GNSS and IMU fusion-based unmanned aerial vehicle high-precision autonomous navigation method and system

Abstract

The invention provides an unmanned aerial vehicle high-precision autonomous navigation method and system based on GNSS and IMU deep fusion, and aims to solve the problems of insufficient navigation precision and poor robustness in a complex electromagnetic environment. Through a tight coupling architecture, GNSS original observed quantity and IMU pre-integration results are jointly modeled in an observation layer, and multi-source constraints are introduced in combination with factor graph optimization, so that the positioning precision and consistency in weak signal and shielding scenes are remarkably improved. For abnormal observation, a robust kernel function is adopted to dynamically adjust the weight, and the influence of electromagnetic interference and a multipath effect is effectively inhibited. Meanwhile, navigation calculation and model prediction control MPC are combined, and sub-meter hovering and high-precision trajectory tracking are achieved. According to the method, in high-voltage transmission line inspection, dependence on a high-cost sensor is reduced, the engineering application value is high, the method can be widely applied to the fields of electric power inspection, disaster emergency, infrastructure monitoring and the like, and reliable technical support is provided for high-precision autonomous navigation of the unmanned aerial vehicle in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of intelligent navigation and control technology for unmanned aerial vehicles (UAVs), specifically to a high-precision autonomous navigation method and system for UAVs based on GNSS and IMU fusion. Background Technology

[0002] In recent years, drones have been increasingly used in power line inspection, disaster emergency response, and infrastructure inspection, and their high-precision autonomous navigation capabilities have gradually become a core factor restricting the effectiveness of mission execution. However, in complex electromagnetic environments, such as high-voltage transmission line corridors, or under highly dynamic operating conditions, existing navigation technologies still have significant limitations and are difficult to meet the requirements of high precision and high robustness.

[0003] 1. Limitations of GNSS standalone navigation:

[0004] Currently, drone navigation primarily relies on the Global Navigation Satellite System (GNSS). In open environments, GNSS can provide relatively accurate positioning information, but its performance degrades significantly in complex environments.

[0005] Electromagnetic Interference: In special scenarios such as high-voltage power transmission corridors, strong electric fields and electromagnetic radiation can interfere with GNSS receivers, leading to carrier phase instability. This manifests as jitter in observations, enhanced multipath signals, and occasional cycle slips, thereby reducing the accuracy of pseudorange and carrier observations. Obstruction Effect: Large metal structures such as power transmission towers and conductors can obstruct and reflect GNSS signals, disrupting direct paths and worsening constellation geometry. When the number of effectively visible satellites decreases or the geometry degrades, the sensitivity of positioning results to individual anomalies increases significantly, causing navigation fluctuations or even short-term loss of lock. Safety Hazards: When UAVs perform close-range inspections or hovering operations, GNSS navigation alone cannot provide continuous, stable, and high-precision position information, posing significant safety risks.

[0006] 2. Limitations of IMU independent navigation:

[0007] Inertial measurement units (IMUs), as another commonly used navigation sensor, have the advantages of fast response speed and high short-time accuracy, but their independent operation has the following problems:

[0008] Cumulative Errors: IMUs obtain velocity and position information by integrating acceleration and angular velocity, but they are inevitably affected by zero-bias instability, scale factor errors, and installation angle errors during long-term operation. These errors accumulate exponentially during integration, causing position estimates to diverge rapidly. Attitude Error Propagation: Attitude estimation errors are transmitted to velocity and position calculations through gravity projection effects, further exacerbating error growth. Especially in high-voltage transmission line inspections, the superposition of strong wind disturbances and electromagnetic interference, along with structural vibrations and dynamic load changes generated by the flight platform, causes the IMU output to contain stronger high-frequency noise and random walks, further compressing its available independent navigation time window. Insufficient Accuracy: Experiments show that low-cost IMUs may exhibit drift deviations of meters or even tens of meters within tens of seconds, failing to meet the sub-meter accuracy requirements for close-range operation.

[0009] 3. Shortcomings of existing loosely coupled GNSS / IMU architectures:

[0010] To overcome the limitations of single sensors, current research generally employs a combined GNSS and IMU navigation mode to achieve complementary advantages: GNSS is used to correct the accumulated errors of the IMU, while the IMU provides short-term continuous positioning when GNSS lock-on is lost or signal quality degrades. However, the current mainstream loosely coupled architecture still has the following problems:

[0011] Information Loss: Loosely coupled architectures typically rely on GNSS and IMU to perform independent calculations, followed by result fusion through filtering algorithms such as Kalman filtering. This result-level fusion cannot fully utilize the rich information in the raw GNSS observations, resulting in limited error suppression capabilities. Insufficient Robustness: When GNSS signals are obstructed or degraded by interference, loosely coupled architectures cannot directly utilize the raw GNSS carrier phase residuals, pseudorange consistency checks, or constellation geometric constraints to suppress noise and eliminate anomalous observations, causing the filter to exhibit hysteresis and instability when dealing with noisy measurements. Loss of Continuity: Once GNSS signals are lost or accuracy deteriorates significantly, loosely coupled systems rely almost entirely on IMU calculations, thus falling into a predicament of rapid drift accumulation, making it difficult to guarantee overall navigation accuracy and continuity.

[0012] 4. International Research Trends and Existing Technological Bottlenecks:

[0013] In recent years, academia and industry have begun to explore tightly coupled or even deeply coupled GNSS / IMU fusion strategies, combining them with methods such as factor graph optimization and extended Kalman filtering to effectively improve robustness and real-time performance under weak signal and interference environments. Typical technical approaches include: Factor graph optimization frameworks: These combine IMU pre-integration, GNSS observations, and environmental sensing constraints such as lidar or vision within a sliding window for joint estimation, maintaining global consistency and robustness, such as LIO-SAM. Filtered tightly coupled methods: These directly fuse IMU and external measurements at the observation level, offering advantages in stability and computational controllability, such as LINS and FAST-LIO. Multi-source collaborative strategies: These utilize environmental sensors to provide geometric constraints during GNSS interference, maintaining track continuity and ground... Figure 1 Consistency, such as LVI-SAM, R2LIVE / R3LIVE. Although the above methods improve navigation performance to some extent, existing technologies still face the following challenges in complex scenarios such as high-voltage transmission lines:

[0014] Strong electromagnetic interference: The strong electric fields in high-voltage transmission corridors not only weaken GNSS signals but may also contaminate the output of IMU and other sensors, further exacerbating navigation errors. Significant obstruction: Structures such as transmission towers and power lines significantly obstruct and reflect GNSS signals, and the limited number of visible satellites makes it difficult to maintain high-precision positioning using traditional methods. Insufficient accuracy and continuity: Existing technologies struggle to meet the requirements for sub-meter accuracy and long-term continuous navigation in close-range alignment and hovering operations.

[0015] In summary, the existing technology has the following key problems that urgently need to be solved:

[0016] 1. Insufficient robustness of GNSS in complex electromagnetic environments: How to maintain the stability and positioning accuracy of GNSS signals under conditions of strong electric fields, multipath interference, and obstruction. 2. Cumulative drift problem of IMU independent navigation: How to suppress the cumulative error of IMUs during long-term operation and extend their available independent navigation time. 3. Fusion limitations of loosely coupled architecture: How to overcome the result-level fusion limitations of traditional loosely coupled architectures and fully explore the information value of raw GNSS observations. 4. High precision and continuity assurance in complex electromagnetic environments: How to achieve robust navigation and sub-meter positioning for UAVs in scenarios with strong interference and high obstruction, such as high-voltage transmission lines. Summary of the Invention

[0017] To address the aforementioned issues, this invention proposes a high-precision autonomous navigation method and system for unmanned aerial vehicles (UAVs) based on GNSS and IMU fusion. The aim is to overcome the shortcomings in accuracy and robustness of existing technologies in complex electromagnetic environments through a tightly coupled architecture, factor graph optimization, multi-constraint introduction, and adaptive weighting mechanisms, thereby providing reliable support for high-precision tasks such as X-ray inspection of power transmission lines.

