JP2007316799A - Autonomous mobile robot having learning function - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an autonomous mobile robot having learning function, capable of correcting and adding a traveling rule, according to the circumstances of a traveling path and utilizing the corrected and added traveling rules, together with preliminarily prepared traveling rules for the next traveling. <P>SOLUTION: This autonomous mobile robot with learning function is provided with an environment map 5, in which the environment information of a traveling path is stored; and a learning function 6 having an SOM 8 in which the traveling rule is preliminarily learned. A self-location is recognized by a self-location recognizing function 2 from the circumstances of the traveling path detected by sensors 1 and is inputted to the environment map 5, and an obstacle is recognized by an obstacle recognizing function 3, from the circumstances of the traveling path detected by the sensors 1 and is inputted to the environment map 5; and when the environment information of the environment map 5 is inputted, the learning function 6 outputs steering information, corresponding to the environment information inputted by using an SOM 8. Thus, move this autonomous mobile robot can be moved to its destination safely and surely. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an autonomous mobile robot having a learning function that has an autonomous movement capability and a learning function, acquires new knowledge every time it travels, and can be used for subsequent traveling.

A conventional autonomous mobile robot is provided with landmarks such as landmarks and electromagnetic fields on a traveling path, and travels along those landmarks, and can move only within a range of predetermined rules. In addition, it is difficult to continue traveling when the rules prepared in advance cannot be used during traveling.
Further, as a conventional autonomous mobile robot, Patent Document 1 describes a robot using a reinforcement learning method. Patent Document 1 describes a technique in which a robot (agent) travels in a traveling environment, defines a reward based on evaluation of the quality, and improves the evaluation.
Patent Document 2 describes a mobile body equipped with a learning function based on a neural network. In Patent Document 2, a travel rule determined by human empirical rules is prepared, SOM (Self-Organizing Map) is learned and used for travel. Basically, travel is performed according to the rules prepared first. It is something to be made.

Japanese Patent Laying-Open No. 2005-293154 (pages 4 to 14, FIG. 2) JP 2000-122991 A (pages 5-9, FIG. 1)

Conventional autonomous mobile robots are equipped with some landmarks such as landmarks, electromagnetic fields, etc. on the traveling road, and travel along those landmarks. There's a problem.
In Patent Document 1, a method is used in which a robot (agent) travels in a traveling environment, and a reward is defined by evaluation of the goodness and the evaluation is improved. In this method, traveling evaluation is performed using the traveling environment as teacher data. Since the rules necessary for driving are generated with priority, when a large calculation load is required and a large number of complicated driving rules are required, problems remain in consideration of practicality and adaptability to the diversity of driving.
Regarding Patent Document 2, although travel rules determined based on human empirical rules are prepared, SOM is learned and used for travel, the travel is basically performed according to the rules prepared first, and cannot cope with changes in the route. There is a problem.

  The present invention has been made to solve the above-described problems. The travel rules are corrected and added according to the conditions of the travel path, and can be used for the next travel together with the travel rules prepared in advance. The purpose is to obtain an autonomous mobile robot with learning function.

  In the autonomous mobile robot having the learning function according to the present invention, the environment map storing the environment information of the travel path, the sensors for detecting the status of the travel path during the travel, the status of the travel path detected by the sensors Self-position recognition means for recognizing the self-position from the vehicle and inputting it to the environment map, obstacle recognition means for recognizing an obstacle from the road conditions detected by the sensors and inputting it to the environment map, and environment information of the environment map And learning means for outputting steering information. The learning means has a self-organizing map that has learned a driving rule having environmental information as an input value, and corresponds to the driving rule of the self-organizing map. Steering information is output.

