Positioning Method for Floor Cleaning Robot

Abstract

The present disclosure provides a positioning method for a sweeping robot, which includes: obtaining a rough global pose of the sweeping robot according to the surrounding environment information sensed by the depth camera of the sweeping robot, and obtaining a positioning particle swarm according to the rough global pose of the sweeping robot; obtaining a displacement increment of the sweeping robot through the odometer and IMU of the sweeping robot, and performing motion update on each particle in the positioning particle swarm according to the initial pose information of the sweeping robot and the displacement increment of the sweeping robot to obtain the prior pose information of each particle in the positioning particle swarm; obtaining the posterior probability density of each particle in the positioning particle swarm according to the measurement information of the lidar of the sweeping robot; and selecting some particles from the positioning particle swarm according to the posterior probability density, and obtaining the position of the sweeping robot in the global grid map according to the prior pose information of the selected particles, and outputting the position of the sweeping robot in the global grid map.

Description

Technical Field

[0001] The present disclosure relates to a positioning method for a floor cleaning robot, and more particularly to an improved particle filter positioning method based on multi-sensor fusion. Background Art

[0002] Currently, floor cleaning robots have been widely used in various scenarios, such as commercial plazas, hotels, families, etc.

[0003] The floor cleaning robot needs to complete its operation tasks within the working scenario and requires accurate positioning information. For example, when the floor cleaning robot cannot accurately know its pose information, it cannot comprehensively clean the ground.

[0004] When the working scenario of the floor cleaning robot is complex, or when the floor cleaning robot is picked up or there is a relatively complex dynamic environment, the floor cleaning robot needs to perform precise positioning in such a complex environment, and needs to match the local environment and the global environment information and use particle filtering for positioning, so that the floor cleaning robot can still complete its operation tasks orderly in the complex environment, and solve the problem of inaccurate positioning when the robot is kidnapped or interfered by dynamic obstacles.

[0005] When the floor cleaning robot performs operation tasks in various scenarios, it is necessary to pre-determine an environmental model or provide external information to the robot through sensors, and the initial pose of the robot is unknown. The pose information of the robot in the map is determined through the sensor data. Ordinary robot positioning may obtain a pose information through a particle rate algorithm or inter-frame matching. This single positioning method is not applicable in a complex environment. For example, the particle filter positioning algorithm based on lidar will fail to position in a narrow passage, and the inter-frame matching positioning based on a vision sensor will fail to position due to the influence of the light source.

[0006] Therefore, there is an urgent need for a new positioning method for floor cleaning robots in the prior art to solve the above technical problems. Summary of the Invention

[0007] To solve one of the above technical problems, the present disclosure provides a positioning method for a floor cleaning robot.

[0008] According to one aspect of the present disclosure, there is provided a positioning method for a floor cleaning robot, which includes:

[0009] Construct a global grid map, control the floor cleaning robot to perform operations in the working scenario, and obtain the initial pose information of the floor cleaning robot;

[0010] Obtain the rough global pose of the floor cleaning robot according to the surrounding environment information sensed by the depth camera of the floor cleaning robot, and obtain a positioning particle swarm according to the rough global pose of the floor cleaning robot;

[0011] The displacement increment of the floor cleaning robot is obtained through the odometer and IMU of the floor cleaning robot, and each particle in the positioning particle swarm is updated in motion according to the initial pose information of the floor cleaning robot and the displacement increment of the floor cleaning robot, so as to obtain the prior pose information of each particle in the positioning particle swarm;

[0012] The posterior probability density of each particle in the positioning particle swarm is obtained according to the measurement information of the lidar of the floor cleaning robot;

[0013] According to the posterior probability density, some particles are selected from the positioning particle swarm, and the position of the floor cleaning robot in the global grid map is obtained according to the prior pose information of the selected particles, and the position of the floor cleaning robot in the global grid map is output.

[0014] According to the positioning method of the floor cleaning robot according to at least one embodiment of the present disclosure, obtaining the displacement increment of the floor cleaning robot through the odometer and IMU of the floor cleaning robot includes:

[0015] After the data obtained by the odometer of the floor cleaning robot and the data obtained by the IMU of the floor cleaning robot are fused through an extended Kalman filter, the displacement increment Δx in the X direction, the displacement increment Δy in the y direction, and the angular displacement increment Δθ in the OXY coordinate system are obtained.

