WO2021112362A1

WO2021112362A1 - Reinforced location recognition method and system through merging of results of multiple location recognition based on covariance matrix

Info

Publication number: WO2021112362A1
Application number: PCT/KR2020/009056
Authority: WO
Inventors: 배기덕; 이정우; 박지현; 엄태영; 최영호
Original assignee: 한국로봇융합연구원
Priority date: 2019-12-06
Filing date: 2020-07-09
Publication date: 2021-06-10
Also published as: KR102259247B1

Abstract

A reinforced location recognition system through merging of results of multiple location recognition, according to an embodiment of the present invention, comprises: a sensor unit including a plurality of sensors that sense a robot to derive different types of sensing values; an observation unit including a plurality of observation modules that calculate, from the sensing values derived by the plurality of sensors, the location of the robot via a plurality of different location recognition algorithms and output a plurality of different types of observation values; a conversion unit including a plurality of conversion modules that convert the plurality of different types of observation values into a covariance matrix that is a Gaussian distribution; a merge unit for merging the Gaussian distribution of the plurality of observation values into one Gaussian distribution; and a prediction unit that derives a prediction value by predicting the location of the robot from the merged one Gaussian distribution.

Description

Reinforced location recognition method and system through covariance matrix-based multi-location recognition result fusion

The present invention relates to location recognition technology, and more particularly, to a method and system for enhanced location recognition through fusion of multiple location recognition results based on a covariance matrix.

In most robot systems, especially mobile robot systems, location recognition to determine where the robot is by itself is an essential technology. Various sensors and methods are being studied for location recognition. Representative sensors used for location recognition include vision sensors that act as human eyes, GPS that can receive absolute coordinates from satellites, and LIDAR that emits lasers in multiple directions and measures the distance based on the degree of reflection. Each of the above sensors has unique characteristics, and a location recognition algorithm method has been developed using these characteristics. However, each sensor has its own weaknesses, which can lead to a decrease in performance depending on the environment in which it is used.

A typical position recognition technology using a vision sensor is ORB. It extracts feature points (corners, boundary points, etc.) from the 2D image, which is the output data of the vision sensor, and stores them in the form of three-dimensional points. The problem with this algorithm is that the distance to a point is calculated by comparing the image in the current frame with the image in the previous frame, so the distance value is inaccurate and the scale changes accordingly.

Location recognition technology using GPS is widely used in car navigation. As is well known, the error value occurs in meters, so it is difficult to use where precise location recognition is required.

As a location recognition technology using light detection and ranging (LIDAR), there is an extended Kalman filter (EKF) localization or AMCL. Both algorithms use the wheel rotation value as the predicted value and then use the LIDAR value to correct the predicted value. Location recognition accuracy is also relatively high due to the characteristic that LIDAR has more accurate measurement values compared to other sensors within 100m. However, due to problems such as lower resolution in the vertical direction compared to the horizontal direction, limitations in an environment where it is difficult to distinguish the difference only with the distance value of the laser, and the fact that complete measurement is impossible if there are obstacles around the LIDAR, all environments and situations It is clear that it is difficult to apply to

An object of the present invention, in consideration of the above points, is to achieve fusion of multi-position recognition results based on covariance matrix that can be used flexibly and universally even when various types and types of sensors are combined, thereby resolving restrictions on sensor configuration. To provide a reinforced location recognition method and system through

A sensor unit including a plurality of sensors for deriving different types of sensing values by sensing a robot is a system for reinforcing position recognition through fusion of multiple position recognition results according to a preferred embodiment of the present invention for achieving the object as described above. and an observation unit comprising a plurality of observation modules for calculating the position of the robot from the sensing values derived by the plurality of sensors through a plurality of different position recognition algorithms and outputting a plurality of observation values in different formats; a transform unit including a plurality of transform modules for transforming the plurality of observation values of different formats into a covariance matrix that is a Gaussian distribution; and a merging unit for merging the Gaussian distribution of the plurality of observation values into one Gaussian distribution; and a predictor for deriving a predicted value by predicting the position of the robot from one merged Gaussian distribution.

