Data processing method, moving body control system, and program

Abstract

A data processing method according to the present disclosure includes: clustering environmental data around a moving body into a plurality of clusters on the basis of clustering parameters; extracting a cluster feature amount of each clustered cluster; calculating a geometric feature amount of the environment on the basis of the cluster feature amount of each cluster; performing feedback control to update the clustering parameters on the basis of the geometric feature amount of the environment; and determining an environmental scene on the basis of the geometric feature amount of the environment.

Description

Data processing method, mobile object control system, and program

[0001] The present disclosure relates to a data processing method, a mobile object control system, and a program.

[0002] Mobile objects, such as autonomous robots, can optimize their behavior and route planning by understanding the surrounding environmental scene. One method for understanding the environmental scene is to use an RGB camera. However, since this method relies solely on image information, it is difficult to accurately represent detailed terrain, such as the degree of terrain undulations and the extent of man-made structures. Meanwhile, when generating an environmental map from environmental data in a mobile object control system, the environmental data may be clustered within a certain range.

[0003] Japanese Patent Application Laid-Open No. 2018-071973

[0004] Patent Document 1 proposes a technology for extracting cluster features by clustering and determining the shape of the road surface, but because the clustering parameters are constant, it is difficult to perform optimal clustering according to the environment.

[0005] Therefore, it is desirable to provide a data processing method, a mobile object control system, and a program that can improve the accuracy of environmental scene determination and perform optimal clustering according to the environmental scene.

[0006] A data processing method according to one embodiment of the present disclosure includes clustering environmental data around a moving object into a plurality of clusters based on clustering parameters, extracting cluster features of each cluster, calculating geometric features of the environment based on the cluster features of each cluster, performing feedback control to update the clustering parameters based on the geometric features of the environment, and determining an environmental scene based on the geometric features of the environment.

[0007] A mobile object control system according to one embodiment of the present disclosure includes a cluster generation unit that clusters environmental data surrounding a mobile object into multiple clusters based on clustering parameters, a cluster feature extraction unit that extracts cluster features of each clustered cluster, an environmental geometric feature calculation unit that calculates geometric features of the environment based on the cluster features of each cluster, a feedback processing unit that performs feedback control to update the clustering parameters based on the geometric features of the environment, an environmental scene determination unit that determines an environmental scene based on the geometric features of the environment, and a behavior plan processing unit that generates a behavior plan for the mobile object based on the environmental scene determination result.

[0008] A program according to an embodiment of the present disclosure causes a computer to execute processes including clustering environmental data around a moving object into multiple clusters based on clustering parameters, extracting cluster features of each cluster, calculating geometric features of the environment based on the cluster features of each cluster, performing feedback control to update the clustering parameters based on the geometric features of the environment, and determining an environmental scene based on the geometric features of the environment.

[0009] In a data processing method, a mobile object control system, or a program according to an embodiment of the present disclosure, geometric features of the environment for determining the environmental scene are calculated based on the cluster features of each cluster, and feedback control of updating the clustering parameters is performed based on the geometric features of the environment.

