A combined navigation positioning method based on state-aware virtual odometer enhancement

Abstract

The application discloses a kind of based on state perception virtual odometer enhanced combination navigation positioning method, by constructing the regression model of convolution neural network and the fusion of Transformer, the original data of inertial measurement unit is modeled, the virtual estimation of vehicle three-dimensional speed is realized;For different motion states of vehicle, state perception mechanism is introduced to adaptively adjust speed regression error to improve the reliability of virtual odometer under complex dynamic conditions.At the same time, the network model hyperparameter is optimized and light weight design by using optimization algorithm;Finally, the virtual speed obtained is introduced into GNSS / INS integrated navigation filter as observation information, and the inertial navigation error is effectively suppressed when GNSS signal fails.The experimental results show that the method can significantly improve the accuracy and robustness of integrated navigation positioning in complex urban environment, and realize continuous and stable vehicle positioning.

Description

Technical Field

[0001] This invention belongs to the field of GNSS (Global Navigation Satellite System) positioning and navigation technology, and relates to a combined navigation and positioning method in complex environments, specifically a combined navigation and positioning method based on state-aware virtual odometry enhancement. Background Technology

[0002] In modern intelligent transportation, autonomous driving, aerospace and other fields, the combination of Global Navigation Satellite System (GNSS) and Inertial Navigation System (INS) has become a core supporting technology. Reliable high-precision positioning is not only a prerequisite for safe vehicle operation and automatic control, but also a key foundation for high-precision map construction and optimal allocation of road resources.

[0003] However, in complex environments such as urban canyons, tunnels, and tree-lined roads, GNSS signals are susceptible to obstruction, interference, and multipath effects, and may even experience signal loss, leading to positioning interruptions. Ensuring the accuracy and stability of vehicle navigation in scenarios where GNSS signals are limited or unavailable is a crucial problem that urgently needs to be solved by the academic and engineering communities.

[0004] To suppress the error divergence of INS when GNSS fails, existing technologies have proposed an auxiliary strategy combining non-holonomic constraints (NHC) and odometry—using vehicle motion constraints to construct three-dimensional velocity constraints to maintain navigation performance. However, these methods typically rely on additional wheel speed encoders or onboard bus speed information, increasing system cost and integration complexity. Furthermore, improper calibration of the inertial measurement unit's mounting angle can reduce the effectiveness of constraints, leading to degraded navigation performance.

[0005] With the development of artificial intelligence technology, deep learning-based methods have been used to overcome the limitations of traditional integrated navigation, including learning-based virtual odometry and learning models directly used for integrated positioning. However, existing solutions have a core shortcoming: the accuracy of virtual speed is crucial to maintaining the performance of integrated positioning. In complex dynamic scenarios such as vehicles moving straight, turning, or stopping, model-based speed constraints such as NHC and learning-based speed constraints are prone to inconsistencies. If both are used in the integrated positioning filter, it can lead to a decrease in positioning performance or even failure.

[0006] Therefore, there is an urgent need for a technical solution that can adapt to different vehicle motion states and coordinate multiple speed constraints in order to improve the stability and accuracy of vehicle integrated navigation in GNSS failure scenarios. Summary of the Invention

[0007] To address the aforementioned issues, this invention discloses a combined navigation and positioning method based on state-aware virtual odometry enhancement. By constructing a regression model fused with a convolutional neural network and a Transformer, the original data from the inertial measurement unit (INS) is modeled to achieve virtual estimation of the vehicle's three-dimensional velocity. For different vehicle motion states, a state-aware mechanism is introduced to adaptively adjust the velocity regression error, thereby improving the reliability of the virtual odometry under complex dynamic conditions. Simultaneously, optimization algorithms are employed to optimize and lightweight the hyperparameters of the network model. Finally, the obtained virtual velocity is used as observation information and introduced into the GNSS / INS combined navigation filter, effectively suppressing inertial navigation errors when the GNSS signal fails.

[0008] To achieve the above objectives, the technical solution of the present invention is as follows:

[0009] A state-aware virtual odometry-enhanced integrated navigation and localization method is proposed. First, a hybrid regression network combining CNN and Transformer is constructed to enhance vehicle navigation. CNN is used to extract local temporal features from the original IMU observations, while Transformer, leveraging its powerful long-range dependency modeling capabilities, can achieve high-precision prediction of the vehicle's three-dimensional velocity.

