KR950005403B1 - Run control method of the automatic run floor cleaner using neural net work - Google Patents

Abstract

dividing a predetermined area near a cleaner into a plurality of cells and designing a local map illustrating an obstacle existance ratio at each cell; structuring a multi-layered neural circuit network which receives the local map and outputs the driving direction and speed of the cleaner; and driving the cleaner along a predetermined path using the neural circuit network.

Description

Driving Control Method of Automatic Travel Cleaner Using Neural Network

1 is a schematic plan view of an automatic traveling cleaner.

2 is a block diagram of a control device of an automatic traveling cleaner.

(A) and (b) of FIG. 3 are diagrams showing a distance relationship between a cleaner and a wall, and a speed and a driving direction relationship.

4 is a configuration diagram of the cleaning area of the automatic travel cleaner and the driving trajectory in this area.

(A) to (d) of Figure 5 is a driving trajectory configuration of the cleaner according to the shape of the funso area.

6 is a main flowchart of the traveling control method of the present invention.

7 is a flowchart of a neural network construction process in the traveling control method of the present invention.

8 is a view showing a local map of the automatic running cleaner in the present invention.

9 is a view showing the structure of a multilayer neural network constructed in the present invention.

10 is a flowchart of a cleaning path planning process in the traveling control method of the present invention.

11 is a diagram showing an example of the planning of the cleaning path in the traveling control method of the present invention.

12 is a diagram showing an example of following a cleaning path in the traveling control method of the present invention.

* Explanation of symbols for main parts of the drawings

1: cleaner body 2: bumper

3A-3K: Distance sensor 4: Caster

5A, 5B: rear wheel 6: inlet

7: dust collection filter 8: suction motor

9A, 9B: Reduction Gear 10A, 10B: Travel Motor

11A, 11B: Encoder Sensor 11E: Angular Speed Sensor

12 control device 13 power supply

14A, 14B: Brush 15: Micom

15A: I / O 15B: Central Processing Unit

15C: memory 16: encoder sensor circuit

17: angular velocity sensor circuit 18: motor drive circuit

19: distance sensor circuit

The present invention relates to an automatic driving cleaner for driving an area to be cleaned by the cleaner to recognize the cleaning area by itself, and by performing a redirection of the planned route by performing a cleaning route to be driven in the recognized cleaning area by itself. To be. In particular, a local map is created by expressing the probability of obstacles in the vicinity of the cleaner as two-dimensional arrangement information of the obstacles, and using the neural network structure as an input and outputting the driving speed and driving direction. The present invention relates to a traveling control method of an automatic traveling cleaner using a neural network for performing a traveling control of a cleaner.

Referring to FIG. 1, a configuration of a general automatic traveling cleaner includes a bumper 2 installed at all sides of the cleaner body 1, and the bumper 2 transmits and receives an ultrasonic wave to identify an obstacle and a distance from the obstacle. Distance sensors (3A-3K) for detecting the corresponding to each direction of the cleaner is installed, the front wheel (caster) (4) is installed on the entire bottom of the cleaner body (1) and the rear left and right rear wheels (5A, 5B) In the center of the bottom of the main body 1, there is a suction inlet 6 for sucking dust or dirt, and a dust collecting chamber having a dust collecting filter 7 on the upper portion thereof, and through the suction hole 6 A suction motor 8, which is a power source to suck in dust and dirt, is installed in the main body 1.

On the other hand, the left and right rear wheels (5A, 5B) is connected to the reduction gears (9A, 9B) consisting of reduction gears and the wheel driving motor (10A, 10B) for driving the wheels through the reduction gear, and the drive of the wheel driving motor The encoder sensors 11A and 11B which detect the travel distance by the said apparatus are provided, and the angular velocity sensor 11C which detects the rotational angular velocity of a cleaner is provided.

And, the control device 12 for controlling the automatic driving cleaning through the decoction of the detection signal of each of the sensors to drive each motor is installed, the suction motor 8, the driving motor (10A, 10B), the control device ( A power supply device 13 for supplying power to 12 is installed, and brushes 14A and 14B are provided to shake off dirt to both front sides of the cleaner body 1.

