Navigation assistance method for mobile carriers

Abstract

The present invention relates to a navigation assistance method for a mobile vehicle (1), the mobile vehicle (1) comprising an inertial navigation unit (10), the inertial navigation unit (10) comprising at least one inertial sensor (12), wherein, within a determined observation window, an estimation unit (11) of the inertial navigation unit (10) performs the following steps: parameterization (E10) of a non-linear system configured to estimate, according to a kinematic model and / or measurement values acquired by at least one inertial sensor (12), the navigation state of the mobile vehicle (1) within a given time interval at iteration n; linearization (E20) of the system such that the system represents the navigation state at iteration n according to the navigation state at iteration n-1 and a correction of the navigation state, the system being initialized by a first prior state; estimation (E21) of a first correction of the navigation state at iteration n by means of a Kalman filter and stochastic cloning; estimation (E22) of a second correction of the navigation state at iteration n by means of a backward-running information filter and stochastic cloning; determination (E30) of a third correction by combining the first correction and the second correction; and correction (E40) of the navigation state at iteration n according to the third correction, the corrected state being used at iteration n+1.

Description

Technical Field

[0001] The present invention relates to the field of tracking the position of a navigation unit.

[0002] More particularly, the present invention relates to a method for assisting navigation of a mobile carrier, and an inertial navigation unit for implementing said method. Background Art

[0003] In the field of estimating the trajectory of an inertial unit based on measurements from the inertial unit and external measurements from different sensors (Global Positioning System (GPS), cameras, lidar, odometers, etc.), the Kalman filter is a well-known tool for tracking the navigation (i.e., its position, speed, acceleration, etc.) of a carrier (such as a ship, an aircraft or a land vehicle, etc.).

[0004] The Kalman filter estimates the navigation state of the carrier in successive iterations via matrix-based and thus linear equations from the noisy measurements provided by the navigation sensors.

[0005] The unit is then considered as a dynamic system controlled by linear equations, which constitutes a restrictive limitation.

[0006] To extend the Kalman filter to a dynamic system regulated by non-linear equations, a method denoted as "Extended Kalman Filter" (EKF) has been provided. This development provides an additional step which consists of linearizing, at each new iteration of the filter, the equations governing the non-linear system at a point in the vector space, which point is generally the state estimated in the previous iteration. Thus, the matrix resulting from this linearization can be used to calculate the new state estimated using the traditional Kalman filter method.

[0007] However, if the linearization point is too far from the actual navigation state of the carrier, the known Extended Kalman Filter does have the drawback of not working properly.

[0008] However, in certain navigation tracking situations, the navigation state of the unit cannot be accurately estimated at the start of the filter, so successive iterations of implementing the Extended Kalman Filter cannot converge to an accurate state estimate.

[0009] A method of the prior art is also known as "smoothing", which consists of weighting the importance given to each measurement by the accuracy of the sensor producing the measurement, so as to calculate a trajectory of measurements that produces a result as close as possible to the observations.

[0010] Compared with the Kalman filter which processes measurements sequentially and uses each measurement once, smoothing can "look back" to correct the calculation based on the latest available observations. This advantage makes this method important for using certain types of sensors (such as cameras).

[0011] However, with respect to the potential use of smoothing in high-precision navigation units, the main drawbacks of smoothing are recognized. Specifically, when a 64-bit computer is replaced by a traditional computer or a 32-bit or even 16-bit electronic control unit commonly used in integrated systems, the performance obtained deteriorates sharply. Due to the accumulation of inaccuracies in numerical calculations, this deterioration phenomenon is very dangerous because it may lead to the approval of an algorithm in the prototype stage, but in fact it can never be implemented on the final product. The cause of the problem has been identified as the use of an inverse matrix with very poor conditions (a known problem that causes considerable numerical errors).

[0012] Therefore, it is necessary to improve the technology in the prior art. Summary of the Invention

[0013] An object of the present invention is to estimate the trajectory of an inertial unit based on the measurement values of the inertial unit and external measurement values from different sensors.

[0014] Another object of the present invention is to provide a method that is more suitable for execution on a high-precision inertial navigation unit than the solution of the above-mentioned prior art.

