Map creation system, method, and program

Abstract

To create a two dimensional map for a robot capable of autonomous traveling.SOLUTION: An input unit configured to input three dimensional point cloud data of a surrounding environment and three dimensional position data of an autonomous mobile robot in an odometry coordinate system, the three dimensional point cloud data and the three dimensional position data being obtained when the autonomous mobile robot moves; A map creation system comprising: a topology construction unit that constructs a topological structure for each of different attributes; a cluster construction unit that constructs a three dimensional travelable cluster from the topological structure using attribute information related to traveling; and a mapping unit that performs mapping on a two dimensional map using the three dimensional travelable cluster.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] An embodiment of the present invention relates to a technique for creating a map for an autonomously moving robot. [Background technology]

[0002] Many autonomously moving robots are equipped with sensors such as laser range finders and are able to detect the surrounding environment, such as structures. Patent Document 1 discloses a system that creates a two-dimensional map that takes into account the height of the cargo being transported, for automatic transport robots used in warehouses where goods are stored.

[0003] On the other hand, Non-Patent Document 1 discloses that spatial information is learned using GNG-DT (Growing Neural Gas with Different Topologies), which constructs a topological structure according to each attribute of an input vector composed of feature quantities with two attributes, spatial information and color information, of each object in image data. This eliminates the need for prior knowledge of the attributes of each object. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2023-023659 [Non-patent literature]

[0005] [Non-Patent Document 1] Yuichiro Toda and three others, "Learning spatial structure based on Growing Neural Gas in feature vectors composed of multiple attributes," Transactions of the Society of Instrument and Control Engineers, Vol. 57, No. 4, pp. 209 / 218 (2021) Summary of the Invention [Problem to be solved by the invention]

[0006] However, Patent Document 1 has a problem in that there is little information that can be taken into consideration when making an obstacle determination.

[0007] Furthermore, Non-Patent Document 1 has a problem in that it must be applied to three-dimensional point cloud data and its effectiveness must be verified in a form that also includes attribute data relating to the shape of an object.

[0008] The present invention was completed through extensive research focused on these issues, and its purpose is to create a two-dimensional map for autonomously moving robots. [Means for solving the problem]

[0009] In order to solve the above problems, the present invention provides a map creation system that includes an input unit that inputs three-dimensional point cloud data of the surrounding environment acquired when the autonomous driving robot is driving and three-dimensional position data of the autonomous driving robot in an odometry coordinate system, a conversion unit that converts the three-dimensional point cloud data into data in the odometry coordinate system that is the same as the three-dimensional position data, a topology construction unit that constructs a topological structure for each different attribute using the three-dimensional point cloud data and three-dimensional position data in the odometry coordinate system, a cluster construction unit that constructs a three-dimensional drivability cluster from the topological structure using attribute information related to driving, and a mapping unit that maps the three-dimensional drivability cluster onto a two-dimensional map using the three-dimensional drivability cluster.

[0010] Another invention is a map creation method comprising: an input step of inputting three-dimensional point cloud data of the surrounding environment acquired while the autonomous mobile robot is traveling, and three-dimensional position data of the autonomous mobile robot in an odometry coordinate system; a conversion step of converting the three-dimensional point cloud data into data in the odometry coordinate system that is the same as the three-dimensional position data; a topology construction step of constructing a topological structure for each different attribute using the three-dimensional point cloud data and three-dimensional position data in the odometry coordinate system; a cluster construction step of constructing three-dimensional drivability clusters from the topological structure using attribute information related to traveling; and a mapping step of mapping the three-dimensional drivability clusters onto a two-dimensional map.

