CN104729500A - Global positioning method of laser-navigated AGV (automatic guided vehicle) - Google Patents

Abstract

The invention discloses a global positioning method of a laser-navigated AGV (automatic guided vehicle). The global positioning of the AGV comprises the determination of direction and position of the AGV in an environmental map. The method comprises the following steps: firstly, determining direction information by reading data of a compass arranged on an AGV body; and secondly, determining position information by using road sign information detected by a laser radar and by adopting a Markov method, wherein the Markov method comprises two stages, namely time updating and observation updating. According to the global positioning method, the problem of accurate positioning of the initial position of the AGV is effectively solved, so that the AGV is not required to start at a known position every time, furthermore, the great inconvenience brought to the application of the AGV is eliminated, namely even if the AGV starts at any position, a relatively accurate initial position of the AGV can be given.

Description

The global localization method of a kind of laser navigation AGV

Technical field

The invention belongs to AGV technical field, particularly relate to the global localization method of a kind of laser navigation AGV.

Background technology

According to research in the past, needing when utilizing characteristics map to position AGV to carry out data correlation, setting up the corresponding relation in observation and map between road sign on this basis.In existing data correlation method, nearest-neighbor method is the more typical independent compatible method of one.Nearest-neighbor method calculates simply, speed is fast, but it have ignored the correlativity in map between road sign, can not ensure to obtain consistent hypothesis, the more important thing is, it is a kind of local association method, needs the initial value that assigned error is less, the hypothesis that the initial value that error is larger very easily leads to errors.

Consider that nearest-neighbor method have ignored the correlativity in map between road sign, have researcher to adopt stricter associating consistent test, propose the compatible branch-bound algorithm of associating.Combine compatible branch-bound algorithm and make use of correlativity in map between road sign, depend on the relative error between road sign instead of absolute error.The reliability of combining compatible branch-bound algorithm is better than nearest-neighbor method.But because the essence of data correlation is that unique match in generation one observation and map between road sign is supposed, this requires that all sub-map that the road sign meeting some in map is formed has uniqueness.But this uniqueness might not be met, at this moment also need the initial pose of a given AGV appropriate accuracy.

In the AGV location of feature based map, EKF method is usually adopted to realize AGV location.EKF method also needs a given relatively accurate initial pose, and initial pose out of true can cause EKF to locate also out of true and even cause location tasks failure.Uncertain due to AGV enable position in the environment, except the sensor that AGV is self-contained, do not have other any positioning system that may be used for determining the initial pose of AGV, unless AGV is each in the place startup that a position is known, this brings great inconvenience to the application of AGV.

Summary of the invention

Be directed to the defect of above-mentioned prior art, the object of the present invention is to provide the global localization method of a kind of laser navigation AGV, with the Global localization problem under the initial pose unknown situation of AGV in the characteristics map solving prior art.

For achieving the above object, the global localization method of a kind of laser navigation AGV of the present invention, the Global localization of described AGV comprises the determination of AGV direction and position in environmental map; Wherein, first, directional information is determined by reading the compass data be arranged on AGV body; Secondly, utilize the mark information that Airborne Lidar measures, and adopt Markov method to determine positional information.

Preferably, when adopting Markov method to determine positional information, first by the continuous position spatial discretization of AGV, be specially: definition AGV coordinate is in the environment (x, y), define AGV and in map, need to carry out the rectangular area of Global localization for [X _min, X _max; Y _min, Y _max], wherein, X _minand X _maxfor minimum value and the maximal value in x direction, Y _minand Y _maxfor minimum value and the maximal value in y direction;

By rectangular area [X _min, X _max; Y _min, Y _max] turn to rectangular element little one by one so that resolution [ResX, ResY] is discrete, wherein, ResX and ResY is respectively the resolution of x and y discrete--direction, usually sets ResX=ResY, and a rectangular element is referred to as a grid, uses represent, wherein, k is the time, and its initial value is zero, i is grid index, is grid give a reliability, be expressed as the grid of whole rectangular area forms a width reliability image.

Preferably, above-mentioned Markov method comprises time renewal and observation two stages of renewal.

