CREATING A MAP OF THE SURROUNDING AREA

Description

[0001] The invention relates to the creation of an environmental map for a mobile household appliance. In particular, the invention relates to the removal of erroneous measurements.

[0002] A floor cleaning robot is designed for use in a household, such as a house or apartment. The robot can move independently and has sensors to scan its surroundings. One sensor can detect a collision with an obstacle, while another enables non-contact scanning of the environment.

[0003] To enable a planned approach to floor cleaning, the robot needs an environmental map that indicates which areas of the floor are accessible to it and which are not. This environmental map can be created by the robot driving through the apartment or house, scanning its surroundings without physical contact. Using a SLAM (Simultaneous Localization and Mapping) method, the robot's position can be determined and obstacles detected by the robot can be mapped simultaneously.

[0004] In particular, non-contact sensors such as radar or LiDAR sensors can produce so-called phantom points in the robot's vicinity, where an obstacle appears to be present, but where there is actually a clear area. Such measurement noise can be caused especially by bright light or in the vicinity of an obstacle made of a transparent or poorly reflective material. Error measurements are frequently found close to the edge of an actual obstacle. This effect is also known as the "mixed pixel effect."

[0005] The publications DE 10 2018 217000 A1, US 2018 / 253107 A, EP 3 514 493 A1, US2019 / 0035090 A1 and BELTER DOMINIK ET AL: "Keyframe-based Local Normal Distribution Transform Occupancy Maps for Environment Mapping", 2018 IEEE 23FiD INTERNATIONAL CONFERENCE ON EMERGING TECHNOLOGIES AND FACTORY AUTOMATION (ETFA), IEEE, Vol. 1, September 4, 2018 (2018-09-04), pages 706-712, XP033426003, DOI: 10.1109 / ETFA.2018.8502517, describe methods for creating an environment map in which certain measurement data are discarded taking predefined assumptions into account.

[0006] One of the problems underlying the present invention is to provide an improved technique for removing phantom points from measurement data of a non-contact sensor, in order to allow improved mapping and / or positioning of a household appliance. The invention solves this problem by means of the subject matter of the independent claims. Dependent claims describe preferred embodiments.

[0007] According to a first aspect of the present invention, a method for creating an environment map for a mobile household appliance comprises the steps of transmitting a measurement signal into the environment of the household appliance and receiving reflections of the measurement signal; determining points in the environment to which a reflection of the measurement signal is associated; determining a local distribution of the points in the environment; discarding the points if the distribution is less than a predetermined straight line; and creating the environment map based on the remaining points, wherein the environment is divided into tiles of a predetermined size; and wherein the points of a tile are discarded if the local distribution of the points of the tile is less than a predetermined straight line.

[0008] It was recognized that for the appliance, points forming a straight line are of particular interest. This straight line can encompass the boundary of an obstacle, such as a wall or a piece of furniture. By discarding points that do not conform to a straight line as defined, phantom measurements can be efficiently removed from the measurement data. Appliance positioning and / or environmental mapping can only be improved by relying on accurate measurements, thus enhancing the accuracy and reliability of the identification.

[0009] The environment is divided into tiles of a predetermined size, with points within a tile being discarded if their spatial distribution deviates from a straight line below the predetermined size. A tile might, for example, be approximately 100 x 100 cm; however, other sizes are also possible. The tiles can be adjacent on all sides, allowing the environment to be completely covered by tiles. Adjacent tiles can also overlap, for example, by 50%. In one embodiment, the method includes the step of discarding points whose distance from an average point location exceeds a predetermined value.

[0010] This allows points that form a dense structure to be retained. Individual points that are far from most other points, however, can be discarded. By appropriately choosing the predetermined value, phantom points that are close to an actual obstacle but not part of it can be effectively removed.

[0011] The procedure first checks whether points from entire tiles should be discarded, and only then whether individual points should be discarded. In one variation, the test criteria can be applied in reverse order.

[0012] The environment is divided into tiles of a predetermined size, as described above. A point assigned to a tile can be discarded if its distance to an average location of the points on that tile exceeds the predetermined value. The average location always lies on the tile and represents the statistical mean of the locations of the points encompassed by the tile. In a particularly preferred embodiment of the invention described herein, a normal distribution transformation (NDT) is determined for a predetermined number of points. The NDT provides an average location and the variance of the points. The first test, which can lead to the discarding of all investigated points, and the second test, which can lead to the discarding of individual points, can advantageously be easily applied with respect to the NDT. The NDT can be performed with manageable complexity.