[0010] Secondly, a state-aware module was designed to adaptively adjust the loss function of the hybrid regression network during the training phase. Compared with fixed weights, the state-aware adaptive weighted loss function allows the hybrid regression network to consider different behavioral characteristics of the vehicle, thereby improving the accuracy and robustness of the virtual speed. Simultaneously, optimization algorithms were employed to optimize and lightweight the hyperparameters of the network model.

[0011] Finally, the virtual velocity based on the hybrid network model is applied to the GNSS / INS combined positioning filter, replacing the original non-integrity constraint (NHC) observation equation to suppress positioning errors during GNSS outages. Furthermore, multiple tests were conducted using open-source data to verify the effectiveness of the proposed method.

[0012] The beneficial effects of this invention are as follows:

[0013] The proposed integrated navigation and positioning method based on state-aware virtual odometry enhances the accuracy and robustness of virtual velocity prediction in GNSS failure scenarios by leveraging the hybrid regression capabilities of CNN and Transformer, along with state-aware adaptive loss adjustment. It adapts to various vehicle motion states, including straight-line driving and turning. In complex environments such as urban canyons and tree-lined roads, this method achieves more stable navigation and positioning. Compared to pure inertial navigation, traditional constraints, and conventional network models, its positioning accuracy is improved by 89.2%, 52.7%, and 21.2%, respectively. Attached Figure Description

[0014] Figure 1 This is a system flowchart of the present invention;

[0015] Figure 2 This is a comparison diagram of the positioning results of linear motion using the method described in this invention under GNSS failure conditions;

[0016] Figure 3 This is a comparison chart of the turning motion positioning results of the method described in this invention under GNSS failure environment;

[0017] Figure 4 This is a comparison chart of the positioning results of continuous turning motion under GNSS failure environment using the method described in this invention. Detailed Implementation

[0018] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0019] like Figure 1 As shown in the figure, this example discloses a combined navigation and positioning method based on state-aware virtual odometry enhancement. The specific steps are as follows:

[0020] Step 1: Obtain the raw triaxial acceleration and triaxial angular velocity data output by the vehicle-mounted inertial measurement unit (IMU), perform time synchronization, noise reduction and sliding window segmentation on the IMU data, and construct an inertial data sequence for network input;

[0021] The IMU data is processed by time synchronization, noise reduction, and sliding window segmentation to form an input sequence of length T.

[0022] Step 2: Construct a hybrid regression network consisting of a Convolutional Neural Network (CNN) and a Transformer network to extract spatial features and perform temporal modeling on the IMU data sequence. Use prepared input samples and labels to learn the nonlinear mapping between IMU data and three-dimensional velocity.

[0023] The construction process of the hybrid regression network includes:

[0024] CNN is used to extract multi-scale features from the raw IMU data within the sliding window to obtain the features of the IMU data; the extracted feature sequence is input into the Transformer encoder, and the long-term temporal dependency of the inertial data is modeled through a multi-head self-attention mechanism; a regression layer is set at the output of the Transformer to output the three-dimensional velocity in the vehicle coordinate system.

[0025] Step 3: Determine the vehicle's motion state information and construct a state perception loss adjustment module. During model training, adaptively weight the regression errors of different speed components to improve the accuracy and stability of virtual odometry estimation under complex motion conditions.

[0026] The state-aware loss adjustment module includes:

[0027] Based on the triaxial acceleration and triaxial angular velocity data output by the inertial measurement unit (IMU), the current motion state of the vehicle is determined, and the three-dimensional velocity regression error is adaptively weighted according to the determination results; in this process, a state-aware weighting coefficient related to the vehicle's motion state is introduced. , ,and These are used to adjust the weights of the regression errors of forward velocity, lateral velocity, and vertical velocity in the loss function, respectively, and are defined as follows:

[0028] ;

[0029] in, , and These correspond to the weighting coefficients for forward, lateral, and vertical velocity errors, respectively. and These represent the acceleration and angular velocity measurements from the six-axis IMU, respectively. Parameters and These are adaptive variables and can be set according to the actual vehicle model. The variables straight, turning, and others represent the state classification results, used to indicate the current motion state of the vehicle.

