Method for localizing a robot in a localization plane

• US 10,197,399 B2
• Filed: 04/14/2015
• Issued: 02/05/2019
• Est. Priority Date: 04/14/2014
• Status: Expired due to Fees

• Current Assignee: Association Pour La Recherche et Le Developpement Des Methodes et Processus Industriels Armines, SoftBank Robotics Europe SAS
• Original Assignee: Association Pour La Recherche et Le Developpement Des Methodes et Processus Industriels Armines, SoftBank Robotics Europe SAS
• Inventors: Wirbel, Emilie, de la Fortelle, Arnaud
• Primary Examiner(s): Figueroa, Jaime
• Application Number:
  US15/128,901
• Publication Number:
  US 20170131102A1
• Time in Patent Office: 1,393 Days
• Field of Search: 700245, 700259
• CPC Class Codes: G01C 21/1652 with ranging devices, e.g. LIDAR or RADAR
• CPC Class Codes: G01C 21/1656 with passive imaging devices, e.g. cameras
• CPC Class Codes: G05D 1/0272 comprising means for registering the travel distance, e.g. revolutions of wheels measuring distance traversed on the ground by vehicles, e.g. using odometers G01C22/00

Abstract

A method for localizing a robot in a localization plane with a bi-dimentional reference with axis x and y comprises: determining by odometry an estimation of coordinates x1 and y1 and orientation θ1 of the robot; determining an estimation θ2 of the orientation of the robot using a virtual compass; determining an estimation θ3 of the orientation of the robot by correlating parts of a reference and a query panorama; determining an estimation x4, y4 of the robot position using Iterative Closest Points; determining standard deviations σ_x1, σ_x2, σ_θ1 σ_θ2, σ_θ3, σ_x4, σ_y4 of the estimations; determining probability distributions G(x1), G(y1), G(θ1), G(θ2), G(θ3), G(x4), G(y4) of each estimation using standard deviations; determining three global distributions GLOB(x), GLOB(y), GLOB(θ) and a global estimation xg, yg of the coordinates of the robot in the localization plane and a global estimation θg of its orientation by applying maximum likelihood to global distributions.

Claims

1. A method for localizing a robot in a localization plane associated with a bi-dimensional reference with two axes x and y comprising the following steps:

determining by odometry an estimation of the coordinates x1 and y1 of the robot in the localization plane as well as an estimation of its orientation θ

1 relatively to a reference direction;

determining an estimation θ

2 of the orientation of the robot by using a virtual compass which identifies at least two pairs of points of interest, first points of each pair being identified in a reference panorama and second point of each pair being identified in a query panorama, this step being initialized with θ

1;

determining an estimation θ

3 of the orientation of the robot by correlating parts of the reference panorama with parts of the query panorama and by identifying when that correlation is maximized, this step being initialized with one of the previous estimations of the orientation;

determining an estimation x4, y4 of the robot position in the localization plane by using an Iterative Closest Points technique, this step being initialized with x1 and y1, the iterative Closest Points techniques using a 3D point cloud as an input and preliminary hypotheses in orientation;

determining the standard deviations σ

_x1, σ

_y1, σ

_θ

1 σ

_θ

2, σ

_θ

3, σ

_x4, σ

_y4 of the aforementioned estimations;

determining Gaussian probability distributions G(x1), G(y1), G(θ

1), G(θ

2), G(θ

3), G(x4) and G(y4) of each available estimation using said standard deviations;

determining three global distributions GLOB(x), GLOB(y) and GLOB(θ

) respectively for the coordinates along the x and y axis and for the orientation θ

of the robot by combining said Gaussian probability distributions and determining a global estimation xg, yg of the coordinates of the robot in the localization plane as well as an global estimation θ

g of its orientation by applying the method of maximum likelihood to the global distributions.

2. The method according to claim 1, wherein the estimations provided by a given step are used by a subsequent step only if considered as reliable.

3. The method according to claim 2, wherein an estimation is considered as reliable when its standard deviation is lower than a predefined threshold.