Preferably, it is provide more new calculation method of a kind of time based on frequency domain process that the time of above-mentioned Markov method upgrades, and detailed process is:

A. by Fourier transform by reliability image from spatial transform to frequency domain, transformation for mula is:

$F(u,v)=\underset{x=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{N-1}{\mathrm{\Σ}}}f(x,y)e^{-i2\mathrm{\π}(\frac{\mathrm{ux}}{M}+\frac{\mathrm{vy}}{N})}$

In formula, f and F is respectively spatial domain picture and frequency domain figure picture, order m and N is respectively the grid number in x and y direction, u and v is respectively the position quantity in x and y direction; E is natural exponential function;

B. carry out multiply operation to frequency domain figure picture, specific formula for calculation is:

$F(u,v)e^{-i2\mathrm{\π}({\mathrm{ux}}_0/M+{\mathrm{vy}}_0/N)}$

X ₀and y ₀be respectively reliability image needs x and the y direction in spatial domain translational movement according to AGV amount of exercise;

C. carry out obfuscation operation to reliability image in a frequency domain, concrete formula is:

F(u,v)·H(u,v)

Wherein, F utilizes the periodicity of Fourier transform low frequency part to be moved to the frequency domain figure picture after picture centre; wherein, D ₀for the filtering parameter artificially arranged;

D. by inverse Fourier transform, reliability image is switched back to spatial domain from frequency domain, complete more new stage time of a Markov method, concrete formula is:

$f(x,y)=\underset{u=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{v=0}{\overset{N-1}{\mathrm{\Σ}}}F(u,v)e^{i2\mathrm{\π}(\frac{\mathrm{ux}}{M}+\frac{\mathrm{vy}}{N})}$

The reliability image that definition time upgrades order $\stackrel{\‾}{\mathrm{Bel}}\left({\hat{X}}_{k,i}\right)=f(x,y).$

Preferably, it is provide a kind of observation level and smooth based on gaussian kernel more new calculation method that the observation of above-mentioned Markov method upgrades, and detailed process is:

A. road markings function of position carries out serialization, and definition AGV at a time observes n road sign, and the position of each road sign is expressed as ((ρ under laser radar coordinate system _i, θ _i), i ∈ (1, n), define the road sign position function in this moment carry out continuous operation to this function, concrete formula is:

$\mathrm{\ρ}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\exp \{-{(\mathrm{\θ}-{\mathrm{\θ}}_i)}^2/{\left({2\mathrm{\σ}}_i\right)}^2\}$

In formula, exp is gaussian kernel function; σ _ithe bandwidth of core, herein, σ _ibe taken as λ/ρ _i; λ be greater than zero constant, undertaken regulating and determining by artificial experience;

B. angularly spaced discrete is carried out to the road sign position function ρ after serialization, will equiangularly Δ θ is discrete turns to τ value, and obtaining the road sign position function after discretize is:

$\mathrm{\ρ}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\exp \{-{({\mathrm{\θ}}_{\min }+\mathrm{r\Δ\θ}-{\mathrm{\θ}}_i)}^2/{\left(2{\mathrm{\σ}}_i\right)}^2\}$

In formula, Δ θ is r ∈ (0, τ-1);

C. carry out observation likelihood to calculate, concrete formula is:

$p\left(z_k\right|{\hat{X}}_{k,i},M)=\underset{i=1}{\overset{\mathrm{\τ}}{\mathrm{\Π}}}\left(\frac{{\mathrm{\ρ}}_{\max }-|{\mathrm{\ρ}}_z\left({\mathrm{\θ}}_i\right)-{\mathrm{\ρ}}_m\left({\mathrm{\θ}}_i\right)|}{{\mathrm{\ρ}}_{\max }}\right)$

In formula, ρ _maxfor the distance of the road sign that AGV can detect farthest; ρ _zfor AGV actual observation level and smooth after observed reading; ρ _mfor AGV is in map observed reading after the observation smoothly at place;

D. upgrade reliability image according to observation likelihood value, concrete formula is:

$\mathrm{Bel}\left({\hat{X}}_{k,i}\right)=\mathrm{\ηp}\left(z_k\right|{\hat{X}}_{k,i},M)\stackrel{\‾}{\mathrm{Bel}}\left({\hat{X}}_{k,i}\right)$

In formula, η=1/p (z _k| Z _0:k-1, U _0:k), be normaliztion constant.

Beneficial effect of the present invention:

Present invention efficiently solves the accurate orientation problem of the initial pose of AGV, AGV is made not need each all to start in a known place, position, eliminate and bring this great inconvenience to the application of AGV, even if namely AGV optional position starts, also can the relatively accurate initial pose of given AGV mono-.