  As described above, according to the present invention, the self-position is determined from the environment map storing the environment information of the travel path, the sensors for detecting the condition of the travel path during traveling, and the condition of the travel path detected by the sensors. Self-recognition means for recognizing and inputting to the environment map, obstacle recognition means for recognizing obstacles from the road conditions detected from the sensors, and input to the environment map, and environment information of the environment map are input, A learning means for outputting steering information is provided. The learning means has a self-organizing map in which driving rules having environmental information as input values have been learned, and outputs steering information according to the driving rules of the self-organizing map. Therefore, it is possible to modify and add the travel rules according to the conditions of the travel route, and use them for the next travel together with the travel rules prepared in advance, so that you can move to the destination safely and reliably. That.

Embodiment 1 FIG.
An autonomous mobile robot (hereinafter also simply referred to as an autonomous mobile robot) having a learning function according to the present invention first searches for a route, recognizes an environment from sensor information, and uses a self-organizing map (SOM) learned in advance. The direction and speed are determined based on the driving method. Even if inconveniences such as hitting an obstacle occur during driving, the driving rules are corrected, the knowledge obtained there is additionally learned as driving rules, and newly acquired driving rules and driving prepared in advance It is an autonomous mobile robot that makes it possible to use rules for the next run and "gets smarter every time you run."
This will be described in detail below.

1 is a block diagram showing an autonomous mobile robot having a learning function according to Embodiment 1 of the present invention.
In FIG. 1, sensors 1 are attached to an autonomous mobile robot and detect landmarks, obstacles, and the like provided on a travel path. Encoders, gyro sensors, ultrasonic sensors, Doppler sensors and the like are included. The self-position recognition function 2 (self-position recognition means) recognizes the self-position by a signal from the sensors 1 and writes it in the environment map 5. The obstacle recognition function 3 (obstacle recognition means) recognizes an obstacle based on the signal from the sensors 1 and the video from the camera 4 and writes it in the environment map 5. The camera 4 captures the road and outputs the image to the obstacle recognition function 3.
The environment map 5 is pre-existing on a traveling path such as a wall, a pillar, or a fixed obstacle in the whole indoor so that one dot becomes 1 [cm] on a personal computer mounted on an autonomous mobile robot having a learning function. It is the map information that wrote the information. When the destination of the autonomous mobile robot is written in the environment map 5 by the touch panel 10 (input means) for specifying the destination, and the route to this destination is determined, the environment information written in the environment map 5 is: This is an input to the learning function 6 (learning means).

The learning function 6 outputs the travel target angle and the action pattern to the drive system input 7 so as to guide the autonomous mobile robot having the learning function to the destination using the input information.
The learning function 6 is an algorithm simulating the information processing method of the human brain, and is a multidimensional data classification / prediction method. This learning function 6 includes a travel rule addition function 61 (travel rule addition means) that corrects and corrects a travel rule and adds a travel rule if a travel rule is inappropriate, a known obstacle, and an unknown travel obstacle. The vehicle has an obstacle avoidance function 62 (obstacle avoidance means) that travels while avoiding an object.
The travel rule addition function 61 and the obstacle avoidance function 62 apply their learning functions using the SOM 8.
The SOM 8 has a travel rule as a node, and can add a travel rule by learning. The environment map 5 is responsible for confirming the route by executing the travel rules by the SOM 8. If this check confirms that travel is appropriate, the autonomous mobile robot will continue to travel. If travel is inappropriate due to the appearance of walls or obstacles (this is input to the environment map 5), the travel rules will be corrected. Then, run rules are added at this time.
The travel target angle and action pattern for guiding the autonomous mobile robot having the learning function, which is input to the drive system input 7, to the destination are transmitted to the drive system 9, and the autonomous mobile robot is guided to the destination.

  The self-position recognition function 2, the obstacle recognition function 3, the environment map 5, and the learning function 6 in FIG. 1 can be configured by software using a microcomputer.

FIG. 2 is a diagram showing an autonomous mobile robot model having a learning function on the environment map according to Embodiment 1 of the present invention.
In FIG. 2, the center positions (X, Y) of the left and right wheels of the autonomous mobile robot model 11 are shown.
FIG. 3 is a flowchart showing a flow of processing in one cycle of the autonomous mobile robot having a learning function according to Embodiment 1 of the present invention.
FIG. 4 is a diagram showing an environment map in which obstacles of the autonomous mobile robot having a learning function according to the first embodiment of the present invention are written.
FIG. 4 shows the autonomous mobile robot model 11 traveling between the walls 14 avoiding the obstacle 13.

