Method for detecting a change in a surrounding area

Abstract

The invention relates to a method for detecting a change in a surrounding area, wherein the mobile device moves or is intended to move in the surrounding area, having the steps of: providing (500) a test set (502) of points in the surrounding area, said points having been detected in the surrounding area at least partly by means of a sensor; providing (510) a comparison data set (512) comprising information relating to a grid (514) comprising cells (516), the distribution of each of which is indicative of the probability with which a point of a comparison set (518) of points in the surrounding area is associated with the cell in question; carrying out (520) a test process for one or each of a plurality of selected test points of the test set, wherein a grid cell to be associated with the selected test point is determined during the test process in each case, and information about a change is determined on the basis of the distribution of the associated cell in order to obtain a test result; determining (570) whether there is a change in the surrounding area and / or determining the change in the surrounding area on the basis of the test result of the one or at least some of the plurality of selected test points, and providing (572) information (574) about the presence of a change and / or information about the change in the surrounding area.

Description

[0001] R.410963 Description Title Method for detecting a in one The present invention relates to a method for detecting a change in an environment, in particular for use in the navigation of a mobile device, in particular of an at least partially automated vehicle or robot, a data processing system and a computer program for carrying out the same, and a mobile device. Background of the invention Mobile devices such as at least partially automated vehicles or robots typically move in an environment, in particular an environment to be processed or a work area, such as a home, in a garden, in a factory hall or on the street, in the air or in water. One of the fundamental problems of such or other mobile devices is to orientate itself, i.e. to know what the environment looks like, in particular where obstacles or other objects are, and where it is (absolutely) located. For this purpose, the mobile device can, for example,B. be equipped with various sensors, such as cameras, lidar sensors, radar sensors or even inertial sensors, with the help of which the environment and the movement of the mobile device are recorded, e.g. two- or three-dimensionally. This enables the mobile device to move locally, detect obstacles in a timely manner and avoid them. Disclosure of the invention R.410963 According to the invention, a method, in particular a computer-implemented method, for detecting a change in an environment, a system for data processing and a computer program for carrying out the same, as well as a mobile device with the features of the independent patent claims are proposed. Advantageous embodiments are the subject of the dependent claims and the following description. The invention generally relates to mobile devices that move or can move in an environment or there, e.g. in a work area.Examples of such mobile devices (or mobile work equipment) include robots and / or drones and / or partially or (fully) automated vehicles (on land, water, or in the air). Robots can include, for example, household robots such as cleaning robots (e.g. in the form of vacuum and / or mop robots), floor or street cleaning equipment, construction robots, or lawnmower robots, as well as other so-called service robots, as well as at least partially automated vehicles, e.g. passenger transport vehicles or goods transport vehicles (also so-called industrial trucks, e.g. in warehouses), but also aircraft such as drones or watercraft. Such a mobile device has, in particular, a control or regulating unit and a drive unit for moving the mobile device, so that the mobile device can be moved in the environment, e.g. along a movement path.For this purpose, navigation information can be determined, for example, specific instructions regarding the direction in which the mobile device should travel in order to follow the movement path. These can then be implemented via the control or regulating unit and the drive unit. In general, this can be referred to as the navigation of the mobile device. In addition, a mobile device can have one or more sensors by means of which the environment or information in the environment can be recorded. As mentioned, these can be, for example, cameras, lidar sensors, radar sensors or inertial measurement units (or inertial sensors) as well as radodometry, with the help of which the environment and the movement of the mobile R.410963 device are recorded, for example, two- or three-dimensionally. Depending on the type of mobile device, other or additional sensors can also be provided. Using sensors such asCameras, lidar sensors, radar sensors, or other depth sensors can capture and provide information, particularly sensor data, about the environment. This includes information about the surroundings of the mobile device, for example, distances of the mobile device (or the sensor in question) to objects in the environment. Such data is particularly present in the form of a set of points (point set). In the case of a lidar sensor, each point indicates distances of the mobile device or the lidar sensor to an object or a specific point on the object. In this context, the set of points is also referred to as a so-called point cloud. Such sets of points can also be captured with other sensors, particularly depth sensors. Typically, such a set of points is generated by scanning such a sensor.A point cloud is captured; this is ultimately a data set with distance information to each of several points. One way to determine the position and orientation, and thus also for navigation, of such a mobile device is to use localization of the mobile device based on SLAM. SLAM ("Simultaneous Localization and Mapping") is a method in robotics in which a mobile device, such as a robot, can or must simultaneously create a map of its surroundings and estimate its spatial position within this map. It thus serves to detect obstacles and thus supports autonomous navigation. In SLAM, there are various approaches to representing maps and positions. Conventional SLAM methods are generally based on geometric information such as nodes and edges. Nodes and edges are typically components of the SLAM graph.The nodes and edges in the SLAM graph can be designed in different ways; traditionally, the nodes correspond, for example, to the pose (position and orientation) of the mobile device or certain environmental features at specific times, while the edges represent relative R.410963 measurements between the mobile device and the environmental feature. SLAM graphs are described in more detail, for example, in “Giorgio Grisetti, Rainer Kümmerle, Cyrill Stachniss, Wolfram Burgard, A Tutorial on Graph-Based SLAM, IEEE Intelligent Transportation Systems Magazine, Vol. 2(4), pp. 31-42, 2010.” Based on such a SLAM graph, a map of the environment (environment map) in which the mobile device moves can be determined or will be determined. With each new data set with information about the environment or, if applicable, the mobile device, which information is obtained from or is based on one or more sensors of the mobile device, the map (orThe SLAM graph can be expanded or updated. When localizing using SLAM, it may be particularly necessary to compare two different sets of points (or point clouds). In particular, an attempt is made to bring both sets of points into agreement, at least within certain tolerances. This can also be referred to as a so-called match or scan match (the latter refers in particular to the case of a lidar sensor that records data as part of a scan). A frequently used term in this context is so-called "point cloud registration." Another aspect that can be relevant for such mobile devices, and especially for navigation in the environment, is detecting changes in the environment, e.g., in a home due to moving furniture, people, animals, other parked objects, or other changes. Outdoors, this could also be, for example, temporarily parked vehicles or the like.This can be relevant for efficient navigation. The map of the environment created, for example, using SLAM should then also be updated accordingly. Alternatively, a map with only static objects could be created by filtering out mobile or moving objects, although this can be cumbersome. R.410963 Within the scope of the present invention, a possibility is now proposed for detecting a change in the environment based on two sets of points (point sets), such as point clouds from a lidar sensor. Although this is explained below using the example of a mobile device such as a robot and its navigation in the environment, it can also be applied