On the other hand, the configuration of the control device 12 is shown in Figure 2, the left and right sides detected by the microcomputer 15 and the encoder sensors (11A, 11B) for controlling each part of the circuit through arithmetic processing and input and output of various information The rotation angle information of the cleaner is calculated by integrating the encoder sensor circuit 16 which converts the latest angle information of the driving motor into digital information and supplies it to the microcomputer 15 and the cleaner rotation angular velocity information detected by the angular velocity sensor 11C. The calculated information is converted into digital information and supplied to the microcomputer 15, and the encoder sensor circuit 16, together with the angular velocity sensor circuit 17 which forms a cleaner position identification means, is controlled under the control of the microcomputer 15. Motor drive circuit 18 for controlling the driving of motors 10A, 10B and suction motor 8, contact sensor 3L (3M) for detecting whether the bumper outside the cleaner contacts an obstacle, and the distance sensor Distance information detected with (3A-3K) A distance sensor circuit 19 for converting a digital signal to the microcomputer 15, a remote controller transmitter 20 for remote control of a vacuum cleaner, and a remote controller receiver 21 for receiving and supplying the remote control signal to the microcomputer 15; The microcomputer 15 includes an input / output unit 15A for input / output processing of all data, a central processing unit 15B for performing calculation and driving control processing of input / output data, and the various data and control programs. A storage unit 15C, an interrupt controller 15D for controlling interrupts of respective circuits, and a clock generator 15E for supplying a clock signal to the central processing unit 15B.

The cleaning operation of the conventional automatic traveling cleaner configured as described above will be described with reference to FIGS. 1 to 4.

First, when a user applies a cleaner power and inputs a cleaner driving command by operating a remote controller transmitter 20, the migration command is output as a remote control signal and received by the remote receiver 21, and the received remote control signal is input / output unit. It is input to the central processing unit 15B of the microcomputer through 15A.

The central processing unit 15B decodes the input remote control signal and performs the following driving control of the cleaner.

That is, since power is applied to the control device 12 so that the control device 12 operates, the microcomputer 15 of the cleaner starts to operate according to the clock supplied from the clock generator 15E, and then through the input / output unit 15A. The motor driving circuit 18 drives the suction motor 8 by outputting the suction motor driving signal. As the suction motor 8 is driven, the dust or dirt on the floor is sucked through the suction port 6 while sweeping the dust or dirt on the floor with the brushes 14A and 14B, and the dust is collected by the dust collecting filter 7. The position data P (coordinate data x, y and direction data e) of the vacuum cleaner stored in the 15C is initialized, and the driving motors 10A, 10B are driven through the motor driving circuit 18 to reduce the speed reduction device 9A. Drive the rear wheels 5A, 5B (traveling wheels) at a speed reduced to 9B).

Accordingly, the cleaner main body 1 starts to move forward, and as the cleaner moves forward, the distance signal traveled by the cleaner through the encoder sensors 11A and 11B is detected, and the digital distance data ΔD is transmitted through the encoder sensor circuit 16. (See also 113, detecting the distance to the wall or furniture (S0-S6), etc.) is stored in the storage unit 15C of the microcomputer 15, and also the rotation angle (turn direction) signal of the cleaner through the angular velocity sensor 11C Is detected and the digital rotation angle data Δθ is stored in the storage unit 15C of the microcomputer 15 through the acceleration sensor circuit 17.

The central processing unit 15B of the microcomputer 15 calculates the position data P of the vacuum cleaner from the distance data ΔD and the rotation angle data Δθ, while using the distance sensor 3A-3K for walls or furniture. The distance sensor circuit 19 receives a signal detecting a distance between the obstacle and the like, and calculates the position data X of the wall from the distance data of the obstacle and the position data P of the cleaner. 20C, and based on distance data with obstacles such as walls or furniture in each direction of the cleaner detected by the distance sensors 3A-3K, the cleaner is kept in parallel with the wall surface while maintaining a constant distance. Parallel driving control is performed by controlling the driving motors 10A and 10B through the motor driving circuit 18 to travel.

The control in which the cleaner travels in parallel with the wall is performed through driving control of the driving motor in the microcomputer 15 using a fuzzy rule.

That is, as shown in (a) of FIG. 3, a distance S0-S6 from obstacles such as walls and furniture is detected from the distance sensors 3A-3K installed in front and side of the cleaner body, and the detected distance is measured. The purge input is made by defining the traveling speed v and the traveling direction w of the cleaner as the purge output, and defining the sebum rule and the purge value so that the cleaner runs along the wall.

For example, the distance S0 (S1) detected by the distance sensor 3A (3B) provided on the front left side of the cleaner body is very small (step 1), small (step 2), slightly smaller (3). Step), appropriate (step 4). Slightly large (5 steps), large (6 steps), very large (7 steps) to have a seven-step purge value, and the remaining distance (3C-3K) detected distance (S2-S6) in two steps ( Small and large) to have a fuzzy value to determine sebum input.

