CN-121977588-A - Vehicle-mounted integrated navigation method based on self-adaptive left-invariable extended Kalman filtering

Abstract

The invention discloses a vehicle-mounted integrated navigation method based on self-adaptive left-invariable extended Kalman filtering, which is used for solving the problems of GNSS signal interruption, multipath interference and inertial navigation error accumulation in a complex urban environment. The method adopts a Litsea SE (3) to represent the motion state of a carrier, constructs a left unchanged error model, ensures estimation consistency from a geometric plane, designs a self-adaptive noise estimation mechanism based on sliding window innovation covariance matching, adjusts the statistical characteristics of the GNSS and zero-speed corrected observation noise on line, realizes multi-source observation tight coupling fusion under a left unchanged filter frame, improves estimation precision and robustness under a dynamic environment, and combines a gravity alignment error monitoring and self-adaptive resetting strategy to ensure reliable operation under extreme observation conditions. The invention obviously improves the positioning precision and the environmental adaptability of the vehicle-mounted GNSS/INS system in high-challenge scenes such as urban canyons, tunnels and the like on the premise of not increasing the hardware cost.

Inventors

• YU ZHILONG
• ZHANG YU

Assignees

• 哈尔滨理工大学

Dates

Publication Date

20260505

Application Date

20260123

Claims (3)

1. 1. The vehicle-mounted integrated navigation method based on the self-adaptive left unchanged expansion Kalman filtering is characterized by comprising the following steps of: step 1, synchronously acquiring and preprocessing multi-source sensor data; Step 2, constructing a left unchanged filtering state model based on a Liqun theory; step 3, performing state prediction based on a left invariant extended Kalman filtering framework; Step 4, designing and executing self-adaptive observation noise covariance estimation; Step 5, updating the left unchanged fusion of the multi-source observation information; and 6, monitoring the running state of the system in real time and recovering the system in a self-adaptive manner.
2. 2. The method of claim 1, wherein the left invariant error in step is defined as: Wherein, the Indicating the state of the real prune group, Representing the estimated state of the lie group, which error can be approximated as a vector over the corresponding lie algebra for small errors The dynamics equation is an autonomous system and a state estimation value The method of claim 1, wherein the adaptive observed noise covariance estimation in step 4 comprises defining the observed innovation at time k as Wherein the method comprises the steps of As a result of the actual observation value, In order to observe the model, the model is, Maintaining a sliding window with the length of N, and calculating an actual sample covariance matrix of the innovation sequence in the window: calculating a theoretical innovation covariance matrix: Wherein the method comprises the steps of In order to observe the matrix, In order to predict the error covariance matrix, For nominal observation noise covariance matrix by matching And (3) with Inverse solution of the observed noise covariance matrix estimated value after current time adjustment And uses a clipping function to ensure that its adjustment amplitude is within a reasonable range.
3. 3. Method according to any of claims 1 and 2, characterized in that it explicitly calibrates and compensates the mounting rotation angle of the inertial measurement unit with respect to the carrier coordinate system and the amount of translational lever arm on-line as part of a state variable, the total dimension of which is not lower than 21 dimensions, during the state estimation.

Description

Vehicle-mounted integrated navigation method based on self-adaptive left-invariable extended Kalman filtering Technical Field The invention relates to the technical fields of vehicle navigation, sensor information fusion and advanced control, in particular to a high-precision and high-robustness combined navigation method suitable for intelligent automobiles and automatic driving vehicles. In particular, the invention provides a vehicle-mounted combined navigation solution of a depth fusion left unchanged filtering theory and a self-adaptive estimation technology aiming at technical challenges of unstable signal quality of a global navigation satellite system, quick accumulation of errors of an inertial navigation system and the like in a complex urban driving environment. Background The positioning navigation system with high precision and high reliability is a basic premise for the intelligent automobile to realize environment sensing, path planning and decision control. At present, the combined navigation of a global navigation satellite system and an inertial navigation system is a mainstream scheme of vehicle-mounted positioning. GNSS can provide global coverage, position, velocity and time information in an absolute coordinate system, but its signals are susceptible to building shadowing, multipath reflection, ionospheric interference, etc., and signal attenuation, discontinuity or even complete failure can occur in typical scenarios such as urban canyons, tunnels, underground parking lots, etc. The INS works autonomously based on the inertial sensor and does not depend on external signals, and the INS has the advantages of being high in short-time precision, high in output frequency and high in anti-interference capability, and can provide continuous navigation information in a GNSS failure period. However, INS, especially inertial measurement units based on mems, integrate over time the sensor zero bias, scale factor error, etc., resulting in unbounded divergence of the navigation error, and inability to independently support accurate navigation for long periods of time. In order to overcome the respective defects, the GNSS and the INS are combined through an information fusion technology to form a system with complementary advantages. The traditional fusion algorithm is mostly based on an extended kalman filter framework. However, EKF requires a first order taylor expansion approximation at the operating point when dealing with strong nonlinear systems such as gesture motion, and this linearization process introduces truncation errors. When the carrier is maneuvered over large angles or the initial errors are large, the linearization errors may lead to reduced filter estimation performance and even to covariance matrix distortion and filter divergence problems, so-called "inconsistency" problems. On the other hand, the observation noise in the actual vehicle-mounted environment is not steady-state white gaussian noise. The observation precision of GNSS changes in real time along with the satellite geometry, signal intensity and multipath effect, and the motion states of vehicle start-stop, jolt and the like also enable the reliability of virtual observation such as zero-speed correction to change dynamically. The adoption of a fixed noise statistical model can not accurately describe the time-varying characteristic, so that the filter can not make optimal balance when observing quality mutation, and the overall robustness of the system in a complex environment is reduced. In the prior art, studies have been attempted to improve the performance of nonlinear estimation or to enhance noise adaptation, respectively. For example, unscented Kalman filtering or particle filtering is adopted to better solve the nonlinear problem, but the calculation load is heavy, and research is also conducted to introduce Sage-Husa adaptive filtering or strong tracking filtering to adjust the noise covariance on line, but the methods are usually built on the state representation of Euclidean space, so that the problem of geometric consistency of state estimation cannot be fundamentally solved. In recent years, the left-invariant extended Kalman filtering theory based on Liqun lie algebra provides a new idea for nonlinear system state estimation. The theory defines the system state on a proper Liqun manifold, and designs the error with unchanged property, so that an error dynamics equation is irrelevant to specific transformation (such as rotation) of the system, thereby theoretically ensuring the convergence and consistency of a filter, and being particularly suitable for the problem of attitude estimation. However, how to deeply integrate the geometric superiority of InEKF with the vital adaptive noise estimation capability in practical engineering, and design a practical vehicle navigation algorithm which not only maintains the theoretical rigor, but also has strong environmental adaptability, is still a technical diff