in other areas; for example, the tracking of changes on a construction site, a comparison of the actual and target state on a construction site (e.g.This could include validating loop closures, relocalization, or map merges (especially in the navigation of mobile devices), detecting moving objects such as doors, detecting changes in the earth's surface, or tracking vegetation growth. A basic approach to detecting a change based on two sets of points is to classify parts of the environment, i.e., points of a set, into one of two classes: "changed" or "not changed." This classification should then be as accurate as possible, meaning there should be neither false positives nor false negative assignments. Any sensor noise leads to inaccuracies in the sets of points and thus also in the classification. Other effects such as disturbances can also have an impact.If entire objects are removed, this can lead to problems, particularly at the edges of the objects where they border other objects. Furthermore, it is often important, especially when navigating mobile devices, that such detection can take place particularly quickly. For example, detection of possible changes with the receipt of each new set of points, e.g. each new scan of the lidar sensor, may be necessary or at least desirable, although it is conceivable that a possible change can only be detected with each so-called "key frame" (e.g. every tenth set of points or each set of points determined according to another specific criterion). R.410963 Against this background, it is proposed within the scope of the present invention that a test set of points in the environment is provided, which have preferably been at least partially detected from the environment by means of a sensor.In addition, a comparison data set comprising information on a grid (e.g. with two or more dimensions) with cells is provided, wherein each cell is assigned a distribution, the distribution of each cell indicating a probability that a point is assigned to a comparison set of points in the environment of the cell. Points of the comparison set have preferably also been recorded at least partially from the environment by means of a sensor, e.g. also by means of the same sensor as the points of the test set. The distribution assigned to each cell is in particular a so-called beta distribution, which will be explained in more detail later. However, the use of a different distribution with, for example, two values, is also conceivable, but without explicitly using a beta distribution. The test set and the comparison set can, for example, be two point clouds recorded by a lidar sensor.The test set represents the set of points for which it is to be checked whether there is a change compared to the reference set or whether such a change can be detected. A grid (or raster) with cells, where each cell is assigned a beta distribution, is also referred to as a beta grid. A beta grid is essentially an occupancy grid, similar to approaches based on so-called octrees. In a classic occupancy grid, there is a single value for each grid cell that indicates whether the cell is occupied, free, or unknown. Probabilistic occupancy grids specify a probability value for each cell that indicates the probability that the cell is occupied. This probability is updated for each observation of this cell. Due to the so-calledHowever, with the Markov assumption, the information about how often a cell has already been seen and how many observations have been made is lost. This means that if a cell R.410963 is seen for the fifth time with a certain probability p (this means, for example, that the occupancy of the cell is checked for the fifth time), this has the same effect as if a cell is seen for the hundredth time with the same probability, even if the hundred previous observations would intuitively indicate that the occupancy probability is already described much more accurately than after only five observations. In other words, for example, if the occupancy of a cell is given as 80% for each observation, then intuitively with each new observation with this probability of 80%, it should become increasingly likely that the cell is actually occupied.However, the typical probabilistic occupancy grid makes no distinction in the number of observations or their history. This is where the beta grid comes into play, modeling the occupancy distribution for each cell. The application of the beta grid is not limited to two dimensions; it can be used with more, e.g., three, dimensions. In 3D, octrees (special data structures) can also be used for a beta grid to avoid storing large amounts of data for cells without observations. In 2D, so-called quadtrees can also be used for a beta grid. To use a beta grid, a probabilistic sensor model should be used. For a lidar sensor, such a model can assume, for example, that a laser beam from a certain direction hits an obstacle at a certain distance.Therefore, the occupancy probability in a cell in which the hit occurred is high, while the occupancy probability in all cells between the lidar sensor and the hit (i.e., at the position in the environment where the laser beam is reflected) is low. These cells can be efficiently determined, for example, using the so-called Bresenham algorithm, as mentioned in "M. Abrash, Michael Abrash's Graphics Programming Black Book, Special Edition. US: Coriolis Group, 2001." In general, more complicated sensor models are also possible, e.g., models in which the laser beams broaden with distance, and models in which the depth measurement is subject to uncertainty, i.e., a hit in a specific cell can, with a certain probability, actually be in another cell that is slightly in front of or behind the expected cell. Such models are particularly important when less accurate sensors, such asUltrasonic sensors are used. To account for any measurement uncertainty, a sensor model can, for example, mark not only the cell of the hit as occupied, but also its neighbors, in particular its left, right, upper, and lower neighbors, but only with half the weight. A more precise approach would be, for example, the use of a Gaussian kernel. A single observation of a cell is modeled, for example, by a Bernoulli distribution. The Bernoulli distribution for ^ ∈ {0,1} with probability ^ ∈[0,1] is given by. The conjugate prior for the Bernoulli distribution is the beta distribution, making it a suitable choice for modeling the occupancy in each cell. The beta distribution for ^ ∈ [0,1] with parameters α, β > 0 is given by with the normalization constant The mean can be calculated as For ^ ∈ ^ observations ^^, … , ^^ of a single cell, a recursive update rule based on Bayes' law can be derived. Assuming independence of the measurements ^^, we obtain: R.410963 ^(^|^^, … , ^^)∝ ^(^^|^)^(^|^^ , … , ^^^^)= ^Bernoulli(^^; ^) ⋅ ^beta(^; α^^^, β^^^) It is therefore known that ^(^|^^, … , ^^) is beta-distributed with the following parameters: ^(^|^^, … , ^^) = ^beta(^; α^, β^) Consequently, the beta distribution in a cell can be updated for each observation simply by, for example, increasing α or β by one, depending on whether the cell is observed as free or occupied. For ^ = 0, i.e., before the first observation, a uniform distribution can be assumed, i.e., α^ = β^ = 1. This procedure can also be generalized, for example, to allow for weighted measurements. For a weight ω > 0, a weighted update can be performed: ^(^|^^, … , ^^) R.410963This is useful, for example, when the sensor model has a certain uncertainty and the updates for different cells should be weighted separately. To classify a point ^ ∈ ^ based on a map ^, the beta grid can be determined or calculated based on all scans in the map. All scans in the map correspond to the point sets underlying the map, whereby these point sets together can then be used, for example, as the comparison set of points. After providing the test set of points and the comparison dataset with the information on the grid, the so-called beta grid, a test process is performed for one or each of several selected test points in the test set.During the test process, a cell of the grid is determined to be assigned to the selected test point, whereby information about a change is determined based on the beta distribution of the assigned cell in order to obtain a test result. Thus, for a test point (or generally a point of the test set), the cell (of the beta grid, which is based on the comparison set) in which the point ^ is located is determined, and the beta distribution associated with this cell is examined. It is then determined whether a change in the environment exists and / or such a change in the environment is identified; this is done based on the test result of one or at least some of the several selected test points. Information about the existence of a change and / or information about the change in the environment is then provided.Based on the information about the presence of a change and / or