Unmanned aerial vehicle obstacle avoidance method and system based on inertial navigation

Abstract

The invention provides an unmanned aerial vehicle obstacle avoidance method and system based on inertial navigation, and belongs to the technical field of unmanned aerial vehicle obstacle avoidance systems. The problem that an existing unmanned aerial vehicle is insufficient in obstacle avoidance capacity under the environment that GPS signals are weak, vision is limited or a sensor detection blind area exists is solved. Acquiring broadband high-frequency dynamic response data output by the inertial navigation system of the unmanned aerial vehicle in real time; establishing or acquiring a reference high-frequency dynamic response model when the unmanned aerial vehicle is in a specific flight state and a power system working mode and is not influenced by an external obstacle; the signal is dynamically responded according to high frequency; identifying whether a specific air medium disturbance echo feature formed by disturbance generated by motion of the unmanned aerial vehicle through an external obstacle exists or not; and finally, generating and executing an obstacle avoidance flight instruction according to the estimation information. The method does not depend on a traditional external environment sensor for initial detection, physical interaction is sensed through the INS, and the autonomous obstacle avoidance capacity and safety of the unmanned aerial vehicle in the complex and challenging environment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of UAV obstacle avoidance systems, and specifically to a UAV obstacle avoidance method and system based on inertial navigation. Background Art

[0002] Currently, UAV obstacle avoidance mainly relies on sensors such as GPS, vision, lidar (LiDAR), ultrasonic, or infrared to sense the environment. However, these technologies have obvious limitations in specific scenarios: GPS fails indoors, in canyons, or when interfered; vision is vulnerable to light, weather, and environmental texture, and has a large amount of computation; although LiDAR has high accuracy, its performance degrades in bad weather such as rain, fog, and dust, it has high cost, and it is difficult to detect specific materials; ultrasonic and infrared sensors have short working distances and are easily interfered by the environment.

[0003] Currently, wireless UAVs highly rely on GPS and other signals for positioning and obstacle avoidance, while the passability of tethered UAVs (optical fiber, cable) has significant problems. When UAVs perform tasks in complex and sensor - restricted environments such as indoors, thick vegetation, tunnels, bad weather, or strong interference, the reliability of the above - mentioned traditional obstacle avoidance methods will be significantly reduced or even completely fail. This technical bottleneck severely restricts the application scope and operation safety of UAVs. Summary of the Invention

[0004] Technical Problems to be Solved In view of the deficiencies of the prior art, the present invention provides a UAV obstacle avoidance method and system based on inertial navigation, which solves the problems of the prior art.

[0005] Technical Solutions

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: A UAV obstacle avoidance method and system based on inertial navigation, including the following steps: Sp1. Drive the power system of the UAV to make it move, and simultaneously collect the broadband high - frequency dynamic response data output by the inertial navigation system. The broadband high - frequency dynamic response data includes three - axis high - frequency linear acceleration signals; Sp2. Establish or obtain a reference high - frequency dynamic response model of the UAV in a specific flight state and power system working mode, considering only its own movement and interaction with the free - space air medium; Sp3. According to the current UAV flight state and power system working mode parameters, use the reference high - frequency dynamic response model to predict the current expected high - frequency dynamic response signal of the UAV in real time; Sp4. Compare the real - time high - frequency dynamic response data collected in step Sp1 with the expected high - frequency dynamic response signal predicted in step Sp3, and extract the residual high - frequency signal representing external environmental abnormal disturbances; Sp5. Analyze the residual high-frequency signal in the time domain, frequency domain, or time-frequency domain to identify whether there are specific air medium disturbance echo characteristics formed by the reflection, diffraction, or interference of the disturbances generated by the movement of the UAV itself after being reflected by external obstacles. Sp6. If the specific air medium disturbance echo characteristics are identified and the energy or correlation index of the characteristics exceeds the preset determination threshold, then it is confirmed that an obstacle is detected. Sp7. Based on the time delay information, frequency components, multi-axis signal correlation, or spatial distribution pattern of the specific air medium disturbance echo characteristics, estimate the information about the relative position, size, or shape of the obstacle. Sp8. Generate and execute an obstacle avoidance flight instruction according to the estimated obstacle information.

[0007] Preferably, the reference high-frequency dynamic response model in Sp2 is established by combining computational fluid dynamics simulation with structural dynamics analysis, or is established by data-driven acquisition through calibration flight in a known obstacle-free large space.

[0008] Preferably, in Sp5, the identification of the specific air medium disturbance echo characteristics includes: detecting a modulation signal related to the passing frequency of the UAV rotor blade or its harmonics, Doppler frequency shift characteristics, or abnormal enhancement of energy in a specific frequency band.

[0009] Preferably, the UAV obstacle avoidance method further includes: In Sp1, actively apply a small disturbance modulation signal with specific spectral characteristics to the UAV power system. In Sp5, focus on analyzing the relevant components in the residual high-frequency signal that are synchronous with or have a specific delay with the disturbance modulation signal to enhance the detection sensitivity and anti-interference ability of the specific air medium disturbance echo characteristics.