Further comprising; a feedback unit for feeding back the prediction value to each of the plurality of transformation modules of the transformation unit, wherein when the prediction value of the prediction unit is fed back from the feedback unit, each of the plurality of transformation modules is a Gaussian of the observed values transformed into the covariance matrix It is characterized in that the covariance matrix of observations is updated by merging the Gaussian distribution of the fed back prediction values into the distribution.

The plurality of sensors include a light detection and ranging (LIDAR) sensor and a vision sensor.

The location recognition algorithm is EKF (extended Kalman filter) Localization, EIF (Eukaryotic Initiation Factors) Localization, PF (Particle filter) Localization and AMCL (adaptive Monte Carlo Localization), Visual SLAM (simultaneous localization and mapping), ORB (Oriented FAST and Rotated BRIEF)-SLAM, Direct SLAM, LSD-SLAM, and characterized in that it includes at least one of Graph-based SLAM.

In the method for reinforcing position recognition through fusion of multiple position recognition results according to a preferred embodiment of the present invention for achieving the object as described above, each of a plurality of sensors of the sensor unit senses a robot to derive sensing values of different types and outputting, by a plurality of observation modules of the observation unit, a plurality of observation values in different formats by calculating the position of the robot through a plurality of different position recognition algorithms from the sensing values derived by the plurality of sensors; A step of transforming, by a plurality of transform modules of a transforming unit, the plurality of observation values of different formats into a covariance matrix that is a Gaussian distribution, and a step of merging, by a merging unit, a Gaussian distribution of the plurality of observation values into a single Gaussian distribution; and deriving a predicted value by predicting the position of the robot from the merged one Gaussian distribution.

In the method, after the step of transforming into the covariance matrix, and before the step of merging into one Gaussian distribution, when the prediction value of the prediction section is fed back from the feedback section, each of the plurality of transform modules is a Gaussian distribution of the observed values converted into the covariance matrix. The method further includes updating a covariance matrix in the form of a mantissa distribution of observed values by merging the Gaussian distribution of the predicted values fed back to .

The location recognition algorithm is EKF (extended Kalman filter) Localization, EIF (Eukaryotic Initiation Factors) Localization, PF (Particle filter) Localization and AMCL (adaptive Monte Carlo Localization), Visual SLAM (simultaneous localization and mapping), ORB (Oriented FAST and Rotated BRIEF)-SLAM, Direct SLAM, LSD-SLAM, and includes at least one of Graph-based SLAM.

According to the present invention, flexibility and versatility are supplemented through the covariance matrix-based integration algorithm to the existing position recognition method through sensor fusion in the position recognition technology essential for robots, and through this, it is easy to apply to various environments and is suitable for sensor configuration. to relieve restrictions.

1 is a diagram for explaining the configuration of an enhanced localization system through fusion of multiple location recognition results based on a covariance matrix according to an embodiment of the present invention.

2 is a flowchart for explaining a method for enhanced localization through fusion of multiple localization results based on a covariance matrix according to an embodiment of the present invention.

3 is a diagram illustrating a computing device according to an embodiment of the present invention.

Prior to the detailed description of the present invention, the terms or words used in the present specification and claims described below should not be construed as being limited to their ordinary or dictionary meanings, and the inventors should develop their own inventions in the best way. For explanation, it should be interpreted as meaning and concept consistent with the technical idea of the present invention based on the principle that it can be appropriately defined as a concept of a term. Therefore, the embodiments described in this specification and the configurations shown in the drawings are only the most preferred embodiments of the present invention, and do not represent all the technical ideas of the present invention, so various equivalents that can replace them at the time of the present application It should be understood that there may be water and variations.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In this case, it should be noted that the same components in the accompanying drawings are denoted by the same reference numerals as possible. In addition, detailed descriptions of well-known functions and configurations that may obscure the gist of the present invention will be omitted. For the same reason, some components are exaggerated, omitted, or schematically illustrated in the accompanying drawings, and the size of each component does not fully reflect the actual size.