Next, the operation will be described.
Considering the traveling of an autonomous mobile robot, safe and reliable operation is a prerequisite. In the present invention, an environment map 5 is introduced as a method for improving the safety and reliability of traveling. The environment map 5 is written with pre-existing information such as walls, pillars, fixed obstacles, etc. of the whole indoor so that 1 dot becomes 1 [cm] on a personal computer mounted on the autonomous mobile robot, and the shape of the map This is map information in which a boundary line for switching an action pattern is also written. In this environment map 5, when the route to the destination is determined for each area surrounded by the boundary line, the environment information is given to the learning function 6, and the learning function 6 guides the autonomous mobile robot to the destination. The driving target angle and the action pattern are input to the drive system input 7 as described above. Here, the autonomous mobile robot model 11 on the environment map 5 is expressed as “the outermost frame of the entire vehicle body including the sensor is rectangular” as shown in FIG.

By checking the position of the autonomous mobile robot with learning function and obstacles including unknown moving objects with the sensors 1, the environment map 5 is constantly updated, and the driving control is performed while switching the driving pattern by the boundary line. Drive to the goal.
In an autonomous mobile robot having a learning function, a self-position recognition function 2 is used to determine a self-position by using a sensor value obtained during traveling, and the obstacle recognition means 3 finds an unknown stationary and moving obstacle. The self-position and the unknown obstacle are written in the environment map 5 and the environment map 5 is updated.
While traveling, the travel rules set in advance are followed, but when the vehicle cannot adaptively travel due to environmental conditions, the travel rules are corrected and adjusted to travel. An SOM algorithm that can be additionally learned is installed as the learning function 6 so that the travel rules obtained here can be used for the next travel. That is, it is an autonomous mobile robot having a learning function of “being smarter every time it travels”.

FIG. 3 shows the flow of travel of the autonomous mobile robot. Here, when the obstacle 13 is recognized during traveling, the obstacle 13 is written in the environment map 5 and an obstacle avoidance action is taken. At this time, at the same time, the autonomous mobile robot updates its own position while taking the avoidance action of the obstacle 13 written in the environment map 5 and travels while writing in the vicinity of the obstacle 13 as shown in FIG.
In FIG. 3, a travel route is determined in step S1. In step S2, the value of each sensor is acquired, and in step S3, information of self position recognition and obstacle recognition is performed using these values. Next, in step S4, the environment map 5 is updated using these. If an obstacle is detected in step S5, an obstacle recognition process is performed in step S3, and the environment map is updated in step S4.
Next, in step S6, an SOM input value is acquired from the environment map 5, SOM inference is executed in step S7, and output to the drive system 9 in step S8 to run the autonomous mobile robot.

Hereinafter, the traveling of the autonomous mobile robot will be described in detail.
First, the self-position recognition of the autonomous mobile robot by the self-position recognition function 2 will be described by means 1 to means 7.
Means 1.
As shown in FIG. 2, the position information of the autonomous mobile robot having the learning function is obtained from the center coordinates X and Y between the left and right wheels of the autonomous mobile robot on the environment map 5 and the angle θ of the vehicle body. The current position is written on the environment map 5 from this value.

