning, the maximum allowable descent speed is reduced to the first preset ratio of the preset safety speed; Under the second level warning, the maximum allowable descent speed is reduced to a second preset ratio of the preset safety speed.

[0011] Preferably, the real-time monitoring of the change rate of the stiffness coefficient and the change rate of the friction coefficient in the tactile feature set and determining successful landing when preset conditions are met includes: Based on the real-time tactile feature set after hierarchical control, perform: The rate of change of the stiffness coefficient per unit time is calculated by the time difference method; The rate of change of friction coefficient per unit time is calculated by time difference method; Landing is considered successful when the following conditions are met simultaneously: The absolute value of the change rate of the stiffness coefficient is continuously lower than the preset stiffness change threshold; The absolute value of the rate of change of the friction coefficient is continuously lower than the preset friction change threshold; The time during which the above two conditions are maintained together exceeds the preset stability duration threshold.

[0012] The present invention also provides a UAV landing control system based on visual-tactile technology, which is used to execute the UAV landing control method based on visual-tactile technology. The system includes: A sensor synchronization module is used to collect visual data streams and tactile data streams through an airborne visual sensor array and an airborne tactile sensor array, and perform spatiotemporal synchronization processing; A feature extraction module is used to extract features from the synchronously processed visual data stream and tactile data stream respectively, and output a visual feature set including slope angle and material type and a tactile feature set including stiffness coefficient, friction coefficient and spectral entropy value; A threshold calculation module is used to perform multi-level calculations based on the slope angle, material type and spectrum entropy value to obtain a safety threshold; a risk assessment module, configured to generate a tactile deviation based on the material type by querying a preset terrain physical database, and generate a risk warning instruction and a risk category when the tactile deviation exceeds the safety threshold; a hierarchical control module, configured to execute hierarchical control according to the risk warning instruction and the risk category, wherein the hierarchical control includes at least one of vertical speed adaptation, dynamic adjustment of admittance parameters, and safe altitude increase; The landing monitoring module is used to monitor the change rate of the stiffness coefficient and the change rate of the friction coefficient in the tactile feature set in real time after the layered control, and determine that the landing is successful when the preset conditions are met.

[0013] Compared with related technologies, the drone landing control method and system based on visual-tactile technology provided by the present invention has the following beneficial effects: This invention significantly improves the safety and robustness of drone landing on complex terrain by using visual-tactile multimodal perception fusion technology. Multimodal feature fusion: Combining visually analyzed slope angle and material type with tactilely perceived stiffness, friction coefficient, and spectral entropy, a dynamic terrain risk assessment model was constructed, overcoming the limitations of a single sensor's perception of physical properties. Adaptive safety threshold: Through piecewise linear calculation of slope angle, material type, and spectral entropy value, the safety threshold is dynamically adjusted to effectively address the risk differences in different terrains; Hierarchical control strategy: Dynamically adjusts safety altitude, admittance parameters, and vertical speed based on risk classification (dynamic instability / slip risk), balancing rapid response and stability; Determination of successful landing: Reliable judgment of landing status is ensured through real-time monitoring of the stiffness / friction coefficient change rate and stability duration threshold constraints.

[0014] Through the deep integration of visual-tactile multimodal perception and dynamic risk assessment, the present invention realizes the full-process safety control of UAV landing in complex terrain, significantly improving landing reliability and environmental adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of a UAV landing control method based on visual-tactile technology provided by the present invention; Figure 2 This is a module structure diagram of a UAV landing control system based on visual-tactile technology provided by the present invention. DETAILED DESCRIPTION

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and examples. It will be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions relevant to the present invention, not all of the structures. Furthermore, the embodiments of the present invention and the features of the embodiments may be combined with one another unless there is a conflict.

[0017] It should also be noted that, for ease of description, only portions relevant to the present invention are shown in the accompanying drawings, rather than all of the contents. Before discussing the exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the various operations (or steps) as sequential processes, many of the operations can be performed in parallel, concurrently, or simultaneously. In addition, the order of the various operations can be rearranged. The process can be terminated when its operations are completed, but may also have additional steps not included in the accompanying drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0018] Example 1 The present invention provides a method for controlling the landing of a UAV based on visual-tactile technology. Figure 1 As shown, the method includes the following steps: S1: The visual data stream and tactile data stream are collected by the airborne visual sensor and airborne tactile sensor array respectively, and then processed synchronously in time and space.