Accompanying drawing explanation

Fig. 1 illustrates the global localization method schematic diagram of a kind of las er-guidance AGV of the present invention.

Fig. 2 illustrates laser navigation AGV system hardware structure figure in embodiments of the invention.

Fig. 3 illustrates localization method AGV locational space discretize schematic diagram of the present invention.

Fig. 4 illustrates the method for updating time process flow diagram based on frequency domain process of the present invention.

Fig. 5 illustrates the level and smooth schematic diagram of gaussian kernel of the present invention.

Fig. 6 illustrates parameter lambda of the present invention affects schematic diagram to sharpening result.

Embodiment

For the ease of the understanding of those skilled in the art, below in conjunction with embodiment and accompanying drawing, the global localization method that AGV guided by a kind of laser radar of the present invention is further described, make object of the present invention, technical scheme and effect clearly, clearly, the content that embodiment is mentioned not is limitation of the invention.

Should be understood that; specific embodiment described in the present invention is only to the explanation for example of the present invention's spirit; for those of ordinary skills; can make various amendment or supplement or adopt similar mode to substitute, all these improve and convert the protection domain that all should belong to claims of the present invention.

With reference to shown in Fig. 2, the hardware configuration of the laser navigation AGV system that the global localization method that AGV guided by a kind of laser radar of the present invention adopts, mainly comprises AGV body, control system, drive system and sensor-based system.Wherein, AGV body dimension is long 100cm, wide 60cm, high 40cm; Be configured with driving wheel and four universal wheels that two differentials drive, driving wheel is positioned at the both sides of car body geometric center, four universal wheels lay respectively at car body just before, just rear, left and right position; The upper control computer of control system by Based PC and the embedded the next computer for controlling based on ARM form; Drive system is made up of two groups of motor drivers, DC servo motor and speed reduction units; Sensor-based system is made up of two scramblers, a laser radar and an electronic compass.

In the present embodiment, arranging Laser Radar Scanning resolution is 0.5 °, and range observation resolution is 1mm, and the BURN-THROUGH RANGE arranging lower laser radar in this measuring distance resolution is 8m.

In the present embodiment, based on ARM embedded the next computer for controlling primary responsibility capturing and coding device data and carry out reckoning, gather electronic compass data, and flying track conjecture result and electronic compass data are sent to upper control computer by serial ports, receive the speed control commands that sends over of upper control computer simultaneously, and by the motion of PWM control AGV.

Move in the rectangular coordinate system of place plane at AGV, the pose of AGV is expressed as the Posterior distrbutionp of the k moment AGV pose being condition with the observation of AGV history, motion control amount and map, i.e. p (X _k| Z _0:k, U _0:k, M) and position problem estimation.

In formula, each variable implication is as follows:

X _k---the AGV pose in k moment;

Z _0:k---from all observations of the AGV road markings in 0 to k moment, Z _0:k={ z ₀, z ₁..., z _k, wherein z _kfor the observation in k moment;

U _0:k---from all motion control inputs of the AGV in 0 to k moment, U _0:k={ u ₀, u ₁..., u _k, wherein u _kfor the control inputs in k moment;

M---environmental map, herein wherein n is road sign (unique point) quantity, m _i=(x _i, y _i) ^t, x _iand y _ifor the coordinate of road sign under two-dimensional world coordinate system.

The probability distribution of the Motion Controlling Model state-transition matrix of AGV represents: p (X _k| X _k-1, u _k).

AGV is when given pose X and map M, and the Posterior distrbutionp of observation model is expressed as: p (z _k| X _k, M).

The Global localization of described AGV comprises determines the position of AGV in environmental map and direction, and wherein, direction is determined by reading the compass data be arranged on AGV body; The mark information that position utilizes Airborne Lidar to measure, the present embodiment adopts Markov method to determine; This Markov method comprises time renewal and observation two stages of renewal, and specific implementation step is as follows:

Under the indoor environment of spaciousness, select to carry out Markov Global localization in environment size 20m × 20m vacant room environment.Coordinate setting is carried out to room integral position, first determines a coordinate origin and X, Y-axis, the route that all AGV pass by is placed on first quartile, guarantee that all position coordinateses are all integers; Asymmetrical layout 20 pieces of baffle plates in indoor, and calculate the correspondence position of each piece of baffle plate, AGV is placed in environment, makes it roam a period of time, before intercepting, the original sensor data of 100s carries out AGV location; Time upgrades the method for updating time based on frequency domain process adopting the present invention to propose, and observation upgrades the observation update method level and smooth based on gaussian kernel adopting the present invention to propose.