[0010] Preferably, in Sp7, the relative azimuth of the obstacle is estimated by analyzing the phase difference or amplitude ratio of the specific air medium disturbance echo characteristics on different axial sensors or different position sensors of the inertial navigation system.

[0011] Preferably, in Sp7, the relative distance of the obstacle is estimated by analyzing the time delay of the specific air medium disturbance echo characteristics relative to the initial disturbance generated by the UAV itself and combining it with the disturbance propagation speed model in the air medium.

[0012] Preferably, the preset determination threshold is dynamically adjusted according to the current environmental background noise level, UAV flight speed, or estimated value of atmospheric turbulence intensity.

[0013] Preferably, the UAV obstacle avoidance method further includes: fusing sensor data, wherein the fused sensor data includes at least one type of sensor data; The fused sensor data is used to confirm the identified echo features in step Sp6, or to calibrate the estimated obstacle information in step Sp7, or to assist in optimizing obstacle avoidance flight instructions in step Sp8, but the initial detection of the obstacle mainly depends on the analysis process based on the inertial navigation system defined in steps Sp1 to Sp6.

[0014] Preferably, the drone obstacle avoidance method further comprises: Continuously record the detected echo characteristics, estimated obstacle information and obstacle avoidance execution effect; By using these recorded data, the reference model in step Sp2, the feature recognition algorithm in step Sp5 or the obstacle information estimation algorithm in step Sp7 are iteratively optimized through an online or offline learning algorithm.

[0015] Preferably, the drone system comprises: An inertial navigation system for broadband high frequency dynamic response data, a processor and a memory storing computer executable instructions configured to be executed by the processor.

[0016] Beneficial Effects

[0017] The present invention provides a method and system for avoiding obstacles for unmanned aerial vehicles based on inertial navigation, which has the following beneficial effects: The present invention uses INS as a high-frequency "vibrator" to detect the air medium disturbance wave information excited by the UAV's own motion and modulated by external obstacles. It no longer relies on the obstacles to generate macroscopic forces or torques on the UAV, but captures more subtle and farther-propagating medium disturbance effects. This provides a potential alternative or supplementary obstacle avoidance approach based on basic physical principles in scenarios where traditional sensors are limited or fail, and achieves a breakthrough in the new technical direction of the original inertial navigation system. It achieves a breakthrough application of technology without increasing hardware costs or even reducing the original hardware costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a system architecture diagram of the present invention; Figure 2 A cloud diagram of the system composition of the present invention; Figure 3 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention. Specific Embodiment 1: INS refers to an inertial navigation system, which is a navigation system that does not rely on external information including GPS satellite signals, ground base stations, etc., and uses an inertial measurement unit to autonomously calculate the position, attitude, and speed of a carrier (including unmanned aerial vehicles, airplanes, missiles, submarines, etc.). The components include: an accelerometer - used to measure the linear acceleration of the carrier along three mutually perpendicular axes. By integrating the acceleration once, the velocity is obtained, and by integrating it twice, the displacement position change is obtained.

[0021] Gyroscope: used to measure the angular velocity of the carrier around three mutually perpendicular axes. By integrating the angular velocity, the attitude angles of the carrier, such as pitch angle, roll angle, yaw angle, etc., are obtained.

[0022] INS continuously calculates the position, velocity, and attitude of the carrier relative to the initial state by frequently reading the data of the accelerometer and gyroscope, and performing integration operations and coordinate system conversions. In the previous discussion, INS can provide high-frequency and high-precision real-time data on the motion state of the UAV itself, including acceleration, angular velocity, etc. Utilize its high-frequency response ability to capture the minute vibrations or dynamic response characteristics generated when the external environment (air disturbances caused by obstacles) acts on the UAV body, thereby indirectly detecting obstacles.

[0023] As Figures 1 - 3 shown, a UAV obstacle avoidance method and system based on inertial navigation includes the following steps: Sp1. Drive the power system of the drone to make it move and simultaneously collect the broadband high-frequency dynamic response data output by the inertial navigation system. The broadband high-frequency dynamic response data includes three-axis high-frequency linear acceleration signals. This step is the basis of the entire method. The drone power system, including the motor and the propeller, drives the drone to move according to the flight control instructions. During this process, the power system not only provides thrust but also becomes a source that continuously radiates energy to the surrounding environment, and these energy forms include sound waves, airframe vibrations, and air pressure fluctuations. The inertial navigation system installed on the airframe, especially the three-axis accelerometer inside it, needs to have broadband characteristics, and its effective measurement frequency range needs to cover the main frequency components generated by the operation of the power system and their harmonics, as well as the characteristic frequencies at which the expected obstacle echoes appear. Usually, it is required that its working bandwidth reaches at least more than 1 kHz, or even several kHz. At the same time, in order to capture weak echo signals, the noise density of the accelerometer needs to be low enough, less than several hundred μg per square root of hertz. The system synchronously collects the three-axis high-frequency linear acceleration signals output by the inertial navigation system at an extremely high sampling rate, 2 kHz to 10 kHz or higher, to form the original broadband high-frequency dynamic response data stream. It is crucial that this data collection must be precisely time-synchronized with the operating state parameters of the drone power system, such as the real-time rotational speed RPM or PWM command value of each motor, and the basic flight state parameters, such as airspeed, attitude angle, angular rate, etc. The synchronization accuracy requirement reaches the microsecond level, which can be achieved through hardware triggering or precise time protocols such as PTP to ensure that the response signal can be accurately associated with the source state that generated it during subsequent signal processing.

