Cost map generation method, mobile robot, storage medium and electronic device

Abstract

The present invention discloses a cost map generation method, a mobile robot, a storage medium and an electronic device, and the method comprises the following steps: a. Obtain obstacle information sensed by an upward depth sensor, a downward depth sensor and a laser radar, and pre-process the obstacle information to realize modeling; b. Update the information within the field of view of the upward depth sensor, the downward depth sensor and the laser radar, and retain the information outside the field of view; c. Project the upward depth sensor, the downward depth sensor and the laser radar point cloud into a map, then mark the corresponding grids, and perform a unified expansion process. The present invention improves the accuracy and reliability of obstacle detection by real-time fusion of multiple sensor data, including the upward depth sensor, the downward depth sensor and the laser radar. The present invention effectively manages data and improves computing efficiency by pre-processing and updating information inside and outside the sensor field of view.

Description

Technical Field

[0001] The present invention belongs to the technical field of mobile robots, and the present invention relates to a cost map generation method, a mobile robot, a storage medium and an electronic device. Background Art

[0002] Costmap is a commonly used perception technology in fields such as autonomous driving and robotics. This technology simply projects obstacles detected by sensors onto a map to achieve the purpose of environmental representation. Existing mobile robots are usually equipped with multiple sensors for environmental perception. For example, the costmap generation method involved in Patent Document 1 (Announcement No.: CN115638800A) obtains point cloud data of obstacles through environmental perception sensors during the operation of the robot; updates the environmental map of the robot's work site based on the accumulated results of the point cloud data; generates an initial costmap based on the point cloud data; and fuses the environmental map and the initial costmap to obtain a target costmap. Patent Document 2 (Announcement No.: CN115542886A) hierarchical costmap generation and fusion method, system, terminal and mobile robot first receive information flow about obstacles during the robot's movement in real time and obtain the robot's posture; discretize and probabilize the sensor information corresponding to each sensor to obtain a two-dimensional discrete probability map corresponding to each sensor.

[0003] However, Patent Documents 1 and 2 have problems of low efficiency and limited processing capabilities in terms of information fusion and obstacle modeling. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a new sensor fusion cost map generation method based on multimodal perception, a mobile robot, a storage medium and an electronic device.

[0005] The technical solution to achieve the purpose of the present invention is:

[0006] The cost map generation method is characterized by comprising the following steps:

[0007] a. Obtain obstacle information perceived by the upward-looking depth sensor, downward-looking depth sensor, and lidar, and pre-process the obstacle information to achieve modeling;

[0008] b. Update the information within the field of view of the upward depth sensor, downward depth sensor, and lidar, and retain the information outside the field of view;

[0009] The information within the field of view of the laser radar is updated using a wire harness model, the information within the field of view of the upward depth sensor is updated using a cone model, and the information within the field of view of the downward depth sensor is updated using a polygonal model.

[0010] c. Project the upward-looking depth sensor, downward-looking depth sensor, and lidar point cloud into the map, then mark the corresponding grids and perform uniform expansion.

[0011] Preferably, in step a, the obstacle information of the upward-view depth sensor is converted into point cloud information through the depth image, and then the upward-view point cloud information is stabilized in the grid and filtered in space in the form of integration in the odometer coordinate system.

[0012] Preferably, in step a, the obstacle information of the downward-looking depth sensor is converted into point cloud information through the depth image, and the ground is extracted using the region growing method to obtain the obstacle information without the ground.

[0013] Preferably, in step a, the laser radar information is converted into a point cloud by removing isolated points, and filtering is used to remove noise points and thin the point cloud.

[0014] Preferably, in step c, a map is established including a static layer, various types of sensor layers and a composite obstacle layer, and the map is updated.

[0015] A mobile robot comprises one or more sensors for executing the above-mentioned cost map generation method.

[0016] A computer-readable storage medium stores a computer program, which is configured to execute the cost map generation method.

[0017] An electronic device comprises a processor and a memory, wherein a computer program is stored in the memory, and the program is configured to execute the cost map generation method.

[0018] A mobile robot system comprises a plurality of sensors and a central processing unit, wherein the central processing unit is configured to execute the cost map generation method.

[0019] An obstacle detection system for a mobile robot comprises one or more sensors and a central processing unit, wherein the central processing unit is configured to execute the cost map generation method.