Mean 2.
In the means 1, the left and right wheels of the autonomous mobile robot model 11 on the environment map as shown in FIG. 2 using the values of the encoder, the gyro sensor, the ultrasonic sensor, the Doppler sensor and the processing values of the camera image. A method for calculating the center coordinates X, Y and the vehicle body angle θ between them is shown below.
The input value to the self-position recognition means 2 is the rotation number of the left and right wheels obtained from the encoder and the attitude angle change value obtained from the gyro sensor. Using these pieces of information, from the self-position of the autonomous mobile robot having the learning function in the previous control operation (previous cycle), the left and right wheels are moved in the direction of the obtained posture angle change value θ ₁ as shown in FIG. The travel distance values e1 and e2 calculated from the number of revolutions are linearly extended, and the center coordinates (X _l , Y _l ) and (X _r , Y _r ) of the left and right wheels of the autonomous mobile robot are calculated. Next, an output value (X ₁ , Y ₁ , θ ₁ ) is obtained by obtaining the midpoint of these two points.

Means 3.
In the first means, the rotational speed of the left and right wheels obtained from the encoder is used as an input value to calculate using a formula. From this input value, using the traveling locus formula derived to calculate the current position from the rotation speed of the left and right wheels in the current cycle from the self position determined in the previous cycle, the output values (X ₂ , Y ₂ , θ _2). )

Means 4.
A method using image processing in the first means. The input value is a two-dimensional image from the camera 4, and output values (X ₃ , Y ₃ , θ ₃ ) are obtained. This can be obtained by fixing the camera to a height of about 1 [m], photographing the front, and processing the photographed image.

Means 5.
A method of using a landmark reader detection signal in the first means. As shown in FIG. 6, the landmark 12 in the facility is installed at the boundary between the straight corridor and the intersection, and part of the boundary line of the environment map 5 is the same as the landmark 12 installed in the facility in advance. Writing in position.
When a landmark detection signal is acquired by a landmark reader, which is one of the sensors 1, while the autonomous mobile robot having a learning function is running, it is parallel to the wall 14 from the current position of the autonomous mobile robot at the time of detection. Then, the coordinates of the landmark 12 on the detected environment map 5 are corrected as shown in FIGS.
That is, the environment map 5 is corrected so that the center coordinates between the left and right wheels of the autonomous mobile robot model 11 are aligned with the landmark 12 as shown in FIG. Thereby, the calibration of the value of the sensor and the confirmation of the path can be performed.

Means 6.
The method using the landmark of the means 5 is not used at the same time as the means 2, means 3 and 4 and is used only when the landmark is detected.

Mean 7
In the case of normal traveling, the self position (X, Y) is determined by means 2, 3 and 4 by using the following conditional expression so as to reduce the error.
For the self-position (X, Y), first, L ₁ and L ₂ are obtained from the following expressions 1 and 2.

Here, if the values of L ₁ and L ₂ are both 20 cm or less, the three predictions are regarded as having little error, and X and Y are

Other than the above, if 50cm or less

Other than the above
X = X ₁
Y = Y ₁
And
As for the value of θ, Δθ ₁ , Δθ ₂ , and Δθ ₃ are first obtained from the equations (3), (4), and (5).

From this, if the values of Δθ ₁ , Δθ ₂ , Δθ ₃ are all within 5 [deg],

Other than the above
θ = θ ₃ (11)
And
The autonomous position of the autonomous mobile robot determined by the above operation is written in the environment map 5.
Each of these expressions is an expression that places a specific gravity on the self-position determined by a technique using a camera image.

Next, the recognition of the obstacle by the obstacle recognition function 3 and the writing to the environment map will be described by means 8 to 11.
Means 8.
The means 8 uses the value of the ultrasonic sensor that is one of the sensors 1 to recognize the traveling area and recognize the obstacle. Using the values obtained from the 10 sets of ultrasonic sensors, the current position of the autonomous mobile robot written on the environment map 5 and the same position as the 10 sets of ultrasonic sensors attached to the body of the autonomous mobile robot From this point, the directivity of the ultrasonic sensor is taken into consideration (spread of ± 23 [deg] with respect to the direction in which the sensor is directed), and written in the environment map 5 so as to draw an arc at an appropriate position.
For example, assuming that an obstacle 13a is present in front of an autonomous mobile robot having a learning function as shown in FIG. 8 and the value of x [cm] is obtained from an ultrasonic sensor on the left in front of the autonomous mobile robot, FIG. As shown in FIG. 9, the measured values up to the obstacle 13 are written in the environment map 5.
Furthermore, the distance, direction, and speed of the moving obstacle are recognized by the compound eye view of the ultrasonic sensor, the Doppler sensor, and the camera.