[0018] To achieve the above objectives, the present invention adopts the following technical solution:

[0019] A high-precision autonomous navigation method for UAVs based on GNSS and IMU fusion is proposed. This method is based on a high-precision autonomous navigation system for UAVs based on GNSS and IMU fusion. The high-precision autonomous navigation system for UAVs based on GNSS and IMU fusion includes: an airborne sensor module, a fusion processing module, a control execution module, a mission payload module, and a power supply module.

[0020] The airborne sensor module includes a GNSS receiver and an inertial measurement unit (IMU). The GNSS receiver receives satellite signals through a multi-frequency, multi-mode antenna and outputs raw pseudorange, carrier phase, and Doppler observations. The IMU consists of a three-axis accelerometer and a three-axis gyroscope, which output the aircraft's linear acceleration and angular velocity in real time. The airborne sensor module collects raw observation data required for navigation and transmits it to the fusion processing module.

[0021] The fusion processing module includes an embedded high-performance processor, which is based on a tightly coupled algorithm framework of Error State Extended Kalman Filter (ESKF) to directly perform joint modeling of raw GNSS observations and IMU integration results at the observation layer. At the same time, factor graph optimization is introduced to jointly estimate GNSS observations, IMU pre-integration, and external constraints within a sliding window, and output the fused positioning results to the control execution module.

[0022] The control execution module includes the flight controller (FC) and the UAV power system; it generates optimal control commands based on navigation calculation results to drive the UAV in flight.

[0023] The mission payload module includes an X-ray imaging device and auxiliary sensors, which communicate with the flight controller via a CAN bus and use the high-precision pose information provided by the navigation system to accurately scan the power transmission line; at the same time, it provides structural feature constraints for navigation as an external constraint factor input to the fusion processing module to enhance the stability of navigation calculation.

[0024] The power module is connected to each module via an independent power supply line, providing a stable DC power supply to the GNSS receiver, IMU, fusion processing module, flight controller and other modules.

[0025] Furthermore, the method includes the following steps:

[0026] Step 1: Data Acquisition: Obtain raw observation data from the Global Navigation Satellite System (GNSS) and the Inertial Measurement Unit (IMU), and introduce a multi-constellation fusion and observation redundancy verification mechanism to eliminate gross errors;

[0027] Step 2: IMU pre-integration: Use IMU data to pre-integrate the motion state of the UAV to generate low-dimensional motion constraints. Pre-integration reduces the state dimension while ensuring real-time performance.

[0028] Step 3: Tightly Coupled Fusion: The raw GNSS measurements and IMU pre-integration results are jointly modeled to construct a joint observation equation, and prediction-update recursion is achieved through Error State Extended Kalman Filter (ESKF).

[0029] Step 4 Factor graph optimization: Within the sliding time window, a factor graph model containing GNSS factors, IMU pre-integration factors and external constraint factors is constructed, and nonlinear least squares optimization is used to solve iteratively, so as to maintain high-precision long-term navigation under the condition of weak GNSS signal or short-term loss of lock.

[0030] Step 5 Anomaly Detection and Adaptive Weighting: By calculating and normalizing the GNSS factor residuals, and combining threshold discrimination and robust kernel function to dynamically adjust the observation weights, the optimization problem is rewritten into a weighted least squares form. In electromagnetic interference or GNSS signal degradation scenarios, the impact of abnormal observations is suppressed, and continuous and stable navigation solutions are maintained.

[0031] Step 6 Control and Execution: The optimized pose and velocity solutions are input into the Model Predictive Controller (MPC). Based on the UAV's six-degree-of-freedom dynamics model and mission objective function, the optimal control input is generated online to complete sub-meter level hovering, path tracking, and anti-disturbance flight, enabling high-precision operations such as alignment imaging and X-ray detection in power transmission line inspection.

[0032] Furthermore, in step 1, the GNSS receiver carried by the UAV synchronously receives signals from multiple constellations through a multi-frequency, multi-mode antenna, and outputs the pseudorange observation ρ of the i-th satellite. i Carrier phase φ i With Doppler frequency shift f D,i Inertial Measurement Unit (IMU) acquires triaxial acceleration a in real time. m With angular velocity ω m ;

[0033] During the data acquisition phase, the observation noise is denoted as: the noise term for pseudorange observations ∈ ρ The noise term of the carrier phase observation ∈ φ The noise term of the Doppler frequency shift observation ∈ f A multi-constellation fusion and observation redundancy verification mechanism was introduced to filter out gross errors.

[0034] Furthermore, in step 2, during the continuous time period [t] k ,t k+1 Within the [unclear context], the motion state of the UAV is pre-integrated using IMU data, and the calculation formula is as follows:

[0035]

[0036] In the formula, ΔR k The attitude change increment; Δv k This represents the increment of velocity change; This represents the original measurement value of the gyroscope at time j. b is the raw accelerometer measurement at time j; g b is the gyroscope's zero bias. a For accelerometer zero bias; n g For gyroscope random noise; n a R represents random noise from the accelerometer; Δt is the sampling time interval; j Let be the attitude rotation matrix at time j; k is the discrete time index on the time axis.

[0037] Furthermore, in step 3, a joint observation equation is constructed, which models the raw GNSS measurements and the IMU pre-integration results together:

[0038]

[0039] In the formula, z i Let ρ be the observation vector at time i; i The pseudorange observation value of the i-th satellite measured by GNSS; φ i f is the carrier phase of the i-th satellite measured by GNSS. D,i The Doppler shift of the i-th satellite measured by GNSS; h(x k ,p i ) is the observation model function, which represents the system state vector x. k and satellite position p i Mapped to predicted observations; x k =[p k ,v k ,q k ,b a ,b g ], is the system state vector, containing the state information of the UAV at time i, p k v represents the location of the drone. k For the speed of the drone, q k Let b be the attitude quaternion of the UAV. a The accelerometer has zero bias, reflecting the systematic error of the IMU sensor; b g The gyroscope has zero bias, reflecting the systematic error of the IMU sensor; p i v represents the spatial position of the i-th satellite; i Let represent the observation noise of the i-th satellite.

[0040] Furthermore, in step 4, the mathematical expression for factor graph optimization is:

[0041] min X ∑‖r GNSS || 2 +∑‖r IMU || 2 +∑‖r constraint || 2 ;

[0042] In the formula, x is the optimization variable, representing the set of pose and deviation states within the sliding window; r GNSS GNSS observation residuals represent the difference between GNSS measurements and current state predictions; r IMU The IMU pre-integration residual represents the difference between the IMU integration result and the current state prediction; r constraint The external constraint residuals include prior knowledge such as altitude limits, flight corridor geometric constraints, and dynamic consistency conditions.

[0043] Furthermore, in step 5, anomaly detection involves identifying outliers in GNSS observations and determining anomalies through residual calculation, normalization, and threshold discrimination.

[0044] First, in each optimization iteration, the residual of the GNSS factor is calculated: r i =z i -h(x,p i );

[0045] Then, the residuals are normalized:

[0046] Finally, threshold discrimination is performed: a confidence interval threshold τ is set, if... The observation is then judged as an outlier with a probability of P > 99.7%.