[0016] According to the positioning method of the floor cleaning robot according to at least one embodiment of the present disclosure, obtaining the posterior probability density of each particle according to the measurement information of the lidar of the floor cleaning robot includes:

[0017] Constructing a likelihood domain map based on the global grid map, and

[0018] According to the measurement information of the lidar, the posterior probability density of each particle is obtained by querying the likelihood domain map.

[0019] According to the positioning method of the floor cleaning robot according to at least one embodiment of the present disclosure, selecting the prior pose information of some particles from the prior pose information according to the posterior probability density includes:

[0020] Set a posterior probability density threshold. When the posterior probability density of a particle is greater than or equal to the posterior probability density threshold, the particle is selected; when the posterior probability density of a particle is less than the posterior probability density threshold, the particle is discarded.

[0021] According to the positioning method of the floor cleaning robot according to at least one embodiment of the present disclosure, it further includes:

[0022] Resample the selected particles and use the resampled particles as a new positioning particle group; or copy and fill the resampled particles as a basis to obtain particles as a new positioning particle group.

[0023] According to at least one embodiment of the present disclosure, a positioning method for a cleaning robot is provided, wherein a DBOW bag of words is constructed using visual information of the current scene obtained by a depth camera of the cleaning robot. When the robot is kidnapped or the current positioning is unreliable, a local map obtained by the depth camera of the cleaning robot is matched with the DBOW bag of words to obtain the relative pose of the cleaning robot. Based on the global pose of the previous frame and the relative pose of the cleaning robot, a coarse global pose of the cleaning robot in the current frame is obtained.

[0024] The rough global pose of the current frame of the sweeping robot is resampled to obtain a localization particle swarm.

[0025] According to the positioning method of the sweeping robot of at least one embodiment of the present disclosure, when the short-term likelihood estimate w of the particle swarm is slow Greater than the long-term likelihood estimate w fast , or the time f that the infrared receiver in the cliff sensor of the sweeping robot receives the infrared ray is greater than the specified time, it is determined that the sweeping robot has been kidnapped or the current positioning is unreliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings illustrate exemplary embodiments of the present disclosure and together with the description serve to explain the principles of the present disclosure. These drawings are included to provide a further understanding of the present disclosure and are incorporated in and constitute a part of this specification.

[0027] Figure 1 It is a structural diagram of a positioning method for a sweeping robot according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0028] The present disclosure will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only intended to explain the relevant content and are not intended to limit the present disclosure. It should also be noted that, for ease of description, only the portions relevant to the present disclosure are shown in the accompanying drawings.

[0029] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in the present disclosure can be combined with each other. The technical solution of the present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0030] Unless otherwise specified, the illustrated exemplary embodiments are to be understood as providing exemplary features of various details of some ways in which the technical concept of the present disclosure can be implemented in practice. Accordingly, unless otherwise specified, features of the various embodiments can be additionally combined, separated, interchanged, and / or rearranged without departing from the technical concept of the present disclosure.

[0031] In the drawings, cross-hatching and / or shading are generally used to clarify the boundaries between adjacent components. Thus, unless otherwise stated, the presence or absence of cross-hatching or shading does not convey or indicate any preference or requirement regarding the specific materials, material properties, dimensions, proportions, commonality between the components shown, and / or any other characteristics, attributes, properties, etc. of the components. Additionally, in the drawings, for clarity and / or descriptive purposes, the sizes and relative sizes of components may be exaggerated. When the exemplary embodiments can be implemented differently, the specific process orders may be performed in an order different from that described. For example, two consecutively described processes may be performed substantially simultaneously or in an order opposite to that described. Further, the same reference numerals denote the same components.

[0032] When a component is referred to as being "on" or "above" another component, "connected to" or "coupled to" another component, the component can be directly on, directly connected to, or directly coupled to the other component, or an intermediate component may be present. However, when a component is referred to as being "directly on" another component, "directly connected to" or "directly coupled to" another component, no intermediate component is present. For this reason, the term "connected" can refer to physical connection, electrical connection, etc., and can have or not have an intermediate component.