First, an enhanced localization system through fusion of multiple location recognition results based on a covariance matrix according to an embodiment of the present invention will be described. 1 is a diagram for explaining the configuration of an enhanced localization system through fusion of multiple location recognition results based on a covariance matrix according to an embodiment of the present invention.

Referring to FIG. 1 , the enhanced location recognition system (10, hereinafter, abbreviated as 'reinforced location recognition system') through the fusion of multiple location recognition results according to an embodiment of the present invention includes a sensor unit 100 and an observation unit 200 ), a transform unit 300 , a merging unit 400 , a prediction unit 500 , and an update unit 600 .

The sensor unit 100 includes a plurality of

sensors

110 and 120 . The plurality of

sensors

110 and 120 are different types of sensors, and have different types of output values. Accordingly, the plurality of

sensors

110 and 120 sense the robot to be observed, and output the sensed values of in different formats, that is, according to the output format of each sensor. For example, the first sensor 110 may be a light detection and ranging (LIDAR) sensor, and the second sensor 120 may be a vision sensor. As described above, in the present embodiment, the sensor unit 100 is described as including a LIDAR sensor and a vision sensor, but the present invention is not limited thereto. The sensor unit 100 of the present invention may use two or more sensors, and there is no limitation on the type thereof.

The observation unit 200 includes a plurality of observation modules 210 , 220 , 230 , and 240 . The plurality of observation modules 210 , 220 , 230 , and 240 operate as a position recognition algorithm, calculate a position through this algorithm, and output an observation value. Each of the plurality of observation modules 210 , 220 , 230 , and 240 corresponds to any one of the plurality of sensors of the sensor unit 100 . Accordingly, the plurality of observation modules (210, 220, 230, 240) each of the observation module (210, 220, 230, 240) from the sensing value output by the corresponding sensor of the sensor unit 100, each of the position recognition algorithm It calculates the position of the robot and outputs a plurality of observation values in different formats.

In an embodiment of the present invention, the first observation module 210 is an extended Kalman filter (EKF) Localization algorithm, the second observation module 220 is a PF (Particle filter) Localization algorithm, and the third observation module 230 is an extended Kalman filter (EKF) algorithm. The Kalman filter) localization algorithm and the fourth observation module 240 apply an Oriented FAST and Rotated BRIEF (ORB)-simultaneous localization and mapping (SLAM) algorithm.

The first observation module 210 and the second observation module 220 correspond to the first sensor 110 . Accordingly, when LIDAR data that is a sensor value of the first sensor 110 is input, the first observation module 210 calculates the position of the robot using the EKF algorithm from the LIDAR data and outputs an observation value. In addition, when LIDAR data that is a sensor value of the first sensor 110 is input, the second observation module 220 calculates the position of the robot from the LIDAR data through a PF algorithm and outputs an observation value. The third observation module 230 and the fourth observation module 240 correspond to the second sensor 120 . Accordingly, when vision sensor data that is a sensor value of the second sensor 120 is input, the third observation module 230 calculates the position of the robot from the vision sensor data through the EKF algorithm and outputs an observation value. In addition, when vision sensor data that is a sensor value of the second sensor 120 is input, the fourth observation module 240 calculates the position of the robot from the vision sensor data through an ORB-SLAM algorithm and outputs an observation value.