[0047] In the formula, r i Let z be the residual of the i-th GNSS observation, representing the difference between the actual observation and the predicted value; i The i-th GNSS observation is directly output by the GNSS receiver; h(x,p) i () is an estimate of x based on the current state and the satellite position p. i The predicted value; x is the current state estimation vector, including the UAV's position, velocity, attitude, and sensor bias; p i Let be the spatial position of the i-th satellite; The normalized residual; σ i The standard deviation of the observation is estimated based on historical data statistics or the accuracy index provided by the GNSS receiver; a common choice is τ=3, which corresponds to the 99.7% confidence interval under a Gaussian distribution.

[0048] Furthermore, in step 5, the adaptive weighting mechanism is as follows: the weights are dynamically adjusted according to the anomaly detection results, and the weighting mechanism is introduced into the optimization problem so that the system can still maintain stable solution under abnormal conditions;

[0049] Introducing a robust kernel function to adjust weights w i :

[0050]

[0051] The final optimization problem is rewritten in weighted least squares form:

[0052]

[0053] In the formula, w i Let be the weight of the i-th observation; x is the optimization variable, representing the set of pose and bias states within the sliding window; w i The adaptive weighting of the i-th GNSS observation; r GNSS,i To represent the difference between the actual and predicted values ​​of the i-th GNSS observation, r GNSS,i =z i -h(x k ,p i ), z i Let h(x) be the observation vector at time i. k ,p i ) is the observation model function, which represents the system state vector x. k and satellite position p i Mapped to predicted observations; r IMU,j For IMU pre-integration residuals; r constraint,k The residuals are external constraints.

[0054] Furthermore, in step 6, the optimized pose and velocity solutions are input to the Model Predictive Controller (MPC). The controller generates the optimal control input based on the UAV's six-degree-of-freedom dynamics model and the mission objective function. The mathematical expression is:

[0055]

[0056] In the formula, u * argmin represents the optimal control input, indicating the optimal control command for the UAV at the current moment and within a certain future timeframe, used to drive the UAV to achieve the desired trajectory tracking and flight mission; u is the control input variable, i.e., the target to be optimized; argmin u To find the control input u;p that minimizes the objective function t p represents the actual position of the UAV at time t. ref v is the reference position of the UAV at time t; tThe actual velocity of the drone at time t; v ref u is the reference velocity of the UAV at time t; t Let t be the control input at time t; Q be the position error weight matrix; R be the velocity error weight matrix; S be the control input weight matrix; and T be the length of the prediction time window.

[0057] Furthermore, in step 5, various robust kernel functions are introduced to handle abnormal residuals under different scenarios;

[0058] Huber kernel function: When the residual is small, it maintains a quadratic penalty; when the residual exceeds a threshold δ, it switches to linear growth, defined as:

[0059]

[0060] Its corresponding weights are:

[0061]

[0062] The Cauchy kernel function suppresses large residuals with slowly decreasing weights and is defined as follows:

[0063]

[0064] The corresponding weighting function is:

[0065]

[0066] Tukey Biweight kernel function: applies a second penalty to residuals within a threshold δ, but sets the weight to zero for residuals exceeding the threshold, thus completely eliminating them. It is defined as follows:

[0067]

[0068] The corresponding weighting function is:

[0069]

[0070] The weighted optimization problem after adopting a robust kernel function is uniformly represented as:

[0071]

[0072] In the formula, ρ(r) is the robust cost function, used to define the contribution of the residuals to the optimization objective, suppressing the influence of outlier residuals through nonlinear design; r is the residual, representing the difference between the actual observed value and the predicted value, reflecting the consistency between the measurement and the model; w(r) is the weight function, derived from the derivative of the robust cost function, used to adjust the contribution of each residual in the optimization; δ is the threshold parameter, used to control the boundary of the residual transitioning from normal values ​​to outlier values, and its value is usually determined based on the actual noise level or experience; c is the scaling parameter, used to adjust the descent rate of the Cauchy kernel function, determining the degree of weight decay as the residual increases; r GNSS,i It is the i-th GNSS observation residual, consisting of the difference between pseudorange, carrier phase, or Doppler measurement and prediction; r IMU,j It is the j-th IMU pre-integration residual, reflecting the deviation between the pose predicted based on inertial measurements and the actual estimate; r constraint,k is the k-th external constraint residual, introduced by prior conditions; X is the set of states within the sliding window, including system state variables such as position, velocity, attitude quaternions, and zero bias of accelerometers and gyroscopes; ρ(·) is any of the above robust kernel functions.

[0073] The high-precision autonomous navigation method and system for UAVs based on GNSS and IMU fusion proposed in this invention significantly improves the navigation performance of UAVs in complex electromagnetic interference and GNSS signal degradation environments through key technologies such as tightly coupled architecture, factor graph optimization, robust weighting mechanism and model predictive control (MPC).

[0074] Its technical effects are reflected in the following aspects:

[0075] 1. Technical aspects: Improving navigation accuracy and robustness

[0076] Deep Fusion at the Observation Layer: This invention directly performs joint modeling of raw GNSS observations and IMU pre-integration results at the observation layer, avoiding error accumulation and information loss caused by independent calculations in traditional loosely coupled methods. Experimental results show that in typical operating scenarios of high-voltage transmission line corridors, even if the GNSS signal is briefly lost for more than 30 seconds, this invention can still maintain a positioning error of less than 0.5 meters, while the error of traditional loosely coupled methods rapidly spreads to the order of several meters or even tens of meters under the same conditions.

[0077] Factor graph optimization enhances consistency: By introducing a factor graph optimization framework to jointly process IMU pre-integration, raw GNSS observations, and external geometric constraints, the navigation solution exhibits stronger consistency and robustness. This global optimization method can maintain continuous navigation for extended periods in weak signal or obstructed environments, significantly improving the system's anti-interference capability.

[0078] 2. Flight control level: Improve hovering stability and trajectory tracking accuracy

[0079] Integrating Navigation and MPC: This invention combines high-precision navigation calculation results with Model Predictive Control (MPC), significantly improving the hovering stability and trajectory tracking accuracy of UAVs under strong wind disturbance conditions. Practical verification shows that the system can maintain UAV attitude fluctuations of less than 2° in strong wind environments and achieve stable sub-meter hovering in close-range alignment imaging tasks. This performance ensures that the UAV can complete high-precision X-ray inspection when close to power transmission lines, avoiding image shifts or detection omissions caused by navigation instability, thereby greatly improving the reliability of the inspection results.

[0080] 3. Engineering application level: Reduce hardware dependence and operation and maintenance costs

[0081] Efficient use of low-cost sensors: This invention reduces reliance on the quality of individual sensors, enabling low-cost IMUs and commercial-grade GNSS receivers to achieve near-professional-level navigation accuracy in complex scenarios. This not only significantly reduces the overall cost of the UAV inspection system but also improves its usability in complex electromagnetic environments in the field.

[0082] Significantly improved detection efficiency: In power line inspection tasks, after applying the solution of this invention, the number of manual reflights required for drones to complete the same scale of line inspection is reduced by about 40%, and the detection efficiency is improved by more than 30%. This improvement provides substantial economic benefits for power grid operation and maintenance, while reducing the need for manual intervention.

[0083] 4. Social and industry level: Ensuring security and expanding application scenarios

[0084] Enhancing operational safety: This invention significantly reduces reliance on manual close-range operations by enhancing the autonomy of drones in disaster emergency response and infrastructure inspection tasks. Particularly in high-voltage transmission line inspections, it avoids the risk of inspection personnel being directly exposed to strong electric fields and complex terrain, significantly improving the safety of power operation and maintenance.

[0085] Broad application prospects: The technical approach of this invention has strong scalability and can be extended to various task scenarios such as post-disaster emergency reconnaissance, traffic infrastructure monitoring, and urban 3D modeling. Whether in high-risk areas or scenarios requiring high precision, this invention can provide reliable navigation support for UAVs, laying a solid technical foundation for their widespread application in more fields.