the information about the change in the environment, navigation information can then be determined for the mobile device, based on which the mobile device can then navigate or be moved. For applications other than mobile device navigation, the information about the presence of a change and / or the information about the change in the environment can be used for other purposes. One advantage of using such a beta grid or the beta distribution is that previous observations are given greater weight, which improves the assessment of whether a change has occurred. For example, if an obstacle has been detected in many scans at a certain location, i.e.that a point of a point set is present there, the probability that a point is present in the respective cell of the beta grid is very high. If no point is detected at the location of the cell in a new scan of the test set, there will very likely be a change there. In one embodiment, the test process comprises, based on the beta distribution of the assigned cell, classifying the selected test point or the assigned cell as either changed, unchanged, or undeterminable in order to obtain the test result. For this purpose, for example,a threshold value ^ ∈ [0,1] can be defined so that the following cases can be distinguished: If α = β = 1 : "not determinable" (no decision) If (α > 1 ∨ β > 1) ^^^ μ(α, β) > ^ : "not changed" If (α > 1 ∨ β > 1) and μ(α, β) ≤ ^ : "changed" In one embodiment, the selected test point or the associated cell in the test process is thus classified as not changed if a comparison value based on the beta distribution, e.g. a mean value of the beta distribution, is greater than a threshold value, and the selected test point or the associated cell in the test process is classified as changed if such a comparison value of the beta distribution is smaller than the threshold value. If the comparison value is equal to the threshold, a classification as "changed" can be made; however, it is also conceivable that a classification as R.410963 "not changed" is made. Instead of an average, for example,A confidence interval can also be used, although using a mean has been shown to be particularly computationally inexpensive. Instead of a single threshold, two different thresholds could also be used, with a first threshold ^. ^ and a second threshold ^ ^with < ^2. In this case, the classification can be, for example, as follows: If μ(α, β) > ^^ : "not changed" If μ(α, β) < ^1 : "changed" If ≤ μ(α, β) ≤ ^2 : "not determinable" (no decision) The cases in which the comparison variable is equal to the first or second threshold value can also be classified differently, for example, as changed or not changed. Otherwise, the above conditions could apply. It should be mentioned that the second case, with the condition that the first threshold value is equal to the second threshold value, would correspond to the first case. In this way, not only two possible test results are used, as is usual for such decisions, but a third possible test result. This avoids an ambiguous situation being accidentally incorrectly classified.So if it is not entirely clear or cannot be said with sufficient certainty whether or not a change has occurred for a cell, no decision is made. This increases the accuracy of detecting whether or what type of change has occurred in the environment. In one embodiment, for one or each of several selected test points classified as unchanged, it is also provided that a check is carried out to determine whether a first normal of the selected test point classified as unchanged contradicts a second normal of the associated cell, and if so, the selected test point classified as unchanged is reclassified as changed. R.410963 Normals can therefore also be included in this beta lattice approach. For example, an approach based on the von Mises distribution can be used for this purpose. This is a probability distribution from the field of circular statistics that is defined on the unit circle.Since the normal direction behaves like a point on the unit circle, the von Mises distribution can be used to model the normal orientation in each cell. The von Mises distribution for ^ ∈ [0,2π) with the position parameter μ ∈ [0,2π) and the concentration parameter κ ≥ 0 is given by:^VM(^; μ, κ) = ^ ⋅ exp(κ cos(^ − μ)) with the normalization constant. where ^ ^is the modified Bessel function of the first kind. Note that a uniform distribution can also be represented by setting ^ = 0. For the beta distribution, a recursive update rule can be derived to incorporate new measurements into the current estimate. Such an update rule is based, for example, on the multiplication of the von Mises probability density functions (or von Mises distributions), as described, for example, in "G. Kurz, I. Gilitschenski, and UD Hanebeck, “Recursive Bayesian filtering in circular state spaces,” IEEE Aerospace and Electronic Systems Magazine, vol. 31, no. 3, pp. 70–87, Mar 2016." Specifically, for example, the prior distribution and the observation probability are both modeled as von Mises distributions, and the posterior, which corresponds to their renormalized product, is also von Mises distributed. For a measurement ^^ ∈ [0,2π) with a concentration κ^ 0 results in:^(^|^^, … , ^^) = ^(^^|^)^(^|^^ , … , ^^^^) R.410963 where the new parameters μ^ , κ^ can be obtained by the equations given in “G. Kurz, I. Gilitschenski, and UD Hanebeck, “Recursive Bayesian filtering in circular state spaces,” IEEE Aerospace and Electronic Systems Magazine, vol. 31, no. 3, pp. 70–87, Mar 2016.” For a practical implementation, it may be more efficient to use the following parameterization where ^μ^, μ^^^ is a unit vector to avoid the computational effort required to calculate trigonometric functions. In this case, the following update equations can be derived:^ = κ^^^ ⋅ μ^,^^^ + κ^ ⋅ μ^,^^ = κ^^^ ⋅ μ^,^^^ + κ^ ⋅ μ^,^ μ^,^ = ^ / κ^ To decide whether a given normal agrees with the normal estimated based on the von Mises distribution, one can consider, for example, a suitable confidence interval of the von Mises distribution. However, the calculation of confidence intervals is usually computationally intensive, as it requires calculating the inverse of the integral of the von Mises probability density function (von Mises distribution), which cannot be done analytically. Therefore, it is particularly advantageous to use the interval μ ± σ, where σ is the circular standard deviation given by: σ = ^−2 log(^), where ^ is the mean resulting length: Even if the well-known property of the Gaussian distribution, that a standard deviation covers approximately 68% of the total probability, does not apply in this case, the proposed method delivers reasonable results as long as the uncertainty is not too large. In one embodiment, it is also provided that the beta distribution—and, if normals are used, in particular also the von Mises distribution—of at least some of the cells is updated based on an additional set of points by adjusting the beta distribution based on information as to whether or not a point from the additional set can be assigned to the cell. The comparison set is then expanded to include the additional set as a comparison set for a subsequent test procedure. In this way, new observations can always be added to the comparison set or the beta grid. For example,The beta distribution can be adjusted depending on a weight assigned to the point of the additional set and / or the cell. The additional set can include the test set and / or another set of points. In the case of the test set, for example, a map is gradually built up. This also explains why, instead of the beta distribution, a different distribution can be used, e.g., with only two values, as already mentioned above. In a practical implementation, the parameters α, β of the beta distribution can be stored for each cell, and these can be incremented by 1 each time the cell is observed to be free or occupied. This counting of free or occupied observations can now also be implemented in R.410963 without explicitly assuming a beta distribution. Instead of the mean or the comparison value of the beta distribution, for example,A decision can be made directly based on the counters as to whether the cell is free or occupied. Another possibility is proposed below, the so-called ray trace method. Here, a test set of points in the environment is provided, which have preferably been at least partially detected by a sensor from the environment. The test set of points is provided in polar or spherical coordinates; the polar or spherical coordinates can comprise two or more dimensions, e.g., three dimensions. In addition, a comparison set of points in the environment is provided. Points in the comparison set have preferably also been at least partially detected by a sensor from the environment, e.g., using the same sensor as the points in the test set.The comparison set of points is also provided in polar or spherical coordinates; the