[0019] In this embodiment, the onboard visual sensor uses a global shutter binocular camera module with a resolution of 1280×800 pixels and a frame rate of 30fps. It is mounted above the central axis of the UAV's landing gear, ensuring that the angle between the optical axis and the body axis is less than 5 degrees. To adapt to low-light environments, an active near-infrared fill light system with a wavelength of 850 nanometers is configured. The tactile sensor uses a 4×4 piezoresistive unit array arranged on the bottom contact surface of the landing gear. The unit force sensing range is 0.05N to 10N (resolution 0.01N), and the dynamic response frequency covers 0-500Hz. It is equipped with a three-axis MEMS accelerometer (range ±16g) and an integrated temperature compensation circuit to achieve a temperature drift compensation rate of over 90%.

[0020] Data acquisition uses a precise clock synchronization mechanism: a synchronous pulse signal (with a time jitter of less than 1 microsecond) generated by the FPGA simultaneously triggers the binocular camera exposure and the tactile array sampling. The underlying IEEE1588v2 protocol is used to achieve cross-sensor clock synchronization. Spatial calibration includes two key steps: (1) visual-tactile coordinate system mapping, which establishes the binocular camera's external parameters based on a checkerboard calibration plate, measures the spatial position relationship between the tactile array and the body's center of mass using a laser tracker, and constructs a multi-source sensor spatial position conversion matrix; (2) dynamic compensation mechanism, which pre-stores the landing gear elastic deformation model (the deformation variable varies as a power function with the load) to correct the mechanical displacement error during landing in real time.

[0021] The visual data processing pipeline sequentially follows: demosaicing using bilinear interpolation, radiometric correction based on a grayscale calibration curve, and lens distortion correction using the Brown-Conrady model. The final output is a binocular disparity map and a microsecond timestamp. Tactile signal conditioning includes a 120Hz band-stop filter to eliminate vibration noise, and an adaptive gain amplifier circuit to dynamically adjust the amplification factor based on contact force. Tactile data and acceleration information are fused using a Kalman filter.

[0022] Spatiotemporal alignment is achieved through a dual-channel approach: the temporal dimension uses cubic spline interpolation to reconstruct the tactile data timing based on the visual frame; the spatial dimension maps the physical coordinates of the tactile units to the image plane to generate a visual-tactile spatial alignment mask. The entire processing pipeline maintains latency within 5 milliseconds, utilizing a DMA transfer mechanism to ensure real-time performance.

[0023] S2: Feature extraction is performed on the synchronously processed visual data stream and tactile data stream respectively, and a visual feature set including slope angle and material type and a tactile feature set including stiffness coefficient, friction coefficient and spectral entropy value are output.

[0024] Specifically, in step S2, the feature extraction of the visual feature set is obtained by performing terrain analysis on the spatiotemporally synchronized visual data stream, wherein the terrain analysis specifically includes the following steps: First, a random sampling consensus algorithm is used to fit the ground plane of the landing area, and the angle between the normal vector of the ground plane and the vertical direction is calculated. The angle is converted using the inverse cosine function to obtain the slope angle.

[0025] In this example, a semi-global matching algorithm is first used to calculate sub-pixel disparity based on a spatiotemporally synchronized binocular disparity map (resolution 1280×800 pixels, frame rate 30 fps). Using preset camera internal parameters (including focal length and baseline distance), the disparity values ​​are converted into 3D point cloud coordinates according to a depth reconstruction formula. A median filter (window size 5×5 pixels) is then applied to eliminate flight noise. A valid depth range of 0.3 to 3 meters is also set to filter out invalid data.

[0026] The generated 3D point cloud is fitted with the ground plane using the RANSAC algorithm: Plane model iterative optimization: Set the maximum number of iterations to 500 (confidence level 99%), and randomly select 3 points each time to calculate the plane equation coefficients; The threshold for determining inliers is 3 cm (based on the mechanical tolerance of the drone landing gear). The plane model with the largest number of inliers is retained, and the plane parameters are optimized using the least squares method based on the entire set of inliers.

