JPH11304489A - Method for position inference - Google Patents

Abstract

PROBLEM TO BE SOLVED: To broaden the range of inference by increasing the number of validation grids and to prevent the possibility of contradiction from increasing during checks, no matter how much the range within which inference can be made is broadened, by enabling the sizes of the validation grids to be freely set. SOLUTION: For landmark information 101, pattern data which could be extracted by a landmark extraction means 102 and statistical data showing the probability of extraction are associated with plural sites of a map, and a position estimated by a position inference means 104 is inputted. The landmark information is provided to a reliability value matrix updating means 105, and an impulse response matrix generation means 106 generates a matrix by use of state values measured by a state measurement means 107. The means 102 extracts the pattern data by processing measurement data obtained by an environment measurement means 103, and the means 105 updates an internally held reliability-value matrix by convoluting the matrix generated by the means 106 with the reliability-value matrix to allow for changes in reliability value, and checking the data extracted by the means 102 with the data provided by the information 101. The means 104 estimates the position of a moving object by use of the position measured by the means 107 and the reliability-value matrix provided by the means 105.

Description

DETAILED DESCRIPTION OF THE INVENTION

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for estimating the position of a moving object such as a mobile robot.

[0002]

2. Description of the Related Art In a conventional position estimation method for a moving object, for example, a mobile robot, a validation gate (Validation gate) is used to manage the reliability of the estimated position.
An error region called “Gate” is used, and a Kalman filter is used to update the estimated value and the estimated error based on the state transition and the observation result of the environment. Since the validation gate is defined as a region that provides an appropriate threshold value for the probability density function of the estimated position and indicates a value equal to or larger than the threshold value, the validation gate may be considered as a region indicating a relatively reliable estimated value. I can do it. In addition, in estimating the position of a mobile robot, it is often assumed that the distribution state of the estimation error follows a normal distribution. In this case, the validation gate is redefined as an inner area of an ellipse, and landmark extraction is performed. Assuming that the mobile robot is located in the area inside the validation gate when matching the landmark pattern data extracted by the device with the landmark pattern data obtained by referring to the map information It is also used as a filter that excludes landmark information in the map that is contrary to.

FIG. 7 is an explanatory diagram of a validation gate used in a conventional position estimation method. Line segment 704,
The estimated position of the mobile robot 701 in the environment map represented by the line segments 705, 706, and 707 becomes gradually uncertain due to an error accompanying the position movement, but is statistically determined by a validation gate. It is highly likely that the user is inside the inside of the validation gate 702, and in particular, it is most likely to be at the center position 703 of the validation gate having an elliptical shape. In many cases, the center position 703 of the validation gate is selected as the estimated position of the mobile robot 701.

[0004]

In such a conventional method, when the size of the validation gate is increased in order to enlarge the estimation range, or when the map information is precise, an erroneous land due to the validation gate may be lost. The filtering of the mark information may not work well, and a plurality of inconsistent matching patterns may occur during the matching process. However, since the distribution of estimation errors that can be managed by the validation gate must follow a normal distribution, it has been difficult to properly deal with the hypothesis that a plurality of inconsistent matching patterns occur. Further, in the conventional method, only information that has been successfully collated is used for updating the estimated value or the estimation error, and the fact that the collation has failed has been ignored. However, in order to further increase the robustness of the system, it is desirable to perform position estimation that can utilize the fact that the matching has failed.

[0005]

According to a first aspect of the present invention, the reliability of an estimated position of a moving object is managed using a reliability value matrix composed of a validation grid. The reliability value matrix is updated by comparing the landmark pattern data generated by the landmark extraction means with landmark information at landmark association points obtained in advance, and is used for position estimation. A position estimation method.

As described above, according to the present invention, the reliability of an estimated position of a robot is evaluated by a validation grid.
d) is managed using a reliability value matrix composed of the landmark pattern data extracted by the landmark extracting means and the landmark pattern data obtained in advance.
Association Point (Landmark AssociationPo)
(int, hereinafter also referred to as LAP)) is a position estimation method characterized by updating the confidence value matrix by collating with landmark information and using it for position estimation. The estimation range can be increased to increase the size of the validation grid, and the size of the validation grid can be set arbitrarily. Therefore, even if the estimation range is increased, there is no particular inconsistency in matching.

According to a second aspect of the present invention, there is provided the position estimating method according to the first aspect, wherein a validation grid in which a distribution state of the reliability value of the estimated position is represented by a plurality of grid-like regions is used. is there.