[0024] Sp2. Establish or obtain a reference high-frequency dynamic response model of the drone under specific flight states and power system operating modes, considering only its own motion and interaction with the free-space air medium. The goal of this step is to accurately predict or characterize the high-frequency dynamic response signals generated corresponding to the operating state of the drone itself under ideal conditions without obstacle interference. This model is the basis for subsequent extraction of abnormal signals. The establishment of the model preferably uses the following two methods: First, physics simulation-driven modeling. This method first uses computational fluid dynamics (CFD) software to simulate the complex unsteady flow fields generated by the rotation of the propeller and the flow around the airframe under different flight states (speed, angle of attack, sideslip angle) and power system operating modes (rotational speed), and accurately calculates the detailed pressure distribution acting on the airframe surface. Subsequently, this time-varying pressure distribution is used as an external load and applied to the refined UAV structural dynamics model established by the finite element method (FEM). By solving the structural dynamics equations, the three-axis high-frequency vibration acceleration response at the installation point of the inertial navigation system is obtained. This process involves a large amount of computation and is usually completed offline. The results are stored in the form of parameterized functions, high-dimensional lookup tables, or reduced-order models for online real-time calls. The model needs to consider the structural material properties, component connection methods, and the dynamic characteristics of sensor installation.

[0025] Second, data-driven modeling. This method requires full-scale calibration flights in a large space that is known to be absolutely safe and free of any reflective obstacles, such as a large anechoic chamber or an open area hundreds of meters above the ground. During the flight, the UAV needs to traverse various typical flight state combinations within its operating envelope, including different speeds, altitudes, attitudes, and different power output levels. At the same time, the complete flight parameters corresponding to the states and the high-frequency inertial navigation system data collected synchronously are recorded with high precision. Using the large amount of data collected, advanced system identification techniques are employed to construct the model. Available models include the non-linear autoregressive exogenous model (NARX), recurrent neural networks with long-term memory capabilities such as the long short-term memory network (LSTM), Gaussian process regression (GPR) that can provide uncertainty estimates, or deep neural networks (DNN) for end-to-end learning. The training objective is to establish an accurate mapping from the input vector of real-time flight and power state parameters to the output time series of the expected three-axis high-frequency acceleration signals. The data-driven model needs to pay attention to the completeness and representativeness of the training data and consider an online adaptive update mechanism to cope with environmental changes or airframe characteristic drifts.

[0026] Regardless of the method used, the final benchmark model can, based on the real-time input state matrix , which are respectively the motor rotational speed, flight speed, and flight mode parameters, and the input state matrix includes but is not limited to the above parameters, predict the benchmark high-frequency dynamic response signal .

[0027] Sp3. According to the current UAV flight state and power system working mode parameters, use the reference high-frequency dynamic response model to predict the current expected high-frequency dynamic response signal of the UAV in real time; this step runs in real time during the UAV flight. The system obtains the current UAV flight state parameters (including speed, attitude, angular rate, altitude) and power system working mode parameters (including the RPM of each motor). Take these real-time parameters as inputs and substitute them into the reference high-frequency dynamic response model established or obtained in step Sp2 . The model then outputs the expected high-frequency dynamic response signal for the current moment or the next extremely short time window . This prediction process requires efficient calculation to ensure that its output rate can match the data acquisition rate of the inertial navigation system and meet the real-time requirement

[0028] Sp4. Compare the real-time high-frequency dynamic response data collected in step Sp1 with the expected high-frequency dynamic response signal predicted in step Sp3, and extract the residual high-frequency signal characterizing the external environmental abnormal disturbance; this step aims to strip off the known part generated by the UAV's own operation from the real-time measurement signal containing all information, so as to highlight the abnormal disturbance caused by external unknown factors, especially the obstacle echo. The specific operation is to compare the high-frequency dynamic response data collected in real time in step Sp1 with the expected high-frequency dynamic response signal predicted in real time in step Sp3 . After accurately aligning them in time, perform a vector subtraction operation . The obtained is the residual high-frequency signal. Ideally, if the model is perfect and there are no external obstacles, this residual signal should be close to white noise with zero mean. In practice, it will contain model errors, unmodeled internal noise sources, atmospheric turbulence disturbances, and the obstacle echo signals we expect to detect. Therefore, the subsequent steps need to further extract the target features from this residual signal