If the running speed (v) and the driving direction (w) of the vacuum cleaner are W = 0, straight ahead, and the positive value is larger positively, the sharper right turn, the larger negative value is sharply left, and so on. The purifier runs in parallel with the wall by defining the purge and the purge value to accurately run along the wall using the purge output having the purge value of the step.

That is, as shown in (a) of FIG. 4, the cleaner first starts traveling from the point A and proceeds toward the wall, and when the point B is reached, the driving track PATH1 is drawn along the wall while keeping parallel to the wall. The vehicle stores the wall position data (X) until it reaches point C.

As such, when the cleaner completes the driving along the outer path PATH1 of the area to be cleaned, the central processing unit 15B of the microcomputer 15 stores the position data X of the wall stored in the above process as shown in FIG. And the inside of the polygonal area AREA1 formed by the connected line segments as the overall area to be cleaned, and the remaining area AREA2 inside the width of the cleaner body 1 from the entire area AREA1 as the width to be cleaned. To be recognized.

Subsequently, the central processing unit 15B of the microcomputer 15 draws a parallel line parallel to the wall surface of the remaining cleaning area AREA2, as shown in FIG. A new path (PATH2) alternately connected (zigzag) at both ends of the LINE) is constructed as (D) of FIG. 4 to plan this path (PATH2) as a cleaning path.

When the planning of the cleaning path PATH2 is completed, the control device 12 drives the traveling motors 10A and 10B through the motor drive path 18 to reduce the speed of the rear wheels 5A, 9A and 9B. 5B) is driven.

Accordingly, the cleaner body 1 starts to move forward, and while the cleaner moves forward, the distance signal traveled by the cleaner through the encoder sensors 11A and 11B is detected, and the digital distance data ΔD is transmitted through the encoder sensor circuit 16. Is stored in the storage unit 15C of the microcomputer 15, and the rotation angle (direction change) signal of the cleaner is detected through the angular velocity sensor 11C, and the digital rotation angle data (Δ) is transmitted through the angular velocity sensor circuit 17. is stored in the storage unit 15C of the microcomputer 15.

The central processing unit 15B of the microcomputer 15 calculates the position data P of the cleaner from the distance data AD and the rotation angle data Δθ, and based on the position data P, the planned cleaning is performed. Calculates the distance and direction data from which the cleaner is displaced in the path PATH2, and uses the calculated distance and direction data to drive the wheels 5A and 5B so as to follow the planned cleaning path PATH2. When the final path following of PATH2 is completed, the automatic driving cleaning is terminated by stopping the suction motor 8 and the driving motors 10A and 10B.

At this time, the cleaner follows the planned cleaning path PATH2 so that the cleaner does not deviate from the path. The control method is as shown in (b) of FIG. 3, and the cleaner position identification circuits 16 and 17 determine the current position of the cleaner. It determines and calculates the angle difference dA between the distance d1 from which the cleaner deviated from the planned path PATH2 FH, and the planned path PATH2.

In the path, the purge rule and the purge value are defined to follow the planned path PATH2 by the cleaner using the calculated distance difference and the angle difference as the purge input, and the cleaner running speed v and the direction w as the purge output. It is to control not to break out.

As described above, the cleaner travels around the outside of the cleaning area surrounded by obstacles such as walls or furniture, recognizes the area to be cleaned by itself, and performs the cleaning while automatically driving the recognized cleaning area. In the current position of the vacuum cleaner does not deviate from the planned path and can be guaranteed high cleaning efficiency and clean state of cleaning only when driving along the path accurately, for this purpose, Precise calculation and the accuracy of cleaner location recognition should be premised.

However, the conventional automatic traveling cleaner which is subjected to the position determination and the driving control as described above has the following problems.

That is, due to the limited directivity of the distance sensors 3A-3K, the obstacle A located in the blind spot of the sensing area as shown in (a) of FIG. 3 cannot be detected, and only the distance to the currently detected obstacle is detected. Because it is used as the sensing information, even if the detected obstacles (B) (C) as in (a) of FIG. 3 do not recognize the characteristics such as the size and shape of the obstacles (B) (C), insufficient sensing information. There is a problem that the error of the determination of the traveling speed and the driving direction due to the acquisition of.