On the other hand, although it has been described that four algorithms are used as the position recognition algorithm, the present invention is not limited thereto, and if it is a position recognition algorithm capable of calculating the position of the robot from a sensor value, there is no limitation of the type. Such location recognition algorithms include, for example, extended Kalman filter (EKF) Localization, Eukaryotic Initiation Factors (EIF) Localization, Particle filter (PF) Localization and adaptive Monte Carlo Localization (AMCL), Visual SLAM (simultaneous localization and mapping), ORB (Oriented). FAST and Rotated BRIEF)-SLAM, Direct SLAM, LSD-SLAM and Graph-based SLAM may be exemplified.

The conversion unit 300 includes a plurality of conversion modules 310 . Each of the transformation modules 310 corresponds one-to-one to the observation modules 210, 220, 230, and 240.

When an observation value that is an output of the corresponding observation module 210 , 220 , 230 , 240 is input, the transformation module 310 transforms the input observation value into a covariance matrix in the form of a Gaussian distribution. At this time, the transformation module 310 determines whether the observed values output from the corresponding observation modules 210 , 220 , 230 , and 240 are covariance matrices in the form of a Gaussian distribution. As a result of this determination, when the corresponding observation value is not the covariance matrix, the transformation module 310 converts the observed value into the covariance matrix. On the other hand, when the corresponding observation value is a covariance matrix, the transformation module 310 bypasses the observation value. As described above, the first observation module 210 and the third observation module 230 use the EKF algorithm, and the second observation module 220 uses the PF algorithm. Accordingly, the observation values output from the first to third observation modules 210 , 220 , and 230 are all output as a covariance matrix in the form of a Gaussian distribution. Accordingly, the transformation module 310 corresponding to the first to third observation modules 210 , 220 , 230 bypasses the observation values output from the first to third observation modules 210 , 220 , and 230 . On the other hand, the fourth observation module 240 uses the ORB-SLAM algorithm. Therefore, the observation value that is the output of the fourth observation module 240 is output as a two-dimensional coordinate value. Accordingly, the transformation module 310 corresponding to the fourth observation module 240 separately transforms the observation value output from the fourth observation module 240 into a covariance matrix in the form of a Gaussian distribution. At this time, the transformation module 310 converts the two-dimensional coordinate value, which is the observation value, which is the output of the fourth observation module 240, into a Gaussian distribution form through a transformation equation using the error of the standard of the second sensor 120 and the algorithm itself as a variable. It can be converted to a covariance matrix.

As described above, the transformation module 310 converts the observation values of the corresponding observation modules 210 , 220 , 230 , and 240 into a covariance matrix in the form of a Gaussian distribution, and then receives the position of the robot from the update unit 600 . When the predicted values are fed back, the covariance matrix in the form of a mantissa distribution of the observed values may be updated by merging the Gaussian distribution of the predicted values with the Gaussian distribution of the observed values. On the other hand, if the transformation module 310 does not have a predicted value for the position of the robot fed back from the update unit 600 , the transformation module 310 outputs the covariance matrix of the observed values as it is.

When the observation values converted from the plurality of transformation modules 310 into the covariance matrix having the mantissa distribution are input, the merging unit 400 merges the mantissa distributions of the plurality of observation values into one Gaussian distribution.

The prediction unit 500 derives a prediction value for predicting the position of the robot from one Gaussian distribution merged by the merging unit 400 .

The update unit 600 feeds back the output of the prediction unit 500 to each of the plurality of transformation modules 310 of the transformation unit 300 . Then, each of the plurality of transformation modules 310 transforms the observation values of the corresponding observation modules 210, 220, 230, 240 into a covariance matrix in the form of a Gaussian distribution, and then adds the Gaussian distribution of the predicted values to the Gaussian distribution of the observation values. By merging, the covariance matrix in the form of mantissa distribution of observations can be updated. Here, the Gaussian distribution of the fed back predicted value may be the probability of the robot's position at time t, and the observed value may be the probability of the robot's position at time t+1. Therefore, the position of the robot can be predicted more precisely by accumulating and merging Gaussian distributions according to the repetition of the above-described feedback. If the noise of the sensor of the sensor unit 100 occurs, the Gaussian distributions of the covariance matrix of the transform unit 300 increase in range again, but if the feedback process is repeated, all the covariance matrices gradually converge to one distribution. do. In other words, it naturally acts as a noise filter.