[0010] FIG. 1 is an explanatory diagram relating to optimization of clustering. FIG. 2 is a block diagram schematically illustrating an example configuration of a data processing device to which a data processing method according to an embodiment of the present disclosure is applied. FIG. 3 is an explanatory diagram illustrating an example of environmental data. FIG. 4 is an explanatory diagram illustrating an example of a type of clustering. FIG. 5 is an explanatory diagram illustrating an example of clustering parameters. FIG. 6 is an explanatory diagram illustrating an example of local geometric information as cluster features obtained from point cloud clusters. FIG. 7 is an explanatory diagram illustrating an example relationship between environmental geometric features and an environmental scene. FIG. 8 is a block diagram schematically illustrating a first example configuration of a mobile object control system to which a data processing method according to an embodiment is applied. FIG. 9 is an explanatory diagram illustrating an example of an environmental map drawn by an environment map drawing device. FIG. 10 is a block diagram schematically illustrating a second example configuration of a mobile object control system to which a data processing method according to an embodiment is applied. FIG. 11 is a block diagram schematically illustrating a third example configuration of a mobile object control system to which a data processing method according to an embodiment is applied. FIG. 12 is a flowchart illustrating an example operation of a data processing device according to an embodiment. FIG. 13 is an explanatory diagram illustrating an example of determination of an environmental scene based on environmental geometric features. FIG. 14 is an explanatory diagram illustrating an overview of specific example 1 of calculation of environmental geometric feature quantities. FIG. 15 is an explanatory diagram illustrating an overview of specific example 2 of calculation of environmental geometric feature quantities. FIG. 16 is a block diagram schematically illustrating an example configuration of a data processing device that realizes specific example 2 of calculation of environmental geometric feature quantities. FIG. 17 is an explanatory diagram illustrating an example of cluster feature quantities when the environmental scene is a natural environment (a cluttered environment). FIG. 18 is an explanatory diagram illustrating an example of cluster feature quantities when the environmental scene is a natural environment (a bush, etc.). FIG. 19 is an explanatory diagram illustrating an example of cluster feature quantities when the environmental scene is an indoor environment (an artificial environment). FIG. 20 is an explanatory diagram illustrating an example of a sub-representation of environmental geometric feature quantities. FIG. 21 is an explanatory diagram illustrating a modified example of a method of representing environmental geometric feature quantities. FIG. 22 is a flowchart illustrating an example of determination by a feedback processing unit of whether feedback control is necessary. FIG. 23 is an explanatory diagram illustrating specific example 1 of updating clustering parameters by feedback control. FIG. 24 is an explanatory diagram illustrating specific example 2 of updating clustering parameters by feedback control.FIG. 25 is an explanatory diagram showing a specific example 3 of updating clustering parameters by feedback control. FIG. 26 is an explanatory diagram showing a specific example 4 of updating clustering parameters by feedback control. FIG. 27 is an explanatory diagram showing a specific example 4 of updating clustering parameters by feedback control. FIG. 28 is an explanatory diagram showing a specific example 5 of updating clustering parameters by feedback control. FIG. 29 is a block diagram schematically showing a modified example of a mobile object control system to which a data processing method according to an embodiment is applied. FIG. 30 is an explanatory diagram showing an example of processing for correcting an environmental scene determination result using time-series environmental scene information. FIG. 31 is an explanatory diagram showing an example of processing for correcting an environmental scene determination result using time-series environmental scene information.

[0011] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. The description will be made in the following order: 0. Comparative Example 1. One Embodiment 1.1 Configuration 1.2 Operation and Specific Examples 1.3 Effects 2. Other Embodiments

[0012] <0. Comparative Example> Mobile objects, such as autonomous robots, can optimize their behavior and path planning by understanding the surrounding environmental scene. One method for understanding the environmental scene is to use an RGB camera for assessment. In this case, assessment typically requires processing RGB images using a deep neural network (DNN) with a high-performance graphics processing unit (GPU), and it is difficult to obtain image information in situations without light sources, such as at night. On the other hand, when a mobile object, such as a robot, generates an environmental map from surrounding environmental data, the environmental data may be clustered within a certain range and processed. In this case, the clustering range, etc., is typically determined by fixed hyperparameters, and is not the optimal value for representing the environment.

[0013] Since environmental scene determination using an RGB camera is based solely on image information, it is difficult to represent detailed terrain, such as the degree of terrain undulations and the extent of man-made structures. Furthermore, in environments where RGB images cannot be acquired, such information cannot be obtained. Furthermore, in situations where such detailed environmental information cannot be obtained, it is difficult to determine clustering parameters for optimal clustering.

[0014] Furthermore, autonomous robots and the like can optimally behave according to the environmental scene when their driving area (driving scene) is predetermined or when information about the predetermined driving scene is input from outside. For example, in an area where there is a high possibility of people appearing, they can behave by driving at a reduced speed while being cautious of people. Conversely, in an environmental scene where there is a low possibility of collision with a person, they can prioritize their target task and move at high speed. In this way, the optimal behavioral pattern changes depending on the environmental scene. In this sense, robots and the like that operate in various environmental scenes can create an optimal behavior plan for the environmental scene by understanding the current environmental scene. By accurately understanding the surrounding environment, the robot's behavior can take danger into consideration, reduce damage to the robot, and optimize driving energy.

[0015] For example, in natural environments, there is no need to consider collisions with people, but the uneven ground makes smooth movement difficult. Also, in semi-natural environments, the possibility of collisions with people is low, people rarely appear suddenly, and the ground is uneven. Also, in artificial environments, people may suddenly appear, so care must be taken not to drive too fast, and the ground is paved, making it easy to drive.