[0015] Therefore, a navigation assistance method for a mobile carrier is provided. The mobile carrier includes an inertial navigation unit, and the inertial navigation unit includes at least one inertial sensor. Wherein, within a determined observation window, the following steps are performed by an estimation unit of the inertial navigation unit:

[0016] Parameterization of a non-linear system, the non-linear system being configured to estimate the navigation state of the mobile carrier at iteration n according to a dynamic model and / or measurement values acquired by at least one inertial sensor;

[0017] Linearize the system around the estimated navigation state;

[0018] Estimate a first correction of the navigation state of the estimated carrier through a Kalman filter and stochastic cloning;

[0019] Estimate a second correction through a backward-running information filter and using stochastic cloning;

[0020] Determine a third correction by fusing the first correction and the second correction; and

[0021] Estimate the estimated navigation state at iteration n + 1 according to the third correction.

[0022] Advantageously, the method further includes one or more of the following features.

[0023] The step of estimating the first correction through the Kalman filter and stochastic cloning is implemented over successive time steps in calculating the correction of the navigation state of the estimated vehicle, and one time step of the filter comprises the following steps:

[0024] Propagate the previous navigation state of the vehicle to a propagated state according to the dynamics model and / or measurements obtained by at least one inertial sensor; and

[0025] Update the propagated state according to direct or relative measurements obtained by at least one additional sensor.

[0026] The step of estimating the second correction through the information filter and stochastic cloning of the first correction is implemented over successive time steps, and for one time step of the information filter, the step comprises the following steps:

[0027] Back-propagate the correction of the later navigation state of the vehicle to a correction of the propagated state according to the dynamics model and / or measurements obtained by at least one inertial sensor;

[0028] Update the correction of the propagated state according to direct or relative measurements obtained by at least one additional sensor.

[0029] In the step of estimating the first correction through the Kalman filter and stochastic cloning, the correction of the navigation state propagated by the Kalman filter comprises: a clone of the correction of the navigation state earlier than the correction of the propagated navigation state, provided that the correction of the earlier navigation state involves a relative measurement of a state correction later than the correction of the propagated navigation state. And in the step of estimating the second correction through the information filter and stochastic cloning operating in reverse, the correction of the navigation state back-propagated by the information filter operating in reverse comprises: a clone of the correction of the navigation state later than the correction of the back-propagated navigation state, provided that the correction of the later navigation state involves a relative measurement of a state correction earlier than the correction of the propagated navigation state.

[0030] The navigation state to be estimated is represented by the following formula

[0031]

[0032] where ψ k is a cost function associated with the measurements of each sensor, P k is the covariance matrix associated with the k-th measurement, i.e., the uncertainty associated with the k-th measurement, the symbol

[0033]

[0034] represents the matrix P kEuclidean norm weighted by the inverse matrix.

[0035] Linearize the non-linear system for estimating the navigation state X of the mobile vehicle with respect to the estimated navigation state to define a correction δX in an estimated state of the following form n *

[0036]

[0037] The propagation step implements an augmented transition matrix of the form where F corresponds to the transition matrix relating the previous state correction to the current state correction k, and Id is the identity matrix with dimensions equal to the number of clones of the past state.

[0038] The update step implements an augmented observation matrix of the form The blocks have indices i and j and the rest consists of zeros.

[0039] The proposed method is able to reformulate the calculation of the "maximum a posteriori probability" (MAP) for smoothing so that the inverse matrices of these matrices only appear implicitly, thus avoiding being affected by numerical approximations.

[0040] The proposed method can also extend the Kalman smoother to the problem of combining inertial measurements and measurements of any other type of direct and / or relative state. The method is based on the joint use of a smoother and the so-called "random cloning" method. This makes it possible to perform the fusion of inertial vision and inertial lidar in particular based on a numerically stable maximum a posteriori probability, even when using a high-precision navigation unit and in an integrated architecture with low computing power.

[0041] According to a second aspect, the present invention also provides an estimation unit for a mobile vehicle, which is configured to implement the foregoing navigation assistance method for a mobile vehicle.