Means 9.
By processing the camera image, the distance from the autonomous mobile robot having a learning function to the obstacle and the width of the obstacle are calculated. In this method, the horizontal edge (the boundary between the floor and the obstacle) is detected from the images obtained from the camera 4 so far, and by using the processed image, the presence or absence of the obstacle, A method for obtaining a distance (reference document [1]) has been proposed. For example, if an obstacle 13a is found in the environment as shown in FIG. 8, the distance of the obtained obstacle from the current position written on the environment map 5 to the environment map 5 as shown in FIG. And write the obstacle 13 by using the width. At this time, the width of the obstacle 13 drawn on the environment map 5 is drawn in consideration of the directivity of the camera 4.

Means 10.
In the method of detecting an obstacle using a touch sensor, first, the detection signal of the micro switch attached to the front and rear total of eight places of the touch sensor is used to detect the front, left and right front, Recognize that there are obstacles in the lateral and backward directions. Then, the obstacle 13 is written as shown in FIG. 11 from the current position of the autonomous mobile robot having the learning function. FIG. 11 shows a state on the environment map when the micro switch in front of the autonomous mobile robot having a learning function detects an obstacle.
When the obstacle is detected by the touch sensor, the traveling is stopped urgently.

Means 11.
Here, in order to compensate for the unstable operation in the distance measurement of the ultrasonic sensor of the means 8, including the obstacle discovered by the camera information of the means 9, a five-stage risk is given as the existence probability (of the obstacle) The probability of existence when it does not exist is 1). On the environment map 5, when the risk reaches the fifth level (when the number indicating the risk reaches 5), it is recognized as an obstacle for the first time.
The risk classification is as follows.
(1) Add a degree of risk 3 to the value of a certain position of an object found by image processing.
(2) When it is confirmed by the image processing that there is no obstacle, -2 is added to the danger level of the position.
(3) Add 1 to the value of a certain position of the object found by the ultrasonic sensor.
(4) Add 5 to the value of a certain position of the object found by the touch sensor.
(5) The risk does not exceed 5 and does not fall below 0.

Next, input from the environment map to the SOM and output from the SOM will be described by means 12 to means 15.
Means 12.
In an autonomous mobile robot having a learning function, the SOM 8 trained in advance using the current position written in the environment map 5 and the speed and direction information obtained by learning the surrounding environment information and moving object information. Get the input value, calculate the best match unit, and determine the next action of the autonomous mobile robot.

Means 13.
As an input value of the SOM 8 learned in advance, the area of the autonomous mobile robot having a learning function as shown in FIG. 12 is divided into 11 regions, and the value of the distance information acquired by the ultrasonic sensor in each region is used as the SOM 8 It is applied as an input value to. The best match unit is calculated for each input value, and the output value is obtained.

Means 14.
Here, the method of obtaining the shortest distance value of each area in FIG. 12 of the means 13 is such that the distance from the vehicle body increases by 1 [cm] for each area from the environment map 5 as shown in FIG. Check if there are any walls 14 or obstacles. If there is an obstacle, the distance value to that point is acquired as the shortest distance value to the obstacle. FIG. 13 shows a method of acquiring values in the areas 5 and 8 in FIG.

Means 15.
The inference output value by the SOM 8 is output as a turning radius value indicating how much turning radius the autonomous mobile robot having a learning function turns straight or left and right.

Next, creation of an environment map as preparation for traveling will be described by means 16 and means 17.
Means 16.
The environment map 5 is created as preparation for traveling. The environment map 5 is created by measuring fixed obstacles such as corridor widths, pillars, and chaise longues in the facility in units of 1 [cm], and the travel method of the autonomous mobile robot having a learning function is switched to the created environment map 5 Write a boundary line.