Next, a method for reinforcing location recognition through fusion of multiple location recognition results based on a covariance matrix according to an embodiment of the present invention will be described. 2 is a flowchart for explaining a method for enhanced localization through fusion of multiple localization results based on a covariance matrix according to an embodiment of the present invention.

Referring to FIG. 2 , the plurality of

sensors

110 and 120 of the sensor unit 100 sense the robot in step S110 to derive sensing values of different types. Here, the plurality of

sensors

110 and 120 are different types of sensors, and have different types of output values. That is, the plurality of

sensors

110 and 120 sense the robot to be observed and output the sensed values according to the respective output formats of the

sensors

110 and 120 . For example, the first sensor 110 may be a light detection and ranging (LIDAR) sensor, and the second sensor 120 may be a vision sensor.

Each of the plurality of observation modules 210 , 220 , 230 , and 240 corresponds to any one of the plurality of sensors of the sensor unit 100 . Accordingly, the plurality of observation modules (210, 220, 230, 240) each of the observation modules (210, 220, 230, S120) from the sensing value output by the corresponding sensor (110, 120) of the sensor unit 100 in step S120, 240) Each position recognition algorithm calculates the position of the robot and outputs a plurality of observation values in different formats. In an embodiment of the present invention, the first observation module 210 is the EKF algorithm, the second observation module 220 is the PF algorithm, the third observation module 230 is the EKF algorithm, and the fourth observation module 240 is ORB-SLAM. Algorithms can be applied. In addition, the first observation module 210 and the second observation module 220 correspond to the first sensor 110 , and the third observation module 230 and the fourth observation module 240 are the second sensor 120 . respond to Accordingly, the first observation module 210 calculates the position of the robot using the EKF algorithm from the LIDAR data that is the sensor value of the first sensor 110 and outputs an observation value, and the second observation module 220 outputs the observation value of the first sensor From the LIDAR data, which is the sensor value of (110), the position of the robot is calculated through the PF algorithm and the observed value is output. In addition, the third observation module 230 calculates the position of the robot from the vision sensor data that is the sensor value of the second sensor 120 through the EKF algorithm and outputs the observation value, and the fourth observation module 240 outputs the second observation value. From the vision sensor data, which is the sensor value of the sensor 120 , the position of the robot is calculated through the ORB-SLAM algorithm, and the observed value is output.

Each of the plurality of transformation modules 310 of the transformation unit 300 corresponds one-to-one to the observation modules 210 , 220 , 230 , and 240 . Accordingly, when an observation value that is an output of the corresponding observation module 210 , 220 , 230 , 240 is input in step S130 , each of the plurality of transformation modules 310 of the transformation unit 300 converts the input observation value into a Gaussian distribution. Convert it to a covariance matrix of the form At this time, the transform module 310 converts the observed values into a covariance matrix when the observed values output from the corresponding observation modules 210, 220, 230, and 240 are not the covariance matrix in the form of a Gaussian distribution. convert On the other hand, when the corresponding observation value is a covariance matrix, the transformation module 310 may bypass the observation value. As described above, the first observation module 210 and the third observation module 230 use the EKF algorithm, and the second observation module 220 uses the PF algorithm. Accordingly, the observation values output from the first to third observation modules 210 , 220 , and 230 are all output as a covariance matrix in the form of a Gaussian distribution. Accordingly, the transformation module 310 corresponding to the first to third observation modules 210 , 220 , 230 bypasses the observation values output from the first to third observation modules 210 , 220 , and 230 . On the other hand, the fourth observation module 240 uses the ORB-SLAM algorithm. Therefore, the observation value that is the output of the fourth observation module 240 is output as a two-dimensional coordinate value. Accordingly, the transformation module 310 corresponding to the fourth observation module 240 separately transforms the observation value output from the fourth observation module 240 into a covariance matrix in the form of a Gaussian distribution.