[0033] For descriptive purposes, the present disclosure may use spatial relative terms such as "under", "below", "beneath", "underneath", "above", "on", "over", "upper", and "side (e.g., as in "sidewall")" to describe the relationship of one component to another (other) component as shown in the drawings. In addition to the orientations depicted in the drawings, the spatial relative terms are also intended to encompass different orientations of the device during use, operation, and / or manufacture. For example, if the device in the drawings is flipped, a component described as "under" or "beneath" another component or feature will then be positioned "above" the other component or feature. Thus, the exemplary term "under" can encompass both "above" and "below" orientations. Additionally, the device may be otherwise positioned (e.g., rotated 90 degrees or at other orientations), and accordingly, the spatial relative descriptors used herein are to be interpreted.

[0034] The terms used herein are for the purpose of describing specific embodiments and are not intended to be restrictive. As used herein, unless the context clearly indicates otherwise, the singular forms "one (kind, person)" and "said (the)" are also intended to include plural forms. In addition, when the terms "comprise" and / or "include" and their variations are used in this specification, the features, integral bodies, steps, operations, parts, assemblies and / or their groups stated are explained, but the presence or addition of one or more other features, integral bodies, steps, operations, parts, assemblies and / or their groups is not excluded. It should also be noted that, as used herein, the terms "substantially", "approximately" and other similar terms are used as approximate terms and not as degree terms, so that they are used to explain the inherent deviations of the measured values, calculated values and / or the values provided that will be recognized by those of ordinary skill in the art.

[0035] Figure 1 It is a structural diagram of a positioning method for a sweeping robot according to an embodiment of the present disclosure.

[0036] like Figure 1 As shown, the positioning method of the sweeping robot disclosed in the present invention includes: 102, constructing a global grid map, controlling the sweeping robot to operate in a working scene, and obtaining initial posture information of the sweeping robot; 104, obtaining a rough global posture of the sweeping robot based on the surrounding environment information perceived by the depth camera of the sweeping robot, and obtaining a positioning particle swarm based on the rough global posture of the sweeping robot; 106, obtaining a displacement increment of the sweeping robot through the odometer and IMU of the sweeping robot, performing motion update on each particle in the positioning particle swarm based on the initial posture information of the sweeping robot and the displacement increment of the sweeping robot, and obtaining prior posture information of each particle in the positioning particle swarm; 108, obtaining a posterior probability density of each particle in the positioning particle swarm based on measurement information of the laser radar of the sweeping robot; and 110, selecting some particles from the positioning particle swarm based on the posterior probability density, obtaining the position of the sweeping robot in the global grid map based on the prior posture information of the selected particles, and outputting the position of the sweeping robot in the global grid map.

[0037] The above steps 102 to 110 will be described in detail below.

[0038] In 102, a global grid map is constructed to control the sweeping robot to operate in the working scene, and the initial position information of the sweeping robot is obtained.

[0039] When a sweeping robot is operating in a work scene, it generally builds a global map of the surrounding work scene in advance.

[0040] In one embodiment, the working scenario of the floor cleaning robot can be represented by a global grid map, and the working scenario can be a dynamic and complex scenario.

[0041] Among them, the global grid map is one of the most commonly used representation methods at present. By discretizing the real environment, the working scenario is divided into grids with a certain resolution, and the grid is also called a cell.

[0042] The cell includes position information and information on whether there are obstacles. Among them, the number in [0, 1] can be used to represent whether there are obstacles in the cell. For example, the closer to 1, the more likely there are obstacles, and the closer to 0, the more likely there are no obstacles.

[0043] The initial pose information of the floor cleaning robot can be the [x y] coordinates and the direction angle θ information on the grid map.

[0044] In

[0044] , according to the surrounding environment information sensed by the depth camera of the floor cleaning robot, the rough global pose of the floor cleaning robot is obtained, and according to the rough global pose of the floor cleaning robot, a positioning particle swarm is obtained.

[0045] In the present disclosure, before the floor cleaning robot works, it is necessary to construct a DBOW dictionary bag; for example, the DBOW dictionary bag can be constructed by the visual information of the current scene obtained by the depth camera of the floor cleaning robot.