[0016] Generally, the most common method for determining environmental scenes is to use RGB camera images and DNN. However, this method requires a huge amount of computation (energy consumption) for DNN processing, and determination becomes difficult in scenes where RGB image information cannot be obtained, such as at night or in fog. It is also difficult to obtain specific texture information for environmental scenes from RGB information. While DNN-based methods can predict terrain, they also depend on training data, and accuracy varies depending on the scene.

[0017] 1. One embodiment> [1.1 Configuration] FIG. 1 is an explanatory diagram regarding optimization of clustering.

[0018] Generally, autonomous robots generate a map of their surroundings based on environmental data and then navigate autonomously. Because autonomous robots run on batteries, they can operate for long periods without charging by reducing power consumption. Environmental data typically requires a large amount of data, so optimizing data representation techniques such as clustering can reduce the amount of data handled, the amount of calculation required to acquire geometric information and generate an environmental map, and the data volume.

[0019] The amount of data to be clustered is a hyperparameter, and is usually set to a fixed value. However, if the amount of data to be clustered is large, the data represented by the clusters will be highly abstract compared to the environment, which may result in poor accuracy in the environmental representation. Conversely, if the amount of data to be clustered is small, the accuracy of the environmental representation will tend to improve, but the amount of data in the clusters will increase, leading to an increase in the amount of data processing and the memory required to store it. In this way, the optimal parameter for the amount of data to be clustered will change depending on the environment. Although it also depends on the type of data, the data obtained from sensors is usually known by the user.

[0020] By optimizing the clustering parameters and setting the amount of data used for clustering to an optimal value for the environmental scene, it is possible to construct a more accurate environmental map with the optimal amount of calculation and data. The advantage of optimizing the clustering parameters is that it is possible to achieve both the optimal amount of calculation and data (low power consumption) and a highly accurate environmental map representation. Below, we will explain a data processing method that improves the accuracy of environmental scene determination and enables optimal clustering according to the environmental scene.

[0021] (Configuration Example of Data Processing Device) FIG. 2 schematically shows a configuration example of a data processing device 1 to which a data processing method according to an embodiment of the present disclosure is applied.

[0022] A data processing device 1 according to one embodiment is configured as an environmental scene determination device that determines an environmental scene around a moving object such as a robot, etc. The data processing device 1 includes a cluster generation unit 11, a cluster feature extraction unit 12, an environmental geometric feature calculation unit 13, a feedback processing unit 14, a cluster parameter determination unit 15, and an environmental scene determination unit 16.

[0023] In the following description, we use the term "environmental geometric features." Environmental geometric features are geometric features of an environment, and are numerical representations of the environment's geometric characteristics. The closest term to environmental geometric features is "scene," but a scene is an abstract human verbalization of an environmental situation that is commonly understood. While it is possible to infer the degree of geometric irregularity of an environment by hearing the name of a scene to some extent, different people imagine different scenes. In this specification, the term "environmental geometric features" refers to the degree of geometric irregularity of an environment. There is a correlation between environmental geometric features and scenes, and it is possible to infer the correlation between environmental geometric features and scenes. Furthermore, it is possible to empirically estimate the frequency of human appearance and the difficulty of traversing a scene, which also affects the robot's behavior planning.

[0024] The cluster generation unit 11 acquires environmental data around the mobile object from the sensor device 10, and clusters the environmental data into a plurality of clusters based on the clustering parameters determined by the cluster parameter determination unit 15. The sensor device 10 is, for example, a three-dimensional distance sensor, and the environmental data is, for example, three-dimensional distance sensor data.

[0025] FIG. 3 is an explanatory diagram showing an example of the environment data.

[0026] There are various types of sensor devices 10 that acquire environmental data, and various types of data are also available. The sensor device 10 may be, for example, a Time of Flight (ToF) sensor, a stereo camera, a ToF camera, or a radar. The data processing device 1 performs clustering optimization and environmental scene determination processing on the environmental data. Examples of environmental data to which such processing can be applied include three-dimensional point cloud data obtained from a ToF sensor, depth images obtained from a stereo camera or a ToF camera, and distribution data obtained from a radar.