polar or spherical coordinates can encompass two or more dimensions, e.g., three dimensions. The test set and the comparison set can, for example, be two point clouds acquired using a lidar sensor. The test set represents the set of points for which it is to be checked whether there is a change compared to the comparison set or whether such a change can be detected. The origins of the test set of points and the comparison set of points in the polar or spherical coordinates are at least essentially the same. The point sets can, for example, have been acquired from the same position, or one can have been transformed to the other point set; a point set can also originate from several individual scans that are matched. R.410963 A test procedure is then carried out for one or each of several selected test points of the test set. During the test procedure, the selected test point is classified based on a set of neighboring points from the comparison set. During the classification, the selected test point is classified either as changed, not changed, or undeterminable in order to obtain a test result. It is then determined whether a change in the environment exists and / or such a change in the environment is determined; this is done based on the test result of one or at least some of the several selected test points. Information about the existence of a change and / or information about the change in the environment is then provided. Based on the information about the existence of a change and / or the information about the change in the environment, for exampleNavigation information for the mobile device can also be determined, based on which the mobile device can then navigate or be moved. For applications other than mobile device navigation, the information about the presence of a change and / or the information about the change in the environment can be used for other purposes. For a more detailed explanation, see “JP Underwood, D. Gillsjö, T. Bailey, and V. Vlaskine, “Explicit 3D change detection using ray-tracing in spherical coordinates,” in Proc. IEEE Int. Conf. Robotics and Automation, 2013, pp. 4735–4741”; there, an algorithm for detecting changes based on so-called ray tracing is proposed. The algorithm is based on the use of polar or spherical coordinates for 2D and 3D scenarios, respectively.In both cases, the coordinate system is expediently centered on the sensor origin of the scan used for classification, i.e., the test set. A point ^ at a specific location means that the line ("ray") from the origin to ^ represents free space and does not touch any other points. R.410963 The proposed procedure is based on this, but introduces the third possibility "undetermined" in addition to the classifications "changed" and "not changed" used there. In this way, not only two possible test results are used, but also a third possible test result. This avoids an ambiguous situation being accidentally incorrectly classified. Therefore, if it is not entirely clear or cannot be said with sufficient certainty whether a change has occurred for a point or not, no decision is made.This increases the accuracy of detecting whether or which change is present in the environment. In one embodiment, the test process comprises determining a set of neighboring points from the comparison set such that the neighboring points lie within an angular range in polar or spherical coordinates around the selected test point. It is then determined whether the selected test point lies within a convex hull around the neighboring points, and if the selected test point is not within the convex hull, the selected test point is classified as undeterminable. For each point ^ in the point cloud to be classified ^ (the test set), the neighborhood ^ (of the set of neighboring points) in polar or spherical coordinates in the other scan ^ (the comparison set) can be calculated, e.g., using a specific window size ^ (angular range).In the polar case, when ϕ(^) = arctan 2 (^^, ^^) calculates the polar angle of ^, the following is obtained: ^= {^ ∈ ^: ϕ(^) ∈ [ϕ(^) − ^, ϕ(^) + ^]} ⊂ ^Similarly, the neighborhood in spherical coordinates can be calculated based on the azimuth and inclination angles. In one embodiment, the checking process further comprises, if the selected check point lies within the convex hull, checking whether a distance of the selected check point from the origin is less than a minimum distance of the neighboring points from the R.410963 origin by more than a first threshold, and if so, classifying the selected check point as changed. It should be noted here that the thresholds used in connection with the ray-trace method have no relationship to the thresholds used in connection with the method based on the beta distribution.If the point lies within the convex hull of its neighbors ^, a distance check for ^ can be performed. For this purpose, | ^ | , the distance of the point to the origin (i.e., the position of the sensor where the scan used for classification ^ was recorded) is determined. This distance is then compared with the distance of the point in ^ that is closest to the origin, which is given by ^ min = min | ^ ^ | ∈^If |^| < ^min − for a threshold ≥ 0, a change is detected, ie, the point is classified as changed. Without the proposed use of the third option "not determinable," otherwise, ie, if the stated condition does not hold, no change would be detected, ie, each point in ^ would be classified into one of two classes. The threshold However, it should always be chosen depending on the sensor noise and alignment accuracy to avoid classifying points as changed that are only slightly closer than points in ^. This approach with only two classes or two possible test results works in some cases, but has some significant limitations. For example, it does not distinguish between points that match the other point cloud and points for which no decision can be made because they were not seen by the sensor that recorded the scan. The neighborhood ^ in polar or spherical coordinates can contain depth jumps and have points at completely different distances from the sensor. The choice of is difficult because values ​​that are too small lead to false positives due to R.410963 noise and imperfect scan alignment, and values ​​that are too large lead to false negatives because small changes are not detected. Therefore, a modified depth test is proposed, with which points can be classified into the three classes mentioned instead of two. Here, too, the following applies: If a point has a smaller distance |^| to the origin than the point in the neighborhood that is closest to the origin (minus a threshold ^ ^), it is considered to have changed. In one embodiment, the checking process further comprises, if this is not the case (i.e., if the distance of the selected check point from the origin is less than the minimum distance of the neighboring points from the origin by more than the first threshold), checking whether a minimum distance of the selected check point from one of the neighboring points is less than a second threshold. If this is the case, the selected check point is classified as not changed, and if this is not the case, the selected check point is classified as undetermined. The distance of the point to its nearest neighbor is calculated in ^ e.g., in relation to the range or the distance ^ min = m ^∈ i^n ^|^| − |^|^is determined. If ^min < ^^ for a threshold ^^ > 0 (the second threshold), ^ is classified as unchanged because it has a close neighbor in the scan^. Note that this is done in addition to the threshold check. If neither of these conditions is met, ^ is classified as undeterminable because it neither creates a contradiction nor closely matches the other point cloud. In some cases, points in the point cloud ^ and the scan ^ (i.e., the test and comparison set) may be in almost the same location but have different normals, e.g., if an object has been rotated or if a horizontal wall in ^ intersects a vertical wall in ^. In this case, without the use of the normals, the above algorithm would still classify the affected points as unchanged, even if it is intuitively clear that this is not the case.It is therefore proposed to extend the algorithm by taking point normals into account. For this purpose, the normals for each point can first be determined or calculated. This can be done based on the direct neighbors within the same point cloud. It may happen that no normal can be calculated for certain points, e.g. for isolated points without close neighbors. In one embodiment, it is therefore further provided that, for one or each of several selected check points classified as unchanged, a check is carried out to determine whether a first normal of the selected check point classified as unchanged contradicts a second normal of one or more neighboring points of the selected check point classified as unchanged. If this is the case, the selected check point classified as unchanged