[0020] Compared with the prior art, the present invention has the following significant advantages:

[0021] (1) Multimodal sensing sensors:

[0022] The present invention improves the accuracy and reliability of obstacle detection by fusing multiple sensor data in real time, including an upward-looking depth sensor, a downward-looking depth sensor and a laser radar.

[0023] (2) Cost map evaluation:

[0024] The cost map evaluation diagram of the present invention can be updated in real time to reflect the current state of the robot's surrounding environment and provide the robot with accurate obstacle information.

[0025] (3) Efficient information processing:

[0026] The present invention effectively manages data and improves computing efficiency by preprocessing and updating information inside and outside the sensor's field of view.

[0027] (4) Accurate model updating:

[0028] The present invention uses a wire harness model, a cone model and a polygonal model to update the laser radar and depth sensor data respectively, thereby ensuring the accuracy of obstacle modeling.

[0029] (5) Optimized map creation and updating:

[0030] The present invention provides a structured method for maintaining and updating map information by establishing and updating a static layer, a sensor layer and a comprehensive obstacle layer.

[0031] (6) Reduce the amount of calculation:

[0032] The present invention integrates cost map preprocessing, including distance discretization and obstacle score preprocessing, which greatly reduces the computational complexity of map processing.

[0033] (7) Strong adaptability:

[0034] The present invention is applicable to different environments and can maintain high performance even when the sensor field of view is limited or blocked.

[0035] (8) Real-time:

[0036] The method of the present invention can process sensor data in real time and update the obstacle assessment map, meeting the real-time requirements of the mobile robot.

[0037] (9) Robustness:

[0038] The present invention generates samples and performs scoring in an iterative manner, and the present invention can resist initial value interference and ensure the stability of posture tracking.

[0039] (10) Innovative sensor data processing:

[0040] The present invention improves the accuracy of obstacle detection by separately processing the internal and external information of the visual field of the upward-looking sensor and the downward-looking sensor and the ray tracing method of the laser radar.

[0041] (11) Flexible system configuration:

[0042] The present invention can be applied to a variety of mobile robotic systems, including robots equipped with different types of sensors and central processing units.

[0043] (12) Easy to integrate:

[0044] The present invention can be stored on a computer-readable storage medium and run on an electronic device, so as to be easily integrated into an existing or new robot system.

[0045] (13) Closed-loop update mechanism:

[0046] The present invention realizes a closed loop of map updating by resetting the static layer when updating sensor information at the next moment, thereby ensuring the continuous accuracy of map information. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a flow chart of the cost map generation method of the present invention;

[0048] Figure 2 A schematic diagram of the downward-looking depth sensor field of view and points within and outside the sensor field of view;

[0049] Figure 3 This is a schematic diagram of ray tracing within a grid map;

[0050] Figure 4 A decay function for the obstacle range affecting the obstacle score;

[0051] Figure 5 Update flow diagrams for each map layer. DETAILED DESCRIPTION

[0052] Example 1

[0053] like Figure 1 As shown, the cost map generation method of the present invention comprises the following steps:

[0054] a. Obtain obstacle information perceived by the upward-looking depth sensor, downward-looking depth sensor, and lidar, and pre-process the obstacle information to achieve modeling;

[0055] b. Update the information within the field of view of the upward depth sensor, downward depth sensor, and lidar, and retain the information outside the field of view;

[0056] The information within the field of view of the laser radar is updated using a wire harness model, the information within the field of view of the upward depth sensor is updated using a cone model, and the information within the field of view of the downward depth sensor is updated using a polygonal model.

[0057] c. Project the upward-looking depth sensor, downward-looking depth sensor, and lidar point cloud into the map, then mark the corresponding grids and perform uniform expansion.

[0058] Example 2

[0059] In step a, real-time multi-sensor information processing includes lidar, upward camera and downward camera processing. Lidar information processing includes removing isolated points and point cloud conversion. The upward camera stabilizes the point cloud by multi-frame comparison, and the downward camera uses the region growing method to extract the ground to obtain the obstacle information after removing the ground. The following section will process the information of each sensor separately.

[0060] Upward depth sensor information preprocessing:

[0061] The upward depth sensor information is converted into point cloud information through the depth image. Since global positioning may jump, the sensor information is processed in the odometer coordinate system. In the robot odometer coordinate system, the upward point cloud information is stabilized in the grid and filtered in space through integration.