[0027] Slope angle calculation: extract the normal vector of the optimized ground plane and calculate the cosine value of the angle between the normal vector and the gravity direction vector; The included angle is converted into an angle value through the arc cosine function, and the slope angle result is output.

[0028] Secondly, a lightweight convolutional neural network is used to classify the materials of the collected RGB images and output the material type.

[0029] In this embodiment, a lightweight convolutional neural network is used to perform material recognition on an RGB image, including: Use the improved MobileNetV3 network (adapted to the embedded platform) and downsample the input image to 320×240 pixels; The network contains an inverted residual structure (expansion ratio 6:1) and a channel attention module (compression ratio 16); Construct a dataset covering four typical terrain types (grass / concrete / sand / ice); Data augmentation strategies were implemented: lighting perturbations (brightness ±30%, contrast ±15%) and random occlusions to simulate shadows.

[0030] The real-time inference process includes: Preprocessing: Extract the landing area region of interest (ROI, center 200×200 pixels) based on the depth map; Histogram equalization: enhances texture features under different lighting conditions; Classification output: It takes 8 milliseconds to process a single frame on the NVIDIA Jetson Nano platform, and the material label is output when the classification confidence is higher than 0.85.

[0031] Finally, the slope angle and material type are combined into a visual feature set.

[0032] In this embodiment, the slope angle (unit: degree) and the material type label are combined into structured data, which is published at a frequency of 30 Hz through the topic communication mechanism of the Robot Operating System (ROS). The data format is encapsulated using the JSON protocol.

[0033] Specifically, in step S2, the feature extraction of the tactile feature set includes the following steps: During tactile feature extraction, a piezoresistive tactile array was used to synchronously collect normal force signals and contact displacement data at a sampling frequency of 500 Hz. Displacement measurement was assisted by a laser displacement meter with an accuracy of ±0.1 mm. The contact process was divided into a pre-load phase with a displacement of less than 2 mm and a main contact phase with a displacement of 2 mm or more. For each phase, a linear fit was performed using the least squares method to establish a linear relationship model between normal force and displacement. A 300-ms sliding data window (containing 150 sampling points) was set. A new fitting cycle was started when the absolute value of the displacement change rate was detected to be greater than 10 mm / s. Finally, the slope of the fitting straight line in the main contact stage was taken as the stiffness coefficient and converted into Newton / mm units (retaining two decimal places).

[0034] Secondly, in the friction coefficient processing, the dynamic ratio of tangential force to normal force is collected synchronously and a 200-ms timing window (10-ms step) is configured. Exponential decay weighting is applied to the friction ratio at each time point in the window (the current data weight is approximately 54%, and the historical data weight decreases exponentially). When the normal force is lower than 0.1 Newton, it is considered as invalid contact and the data is discarded, and the weighted average friction coefficient is output (retain three decimal places).

[0035] Finally, for the calculation of spectral entropy, a 20-400Hz bandpass filter was performed on the accelerometer's vertical signal to eliminate vibration interference, and a 1024-point data window (50% overlap) was standardized. The power spectrum was estimated with a 0.49Hz resolution using a windowed Fourier transform (Hanning window function). After normalizing the power spectral density to a probability distribution, the spectral uncertainty measure was calculated according to the Shannon entropy definition. In particular, the 100-200Hz characteristic frequency band was given a double weight to enhance the surface response, and the result was calibrated to a range of 0-8 bits (retaining one decimal place). A pre-stored temperature drift curve was used to compensate for ambient temperature effects in real time, and the entropy value baseline was calibrated using a 100Hz sinusoidal excitation signal at startup. The characteristic data is encapsulated in a four-tuple consisting of a timestamp (microsecond level), stiffness coefficient, friction coefficient, and spectral entropy value. Combined with a dynamic compensation mechanism (stiffness temperature drift compensation rate of 0.08% / °C), this ensures that the stiffness accuracy reaches ±5%FS, the glass surface friction repeatability error reaches ±0.02, and the white noise spectral entropy resolution reaches 0.1 bit, meeting industrial-grade accuracy requirements in silicone array testing.