[0046] Specifically, all the obstacle points in the global grid map are inserted into the KD obstacle tree, and then the global grid map is traversed, and the nearest neighbor obstacle point of the free point is found through the optimized nearest neighbor search. When the nearest neighbor distance (for example, the distance between the free point and the nearest obstacle point) exceeds the maximum distance set by the lidar scan, it is directly replaced with the maximum lidar scan distance.

[0047] The optimized nearest neighbor search has a limit on the search time compared with the original algorithm. When the set time is exceeded, the current approximate nearest distance is directly used as the nearest obstacle point. Secondly, it is sorted according to the distance from the target point to the hyperplane determined by the node. Then, the node with the highest priority is traversed first each time. That is to say, the backtracking check always starts from the tree node with the highest priority. In the offline stage, the floor cleaning robot walks around in the working environment for several laps to collect visual information and pose information, and uses K-means clustering to train the feature information into a DBOW dictionary bag.

[0048] When performing rough global positioning on the floor cleaning robot, the local map constructed by the depth camera of the floor cleaning robot is matched with the DBOW dictionary bag to obtain the relative pose of the floor cleaning robot; according to the global pose of the previous frame and the relative pose of the floor cleaning robot, the rough global pose of the current frame of the floor cleaning robot is obtained.

[0049] Resample the rough global pose of the current frame of the floor cleaning robot to obtain a positioning particle swarm; thereby enabling the positioning particle swarm to be distributed around the floor cleaning robot, which can greatly reduce the number of iterations and the calculation time during the positioning of the floor cleaning robot.

[0050] 106. Obtain the displacement increment of the floor cleaning robot through the odometer and IMU of the floor cleaning robot, and update the motion of each particle in the positioning particle swarm according to the initial pose information of the floor cleaning robot and the displacement increment of the floor cleaning robot to obtain the prior pose information of each particle in the positioning particle swarm.

[0051] In this disclosure, the floor cleaning robot includes an odometer and an IMU (Inertial Measurement Unit). Among them, the position changes in the X and Y directions of the floor cleaning robot can be obtained through the odometer, and the change in the heading angle of the floor cleaning robot can be obtained through the IMU.

[0052] At this time, after the data obtained by the odometer of the floor cleaning robot and the data obtained by the IMU of the floor cleaning robot are fused through an extended Kalman filter, the displacement increment Δx in the X direction, the displacement increment Δy in the y direction, and the angular displacement increment Δθ in the OXY coordinate system are obtained.

[0053] Then, add the above displacement increment to each particle in the positioning particle swarm, so that each particle in the positioning particle swarm simulates the movement of the floor cleaning robot. At this time, the prior pose information of the obtained particles is obtained.

[0054] 108. Obtain the posterior probability density of each particle in the positioning particle swarm according to the measurement information of the lidar of the floor cleaning robot.

[0055] In this disclosure, before the floor cleaning robot actually operates, it is also necessary to construct a likelihood domain map and enable the likelihood domain map to be used offline, thereby reducing the time for optimizing the posterior distribution and improving the positioning efficiency.

[0056] In the likelihood domain map, no longer consider what happens along the entire ray of the laser sensor, but only need to consider its end point.

[0057] First, map the obstacle points detected by the laser sensor to the known map:

[0058]

[0059] Where x t =(x y θ) is the pose state of the floor cleaning robot, (x k,sens y k,sens ) represents the local coordinate position of the installed sensor, θ k,sens represents the deflection angle of the sensor beam relative to the heading angle of the robot, and the measurement end point of the sensor

[0060] Of course, this process requires that the laser sensor actually detects the obstacle, that is, the measured value cannot be the maximum value. If the measured value is the maximum value, the measured value will be discarded.

[0061] After calculating the obstacle point corresponding to the sensor data, find the point closest to the obstacle on the grid map. dist represents the point The distance to the nearest obstacle, so the posterior measurement probability of the sensor can be expressed as a Gaussian function ε with a mean of 0 σhit express:

[0062]

[0063] At this point, the likelihood domain map of the sweeping robot is completed.