[0086] 5. Overall Benefits: Overcoming technological bottlenecks and creating economic and social value.

[0087] Technological Breakthrough: This invention solves the core bottleneck problems of existing integrated navigation technologies in complex electromagnetic environments, including the susceptibility of GNSS signals to interference, IMU drift accumulation, and the limitations of loosely coupled architecture, and achieves high-precision and highly robust autonomous navigation.

[0088] Economic value: By reducing hardware costs, minimizing manual re-flight operations, and improving detection efficiency, this invention brings significant economic benefits to industries such as power line inspection.

[0089] Social benefits: This invention not only enhances the operational capabilities of drones in complex environments, but also makes significant contributions to the intelligent development of related industries and social security by reducing human intervention and ensuring operational safety.

[0090] In summary, the high-precision autonomous navigation method and system for UAVs based on GNSS and IMU fusion proposed in this invention not only breaks through the bottlenecks of existing integrated navigation in terms of technical performance, but also demonstrates significant value in economic and social applications. By improving navigation accuracy and robustness, optimizing flight control effects, reducing hardware dependence, expanding application scenarios, and ensuring operational safety, this invention provides a reliable solution for high-precision autonomous navigation of UAVs in complex electromagnetic environments, possessing important engineering application significance and broad development prospects. Attached Figure Description

[0091] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0092] Figure 1 This is a flowchart of the high-precision autonomous navigation method for UAVs based on deep fusion of GNSS and IMU according to the present invention. Detailed Implementation

[0093] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0094] To achieve high-precision autonomous navigation for unmanned aerial vehicles (UAVs) in complex electromagnetic environments, this invention proposes a method and system for high-precision autonomous navigation of UAVs based on GNSS and IMU fusion. This scheme comprehensively utilizes the global positioning capability of GNSS and the high dynamic short-time accuracy characteristics of IMU, employing a tightly coupled / deeply coupled architecture to achieve signal-level data fusion, and enhancing the system's robustness in signal degradation scenarios through improved filtering and optimization algorithms. The system as a whole consists of an onboard sensor module, a fusion processing module, a control execution module, and a mission payload module, with information exchange between the modules achieved through a high-speed data bus.

[0095] The overall architecture of a high-precision autonomous navigation system for UAVs based on GNSS and IMU fusion is as follows:

[0096] Airborne sensor module: The airborne sensor module includes a GNSS receiver and an IMU; the airborne sensor module is responsible for collecting the raw observation data required for navigation and providing input for subsequent fusion algorithms.

[0097] GNSS receivers receive satellite signals through multi-frequency, multi-mode antennas and output pseudorange, carrier phase, Doppler shift, and C / N0 observations. GNSS receivers provide global positioning information and offer a high-precision reference for primary navigation calculations in open environments. GNSS receivers support joint calculations with multiple constellations including GPS, BeiDou, GLONASS, and Galileo, and possess anti-interference capabilities such as RF front-end suppression and cycle slip detection.

[0098] The inertial measurement unit (IMU) acquires triaxial acceleration and angular velocity data in real time. The IMU provides high-dynamic short-time motion state estimation, maintaining continuous positioning even when GNSS signals are weak or lost. As a high dynamic range device, the IMU is adapted to strong disturbance flight conditions; it incorporates zero-bias compensation and temperature drift correction mechanisms.

[0099] The GNSS receiver and IMU align their sampling times through a time synchronization module to ensure spatiotemporal consistency; the data is preprocessed and then input into the fusion processing module.

[0100] Fusion Processing Module: The fusion processing module includes a high-performance embedded processor such as ARM+FPGA, which executes tightly coupled filtering, factor graph optimization, and robust weighting mechanisms, serving as the core computing unit of the entire system. The high-performance ARM processor executes upper-level algorithms such as factor graph optimization and robust weighting; the FPGA co-processing unit accelerates lower-level calculations, such as IMU pre-integration and matrix operations, enabling deep coupling processing of GNSS and IMU data to generate high-precision pose calculations, achieve anomaly detection and adaptive weighting, and improve system robustness.

[0101] The main algorithms of the fusion processing module are as follows: 1. By using a tightly coupled error state extended Kalman filter (ESKF), the raw GNSS observations and IMU pre-integration results are jointly modeled to avoid information loss due to loose coupling; simultaneously, the state vectors such as position, velocity, attitude, and sensor bias are dynamically updated. 2. Through sliding window factor graph optimization, GNSS factors, IMU pre-integration factors, and external constraint factors, such as altitude restrictions and flight corridor geometric constraints, are combined; simultaneously, a nonlinear least squares problem is constructed, and global consistency is maintained through iterative optimization. 3. Through a robust weighting mechanism, Huber, Cauchy, or Tukey Biweight kernel functions are introduced based on residual statistics to dynamically adjust the weights of abnormal observations.

[0102] The function of the fusion processing module is to maintain continuous and stable navigation calculations under conditions of weak GNSS signals, obstruction, or loss of lock; and to improve the positioning accuracy and robustness of the system in complex electromagnetic environments. The fusion processing module receives preprocessed data from the GNSS receiver and IMU, and outputs the navigation calculation results to the flight controller.

[0103] Control Execution Module: The control execution module includes the flight controller (FC) and the UAV power system, such as propellers and motors; it generates optimal control commands based on navigation calculation results to drive the UAV in flight.

[0104] The flight controller receives pose and velocity information from the fusion processing module; it generates optimal control inputs, such as propeller speed allocation, based on the Model Predictive Control (MPC) algorithm. The flight controller enables trajectory tracking, hovering, and disturbance-resistant flight, ensuring the UAV can perform sub-meter level alignment imaging and X-ray inspection tasks during power line inspections. The flight controller possesses high dynamic response capabilities and supports external disturbance estimation and compensation.

[0105] The flight controller adjusts the propeller speed according to control commands, changing the attitude and position of the UAV to achieve precise flight control and cope with external disturbances such as strong winds and electromagnetic interference. The flight controller receives navigation calculation results from the fusion processing module and outputs control commands to the power system.

[0106] Mission payload module: The mission payload module includes mission payloads such as X-ray imaging equipment, lidar, and vision sensors; it enhances navigation performance through feedback auxiliary information such as geometric constraints.

[0107] Among them, X-ray imaging equipment detects defects in power transmission lines, providing high-precision image data to support power operation and maintenance decisions. Auxiliary sensors, including lidar and vision sensors, are used for environmental perception and map building; their feedback geometric constraints, such as the location of power transmission lines and terrain features, enhance the stability and accuracy of navigation calculations.

[0108] The mission payload module communicates with the flight controller via the CAN bus, and auxiliary sensor data can be used as external constraint factors input to the fusion processing module.

[0109] Power Module: The power module provides a stable power supply to ensure the normal operation of all modules. It provides a stable DC power supply to the GNS S receiver, IMU, fusion processing module, flight controller, and other modules, ensuring their normal operation in complex electromagnetic environments. It is also equipped with a voltage regulator to cope with power fluctuations. The power module is connected to each module via independent power supply lines, providing overcurrent protection and electromagnetic compatibility design.

[0110] A high-precision autonomous navigation system for unmanned aerial vehicles (UAVs) based on GNSS and IMU fusion achieves high-precision autonomous navigation in complex electromagnetic environments through the collaborative work of onboard sensor modules, fusion processing modules, control execution modules, payload modules, and power supply modules. Each module has a clear division of labor and complementary functions, jointly solving the navigation challenges of UAVs in scenarios such as high-voltage power line inspection. GNSS and IMU provide multi-source observation data; the fusion processing module is responsible for data fusion and optimization; the control execution module achieves precise flight control; and the payload module supports high-precision operations and navigation enhancement. This system is modular and highly scalable, suitable for various high-risk, high-precision application scenarios such as power line inspection, disaster emergency response, and infrastructure monitoring.