[0029] Sp5. Analyze the residual high-frequency signal in the time domain, frequency domain or time-frequency domain to identify whether there are specific air medium disturbance echo features formed by the reflection, diffraction or interference of the disturbances generated by the UAV's own movement by external obstacles; in the signal processing link, the goal is to detect and identify the weak obstacle echo features from the background noise and interference. It is necessary to deeply analyze the residual high-frequency signal obtained in step Sp4 and use the fast Fourier transform FT or improved spectral estimation algorithm to calculate the power spectral density of the residual signal. Focus on the characteristic frequencies related to the UAV power system, especially the blade passing frequency BladePassFrequency, BPF and its harmonic frequencies 。The presence of obstacles results in abnormal energy enhancement peaks near these frequency points, or a slight frequency shift due to the Doppler effect. The BPF can be calculated from where is the number of blades. Time-frequency domain analysis: For non-stationary echo signals or scenarios where it is necessary to locate the time of their occurrence, time-frequency analysis tools such as the short-time Fourier transform (STFT), wavelet transforms such as the continuous wavelet transform (CWT) or discrete wavelet transform (DWT), or the Hilbert-Huang transform (HHT) are used. These tools can show the distribution of signal energy in time and frequency, helping to identify transient echo characteristics or the variation patterns of specific frequency components over time.

[0030] Active modulation signal analysis: If an active perturbation modulation signal with specific spectral characteristics is applied to the dynamic system in step Sp1, such as a sinusoidal RPM fluctuation with a frequency of . Then in this step, the residual signal should be analyzed for components that are synchronized with this modulation signal and the interaction frequencies (such as ) generated by it and the natural frequencies such as the BPF. Digital implementation is carried out using the principle of a lock-in amplifier, or the cross-correlation function of the residual signal and a known modulation signal template (considering the propagation delay) is calculated. The presence of high-correlation peaks or specific frequency-locked components will greatly enhance the credibility of detection and effectively suppress irrelevant noise.

[0031] Pattern recognition and machine learning: Design or train pattern recognition algorithms, detectors based on template matching, or use machine learning models including convolutional neural networks (CNN) or recurrent neural networks (RNN) to directly learn and identify complex echo signature patterns from the original segments of the residual signal or its time-frequency spectrogram.

[0032] The output of this step is a decision on the presence of echo characteristics and the extracted characteristic parameters, including the center frequency, bandwidth, energy intensity, signal-to-noise ratio (SNR), occurrence time, duration, and correlation with the active modulation signal, etc., of the echo signal.

[0033] Sp6. If specific air medium disturbance echo features are identified and the energy or correlation index of these features exceeds a preset determination threshold, then an obstacle is confirmed to be detected; this step is based on the features identified in step Sp5 for the final confirmation of the existence of an obstacle. Key quantitative indicators of the extracted echo features, such as the SNR of a specific frequency peak, the height of the cross-correlation peak, the output intensity of the matched filter, or the confidence score of the machine learning classifier, are compared with a preset determination threshold. Only when the indicator significantly exceeds the threshold is an obstacle confirmed to be detected to control the false alarm rate. The preset determination threshold should not be fixed. The preferred solution is to use a dynamic adaptive threshold. The bases for adjusting the threshold include: Real-time background noise level: By analyzing the energy level of the residual signal in the non-expected echo frequency region, the background noise power is estimated in real time , threshold setting , where is the base threshold, is the proportionality coefficient. UAV flight speed: When flying at high speed, the aerodynamic noise and structural vibration generated by itself are usually stronger, and the threshold needs to be appropriately increased. Atmospheric turbulence intensity: The turbulence intensity is estimated using low-frequency inertial data or atmospheric sensors. In a strong turbulence environment, more interference is introduced and the threshold needs to be increased. Sensor fusion confirmation: When step Sp5 initially identifies suspicious features, other low-cost, short-range sensors such as ultrasonic or infrared sensors are queried. If these sensors also report potential objects in the corresponding direction, even if the information is ambiguous, the confirmation threshold is appropriately reduced , to improve the detection sensitivity; conversely, if other sensors clearly show that there is no object in this direction, the threshold is increased to reduce false alarms. The confirmation logic is designed as "continuous confirmation", that is, it is required that the feature stably exceeds the threshold in multiple consecutive analysis windows, or "multiple feature confirmation", which requires multiple different types of echo feature indicators to meet the standards simultaneously.

[0034] Sp7. Estimate information on the relative position, size, or shape of an obstacle based on the time delay information, frequency components, multi-axis signal correlation, or spatial distribution pattern of the echo characteristics of a specific air medium disturbance; after confirming the detection of an obstacle, this step uses the echo characteristics extracted in step Sp5 to estimate the relevant information of the obstacle, providing a basis for obstacle avoidance. Single INS three-axis analysis: Analyze the distribution ratio of the energy of the confirmed echo signal on the three measurement axes X, Y, and Z of the inertial navigation system. If the energy on the Z-axis is much greater than that on the X and Y axes, it indicates that the obstacle is mainly above or below. This can only provide rough direction information. If multiple (at least two) inertial measurement units with accurately calibrated relative positions and attitudes are installed on the UAV, calculate the time difference of arrival (TDOA) or phase difference of arrival (PDoA) of the same echo signal at different units to more accurately calculate the direction of arrival of the echo, that is, the relative azimuth and pitch angles of the obstacle. Calculate the TDOA using the peak position of the cross-correlation function, or obtain the PDoA by calculating the cross-power spectrum of the signals between sensors. This requires extremely high time synchronization accuracy between sensors. For distributed INS, also analyze the amplitude difference of the same echo signal measured on different sensors, and infer the direction in combination with the signal attenuation model.