In addition, since the automatic travel cleaner draws a horizontal line only in the direction where it is located after circumnavigating along the wall as shown in (a) or (c) of FIG. The path following cannot be planned as the path that is not parallel to the wall as shown in (D) for the cleaning area that cannot be planned with the same shortest cleaning path and the straight wall which is parallel or perpendicular to the direction of the path as shown in (C) is not planned. There is a problem that the position measurement error of the cleaner generated in the liver cannot be corrected through the distance sensor as shown in FIG. 3B.

The present invention creates a local map expressing the existence probability of obstacles around the cleaner by two-dimensional obstacle arrangement information, constructs a multi-neural neural network using the prepared local map, and uses the constructed neural network to construct a wall Run in parallel to recognize the area to be cleaned. Necessary sufficient information about the environment around the cleaner by planning the shortest optimal cleaning path from the recognized coordinates of the cleaning area, and controlling the cleaner to follow the planned route using the neural network input as the distance and direction of the cleaner. By using this information, the neural network is used to overcome the cleaner driving error, to precisely measure the position of the cleaner, and to correct the measured position error by using the distance sensing information during driving. An object of the present invention is to provide a driving control method for an automatic traveling cleaner, and the present invention will be described in detail below with reference to the accompanying drawings.

The automatic performing cleaner which performs the driving control method of the automatic traveling cleaner using the neural network of the present invention for achieving the above object has the same configuration as in FIGS. 1 and 2 and has a distance detected by each sensor. According to the present invention performs the driving control process as follows, the same components as in the prior art with reference to FIGS. 1 to 4 will be omitted overlapping, 6 to 12 The driving control method of the present invention will be described with reference to FIG.

first. Referring to FIG. 6, the present invention divides a predetermined area around a cleaner into a plurality of cells, prepares a local map representing an obstacle existence probability for each cell, inputs the local map, and determines the driving direction of the cleaner. Building a multi-layer neural network with speed as the output and using the constructed neural network to run in parallel with the wall, and cleaning the area after traveling around the outside of the area to be cleaned along the wall. A cleaning path planning step for recognizing a cleaning area for recognizing the cleaning area and a cleaning path for receiving the shortest cleaning path and position correction using the outer position coordinates of the cleaning area recognized in the step; and from the cleaning path planned in the step The distance and direction from which the cleaner is displaced, and the distance and direction are input to the fuzzy control. A path tracking step of following the planned cleaning path by adding the driving direction and distance of the cleaner as an output value as an input of the constructed multi-layer neural network.

The traveling control method of the automatic traveling cleaner using the neural network of the present invention made as described above will be described with reference to FIGS. 7 to 12 in each step.

7 shows a step in which the cleaner runs parallel to the wall along the outer wall of the cleaning station.

That is, when the power is supplied to the cleaner and the user starts driving by using the remote controller transmitter 20, the cleaner moves by the microcomputer 15 control operation of the controller 12 as described above. Create a local map that expresses the probability of the presence of obstacles around the cleaner.

The local map divides the divided space into a cell structure by dividing the surrounding space of a square of a certain size around the current position of the cleaner, as in the case of a checkerboard. In addition, each cell displays the probability that an obstacle such as a wall or a furniture exists (the cell marked with a black dot in the drawing is a cell having a relatively high probability value), and this is expressed as two-dimensional layout information of the obstacle in the space around the cleaner. Defined as (LOCAL MAP).

In other words, in condition probability theory, the probability of occupied area and non-occupied area by obstacle is calculated and displayed by using the occupation probability of a cell, the information so far, and newly acquired information, and the cells belonging to the occupation area for each distance sensor are displayed. The probability is updated and expressed as two-dimensional arrangement information of obstacles in the space around the cleaner.

The local map is determined that the center of the automatic driving station coincides with the center of the local map and that the center moves as the automatic driving cleaner moves, and the local map does not rotate about the reference coordinate system but only translates. Should be.

Such a local map is created for each driving cycle of the cleaner, and FIG. 8B shows the overall state of the created local map.

A local map is created as described above, and the local map is divided into n regions (In, 2n, 3n ..., nn) centered on the cleaner as shown in FIG. N neural networks (N / N1, N / N2, N / N3, ... N / Nn) that output the running speed (v) and the quantity (w) of the vacuum cleaner are constructed.

The neural network is a straight forward when the running speed (v) and the driving direction (w) of the vacuum cleaner is w = 0, and a rapid right turn is made when the absolute value of the positive value is larger, and a sharp left turn is made when the absolute value of the negative value is large. Construct with neural network as output.