[0036] S3: Perform multi-level calculations based on the slope angle, material type, and spectrum entropy value to obtain a safety threshold.

[0037] Specifically, step S3 includes the following steps: S31: Based on the slope angle and material type, query a predefined safety reference database to obtain a safety reference value corresponding to the slope angle and material type.

[0038] In step S31, when obtaining safety benchmark values, a predefined safety benchmark database is constructed, containing three fields: material type, slope range, and corresponding safety benchmark values. This database is derived from the statistical results of 500 repeated landing tests on four typical terrain types. Physically defined, it represents the maximum safety deviation tolerance for a specific material-slope combination (e.g., a 5° slope benchmark value of 0.28 for dry concrete). During real-time queries, the slope range (0-5° / 5-15° / 15-30°) is matched based on the current slope angle. The benchmark value is dynamically adjusted based on the material classification confidence level: when the confidence level is between 0.85 and 0.95, the benchmark value is increased by 5%. When the confidence level is ≥0.95, the original value is retained. Exception handling includes returning a mandatory safety value of 0.60 for slopes greater than 30°. Unmatched records return a default value of 0.35 and an unknown material warning.

[0039] S32: Based on the spectrum entropy value, calculate an adjustment factor using a preset piecewise linear function.

[0040] In this embodiment, the adjustment factor calculation in step S32 utilizes a three-stage rule for spectral entropy input: when the entropy value is between 0 and 2 bits (rigid terrain), a fixed adjustment factor of 0.8 is output; when the entropy value is between 2.0 and 4.5 bits (transitional terrain), linear interpolation is performed using the formula 0.8 + 0.1 × (current entropy value - 2); and when the entropy value exceeds 4.5 bits (soft terrain), an upper limit of 1.0 is output. A dynamic compensation mechanism is embedded in the implementation: the entropy value is temperature-corrected based on temperature sensor data (a 0.1% compensation is applied for every 25°C ± 1°C deviation from the baseline temperature). Entropy updates are frozen when Z-axis vibration acceleration > 0.5g is detected to prevent misjudgment.

[0041] S33: Linearly combine the safety reference value and the adjustment factor to calculate a safety threshold.

[0042] The safety threshold synthesis in step S33 is performed using a linear combination operation: safety threshold = baseline weight factor × safety baseline value + adjustment factor weight factor × adjustment factor. Weight allocation uses a dual-mode strategy: in normal terrain, the baseline weight is set to 0.7 and the adjustment factor weight is set to 0.3. When the terrain slope is greater than 15° or the spectral entropy value is greater than 5 bits, the high-risk mode is switched, the baseline weight is reduced to 0.6, and the adjustment factor weight is increased to 0.4.

[0043] The calculation results are processed with triple constraints: the threshold validity range is set to 0.22-0.55, the proximity protection mandatory upper limit of 0.45 is activated when the drone altitude is lower than 5 meters, and finally the output is smoothed by a first-order lag filter (historical value accounts for 80% and current value accounts for 20%).

[0044] S4: Based on the material type, a tactile deviation is generated by querying a preset terrain physical database, and a risk warning instruction and a risk category are generated when the tactile deviation exceeds the safety threshold.

[0045] Specifically, step S4 includes the following steps: S41: querying a preset terrain physical database to obtain a standard stiffness coefficient range and a standard friction coefficient range corresponding to the material type.

[0046] In step S41, during the terrain physical database query, a data structure is constructed containing material type, standard stiffness coefficient range, standard friction coefficient range, and environmental sensitivity coefficient. This data is derived from laboratory material testing (200×200mm specimens with 0-100N pressure loading) and 200 field slip measurements under 10 temperature and humidity conditions. During real-time queries, the reference range is dynamically adjusted based on temperature and humidity sensor data (for example, wet grass at 5°C activates an environmental sensitivity coefficient multiplier of 1.25). A database version check (CRC32 algorithm) is performed every 30 seconds, and the wet state parameter table is automatically switched when precipitation is detected. Typical parameters include a standard stiffness range of 10.5-14.2 N / mm for concrete in an environment of 5-35°C, and a friction coefficient range of 0.75-0.85.