[0011] Another invention is a map creation program executable by a computer, comprising: an input step for inputting three-dimensional point cloud data of the surrounding environment acquired when the autonomous mobile robot is traveling, and three-dimensional position data of the autonomous mobile robot in an odometry coordinate system; a conversion step for converting the three-dimensional point cloud data into data in the odometry coordinate system that is the same as the three-dimensional position data; a topology construction step for constructing a topological structure for each different attribute using the three-dimensional point cloud data and three-dimensional position data in the odometry coordinate system; a cluster construction step for constructing three-dimensional drivability clusters from the topological structure using attribute information related to traveling; and a mapping step for mapping the three-dimensional drivability clusters onto a two-dimensional map. [Effects of the Invention]

[0012] According to the present invention, a two-dimensional map can be created for an autonomously moving robot. [Brief explanation of the drawings]

[0013] [Figure 1] 1 is a functional block diagram of a map creation system 100 according to an embodiment of the present invention. [Figure 2] 1 is a schematic diagram showing the appearance of an autonomous mobile robot 200 according to the present embodiment. [Figure 3] 10 is a flowchart illustrating mapping of three-dimensional point cloud data onto two-dimensional map data according to the present embodiment. [Figure 4] 10 is a flowchart for converting data in a LiDAR coordinate system into data in an odometry coordinate system according to the present embodiment. [Figure 5] 1A to 1C are diagrams for explaining various coordinate systems according to the present embodiment. [Figure 6] 10 is a flowchart for constructing a travel possibility cluster from a topological structure having multiple attribute information according to this embodiment. [Figure 7] 5 is a flowchart illustrating a process of mapping data in the odometry coordinate system onto two-dimensional map data according to the present embodiment. [Figure 8] 1A and 1B are schematic diagrams showing an example of a three-dimensional environment and diagrams for explaining corresponding two-dimensional map data. [Figure 9] 1A to 1C are schematic diagrams of a three-dimensional environment when an empty box according to the present embodiment is placed, and diagrams for explaining two-dimensional map data in which the empty box is reflected. [Figure 10] FIG. 10 is a functional block diagram of a map creation system 100' according to another embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0014] An embodiment of the present invention will be described with reference to the drawings, assuming that the map creation system according to the present embodiment is an online processing system.

[0015] (Map creation system according to this embodiment) 1 is a functional block diagram of a map creation system 100 according to an embodiment of the present invention. The map creation system 100 includes an autonomous mobile robot 200, an information processing device 300, and a two-dimensional map database 400.

[0016] The autonomous mobile robot 200 includes a traveling unit 210 having a mechanism for traveling, a data input unit 220 that inputs the surrounding environment of the robot 200 as three-dimensional point cloud data and inputs three-dimensional position data in an odometry coordinate system while the robot 200 is traveling, a point cloud data conversion unit 230 that converts the three-dimensional point cloud data into data in the same odometry coordinate system as the three-dimensional position data, a control unit 240 that controls the entire robot 200, and a communication unit 250 that can communicate with an information processing device 300. The information processing device 300 includes a point cloud data processing unit 310 that processes the point cloud data converted into the odometry coordinate system, a mapping unit 320 that maps the processed point cloud data onto two-dimensional map data stored in a two-dimensional map database 400, a control unit 330 that controls the entire device 300, and a communication unit 340 that can communicate with the robot 200 and the database 400.

[0017] In the case of online processing, the device 300 and database 400 are mounted on the robot 200. Therefore, the control units 240 and 330 can be implemented in a single CPU. The communication units 240 and 250 communicate information between the robot 200, the device 300, and the database 400 via signal cables or the like. Note that, depending on the data sampling frequency, data transfer speed, and offline processing, information communication is not limited to signal cables and may be performed via a wired or wireless local area network or the Internet.

[0018] FIG. 2 is a schematic diagram showing the appearance of the autonomous mobile robot 200 according to this embodiment. The autonomous mobile robot 200 is a roughly rectangular parallelepiped with a length of 90 cm, a width of 60 cm, and a height of 120 cm. A battery for driving the running unit 210 is built in the robot. The running unit 210 has six wheels, but is not limited to this and may have four wheels or other mechanisms that allow it to travel indoors or outdoors. Furthermore, because the robot 200 is battery-powered, when the device 300 is also mounted on the robot 200 and online processing is performed, the point cloud data processing unit 310 needs to perform real-time processing (approximately 10 Hz) using a low-power CPU rather than a high-power GPU.