[0062] In the depth sensor coordinate system, a pixel point in the depth map is converted to a point in the depth sensor coordinate system. for:

[0063] (1)

[0064] In the formula, , , are the coordinates of point n on the α-axis, β-axis, and γ-axis in the depth sensor coordinate system, is the length of the image along the α axis, is the length of the image along the β axis, is the coordinate value of pixel point n on the α axis of the image, is the coordinate value of pixel point n on the β axis of the image, is the depth of pixel n, , are the camera internal parameters respectively.

[0065] Current moment All pixels in the image The set in the upward camera coordinate system is , the current depth image point cloud set is converted to the point cloud set in the odometer coordinate system for:

[0066] (2)

[0067] in, is the coordinate matrix, , , are the coordinates of the α-axis, β-axis, and γ-axis in the odometer coordinate system, are matrix coefficients.

[0068] Last moment The depth image point cloud set is converted to the point cloud set in the odometer coordinate system as By analyzing and The point cloud in the odometer coordinate system is used to stabilize the information. A three-dimensional grid map is established in the odometer coordinate system with a resolution of , the origin is , the three-dimensional grid occupied by each point It can be expressed as:

[0069] (3)

[0070] To round up, , , It is the value of each point on the α, β, and γ axes in the mileage coordinate system.

[0071] and The rasterized set is and , through and Compare the inner elements to obtain the common set of elements Other elements may contain noise, so they are not included in the map processing for the time being. The collection is compared and added to the map at the next moment. The effective point set for:

[0072] (4)

[0073] To round up, , , It is the opposite number of the value of each point on the α, β, and γ axes in the mileage coordinate system. is the resolution of the raster map. , , are the values ​​on the origin α, β, and γ axes, is the coordinate of the conjugate point of the point in the sensor coordinate system, so at the current moment Point cloud collection of the upward camera point cloud in the global coordinate system for:

[0074] (5)

[0075] are the coordinates of each point in the global coordinate system, , , is the value of each point on the α, β, and γ axes in the global coordinate system, is the coefficient matrix in the global coordinate system, The set of valid points.

[0076] Preprocessing of downward depth sensor information:

[0077] The downward-looking depth sensor information is converted into point cloud information through the depth image, and the sensor information is also processed in the odometer coordinate system. Through formula (1), we can get a pixel point in the depth image converted to the point in the depth sensor coordinate system: , the point cloud set is , the ground point cloud set is obtained by region growing method , so the ground obstacle point cloud set is removed for:

[0078] (6)

[0079] In formula (6), is the position of the robot in the global coordinate system, is the position of the downward-looking camera in the robot, is a collection of point clouds without ground obstacles in global coordinates. , , It is the value of each point on the α-axis, β-axis, and γ-axis in the global coordinate system after removing the ground obstacle point cloud.

[0080] LiDAR information preprocessing:

[0081] The laser radar information contains distance information and angle information, so the position of the laser radar end point in the global map can be calculated. The coordinates of the laser point on the laser sensor for:

[0082] (7)

[0083] In formula (7), For the The distance measured by the laser beam, For the The rotation angle of the laser beam relative to the origin of the laser radar. The isolated points are filtered out by establishing a grid map in the laser sensor coordinate system. The origin of the laser grid map is , the resolution is , the coordinate value of the laser point in the map grid is :

[0084] (8)

[0085] In formula (8), The function is a ceiling function. , For the The values ​​of the laser beam on the α-axis and β-axis.

[0086] After projecting all laser points onto the map, search Nearby grids, specifically top, bottom, left, right, upper left, lower left, upper right, and lower right grids. If none of these grids are occupied, then occupy The laser point is an isolated point and will be deleted in subsequent processing.

[0087] Convert the filtered laser point cloud into a collection of point clouds in the world coordinate system through coordinate conversion :

[0088] (9)

[0089] In formula (9) is the transformation from the robot coordinate system to the world coordinate system, is the transformation from the laser radar coordinate system to the robot coordinate system, is the set of lidar points in the sensor coordinate system after removing isolated points. is the set of laser points in the world coordinate system. , is the value of the laser point on the α-axis and β-axis in the world coordinate system. So far, the laser radar point cloud information in the world coordinate system has been obtained.

[0090] Example 3

[0091] Step b includes sensor fusion cost map preprocessing and sensor information update. Each step will be described in detail below.