[0111] Based on the aforementioned GNSS and IMU fusion-based high-precision autonomous navigation system for UAVs, this embodiment also provides a GNSS and IMU fusion-based high-precision autonomous navigation method for UAVs, such as... Figure 1 As shown, the method includes the following steps:

[0112] Step 1: Data Acquisition: Obtain raw observation data from the Global Navigation Satellite System (GNSS) and the Inertial Measurement Unit (IMU), introduce multi-constellation fusion and observation redundancy verification mechanisms, and eliminate gross errors.

[0113] Specifically, the GNSS receiver carried by the UAV synchronously receives signals from multiple constellations through a multi-frequency, multi-mode antenna, and outputs the pseudorange observation ρ of the i-th satellite. i Carrier phase φ i With Doppler frequency shift f D,i Simultaneously, the inertial measurement unit (IMU) acquires triaxial acceleration a in real time. m With angular velocity ω m ;

[0114] During the data acquisition phase, GNSS observations are affected by electromagnetic interference and multipath effects, while IMU data suffers from zero bias and random walk errors. Therefore, the observation noise is denoted as follows: the noise term of pseudorange observations ∈ ρ The noise term of the carrier phase observation ∈ φThe noise term of the Doppler frequency shift observation ∈ f To enhance robustness, a multi-constellation fusion and observation redundancy verification mechanism was introduced to filter out gross errors. For time synchronization, GNSS and IMU data were mapped to a unified time base using PPS pulses per second, hard triggering, or interpolation alignment methods to ensure spatiotemporal consistency.

[0115] Step 2: IMU pre-integration: Use IMU data to pre-integrate the motion state of the UAV to generate low-dimensional motion constraints. Pre-integration reduces the state dimension while ensuring real-time performance.

[0116] Specifically, in the continuous time period [t] k ,t k+1 Within the [unclear context], the motion state of the UAV is pre-integrated using IMU data, and the calculation formula is as follows:

[0117]

[0118] In the formula, ΔR k The attitude change increment; Δv k This represents the increment of velocity change; This represents the original measurement value of the gyroscope at time j. b is the raw accelerometer measurement at time j; g b is the gyroscope's zero bias. a For accelerometer zero bias; n g For gyroscope random noise; n a R represents random noise from the accelerometer; Δt is the sampling time interval; j Let be the attitude rotation matrix at time j; k is the discrete time index on the time axis.

[0119] By pre-integrating, the state dimension can be reduced while ensuring real-time performance, thereby effectively utilizing high-frequency IMU measurements in subsequent factor graph optimization.

[0120] Step 3: Tightly Coupled Fusion: The raw GNSS measurements and IMU pre-integration results are jointly modeled to construct a joint observation equation, and prediction-update recursion is achieved through Error State Extended Kalman Filter (ESKF).

[0121] Specifically, a joint observation equation is constructed to model the data using both raw GNSS measurements and IMU pre-integration results:

[0122]

[0123] In the formula, z i Let ρ be the observation vector at time i; i The pseudorange observation value of the i-th satellite measured by GNSS; φ if is the carrier phase of the i-th satellite measured by GNSS. D,i The Doppler shift of the i-th satellite measured by GNSS; h(x k ,p i ) is the observation model function, which represents the system state vector x. k and satellite position p i Mapped to predicted observations; x k =[p k ,v k ,q k ,b a ,b g ], is the system state vector, containing the state information of the UAV at time i, p k v represents the location of the drone. k For the speed of the drone, q k Let b be the attitude quaternion of the UAV. a The accelerometer has zero bias, reflecting the systematic error of the IMU sensor; b g The gyroscope has zero bias, reflecting the systematic error of the IMU sensor; p i v represents the spatial position of the i-th satellite; i Let represent the observation noise of the i-th satellite.

[0124] Finally, the Error State Extended Kalman Filter (ESKF) is used to complete the prediction-update recursion, avoiding the nonlinear divergence of state drift.

[0125] Step 4: Factor graph optimization: Within the sliding time window, a factor graph model containing GNSS factors, IMU pre-integration factors, and external constraint factors is constructed, and nonlinear least squares optimization is used to solve the problem iteratively, so as to maintain high-precision long-term navigation under the condition of weak GNSS signal or short-term loss of lock.

[0126] Specifically, within the sliding time window, a factor graph model is constructed that includes GNSS factors, IMU pre-integration factors, and prior constraint factors, forming a nonlinear least squares optimization problem:

[0127] min X ∑‖r GNSS || 2 +∑‖r IMU || 2 +∑‖r constraint || 2 ;

[0128] In the formula, x is the optimization variable, representing the set of pose and deviation states within the sliding window; r GNSS GNSS observation residuals represent the difference between GNSS measurements and current state predictions; r IMUThe IMU pre-integration residual represents the difference between the IMU integration result and the current state prediction; r constraint The external constraint residuals include prior knowledge such as altitude limits, flight corridor geometric constraints, and dynamic consistency conditions.

[0129] Through iterative optimization, the consistency and stability of the global solution can be maintained even under conditions of weak GNSS signals or short-term loss of lock.

[0130] Step 5 Anomaly Detection and Adaptive Weighting: By calculating and normalizing the GNSS factor residuals, and combining threshold discrimination and robust kernel function to dynamically adjust the observation weights, the optimization problem is rewritten into a weighted least squares form. This suppresses the impact of abnormal observations in electromagnetic interference or GNSS signal degradation scenarios, and maintains continuous and stable navigation solutions.

[0131] In actual operation, GNSS observations may produce outliers due to electromagnetic interference, multipath effects, or satellite obstruction. Without intervention, these outliers can significantly affect the global solution results of factor map optimization. Therefore, an adaptive weighting mechanism based on residual statistics is proposed, with the following steps:

[0132] (1) Residual calculation:

[0133] In each optimization iteration, the residual of the GNSS factor is calculated:

[0134] r i =z i -h(x,p i );

[0135] In the formula, r i Let z be the residual of the i-th GNSS observation, representing the difference between the actual observation and the predicted value; i The i-th GNSS observation is directly output by the GNSS receiver; h(x,p) i () is an estimate of x based on the current state and the satellite position p. i The predicted value; x is the current state estimation vector, including the UAV's position, velocity, attitude, and sensor bias; p i Let be the spatial position of the i-th satellite.

[0136] (2) Residual normalization:

[0137] To improve the comparability between different measurement channels, the residuals are normalized:

[0138]

[0139] In the formula, The normalized residual; σ iThe standard deviation of the observation is estimated based on historical data statistics or accuracy indicators provided by the GNSS receiver.

[0140] (3) Threshold discrimination:

[0141] Set a confidence interval threshold τ, if The observation is then considered a possible outlier. A common choice is τ = 3, which corresponds to the 99.7% confidence interval under a Gaussian distribution.

[0142] (4) Weight adjustment:

[0143] To address potential anomaly observations, a robust kernel function is introduced to adjust the weights w. i :

[0144]

[0145] In the formula, w i Let be the weight of the i-th observation. This rule weakens the impact of large residual observations on the optimization results, while normal observations retain their original weights.

[0146] (5) Weighted optimization objective:

[0147] The final optimization problem is rewritten in weighted least squares form:

[0148]

[0149] In the formula, x is the optimization variable, representing the set of pose and deviation states within the sliding window; w i The adaptive weighting of the i-th GNSS observation; r GNSS,i To represent the difference between the actual and predicted values ​​of the i-th GNSS observation, r GNSS,i =z i -h(x k ,p i ), z i Let h(x) be the observation vector at time i. k ,p i ) is the observation model function, which represents the system state vector x. k and satellite position p i Mapped to predicted observations; r IMU,j For IMU pre-integration residuals; r constraint,k The residuals are external constraints.