[0019] On the front of the autonomous mobile robot 200, a Livox MID-360 is installed at a height of 70 cm from the ground as a data input unit 220. The MID-360 is equipped with a 3D-LiDAR sensor and an IMU (inertial measurement unit). The 3D-LiDAR sensor can acquire the entire 360-degree surrounding environment in front of the autonomous mobile robot 200 as three-dimensional point cloud data. The IMU can detect three-dimensional inertial motion (translational motion and rotational motion in three orthogonal axial directions) and can acquire position data of the installation location on the front of the autonomous mobile robot 200.

[0020] An RGB-D camera may be installed in the same location as the MID-360 as an additional function of the data input unit 220. The RGB-D camera is a depth camera that outputs both color (RGB) and depth (D) data in real time. By installing an RGB-D camera, the data input unit 220 can acquire 3D point cloud data with color information.

[0021] 3 is a flowchart for mapping 3D point cloud data onto 2D map data according to this embodiment. The data input unit 220 inputs 3D point cloud data during driving, and the point cloud data conversion unit 230 converts the 3D point cloud data into an odometry coordinate system (S100). The point cloud data processing unit 310 performs GNG-DT processing on the 3D point cloud data converted by the point cloud data conversion unit 230 (S200). The mapping unit 320 maps the data after GNG-DT processing onto a 2D map (S300). In this embodiment, an occupancy grid map is used as the 2D map. The occupancy grid map is a map in which a 2D environment is divided into a grid and whether or not each cell is occupied (i.e., whether or not there is an obstacle) is expressed in terms of occupancy probability.

[0022] 4 is a flowchart for converting data in the LiDAR coordinate system into data in the odometry coordinate system according to this embodiment. This corresponds to the step of converting 3D point cloud data, which is part of S100 in FIG. 3. In this embodiment, the point cloud data conversion unit 230 unifies the coordinate system of the input point cloud using odometry information as preprocessing for GNG-DT.

[0023] FIG. 5 is a diagram for explaining various coordinate systems according to this embodiment. The relationship between the map coordinate system, the odometry coordinate system, and the LiDAR coordinate system will be explained. A two-dimensional map can represent on a plane the surrounding environment in which the autonomous mobile robot 200 travels. The map coordinate system is a coordinate system based on the origin of the two-dimensional map.

[0024] Odometry is generally a technology that calculates the rotation angles of the wheels and steering of an autonomous mobile robot, determines the amount of movement of each, and estimates the robot's current position from the cumulative calculation. This type of estimation technology is also called self-position estimation. In this embodiment, an IMU (corresponding to the position data acquisition unit 230) installed in front of the autonomous mobile robot 200 acquires current position data using the acceleration and angular acceleration of three-dimensional inertial motion. The position data includes not only position information about the robot but also attitude information. The odometry coordinate system is a three-dimensional coordinate system with the origin at the position where odometry calculation began (the position where the robot was started). The robot's position data at each time t can then be plotted on a map coordinate system.

[0025] The 3D-LiDAR sensor is a technology that irradiates light three-dimensionally and detects the shape of a location as 3D point cloud data from the distance around the sensor. In this embodiment, the 3D-LiDAR sensor (corresponding to the point cloud data acquisition unit 220) installed in front of the autonomous mobile robot 200 acquires 3D point cloud data of the entire 360-degree periphery in front of the robot 200. The LiDAR coordinate system is a coordinate system that marks the 3D point cloud data at each time t.

[0026] Returning to the description of the flowchart in Fig. 4, in this embodiment, the point cloud data conversion unit 230 executes each processing step in Fig. 4. In S110, the process waits for 3D-LiDAR data (X(t)) in the LiDAR coordinate system to be input from the data input unit 220 as three-dimensional point cloud data around the robot 200. In S120, the process waits for odometry data (O(t)) in the odometry coordinate system to be input from the data input unit 220 as position data of the robot 200 (S120). In S130, X(t) is converted into the odometry coordinate system to obtain 3D-LiDAR data (Y(t)) in the odometry coordinate system (S130). This allows both O(t) and Y(t), which are input data of the map creation system, to be unified into the odometry coordinate system.