[0047] S42: Calculating and obtaining a stiffness deviation rate and a friction deviation rate based on the standard stiffness coefficient range and the standard friction coefficient range, and taking a larger value of the stiffness deviation rate and the friction deviation rate as the tactile deviation.

[0048] The tactile deviation calculation in step S42 first generates dynamic baseline values: the median of the standard stiffness coefficient range is used as the stiffness baseline value, and the median of the friction coefficient range is used as the friction baseline value. For non-uniform materials (such as sand), a 3×3 neighborhood spatial mean filter is used to optimize the baseline values. The stiffness deviation rate (the absolute percentage deviation between the real-time stiffness and the baseline value) and the friction deviation rate (the absolute percentage deviation between the real-time friction coefficient and the baseline value) are calculated in real time, and the larger value is taken as the comprehensive tactile deviation.

[0049] S43: Perform risk grading based on the tactile deviation and the safety threshold to generate risk warning instructions and risk categories, wherein the risk warning instructions include a level one warning and a level two warning, and the risk categories include dynamic instability and slip risks corresponding to the level one warning and the level two warning, respectively.

[0050] The risk grading process in step S43 establishes a three-level decision-making mechanism: A Level 1 warning and dynamic instability risk classification are generated when the comprehensive tactile deviation exceeds the product of the margin coefficient and the safety threshold; a Level 2 warning and slip risk classification are generated when the deviation falls between 70% and 100% of the product of the margin coefficient and the safety threshold. The margin coefficient is dynamically configured based on the material and environment (e.g., 1.25 for rainy and foggy concrete, 0.6 for cold ice surfaces), and a multi-source verification mechanism is embedded. A spectral entropy value greater than 6 bits elevates the alert to a low-risk level, and a slope angle greater than 15° mandates a dynamic instability risk.

[0051] S5: Execute hierarchical control according to the risk warning instruction and the risk category, wherein the hierarchical control includes at least one of vertical speed adaptation, dynamic adjustment of admittance parameters, and safe altitude increase.

[0052] Specifically, in step S5, the execution of hierarchical control includes: First, a control combination method is selected according to the risk category: when the risk category is dynamic instability, safety height increase, dynamic adjustment of admittance parameters and vertical speed adaptation are performed simultaneously.

[0053] Secondly, when the risk category is slip risk, dynamic adjustment of admittance parameters and vertical speed adaptation are performed simultaneously.

[0054] Next, an adjustment range is set based on the risk warning instruction level, wherein the safety height improvement includes: In the first level warning, the safety altitude threshold is increased by the first offset; Under the second-level warning, the safety height threshold is increased by a second offset, and the first offset is a preset multiple of the second offset.

[0055] The dynamic adjustment of the admittance parameters includes: For dynamic instability: increase the admittance damping coefficient to a first percentage range of the reference value, and reduce the inertia coefficient to a second percentage range of the reference value; To address the risk of slippage: Increase the admittance damping coefficient to the third percentage range of the baseline value.

[0056] The vertical speed adaptation includes: Under the first level warning, the maximum allowable descent speed is reduced to the first preset ratio of the preset safety speed; Under the second level warning, the maximum allowable descent speed is reduced to a second preset ratio of the preset safety speed.

[0057] In the hierarchical control implementation of step S5, a control combination is first selected based on the risk category: if the risk category is dynamic instability (slope > 15° or stiffness deviation > 45%), the three-module coordinated control (refresh period 50ms) of safe height increase, dynamic adjustment of admittance parameters, and vertical speed adaptation is simultaneously executed; If there is a risk of slip (friction deviation > 30%), the dual-module cross-coupling control of admittance regulation and speed adaptation is executed, with admittance regulation having a higher priority.

[0058] The safety height is increased using a graded offset strategy: the first-level warning calculates the height increment through the material sensitivity coefficient (such as the incremental coefficient of 0.15 in a concrete environment), sets the new safety height as the sum of the current height and the dynamic offset (upper limit constraint is 80 meters), and the offset is reduced by 30% when the height is less than 3 meters from the ground to prevent ground rebound; the second-level warning offset is reduced by the same coefficient and the first-level / second-level offset ratio is guaranteed to be ≥1.5.