is reclassified as changed.Preferably, it is determined whether the first normal contradicts the second normal by comparing a dot product of the first normal and the second normal with a third threshold. For example, a mean value of individual normals of the plurality of neighboring points or a normal of a plane defined by the plurality of neighboring points is used as the second normal, wherein it is determined that the first normal contradicts the second normal if the dot product is smaller than the third threshold, or wherein one of the individual normals of the plurality of neighboring points that is closest to the first normal is used as the second normal, and wherein it is determined that the first normal contradicts the second normal if the dot product is smaller than the third threshold. Methods for calculating normals for a given point cloud can be found, for example, in “S. Holzer, RB Rusu, M. Dixon, S. Gedikli, and N. Navab, “Adaptive R.410963 neighborhood selection for realtime surface normal estimation from organized point cloud data using integral images,” in 2012 IEEE / RSJ International Conference on Intelligent Robots and Systems. IEEE, oct 2012.”, in “C. Mura, G. Wyss, and R. Pajarola, “Robust normal estimation in unstructured 3d point clouds by selective normal space exploration,” The Visual Computer, vol.34, no.6-8, pp. 961–971, May 2018.” Or “G. Liu, Depending on whether the point clouds are present as organized point clouds or not and how much computing power is available, a suitable method can be selected by finding a compromise between computational efficiency and accuracy. Assuming there is a point ^ and its normal ^^, it is first classified as described.If it is classified as unchanged, its normal is also compared with the neighborhood ^. If the normal contradicts the neighborhood, the point is classified as changed. To the normal ^. ^ To compare with the neighborhood, various alternative methods can be used. For example, a line or plane fitting can be performed according to the principle of least squares through all points of the neighborhood ^. Then, the normal ^^ of the line or plane is compared with ^ ^ compared, and a contradiction is detected if ^ ^ ^ ⋅ ^^ < ^^ . This works particularly well, for example, when the points are somewhat noisy, but depth jumps and corners may not be handled correctly; it may also require a bit more computing power. Assuming that individual normals are already given for all points in the neighborhood ^, the mean normal can be calculated as follows: R.410963 A contradiction is recognized when ⋅ ^^ < ^^. This method is more robust against depth jumps and faster to calculate, but may be less resistant to noise than the line or plane fitting method. If it is assumed that individual normals are already given for all points in the neighborhood ^, then a point with the most similar normal to ^ ^ can be found as follows: ^best = arg max^ ^ ^ ⋅ ^^ {^ ^ :^∈^} A contradiction is recognized if ^ ^ ^^^^ ⋅ ^^ < ^^. This approach is fast to compute and works well even for depth jumps and corners. However, it may not be very robust against noise. In all cases, ^ ^(the third threshold mentioned above) can be chosen depending on the expected accuracy of the normal calculation and the trade-off between false positives and false negatives. It should be noted that in all cases, the situation in which the dot product equals the third threshold can also be assumed to be a contradiction. With both proposed approaches, point clouds recorded by a suitable sensor, e.g., a 2D / 3D lidar sensor, a time-of-flight camera, a stereo camera, etc., can be used to detect changes. A point cloud recorded from a single location can be referred to as a scan, while a set of aligned scans, each recorded from a different position, can be referred to as a map. This makes it possible, in particular, to distinguish between different scenarios.1) Classifying a scan based on the scan: This involves classifying the points of a single scan based on the information of another single scan R.410963 2) Classifying a map based on the scan: This involves classifying the points of an entire map based on the information of another single scan 3) Classifying a scan based on a map: This involves classifying the points of a single scan based on the information of an entire map 4) Classifying a map based on a category: This involves classifying the points of an entire map based on the information of another entire map To achieve scenarios 2) and 4), it is possible to classify each scan in the map using scenarios 1) and 3) and then aggregate the results.One difference between scenarios 1) and 3) is the amount of information available. The information from a map with many partially overlapping scans enables probabilistic considerations that are difficult or impossible based on a single scan. With the ray trace approach mentioned above, scenarios 1) and 2) in particular are possible, whereby the scan or map to be classified corresponds to the test set of points, and the basis used for this, the scan or map, corresponds to the comparison set of points. With the beta grid approach mentioned above, all of these scenarios are possible, whereby the scan or map to be classified corresponds to the test set of points, and the basis used for this, the scan or map, corresponds to the comparison set of points.With both the ray tracing and beta grid approaches, it can be assumed that the sensor detects or measures a point ^ at a specific location, meaning that the line between the sensor origin and the point ^ only traverses free space and encounters no other obstacles. For this, the location of the sensor's origin should be known. This is relatively simple for a single static sensor, but may be somewhat more difficult if more than one sensor is involved or if the sensor is moving. R.410963 With multiple sensors, e.g., multiple lidar sensors or multiple depth cameras, a common practice is to use extrinsic calibration to transform all point clouds into the same coordinate frame and merge them into a single point cloud.In this way, algorithms that work with the point cloud do not have to consider the various sensors and their calibration, but can work as if the point cloud originated from a single sensor. This approach, however, is not, or not easily, applicable to the presented change detection algorithms. Instead, it is useful to either keep the point clouds separate and process them sensor-by-sensor, or to store the information about the corresponding sensor origin for each point in the point cloud. This can be achieved, for example, by storing a sensor identifier for each point and storing the sensor origin separately for each sensor identifier. Another aspect is moving sensors. This particularly applies to sensors that require some time to record their measurements, e.g., rotating lidar sensors.In this case, a single scan is recorded over a specific period of time, typically a few tens to hundreds of milliseconds, and each point of the scan is recorded at a slightly different time and thus from a slightly different location. For example, motion dewarping algorithms can be used to correct the point positions so that the overall scan is consistent. To do this, a sensor position is selected, e.g., at the beginning or end of the scan, and all points are transformed into this image based on an estimate of the sensor trajectory. In practice, using the undistorted scan with the change detection algorithms presented above usually leads to quite satisfactory results. However, for scanners that rotate slowly or move quickly, the change detection results may differ.deteriorate because the actual sensor origin is not used. In this case, the rectification step can be modified so that the true origin for each point is stored based on the sensor trajectory and the point's timestamp. This R.410963 means a certain additional effort in storage and calculation, but can lead to more accurate results. A data processing system according to the invention or a computing unit, e.g., a control unit or a control unit of a mobile device, or a server or other computer, is configured, in particular in terms of programming, to carry out a method according to the invention or another proposed method, e.g., in one of the described embodiments. The invention also relates to a mobile device having such a data processing system or configured to receive navigation information determined as described above.The mobile device preferably also has a drive system and a control or regulating unit for moving the mobile device according to the navigation information. The mobile device is preferably also configured to carry out processing; in particular, the mobile device can be one as described above, e.g. a cleaning robot or a robotic lawnmower. The implementation of a method according to the invention in the form of a computer program or computer program product with program code for carrying out all method steps