[0027] There are various possible types (methods) and clustering parameters for clustering environmental data by the cluster generation unit 11. Clustering is a method for grouping all data into smaller data. There are various possible grouping methods and clustering parameters that control the data within the groupings. Examples of types of clustering and clustering parameters that determine clustering are shown below.

[0028] Fig. 4 is an explanatory diagram showing an example of the type of clustering, which shows an overview of k-means and OctTree as an example of the type of clustering.

[0029] Examples of types of clustering include the following: Hierarchical clustering: top-down clustering, bottom-up clustering, Diana method Non-hierarchical clustering: k-means, DBSCAN Density-based clustering: DBSCAN Model-based clustering: Gaussian mixture distribution model Grid-based clustering: OctTree

[0030] FIG. 5 is an explanatory diagram illustrating an example of the clustering parameters.

[0031] Examples of clustering parameters include the following: - Maximum cluster size - Minimum cluster size - Maximum number of data in a cluster - Minimum number of data in a cluster - (Average) number of data in a cluster - Maximum variance - Minimum variance - Outlier determination threshold

[0032] The cluster feature extraction unit 12 extracts cluster features of each cluster generated by the cluster generation unit 11 .

[0033] 6 is an explanatory diagram showing an example of local geometric information as cluster features obtained from a point cloud cluster. Fig. 6 shows an overview of the normal direction of a cluster, variance information of the cluster, and size of the cluster as the local geometric information.

[0034] The cluster feature extraction unit 12 extracts, for example, local geometric information of the point cloud cluster as the cluster feature. For example, the cluster feature extraction unit 12 may extract, as the cluster feature, at least one of local geometric information of the normal direction of each cluster, variance information of each cluster, and size of each cluster.

[0035] The normal direction of a cluster is the thickness direction of the local space formed by the point cloud that makes up the cluster. If the point cloud forms a plane, the normal direction represents the orientation of the plane.

[0036] The variance information of a cluster is information about the thickness of the local space formed by the constituent point cloud. When the point cloud forms a surface such as a floor or wall, the variance usually shows a small value. On the other hand, in areas such as uneven objects or areas where objects of different shapes intersect, there is variation in the thickness direction of the points, so the variance value becomes large.

[0037] The size of a cluster (the length in the major axis direction of the cluster) is the length in the major axis direction of the distribution of the local space constituted by the point cloud. When the number of point clouds contained in a cluster is the same, it can be determined that the larger the cluster size, the lower the density of the point clouds that constitute it.

[0038] The environment geometric feature calculation unit 13 calculates an environment geometric feature α based on the cluster feature of each cluster extracted by the cluster feature extraction unit 12 .

[0039] The feedback processing unit 14 performs feedback control to update the clustering parameters based on the environment geometric feature amount α calculated by the environment geometric feature amount calculation unit 13. The feedback processing unit 14 determines whether or not feedback control of the environment geometric feature amount α to the clustering parameters is necessary, based on the environment geometric feature amount α, and performs the feedback control.

[0040] When the feedback processing unit 14 determines that feedback control needs to be performed, the cluster parameter determination unit 15 updates the clustering parameters based on the feedback control performed by the feedback processing unit 14. The feedback control for updating the clustering parameters is repeatedly performed until the feedback processing unit 14 determines that feedback control does not need to be performed.

[0041] When the feedback processing unit 14 determines that there is no need to perform feedback control, the environmental scene determination unit 16 determines the environmental scene based on the final environmental geometric feature amount α.

[0042] FIG. 7 is an explanatory diagram showing an example of the relationship between the environmental geometric feature amount α and the environmental scene.

[0043] 7, the environmental geometric feature value α is expressed as a one-dimensional numerical value (0 to 1). For example, the environmental scene determination unit 16 determines that a completely natural environment is present when the value of the environmental geometric feature value α is 0, and that a completely artificial (flat) environment is present when the value of the environmental geometric feature value α is 1.

[0044] (Configuration Example of a Mobile Object Control System) Fig. 8 shows a first configuration example of a mobile object control system to which a data processing method according to an embodiment is applied. Fig. 8 shows an example of a mobile object control system that controls a robot 20 as a mobile object.