With reference to shown in Fig. 1, first, by the continuous position spatial discretization of AGV, AGV needs the rectangular area of carrying out Global localization to be [0,20 in map; 0,20], by rectangular area [0,20; 0,20] turn to rectangular element little one by one so that resolution [0.1,0.1] is discrete, wherein, 0.1 and 0.1 resolution being respectively x and y discrete--direction, sets equivalence usually.A rectangular element is referred to as a grid, uses represent, wherein, k is the time, and i is grid index.For grid give a reliability, be expressed as the grid of whole rectangular area forms a width reliability image, as shown in Figure 3.

AGV exists the reliability of place's prediction is reliability after corresponding correction is time update equation formula then after discretize and observation renewal equation formula can be expressed as:

$\begin{array}{c}\stackrel{\‾}{\mathrm{Bel}}\left({\hat{X}}_{k,i}\right)=\underset{j\∈\mathrm{\Θ}}{\mathrm{\Σ}}p\left({\hat{X}}_{k,i}\right|{\hat{X}}_{k-1,j},u_k)\mathrm{Bel}\left({\hat{X}}_{k-1,j}\right)\\ \mathrm{Bel}\left({\hat{X}}_{k,i}\right)=\mathrm{\ηp}\left(z_k\right|{\hat{X}}_{k,i},M)\stackrel{\‾}{\mathrm{Bel}}\left({\hat{X}}_{k,i}\right)\end{array}$

Utilize the AGV data obtained in the present embodiment to carry out AGV Global localization, the time that have employed Markov method upgrades, i.e. more new calculation method of a kind of time based on frequency domain process, and the detailed process of the method is:

A. by Fourier transform by reliability image from spatial transform to frequency domain, utilize two dimensional discrete Fourier transform formula exactly: the spatial domain picture f (x, y) of reliability is transformed to frequency domain through Fourier transform, obtains the frequency domain figure of reliability as F (u, v); In formula, f and F is respectively spatial domain picture and frequency domain figure picture; The grid number in x and y direction is 100;

B. the non-integral multiple elementary cell translation of reliability image in spatial domain, utilizes the relation between spatial domain picture and frequency domain figure picture: obtain new reliability frequency domain figure picture, the x in formula ₀and y ₀be any real number, realize the non-integral multiple elementary cell translation of reliability image;

C. the obfuscation in a frequency domain of reliability image is operating as: F (u, v) H (u, v); Wherein, F utilizes the periodicity of Fourier transform low frequency part to be moved to the frequency domain figure picture after picture centre; wherein, D ₀for the filtering parameter artificially arranged;

D. by inverse Fourier transform, reliability image is switched back to spatial domain from frequency domain, complete more new stage time of a Markov method, as shown in Figure 4.

The present embodiment proceeds the level and smooth observation of gaussian kernel more new calculation method, and the detailed process of the method is:

A.AGV at a time observes 3 pieces of baffle plates i.e. 3 road signs, and any one road sign can be expressed as function ρ _i=ρ _iδ ((θ-θ _i), by the function summation corresponding to all road signs observed, obtain a Non-smooth surface, discrete road sign position function model: as shown in Figure 5;

B. serialization is carried out to Non-smooth surface, discrete road sign position function, replace the ρ in above formula with gaussian kernel function _iδ (θ-θ _i), obtaining continuous print road sign position function is:

In formula, exp is gaussian kernel function; σ _ithe bandwidth of core, herein, σ _ibe taken as λ/ρ _i; λ be greater than zero constant, λ is on the impact of sharpening result with reference to shown in Fig. 6, and λ is larger, and gained road sign position function is more level and smooth;

C. angularly spaced discrete is carried out to the road sign position function after serialization: by independent variable θ ∈ (θ _min, θ _max) equiangularly Δ θ be divided into τ section, then any one road sign position can be expressed as:

$\mathrm{\ρ}=\underset{i=1}{\overset{3}{\mathrm{\Σ}}}\exp \{-{({\mathrm{\θ}}_{\min }+\mathrm{r\Δ\θ}-{\mathrm{\θ}}_i)}^2/{\left(2{\mathrm{\σ}}_i\right)}^2\},$ In formula, Δ θ is