[0030] The determination of vehicle motion state is based on the statistical characteristics of IMU angular velocity measurement data, and is achieved by calculating the variance of the observed data within the sliding time window. The calculation method is shown in the following formula:

[0031] ;

[0032] in, The window length can be set to the number of IMU data samples taken within 1 second. This is the angular velocity measurement value from the IMU. This represents the observation noise level for that axis; The hypothesis test value is 1. When the value is close to 1, the direction is straight; when it is far above 1, the direction is turning.

[0033] Based on this, a state-aware weighted loss function is constructed to train the virtual odometry speed prediction network, and its definition is shown in the following formula:

[0034] ;

[0035] Where N represents the number of samples. , and This represents the difference between the speed predicted by the network and the corresponding true value. The design of the loss function reflects the different accuracy requirements of each speed component under different vehicle motion states, thereby improving the speed prediction accuracy of the virtual odometry network.

[0036] Step 4: Introduce the virtual odometer's three-dimensional velocity as an observation into the GNSS (Global Navigation Satellite System) / INS (Inertial Navigation System) combined navigation filter. When the GNSS signal fails or degrades, the INS error is constrained and corrected to obtain continuous and stable vehicle positioning results.

[0037] The process of introducing a GNSS / INS integrated navigation filter into the virtual odometry. First, the state vector is selected:

[0038] ;

[0039] in, , and These are position, velocity, and attitude errors, respectively. and Zero bias error of gyroscope and accelerometer respectively and These are the proportional factor errors for the gyroscope and accelerometer, respectively. Then, integrated navigation state prediction is performed:

[0040] ;

[0041] in, Let be the state vector from the previous moment. The state transition matrix is ​​the state transition matrix between adjacent time steps. This represents state process noise. Next, using the virtual three-dimensional velocity as a pseudo-observation, a velocity observation equation is constructed:

[0042] ;

[0043] in, To observe the residual vector, the velocity predicted by the network model and inertial recursive velocity Subtract, and take into account the installation arm. Influence; This is the velocity observation coefficient matrix; For velocity observation noise. When GNSS fails, based on the virtual velocity observations predicted by the proposed network model, the cumulative error constraint and update of the GNSS / INS combined system can be achieved according to the following formula, resulting in the updated state as follows:

[0044] ;

[0045] in, This is the updated state vector; The filter gain coefficient is obtained by weighing the system's prediction variance and observation noise. In summary, the virtual velocity observation is used to suppress inertial navigation error divergence, thereby improving the positioning continuity and robustness of the integrated navigation system.

[0046] Through the above implementation methods, this invention, by leveraging the hybrid regression capabilities of CNN and Transformer, and state-aware adaptive loss adjustment, can improve the prediction accuracy and robustness of virtual velocity in GNSS failure scenarios, adapting to various vehicle motion states such as straight-line travel and turning. In complex environments such as urban canyons and tree-lined roads, this method can achieve more stable navigation and positioning.

[0047] like Figure 2 , 3 As shown in Figure 4, compared with pure inertial navigation (Scheme 1), traditional constraints (Scheme 2), and conventional network models (Scheme 3 and 4), the positioning accuracy of this scheme (Scheme 5) is improved by 89.2%, 52.7%, and 21.2%, respectively.

[0048] It should be noted that the above content merely illustrates the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. For those skilled in the art, various improvements and modifications can be made without departing from the principle of the present invention, and all such improvements and modifications fall within the scope of protection of the claims of the present invention.

Claims

1. A combined navigation and positioning method based on state-aware virtual odometry enhancement, characterized in that, include: Step 1: Obtain the raw triaxial acceleration and triaxial angular velocity data output by the vehicle-mounted inertial measurement unit (IMU), perform time synchronization, noise reduction, and sliding window segmentation on the IMU data, and construct an inertial data sequence for network input; Step 2: Construct a hybrid regression network consisting of a convolutional neural network (CNN) and a Transformer network to extract spatial features and perform temporal modeling on the IMU data sequence. Use prepared input samples and labels to learn the nonlinear mapping between IMU data and three-dimensional velocity. Step 3: Determine the vehicle's motion state information and construct a state perception loss adjustment module. During model training, adaptively weight the regression errors of different speed components to improve the accuracy and stability of virtual odometry estimation under complex motion conditions. Step 4: Introduce the virtual odometer's three-dimensional velocity as an observation into the GNSS / INS integrated navigation filter. When the GNSS signal fails or degrades, constrain and correct the INS error to obtain continuous and stable vehicle positioning results.