[0013] Effort can be determined for a large number of points. It is particularly preferred to determine NDTs for each point of a tile, as described herein.

[0014] The points are typically defined in two dimensions, which usually lie in a horizontal plane with respect to the appliance. The variance of a set of points then comprises two eigenvectors. In general, the variance includes as many eigenvectors as the included points have dimensions. An eigenvalue can be determined for each eigenvector of the variance. The points under consideration can be discarded if the ratio between the larger and smaller eigenvalues ​​deviates from one by more than a predetermined amount.

[0015] The points under consideration are organized much more strongly in the manner of a straight line when the ratio is significantly different from one. Typically, this ratio is several hundred, several thousand, or even more for a straight line in the tile. If the eigenvalues ​​are very similar, the ratio can be close to one. If, for example, there is no obstacle at all in the area under consideration, the ratio can drop below 10, below 2, or even below 1.1.

[0016] In another embodiment, the points of a tile can be discarded if one of the eigenvalues ​​exceeds or falls below a predetermined value. For example, it can be determined whether the smaller of the eigenvalues ​​falls below a predetermined minimum value or exceeds a predetermined maximum value. Similarly, it can also be checked whether the larger of the eigenvalues ​​falls below a predetermined minimum value or exceeds a predetermined maximum value. This allows for the advantageous evaluation of further indications of a highly noisy measurement. This test can be performed, in particular, in addition to the consideration of the ratio of the eigenvalues ​​described above.

[0017] The distance can be determined based on the Mahalanobis distance of the point to the average point. This distance is defined between points in a multidimensional vector space and intuitively gives the distance between the points in standard deviations. The Mahalanobis distance can be easily determined and is meaningful in the vector space under consideration.

[0018] According to a second aspect of the invention, a first device for determining an environmental map for a mobile household appliance comprises a transmitting device for emitting a measurement signal into the environment of the household appliance; a receiving device for receiving reflections of the measurement signal; and a processing device. The processing device is configured to determine points in the environment to which a reflection of the measurement signal is assigned; to determine a spatial distribution of the points in the environment; to discard the points if the distribution deviates from a straight line to a lesser extent than predetermined; to determine the environmental map based on the remaining points; and to divide the environment into tiles of a predetermined size, wherein the processing device is configured to discard the points of a tile if the spatial distribution of the points of the tile deviates from a straight line to a lesser extent than predetermined.

[0019] In one embodiment, the device comprises the processing unit which is configured to reject points whose distances to an average location of the points exceed a predetermined value.

[0020] The two embodiments of the device can also be integrated together, such that the processing unit is configured to discard the points under consideration if the distribution is less than predetermined and does not correspond to a straight line; and to discard individual points whose distances to an average location of the points exceed a predetermined value.

[0021] The processing unit of the device may be configured to execute all or part of one of the methods described herein. For this purpose, the processing unit may include a programmable microcomputer or microcontroller, and the method may be in the form of a computer program product with program code. The computer program product may also be stored on a computer-readable data carrier. Features or advantages of the method may be transferred to the device and vice versa.

[0022] In a preferred embodiment, the device includes the transmitting and receiving units of a LiDAR sensor. The LiDAR sensor can measure, in particular, in the horizontal direction and have a detection angle of up to 360°. In a first variant, it operates according to the time-of-flight (TOF) principle by determining the distance to a point on an obstacle based on the time between the emission of light and the arrival of the light reflected from the obstacle, and multiplying this time by the speed of light. Alternatively, the sensor can use a correlation of a modulated transmitted signal with a received signal to determine the distance to a reflection point. Multiple modulation frequencies can be used to increase the unambiguous measurement range.In a second variant, the sensor operates on the triangulation principle, being configured to determine the direction of a point on an obstacle relative to the point's directional angle from two distanced locations. Both variants can be cost-effective and provide sufficiently accurate scanning of the area around a typical household appliance.

[0023] According to another aspect of the invention, the household appliance comprises a floor cleaning robot with the device described herein. The floor cleaning robot can be configured to clean the floor area of ​​a household, for example, by means of a pneumatic suction device or a wet cleaning device. In a further embodiment, the household appliance can comprise virtually any type of mobile, floor-based robot, for example, a robotic lawnmower or a robotic security guard.