[0045] A mobile object control system according to one embodiment includes the data processing device 1, an environmental map generation unit 21, a behavior plan processing unit 23, and a terminal device 50 shown in FIG. 2 . The robot 20 includes a robot driving unit 24. The terminal device 50 includes an environmental map drawing device 22 and an input IF (interface) 25. In the mobile object control system according to one embodiment, the sensor device 10 may be provided in the robot 20. Furthermore, some or all of the other components may be provided in the robot 20. For example, some or all of the components in the data processing device 1 may be provided in the robot 20. Furthermore, at least one of the environmental map generation unit 21 and the behavior plan processing unit 23 may be provided in the robot 20. Furthermore, some of these components may be configured separately from the robot 20. The same applies to the other configuration examples of mobile object control systems described below.

[0046] The clustering parameters and environmental scene information determined by the data processing device 1 shown in Fig. 2 can be used to generate an environmental map and to plan the behavior of the robot 20. In addition, in the process of developing the behavior plan processing unit 23 based on the environmental map, an operator 40 may monitor the environmental map drawn by the environment map drawing device 22 and create a program for the behavior plan processing unit 23 to generate a path. The environment map drawing device 22 may be used not only in the development process but also when debugging the product.

[0047] The behavior plan processing unit 23 generates a behavior plan for the robot 20 based on the environmental scene determination result by the environmental scene determination unit 16. The behavior plan processing unit 23 uses the environmental geometric feature α and the environmental scene information according to the created program to utilize them as environmental information, such as to change the behavior of the robot 20.

[0048] The environmental map generating unit 21 generates an environmental map based on the clustered environmental data.

[0049] The environment map drawing device 22 draws the information of the environment map generated by the environment map generating unit 21 so that the operator 40 can visually recognize it.

[0050] The input IF 25 is an input unit that receives input of control information for controlling the action plan, which is generated by the operator 40 based on the information of the environmental map.

[0051] FIG. 9 is an explanatory diagram showing an example of an environment map drawn by the environment map drawing device 22. As shown in FIG.

[0052] The input IF 25 accepts input of control information for controlling a behavior plan for a specific area in which the robot 20 can move, determined based on the information on the environmental map. When an operator 40 of the robot 20 checks the environmental map, the environmental map data from the environmental map generation unit 21 is transferred to the environmental map drawing device 22 via a wired or wireless connection, and the environmental map is projected onto a display or the like in a form that the operator 40 can visually confirm. For example, when programming a path plan, the operator 40 defines driving rules for driving at an optimal speed and route in an area determined to be a drivable area in the environmental map (a specific area in which the robot 20 can move). The operator 40 also adjusts the completed program by drawing the environmental map and debugging it. The operator 40 may also check the environmental map from the perspective of analysis, such as whether the robot 20 can move without any problems as a product.

[0053] FIG. 10 shows a second example of the configuration of a mobile object control system to which a data processing method according to an embodiment of the present invention is applied.

[0054] The input IF 25 may be configured to receive an input of a setting of a determination condition for whether or not to perform feedback control in the feedback processing unit 14. The input IF 25 may also be configured to receive an input of an update setting of the clustering parameters in the cluster parameter determination unit 15.

[0055] The operator 40 may refer to the environmental map drawn by the environmental map drawing device 22 to set the conditions for determining whether or not to implement feedback control, or input update settings for the clustering parameters in the cluster parameter determination unit 15 (creating and modifying an update program, etc.).

[0056] FIG. 11 shows a schematic diagram of a third example of the configuration of a mobile object control system to which a data processing method according to an embodiment of the present invention is applied.

[0057] The environment map generation unit 21 may generate an environment map based on the environmental data before it is clustered by the cluster generation unit 11. The environment map generation unit 21 may change the drawing method of the environment map to a method suitable for the environmental scene based on the determination result of the environmental scene. This allows the environment map generation unit 21 to automatically optimize the generation parameters used in generating the environment map according to the environmental scene.

[0058] The main components of the data processing device 1 and the mobile object control system according to an embodiment may be configured as a computer including, for example, one or more processors (e.g., CPUs (Central Processing Units)) and one or more memories (e.g., ROMs (Read Only Memory) and RAMs (Random Access Memory)) communicatively connected to the one or more processors. In this case, the processing of each block in the data processing device 1 and the mobile object control system according to an embodiment may be realized by the one or more processors executing processing based on a program stored in the one or more memories. Furthermore, the processing of each block in the data processing device 1 and the mobile object control system according to an embodiment may be realized by the one or more processors executing processing based on a program supplied from an external device via, for example, a wired or wireless network. The program may be recorded on a non-transitory recording medium readable by a computer.