[0059] The admittance parameter adjustment is dynamically configured for different risks: in the dynamic instability state, the damping coefficient increment is calculated based on the tactile deviation (for example, it is increased to 130% of the baseline value when the deviation is 0.4), and the inertia coefficient is simultaneously reduced to 75% of the baseline value. The parameter change rate is limited to ±20% / second to prevent actuator saturation; in the slip risk state, the damping coefficient is only increased to 115%-130% of the baseline value (linearly adjusted according to the friction deviation).

[0060] Vertical speed adaptively performs gradient suppression: the first-level warning reduces the maximum descent speed to 50% of the preset value and starts a 0.5g deceleration curve; the second-level warning reduces it to 70% and executes a 0.3g deceleration curve; when the safety altitude increase is activated, an additional 20% speed suppression amount is applied to achieve double protection.

[0061] S6: After the hierarchical control, the rate of change of the stiffness coefficient and the rate of change of the friction coefficient in the tactile feature set are monitored in real time, and a successful landing is determined when preset conditions are met.

[0062] Specifically, step S6 includes the following steps: S61: Based on the real-time tactile feature set after hierarchical control, execute: The rate of change of the stiffness coefficient per unit time is calculated by the time difference method; The rate of change of the friction coefficient per unit time is calculated by the time difference method.

[0063] In step S61, the haptic parameter change rate calculation is performed based on the real-time haptic feature set (updated at 30Hz) after hierarchical control. For the stiffness coefficient, the rate of change is calculated using first-order central differences within a 300ms time window (9 sampling points): the difference between the current stiffness value and the value 0.3 seconds prior, divided by the time interval. For the friction coefficient change rate, backward differences are used within a 500ms window (15 sampling points) and output as percentages per minute (with a unit conversion factor of 60 for a 33ms sampling period). After input data is validated (stiffness limited to 0.1-50 N / mm, friction coefficient limited to 0.05-1.2), invalid values ​​are replaced by a sliding mean over the preceding three seconds. The calculation incorporates median filtering (3-point window) and dynamic compensation. The stiffness change rate is corrected for temperature drift based on temperature sensor data (a 0.1% adjustment is applied for every 25°C ± 1°C deviation). When the normal force exceeds 5N, the friction change rate threshold is automatically increased by 20% to accommodate high-load conditions.

[0064] S62: Landing is considered successful when the following conditions are met simultaneously: The absolute value of the change rate of the stiffness coefficient is continuously lower than the preset stiffness change threshold; The absolute value of the rate of change of the friction coefficient is continuously lower than the preset friction change threshold; The time during which the above two conditions are maintained together exceeds the preset stability duration threshold.

[0065] In this embodiment, the landing state determination in step S62 is verified by three conditions: first, dynamically setting the material-related threshold: Map stiffness change thresholds (0.8% / s for concrete, 1.5% / s for grass, and 2.0% / s for sand) and friction change thresholds (15% / min for concrete, 30% / min for grass, and 45% / min for sand) against the current material type, and implement environmental corrections (friction threshold ×1.5 for rainfall, stiffness threshold ×1.3 for temperatures >40°C). Secondly, dual-parameter joint monitoring is performed, requiring that the absolute value of the stiffness change rate and the absolute value of the friction change rate are continuously lower than the corresponding thresholds; Finally, a time continuity verification is performed, and the stability duration threshold is determined by the material type (2 seconds for rigid ground and 3 seconds for soft ground). When both conditions are met simultaneously, the cumulative timing is started (allowing instantaneous exceeding of ≤0.5 seconds without interruption of timing). If any condition exceeds the limit, the timer is immediately cleared.

[0066] After achieving sustained stability, the aircraft enters the attitude verification phase: IMU data is read to verify that both pitch and roll angles are less than 3 degrees. If these attitude conditions are met, a successful landing signal is triggered. A timeout mechanism (a forced landing is triggered if monitoring exceeds 60 seconds) and false trigger protection (three consecutive false positives trigger a system self-check) are also included.