[0092] The fusion cost map preprocessing includes distance discretization, obstacle scoring, and static map preprocessing. Preprocessing the fusion cost map can significantly reduce the amount of map processing computation.

[0093] The fused cost map is represented as a grid with the origin at , the map width is , the height is , the resolution is Meters / grid, and the distance affected by obstacles sensed by each sensor in the global coordinate system is also finite, so the map distance that needs to be processed is also the maximum grid distance of discretization. The distance set to be processed for:

[0094] (10)

[0095] In formula (10), and are the number of grids in the X-axis and Y-axis directions from the grid affected by the obstacle to the source grid. The corresponding distance And it is the only corresponding Represents the maximum distance among all points in the map coordinate system.

[0096] After preprocessing the distances affected by obstacles, it is also necessary to preprocess the obstacle scores of these distances. The closer the distance to the obstacle, the higher the cost of passing. The closer the distance to the obstacle point is to the robot, the higher the cost of passing. When , the obstacle score of this area is the highest, indicating that this area is impassable. The attenuation function of the obstacle range affecting the obstacle score is as follows: Figure 4 Shown: From Figure 4 It can be analyzed that the closer to the obstacle center, the higher the obstacle score, and vice versa. The grid obstacle score of is the highest, indicating that this area intersects with the robot body, so the robot is prohibited from passing through this area. Distance from the obstacle center Within the range, the obstacle score gradually decreases until it reaches 0. Therefore, the preprocessed obstacle score set for:

[0097] (11)

[0098] ， is the value of the obstacle on the α-axis and β-axis in the coordinate system, sr is the resolution, is the maximum value of the coordinate system.

[0099] Since the static map of the sensor fusion cost map is not updated, its expansion can be preprocessed. In the initial state, there are only occupied grids and unoccupied grids in the static map. The global map grid set is each grid coordinate is , the occupied grid constitutes a set , assign values ​​to occupied and unoccupied grids:

[0100] (12)

[0101] in, Indicates the extraction of map raster. Represents the assignment of values ​​to a static map grid.

[0102] Sensor information update: Sensor information update involves the information of upward-looking sensors, downward-looking sensors, and lidar. The update of sensor information involves marking and retaining data in the grid map. The marking and retention of sensor information of upward-looking sensors, downward-looking sensors, and lidar involves the corresponding sensor models, while retaining the distance threshold from the robot. The information in the robot’s global coordinate system is .

[0103] The upward sensor uses the cone model to back-project the spatial point cloud back to the camera coordinate system to determine whether it is in the depth image, so as to determine whether the point is in the sensor's field of view. The field of view of the downward sensor is projected as a trapezoid inside, so whether the detection point is within the trapezoid range can determine whether it is in the field of view of the downward sensor. LiDAR uses ray tracing to clear obstacles.

[0104] Example 4

[0105] Project the upward-looking depth sensor, downward-looking depth sensor, and lidar point cloud into the map, then mark the corresponding grids and perform uniform expansion.

[0106] Upward depth sensor point cloud projection:

[0107] At the current moment, the upward sensor point cloud information is , the cached point cloud information of the upward sensor is Before adding the cached information to the map, it is necessary to judge it, retain the information outside the field of view, and remove the information within the field of view. The position of the upward sensor in the global coordinate system is Therefore, the set in the sensor coordinate system for:

[0108] (13)

[0109] Depth Image The number of pixels in the axial direction is , The number of pixels in the axial direction is . , the coordinate values ​​in the image It can be calculated as:

[0110] (14)

[0111] In formula (14), and is the internal parameter of the depth camera. and illustrate In the depth image, that is, In the field of view of the upward camera. Remove all points within the field of view and obtain the point cloud set outside the field of view that needs to be retained The collection of cached sensor point clouds to be processed for:

[0112] (15)

[0113] is any point in the field of view, so for the upward-looking sensor, the point cloud set to be processed is .

[0114] Downward-looking depth sensor point cloud projection:

[0115] At the current moment, the downward sensor point cloud information is , the cached point cloud information of the downward sensor is Before adding the cached information to the map, it is necessary to judge it, retain the information outside the field of view, and remove the information within the field of view. Since the point cloud of the downward-looking sensor is projected as a trapezoid on the ground, it is possible to determine whether a point in the point cloud is within the range of the trapezoid to determine whether it is within the field of view. According to the polygon theorem, if the sum of the areas of the triangles formed by connecting a point with each side of the trapezoid is greater than the area of ​​the trapezoid, it means that the point is outside the trapezoid. Otherwise, it is inside the trapezoid. Figure 2 shown.