[0042] According to a third aspect, the present invention provides an inertial navigation unit for a mobile vehicle, which includes an interface for receiving inertial measurements acquired by at least one inertial sensor, an interface for receiving additional measurements acquired by at least one additional sensor, and an estimation unit according to the second aspect, which estimates the navigation state of the unit based on the measurements acquired by the interface for receiving inertial measurements and the interface for receiving additional measurements.

[0043] There is also provided a computer program product, which includes program code instructions for performing the steps of the foregoing method when the program is executed by an estimation unit of a mobile vehicle trajectory according to the second aspect. Description of the Drawings

[0044] Other features, objects, and advantages of the present invention will become apparent from the following description, which is illustrative only and not restrictive and must be read with reference to the accompanying drawings, in which:

[0045] Figure 1 shows a navigation unit for a mobile carrier according to an embodiment of the present invention;

[0046] Figure 2 and Figure 3 illustrates steps of a method for tracking the position of a navigation unit for a mobile carrier according to an embodiment of the present invention;

[0047] Figure 4A 、 Figure 4B and Figure 4C shows a time-domain graph illustrating the time of arrival of measurement values acquired by the navigation unit and the processing time taken, according to an embodiment of the present invention; and

[0048] Figure 4D and Figure 4E shows a time-domain graph of the time of arrival of measurement values obtained by the navigation unit and the processing time taken, according to different embodiments of the present invention. DETAILED DESCRIPTION

[0049] The term "random cloning" should be understood to mean increasing the system state by replicating past states relevant to future observations.

[0050] The term "correction of the navigation state" should be understood to mean an estimation of the difference between the estimated state and the actual state of the system.

[0051] The term "fusion of two corrections" should be understood to mean calculating a combined correction resulting from two previously obtained corrections, typically by applying a weighted average. The weights are preferably obtained from the information matrix from the information filter and the inverse matrix of the covariance matrix from the Kalman filter.

[0052] The term "backward-running information filter" should be understood to mean recursively calculating the vector and information matrix of the Gaussian law representing information from future states from the last time step to the first time step of a window.

[0053] The term "navigation state" should be understood to mean a set of variables representing at least the direction and position or direction and velocity of a carrier at a given time or during a series of times.

[0054] In the remainder of this document, the covariance of measurement values is written as P with a subscript, and the covariance returned by the Kalman filter is written as P with a superscript.

[0055] Reference Figure 1, the inertial unit 10 is integrated on a mobile carrier 1, such as a land vehicle, a helicopter, an airplane, etc. The inertial unit 10 includes the following parts: an inertial sensor 12, an additional sensor 13, and an estimation unit 11 for performing estimation calculations. These parts can be physically separated from each other.

[0056] The inertial sensor 12 is typically an accelerometer and / or a gyroscope that separately measures the specific forces and rotational speeds experienced by the carrier relative to an inertial reference frame. This specific force is equivalent to the original non-gravitational acceleration. When these sensors are fixed relative to the carrier, the unit is referred to as "strapdown".

[0057] The additional sensor 13 is variable according to the type of the carrier, its dynamic range, and the intended application. The inertial unit typically uses a Global Navigation Satellite System (GNSS) receiver (such as GPS). For a land vehicle, the additional sensor 13 can also be one or more odometers. For a ship, the additional sensor 13 can be a "log (loch)" that gives the speed of the ship relative to the water or the seabed. For example, a camera or a radar of the Light Detection and Ranging (LiDar) type is another example of the additional sensor 13.

[0058] The output data of the estimation unit 11 is a state representing the navigation of the carrier, which will be referred to as the navigation state x in the remainder of this text, and as the internal state of the inertial unit 10 when applicable.

[0059] Depending on the type of sensors that generate the measurement values, three different types of measurement values can be found. Therefore, the following can be found:

[0060] Inertial measurement values related to two consecutive states x i and x i+1 ;

[0061] Direct measurement values that involve only one of x i at a time, such as GPS measurement values; and

[0062] Correlated measurement values, that is, the correlated measurement values involve at least two states x i and x j that may not be continuous. For example, the measurement values obtained using a camera or LiDar are such cases.

[0063] The navigation state can include at least one navigation variable of the carrier (position, speed, acceleration, direction, etc.). The navigation state can be represented in the form of a vector in any case, and each of its components is a navigation variable of the carrier.