Next, it is determined whether feedback exists in step S140, and if there is feedback, each of the plurality of transform modules 310 of the transform unit 300 is applied to the Gaussian distribution of the observation values converted into the covariance matrix in step S150. The covariance matrix in the form of a mantissa distribution of observations is updated by merging the Gaussian distribution of the fed back prediction values. If the observed value is the probability of the robot's position at time t, then the Gaussian distribution of the fed back predicted value can be the probability of the robot's position at time t-1.

Next, when an observation value that is a covariance matrix having a mantissa distribution is input from the plurality of transformation modules 310, the merging unit 400 merges the mantissa distribution of the plurality of observation values into one Gaussian distribution in step S160. .

Then, the prediction unit 500 derives a prediction value for predicting the position of the robot from one Gaussian distribution merged by the merging unit 400 in step S170 .

The update unit 600 feeds back the output of the prediction unit 500 to each of the plurality of transformation modules 310 of the transformation unit 300 in step S180 .

Since steps S110 to S180 described above are for continuously tracking the position of the robot, it is continuously repeated until the process of tracking the position of the robot is terminated.

3 is a diagram illustrating a computing device according to an embodiment of the present invention. The computing device TN100 of FIG. 3 may be an apparatus for the enhanced location recognition system and method described herein.

The computing device TN100 may include at least one processor TN110 , a transceiver device TN120 , and a memory TN130 . Also, the computing device TN100 may further include a storage device TN140 , an input interface device TN150 , an output interface device TN160 , and the like. Components included in the computing device TN100 may be connected by a bus TN170 to communicate with each other.

The processor TN110 may execute a program command stored in at least one of the memory TN130 and the storage device TN140. The processor TN110 may mean a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to an embodiment of the present invention are performed. The processor TN110 may be configured to implement procedures, functions, methods, and the like described in connection with an embodiment of the present invention. The processor TN110 may control each component of the computing device TN100 .

Each of the memory TN130 and the storage device TN140 may store various information related to the operation of the processor TN110 . Each of the memory TN130 and the storage device TN140 may be configured as at least one of a volatile storage medium and a non-volatile storage medium. For example, the memory TN130 may include at least one of a read only memory (ROM) and a random access memory (RAM).

The transceiver TN120 may transmit or receive a wired signal or a wireless signal. The transceiver TN120 may be connected to a network to perform communication.

Meanwhile, the method according to the embodiment of the present invention described above may be implemented in the form of a program readable by various computer means and recorded in a computer readable recording medium. Here, the recording medium may include a program command, a data file, a data structure, etc. alone or in combination. The program instructions recorded on the recording medium may be specially designed and configured for the present invention, or may be known and available to those skilled in the art of computer software. For example, the recording medium includes magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, and magneto-optical media such as floppy disks ( magneto-optical media) and hardware devices specially configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions may include high-level languages that can be executed by a computer using an interpreter or the like as well as machine language such as generated by a compiler. Such hardware devices may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

On the other hand, the conventional location recognition method through the fusion of sensors has the advantage that various sensor data can be easily used, but if the configuration of the sensor is changed, the location recognition method may also be different. That is, there is a disadvantage that the sensor configuration is not flexible. However, the present invention transforms the position recognition result using each sensor into a covariance matrix form, and based on this, it is possible to provide a fusion position recognition method and system that is robust to various environments and can be used in any sensor configuration.

Although the present invention has been described above using several preferred embodiments, these examples are illustrative and not restrictive. As such, those of ordinary skill in the art to which the present invention pertains will understand that various changes and modifications can be made in accordance with the doctrine of equivalents without departing from the spirit of the present invention and the scope of rights set forth in the appended claims.