[0059] Furthermore, the main components of the data processing device 1 and the mobile object control system according to an embodiment may be configured with multiple computers, each having one or more processors and one or more memories. Furthermore, the data processing device 1 and the mobile object control system according to an embodiment may be configured with multiple devices (such as server devices) each having a computer.

[0060] [1.2 Operation and Specific Examples] The data processing device 1 performs clustering of environmental data using initial clustering parameters that serve as a guide for environmental scene discrimination, extracts cluster features for each cluster, and then calculates an environmental geometric feature value α using the extracted cluster features. Next, the environmental geometric feature value α is fed back to a control unit (feedback processing unit 14, cluster parameter determination unit 15) that controls the clustering parameters using the environmental geometric feature value α, and the clustering parameters optimal for the environmental scene are updated. The above process is repeated one or more times, and the environmental scene is determined from the final environmental geometric feature value α.

[0061] FIG. 12 is a flowchart showing an example of the operation of the data processing device 1 according to an embodiment.

[0062] First, the cluster generation unit 11 acquires environmental data (e.g., three-dimensional distance sensor data) from the sensor device 10 (step S101). Next, the cluster generation unit 11 loads clustering parameters from the cluster parameter determination unit 15 (step S102). Next, the cluster generation unit 11 performs clustering of the environmental data based on the clustering parameters (step S103).

[0063] Next, the cluster feature extraction unit 12 extracts cluster features (step S104), and then the environment geometric feature calculation unit 13 calculates an environment geometric feature α based on the cluster features (step S105).

[0064] Next, the feedback processing unit 14 performs a process of determining whether or not feedback control is necessary based on the environmental geometric feature value α (step S106). If it is determined that feedback control is necessary, the cluster parameter determination unit 15 updates the clustering parameters based on the feedback control performed by the feedback processing unit 14 (step S107), and then the process returns to step S102 performed by the cluster generation unit 11. On the other hand, if it is determined that feedback control is not necessary, the environmental scene determination unit 16 determines the environmental scene based on the environmental geometric feature value α (step S108), and the process ends.

[0065] 13 is an explanatory diagram showing an example of determining an environmental scene based on the environmental geometric feature α. The environmental geometric feature calculation unit 13 calculates the environmental geometric feature α based on the local geometric information extracted as the cluster feature by the cluster feature extraction unit 12.

[0066] The environmental geometric feature calculation unit 13 expresses the local geometric features of each cluster extracted as cluster features as some kind of environmental geometric feature α (quantitative value) representing the geometric shape of the environmental scene. In this way, the data processing device 1 according to one embodiment reduces the environmental geometric feature α to a discrete environmental scene determination and updates the clustering parameters by feeding back the environmental geometric feature α to the clustering parameters. The environmental geometric feature α may be expressed as a one-dimensional scalar value or a two- or higher-dimensional vector. Various representations are possible, such as reducing the geometric feature of the cluster to a one-dimensional value or to two or more elements. Environmental scene determination involves classifying scenes to enable human understanding, and is difficult to handle numerically. Therefore, expressing the environmental geometric feature α numerically as intermediate data makes it easier to use in the final environmental scene determination. Furthermore, it can be used for feedback control by the feedback processing unit 14 to easily update the clustering parameters.

[0067] (Specific Example of Calculation of Environmental Geometric Feature Amount) (Specific Example 1 of Calculation) FIG. 14 is an explanatory diagram showing an outline of specific example 1 of calculation of the environmental geometric feature amount α.

[0068] The environment geometric feature calculation unit 13 may calculate the environment geometric feature α based on vector information on the normal direction of each cluster extracted as a cluster feature by the cluster feature extraction unit 12. For example, the environment geometric feature calculation unit 13 calculates the axis with the largest component among the X, Y, and Z axis components of the extracted normal direction vector and its vector value across all clusters, and calculates a statistical value as the environment geometric feature α. The vector value may be normalized to a value between 0 and 1 for each axis. As a result, for example, in an artificial environment (floor, wall, ceiling, etc.), many clusters with normal directions facing a certain direction are generated, resulting in a large vector value in a specific direction, as shown in FIG. 14 . On the other hand, in a natural environment, the direction of the normal direction varies from place to place, and the largest vector component does not often occur in clusters facing a specific direction, resulting in a small vector value. In this way, the environment geometric feature calculation unit 13 extracts features corresponding to the geometric shape of the environment (scene) as the environment geometric feature α. 14 shows an example in which vector extraction is performed using the three directions of the X, Y, and Z axes as representative values, but vector extraction may also be performed using six axes (X, Y, Z, XY, YZ, and ZX) including the X, Y, Y, Z, and ZX directions, which allows for more detailed extraction of environmental features.