Means 17.
The initial position and destination are input to the created environment map 5, the route to the target value is determined, and the target angle and travel so that an autonomous mobile robot having a learning function in each area of the environment map 5 is guided to the destination. Enter the direction.

Next, the update of the environment map by the traveling of the autonomous mobile robot will be described by means 18 to 21.
Means 18.
Start running of an autonomous mobile robot with a learning function. During traveling, first the values from the encoder and gyro sensor and the camera image from the camera are acquired. The position of the autonomous mobile robot having a learning function is recognized using the formula obtained by formulating the acquired value and the running characteristic.

Means 19.
Using the obtained position information of the autonomous mobile robot having the learning function, the information is written on the environment map 5.

Next, the creation of the SOM will be described by means 20.
Means 20.
When the current position of the autonomous mobile robot having a learning function is written in the environment map 5, an unknown obstacle is recognized using an ultrasonic sensor and a camera image.

When an obstacle such as an object placed on the road, a moving object, a person, or another wheelchair is encountered, the autonomous mobile robot that is traveling is measuring its position, speed, and direction in real time, It is necessary to control the traveling direction and speed.
In this system, as shown in FIG. 14, by using SOM8, the steering information (traveling speed and direction) is estimated from environmental information and obstacle movement information (position, speed and direction) around the autonomous mobile robot, and safely. The autonomous mobile robot will be moved to the destination.
The position, speed, and direction of the moving obstacle with respect to the traveling autonomous mobile robot are detected by sensors mounted on the autonomous mobile robot. The position of the obstacle is measured by an ultrasonic sensor, the speed is measured by a Doppler sensor, and the direction is measured by an azimuth angle sensor by binocular vision using two cameras.
Signals from these sensors are recognized as obstacles by the obstacle recognition function 3 and written to the environment map 5. And it inputs into SOM8 prepared beforehand from the environment map 5, and obtains the steering information at that time. However, the direction is determined while matching with the route information to the destination obtained by the route search.

For this reason, it is necessary to create the SOM 8 for learning the running state.
for that reason,
(A) Go to the site, move the autonomous mobile robot in a specific situation, and collect data of surrounding information and steering information.
(B) Collect data of surrounding information and steering information by computer simulation in the same situation as above.
(C) (a) (b) Match both.

The autonomous mobile robot is run at various sites, the surrounding information for the autonomous mobile robot that is running and the position, speed, and direction data of the obstacle are taken, and the appropriate travel direction and travel speed in that case are determined in advance. SOM8 like 15 needs to be learned and created. In order to learn SOM8, as shown in FIG. 17A, environmental information, obstacle movement information, destination information, and steering information around the autonomous mobile robot are used. The completed (learned) SOM 8 is mounted on an autonomous mobile robot, and sensor information (obstacle movement information), destination information, and surrounding information are stored in the SOM 8 according to a specific traveling state as shown in FIG. The steering information (azimuth and speed) of the autonomous mobile robot is obtained from the input, and output to the drive system 9 for use in traveling in the actual environment of the autonomous mobile robot.
The sensor group 1 is mounted on the autonomous mobile robot, the position, speed, and direction of the obstacle with respect to the traveling autonomous mobile robot are obtained, and the autonomous mobile robot model travels while writing on the environment map 5.
Safe driving is possible by matching the driving information in the real environment and the driving information on the environment map 5.
FIG. 16 shows an image diagram of the SOM in the traveling system of the autonomous mobile robot by learning.

Means 21.
If an unknown obstacle is found, the environment map 5 is updated while referring to the danger level of the means 11.

Next, the correction and addition of the SOM traveling rules by traveling will be described by means 22.
Means 22.
When the processing so far is completed, the latest environment map 5 during traveling of the autonomous mobile robot having a learning function is completed. Using this environment map 5, an input value for determining operation information by the SOM 8 is acquired, and the next operation is determined using the inference ability of the SOM 8.
The autonomous mobile robot having a learning function travels to the destination by repeating the acquisition of the sensor value, self-position recognition, obstacle recognition, and determination of the next operation using the inference ability of SOM8.