[0064] The estimation unit 11 specifically includes an algorithm that is configured to fuse the information given by the additional sensor 13 and the inertial sensor 12 in order to provide an optimal estimate of the navigation information. This fusion is accomplished as follows: a continuous or discrete dynamic system is used as a model to predict the state at each moment based on the state at the previous moment by using a non-linear propagation function f, and the way of observing using the observation function h can also be non-linear. Such a system is non-linear.

[0065] The estimate is then written using a circumflex The actual quantity is written without a circumflex (x).

[0066] The estimation unit 11 includes a main interface 21 for receiving the measurements obtained by the inertial sensor 12, an auxiliary interface 22 for receiving the measurements obtained by the additional sensor 13, and at least one processor 20 that is configured to implement the method described below.

[0067] Smoothing is an algorithm that can be encoded in the form of a computer program executable by the processor 20.

[0068] The estimation unit 11 further includes an output 23 for transmitting the output data calculated by the processor 20.

[0069] Reference Figure 2 and Figure 3 , illustrate certain steps of a method for navigation assistance of a mobile vehicle including an inertial navigation unit, implemented by the estimation unit 11 according to an embodiment of the present invention.

[0070] As part of this method, estimate the trajectory followed by the vehicle, the navigation state of which is the last state of the trajectory. Based on the trajectory estimated in any way Use a method to determine the correction to be applied to the trajectory to obtain a corrected trajectory, which can take into account the information items given by the additional sensor 13 and the inertial sensor 12 at any moment during navigation to provide an optimal estimate of the navigation information.

[0071] Due to the trajectory is estimated by a non-linear problem, so in the so-called parameterization step E10, its implementation is that a non-linear system is configured to describe the change of the navigation state of the mobile vehicle 1. Therefore, the linearized system forms a linear least squares problem that needs to be solved to determine the correction δX of the estimated trajectory n * of a linear least squares problem.

[0072] The trajectory associated with MAP is defined by a non-linear optimization problem of the following form:

[0073]

[0074] Wherein:

[0075] ψ k represents a cost function associated with the measured value of each sensor,

[0076] P k represents the covariance matrix associated with the k-th measured value, i.e., the uncertainty associated with the k-th measured value,

[0077] The symbol represents the Euclidean norm weighted by the inverse matrix of matrix P k of.

[0078] Then the estimated trajectory can be corrected by iteration of this method This method relies on iteratively solving this non-linear optimization problem by successive linearization of the non-linear optimization problem. Therefore, for a series of solutions of the form are searched while seeking to minimize the linear system approximating the MAP optimization problem, which results in solving a series of linear least squares problems of the following form:

[0079]

[0080] where the matrix A k and the vector b k depend on the measured values, and the selected parameterization. A method considering a trajectory δX n consisting of several consecutive states or a part thereof actually consists of several blocks, each block representing one of these states:

[0081] The problem encountered stems from the fact that the standard method for solving these linear problems requires explicit calculation of P k -1 which leads to serious numerical problems in the case of high-precision inertial units, especially in an integrated architecture operating with a reduced computing device (with a 32-bit or even 16-bit card).

[0082] An alternative that can also be applied in the proposed method is to not consider P k but its square root, i.e., the matrix S k such that P k = S k S k T which also avoids inversion.

[0083] Therefore, the non-linear system is linearized with respect to the navigation state (step E20). The navigation state at iteration n is given by the state The sequence composition. The system is considered to be initialized to the first iteration by the first prior state. This state can be any state.

[0084] Once the problem has been linearized (step E20), the solution steps E21, E22, E30 of the system are performed. A solution algorithm is provided, which can find the exact solution of the linear least squares problem, but can avoid the numerical stability problems that may occur in units with too high precision.

[0085] For this purpose, the fact is utilized that the least squares problem can be solved without numerical instability by a method called the Kalman smoother, which is accomplished using the so-called "stochastic cloning" method, so that the number of states of the system can be changed and written in a form compatible with the Kalman smoother.