4. The method according to claim 1, wherein the global probability distributions are derived as follow:

GLOB(x)=G(x1)*G(x4)

GLOB(y)=G(y1)*G(y4)

GLOB(θ

)=G(θ

1)*G(θ

2)*G(θ

3).

5. The method according to claim 1, wherein θ

3 value is estimated based on an image template matching which is performed over two pyramids of images, a first pyramid of images being generated from a single reference image by downscaling it using several scaling steps, the second pyramid of images being generated from a single query image by downscaling it using several scaling steps.

6. A humanoid robot comprising at least:

2D RGB camera in order to construct a query panorama comprising at least one reference image;

processing capabilities to implement a method for localizing said robot, based on said query panorama, in a localization plane associated with a bi-dimensional reference with two axes x and y comprising the following steps;

determining by odometry an estimation of the coordinates x1 and y1 of the robot in the localization plane as well as an estimation of its orientation θ

1 relatively to a reference direction;

determining an estimation θ

2 of the orientation of the robot by using a virtual compass which identifies at least two pairs of points of interest, first points of each pair being identified in a reference panorama and second point of each pair being identified in said query panorama, this step being initialized with θ

1;

determining an estimation θ

3 of the orientation of the robot by correlating parts of the reference panorama with parts of the query panorama and by identifying when that correlation is maximized, this step being initialized with one of the previous estimations of the orientation;

determining an estimation x4, y4 of the robot position in the localization plane by using an Iterative Closest Points technique, this step being initialized with x1 and y1, the iterative Closest Points techniques using a 3D point cloud as an input and preliminary hypotheses in orientation;

determining the standard deviations σ

_x1, σ

_y1, σ

_θ

1 σ

_θ

2, σ

_θ

3, σ

_x4, σ

_y4 of the aforementioned estimations;

determining Gaussian probability distributions G(x1), G(y1), G(θ

1), G(θ

2), G(θ

3), G(x4) and G(y4) of each available estimation using said standard deviations;

determining three global distributions GLOB(x), GLOB(y) and GLOB(θ

) respectively for the coordinates along the x and y axis and for the orientation θ

of the robot by combining said Gaussian probability distributions and determining a global estimation xq, yg of the coordinates of the robot in the localization plane as well as an global estimation θ

g of its orientation by applying the method of maximum likelihood to the global distributions.

7. The humanoid robot according to claim 6, wherein a 3D sensor is used to compute point clouds in order to implement the Iterative Closest Point Technique.

8. A computer program product, stored on a non-transitory computer readable medium comprising code instructions for causing a computer to implement a method of for localizing a robot in a localization plane associated with a bi-dimensional reference with two axes x and y comprising the following steps:

determining by odometry an estimation of the coordinates x1 and y1 of the robot in the localization plane as well as an estimation of its orientation θ

1 relatively to a reference direction;

determining an estimation θ

2 of the orientation of the robot by using a virtual compass which identifies at least two pairs of points of interest, first points of each pair being identified in a reference panorama and second point of each pair being identified in said query panorama, this step being initialized with θ

1;

determining an estimation θ

3 of the orientation of the robot by correlating parts of the reference panorama with parts of the query panorama and by identifying when that correlation is maximized, this step being initialized with one of the previous estimations of the orientation;

determining an estimation x4, y4 of the robot position in the localization plane by using an Iterative Closest Points technique, this step being initialized with x1 and y1, the iterative Closest Points techniques using a 3D point cloud as an input and preliminary hypotheses in orientation;

determining the standard deviations σ

_x1, σ

_y1, σ

_θ

1 σ

_θ

2, σ

_θ

3, σ

_x4, σ

_y4 of the aforementioned estimations;

determining Gaussian probability distributions G(x1), G(y1), G(θ

1), G(θ

2), G(θ

3), G(x4) and G(y4) of each available estimation using said standard deviations;

determining three global distributions GLOB(x), GLOB(y) and GLOB(—

0) respectively for the coordinates along the x and y axis and for the orientation θ

of the robot by combining said Gaussian probability distributions and determining a global estimation xg, yg of the coordinates of the robot in the localization plane as well as an global estimation θ

g of its orientation by applying the method of maximum likelihood to the global distributions.