[0150] By using an adaptive weighting mechanism based on residual statistics, it can be ensured that even in scenarios with significant electromagnetic interference and degraded GNSS signal quality, continuous and stable navigation solutions can still be maintained by relying on the IMU and constraint factors.

[0151] Step 6 Control and Execution: The optimized pose and velocity solutions are input into the Model Predictive Controller (MPC). Based on the UAV's six-degree-of-freedom dynamics model and mission objective function, the optimal control input is generated online to complete sub-meter level hovering, path tracking, and anti-disturbance flight, enabling high-precision operations such as alignment imaging and X-ray detection in power transmission line inspection.

[0152] Specifically, the optimized pose and velocity solutions are input into the Model Predictive Controller (MPC), which generates the optimal control input based on the UAV's six-degree-of-freedom dynamics model and the mission objective function.

[0153]

[0154] In the formula, u * argmin represents the optimal control input, indicating the optimal control command for the UAV at the current moment and within a certain future timeframe, used to drive the UAV to achieve the desired trajectory tracking and flight mission; u is the control input variable, i.e., the target to be optimized; argmin u To find the control input u;p that minimizes the objective function t p represents the actual position of the UAV at time t. ref v is the reference position of the UAV at time t; t The actual velocity of the drone at time t; v ref u is the reference velocity of the UAV at time t; t Let t be the control input at time t; Q be the position error weight matrix; R be the velocity error weight matrix; S be the control input weight matrix; and T be the length of the prediction time window.

[0155] Through online optimization by the MPC controller, the drone can achieve sub-meter level hovering, path tracking, and anti-interference flight, ensuring the accuracy of alignment imaging and X-ray detection during power transmission line inspection.

[0156] like Figure 1 As shown, the overall process of this method covers the entire process from task configuration, sensor data acquisition and preprocessing, to navigation calculation, factor graph optimization, robust weighting mechanism, flight control execution, and health monitoring and failure handling. Specifically, the process first involves setting mission parameters and performing pre-flight self-checks on the ground side. Then, the UAV's onboard GNSS receiver and inertial measurement unit (IMU) complete the acquisition and time synchronization of multi-source measurements. Through the data preprocessing module, GNSS observations are processed for gross error removal and cycle slip detection, while IMU measurements undergo bias compensation and temperature drift correction, and low-dimensional motion constraints are formed through pre-integration.

[0157] In the tightly coupled error state extended Kalman filter (ESKF) stage, the raw GNSS observations and IMU pre-integrations jointly construct the observation model and generate residuals. After normalization and robust weighting, the state is updated to ensure stable solution even under electromagnetic interference and weak signal conditions. Furthermore, the sliding window factor graph optimization module combines GNSS factors, IMU pre-integration factors, and prior constraint factors to perform nonlinear least squares solutions, ensuring global consistency and high-precision positioning of the system in complex environments.

[0158] At the control execution level, this invention utilizes a Model Predictive Controller (MPC) to generate optimal control inputs, enabling real-time adjustments to trajectory tracking, hovering, and anti-disturbance flight. Simultaneously, a health monitoring module assesses the reliability of navigation calculations and adaptively switches between normal tightly coupled, weak GNSS-enhanced IMU-weighted, and pure inertial navigation modes based on signal level discrimination results, triggering safety alarms and emergency return-to-home when necessary. Finally, the system features status log recording and edge-to-edge collaborative update capabilities, facilitating post-task processing and model optimization.

[0159] Through the above steps, this high-precision autonomous navigation method for UAVs based on GNSS and IMU fusion achieves tight coupling between GNSS and IMU at the observation level. Combined with factor graph optimization and adaptive weighting mechanisms, it significantly improves the robustness and high-precision autonomous navigation capabilities of UAVs in complex electromagnetic environments. This method is particularly suitable for high-voltage transmission line detection scenarios, maintaining track continuity and positioning accuracy even under weak GNSS signal conditions or short-term loss of lock, providing reliable assurance for UAVs to perform high-precision operations.

[0160] This high-precision autonomous navigation method for UAVs based on GNSS and IMU fusion differs from loosely coupled result-level fusion by directly modeling and fusing raw GNSS observations, thus maintaining stronger robustness in weak signal and interference environments. By introducing prior knowledge such as power transmission corridor structural constraints and flight altitude limitations, the method improves solution stability in GNSS degradation scenarios. Simultaneously, the fusion of positioning results and predictive control works synergistically to achieve dynamic, disturbance-resistant hovering and path planning in complex environments. This method can be further fused with lidar, visual sensors, etc., to achieve multimodal cooperative navigation, demonstrating strong adaptability.

[0161] Furthermore, as a preferred technical solution in this embodiment, to further enhance the system's ability to suppress gross errors, in addition to simple threshold weighting, various robust kernel functions can be introduced in the weight adjustment stage to handle abnormal residuals under different scenarios. Specifically:

[0162] 1. Huber kernel function:

[0163] When the residual is small, maintain the double penalty; when the residual exceeds the threshold δ, switch to linear growth.

[0164]

[0165] Its corresponding weights are:

[0166]

[0167] In the formula, ρ(r) is the robust cost function, which defines the contribution of the residual to the optimization objective and suppresses the influence of abnormal residuals through nonlinear design; r is the residual, which represents the difference between the actual observed value and the predicted value, reflecting the consistency between the measurement and the model; w(r) is the weight function, which is derived from the derivative of the robust cost function and is used to adjust the contribution of each residual in the optimization; δ is the threshold parameter, which controls the boundary of the residual from the normal value to the abnormal value, and its value is usually determined according to the actual noise level or experience.

[0168] The Huber kernel function is suitable for scenarios with mild disturbances, and can reduce the impact of large residuals while preserving the weights of normal observations.

[0169] 2. Cauchy kernel function:

[0170] For large residuals, a slowly decreasing weight is used for suppression, defined as:

[0171]

[0172] The corresponding weighting function is:

[0173]

[0174] In the formula, c is a scaling parameter used to adjust the descent rate of the Cauchy kernel function and determine the degree of weight decay as the residual increases.

[0175] The Cauchy kernel function is more robust to outliers caused by strong electromagnetic interference, and can avoid the excessive influence of a single gross error on the optimization results.

[0176] 3. Tukey Biweight kernel function:

[0177] A secondary penalty is applied to residuals within the threshold δ range, but residuals exceeding the threshold are directly assigned a weight of zero to achieve complete removal.

[0178]

[0179] The corresponding weighting function is:

[0180]

[0181] The Tukey Biweight kernel function is particularly suitable for situations with severe electromagnetic interference and frequent gross errors, and can completely shield abnormal observations.

[0182] 4. Unified optimization format:

[0183] The weighted optimization problem after adopting a robust kernel function is uniformly represented as:

[0184]

[0185] In the formula, r GNSS,i It is the i-th GNSS observation residual, consisting of the difference between pseudorange, carrier phase, or Doppler measurement and prediction; r IMU,j It is the j-th IMU pre-integration residual, reflecting the deviation between the pose predicted based on inertial measurements and the actual estimate; r constraint,k is the k-th external constraint residual, introduced by prior conditions; X is the set of states within the sliding window, including system state variables such as position, velocity, attitude quaternions, and zero bias of accelerometers and gyroscopes; ρ(·) is any of the above robust kernel functions.

[0186] By introducing different robust kernel functions, an appropriate outlier handling mechanism can be flexibly selected based on the mission environment. The Huber kernel function can be used in scenarios with mild electromagnetic interference, the Cauchy kernel function in scenarios with strong but discontinuous interference, and the Tukey kernel function for complete outlier removal in extremely complex environments. This alternative design ensures that the patented solution maintains good navigation robustness and accuracy under various complex electromagnetic environments.