[0064] Then, when the sweeping robot is used offline, the posterior probability density of each particle is obtained by querying the likelihood domain map based on the measurement information of the lidar.

[0065] 110. Select some particles from the positioning particle swarm according to the posterior probability density, obtain the position of the sweeping robot in the global grid map according to the prior pose information of the selected particles, and output the position of the sweeping robot in the global grid map.

[0066] In this disclosure, a laser sensor is used to precisely locate particles. Specifically, in the process of obtaining the precise positioning of a sweeping robot, the particles' prior pose information is screened, and particles with low confidence are eliminated. Then, particles with higher confidence are selected from the positioning particle group. The position of the sweeping robot on the global grid map is output through weighted calculation of these particles with higher confidence.

[0067] Specifically, a posterior probability density threshold can be set, and all particles in the positioning particle swarm are traversed through the posterior probability density threshold. When the posterior probability density of a particle is greater than or equal to the posterior probability density threshold, the particle is selected; when the posterior probability density of a particle is less than the posterior probability density threshold, the particle is discarded.

[0068] In actual use, the posterior probability density threshold varies during the calculation process of each measurement update. For example, the posterior probability density threshold may be randomly sampled from a 0-1 uniform distribution.

[0069] In the present disclosure, after calculating the posterior probability density for the particles in each particle cluster, the weights of the particles are normalized, and the selected particles are resampled. Based on these resampled particles, or based on these resampled particles, the particles obtained by replication and filling are used as a new positioning particle swarm, thereby completing the update of the positioning particle swarm.

[0070] When the short-term likelihood estimate w of the particle swarm slow is greater than the long-term likelihood estimate w fast , or the time f when the infrared receiver in the cliff sensor of the sweeping robot receives infrared rays is greater than the specified time, it is determined that the sweeping robot has been kidnapped or the current positioning is not credible.

[0071] Correspondingly, when the sweeping robot has been kidnapped or the current positioning is not credible, it is necessary to perform rough global positioning on the sweeping robot, and based on the rough global positioning information of the sweeping robot, reconstruct the positioning particle swarm.

[0072] For example, the visual information of the current key frame obtained by the depth camera of the sweeping robot can be matched with the DBOW dictionary bag to obtain the dictionary bag vector of the posterior observation frame, and the position of the posterior observation frame is estimated to obtain the relative pose information of this frame relative to the previous frame, and thus obtain the rough global positioning information of the current frame of the sweeping robot, and based on this rough global positioning information, obtain the positioning particle swarm.

[0073] The positioning method of the sweeping robot of the present disclosure can solve the positioning problem of the sweeping robot in a complex dynamic environment. For example, when the robot is kidnapped and under the interference of dynamic obstacles, using the AMCL algorithm for positioning not only consumes a long time but also the positioning may fail.

[0074] In order to ensure that the robot can be accurately positioned in a complex environment and improve the positioning efficiency, the present disclosure collects the visual information of the current map and clusters to construct an offline dictionary bag in the offline stage. When the robot is kidnapped, the visual information of the key frame is matched with the dictionary bag to obtain a relative pose information, and a rough positioning information is obtained according to the relative pose change. Then, the particle filter algorithm is based on the particles obtained by resampling with this rough positioning information and iterates continuously to obtain an accurate positioning.

[0075] When the robot is operating normally, the fusion of the IMU and the odometer is used to improve the prior pose accuracy during motion update, and the pose distribution is optimized during measurement update to improve the posterior positioning accuracy, thereby outputting an accurate pose information.

[0076] Therefore, the positioning method of the sweeping robot disclosed in this disclosure can effectively solve the positioning problem of the sweeping robot in complex environments, and its positioning accuracy can reach within 5 cm and effectively improve the positioning efficiency. This method can be applied to different sweeping robots to solve the positioning problem in complex environments.

[0077] In the description of this specification, the description with reference to the terms "one embodiment / method", "some embodiments / methods", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment / method or example are included in at least one embodiment / method or example of the present application. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment / method or example. Moreover, the specific features, structures, materials or characteristics described may be combined in an appropriate manner in any one or more embodiments / methods or examples. In addition, those skilled in the art may combine and combine different embodiments / methods or examples described in this specification and the features of different embodiments / methods or examples, unless they are contradictory.