[0035] Relative distance estimation is based on echo time delay: This is the main distance estimation method. It is necessary to identify the echo characteristics (a pulse or a related peak in the residual signal) associated with a specific energy emission event of the UAV itself (the moment when a specific blade passes through a fixed reference point on the fuselage, or a specific phase point of an actively modulated signal). Measure the time difference from the emission event to the reception of the corresponding echo characteristics. . Using the known propagation speed of the air medium disturbance (the speed of sound, which needs to be corrected according to environmental parameters such as temperature), the distance to the obstacle can be estimated as . The accuracy of this method highly depends on the time delay The measurement accuracy and the precise definition of the initial emission time. In the case of continuous wave or narrowband signals, the rate of change of distance is deduced from the phase delay information rather than the absolute distance. Size or shape estimation is initially inferred by analyzing features such as the spectral width of the echo signal (larger objects reflect a wider frequency band), the variation of energy attenuation with frequency (scattering characteristics of different materials or shapes), the duration of the echo signal, or the Doppler spread, in combination with a priori databases or complex scattering models. This generally requires more complex signal processing and machine learning models. Sensor fusion calibration: If sensors such as LiDAR or structured light vision that can provide precise three-dimensional point clouds are fused, even if the data is sparse or discontinuous, they are used to calibrate or refine the azimuth and distance results estimated from the INS echo features, improving the overall estimation accuracy and reliability. This step finally outputs a state vector containing information such as the confidence level of the presence of obstacles, the estimated relative azimuth (such as horizontal angle, pitch angle), distance (or distance level, approach rate), etc.

[0036] Sp8. Generate and execute obstacle avoidance flight instructions based on the estimated obstacle information.

[0037] The reference high-frequency dynamic response model in Sp2 is established through computational fluid dynamics simulation combined with structural dynamics analysis, or data-driven established by calibrating flights in a known obstacle-free large space. After receiving the obstacle information estimated in step Sp7, this step is responsible for making decisions and executing obstacle avoidance actions.

[0038] Inputs: Obstacle state vector, the current state of the UAV (precise position, speed, attitude), predetermined mission waypoints or trajectories, flight performance constraints (maximum acceleration, angular rate, etc.).

[0039] The strategies are divided into the following cases: 1. Reactive: Simple logic and fast response. If an obstacle is within a certain distance directly ahead, execute an emergency brake or avoid upward / downward / left / right. The avoidance direction is opposite to the azimuth of the obstacle, and the avoidance amplitude is inversely proportional to the distance.

[0040] 2. Geometric planning-based: Adopt the velocity obstacle method or the ORCA method to calculate the set of velocities to avoid collisions in the velocity space and select an optimal avoidance velocity.

[0041] 3. Optimization-based: Model Predictive Control (MPC) is adopted to predict the motion trajectories of the UAV and obstacles within a finite future time domain, solve an optimization problem, and find a series of control inputs that can maximize the obstacle avoidance safety margin and minimize the path deviation while satisfying the constraints. The cost function of MPC needs to carefully design the weights to balance safety, efficiency, and comfort. This method can be developed specifically according to the actual application environment, including different flight situations of racing drones, daily entertainment drones, or functional drones, and different threshold control schemes can be developed.

[0042] The output of the decision-making logic is converted into commands acceptable to the underlying flight controller, including target attitude angles / angle rates, target velocity vectors, target accelerations, or target thrust / throttle commands, and the commands are sent to the flight controller for execution. During the obstacle avoidance process, Sp1-Sp7 are continuously run to monitor the changes in the obstacle state and dynamically adjust the avoidance strategy until the threat is determined to be lifted.

[0043] Sensor fusion optimization: The obstacle avoidance path planning makes full use of the environmental map information provided by the fusion sensors to select a more globally optimal avoidance path, detouring instead of simply stopping or turning.

[0044] In Sp5, the identification of the echo characteristics of specific air medium disturbances includes: detecting modulation signals related to the passing frequency of the UAV rotor blades or their harmonics, Doppler frequency shift characteristics, or abnormal enhancement of energy in a specific frequency band.

[0045] The UAV obstacle avoidance method also includes: In Sp1, actively apply a small disturbance modulation signal with specific spectral characteristics to the UAV power system; In Sp5, focus on analyzing the relevant components in the residual high-frequency signals that are synchronized with or have a specific delay with the disturbance modulation signal to enhance the detection sensitivity and anti-interference ability of the echo characteristics of specific air medium disturbances.

[0046] In Sp7, the relative azimuth of the obstacle is estimated by analyzing the phase difference or amplitude ratio of the echo characteristics of specific air medium disturbances on different axial sensors or different position sensors of the inertial navigation system.