Next, the travel evaluation and acquisition of additional learning data are considered on the environment map. FIG. 18 shows driving evaluation and acquisition of additional learning data.
If a collision between the autonomous mobile robot and an obstacle is predicted during traveling, the traveling direction is adjusted. Thus, the obtained travel rule data is used for additional learning. First, environmental information is input from the SOM 8 prepared by learning, and travel prediction is performed based on the obtained travel rules. If the risk (proximity) is low, the vehicle travels as it is. When the degree of risk exceeds a certain value, environmental information is input and the driving rule is corrected based on the obtained driving rule and obstacle information. This correction is performed by fuzzy control using the travel rules and obstacle information obtained in advance as inputs (see FIG. 18).

Next, a flowchart of the travel evaluation and acquisition of additional learning data in FIG. 19 will be described with reference to FIG.
In step S11, environmental information is input to the learned SOM 8, and outputs of steering angle and speed are obtained. Next, in step S12, this output is given to the drive system 9 for trial. In step S13, the travel rule is evaluated (see FIG. 15). If it is dangerous, the travel rule is corrected in step S14, the process returns to step S12 and is retried, and additional learning to SOM 8 is performed in step S15.
If the evaluation is OK in step S13, the vehicle travels in step S16, and the process ends if the target point is reached. If it does not reach the target point, it repeats from step S1.
FIG. 20 shows the obstacle avoidance behavior in the actual environment and on the environment map by running according to FIG.

  According to the first embodiment, it is possible to provide an autonomous mobile robot having a learning function that can be used for the next travel along with the travel rules prepared and corrected in advance according to the conditions of the travel path. It is possible to construct an autonomous mobile robot having a learning function that can safely and reliably move to a destination in a factory, a welfare facility, a hospital lobby, or the like.

References [1] Yoshinobu Kanno, Hisashi Yamamoto, Makoto Oki, Masaaki Ohkita: Position recognition by matching camera data and current location data in indoor environment, Proceedings of the IEEJ National Conference, 3-043, p. 4 (2002).

  Control technology of autonomous mobile robot with learning function can be applied in various fields. For example, an autonomous mobile bed having a learning function in a hospital, an autonomous mobile wheelchair having a learning function in a welfare facility, a cleaning robot, a guide / guide robot, and the like.

It is a block diagram which shows the autonomous mobile robot with a learning function by Embodiment 1 of this invention. It is a figure which shows the autonomous mobile robot model which has a learning function on the environment map by Embodiment 1 of this invention. It is a flowchart which shows the flow of the process of 1 period of the autonomous mobile robot with a learning function by Embodiment 1 of this invention. It is a figure which shows the environment map in which the obstacle of the autonomous mobile robot with a learning function by Embodiment 1 of this invention was written. It is a figure which shows the determination of the position and attitude | position angle of the autonomous mobile robot model which has a learning function by the encoder and gyro sensor of the autonomous mobile robot which has a learning function by Embodiment 1 of this invention. It is a figure which shows the correction 1 of the self-position by the landmark of the autonomous mobile robot with a learning function by Embodiment 1 of this invention. It is a figure which shows the correction 2 of the self position by the landmark of the autonomous mobile robot with a learning function by Embodiment 1 of this invention. It is a figure which shows the environment where the obstruction which the autonomous mobile robot with a learning function by Embodiment 1 of this invention drive | works exists. It is a figure which shows the update of the environment map by the ultrasonic sensor of the autonomous mobile robot with a learning function by Embodiment 1 of this invention. It is a figure which shows the obstruction recognition by the camera image of the autonomous mobile robot with a learning function by Embodiment 1 of this invention. It is a figure which shows the update of the environment map with the touch sensor of the autonomous mobile robot with a learning function by Embodiment 1 of this invention. It is a figure which shows the division | segmentation of the vehicle body periphery area | region for the distance measurement of the autonomous mobile robot with a learning function by Embodiment 1 of this invention. It is a figure which shows the input value distance information acquisition example of the SOM inference in the area | region (5, 8) where the autonomous mobile robot with a learning function by Embodiment 1 of this invention was defined. It is an image figure which shows the driving | running | working system by learning of the autonomous mobile robot with a learning function by Embodiment 1 of this invention. It is a figure which shows the acquisition process of the knowledge by the additional learning of SOM of the autonomous mobile robot with the learning function by Embodiment 1 of this invention. It is an image figure which shows the traveling system SOM map of the autonomous mobile robot by learning of the autonomous mobile robot with the learning function by Embodiment 1 of this invention. It is a figure which shows the learning of SOM of the autonomous mobile robot with a learning function by Embodiment 1 of this invention, and utilization of the learned SOM. It is a figure which shows driving | running | working evaluation and acquisition of additional learning data of the autonomous mobile robot with a learning function by Embodiment 1 of this invention. It is a flowchart which shows driving | running | working evaluation and acquisition of additional learning data of the autonomous mobile robot with a learning function by Embodiment 1 of this invention. It is a figure which shows the obstacle avoidance action on the real environment and environment map of the autonomous mobile robot with a learning function by Embodiment 1 of this invention.