[0027] For conversion to the odometry coordinate system, in the case of 3D-LiDAR data, software LiDAR Odometry or LiDAR Inertial Odometry can be used, and in the case of 3D-LiDAR data with color information, software RGB-D SLAM (Simultaneous Localization and Mapping, which simultaneously estimates self-position and creates an environmental map) can be used.

[0028] Figure 6 is a flowchart for constructing a driving possibility cluster from a topological structure having multiple attribute information according to this embodiment. This corresponds to the GNG-DT processing step (S200) in Figure 3. In this embodiment, the point cloud data processing unit 310 constructs a topological structure using GNG-DT and performs clustering using the position data O(t) standardized in the odometry coordinate system and the three-dimensional point cloud data Y(t) as input data from the point cloud data conversion unit 230.

[0029] After the input data is standardized in the odometry coordinate system, general GNG-DT processing can be performed. In this embodiment, the input vector on the odometry coordinate system is composed of multiple attributes. Then, for this input vector, a topological structure for each different attribute is constructed while learning the spatial structure of the position data, and clustering is performed according to each attribute information. Below, we will explain how to learn the spatial structure based on Growing Neural Gas (GNG) for feature vectors composed of each attribute.

[0030] There are methods based on self-organizing maps that learn spatial structures from visual sensors such as 3D-LiDAR, while also learning topological structures composed of nodes and edges that describe the shapes and spatial relationships of objects. Among these, GNG includes an algorithm that adds nodes while deleting redundant nodes and edges, making it possible to appropriately learn spatial structures for unknown data distributions (for example, the spatial structure of unknown objects).

[0031] GNG uses only given data to learn reference vectors, and is therefore called unsupervised learning, performing competitive learning using winner-take-all. Furthermore, GNG is an algorithm that can dynamically change the topological structure based on the connections held by the referenced nodes. Furthermore, GNG can dynamically remove nodes and edges based on the concept of age. Therefore, GNG does not have redundant nodes or edges, and can arrange nodes based on an appropriately given spatial structure.

[0032] However, GNG only constructs a single topological structure, which means that it cannot simultaneously learn the topological structure for each attribute for an input vector consisting of multiple features (e.g., two-dimensional position coordinates and CIE Lab color coordinate values).

[0033] In this embodiment, we use GNG-DT (Growing Neural Gas with Different Topologies), an improved version of GNG. GNG-DT is robust to differences in scale between attributes and is capable of learning spatial information by defining and constructing multiple topological structures for each attribute. This is because GNG-DT can maintain approximately the same spatial structure in data sets of different scales for both the topological structure of position data and the topological structure of color information.

[0034] GNG-DT can also simultaneously learn feature vectors composed of multiple attributes and perform online clustering according to attribute information, making it possible to perform online clustering that can be applied when the robot 200 is moving and unknown objects (such as obstacles not shown on a map) in the environment surrounding the robot 200 change over time.

[0035] 6, S210 waits for the odometry data (O(t)) in the odometry coordinate system to be input from the point cloud data conversion unit 230. S220 waits for the 3D-LiDAR data (Y(t)) in the odometry coordinate system to be input from the point cloud data conversion unit 230.

[0036] In step S230, a topological structure having multiple attribute information is constructed, where the multiple attribute information includes position data, surface roughness, reflection intensity, color information, and the like.

[0037] S240 constructs a travel possibility cluster from the topological structure using attribute information related to travel. The travel possibility cluster is three-dimensional position data that classifies areas where the robot 200 can travel and areas where it cannot.

[0038] First, PCA (Principal Component Analysis) is performed on the three-dimensional position data to obtain normal information. Next, tilt angle information of the topological structure (nodes and edges), three-dimensional shape information (surface roughness), etc. are obtained from the normal information. In this embodiment, the tilt angle information and three-dimensional shape information are set as attribute information related to driving.

[0039] In the determination based on the tilt angle, if the tilt angle of a node is smaller than the maximum tilt angle at which the robot 200 can move, the node is determined to be drivable, and if it is equal to or greater than the maximum tilt angle, the node is determined to be drivable. In the determination based on the three-dimensional shape feature, if the three-dimensional shape feature of a node is smaller than the threshold value of the three-dimensional shape feature, the node is determined to be drivable, and if it is equal to or greater than the threshold, the node is determined to be drivable. In this way, node determination is performed based on the tilt angle information and the threshold value of the three-dimensional shape feature. Here, the threshold value of the three-dimensional shape feature is calculated from the eigenvalue based on the normal information of the three-dimensional shape.