[0078] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of such features. Throughout the description of this application, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0079] Those skilled in the art will appreciate that the above embodiments are merely intended to clearly illustrate the present disclosure and are not intended to limit the scope of the present disclosure. Other changes or modifications may be made based on the above disclosure, and such changes or modifications are still within the scope of the present disclosure.

Claims

1. A positioning method for a floor cleaning robot, characterized in that, Including: Construct a global grid map, control the sweeping robot to operate in the working scenario, and obtain the initial pose information of the sweeping robot; Obtain the rough global pose of the sweeping robot according to the surrounding environment information sensed by the depth camera of the sweeping robot, and obtain the positioning particle swarm according to the rough global pose of the sweeping robot; Obtain the displacement increment of the sweeping robot through the odometer and IMU of the sweeping robot, and update the motion of each particle in the positioning particle swarm according to the initial pose information of the sweeping robot and the displacement increment of the sweeping robot, and obtain the prior pose information of each particle in the positioning particle swarm; Obtain the posterior probability density of each particle in the positioning particle swarm according to the measurement information of the lidar of the sweeping robot; Select some particles from the positioning particle swarm according to the posterior probability density, and obtain the position of the sweeping robot in the global grid map according to the prior pose information of the selected particles, and output the position of the sweeping robot in the global grid map; Among them, obtaining the posterior probability density of each particle according to the measurement information of the lidar of the sweeping robot includes: constructing a likelihood domain map based on the global grid map, and obtaining the posterior probability density of each particle by querying the likelihood domain map according to the measurement information of the lidar; Among them, in the likelihood domain map, map the obstacle points detected by the laser sensor to the known map: Among them is the pose state of the floor cleaning robot, represents the local coordinate position of the installed sensor, represents the deflection angle of the sensor beam relative to the robot's heading angle, and the measurement end point of the sensor ; After calculating the obstacle points corresponding to the sensor data, find the point on the grid map that is closest to the obstacle from this point. dist represents the point distance to the closest obstacle. The posterior probability density of the sensor can be represented by a Gaussian function with a mean of 0 as follows: ； Among them, resample the selected particles, and use the resampled particles as the new positioning particle swarm; or based on the resampled particles, copy and fill the obtained particles as the new positioning particle swarm.

2. The positioning method of the floor sweeping robot according to claim 1, characterized in that, Obtaining the displacement increment of the sweeping robot through the odometer and IMU of the sweeping robot includes: After the data obtained by the odometer of the floor cleaning robot and the data obtained by the IMU of the floor cleaning robot are fused through an extended Kalman filter, the displacement increment in the X direction in the OXY coordinate system is obtained , the displacement increment in the y direction and the angular displacement increment .

3. The positioning method of the floor sweeping robot according to claim 1, wherein, Selecting the prior pose information of some particles from the prior pose information according to the posterior probability density includes: Set the posterior probability density threshold. When the posterior probability density of a particle is greater than or equal to the posterior probability density threshold, the particle is selected; when the posterior probability density of a particle is less than the posterior probability density threshold, the particle is discarded.

4. The positioning method of the floor sweeping robot according to claim 1, characterized in that, Construct a DBOW dictionary bag through the visual information of the current scene obtained by the depth camera of the sweeping robot; when the robot is kidnapped or the current positioning is untrustworthy, match the local map obtained by the depth camera of the sweeping robot with the DBOW dictionary bag to obtain the relative pose of the sweeping robot; according to the global pose of the previous frame and the relative pose of the sweeping robot, obtain the rough global pose of the current frame of the sweeping robot; Resample the rough global pose of the current frame of the sweeping robot to obtain the positioning particle swarm.

5. The positioning method of the floor-sweeping robot according to claim 4, characterized in that, When the short-term likelihood estimate of the particle swarm is greater than the long-term likelihood estimate , or the time f when the infrared receiver in the cliff sensor of the floor sweeping robot receives infrared rays is greater than the specified time, it is determined that the floor sweeping robot has been kidnapped or the current positioning is untrustworthy.

Images