D. the calculating observing likelihood is carried out: $p\left(z_k\right|{\hat{X}}_{k,i},M)=\underset{i=1}{\overset{\mathrm{\τ}}{\mathrm{\Π}}}\left(\frac{{\mathrm{\ρ}}_{\max }-|{\mathrm{\ρ}}_z\left({\mathrm{\θ}}_i\right)-{\mathrm{\ρ}}_m\left({\mathrm{\θ}}_i\right)|}{{\mathrm{\ρ}}_{\max }}\right),$

Wherein, ρ _maxfor the distance of the road sign that AGV can detect farthest; ρ _zfor AGV actual observation level and smooth after observed reading; ρ _mfor AGV is in map observed reading after the observation smoothly at place; So just complete and once observe renewal, now may obtain the possible position of multiple AGV state reliability.

In order to obtain AGV locating information accurately, we utilize the original laser sensing data collected, repeatedly carry out time renewal and observation upgrades, finally obtain the convergency value of an AGV position reliability accurately, the specific algorithm that the AGV based on Markov locates is as follows:

Initialization k=0;

Each grid of For

$\mathrm{Bel}\left({\hat{X}}_{k,i}\right)=p\left({\hat{X}}_{k,i}\right);$ // initialization probability grid

End

While TRUE

k＝k+1；

Calculated by the time Renewal model based on frequency domain process

w＝0；

Each grid of For

Calculating observation likelihood

$\mathrm{Bel}\left({\hat{X}}_{k,i}\right)=p\left(z_k\right|{\hat{X}}_{k,i},M)\stackrel{\‾}{\mathrm{Bel}}\left({\hat{X}}_{k,i}\right);$

$w=w+\mathrm{Bel}\left({\hat{X}}_{k,i}\right);$

End

Each grid of For

$\mathrm{Bel}\left({\hat{X}}_{k,i}\right)=w^{-1}\·\mathrm{Bel}\left({\hat{X}}_{k,i}\right);$

End

End

Embody rule approach of the present invention is a lot, and the above is only the preferred embodiment of the present invention, should be understood that; for those skilled in the art; under the premise without departing from the principles of the invention, can also make some improvement, these improvement also should be considered as protection scope of the present invention.

Claims (10)

1. a global localization method of laser navigation AGV, is characterized in that, the Global localization of described AGV comprises the determination of AGV direction and position in environmental map; First, directional information is determined by reading the compass data be arranged on AGV body; Secondly, utilize the mark information that Airborne Lidar measures, and adopt Markov method to determine positional information.

2. the global localization method of a kind of laser navigation AGV as claimed in claim 1, it is characterized in that, when adopting Markov method to determine positional information, first by the continuous position spatial discretization of AGV, be specially: definition AGV coordinate is in the environment (x, y), define AGV and in map, need to carry out the rectangular area of Global localization for [X _min, X _max; Y _min, Y _max], wherein, X _minand X _maxfor minimum value and the maximal value in x direction, Y _minand Y _maxfor minimum value and the maximal value in y direction;

By rectangular area [X _min, X _max; Y _min, Y _max] turn to rectangular element little one by one so that resolution [ResX, ResY] is discrete, wherein, ResX and ResY is respectively the resolution of x and y discrete--direction, usually sets ResX=ResY, and a rectangular element is referred to as a grid, uses represent, wherein, k is the time, and its initial value is zero, i is grid index, is grid give a reliability, be expressed as the grid of whole rectangular area forms a width reliability image.

3. the global localization method of a kind of laser navigation AGV as claimed in claim 1, is characterized in that, described Markov method comprises time renewal and observation two stages of renewal.

4. the AGV global localization method of a kind of laser navigation as claimed in claim 3, it is characterized in that, the described time, more the new stage adopted a kind of method for updating time based on frequency domain process, the method by Fourier transform by reliability image from spatial transform to frequency domain, utilize the non-integral multiple grid translation of the relational implementation reliability image between spatial domain picture and frequency domain figure picture and the obfuscation of reliability image is operated, then by inverse Fourier transform, reliability image being switched back to spatial domain from frequency domain.