According to a third aspect of the present invention, the landmark information comprises pattern data constituting a landmark and statistical data indicating the probability of occurrence of each pattern data. And a landmark association point for which the landmark information is determined in advance.
It is the position estimation method described.

According to a fourth aspect of the present invention, a matrix obtained by approximating a probability density distribution representing an estimation error due to a position movement from a previous process to a current process by a finite grid area and the confidence value matrix are combined. A lution operation is performed, and the pattern data of the landmark information generated by the landmark extracting means is compared with the pattern data of the landmark information at the landmark association point. 2. The position estimating method according to claim 1, wherein the confidence value matrix is updated using the statistical data.

Furthermore, in the invention of claims 2 to 4 of the present invention, since the reliability value of the estimated position is managed for each validation grid, there is no restriction on the distribution method. This means that even if multiple conflicting hypotheses occur, it is possible to perform flexible processing considering all the hypotheses, and even if the validation is incorrectly performed on one validation grid, correct verification can be performed on the other validation grids. If done, it will not have a significant adverse effect on later estimates. In addition, since the fact that the collation has failed is also evaluated based on the statistical data of the landmark information, a remarkable effect that it is possible to realize optimal estimation using information efficiently can be achieved.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Specific examples of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a system configuration according to the embodiment of the present invention. Landmark information 10
1 includes, in advance, pattern data that may be extracted by the landmark extracting means 102 and statistical data indicating the probability of extracting the pattern data in a plurality of environmental maps for performing position estimation. And the latest estimated position estimated by the position estimating means 104 is input online.
The landmark information required by the reliability value matrix updating means 105 is provided. The impulse response matrix generation means 106 generates an impulse response matrix using the state values (position x, position y, direction θ, velocity v) of the mobile robot measured by the state measurement means 107.

The landmark extracting means 102 processes the measurement data obtained by the environment measuring means 103 to extract landmark pattern data. The reliability value matrix updating means 105 includes an impulse response matrix generating means 1
Convolution of the impulse response matrix generated in 06 and the internally held confidence value matrix (convo-luti
on), after considering the change in the reliability value due to the position movement, each pattern is extracted in accordance with the result of the comparison between the pattern data extracted by the landmark extracting means 102 and the pattern data provided by the landmark information 101. The reliability value matrix is updated in consideration of the statistical data belonging to the data. The position estimating means 104 includes a position (x, y) measured by the state measuring means 107 and the confidence value matrix updating means 1
A more accurate position (x, y) is estimated using the confidence value matrix that is sequentially updated by 05.

FIG. 2 is a flowchart illustrating the operation procedure of the position estimation method according to the present invention. Hereinafter, FIG. 2 will be described for each of the steps (A1 to A10). A1: Initialize the state values (position x, position y, direction θ, velocity v) and values of the reliability value matrix. An appropriate value is set as the initial value of the reliability value matrix based on the degree of uncertainty of the initial position of the mobile robot. If the initial value of the mobile robot can be accurately grasped, the confidence value of the validation grid corresponding to the center position of the confidence value matrix is calculated.
The value may be set to 100%, and if it can be grasped only in an uncertain manner, a normal distribution may be allocated to the validation grid around the center position so that the total value becomes 100%. A2: The shape and the like of the surrounding environment of the mobile robot are measured by environment measuring means 103 such as a laser sensor or an ultrasonic sensor.

A3: State values of the mobile robot (position x, position
y, direction θ, velocity v) are measured by the state measuring means 107 such as an encoder or a gyro. A4: Using the state value of the mobile robot measured in step A3, the estimated position of the mobile robot is updated by the following equation.

[0015]

(Equation 1)

A5: The landmark pattern data is extracted by the landmark extracting means 102. A6: The impulse response matrix generation means 106 generates an impulse response matrix.

[0017]

(Equation 2)

A7: Step the confidence value of the confidence value matrix
The transition is made by taking a convolution with the impulse response matrix obtained in A6. A8: The pattern data extracted by the landmark extracting means 102 is compared with the pattern data provided by the landmark information 101, and each pattern is extracted according to the result.
The reliability value matrix is updated in consideration of the statistical data belonging to the data. A9: If the result of estimating the position of the mobile robot by analyzing the distribution of the reliability values in the reliability value matrix (for example, detecting the maximum value) and deviating from the position obtained in step A4, The value of the value matrix is corrected according to the correction amount. A10: If there is an end request, the process ends, and if not, the process returns to step A2.

FIG. 3 is an explanatory diagram showing an example of a validation grid when a plurality of conflicting hypotheses of the present invention occur. Regions 301 and 3 with high confidence values as estimated positions
02 indicates a state where it has occurred and cannot be expressed by the conventional method.