The present invention supplements the flexibility and versatility of a position recognition method essential for a robot through a covariance matrix-based integration algorithm, and through this, it is easy to apply to various environments and eliminates restrictions on sensor configuration. Accordingly, the present invention has industrial applicability because it has sufficient potential for commercialization or business, as well as to the extent that it can be clearly implemented in reality.

Claims

In the reinforced location recognition system through the fusion of multiple location recognition results,

a sensor unit including a plurality of sensors for sensing the robot and deriving different types of sensing values;

an observation unit including a plurality of observation modules for calculating the position of the robot from the sensing values derived by the plurality of sensors through a plurality of different position recognition algorithms and outputting a plurality of observation values in different formats;

a transform unit including a plurality of transform modules for transforming the plurality of observation values of different formats into a covariance matrix having a Gaussian distribution;

a merging unit merging the Gaussian distributions of the plurality of observation values into one Gaussian distribution; and

A prediction unit for deriving a predicted value by predicting the position of the robot from the merged one Gaussian distribution;

Enhanced location awareness system.
According to claim 1,

It further comprises; a feedback unit for feeding back the predicted value to each of the plurality of transformation modules of the transformation unit,

When the prediction value of the prediction unit is fed back from the feedback unit,

Each of the plurality of transformation modules updates the covariance matrix of observations by merging the Gaussian distribution of the fed back prediction values with the Gaussian distribution of the observation values transformed into the covariance matrix.

Enhanced location awareness system.
According to claim 1,

The plurality of sensors

comprising a light detection and ranging (LIDAR) sensor and a vision sensor

Enhanced location awareness system.
According to claim 1,

The location recognition algorithm is

EKF (extended Kalman filter) Localization, EIF (Eukaryotic Initiation Factors) Localization, PF (Particle filter) Localization and adaptive Monte Carlo Localization (AMCL), Visual SLAM (simultaneous localization and mapping), ORB (Oriented FAST and Rotated BRIEF)-SLAM , Direct SLAM, LSD-SLAM and Graph-based SLAM characterized in that it comprises at least one

Enhanced location awareness system.
In the reinforced location recognition method through fusion of multiple location recognition results,

deriving different types of sensing values by sensing the robot by each of the plurality of sensors of the sensor unit;

outputting, by each of the plurality of observation modules of the observation unit, the plurality of observation values in different formats by calculating the position of the robot through a plurality of different position recognition algorithms from the sensing values derived by the plurality of sensors;

converting, by a plurality of transformation modules of a transformation unit, the plurality of observation values of different formats into a covariance matrix having a Gaussian distribution;

merging, by a merging unit, Gaussian distributions of the plurality of observation values into one Gaussian distribution; and

Deriving a predicted value by a prediction unit predicting the position of the robot from the merged one Gaussian distribution;

Reinforced position recognition method.
6. The method of claim 5,

After the step of converting to the covariance matrix, before the step of merging into the single Gaussian distribution,

When the prediction value of the prediction unit is fed back from the feedback unit,

The step of updating the covariance matrix in the form of a mantissa distribution of the observation values by merging the Gaussian distribution of the predicted values fed back to the Gaussian distribution of the observation values transformed into the covariance matrix by each of the plurality of transformation modules;

Reinforced position recognition method.
6. The method of claim 5,

The plurality of sensors

comprising a light detection and ranging (LIDAR) sensor and a vision sensor

Reinforced position recognition method.
6. The method of claim 5,

The location recognition algorithm is

EKF (extended Kalman filter) Localization, EIF (Eukaryotic Initiation Factors) Localization, PF (Particle filter) Localization and adaptive Monte Carlo Localization (AMCL), Visual SLAM (simultaneous localization and mapping), ORB (Oriented FAST and Rotated BRIEF)-SLAM , Direct SLAM, characterized in that it comprises at least one of LSD-SLAM and Graph-based SLAM

Reinforced position recognition method.