[0116] Trapezoidal area formed by downward-looking sensors For triangle area With triangle Area sum:

[0117] (16)

[0118] In formula (16), It means to find the absolute value of the determinant. Figure 2 In the figure, the sum of the areas of the triangles formed by the points in the bottom view point cloud and the edges is, , , , , , , , are the coordinates of A, B, C, and D in the plane coordinate system

[0119] (17)

[0120] In formula (17), for The point and the sides of the trapezoid form the sum of the areas of the triangle. for:

[0121] (18)

[0122] , , It is the value of point o on the x-axis, y-axis, and z-axis in the three-dimensional coordinate system.

[0123] The collection of cached sensor point clouds to be processed for:

[0124] (19)

[0125] is any point in the field of view, so for the downward-looking sensor, the point cloud set to be processed is .

[0126] LiDAR point cloud projection:

[0127] At this moment, the current point cloud information of the laser radar is , the cache information is The cache information is projected into the map grid, and the projected grid collection for:

[0128] (20)

[0129] In the formula , is the coordinate of point g, , is the origin coordinate, , is the coordinate of the laser point in the coordinate system, is a constant and the position of the laser radar in the global , using the current point cloud collection Ray tracing is performed for each point in the Figure 3 shown.

[0130] The grid that the ray trace passes through is passed by the current point cloud, so there is no obstacle. The corresponding cached point cloud needs to be removed from the collection. Remove from the list and get a new set to be processed . Therefore, the point cloud set that the laser radar needs to process is: .

[0131] So far, we have obtained the point cloud set that needs to be added to the map processing at the current moment , the points in the point cloud set are placed in the composite obstacle layer for processing. The resolution and number of grids of the composite obstacle layer are exactly the same as those of the static layer. The initial obstacle score of each grid is 0, and the obstacle score of each grid is updated according to the obstacle score affected by the sensor. Medium Element , which constitutes the current moment Occupancy grid collection for:

[0132] (twenty one)

[0133] Similar to static map preprocessing, , the set of grids it affects for:

[0134] (twenty two)

[0135] Iterate over a raster collection For all elements in , calculate their obstacle scores in the composite obstacle layer.

[0136] (twenty three)

[0137] In formula (23), To obtain the obstacle score for a grid in the composite obstacle layer, , To iterate over a raster collection The coordinates of all elements in on the α-axis and β-axis, , is the coordinate of each point in the grid set on the α, β axis. The grid-obstacle score pair for each updated obstacle score in the composite obstacle layer can be obtained by formula (23): . Iterate over the collection , and find the barrier score of the corresponding raster in the static layer to form a static raster-barrier score pair: .if , then Add to static update collection Middle and rear general Assign to the static layer. When updating sensor information, traverse Elements in , reset the static layer, thus achieving an updated closed loop, such as Figure 5 shown.

[0138] Example 5

[0139] This embodiment proposes a mobile robot, which includes one or more sensors and is used to execute the cost map generation method described in Embodiments 1-4.

[0140] Example 6

[0141] This embodiment provides a computer-readable storage medium on which a computer program is stored. The program is configured to execute the cost map generation method described in Embodiments 1-4.

[0142] Example 7

[0143] This embodiment proposes an electronic device, including a processor and a memory, wherein the memory stores a computer program, and the program is configured to execute the cost map generation method described in Embodiments 1-4.

[0144] Example 8

[0145] This embodiment proposes a mobile robot system, including multiple sensors and a central processing unit, wherein the central processing unit is configured to execute the cost map generation method described in embodiments 1-4.

[0146] Example 9

[0147] This embodiment proposes an obstacle detection system for a mobile robot, including one or more sensors and a central processing unit, wherein the central processing unit is configured to execute the cost map generation method described in embodiments 1-4.