[0024] The invention will now be described in more detail with reference to the accompanying figures, in which: Figure 1: A mobile household appliance; Figure 2: A flowchart of a procedure; Figure 3: Exemplary variances; and Figure 4: Exemplary samples of a household represents.

[0025] Figure 1 Figure 1 shows a mobile household appliance 100 in an environment 105. The household appliance 100 can, in particular, include a floor-cleaning machine, for example, a robotic vacuum cleaner. The household appliance 100 is designed to be used in a household where it can move autonomously across a floor area. The environment 105 can comprise a section of the household or the entire household. The environment 105 can be subdivided into individual sections, which are divided into Figure 1The adjacent square tiles 110 are indicated by dashed lines. The tiles 110 are preferably arranged in a checkerboard pattern as shown; however, other layouts are also possible. For clarity, the overlap of adjacent tiles 110 is not shown, but this can be used in practice and can amount to approximately 50% of the area of ​​each tile 110.

[0026] The household appliance 100 typically comprises one or more components not shown, such as a power supply, a drive motor, a wireless communication device, a floor surface treatment device, or a collision sensor. The household appliance 100 also comprises a device 112, which includes a processing unit 120 connected to a transmitter 125 and a receiver 130. The transmitter 125 and the receiver 130 may be comprised of a common sensor 135, which may be configured, in particular, as a LiDAR sensor. The transmitter 125 is configured to emit a measurement signal, in particular light, into the environment 105, and the receiver 130 is configured to receive the signal reflected from the obstacle 115. Optionally, the device 112 further comprises a memory 140 for storing an environmental map of the environment 105.The household appliance 100 can navigate within its environment 105 based on the surrounding map and, if necessary, move autonomously.

[0027] The processing unit 120 is configured to determine the environmental map based on points 145 detected by the sensor 135 in the environment 105. A point 145 is located at a point in the environment 105, typically at the edge of an obstacle 115. At point 145, the measurement signal from the transmitting unit 125 is reflected such that it reaches the receiving unit 130. The distance and direction of point 145 relative to the household appliance 100 can be determined based on the transmitted and received signals, for example, by triangulation or based on the transmission direction and the time between transmission and reception of the measurement signal. In this way, the location of point 145 relative to the household appliance 100 can be determined. If the position of the household appliance 100 in the environment 105 is known, the location of point 145 in the environment 105 can be determined.

[0028] Measurement noise or interference can cause the sensor 135 to determine a point 145 that is not actually located at an obstacle 115. For example, a reflection may occur at an object that is not within the movement range of the household appliance 100, or the signal detected by the receiving device 130 may not be based on a simple reflection of the measurement signal emitted by the transmitting device 125, but rather on sunlight or multiple reflections of the measurement signal. Such a determined point 145 is also called a phantom point.

[0029] The processing unit 120 is preferably configured to perform a SLAM procedure to determine, on the one hand, the position of the household appliance in the environment 105 and, on the other hand, the environment map. It is proposed to examine certain points 145 to determine whether they are phantom points. Phantom points can be discarded, and the determination of the environment map and / or the position can be improved by relying on points 145 that are not phantom points.

[0030] Figure 2 Figure 1 shows a flowchart of a method 200 for checking certain points 145. The method 200 can be carried out, in particular, on board a household appliance 100, preferably by means of a processing device 120. The sequence in which the steps shown are carried out can be varied. Different methods 200 can be formed by omitting steps, in particular step 235.

[0031] In step 205, a measurement signal can be transmitted by means of the transmitter 125. If the sensor 135 is, for example, a LiDAR sensor, the measurement signal can include light; in the case of a radar sensor 135, radar waves; and in the case of an ultrasonic sensor, sound waves. Other types of sensors 135 can use other measurement signals. In step 210, a signal can be received by means of the receiver, which may, in particular, include a reflection of the previously transmitted measurement signal.

[0032] In step 215, a location in the vicinity 105 of the household appliance 100 can be determined where the presumed reflection of the received signal occurred. At this location, a point 145 can be identified. Initially, however, it is unclear whether point 145 is actually part of an obstacle 115 or a phantom point to which no obstacle 115 is assigned. In step 220, point 145 can be assigned to a tile 110 based on its location. Assigning points 145 to tiles 110 has the advantage that not all points 145 detectable from a position of the household appliance 100 need to be processed, but only those that are in a predetermined spatial relationship.