[0086] This smoother applicable to linear systems is built on the output of the Kalman filter, that is, the smoother is used to correct the estimate generated by the Kalman filter by including the information contained in the observations from the future. Therefore, in order to apply this smoother, the data must first be scanned in a "direct" sense by applying the Kalman filter, that is, from time i = 1 to time i = T (where T is the duration of the observation window). Then the information filter running in reverse (i.e., from time k = T to time k = 1) can be applied. Therefore, after the smoother, the estimate of the state at time i takes into account not only the past observations Y1,..., Yi, but also all the observations Y1,..., YT contained in the observation window.

[0087] In a known manner, the Kalman filter KF is a recursive estimator described by a linear system. In this article, these linearized states are for the correction of the navigation states to be applied to the trajectory.

[0088] The filter is initialized with an initial state, which will be used as the input for the first time step of the filter. Each subsequent time step of the filter will use the state estimated by the previous time step of the filter as the input and provide a new estimate (or correction) of the linearized state of the vehicle.

[0089] The time steps of the Kalman filter usually include two steps: propagation and update.

[0090] The propagation step is based on the previous linearized state (or the initial linearized state of the first iteration) and determines the propagated linearized state of the vehicle with the aid of the linearized propagation function.

[0091] The Kalman filter enables the approximation of the mean and variance of the conditional probability distribution of the linearized state to be performed knowing all the past observations up to this moment.

[0092] There exists a very special set of circumstances in which it is not necessary to compute the inverse matrix of the matrix P associated with the measurements of the inertial unit k to obtain δX n * , and the measurements of the inertial unit are divided into the following two categories:

[0093] The inertial measurements from the inertial sensor 12, which are related to δX n i and δX n i+1 and whose covariance is denoted as Q, and

[0094] the direct measurements from the additional sensor 13, which involve only one of x i at a time, such as GPS measurements.

[0095] Specifically, in this case, a Kalman smoother can be used to solve the problem, and one of its most recent formulas can avoid having to invert the problematic matrix.

[0096] This uses a second estimation by the Kalman filter, and this time the Kalman filter applied in the reverse direction of the observations in a known form is also called an "information filter", which provides a second correction. Then it is fused with the result of the first Kalman filter.

[0097] The Kalman smoother can obtain the desired mean x 1 、…、x p ,

[0098] However, the Kalman smoother is not applicable in this way to other cases, especially when the measurements other than the inertial measurements are correlated measurements, i.e., they are related to at least two states δX n i and For example, this is the case for measurements obtained using a camera or LiDar.

[0099] For all additional sensors 13, the covariance of the associated measurements is denoted as R.

[0100] To be able to extend the Kalman smoother to the problem of combining inertial measurements and measurements of any type of direct and / or correlated states, the method is based on the joint use of the smoother and the so-called "stochastic cloning" method, which is detailed below. This enables the implementation of especially inertial vision and inertial LiDar fusion based on numerically stable maximum a posteriori probability, even when using a high-precision navigation unit and in an integrated architecture with low computational power.

[0101] Reference Figure 4A, which shows an example of a trajectory to be estimated by smoothing over a determined time period. The straight arrows between two consecutive states represent inertial measurement values, and the curved arrows represent correlated measurement values between two states that may be discontinuous.

[0102] Figure 4B Shows the use of stochastic cloning. Thus, the state evolves by cloning the past states involved in retrieving later measurements (e.g., δX0 which will have a direct impact on δX3). The use of stochastic cloning is represented by the alternative states V 1 ,..., V p where the size of the alternative states varies over time because as long as they are involved in measurements of later states, they will contain clones of past states in memory. Thus, the alternative states V i are composed of δX i and the clones of earlier states δX j1 ,..., δX jm .

[0103] Figure 4C Shows the advantages of the combined use of a smoother and the so-called "stochastic cloning" method with high linearization error. Thus, the propagation of information is relatively stable, and thus the correction of the initial error brought by smoothing is also duly considered to obtain a high-quality estimate.

[0104] By comparison, Figure 4D shows the limitations of the Kalman filter applying stochastic cloning. Thus, the high linearization error that may exist in inertial-visual fusion is propagated and never corrected afterwards, which may lead to low-quality estimates, or even illogical estimates.

[0105] Figure 4E Shows the theoretical advantages and practical limitations of traditional smoothing relative to Figure 3 and the method shown in the example of Figure 4. Theoretically, since later measurements must correct the initial linearization error. However, in practice, the numerical errors in traditional implementations are inherited and seriously degrade the estimate.