[0047] In Sp7, the relative distance of the obstacle is estimated by analyzing the time delay of the echo characteristics of specific air medium disturbances relative to the initial disturbance generated by the UAV itself and combining the disturbance propagation speed model in the air medium.

[0048] The preset decision threshold is dynamically adjusted according to the current environmental background noise level, the estimated UAV flight speed, or the estimated atmospheric turbulence intensity.

[0049] The UAV obstacle avoidance method also includes: Fuse sensor data, where the fused sensor data includes at least one type of sensing data; The fused sensor data is used to confirm the identified echo features in step Sp6, or calibrate the estimated obstacle information in step Sp7, or assist in optimizing the obstacle avoidance flight instruction in step Sp8. However, the initial detection of obstacles mainly relies on the analysis process based on the inertial navigation system defined in steps Sp1 to Sp6.

[0050] The UAV obstacle avoidance method further includes: Continuously record the detected echo features, estimated obstacle information, and the execution effect of obstacle avoidance; Using these recorded data, through online or offline learning algorithms, iteratively optimize the benchmark model in step Sp2, the feature recognition algorithm in step Sp5, or the obstacle information estimation algorithm in step Sp7. Specific Embodiment Two: As Figures 1 - 3 shown, the UAV system includes: An inertial navigation system, a processor, and a memory for broadband high-frequency dynamic response data. The memory stores computer-executable instructions. When the computer-executable instructions are configured to be executed by the processor, a high-bandwidth, low-noise INS sensor needs to be equipped. The processor needs to have powerful real-time computing capabilities and requires a multi-core CPU, GPU, or FPGA to undertake complex signal processing, model prediction, and optimization calculation tasks. A high-speed data bus and sufficient memory are required. The real-time operating system RTOS is used to ensure the determinism of task scheduling. Software modular design, including data acquisition and synchronization module, benchmark model prediction module, signal processing and feature extraction module, obstacle estimation module, decision and control module, sensor fusion module, data recording module, human-computer interaction and monitoring module, etc.

[0052] The system should have the ability to record data, recording detailed data for each detection event, including the original signal, processing results, decision-making process, flight response, and task scenario information. Using these data, offline train and optimize the benchmark model, feature extractor, estimator, and decision-making logic, and even use reinforcement learning methods to optimize the obstacle avoidance strategy. Also design online real-time adjustment of the noise model and decision threshold. For a safety-critical obstacle avoidance system, failure mode analysis needs to be considered, and necessary redundancy and backup mechanisms are designed. When this INS detection method fails or has a low confidence level, it can smoothly switch to an obstacle avoidance mode based on other sensors or execute a preset safe hover / return-to-base procedure. Specific Embodiment Three: Based on the technical solutions of Specific Embodiment One and Specific Embodiment Two, further give an application description in combination with the environment: The drone needs to fly autonomously in a large warehouse filled with shelves, and automatically take inventory of the stock on the high shelves through the equipped scanning devices (such as RFID readers or barcode scanners).

[0054] GPS signals are completely unavailable, relying on an internal positioning system (such as SLAM or UWB), but there are cumulative errors or positioning drifts. The warehouse layout is complex, the aisles between the shelves are narrow, and there are a large number of metal structures, causing problems such as lack of texture or repetitive areas for visual SLAM, and multipath reflection interference for LiDAR. The inventory is stacked irregularly, and there are sometimes temporarily placed forklifts, cargo stacks, or equipment that has not been put back in place in time, forming static or dynamic obstacles not marked on the map. The lighting conditions are uneven.

[0055] A quadrotor drone is used, equipped with a high-precision positioning module, a mission payload, and the obstacle avoidance system based on inertial navigation described in this solution. The INS selects an industrial-grade MEMS inertial measurement unit, with a sampling rate of at least 4kHz and an accelerometer noise density of less than 150 micro-g / sqrt(Hz). The reference high-frequency dynamic response model has been established through calibration flights in the safe areas of the warehouse. The system enables an active modulation strategy, applying a weak sinusoidal perturbation to the motor RPM.

[0056] The drone slowly flies at a speed of 0.5 m / s in the shelf aisle according to the predetermined path. The main navigation system provides positioning and path tracking control.

[0057] Sp1-Sp3: The obstacle avoidance system continuously collects INS acceleration data at 4kHz, synchronously records the motor RPM and the phase of the active modulation signal. According to the current flight speed, attitude, and RPM, the data-driven reference model is called in real time to predict the expected high-frequency acceleration signal when there is no obstacle.

[0058] Sp4: Subtract the predicted reference signal from the real-time INS data to obtain the residual high-frequency signal.

[0059] The drone approaches a corner of an aisle, there are certain specific obstacles that are not in the pre-installed map. The vision or LiDAR of the drone has not been clearly identified due to the angle or distance.

[0060] Sp5: The signal processing module analyzes the residual signal. Due to the active modulation enabled, the system pays special attention to the components related to the modulation frequency and its sideband frequencies generated by the interaction with the blade passing frequency BPF. The analysis finds that at one of the sideband frequencies, there is a significant and stable peak in the energy of the residual signal, and its signal-to-noise ratio SNR reaches 8dB. At the same time, through cross-correlation calculation, it is found that there is a stable time delay between this frequency component and the transmitted modulation signal.