[0067] Example 2 The present invention also provides a UAV landing control system based on visual tactile technology, which is used to execute the UAV landing control method based on visual tactile technology. Figure 2 As shown, the system includes: The sensor synchronization module 100 is used to collect visual data streams and tactile data streams through an airborne visual sensor array and an airborne tactile sensor array, and perform spatiotemporal synchronization processing.

[0068] The feature extraction module 200 is used to extract features from the synchronously processed visual data stream and tactile data stream respectively, and output a visual feature set including slope angle and material type and a tactile feature set including stiffness coefficient, friction coefficient and spectral entropy value.

[0069] The threshold calculation module 300 is used to perform multi-level calculations based on the slope angle, material type and spectrum entropy value to obtain a safety threshold.

[0070] The risk assessment module 400 is configured to generate a tactile deviation based on the material type by querying a preset terrain physical database, and generate a risk warning instruction and a risk category when the tactile deviation exceeds the safety threshold.

[0071] The hierarchical control module 500 is configured to execute hierarchical control according to the risk warning instruction and the risk category, wherein the hierarchical control includes at least one of vertical speed adaptation, dynamic adjustment of admittance parameters, and safe altitude increase.

[0072] The landing monitoring module 600 is used to monitor the change rate of the stiffness coefficient and the change rate of the friction coefficient in the tactile feature set in real time after the hierarchical control, and determine that the landing is successful when the preset conditions are met.

[0073] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0074] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0075] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

Claims

1. A UAV landing control method based on visual tactile technology, characterized in that: The method comprises the following steps: The visual data stream and tactile data stream are collected by the onboard visual sensor and the onboard tactile sensor array respectively, and are processed synchronously in time and space; Feature extraction is performed on the synchronously processed visual data stream and tactile data stream respectively, and a visual feature set including slope angle and material type and a tactile feature set including stiffness coefficient, friction coefficient and spectral entropy value are output; Perform multi-level calculations based on the slope angle, material type, and spectrum entropy value to obtain a safety threshold; Based on the material type, generating a tactile deviation by querying a preset terrain physical database, and generating a risk warning instruction and a risk category when the tactile deviation exceeds the safety threshold; executing hierarchical control according to the risk warning instruction and the risk category, wherein the hierarchical control includes at least one of vertical speed adaptation, dynamic adjustment of admittance parameters, and safe altitude increase; After the hierarchical control, the change rate of the stiffness coefficient and the change rate of the friction coefficient in the tactile feature set are monitored in real time, and a successful landing is determined when preset conditions are met.

2. The method for controlling the landing of a UAV based on visual-tactile technology according to claim 1, characterized in that: The feature extraction of the visual feature set includes: Performing terrain analysis on the spatiotemporally synchronized visual data stream, wherein the terrain analysis includes: A random sampling consensus algorithm is used to fit the ground plane of the landing area, the angle between the normal vector of the ground plane and the vertical direction is calculated, and the angle is converted using an inverse cosine function to obtain the slope angle; Use lightweight convolutional neural networks to classify the materials of the collected RGB images and output the material type; The slope angle and material type are combined into a visual feature set.

3. The method for controlling the landing of a UAV based on visual-tactile technology according to claim 2, characterized in that: The feature extraction of the tactile feature set includes: The stiffness coefficient is calculated by linearly fitting the force-displacement curve of the contact process using the least squares method; The dynamic ratio of the tangential force to the normal force is calculated within a preset sliding window, and the dynamic ratio within the preset window is weighted averaged using a time exponential to obtain the friction coefficient; Band-limited spectrum analysis is performed on the acceleration signal, and the Shannon entropy is calculated based on the spectrum energy distribution obtained by the analysis as the spectrum entropy value.

4. The method for controlling the landing of a UAV based on visual-tactile technology according to claim 3, characterized in that: The step of performing multi-level calculations based on the slope angle, material type, and spectrum entropy value to obtain a safety threshold comprises: Based on the slope angle and material type, query a predefined safety reference database to obtain a safety reference value corresponding to the slope angle and material type; Based on the spectrum entropy value, calculating the adjustment factor by a preset piecewise linear function; The safety reference value and the adjustment factor are linearly combined to calculate a safety threshold.