is also advantageous, since this entails particularly low costs, in particular if an executing control device is also used for other tasks and is therefore already present. Finally, a machine-readable storage medium is provided with a computer program stored thereon as described above. Suitable storage media orData carriers for providing the computer program are, in particular, magnetic, optical, and electrical storage devices, such as hard disks, flash memories, EEPROMs, DVDs, and others. Downloading a program via computer networks (Internet, intranet, etc.) is also possible. Such a download can be wired or wireless (e.g., via a WLAN network, a 3G, 4G, 5G, or 6G connection, etc.). R.410963 Further advantages and embodiments of the invention emerge from the description and the accompanying drawings. The invention is schematically illustrated in the drawings using an exemplary embodiment and is described below with reference to the drawings. Brief description of the drawings Figure 1 schematically shows a mobile device in an environment to explain the invention. Figures 2a and 2b show sets of points for explanation. Figure 3 schematically shows a sequence of a method.Figures 4a, 4b, 4c, 4d show sets of points for explanation. Figure 5 schematically shows a sequence of a method in one embodiment. Figure 6 shows beta distributions for explanation. Embodiment(s) of the invention Figure 1 schematically and by way of example shows a mobile device 100 in an environment 120, in particular a work area, for explaining the invention. The mobile device 100 is, for example, a robot vacuum cleaner with a control or regulating unit 102 and a drive unit 104 (with wheels) for moving the robot vacuum cleaner 100, e.g., along a movement path 130. Furthermore, the robot vacuum cleaner 100 has, for example, a sensor 106 designed as a lidar sensor with a detection range. For better R.410963 illustration, the detection range is chosen to be relatively small here; In practice, the detection range can also be up to 360° (e.g.but at least at least 180° or at least 270°). The environment 120 can be detected by means of the sensor 106, i.e., specific sets of points can be generated as environmental sensor data. In a scan using the lidar sensor, for example, a set of points, a so-called point cloud, can be generated, with each point indicating a distance from the sensor of an object at which the laser beam is reflected. The robot vacuum cleaner 100 also has a computing unit 108, e.g., a control unit, by means of which data can be exchanged with a higher-level system 110, e.g., via an indicated radio connection. In the system 110, for example, movement paths (or general navigation information) can be determined, which are then transmitted to the system 108 in the robot vacuum cleaner 100, which the robot vacuum cleaner is then to follow.However, it can also be provided that a movement path (or general navigation information) is determined in the system 108 itself or is otherwise received there. Instead of a movement path or the navigation information, the system 108 can, for example, also receive control information that has been determined based on a movement path or the navigation information, and according to which the control or regulating unit 102 can move the robot vacuum cleaner 100 via the drive unit 104 to follow a movement path. The movement path 130 is indicated here only as an example. The robot vacuum cleaner 100 is intended, for example, to move independently in the environment 120 or navigate there and, for example, clean a floor. Furthermore, several different objects or obstacles are shown in the environment as examples, namely a wall 140 and a cupboard 142.Although the invention is explained here and in the following particularly using the example of the robot vacuum cleaner, this also applies to other mobile devices such as robotic lawnmowers or other self-driving vehicles. However, R.410963 also applies to other types of applications, as mentioned at the beginning. Instead of the lidar sensor, a camera, a radar sensor, or another depth sensor can be provided. Figure 2a schematically shows a set 200a of points (point set) that have been detected from a position 204. The points are labeled 202a; laser beams from position 204 to points are shown only as an example and labeled 206. The set 200a shown here can, for example,can be obtained in the situation shown in Figure 1, where the cabinet can be seen at the top right in Figure 2a (where points 202a are closer to position 204), while the wall 140 can be seen at the bottom right in Figure 2a (where points 202a are further away from position 204). Figure 2b schematically shows a set 200b of points (point set) which, for example, were also acquired from position 204; however, it is also conceivable that the set 200b of points was acquired from a different position but was then transformed to position 204. The points are labeled 202b; laser beams from position 204 to points are shown only as an example and are labeled 206.The set 200b shown here differs from set 200a in particular in that the points 202b, which are placed closer to position 204, i.e., have a smaller distance, are more numerous and extend further down (further to the right as seen from position 204). This can be due, for example, to the fact that the bar has been moved accordingly according to Figure 1, i.e., there is a change in the environment. The following explains ways in which such a change can be detected using two such sets of points. Set 200a can be used as the comparison set, and set 200b as the test set, i.e., it is checked whether there are any points in the test set that have changed compared to the comparison set. R.410963 Figure 3 schematically shows the flow of a method, namely the so-called ray trace method.For this purpose, in a step 300, a test set 302 of points in the environment is provided, the points being at least partially detected from the environment by means of a sensor. In step 310, a comparison set 312 of points in the environment is provided, which, for example, have also been at least partially detected from the environment by means of a sensor. In both cases, for example, the sensor of a mobile device as shown in Figure 1 can have been used. These point sets can, for example, have been detected as part of a movement of the mobile device in the environment, e.g. as scans. In addition, both the test set of points and the comparison set of points are provided in polar or spherical coordinates. In Figures 2a, 2b, the point sets are shown, for example, in polar coordinates, or such a representation is at least indicated.The origins of the test set of points and the comparison set of points in the polar coordinates at least essentially agree, as was also explained for the case of Figures 2a, 2b. There, the origins correspond to position 204. In step 320, a test process is then carried out for one or each of several selected test points in the test set. The test process each comprises: classifying the selected test point, based on a set of neighboring points from the comparison set, either as changed, not changed, or undeterminable. For this purpose, in step 330, a set 332 of neighboring points from the comparison set is first determined such that the neighboring points lie within an angular range in polar coordinates around the selected test point. In step 334, it is then determined whether the selected test point lies within a convex hull around the neighboring points. If the selected test point R.410963 does not lie within the convex hull, the selected test point is classified as undeterminable in step 336, i.e., assigned to a class 393 "undeterminable"; this is then a test result. If, on the other hand, the selected test point lies within the convex hull, then in step 340 a check is made as to whether a distance of the selected test point from the origin is less than a minimum distance of the neighboring points from the origin by more than a first threshold value. If this is the case, the selected test point is classified as changed in step 342, i.e., assigned to a class 391 "changed"; this is then a test result. If, on the other hand, this is not the case, then in step 350 a check is made as to whether a minimum distance of the selected test point from one of the neighboring points is less than a second threshold value. If this is the case, the selected test point is classified as not changed in step 352, i.e.,classified into a class 392 "not changed"; this is then a test result. If this is not the case, the selected test point is classified as undetermined in step 354, i.e. classified into class 393 "undetermined". Examples of this are shown in Figures 4a, 4b, and 4c. Each diagram shows a test point 400 with a set of neighboring points 402. The test point comes from a test set, and the neighboring points from the comparison set. Both sets of points are shown in polar coordinates with the same origin. An angular range within which