2. The integrated navigation and positioning method based on state-aware virtual odometry enhancement as described in claim 1, characterized in that, Step 1 involves performing time synchronization, noise reduction, and sliding window segmentation on the IMU data to form an input sequence of length T.

3. The integrated navigation and positioning method based on state-aware virtual odometry enhancement as described in claim 1, characterized in that, Step 2 involves constructing the hybrid regression network, which includes: CNN is used to extract multi-scale features from the raw IMU data within the sliding window to obtain the features of the IMU data; the extracted feature sequence is input into the Transformer encoder, and the long-term temporal dependency of the inertial data is modeled through a multi-head self-attention mechanism; a regression layer is set at the output of the Transformer to output the three-dimensional velocity in the vehicle coordinate system.

4. The integrated navigation and positioning method based on state-aware virtual odometry enhancement as described in claim 1, characterized in that, The state-aware loss adjustment module described in step 3 includes: Based on the triaxial acceleration and triaxial angular velocity data output by the inertial measurement unit (IMU), the current motion state of the vehicle is determined, and the three-dimensional velocity regression error is adaptively weighted according to the determination results; in this process, a state-aware weighting coefficient related to the vehicle's motion state is introduced. , ,and These are used to adjust the weights of the regression errors of forward velocity, lateral velocity, and vertical velocity in the loss function, respectively, and are defined as follows: ； in, , ,and These correspond to the weighting coefficients for forward, lateral, and vertical velocity errors, respectively. and These represent the angular velocity and acceleration measurements from the six-axis IMU, respectively; parameters and These are adaptive variables, set according to the actual vehicle model; the variables straight, turning, and others represent the state classification results, used to indicate the current motion state of the vehicle. The determination of vehicle motion state is based on the statistical characteristics of IMU angular velocity measurement data, and is achieved by calculating the variance of the observed data within the sliding time window. The calculation method is shown in the following formula: ； in, The window length is the number of IMU data samples taken within 1 second. This is the angular velocity measurement value from the IMU. This represents the observation noise level for that axis; This is the hypothesis test value; a value close to 1 indicates going straight, while a value far exceeding 1 indicates turning. Based on this, a state-aware weighted loss function is constructed to train the virtual odometry speed prediction network, and its definition is shown in the following formula: ； Where N represents the number of samples, , and This represents the difference between the speed predicted by the network and the corresponding true value; the design of the loss function reflects the different accuracy requirements of each speed component under different vehicle motion states, thereby improving the speed prediction accuracy of the virtual odometer network.

5. The integrated navigation and positioning method based on state-aware virtual odometry enhancement as described in claim 1, characterized in that, Step 4 describes the process of introducing the GNSS / INS integrated navigation filter into the virtual odometry; first, the state vector is selected: ； in, , and These are position, velocity, and attitude errors, respectively. and Zero bias error of gyroscope and accelerometer respectively and The errors are calculated for the gyroscope and accelerometer scaling factors, respectively; then, integrated navigation state prediction is performed. ； in, Let be the state vector from the previous moment. The state transition matrix is ​​the state transition matrix between adjacent time steps. To address the noise in the state process, we then use the virtual three-dimensional velocity as a pseudo-observation to construct the velocity observation equation: ； in, To observe the residual vector, the velocity predicted by the network model and inertial recursive velocity Subtract, and take into account the installation arm. Influence; This is the velocity observation coefficient matrix; To account for velocity observation noise; when GNSS fails, based on the virtual velocity observations predicted by the proposed network model, the cumulative error constraint and update of the GNSS / INS combined system are implemented according to the following formula, resulting in the updated state as follows: ； in, This is the updated state vector; The filter gain coefficient is obtained by weighing the system's prediction variance and observation noise.

Images