Explanation of symbols

1 Sensors 2 Self-position recognition function 3 Obstacle recognition function 4 Camera 5 Environment map 6 Learning function 7 Drive system input 8 SOM
9 Driving system 10 Touch panel 11 Autonomous mobile robot model 11a on the environment map Autonomous mobile robot 12 in the real environment 12 Landmark on the environmental map 13 Obstacle on the environmental map 13a Obstacle in the real environment 14 Wall 61 on the environmental map Travel rule addition function 62 Obstacle avoidance function

Claims (10)

  An environment map that stores environmental information of the road, sensors that detect the status of the road during driving, and the self-position is recognized from the conditions of the road detected by the sensors and input to the environment map. Self-position recognizing means for recognizing obstacles from the road conditions detected by the sensors and inputting the obstacle information to the environment map, and environment information of the environment map is inputted, and steering information is obtained. A learning means for outputting, and the learning means has a self-organizing map having learned the driving rules having the environment information as an input value, and outputs steering information according to the driving rules of the self-organizing map. An autonomous mobile robot with a learning function characterized by   2. The autonomous mobile robot having a learning function according to claim 1, wherein the learning means includes obstacle avoidance means for avoiding the obstacle when an obstacle is input to the environment map during traveling.   The said learning means has the driving rule addition means which adds the driving rule which avoids the said obstacle to the said self-organization map, when the obstacle is input into the said environment map during driving | running | working. An autonomous mobile robot having the learning function according to 1.   2. The autonomous movement having a learning function according to claim 1, wherein the self-organizing map has learned the driving rule by inputting environmental information of the environmental map and steering information corresponding thereto. robot.   The obstacle avoiding means evaluates whether or not the vehicle can run according to the driving rules of the self-organizing map based on the environmental information of the environmental map, and when the vehicle cannot run, corrects the driving rules to avoid the obstacle. An autonomous mobile robot having a learning function according to claim 2.   2. The autonomous mobile robot having a learning function according to claim 1, further comprising input means for inputting a destination to the environment map.   2. The autonomous mobile robot having a learning function according to claim 1, wherein the sensors include a plurality of sensors, and the self-position recognition means recognizes the self-position using a plurality of sensor information.   2. The autonomous mobile robot having a learning function according to claim 1, wherein the sensors include a plurality of sensors, and the obstacle recognition means recognizes an obstacle using a plurality of sensor information.   2. The autonomous mobile robot having a learning function according to claim 1, wherein a landmark is defined on the travel path, and the self-position recognition means confirms the self-position based on the landmark. The autonomous mobile robot having a learning function according to claim 1, wherein the sensors include a touch sensor, and the travel is stopped urgently according to information from the touch sensor.

Images