[0187] Furthermore, to verify the feasibility and effectiveness of this high-precision autonomous navigation method for UAVs based on GNSS and IMU fusion, the following example illustrates the process using the "UAV inspection of a 500kV high-voltage transmission line corridor in Heilongjiang Province" mission:

[0188] In terms of hardware configuration, the UAV platform for this UAV inspection mission is a hexacopter electric UAV, equipped with a GNSS receiver, inertial measurement unit (IMU), flight controller (FC), embedded fusion processing module, power supply module, and mission payload module. The GNSS receiver is connected to a multi-frequency, multi-mode antenna via an RF cable, capable of simultaneously receiving GPS, BeiDou, GLONASS, and Galileo satellite signals, and outputting pseudorange, carrier phase, and Doppler shift observations. The IMU consists of a three-axis accelerometer and a three-axis gyroscope, connected to the fusion processing module via a high-speed SPI bus, outputting acceleration and angular velocity in real time. Observational data from both the GNSS receiver and IMU are input to the embedded fusion processing module, which integrates a high-performance ARM processor and an FPGA coprocessor unit for executing Error State Extended Kalman Filter (ESKF) and factor graph optimization algorithms. The output of the fusion processing module is transmitted to the flight controller via a serial interface as navigation input for the control law. The power supply module provides a stable DC power supply to the GNSS receiver, IMU, and fusion processing module, and ensures power stability under electromagnetic interference environments through a voltage regulator. The mission payload module includes an X-ray imager and a high-definition camera, which are connected to the flight controller via a CAN bus for circuit defect detection and image assistance.

[0189] In terms of implementation, the navigation process of this UAV inspection mission includes the following key steps:

[0190] First, during the UAV takeoff phase, the GNSS receiver begins receiving signals from multiple constellations, the IMU collects initial acceleration and angular velocity data in real time, and the fusion processing module completes initial attitude alignment. During the flight inspection phase, the IMU continuously integrates to obtain short-term velocity and displacement estimates for the aircraft, and the raw pseudorange and carrier phase observations output by the GNSS are fed into the fusion module to establish a tightly coupled observation model. Taking a typical operation scenario in a power transmission corridor in the suburbs of Harbin as an example, when crossing woodlands and power transmission towers, some GNSS satellite signals are blocked, and the calculation error of traditional loosely coupled methods quickly expands to several meters. However, this UAV inspection task, through factor graph optimization, jointly constrains GNSS, IMU, and corridor geometric prior information within a sliding window, maintaining a positioning accuracy within 0.3 meters.

[0191] In handling anomaly observations, this UAV inspection mission introduces a robust kernel function mechanism: when the GNSS pseudorange residual exceeds the normalization threshold, the fusion module automatically assigns a lower weight to the observation or removes it directly, while using IMU pre-integration and altitude constraints to maintain continuous solution. When encountering strong electromagnetic interference, such as when the UAV is close to a high-voltage power line for X-ray imaging, cycle slips may occur in some carrier phase observations. An adaptive weighting mechanism effectively suppresses the impact of abnormal measurements, ensuring that the overall navigation solution is not disrupted by single-point interference.

[0192] In the control and execution phase, the high-precision pose information output by the fusion processing module is sent to the model predictive controller (MPC), which, combined with the UAV's six-degree-of-freedom dynamic equations, generates the optimal control input to achieve real-time adjustment of the propeller speed.

[0193] Experimental results show that the UAV inspection mission can maintain a stable attitude within ±2° even in wind speeds of 8 m / s, and achieve close-range hovering of power transmission lines, ensuring the alignment accuracy of X-ray images.

[0194] In summary, this application example verifies that the high-precision autonomous navigation method for UAVs based on GNSS and IMU fusion can achieve continuous, stable and high-precision autonomous navigation in the complex electromagnetic environment of power transmission lines, providing reliable technical support for high-risk scenarios such as power line inspection.

[0195] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A high-precision autonomous navigation method for unmanned aerial vehicles based on GNSS and IMU fusion, characterized in that: The method is based on a GNSS and IMU fusion unmanned aerial vehicle high-precision autonomous navigation system, which comprises an airborne sensor module, a fusion processing module, a control execution module, a task load module and a power module. The airborne sensor module comprises a GNSS receiver and an inertial measurement unit (IMU). The GNSS receiver receives satellite signals through a multi-frequency multi-mode antenna and outputs raw pseudorange, carrier phase and Doppler observation. The IMU is composed of a three-axis accelerometer and a three-axis gyroscope, which outputs the linear acceleration and angular velocity of the aircraft in real time. The airborne sensor module collects the raw observation data required for navigation and transmits it to the fusion processing module. The fusion processing module comprises an embedded high-performance processor, which is based on the tight coupling algorithm framework of error state extended Kalman filter (ESKF) and directly models the GNSS raw observation and IMU integral results in the observation layer. It also introduces factor graph optimization to jointly estimate the GNSS observation, IMU pre-integral and external constraints within a sliding window, and outputs the fusion positioning results to the control execution module. The control execution module comprises a flight controller (FC) and a UAV power system. It generates optimal control commands based on the navigation solution and drives the UAV to fly. The task load module comprises an X-ray imaging device and auxiliary sensors, which communicate with the flight controller through CAN bus and use the high-precision position and attitude information provided by the navigation system to scan the power transmission line accurately. At the same time, it provides structural feature constraints to the navigation as external constraint factors, which are input into the fusion processing module to enhance the stability of the navigation solution. The power module is connected to each module through independent power supply lines and provides stable DC power for the GNSS receiver, IMU, fusion processing module, flight controller and other modules.

2. The GNSS and IMU fusion based high-precision autonomous navigation method for unmanned aerial vehicles according to claim 1, characterized in that: The method comprises the following steps: Step 1: Data acquisition: Obtain the raw observation data of global navigation satellite system (GNSS) and inertial measurement unit (IMU), introduce multi-constellation fusion and observation redundancy checking mechanism, and eliminate gross errors; Step 2: IMU pre-integration: Use IMU data to pre-integrate the motion state of the UAV, generate low-dimensional motion constraints, and reduce the state dimension under the premise of ensuring real-time; Step 3: Tight coupling fusion: Model the GNSS raw measurement and IMU pre-integration results together, construct the joint observation equation, and realize the prediction-update recursion through error state extended Kalman filter (ESKF); Step 4: Factor graph optimization: Within a sliding time window, construct a factor graph model containing GNSS factors, IMU pre-integration factors and external constraint factors, and solve it iteratively using nonlinear least squares optimization to maintain high-precision long-term navigation in GNSS weak signal or short-time lock loss situations; Step 5: Abnormality detection and adaptive weighting: Calculate the GNSS factor residual and perform normalization, combine threshold discrimination and robust kernel function to dynamically adjust the observation weight, rewrite the optimization problem as a weighted least squares form, suppress the influence of abnormal observations in electromagnetic interference or GNSS signal degradation scenarios, and maintain continuous and stable navigation solution. Step 6 control and execution: the optimized pose and velocity solution is input into the model predictive controller MPC, which generates optimal control input based on the six-degree-of-freedom dynamics model of the UAV and the task objective function, to complete sub-meter hovering, path tracking and disturbance rejection flight, and to complete high-precision tasks such as image mapping and X-ray detection in power line inspection. 3.The GNSS and IMU fusion based high-precision autonomous navigation method for UAVs according to claim 2, characterized in that: In step 1, the GNSS receiver carried by the UAV synchronously receives multi-constellation signals through a multi-frequency multi-mode antenna, and outputs the pseudo-range observation ρ i , carrier phase φ i and Doppler frequency shift f D,i of the i-th satellite; the inertial measurement unit IMU collects three-axis acceleration a m and angular velocity ω m in real time; In the data acquisition phase, the observation noise is denoted as: the noise term of pseudorange observation ρ , the noise term of carrier phase observation φ , the noise term of Doppler shift observation f , multi-constellation fusion and observation redundancy checking mechanism are introduced to eliminate gross errors.