[0148] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. The information fusion cost map generation method is characterized by: The following steps are involved: a. Obtain obstacle information perceived by the upward-looking depth sensor, downward-looking depth sensor, and lidar, and pre-process the obstacle information to achieve modeling; b. Update the information within the field of view of the upward depth sensor, downward depth sensor, and lidar, and retain the information outside the field of view; The information within the field of view of the laser radar is updated using a wire harness model, the information within the field of view of the upward depth sensor is updated using a cone model, and the information within the field of view of the downward depth sensor is updated using a polygonal model. c. Project the upward-looking depth sensor, downward-looking depth sensor, and lidar point cloud into the map, then mark the corresponding grids and perform unified expansion processing; Upward-looking depth sensor point cloud projection includes: The upward depth sensor point cloud information at the current moment is obtained as , the cached point cloud information of the upward depth sensor is , before adding cached information to the map, it is necessary to judge it, retain the information outside the field of view, and remove the information within the field of view; Get the set in the sensor coordinate system : ； Depth Image The number of pixels in the axial direction is , The number of pixels in the axial direction is , is the position and posture of the upward-looking sensor in the global coordinate system, , the coordinate values ​​in the image for: ； In the formula and is the internal parameter of the depth camera, and Explanation points In the depth image, that is, Within the field of view of the upward camera; In the collection Remove all points within the field of view and obtain the point cloud set outside the field of view that needs to be retained , get the cached sensor point cloud set to be processed ; For the upward-looking depth sensor, the point cloud set to be processed ; Downward-looking depth sensor point cloud projection includes: The point cloud information of the downward depth sensor at the current moment is obtained as , the cached point cloud information of the downward depth sensor is , before adding cached information to the map, it is necessary to judge it, retain the information outside the field of view, and remove the information within the field of view; By judging whether a point in the point cloud is within the range of the trapezoid, it is determined whether it is within the field of view, and the downward depth sensor point cloud set outside the field of view is obtained. : ； , , is the value of point o on the x-axis, y-axis, and z-axis in the three-dimensional coordinate system. for The sum of the areas of the triangles formed by the point and the sides of the trapezoid, The trapezoidal area formed by the downward-looking depth sensor; Get the cached sensor point cloud set to be processed : ； For any point in the field of view, is the distance threshold; For the downward-looking depth sensor, the point cloud set to be processed ; LiDAR point cloud projection includes: At the current moment, the current point cloud information of the laser radar is , the cache information is , the cache information is projected into the map grid, and the projected grid set is obtained ; Use Current Point Cloud Collection Perform ray tracing for each point in the grid, and extract the corresponding cached point cloud from the collection Remove from the list and get a new set to be processed ; The point cloud set to be processed by the laser radar is: ; Get the point cloud set that needs to be added to the map processing at the current moment ; The points in the point cloud set that need to be added to the map processing are placed in the composite obstacle layer for processing. The resolution and number of grids of the composite obstacle layer are exactly the same as those of the static layer. The initial obstacle score of each grid is 0, and the obstacle score of each grid is updated according to the obstacle score affected by the sensor.

2. The information fusion cost map generation method according to claim 1, characterized in that: In step a, the obstacle information of the upward-view depth sensor is converted into point cloud information through the depth image, and then the upward-view point cloud information is stabilized in the grid and filtered in space in the form of integration in the odometer coordinate system.

3. The information fusion cost map generation method according to claim 1, characterized in that: In step a, the obstacle information of the downward-looking depth sensor is converted into point cloud information through the depth image, and the ground is extracted using the region growing method to obtain the obstacle information without the ground.

4. The information fusion cost map generation method according to claim 1, characterized in that: In step a, the lidar information is converted into a point cloud by removing isolated points, and filtering is used to remove noise and thin the point cloud.

5. The information fusion cost map generation method according to claim 1, characterized in that: In step c, a map is established including a static layer, various types of sensor layers, and a composite obstacle layer, and the map is updated.

6. A mobile robot based on the method according to any one of claims 1 to 5, characterized in that: The robot comprises one or more sensors for performing the method.

7. A computer-readable storage medium having a computer program stored thereon, wherein the program is configured to execute the method according to any one of claims 1 to 5.

8. An electronic device, characterized in that: The invention comprises a processor and a memory, wherein a computer program is stored in the memory, and the program is configured to execute the method according to any one of claims 1 to 5.

9. A mobile robot system, characterized in that: The system comprises a plurality of sensors and a central processing unit, wherein the central processing unit is configured to execute the method according to any one of claims 1 to 5.

10. An obstacle detection system for a mobile robot, characterized in that: The device comprises one or more sensors and a central processing unit, wherein the central processing unit is configured to execute the method according to any one of claims 1 to 5.

Images