[0061] Sp6: Real-time background noise estimation shows that the current environmental noise level is normal. The calculated SNR (8 dB) exceeds the confirmation threshold dynamically adjusted according to the current noise level (set to 6 dB for general environment). The system confirms the detection of an obstacle.

[0062] Sp7: By analyzing the energy distribution ratio of the echo characteristic signal in the three axes of the INS, it is preliminarily determined that the obstacle is mainly on the right side in front. By accurately measuring the time delay and combining with the air sound speed model, the distance to the obstacle is estimated to be about 2.5 meters.

[0063] Sp8: The obstacle avoidance decision-making module receives the obstacle information (right side in front, about 2.5 meters). Since the UAV is flying in the channel and the space on the left and right is limited, the system generates an instruction: immediately stop moving forward, laterally shift 0.5 meters to the left, and at the same time issue an alarm prompt or pause the task. The UAV executed this instruction smoothly.

[0064] The UAV successfully avoided a collision with an unexpected ladder, ensuring the safety of the equipment and the continuity of the inventory task. In the complex indoor environment with GPS failure, uneven illumination, and sensor blind spots, this INS-based method does not rely on external positioning and traditional vision / LiDAR perception. By analyzing the disturbance echoes generated by itself, it successfully detected the unmodeled static obstacles, demonstrating its unique environmental adaptability. The active modulation strategy improves the sensitivity and reliability of detection. Specific Embodiment 4: Based on the technical solutions of Specific Embodiment 1 and Specific Embodiment 2, further application explanations in combination with the environment are given: The UAV needs to closely inspect the concrete surface corrosion and cracks in the area near the water line of the pier of a large cross-sea bridge. The pier is a huge concrete structure, which causes serious occlusion and multipath interference to the GPS signal, and the positioning accuracy is unreliable. The reflection of the water surface, waves, as well as fog and water vapor will interfere with the detection performance of vision and LiDAR sensors. The sea breeze is strong and unstable, increasing the difficulty of flight control and background noise. The inspection distance is very close (less than 2 meters), and it is necessary to accurately perceive the distance and relative attitude to the pier surface to avoid collision.

[0066] A multi-rotor UAV with a certain wind resistance is used, equipped with a high-definition zoom camera, a laser rangefinder (as an auxiliary confirmation and distance calibration), and the obstacle avoidance system of this solution. The INS selects a high-performance MEMS or fiber optic gyro level, with a sampling rate of 6 kHz and low noise of the accelerometer. The reference model combines physical simulation (considering the special wind field near the pier) and calibration data of the actual open area of the bridge.

[0067] Under the remote monitoring of the operator or based on the feedback of the laser rangefinder, the UAV slowly approaches the pier surface, attempting to maintain an inspection distance of 1.5 meters.

[0068] Sp1 - Sp4: The system collects INS data at 6 kHz, predicts the reference signal (the influence of ground effect / wall effect when approaching large structures has been considered in the model), and extracts the residual signal.

[0069] Suddenly, a lateral sea breeze blows, pushing the UAV closer to the bridge pier. At the same time, due to the rapid change in the UAV's attitude and the distance from the wall, the prediction accuracy of the reference model drops briefly. The update frequency of the laser rangefinder is low and fails to promptly reflect the most dangerous instantaneous approach.

[0070] Sp5: The residual signal analysis module detects that within a wide frequency range (especially in the frequency bands related to the natural frequency of the airframe structure and the BPF), the energy level increases sharply, far exceeding the fluctuations caused by normal wind disturbances. The signal characteristics show asymmetry, indicating strong reflections from a specific direction.

[0071] Sp6: The overall energy integral value of this residual signal or the peaks at multiple key frequencies simultaneously break through the dynamic threshold (which has been increased according to the wind speed estimate). The system determines that a close - range obstacle with high confidence (i.e., the surface of the bridge pier) is approaching rapidly.

[0072] Sp7: By analyzing the energy distribution of the echo signal in the body coordinate system (the signal along the axial direction of the right - hand sensor is significantly stronger than that of the left - hand sensor), it is confirmed that the threat mainly comes from the right side. Time - delay analysis (if distinguishable) or simply the sharp increase rate of the energy intensity indicates that the distance is rapidly shortening and the danger level is extremely high.

[0073] Sp8: The obstacle avoidance system triggers the highest - priority avoidance instruction: immediately apply the maximum lateral thrust to the left (away from the bridge pier direction), and at the same time slightly increase the total thrust to resist the height loss caused by wind pressure and side - shift.

[0074] The UAV is forcibly pushed away before colliding with the bridge pier and returns to a safe distance. Although the inspection task is temporarily interrupted, expensive equipment damage is avoided. In the case where GPS is unreliable and other sensors do not respond promptly or fail due to environmental factors (reflection, water vapor, gusts, low update rate), the INS - based method provides the ultimate and rapid safety guarantee by sensing the abnormal air - medium interaction (strong reflections at extremely close range) caused by the combined action of rapid distance change and strong wind disturbance. It is extremely sensitive to dynamic changes in distance and relative attitude changes and is suitable as the last line of defense for close - range operations.