FIG. 4 is a block diagram showing a circuit configuration of an embodiment in which the present invention is actually applied to a mobile robot. The landmark information 401 includes a linear component extraction device 402
An environment in which position estimation is performed by using a parameter value (hereinafter, referred to as a straight line parameter value) representing a linear component that may be extracted as pattern data, and further using the probability of extracting the linear parameter value as statistical data. Are associated with a plurality of locations (coordinate values) of the two-dimensional planar map of the mobile robot, and the latest position (x, y) of the mobile robot estimated by the position estimation device 404 is input to the confidence value matrix updating device 405. Provide the required landmark information.

Impulse response matrix generator 406
Generates an impulse response matrix using the position (x, y), direction θ, and velocity V of the mobile robot measured by the state measurement device 407. The linear component extraction device 402 is an image / environment image of the environment measured by the laser scanner 403.
The data is subjected to Hough transform processing to extract straight line parameter values in the image data. The confidence value matrix updating device 405 calculates the change in the confidence value of the confidence value matrix accompanying the position movement of the robot by the convolution of the impulse response matrix generated by the impulse response matrix generation device 406 and the internally held confidence value matrix. After that, according to the result of matching between the straight line parameter value extracted by the straight line component extraction device 402 and the straight line parameter value provided by the landmark information 401, a confidence value matrix internally held in consideration of statistical data belonging to each straight line parameter value To update. The position estimating device 404 uses the position (x, y) of the mobile robot measured by the state measuring device 407 and the confidence value matrix sequentially updated by the confidence value matrix updating device 405 to obtain a more accurate position of the mobile robot.
Estimate (x, y). Next, a method of updating the confidence value matrix, which is the process of step A8, will be described.

FIG. 5 is a flowchart of a method of updating a confidence value matrix processed by the confidence value matrix updating means of the present invention. Hereinafter, the reliability value matrix updating means of FIG.
Steps 105 and 405 will be described step by step. B1: Set the number of validation grids in Counter I. B2: Find the coordinates (x, y) of the No. I validation grid in the environment map. Here, the coordinates of the validation grid located at the center of the estimable area formed in a two-dimensional rectangular shape by the validation grid are set to be the same as the estimated position of the mobile robot at the present time. B3: 1st success rate on the 1st validation grid
(100%), and initialize the number of times of collation to 0. The value of the success rate is reduced based on the statistical data every time the verification of the landmark information fails. In contrast, the number of matches is
It is incremented by one each time landmark information is successfully collated, and is used to redistribute confidence values in step B21.

B4: The LAP located closest to the I-th validation grid is searched from the landmark information. B5: The counter J is set with the number of pieces of landmark information associated with the LAP searched in step B4. B6: The flag indicating the fact that the J-th landmark information has been collated with the straight-line parameter value extracted by the straight-line component extraction device is initialized to "failure".

[0024]

(Equation 3)

B8: The counter K has a linear component extracting device 502
Set the number of straight lines extracted in. B9: The straight-line parameter value obtained by correcting the J-th straight-line parameter value in step B7 and the K-th straight-line parameter value obtained by the straight-line component extraction device 402 are compared. The determination as to whether collation is possible is made based on whether or not the absolute value of the difference between the two parameter values is within a predetermined threshold.

B10: If the collation in step B9 is successful, the process proceeds to step B11; otherwise, the process proceeds to step B12. B11: A flag indicating the fact that the straight-line parameter value obtained by correcting the J-th straight-line parameter value in step B7 is compared with the straight-line parameter value extracted by the straight-line component extraction device 502 is set to “success”, Increment the number of checks in the 1st validation grid by 1.

B12: Decrease the counter K indicating the number of the straight line extracted by the straight line component extracting device 502 by one. B13: If the value of the counter K is positive, the process returns to step B9; otherwise, the process proceeds to step B14. B14: If the straight-line parameter value obtained by correcting the J-th straight-line parameter value in Step B7 matches the straight-line parameter value obtained by the straight-line component extraction device 402, Step B15
Otherwise, to step B16. B15: Update the success rate corresponding to the 1st validation grid by the following formula.

[0028]

(Equation 4)

B16: Decrease the counter J indicating the landmark information number by one. B17: If the value of the counter J is positive, the process returns to step B6; otherwise, the process proceeds to step B18. B18: Update the confidence value corresponding to the 1st validation grid by the following formula.

[0030]

(Equation 5)

B19: Decrement the counter I indicating the number of the validation grid by one. B20: If the value of the counter I is positive, the process returns to step B2; otherwise, the process proceeds to step B21. B21: all decrease confidence value confidence value matrix, by matching the number n _i the following formula using the corresponding to validation grid of the I-th and reallocated to the validation grid confidence value matrix.