[0040] The construction of the drivability cluster (C(t)) in the odometry coordinate system is performed within the GNG-DT algorithm based on the topological structure constructed in S230. The GNG-DT algorithm mainly updates the winning node and edges. While the winning node is updated using only three-dimensional position data, the edge update builds the connection relationships for each attribute while maintaining the spatial structure.

[0041] When there are two drivable nodes, the edge connecting the two nodes is determined to be drivable, and the edge is updated within the GNG-DT algorithm. When one node is drivable but the other is not, the edge connecting those nodes is determined to be not drivable, and the edge is updated within the GNG-DT algorithm. In this way, a drivability cluster (C(t)) can be constructed by setting the nodes on both sides of the edge determined to be drivable as drivable nodes (N_OK(t)) and setting the other nodes as not drivable nodes (N_NG(t)).

[0042] Fig. 7 is a flowchart showing how to map data in the odometry coordinate system to two-dimensional map data according to this embodiment. This corresponds to the step of mapping to the occupancy grid map (S300) in Fig. 3. In this embodiment, the mapping unit 320 executes the processing steps in Fig. 6 as post-processing of the GNG-DT.

[0043] In S310, the process waits for the drivability cluster (C(t)) in the odometry coordinate system to be input. In S320, the drivable node (N_OK(t)) is obtained from C(t). In S330, the Z coordinate is removed from N_OK(t) and projected onto a two-dimensional plane of XY to obtain the position (N_OK(t)_XY). In S340, the occupancy probability of each cell in the occupancy grid map is reduced using N_OK(t)_XY.

[0044] In S350, the non-drivable node (N_NG(t)) is obtained from C(t). In S360, the Z coordinate is removed from N_NG(t) and projected onto a two-dimensional plane of XY to obtain its position (N_NG(t)_XY). In S370, N_NG(t)_XY is used to increase the occupancy probability of each cell in the occupancy grid map.

[0045] 8A and 8B are schematic diagrams showing an example of a three-dimensional environment and diagrams for explaining corresponding two-dimensional map data. Fig. 8A shows an outline of the actual three-dimensional environment in which the robot 200 travels, using lines. Fig. 8B shows an occupancy grid map as an example of a two-dimensional map corresponding to Fig. 8A. In Figs. 8A and 8B, 810, 820, 830, and 840 indicate corridors extending from the center to the periphery.

[0046] 9A and 9B are schematic diagrams of a three-dimensional environment when an empty box according to this embodiment is placed, and diagrams for explaining two-dimensional map data that reflects the empty box. FIG. 9A shows a state in which an empty box is placed as an unknown object 850 in FIG. 8A, drawn with lines. FIG. 9B shows that the position (N_NG(t)_XY) of the empty box 850 has been mapped as an obstacle onto an occupancy grid map by the map creation system 100 according to this embodiment.

[0047] 9(b), the occupancy probability of each cell in the occupancy grid map is increased using N_NG(t)_XY by S370 in FIG. 7. Therefore, the robot 200 determines that the passages 810 and 820 with many black dots are impassable, and can automatically create no-entry areas. In the case of online processing where the control units 240 and 330 are implemented in a single CPU, obstacles can be detected in real time (10 to 20 Hz).

[0048] (Action and effect) The map creation system of this embodiment converts 3D point cloud data into an odometry coordinate system, i.e., unifies the 3D point cloud data into the same odometry coordinate system as the autonomous robot's 3D position data, thereby improving the tracking performance of GNG-DT learning. Furthermore, even when using 3D point cloud data and 3D position data from an autonomous robot performing sudden movements, noise in the topological structure can be reduced when constructing a topological structure with multiple attribute information. For example, if the sampling frequency of the 3D-LiDAR input is 10 Hz and the robot makes a sudden turn on the spot, resulting in little overlap with the previous viewpoint within 100 ms, the coordinates of the input point cloud will change significantly. In this case, if odometry is not used, the topological structure learning will not be able to keep up, and noise may occur. Using odometry reduces the coordinate changes in the input point cloud, allowing real-time topological structure learning to keep up, leading to reduced noise.