Subsequently, the final neural network (N / N) is constructed by using the outputs of the n neural networks as the inputs and the final outputs of the traveling speed (v) and the direction (w) of the cleaner. do.

9 shows the structure of the neural network constructed as described above, and the portion indicated by the dotted line in this neural network becomes the network structure to be added upon following the cleaning path to be described later.

The neural network learns that the cleaner can travel correctly along the wall according to the distribution of probability values of each divided cell, so that the cleaner runs in parallel with the wall along the wall forming the outer edge of the area to be cleaned.

In this way, when the cleaner completes driving along the outer path PATH1 of the area to be cleaned, the central processing unit 15B of the microcomputer 15 connects the position data X of the wall stored in the above process and connects to the connected line segments. The inside of the polygonal area AREA1 is recognized as the area to be cleaned as a whole, and the area AREA2 inside the width of the cleaner body 1 from the entire cleaning area AREA1 is recognized as the remaining cleaning area.

When the recognition of the cleaning area is completed, the optimum path for cleaning will be the shortest path in the recognized cleaning area, and the positioning error of the cleaner can be corrected by the distance information detected by the distance sensor 3A-3K. .

10 shows the planning stage of the cleaning path described above.

That is, from the coordinates X of the cleaning area recognized in the cleaning area recognition step, the midpoint coordinates P _{o ...} P _n of two adjacent position coordinates are obtained as shown in FIG. After connecting, select the three consecutive points (P ₁ , P _{1 + 1} , P _{1 + 2} ) and the slope of the straight line connecting two adjacent points [(P ₁ , P _{1 + 1} ) (P _{1+ 1} , P _{1 + 2} )].

The degree of similarity is judged by comparing the inclinations of two adjacent straight lines obtained in the above step, and when the determination result has a similar inclination more than a predetermined value, these three points are stored as the same straight line group SC, and the inclination of the two straight lines In the case of dissimilarity, a series of processes for storing them as a new straight line group SC is performed for all midpoint coordinates (n coordinates) to find and store the same straight line group having similar slopes.

Subsequently, a representative straight line (effective linear line) EL that best represents the coordinate value among points in the same straight line group SC in the straight line group SC is obtained by a least square method, and the effective straight line EL is obtained. An imaginary line segment is drawn at regular intervals in the same direction as the road, and the number NP (number of paths) of the line segment is obtained.

And, the number of straight lines NSP (parallel or perpendicular to the effective straight line EL) is compared by comparing the effective straight lines EL obtained in the above process and the straight lines CL connecting two consecutive midpoint coordinates. Number).

The number NSP of the straight lines represents the extent to which the position error of the cleaner can be corrected, and the larger the number, the higher the possibility of correcting the position error.

Subsequently, the selectivity WEL of the effective straight line is calculated by weighting the number NP of the path and the number NSP of the slope with respect to the effective straight line EL.

In other words, the number of paths (NP) is multiplied by a negative weight, and the number of paths (NSP) is multiplied by a positive weight (+), and these values are summed together to select the effective linear selectivity (WEL). This selectivity WE has a larger value as the number NP of paths is smaller and the number NSP of slopes indicating the degree of correction is large.

The above-described process of calculating the selectivity of the effective straight line is repeated for all the effective straight lines EL, and the effective straight line EL having the largest value is selected from the selected selectivity of the effective straight lines WEL. Plan the path consisting of these segments as the optimal cleaning path by drawing the segments with a constant width parallel to the selected effective straight line (EL) and connecting the starting wave end points of these segments with each other (zigzag). do.

As described above, when the cleaning path is planned, cleaning is performed by estimating the planned cleaning path. The cleaning path following step avoids the obstacle even when it encounters the obstacle while following the path as shown in FIG. In order to continue following the route, a local map around the cleaner is prepared as in (b) of FIG. 12 in the same manner as the driving control in the cleaning area driving step, and the probability value of each cell of the created local map is inputted. A multi-layer neural network is followed to follow the planned cleaning path.

That is, the purge input sets the angle difference dA (direction difference value) between the distance D1 separated from the cleaning path that the cleaner is associated with the planned cleaning path, and the driving direction W and the traveling speed v of the cleaner. The traveling speed v and the traveling direction w are obtained by applying a fuzzy rule outputting?, And the traveling speed v and the direction w, which are the fuzzy output values, are added to the final neural network N / N.

9 shows the structure of the multilayer neural network in which the traveling speed v and the direction w, which are fuzzy control outputs, are added as inputs to the final network N / N.