[0075] It should be noted that, in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising a reference structure" does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0076] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it is understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An obstacle avoidance method for an unmanned aerial vehicle based on inertial navigation, characterized in that, Including the following steps: Sp1. Drive the power system of the drone to make it move, and synchronously collect the broadband high-frequency dynamic response data output by the inertial navigation system. The broadband high-frequency dynamic response data includes three-axis high-frequency linear acceleration signals; Sp2. Establish or obtain a reference high-frequency dynamic response model of the drone under specific flight states and power system operating modes, considering only its own motion and interaction with the free-space air medium; Sp3. According to the current drone flight state and power system operating mode parameters, use the reference high-frequency dynamic response model to predict the current expected high-frequency dynamic response signal of the drone in real time; Sp4. Compare the real-time high-frequency dynamic response data collected in step Sp1 with the expected high-frequency dynamic response signal predicted in step Sp3, and extract the residual high-frequency signal characterizing external environmental abnormal disturbances; Sp5. Analyze the residual high-frequency signal in the time domain, frequency domain or time-frequency domain to identify whether there are specific air medium disturbance echo characteristics formed by the reflection, diffraction or interference of the disturbance generated by the drone's own motion after passing through external obstacles; Sp6. If the specific air medium disturbance echo characteristics are identified and the energy or correlation index of the characteristics exceeds the preset determination threshold, it is confirmed that an obstacle is detected; Sp7. Based on the time delay information, frequency components, multi-axis signal correlation or spatial distribution pattern of the specific air medium disturbance echo characteristics, estimate the relative position, size or shape information of the obstacle; Sp8. Generate and execute an obstacle avoidance flight instruction according to the estimated obstacle information.

2. The obstacle avoidance method for an unmanned aerial vehicle based on inertial navigation according to claim 1, characterized in that, The reference high-frequency dynamic response model in Sp2 is established through computational fluid dynamics simulation combined with structural dynamics analysis, or established by data-driven calibration flight in a known obstacle-free large space.

3. The obstacle avoidance method for an unmanned aerial vehicle based on inertial navigation according to claim 1, characterized in that, In Sp5, the identification of the specific air medium disturbance echo characteristics includes: detecting modulation signals related to the passing frequency of the drone's rotor blades or their harmonics, Doppler frequency shift characteristics, or abnormal enhancement of energy in a specific frequency band.

4. The obstacle avoidance method for an unmanned aerial vehicle based on inertial navigation according to claim 1, characterized in that, The drone obstacle avoidance method further includes: In Sp1, actively apply a small disturbance modulation signal with specific spectral characteristics to the drone's power system; In Sp5, focus on analyzing the relevant components synchronized with or having a specific delay with the disturbance modulation signal in the residual high-frequency signal to enhance the detection sensitivity and anti-interference ability of the specific air medium disturbance echo characteristics.

5. The obstacle avoidance method for an unmanned aerial vehicle based on inertial navigation according to claim 1, characterized in that, In Sp7, the relative azimuth of the obstacle is estimated by analyzing the phase difference or amplitude ratio of the specific air medium disturbance echo characteristics on different axial sensors or different position sensors of the inertial navigation system.

6. The obstacle avoidance method for an unmanned aerial vehicle based on inertial navigation according to claim 1, characterized in that, In Sp7, the relative distance of the obstacle is estimated by analyzing the time delay of the specific air medium disturbance echo characteristics relative to the initial disturbance generated by the drone itself and combining with the disturbance propagation speed model in the air medium.

7. The obstacle avoidance method for an unmanned aerial vehicle based on inertial navigation according to claim 1, characterized in that, The preset determination threshold is dynamically adjusted according to the current environmental background noise level, drone flight speed or estimated value of atmospheric turbulence intensity.

8. The obstacle avoidance method for an unmanned aerial vehicle based on inertial navigation according to claim 1, characterized in that, The drone obstacle avoidance method further includes: Fuse sensor data, where the fused sensor data includes at least one type of sensing data; The fused sensor data is used to confirm the identified echo features in Sp6, or calibrate the estimated obstacle information in Sp7, or assist in optimizing the obstacle avoidance flight instructions in Sp8.

9. The obstacle avoidance method for an unmanned aerial vehicle based on inertial navigation according to claim 1, characterized in that, The drone obstacle avoidance method further includes: Continuously record the detected echo features, estimated obstacle information, and obstacle avoidance execution effects; Using these recorded data, through online or offline learning algorithms, iteratively optimize the reference model in Sp2, the feature recognition algorithm in Sp5, or the obstacle information estimation algorithm in Sp7.

10. An unmanned aerial vehicle system corresponding to the obstacle avoidance method for an unmanned aerial vehicle based on inertial navigation according to any one of claims 1-9, characterized in that, The drone system includes: An inertial navigation system, a processor, and a memory for broadband high-frequency dynamic response data. The memory stores computer-executable instructions, and the computer-executable instructions are configured to be executed by the processor.

Images