the neighboring points lie is shown as 404. In Figure 4a, for example, the upper diagram shows a convex hull 408 around the neighboring points, and the test point 400 lies within this convex hull. In the middle diagram, the convex hull around the neighboring points is not shown, but here too the test point 400 would lie within it.In the lower diagram, the convex hull around the neighboring points is also not shown, but here too the test point 400 would lie within it. R.410963 In this case, the distance of the test point 400 to the origin is determined, designated 406 in the upper diagram in Figure 4a as an example. A minimum distance of the neighboring points from the origin is designated 410 in the middle diagram. If this distance 406 is less than the distance 410 by more than a first threshold, then the point is classified as changed. In Figure 4a, this is not the case in any of the diagrams. A check is then carried out to determine whether a minimum distance of the selected test point from one of the neighboring points is less than a second threshold. Such a minimum distance is designated 412 in the lower diagram in Figure 4a. Depending on the choice of the second threshold, this can be regarded as met in all diagrams in Figure 4a. The test points would therefore be classified as unchanged.Otherwise, the test points would be classified as undetermined. In Figure 4b, for example, a convex hull 408 around the neighboring points is indicated in the diagram, but the test point 400 is not located within it. In Figure 4c, for example, a convex hull 408 around the neighboring points is indicated in the diagram, but the test point 400 is not located within it here either. The distance of the test point in Figure 4b from the origin is significantly smaller than the average distance of the neighboring points, but in Figure 4c it is significantly larger. In both cases, however, the test point is classified as undetermined. Furthermore, for example, for one or each of several selected test points classified as unchanged, a check can now be carried out in step 360 to determine whether a first normal of the selected test point classified as unchanged contradicts a second normal of one or more neighboring points of the selected test point classified as unchanged.For this purpose, the first normal of the selected test point classified as unchanged can be determined in step 362, and a second normal of the neighboring points can be determined in step 364. R.410963 If ​​this is the case, i.e. the first normal contradicts the second normal, in step 366 the selected test point classified as unchanged is reclassified as changed, i.e. assigned to class 391. In Figure 4d, an object or obstacle 412 is shown in an environment which is said to have been added. Points 491 are then those that are classified as changed, and points 492 are those that are not changed. It can be seen that, for example, in the area designated 414, points are classified as unchanged even though they should actually be classified as changed. The area 414 is also shown enlarged.For a point 492 that is classified as unchanged and for which it is clear that it has changed, a first normal 416 is shown, and for a point 492 that is classified as unchanged and for which it is clear that it has not changed, a second normal 418 is shown; the latter can represent an average normal of several neighboring points. If the first normal contradicts this second normal, the point initially classified as unchanged can be reclassified as changed. Based on the test result of one or at least some of the several selected test points, i.e., the classifications into the classes, it is then determined in step 370 whether a change in the environment exists and / or any change that may exist in the environment is determined, i.e., what type of change it is. In particular, this is done based on the points classified as changed.In step 372, information 374 about the presence of a change and / or information about the change in the environment is then provided. Figure 5 shows a flow of a method in one embodiment, the approach with the beta grid. For this purpose, in a step 500, a test set 502 of points in the environment is provided, the points that have been at least partially detected from the environment by means of an R.410963 sensor. In step 510, a comparison data set 512 is provided. This comprises information about a grid 514 with cells 516, each cell being assigned a beta distribution, the beta distribution of each cell indicating a probability that a point is assigned to a comparison set 518 of points in the environment of the cell. The points of the comparison set have, for example, also been at least partially detected from the environment by means of a sensor. In both cases, e.g.B. the sensor of a mobile device as shown in Figure 1 may have been used. These point sets may, for example, have been recorded as part of a movement of the mobile device in the environment, e.g. as scans. Figure 5 shows exemplary beta distributions, with the various distributions 501 to 509 having the following parameters in this order: α = 1, β = 1; α = 1, β = 2; α = 1, β = 2; α = 2, β = 1; α = 2, β = 2; α = 2, β = 3; α = 3, β = 1; α = 3, β = 2; α = 3, β = 3. The function already mentioned above is used here:. where ^ is plotted on the easting axis, the function value on the northing axis. In a grid, such a function can then be assigned to each cell; for example, each point in Figures 2a, 2b can correspond to a cell in which the probability that a point is in such a cell is high, or even 1. In step 320, a test process is then carried out for one or each of several selected test points of the test set. In each case, the test process comprises determining a cell of the grid to be assigned to the selected test point, wherein information about a change is determined based on the beta distribution of the assigned R.410963 cell in order to obtain a test result.In particular, the test process, step 530, may also include classifying the selected test point, based on the beta distribution of the associated cell, the selected test point, or the associated cell, either as changed, not changed, or undeterminable. In other words, the selected test point is assigned, for example, to a class 592 "changed" or a class 591 "not changed" or a class 593 "undeterminable"; this then constitutes a test result. The selected test point or the associated cell is classified, for example, as not changed, 591, in the test process if a mean value of the beta distribution is greater than a threshold; however, the selected test point or the associated cell is classified as changed, 592, in the test process if a mean value of the beta distribution is less than the threshold. If the mean of the beta distribution is equal to the threshold, for example,can be decided depending on the specification. If there is no probability at all, e.g., because there is no value for it in the comparison set yet, or e.g., if α = 1, β = 1, then the class "undeterminable" can be selected. As mentioned above, when using two thresholds, a classification as "undeterminable" can be made if the mean lies between them. The value of the beta distribution of the cell indicates whether or not there was a point in the comparison set here—at least with a certain probability. If there was a point there, the probability is higher than if there was no point there. Therefore, if the probability is high for a cell where there is definitely a check point, there was probably a point there previously, thus there is no change. In addition, it can be provided that for one or each of several selected check points classified as unchanged, 591, R is checked in step 540.410963 whether a first normal 542 of the selected test point classified as unchanged contradicts a second normal 544 of the assigned cell. If so, in step 546 the selected test point classified as unchanged is reclassified as changed 592. Similar to the ray trace approach, the normals can also be used here to check whether a change has actually occurred. Furthermore, it can be provided that in step 550 the beta distribution of at least some of the cells is updated based on an additional set—this can be, for example, or include the test set—by adjusting the beta distribution based on information as to whether a point of the additional set can be assigned to the cell or not. In step 560, the comparison set, expanded by the additional set, can then be provided as a comparison set for a subsequent test process.Based on the test result of one or at least some of the multiple selected test points, it is then determined in step 570 whether a change in the environment exists and / or a possible change in the environment is determined, for example, the type of change. In particular, this is done based on the points classified as changed. In step 572, information 574 about the presence of a change and / or information about the change in the environment is then provided.