In this neural network, the neural network learning is trained to follow the planned path according to the obstacle pattern of each area and the driving speed (vt) and the direction (wt) which are the fuzzy control output when following the planned cleaning path, so that the cleaner cleans the planned cleaning path. Make sure you follow the path correctly without leaving.

When following the cleaning path is completed, since the automatic driving cleaning of the recognized cleaning area is completed, the control device 12 stops the driving of the suction motor and the driving motor through the motor driving circuit and ends the cleaning.

As described above, according to the present invention, the cleaner recognizes the environmental information related to the presence of obstacles around the user using a local map, and accurately recognizes the optimal cleaning path through the multilayer neural network. By following the learning and performing the automatic driving control, it is possible to reduce the error of the cleaner position calculation, to precisely calculate the position by precisely following the cleaning path, and to ensure the clean cleaning result of the area to be cleaned. .

Claims (7)

Local map creation step of dividing a certain area around the cleaner into a plurality of cells (CELL) and creating a local map expressing the probability of obstacle existence for each cell, and using the local map as an input and outputs the driving direction and speed of the cleaner A neural network construction step of constructing a multi-layer neural network, and a traveling control method of the automatic running cleaner using a neural network consisting of a cleaner driving step of driving a cleaner along a predetermined path using the constructed neural network. The method of claim 1, wherein the neural network construction step comprises: dividing the created local map into n areas around the cleaner, and inputting the probability values of the cells of each area divided in the step as inputs. v) constructing n neural networks with outputs of v) and direction (w); and inputting the output of the neural network constructed in the above step, and outputting the running speed (v) and the direction (W) of the cleaner. Running control method of the automatic running cleaner using a neural network consisting of a step of constructing a neural network of the final stage to form a neural network of a multi-layer structure. The vacuum cleaner of claim 1 or 2, wherein the cleaner driving step comprises inputting a local map, which is the surrounding obstacle arrangement information of the cleaner, and outputting a driving direction (w) and a speed (v), thereby cleaning the neural network. A traveling control method of an automatic traveling cleaner using a neural network that performs a parallel driving with a wall when traveling around a cleaning area by learning to travel in parallel along a wall. The traveling speed of claim 1, wherein the cleaner driving step comprises: obtaining a distance and a direction in which the cleaner is separated from a planned path, and a purge control output using the distance and the direction obtained in the step as a purge control input. And driving the driving speed and the driving direction obtained in the step as the input of the multilayer neural network, and learning the driving direction so as to control the cleaner to follow the planned route. Control method. A cleaning area driving step of driving the cleaner along the outside of the cleaning area to obtain position coordinates at each point; and a cleaning area recognition step of circumscribing the outside of the area to be cleaned along the wall in the step and recognizing the area as the area to be cleaned; A cleaning path planning step of planning a cleaning path capable of receiving the shortest cleaning path and position correction using the outer position coordinates of the cleaning area recognized in the step; and a distance from which the cleaner is separated from the cleaning path planned in the step Driving of an automatic traveling cleaner using a neural network consisting of a path following step of following the planned cleaning path by using the driving direction and distance of the cleaner which is an output value of the purge control obtained by obtaining a direction and using this distance and the direction as an input of the fuzzy control. Control method. The method of claim 5, wherein the cleaning path planning step comprises: obtaining a midpoint coordinate of an adjacent point from the recognized position coordinates of the cleaning area, connecting the midpoints obtained in the step with an adjacent point, and connecting the connected lines. Obtaining a slope of the step; and storing the slope obtained in the step in the same straight line group having similar slopes by retrieving the similar information; and a straight line representing each coordinate value in the straight line group obtained in the step; EL), drawing an imaginary line at regular intervals in parallel with the effective straight line EL obtained in the above step, and calculating the number as the number of paths NP, and the effective straight line EL obtained in the step Comparing the straight lines connecting adjacent midpoints to obtain the number of straight or parallel straight lines as the number of slopes (NSP), and the number of paths obtained in the step ( Selecting an effective straight line based on NP) and the number of slopes (NSP), and drawing a line segment with a constant width in parallel with the selected effective straight line, and connecting the start and end points of the line segment to each other to determine a cleaning path. Running control method of the automatic running cleaner using the neural network. The method of claim 6. In selecting the effective straight line, a negative weight is given to the number NP of the obtained paths, and a positive weight is given to the number of slopes NSP, which is the sum of these two values. Running control method of the automatic running cleaner using a neural network comprising the step of selecting an effective straight line having a value.