[0033] Steps 205 to 220 can be repeated multiple times to perform a large number of scans of the environment 105, which can result in a large number of points 145 that may be located in different tiles 110. The household appliance 100 can be stationary or moving on the floor area of ​​the environment 105 during the scan.

[0034] In step 225, a normal distribution transform (NDT) can be determined for the points of a tile 110. The NDT was proposed by Peter Biber and Wolfgang Straßer, University of Tübingen, in "The Normal Distributions Transform: A New Approach to Laser Scan Matching" and comprises a normal distribution of points 145 that reflects a local probability of determining a point 145.

[0035] The NDT can be determined as follows for tiles 110 that contain at least three points 145: 1.) Collect all points xi=1..n in tile 110; 2.) Determine an average location 3.) Determining the covariance matrix

[0036] The probability of determining a point 145 at the two-dimensional point x of the considered tile 110 can then be modeled by the normal distribution N(q, Σ):

[0037] This allows for a piecewise continuous and differentiable description of the two-dimensional (horizontal) plane in the form of a probability density function. Further details on the determination can be found in the aforementioned publication.

[0038] In step 230, it can be determined whether tile 110 mainly contains noisy points 145. This is particularly the case if the points 145 do not form a straight line in tile 110. The determined NDT contains the average location q and a variance of the points 145 encompassed by tile 110. In two dimensions, the variance comprises two eigenvectors, for each of which an eigenvalue can be determined. An eigenvalue describes how much variance is contained in that eigenvector. A quotient of the larger eigenvalue divided by the smaller eigenvalue describes whether the variances are more or less evenly distributed. If they are equally distributed, the quotient is close to one, and tile 110 does not contain a straight line along which points 145 are concentrated.On the other hand, if the quotient is, for example, 10 or more (typically several orders of magnitude more than one), then a significant number of points 145 of tile 110 may lie along a straight line. In step 230, it can therefore be determined whether the quotient is closer than a predetermined measure to one. If this is the case, the points 145 of the considered tile 110 can be discarded. Otherwise, the points 145 of tile 110 can be retained.

[0039] In step 235, it can be determined for each individual point 145 of a tile 110 whether it is closer to the average location q than a further predetermined measure. This check is preferably performed for all points 145 of a tile 110. The distance between a point 145 and the average location q can be determined as the Mahalanobis distance. If the determined distance exceeds the further predetermined measure, the point 145 can be discarded; otherwise, it can be retained.

[0040] In step 240, based on the undisclosed points 145, the position of the household appliance 100 and / or the obstacle 115 can be determined. The obstacle 115 can be entered into the environment map, taking into account any existing information about the obstacle 115.

[0041] Figure 3 shows exemplary variances 305 of points 145 in tiles 110.

[0042] Figure 3aThis shows a variance 305, which corresponds to points 145 that tend to be evenly distributed in a tile 110. A first eigenvector 310 and a second eigenvector 315 (which are perpendicular to each other) have similar eigenvalues, as can be seen from the similar lengths of the eigenvectors 315. A quotient of these eigenvalues ​​is close to one.

[0043] Figure 3b This shows a variance 305, which corresponds to points 145 that tend to lie on a straight line. A first eigenvector 310 and a second eigenvector 315 have different eigenvalues, visible from the different lengths of the eigenvectors 315, and the quotient of the eigenvalues ​​is significantly different from one.

[0044] Figure 4shows exemplary scans of an environment 105, comprising a sample household. The dark background represents a floor area under consideration in a top view; a subdivision into tiles 110 is not shown. Each specific point 145 is represented by a light spot.

[0045] Figure 4a shows points 145 in the vicinity of 105 before the rejection; Figure 4b after the rejection of points 145, which were identified as phantom points. The rejection can be carried out in particular after steps 230 and 235 of procedure 200 of Figure 2 take place.

[0046] Although the scanning of Figure 4b Apparently only a few points (145) less than the scan of Figure 4a The edges of obstacles 115 are still clearly recognizable by the points 145 that accumulate along them.

[0047] Individual points 145, which are in the sampling of Figure 4anear an edge of an obstacle 115 are visible, but are not part of the edge, are in the second scan of Figure 4b significantly reduced. This effect may be caused in particular by step 235.