[0106] The combined use of a smoother and the so-called "stochastic cloning" method can be implemented by the following algorithm.

[0107] In particular, it is sought to calculate a first correction (step E21) and a second correction (step E22) of the navigation state in the considered iteration at each iteration.

[0108] In step E21, in order to apply this smoother, the data is scanned in the "direct" direction (i.e., from time i = 0 to time i = T (where T is the duration of the observation window)) by applying a Kalman filter to determine the vector x i and the matrix Pi , representing the information up to time i, i.e.,

[0109]

[0110] In particular, the correction associated with state i will be given by the last block of V i , so its mean is given by the last block of x i , and its covariance is given by the block in the lower right corner of P i :

[0111]

[0112] [1] Initialization;

[0113]

[0114] [2] For each iteration i;

[0115] [3] If δX n i-1 is involved in a later measurement, then clone δX n i-1 into the mean x of the alternative state i-1 . Extend the covariance P by copying the last row and then the last column i-1 ;

[0116] [4] Propagation of the extended state:

[0117] where

[0118] the dimension of the identity matrix is equal to the number of clones stored in the alternative state. The covariance is propagated in the traditional way.

[0119] [5] If there is a measurement between i and j, where j < i;

[0120] [6] Update according to the known Kalman gain equation as follows:

[0121]

[0122] The block has indices i and j.

[0123] [7] If is not involved in a later measurement, then delete in the alternative state Delete the associated blocks, rows, and columns in x i and P i .

[0124] In step E22, apply the smoother to recursively calculate the posterior distribution:

[0125] P(V i |Y i+1 ,…,Y T ),

[0126] in the form of representing the information starting from time i, that is, encoding a normal distribution via the vector y i and the information matrix J i such that the matrix J

[0127]

[0128] corresponds to the inverse matrix of the covariance matrix associated with this distribution. i

[0129] [1] Initialize y 0 = 0, j 0 = 0;

[0130] [2] For each iteration i, where i ranges from n to 0;

[0131] [3] If δX n i involves previous measurement values, extend y with as many zeros as the dimension of δX n i , and add as many rows of zeros and then as many columns of zeros to j i ; i

[0132] [4] If there is a measurement value Y n i between δX and ij , where j < i;

[0133] [5] Update in the form of the following information:

[0134]

[0135] The block has indices i and j:

[0136]

[0137] and

[0138] [6] If does not involve earlier measurement values, then extend the fusion in the information term :

[0139] and

[0140] where represents y​​i associated with cloning δX n k the corresponding part corresponding to the one associated with δX n k and δX n l associated information blocks, and then delete the blocks associated with the associated blocks

[0141] [7] Backpropagation of extended information items:

[0142] where

[0143] the dimension of the identity matrix is equal to the number of clones stored in the alternative state, and

[0144]

[0145] In step E30, the "direct" and "reverse" estimates are then fused. So the problem lies in fusing the first correction and the second correction obtained in step E21 and step E22 respectively.

[0146] Therefore, for each iteration i, the following calculation of the final correction can now be carried out:

[0147]

[0148] Then, for each i, the correction of the navigation state δX n i* is obtained as the last block of the extended state

[0149] Next, once the correction is determined, the correction of the navigation state is then performed in the following form (step E40):

[0150]

[0151] Therefore, the proposed method can extend the Kalman smoother to the problem of combining inertial measurements and any type of direct and / or related state measurements. The method is based on the joint use of the smoother and the so-called "stochastic cloning" method detailed below. This enables the implementation of the fusion of inertial vision and inertial LiDar in particular based on numerically stable maximum a posteriori probability, even when using a high-precision navigation unit and in an integrated architecture with reduced computing power.​

Claims

1. A navigation assistance method for a mobile vehicle (1), the mobile vehicle (1) comprising an inertial navigation unit (10), the inertial navigation unit (10) comprising at least one inertial sensor (12), wherein, Within a determined observation window, the estimation unit (11) of the inertial navigation unit (10) performs the following steps: Step E10: Parametrization of the non-linear system configured to estimate the navigation state of the mobile carrier (1) over a given time interval at iteration n based on a dynamics model and / or measurements obtained by the at least one inertial sensor (12); Step E20: Linearization of the system such that the system represents the navigation state at iteration n based on the navigation state at iteration n - 1 and a correction to the navigation state, the system being initialized with a first prior state; Step E21: Estimation of a first correction to the navigation state at iteration n by means of a Kalman filter and stochastic cloning; Step E22: Estimation of a second correction to the navigation state at iteration n by means of a backward-running information filter and stochastic cloning; Step E30: Determination of a third correction by fusing the first correction and the second correction; and Step E40: Correction of the navigation state at iteration n based on the third correction, the navigation state corrected based on the third correction being used at iteration n + 1.