[0049] Furthermore, a 3D drivability cluster is constructed using attribute information related to driving from the topological structure. This reduces noise in the 3D drivability cluster. It also improves the accuracy of mapping unknown objects onto a 2D map. For example, GNG-DT outputs a topological structure that reflects each 3D LiDAR input immediately. This output includes sporadic noise and moving objects other than autonomous robots (people, cars, etc.). By accumulating and reflecting the topological structure in a 2D occupancy grid map, it becomes possible to remove temporary obstacles, such as sporadic noise and moving objects, from the 2D map. While it is desirable to immediately reflect detected obstacles on the map, if only immediate reflection were used, the detected obstacles would be overwritten by the latest input, discarding past information. By accumulating past drivability information as occupancy probabilities, it is possible to obtain both the detection of temporary obstacles and a map that removes temporary obstacles, which is desired for offline processing, as described below.

[0050] (Map Creation System According to Another Embodiment: Offline Processing) In the case of the above-mentioned online processing, the information processing device 300 needs to be mounted on the battery-powered autonomous mobile robot 200. For this reason, the point cloud data processing unit 310 needs to perform real-time processing (approximately 10 Hz) using a CPU with low power consumption, rather than a GPU with high power consumption. In contrast, in the case of offline processing, a GPU with high power consumption can be used. The map creation system 300' for offline processing will be described below.

[0051] 10 is a functional block diagram of a map creation system 100' according to another embodiment (offline processing). In the case of offline processing, a battery-powered autonomous mobile robot 200 is equipped with a point cloud and position data storage unit 500 instead of an information processing device 300'. The robot 200 autonomously travels according to a 2D map database 400. A point cloud data conversion unit 230 stores the 3D point cloud data and 3D position data converted into an odometry coordinate system in the point cloud and position data storage unit 500.

[0052] The information processing device 300' is connected to a power outlet and receives the power necessary for the GPU, which consumes a lot of power, to operate. In offline processing, the point cloud data processing unit 310' running on the GPU receives the 3D point cloud data and 3D position data in the odometry coordinate system from the point cloud / position data storage unit 500, and performs construction of a topological structure and clustering using GNG-DT as described in Figure 6.

[0053] The learning parameters of the point cloud data processing unit 310' can be changed for offline processing. Specifically, since there are no restrictions on the learning time of GNG-DT, parameters such as the number of nodes and the number of learning times can be increased to their maximum values. By applying GNG-DT, which places emphasis on learning accuracy in this way, it is possible to automatically generate no-entry areas for the entire driving area of ​​the autonomous mobile robot 200.

[0054] (Variation) The GNG-DT described above can adjust real-time performance and accuracy by adjusting the number of nodes and the number of learning times. For example, it is possible to implement online processing that applies GNG-DT to point cloud and position data input in real time and reflects this in an occupancy grid map to detect obstacles in real time, or offline processing that creates a set of point cloud and position data for each scan of a point cloud map created in advance using SLAM or similar, and applies GNG-DT, which emphasizes accuracy, to automatically generate no-entry areas for the entire driving area and update them in real time.

[0055] Although the embodiments of the present invention (including modified examples) have been described above, two or more of these embodiments may be combined and implemented. Alternatively, one of these embodiments may be partially implemented. Furthermore, two or more of these embodiments may be partially combined and implemented. For example, by converting the 3D point cloud data of this embodiment into an odometry coordinate system, it is possible to achieve both immediate online obstacle detection that emphasizes real-time performance and automatic offline no-entry area generation that emphasizes accuracy. Furthermore, in the case of online processing, the point cloud data processing unit 310 may perform some of the processing steps in FIG. 4 that are performed by the point cloud data acquisition unit 220.

[0056] Furthermore, the present invention is not limited to the above-described embodiments. Various modifications within the scope of the claims and within the scope of those skilled in the art are also included in the present invention. For example, color information obtained from a camera may be added to LiDAR point cloud data to determine the drivable area.