[0048] Furthermore, in the sampling of Figure 4b Free areas between obstacles 115 significantly fewer individual points 145. This effect may be caused in particular by step 230. Reference sign

[0049] 100 Household appliance 105 Environment 110 Tile 112 Device 115 Obstacle 120 Processing device 125 Transmitting device 130 Receiving device 135 Sensor 140 Memory 145 Point 200 Procedure 205 Send measurement signal 210 Receive reflection 215 Assign point 220 Assign point to tile 225 Determine NDT for tile 230 Discard tile with noisy points 235 Discard points far from average point 240 Determine environment map 305 Variance 310 First eigenvector 315 Second eigenvector

Claims

1. Method (200) for creating a map of the surroundings for a mobile household appliance (100), wherein the method (200) comprises the following steps: - transmitting (205) a measurement signal into the surroundings (105) of the household appliance (100) and receiving (210) reflections of the measurement signal; - determining (215) points (145) in the surroundings (105) which are allocated a reflection of the measurement signal; - determining (225) a spatial distribution of the points (145) in the surroundings (105); - discarding (230) points if the distribution corresponds less than predetermined to a straight line; and - creating (240) the map of the surroundings on the basis of the remaining points (145), wherein the surroundings (105) are divided into tiles (110) of a predetermined size; - characterised in that the points (145) of a tile (110) are discarded if the spatial distribution of the points (145) of the tile (110) corresponds less than predetermined to a straight line.

2. Method (200) according to claim 1, wherein the method (200) comprises the following step: - discarding (235) points (145) whose distance to an average location of the points (145) exceeds a predetermined value.

3. Method (200) according to claim 2, wherein the surroundings (105) are divided into tiles (110) of a predetermined size; and wherein a point (145), which is assigned to a tile (110), is discarded if its distance to an average location of the points of the tile (110) exceeds the predetermined value.

4. Method (200) according to one of claims 1 to 3, wherein a normal distribution transformation NDT is determined (235) for the points of a tile, wherein the NDT indicates an average location of the points (145) and the variance of the points (145).

5. Method (200) according to claims 1 and 4, wherein the points (200) are each determined in two dimensions; wherein the variance of the points (145) of a tile includes two eigenvectors (310, 315); wherein eigenvalues of the eigenvectors (310, 315) are determined; wherein the points (145) of a tile (110) are discarded if a relationship between the larger and the smaller of the eigenvalues deviates more than predetermined from one.

6. Method (200) according to claims 1 and 4, wherein the points (200) are each determined in two dimensions; wherein the variance of the points (145) of a tile includes two eigenvectors (310, 315); wherein eigenvalues of the eigenvectors (310, 315) are determined; wherein the points (145) of a tile (110) are discarded if one of the eigenvalues exceeds or is below a predetermined value.

7. Method (200) according to claims 3 and 4, wherein the distance is determined on the basis of the Mahalanobis distance of the point (145) to the average point.

8. Apparatus (112) for determining a map of the surroundings for a mobile household appliance (100), wherein the apparatus (112) comprises the following: - a transmission facility (125) for transmitting a measurement signal into the surroundings (105) of the household appliance (100); - a receive facility (130) for receiving reflections of the measurement signal; and - a processing facility (120) that is configured so as: o to determine points (145) in the surroundings (145) which are allocated a reflection of the measurement signal; o to determine (225) a spatial distribution of the points (145) in the surroundings (105); o to discard points (145) if the distribution corresponds less than predetermined to a straight line; and o to determine the map of the surroundings on the basis of remaining points (240); o to divide the surroundings (105) into tiles (110) of a predetermined size; characterised in that o the processing facility (120) is configured so as to discard the points (145) of a tile (110) if the local distribution of the points (145) of the tile (110) corresponds less than predetermined to a straight line.

9. Apparatus (112) according to claim 8, wherein the apparatus (112) comprises the following: - the processing facility (120), which is configured so as o to discard points (145), whose distances to an average location of the points (145) exceed a predetermined value.

10. Apparatus (112) according to claim 8 or 9, wherein the transmission facility (125) and the receive facility (130) are comprised by a LiDAR sensor (135).

11. Floor cleaning robot (100), comprising an apparatus (112) according to one of claims 8 to 10.