5. the global localization method of a kind of laser navigation AGV as claimed in claim 4, it is characterized in that, the pass between described spatial domain picture and frequency domain figure picture is:

$f(x-x_0,y-y_0)\⇔F(u,v)e^{-i2\mathrm{\π}({\mathrm{ux}}_0/M+{\mathrm{vy}}_0/N)}$

Wherein, f is spatial domain picture; F is frequency domain figure picture; M and N is respectively the grid number in x and y direction; U and v is respectively the position quantity in x and y direction; x ₀and y ₀be respectively reliability image needs x and the y direction in spatial domain translational movement according to AGV amount of exercise.

6. the global localization method of a kind of laser navigation AGV as claimed in claim 4, is characterized in that, described is operating as the obfuscation of reliability image:

F(u,v)·H(u,v)

Wherein, F utilizes the periodicity of Fourier transform low frequency part to be moved to the frequency domain figure picture after picture centre; wherein, D ₀for the filtering parameter artificially arranged.

7. the global localization method of a kind of laser navigation AGV as claimed in claim 3, it is characterized in that, more the new stage adopts a kind of observation update method level and smooth based on gaussian kernel in described observation, the method carries out serialization by gaussian kernel function to the road sign extracted, angularly spaced discrete is carried out to the virtual environment profile after serialization, finally sets up the likelihood of observation likelihood model calculating observation.

8. the global localization method of a kind of laser navigation AGV as claimed in claim 7, is characterized in that, the described road sign to extracting carries out serialization, and detailed process is:

A. define AGV and at a time observe n road sign, then the position of this n road sign under laser radar polar coordinate system is respectively (ρ ₁, θ ₁), (ρ ₂, θ ₂) ..., (ρ _n, θ _n), θ is considered as independent variable, and ρ is considered as dependent variable, then any one road sign can be expressed as function ρ _i=ρ _iδ ((θ-θ _i), wherein δ function is:

$\mathrm{\δ}\left(x\right)=\left\{\begin{array}{cc}1& x=0\\ 0& x\≠0\end{array}\right.$

By the function summation corresponding to all road signs observed, obtain a Non-smooth surface, discrete road sign position function:

$\mathrm{\ρ}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ρ}}_i\mathrm{\δ}(\mathrm{\θ}-{\mathrm{\θ}}_i)$

B. serialization is carried out to Non-smooth surface, discrete road sign position function, replace the ρ in above formula with gaussian kernel function _iδ (θ-θ _i), obtaining continuous print road sign position function is:

$\mathrm{\ρ}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\exp \{-{(\mathrm{\θ}-{\mathrm{\θ}}_i)}^2/{\left({2\mathrm{\σ}}_i\right)}^2\}$

In formula, exp is gaussian kernel function; σ _ithe bandwidth of core, herein, σ _ibe taken as λ/ρ _i; λ be greater than zero constant, λ is larger, and gained road sign position function is more level and smooth.

9. the global localization method of a kind of laser navigation AGV as claimed in claim 7, is characterized in that, describedly carries out angularly spaced discrete to the road sign position function after serialization, and concrete grammar is:

By independent variable θ ∈ (θ _min, θ _max) equiangularly Δ θ be divided into τ section, wherein, θ _minand θ _maxbe respectively minimum value and the maximal value of θ, the road sign position function of discretize can be obtained thus:

$\mathrm{\ρ}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\exp \{-{({\mathrm{\θ}}_{\min }+\mathrm{r\Δ\θ}-{\mathrm{\θ}}_i)}^2/{\left({2\mathrm{\σ}}_i\right)}^2\}$

In formula, Δ θ is

10. the global localization method of a kind of laser navigation AGV as claimed in claim 7, is characterized in that, the computing method of described observation likelihood are:

$p\left(z_k\right|{\hat{X}}_{k,i},M)=\underset{i=1}{\overset{\mathrm{\τ}}{\mathrm{\Π}}}\left(\frac{{\mathrm{\ρ}}_{\max }-|{\mathrm{\ρ}}_z\left({\mathrm{\θ}}_i\right)-{\mathrm{\ρ}}_m\left({\mathrm{\θ}}_i\right)|}{{\mathrm{\ρ}}_{\max }}\right)$

Wherein, ρ _maxfor the distance of the road sign that AGV can detect farthest; ρ _zfor AGV actual observation level and smooth after observed reading; ρ _mfor AGV exists observed reading after the observation smoothly at place.