2. The navigation assistance method for a mobile carrier (1) according to claim 1, wherein the step E21 of estimating the first correction by means of the Kalman filter and stochastic cloning is performed over consecutive time steps, one time step comprising the following steps: propagating the previous navigation state of the carrier as a propagated state based on a dynamics model and / or measurements obtained by the at least one inertial sensor (12); and updating the propagated state based on direct or correlated measurements obtained by at least one additional sensor (13), the step E22 of estimating the second correction by means of an iterative information filter and stochastic cloning is performed over consecutive time steps, and one time step comprises the following steps: backpropagating a correction to the later navigation state of the carrier as a backpropagated state correction based on a dynamics model and / or measurements obtained by the at least one inertial sensor (12); and updating the correction of the backpropagated state based on direct or correlated measurements obtained by the at least one additional sensor (13).

3. The navigation assistance method for a mobile carrier (1) according to claim 2, wherein in step E21 of estimating the first correction by means of the Kalman filter and stochastic cloning, the correction of the navigation state propagated by the Kalman filter includes cloning of a correction of the navigation state earlier than the correction of the propagated navigation state, provided that the correction of the earlier navigation state relates to correlated measurements of a state correction later than the correction of the propagated navigation state; and wherein In the step E22 of estimating the second correction by means of the information filter running in reverse and random cloning, the correction of the navigation state propagated by the information filter running in reverse includes cloning of the correction of the navigation state that is later than the correction of the navigation state propagated, provided that the correction of the later navigation state involves measurement values of the correction of the state that is earlier than the correction of the propagated navigation state.

4. The navigation assistance method for a mobile vehicle (1) according to claim 3, wherein, The non-linear system configured to estimate the navigation state is represented as follows: where ψ k is a cost function associated with the measurement value of each sensor, and P k is the covariance matrix associated with the k-th measurement value, that is, the uncertainty associated with the k-th measurement value Notation Denotes the Euclidean norm weighted by the inverse matrix of matrix P k .

5. The navigation assistance method for a mobile vehicle (1) according to claim 4, wherein, Linearize the non-linear system for estimating the navigation state X* of the mobile vehicle (1) to define a correction δX in the estimated state in the following form n * 6. The navigation assistance method for a mobile vehicle (1) according to any one of claims 2 to 5, wherein, The step of propagating the previous navigation state of the vehicle as a propagated state is implemented with an augmented transition matrix of the following form: where F corresponds to the transition matrix associating the previous state correction with the current state correction k, and Id is the identity matrix with a dimension equal to the number of clones of the past state.

7. A navigation assistance method for a mobile vehicle (1) according to any one of claims 2 to 5, wherein, Each step of updating the propagation state and updating the correction of the backpropagation state implements an augmented observation matrix of the form where the blocks have indices i and j and the rest consists of zeros.

8. An estimation unit (11) of a mobile vehicle (1), the estimation unit (11) being configured to implement the method according to any one of claims 1 to 7.

9. An inertial navigation unit (10) of a mobile vehicle (1), comprising: an interface for receiving inertial measurement values (21) acquired by at least one inertial sensor (12); an interface for receiving additional measurement values (22) acquired by at least one additional sensor (13); The estimation unit (11) according to claim 8, the estimation unit (11) estimating the navigation state of the unit based on the measurement values acquired by the interface for receiving inertial measurement values (21) and the interface for receiving additional measurement values (22).

10. A computer program product, the computer program product comprising program code instructions which, when the program is executed by an estimation unit (11) of the trajectory of a mobile vehicle (1), execute the steps of the method according to any one of claims 1 to 7.

Images