5. The method for controlling the landing of a UAV based on visual-tactile technology according to claim 4, characterized in that: The step of generating a tactile deviation based on the material type by querying a preset terrain physical database, and generating a risk warning instruction and a risk category when the tactile deviation exceeds the safety threshold, includes: Querying a preset terrain physical database to obtain a standard stiffness coefficient range and a standard friction coefficient range corresponding to the material type; Calculating a stiffness deviation rate and a friction deviation rate based on the standard stiffness coefficient range and the standard friction coefficient range, and taking the larger value of the stiffness deviation rate and the friction deviation rate as the tactile deviation; Risk grading is performed based on the tactile deviation and the safety threshold to generate risk warning instructions and risk categories, wherein the risk warning instructions include a first-level warning and a second-level warning, and the risk categories include dynamic instability and slip risks corresponding to the first-level warning and the second-level warning, respectively.

6. The method for controlling the landing of a UAV based on visual-tactile technology according to claim 5, characterized in that: The execution of the hierarchical control includes: Select a control combination method according to the risk category: when the risk category is dynamic instability, simultaneously perform safety altitude increase, dynamic adjustment of admittance parameters and vertical speed adaptation; When the risk category is slip risk, dynamic adjustment of admittance parameters and vertical speed adaptation are performed simultaneously; The adjustment range is set based on the risk warning instruction level, wherein the safety height improvement includes: In the first level warning, the safety altitude threshold is increased by the first offset; In the second level warning, the safety height threshold is increased by a second offset, wherein the first offset is a preset multiple of the second offset; The dynamic adjustment of the admittance parameters includes: For dynamic instability: increase the admittance damping coefficient to a first percentage range of the reference value, and reduce the inertia coefficient to a second percentage range of the reference value; To address the risk of slippage: increase the admittance damping coefficient to the third percentage range of the baseline value; The vertical speed adaptation includes: Under the first level warning, the maximum allowable descent speed is reduced to the first preset ratio of the preset safety speed; Under the second level warning, the maximum allowable descent speed is reduced to a second preset ratio of the preset safety speed.

7. The method for controlling the landing of a UAV based on visual-tactile technology according to claim 6, characterized in that: The real-time monitoring of the change rate of the stiffness coefficient and the change rate of the friction coefficient in the tactile feature set and determining successful landing when a preset condition is met includes: Based on the real-time tactile feature set after hierarchical control, perform: The rate of change of the stiffness coefficient per unit time is calculated by the time difference method; The rate of change of friction coefficient per unit time is calculated by time difference method; Landing is considered successful when the following conditions are met simultaneously: The absolute value of the change rate of the stiffness coefficient is continuously lower than the preset stiffness change threshold; The absolute value of the rate of change of the friction coefficient is continuously lower than the preset friction change threshold; The time during which the above two conditions are maintained together exceeds the preset stability duration threshold.

8. A UAV landing control system based on visual tactile technology, used to execute the UAV landing control method based on visual tactile technology according to any one of claims 1 to 7, characterized in that: The system comprises: A sensor synchronization module is used to collect visual data streams and tactile data streams through an airborne visual sensor array and an airborne tactile sensor array, and perform spatiotemporal synchronization processing; A feature extraction module is used to extract features from the synchronously processed visual data stream and tactile data stream respectively, and output a visual feature set including slope angle and material type and a tactile feature set including stiffness coefficient, friction coefficient and spectral entropy value; A threshold calculation module is used to perform multi-level calculations based on the slope angle, material type and spectrum entropy value to obtain a safety threshold; a risk assessment module, configured to generate a tactile deviation based on the material type by querying a preset terrain physical database, and generate a risk warning instruction and a risk category when the tactile deviation exceeds the safety threshold; a hierarchical control module, configured to execute hierarchical control according to the risk warning instruction and the risk category, wherein the hierarchical control includes at least one of vertical speed adaptation, dynamic adjustment of admittance parameters, and safe altitude increase; The landing monitoring module is used to monitor the change rate of the stiffness coefficient and the change rate of the friction coefficient in the tactile feature set in real time after the layered control, and determine that the landing is successful when the preset conditions are met.

Images