Claims

R.410963 Claims1. A method for detecting a change in an environment, in particular for use in navigation of a mobile device, in particular an at least partially automated vehicle or robot, wherein the mobile device is moving or is intended to move in the environment, comprising: providing (500) a test set (502) at points in the environment, which have preferably been at least partially detected by a sensor from the environment;Providing (510) a comparison data set (512) comprising information about a grid (514) with cells (516), wherein each cell is assigned a distribution, in particular a beta distribution, wherein the distribution of each cell indicates a probability that a point of a comparison set (518) is assigned to points in the environment of the cell, wherein points of the comparison set have preferably been at least partially detected from the environment by means of a sensor; Performing (520) a test process for one or each of several selected test points of the test set, wherein during the test process a cell of the grid to be assigned to the selected test point is determined, and wherein, based on the distribution of the assigned cell, information about a change is determined in order to obtain a test result;Determining (570) whether a change in the environment exists and / or determining the change in the environment based on the test result of the one or at least part of the plurality of selected test points, and providing (572) information (574) about the existence of a change and / or information about the change in the environment; R.4109632. The method of claim 1, wherein the testing process comprises: classifying (530), based on the distribution of the assigned cell, the selected test point, or the assigned cell, either as changed (591), not changed (592), or not determinable (593), to obtain the test result.

3. The method of claim 2, wherein the selected test point or the assigned cell is classified as changed in the testing process if a comparison variable based on the distribution is less than a first threshold, and wherein the selected test point or the assigned cell is classified as not changed in the testing process if the comparison variable based on the distribution is greater than a second threshold, wherein the first threshold is equal to or less than the second threshold.4.The method of claim 3, wherein the first threshold is less than the second threshold, and wherein the selected checkpoint or the associated cell is classified as undeterminable if the comparison value based on the distribution is greater than or equal to the first threshold and less than the second threshold.

5. The method of any one of claims 2 to 4, further comprising, for one or each of a plurality of selected checkpoints classified as unchanged: checking (540) whether a first normal (542) of the selected checkpoint classified as unchanged contradicts a second normal (544) of the associated cell, and if so, reclassifying (546) the selected checkpoint classified as unchanged as changed.

6. The method of any one of the preceding claims, further comprising:. R.410963 Updating (550) the distribution of at least some of the cells based on an additional set of points by adjusting the distribution based on information as to whether or not a point of the additional set can be assigned to the cell, and providing (560) the comparison set expanded by the additional set as a comparison set for a subsequent test process.

7. Method according to claim 6, wherein the distribution is adjusted depending on a weight assigned to the point of the additional set and / or the cell.

8. Method according to claim 6 or 7, wherein the additional set comprises the test set and / or another set of points.

9. Method according to one of the preceding claims, wherein the sensor comprises a depth sensor, in particular a lidar sensor.10.Method according to one of the preceding claims, further comprising: determining navigation information for the mobile device based on the information about the presence of a change and / or the information about the change in the environment.

11. System for data processing, comprising means for carrying out the method according to one of the preceding claims.

12. Mobile device which has a system according to claim 11 and / or which is configured to receive navigation information which has been determined according to a method according to claim 10, and which is configured to navigate based on the navigation information, preferably with a control or regulating unit and a drive unit for moving the mobile device according to the navigation information.

13. Mobile device according to claim 12, which is designed as an at least partially automated moving vehicle, in particular as a. R.410963 passenger transport vehicle or as a goods transport vehicle, and / or as a robot, in particular as a household robot, e.g. cleaning robot, floor or street cleaning device or lawnmower robot, and / or as a drone.

14. Computer program, comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method steps of a method according to one of claims 1 to 10 when it is executed on the computer.

15. Computer-readable storage medium on which the computer program according to claim 14 is stored.

Images