4. The GNSS and IMU fusion based high-precision autonomous navigation method for unmanned aerial vehicles according to claim 2, characterized in that: In step 2, the motion state of the UAV is pre-integrated using the IMU data within a continuous time period [t k ,t k+1 ], and the calculation formula is as follows: where ΔR k is the attitude change increment; Δv k is the velocity change increment; is the gyroscope raw measurement at the jth time instant; is the accelerometer raw measurement at the jth time instant; b g is the gyroscope bias; b a is the accelerometer bias; n g is the gyroscope random noise; n a is the accelerometer random noise; Δt is the sampling time interval; R j is the attitude rotation matrix at the jth time instant; k is the discrete time instant index on the time axis.

5. The GNSS and IMU fusion based high-precision autonomous navigation method for UAV according to claim 2, characterized in that: In step 3, the joint observation equation is constructed to model the GNSS raw measurements and the IMU pre-integration results together: where z i is the observation vector at the i-th time instant; ρ i is the pseudorange observation of the i-th satellite of GNSS measurements; φ i is the carrier phase of the i-th satellite of GNSS measurements; f D,i is the Doppler shift of the i-th satellite of GNSS measurements; h(x k , p i ) is the observation model function that maps the system state vector x k and the satellite position p i to the predicted observation; x k = [p k , v k , q k , b a , b g ] is the system state vector, containing the state information of the UAV at the i-th time instant, p k is the position of the UAV, v k is the velocity of the UAV, q k is the attitude quaternion of the UAV, b a is the accelerometer bias, reflecting the systematic error of the IMU sensor; b g is the gyroscope bias, reflecting the systematic error of the IMU sensor; p i is the spatial position of the i-th satellite; v i is the observation noise of the i-th satellite.

6. The GNSS and IMU fusion based high-precision autonomous navigation method for UAV according to claim 2, characterized in that: In step 4, the mathematical expression of the factor graph optimization is: min X ∑‖r GNSS ‖ 2 +∑‖r IMU ‖ 2 +∑‖r constraint ‖ 2 ; where x is the optimization variable, representing the state set of poses and biases within the sliding window; r GNSS is the GNSS observation residual, representing the difference between GNSS measurements and the current state prediction; r IMU is the IMU pre-integration residual, representing the difference between IMU integration results and the current state prediction; r constraint is the external constraint residual, including prior knowledge such as height limit, flight corridor geometry constraint, dynamics consistency condition, etc.

7. The GNSS and IMU fusion based high-precision autonomous navigation method for UAV according to claim 2, characterized in that: In step 5, the abnormality detection is to identify the outliers in the GNSS observations, and the abnormality determination is completed through residual calculation, normalization processing and threshold judgment; First, in each optimization iteration, the residual of the GNSS factor is computed: r i = z i - h(x, p i ); Then, the residual is normalized: Finally, threshold discrimination is performed: set a confidence interval threshold τ, if then the observation is determined to be an outlier with a probability of P>99.7%. wherein r i is the residual of the i-th GNSS observation, representing the difference between the actual observation and the predicted value; z i is the i-th GNSS observation, output directly by the GNSS receiver; h(x, p i ) is the predicted value based on the current state estimate x and satellite position p i ; x is the current state estimate vector, including the position, velocity, attitude of the UAV and sensor bias; p i is the spatial position of the i-th satellite; is the normalized residual; σ i is the standard deviation estimate of the observation, calculated from historical data statistics or provided by the GNSS receiver accuracy indicator; the common choice is τ = 3, corresponding to the 99.7% confidence interval under the Gaussian distribution.

8. The GNSS and IMU fusion based high-precision autonomous navigation method for UAV according to claim 7, characterized in that: In step 5, the adaptive weighting mechanism is to dynamically adjust the weight according to the abnormality detection results, and to introduce the weighting mechanism into the optimization problem, so that the system can still maintain stable solution under abnormal conditions; Introducing a robust kernel function to adjust the weights w i : The final optimization problem is rewritten as a weighted least squares form: where w i is the weight of the i-th observation; x is the optimization variable, representing the state collection of poses and biases within the sliding window; w i is the adaptive adjustment weight of the i-th GNSS observation; r GNSS,i is the difference between the actual value and the predicted value of the i-th GNSS observation, r GNSS,i = z i - h(x k , p i ), z i is the observation vector at the i-th time, h(x k , p i ) is the observation model function, mapping the system state vector x k and satellite position p i to the predicted observation value; r IMU,j is the IMU pre-integration residual; r constraint,k is the external constraint residual.

9. The GNSS and IMU fusion based high-precision autonomous navigation method for UAV according to claim 2, characterized in that: In step 6, the optimized pose and velocity solution is input into the model predictive controller MPC, which generates optimal control input based on the six-degree-of-freedom dynamics model of the UAV and the task objective function, and the mathematical expression is: In the formula, u * is the optimal control input, representing the optimal control command of the UAV at the current time and in the future for a period of time, used to drive the UAV to achieve the desired trajectory tracking and flight tasks; u is the control input variable, that is, the target to be optimized; argmin u is to find the control input u that minimizes the objective function; p t is the actual position of the UAV at the t time; p ref is the reference position of the UAV at the t time; v t is the actual speed of the UAV at the t time; v ref is the reference speed of the UAV at the t time; u t is the control input at the t time; Q is the position error weight matrix; R is the speed error weight matrix; S is the control input weight matrix; and T is the length of the prediction time window.

10. The GNSS and IMU fusion based high-precision autonomous navigation method for UAV according to claim 2, characterized in that: In step 5, multiple robust kernel functions are introduced to process abnormal residuals in different scenarios; Huber kernel function: when the residual is small, keep quadratic penalty; when the residual exceeds the threshold δ, switch to linear growth, defined as: The corresponding weight is: Cauchy kernel function: slow weight suppression for large residuals, defined as: The corresponding weight function is: Tukey Biweight kernel function: quadratic penalty for residuals within the threshold δ, but for residuals exceeding the threshold, the weight is directly set to zero to completely eliminate, defined as: The corresponding weight function is: The weighted optimization problem after using the robust kernel function is uniformly represented as: In the formula, ρ(r) is the robust cost function, which is used to define the contribution of the residual to the optimization objective, and the influence of abnormal residuals is suppressed through nonlinear design; r is the residual, representing the difference between the actual observation and the prediction, reflecting the consistency between the measurement and the model; w(r) is the weight function, derived from the derivative of the robust cost function, used to adjust the contribution of each residual in the optimization; δ is the threshold parameter, used to control the boundary of the transition from normal to abnormal value of the residual, which is usually determined according to the actual noise level or experience; c is the scale parameter, used to adjust the falling speed of the Cauchy kernel function, determining the attenuation degree of the weight with the increase of the residual; r GNSS,i is the i-th GNSS observation residual, composed of the difference between the pseudo-range, carrier phase or Doppler measurement and the predicted value; r IMU,j is the j-th IMU pre-integration residual, reflecting the deviation between the predicted pose based on inertial measurement and the actual estimation; r constraint,k is the k-th external constraint residual, introduced by the prior condition; X is the state set in the sliding window, including position, velocity, attitude quaternion, accelerometer and gyroscope bias, etc. system state variables; ρ(·) is any of the above robust kernel functions.

Images