[0032]

(Equation 6)

Now, a detailed description will be given of the estimated position correction. The correction based on the positional deviation ΔZ in step A9 of the flowchart of FIG. 2 showing the basic operation procedure of the present invention is as follows.

[0034]

(Equation 7)

Next, in the correction of the confidence value matrix, a memory area that overlaps before and after the correction is obtained, and the value before the correction is copied to the corresponding memory area of the confidence value matrix after the correction, without overlapping. An appropriate initial value (for example, 0) is set in the area.

As a method for estimating the position of the mobile robot itself by the conventional means, a method for statistically obtaining the estimated position is used, and it is assumed that the distribution of the confidence values representing the estimated position follows a normal distribution. Therefore, the estimated position is limited to an elliptical shape, and it is difficult to cope with inconsistency in the collation process. However, in the present invention, in order to eliminate the conventional difficulties, the reliability distribution of the estimated position is changed from a conventional validation gate to a validation grid, and is represented by a plurality of grid-like regions in a planar map. This facilitated the expansion of the estimation range, made it possible to deal with inconsistencies in the matching process without any restriction on the shape of the distribution, and achieved the initial purpose.

[0037]

As described above, according to the present invention, the range of estimation can be expanded by increasing the number of validation grids, and the size of the validation grid can be set arbitrarily. No matter how wide the range is, the likelihood of inconsistency in matching does not increase. Further, since the reliability value of the estimated position is managed for each validation grid, there is no restriction on the distribution method. This means that even if multiple conflicting hypotheses occur, it is possible to perform flexible processing considering all the hypotheses, and even if the validation is incorrect in one validation grid, other validations -If the grid is correctly collated, the possibility of having a large adverse effect on later estimation is reduced. In addition, since the fact that the matching has failed is also evaluated based on the statistical data of the landmark information, it is possible to realize an optimal estimation using the information efficiently.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a system configuration according to an embodiment of the present invention.

FIG. 2 is a flowchart illustrating an operation procedure of a position estimation method according to the present invention.

FIG. 3 is an explanatory diagram showing an example of a validation grid when a plurality of conflicting hypotheses according to the present invention occur.

FIG. 4 is a block diagram showing a system configuration of an embodiment when the present invention is actually applied to a mobile robot;

FIG. 5 is a flowchart of a method of updating a confidence value matrix processed by the confidence value matrix updating means of the present invention.

FIG. 6 is a flowchart showing the basic operation procedure of the present invention, in which the correction based on the positional deviation in step A9 in the flowchart of FIG. 2 is performed by correcting (updating) the estimated position and the reliability value matrix of the mobile robot due to the positional deviation at each time. Explanatory diagram showing an embodiment implemented as

FIG. 7 is an explanatory diagram of a validation gate used in a conventional position estimation method.

[Description of sign]

 101,401 Landmark information 102 Landmark extraction means 103 Environmental measurement means 104 Position estimation means 105 Reliability value matrix updating means 106 Impulse response matrix generation means 107 State measurement means 300 Validation grid 301,302 Estimated position (moving object) 402 Linear component extraction device 403 Laser scanner 404 Position estimation device 405 Reliability value matrix update device 406 Impulse response matrix generation device 407 State measurement device 701 Mobile robot 702 Validation gate 703 Center of validation gate 704,705,706,707 line segment

Claims (4)

[Claims] 1. The reliability of an estimated position of a moving object is managed using a confidence value matrix composed of a validation grid, and landmark pattern data generated by landmark extraction means is obtained in advance. A position estimating method comprising: updating the confidence value matrix by collating with landmark information at a landmark association point that has been used; 2. The position estimating method according to claim 1, wherein a validation grid in which the reliability value distribution state of the estimated position is represented by a plurality of grid regions is used. 3. The landmark information includes pattern data constituting a landmark and statistical data indicating the probability of occurrence of each pattern data. The landmark information is obtained in advance at a plurality of locations on an environment map. 2. The position estimating method according to claim 1, wherein landmark association points are used. 4. A convolution operation is performed between a matrix obtained by approximating a probability density distribution representing an estimation error due to a position movement from a previous process to a current process by a finite grid region and the confidence value matrix, Based on the result of matching between the pattern data of the landmark information generated by the landmark extracting means and the pattern data of the landmark information at the landmark association points, a reliability value matrix is obtained using statistical data belonging to each pattern data. The position estimating method according to claim 1, wherein the position is updated. [0001]