[0057] (Application example) An application example of the present invention will be described. To put a robot capable of autonomously traveling over a wide outdoor area into practical use, multiple tasks are required, such as map creation and revision, route creation and revision, and test driving. Among these, the task that requires the most time is the establishment of no-entry areas during route creation and revision. Because the final output of map creation in this embodiment is a two-dimensional occupancy grid map, it is possible to reduce the computational load on the CPU and power consumption in the subsequent route planning process that uses the map during autonomous traveling. Furthermore, it is possible to integrate no-entry areas with a previously created two-dimensional environmental map, which can be used as map data for route planning for other autonomous traveling robots.

[0058] In addition, compared to conventional downsampling, it is possible to compress data while retaining shape information, which reduces communication volume and enables map updates and remote control on multiple devices on the cloud. [Explanation of symbols]

[0059] 100, 100' Mapping System 200 Autonomous Driving Robot 210 Running part 220 Data Entry Section 230 Point Cloud Data Conversion Unit 240, 330 Control section 250, 340 Communications Department 300, 300' information processing device 310, 310' Point cloud data processing unit 320 Mapping Department 400 2D map databases 500 Point cloud and position data storage unit Aisles 810, 820, 830, 840 850 Unknown Object

Claims

1. an input unit for inputting three-dimensional point cloud data of the surrounding environment acquired while the autonomous mobile robot is traveling and three-dimensional position data of the autonomous mobile robot in an odometry coordinate system; a conversion unit that converts the three-dimensional point cloud data into data in the same odometry coordinate system as the three-dimensional position data; a topology constructing unit that constructs a topology structure for each different attribute using the three-dimensional point cloud data and the three-dimensional position data of the odometry coordinate system; a cluster construction unit that constructs a three-dimensional travel possibility cluster from the topological structure using attribute information related to travel; a mapping unit that uses the three-dimensional drivability clusters to map onto a two-dimensional map; A map creation system comprising:

2. 2. The map creation system according to claim 1, wherein the topology construction unit and the cluster construction unit are performed within a GNG-DT algorithm.

3. If there are two nodes that can be traveled, a first edge connecting the two nodes is determined to be drivable, and the first edge is updated within the GNG-DT algorithm; The map creation system described in claim 2, characterized in that if one node is drivable but the other node is not drivable, the second edge connecting the one and the other node is determined to be not drivable, and the second edge is updated within the GNG-DT algorithm.

4. 2. The map creation system according to claim 1, wherein the travel-related attribute information is inclination angle information and three-dimensional shape information.

5. The map creation system according to claim 1 , further comprising a storage unit that stores the three-dimensional point cloud data and the three-dimensional position data of the odometry coordinate system.

6. an input step of inputting three-dimensional point cloud data of the surrounding environment acquired while the autonomous mobile robot is traveling and three-dimensional position data of the autonomous mobile robot in an odometry coordinate system; a conversion step of converting the three-dimensional point cloud data into data in the same odometry coordinate system as the three-dimensional position data; a phase construction step of constructing a phase structure for each different attribute using the three-dimensional point cloud data and the three-dimensional position data of the odometry coordinate system; a cluster construction step of constructing a three-dimensional travel possibility cluster from the topological structure using travel-related attribute information; a mapping step of mapping the three-dimensional drivability clusters onto a two-dimensional map; A map creation method comprising:

7. an input step of inputting three-dimensional point cloud data of the surrounding environment acquired while the autonomous mobile robot is traveling and three-dimensional position data of the autonomous mobile robot in an odometry coordinate system; a conversion step of converting the three-dimensional point cloud data into data in the same odometry coordinate system as the three-dimensional position data; a phase construction step of constructing a phase structure for each different attribute using the three-dimensional point cloud data and the three-dimensional position data of the odometry coordinate system; a cluster construction step of constructing a three-dimensional travel possibility cluster from the topological structure using travel-related attribute information; a mapping step of mapping the three-dimensional drivability clusters onto a two-dimensional map; A computer-executable map creation program comprising:

Images