[0069] (Specific Example 2 of Calculation) Fig. 15 is an explanatory diagram showing an overview of specific example 2 of calculation of the environmental geometric feature amount α. Fig. 16 is a block diagram showing an outline of an example of the configuration of a data processing device 1 that realizes specific example 2 of calculation of the environmental geometric feature amount α.

[0070] The data processing device 1 according to an embodiment may further include a nearby cluster extraction unit 17. The nearby cluster extraction unit 17 extracts information about nearby clusters that exist around any cluster whose cluster features have been extracted by the cluster feature extraction unit 12. Methods for searching for nearby clusters include methods that use kd-trees and Mahalanobis distances.

[0071] In the configuration example of Fig. 2, the environment geometric feature calculation unit 13 calculates the environment geometric feature α using the cluster feature of each cluster extracted by the cluster feature extraction unit 12. In contrast to this, as in the configuration example of Fig. 16, the environment geometric feature calculation unit 13 may calculate the environment geometric feature α using the cluster feature of each cluster extracted by the cluster feature extraction unit 12 and information on neighboring clusters extracted by the neighboring cluster extraction unit 17. This makes it possible to define the environment geometric feature α from an even macroscopic perspective, and may enable the environment geometric feature α for each environmental scene to be more clearly expressed.

[0072] The environment geometric feature calculation unit 13 calculates features using the features of each cluster extracted by the cluster feature extraction unit 12 and the features of its neighboring clusters. The environment geometric feature calculation unit 13 performs these calculations for all clusters to calculate the environment geometric feature α.

[0073] The environmental geometric feature calculation unit 13 searches for cluster similarity between the feature of each cluster extracted by the cluster feature extraction unit 12 and its neighboring cluster. The cluster similarity φ is calculated, for example, by the following formula (1). Na and Nb are two clusters whose similarity is to be determined. φ=dot(Na, Nb) (1)

[0074] The environmental geometric feature calculation unit 13 performs the calculation of equation (1) on all neighboring clusters (equation (2)). Φ=Σφ (2)

[0075] The environment geometric feature calculation unit 13 performs the above calculation for all clusters to calculate the environment geometric feature α (equation (3)). α=ΣΦ / Z (3)

[0076] The environment geometric feature calculation unit 13 normalizes the environment geometric feature α to a value between 0 and 1. For example, it divides by the total number of calculations Z, as shown in equation (3). Because each value of Φ is between 0 and 1, the calculations for all clusters become an average value, which can be normalized to between 0 and 1. The environment geometric feature calculation unit 13 sets the normalized value as the environment geometric feature α. For example, in an environment made entirely of flat surfaces, the environment geometric feature α will be a value close to 1, and even in an environment with many artificial objects such as floors and walls, the value will be close to 1. On the other hand, in a natural environment, the calculation results of equation (1) above will be small, so the environment geometric feature α will also be a small value, and it can be determined that the environment is a natural environment.

[0077] (Specific Example 3 of Calculation) The environmental geometric feature calculation unit 13 may calculate the cluster similarity φ using the following equation (1)' instead of the above equation (1) in the above specific example 2. As shown in the following equation (1)', by using a function f(λa, λb) for calculating a correction amount using the variance values ​​λa and λb of each cluster for the above equation (1), the variance values ​​λa and λb of a surface closer to an artificial object, for example, become smaller, and the variance values ​​λa and λb become larger in a more natural environment. Therefore, the environmental geometric feature calculation unit 13 can calculate an environmental geometric feature α that improves the discrimination between an artificial environment and a natural environment. φ=dot(Na, Nb)*1 / f(λa, λb) ... (1)' Two examples of the correction function f(λa, λb) are shown below. f (λa, λb) = 1 / exp (-[λa + λb]) f (λa, λb) = 1 / exp (- [λa * λb])

[0078] (Effects of Specific Examples 2 and 3 of Calculation) Fig. 17 is an explanatory diagram showing an example of cluster features when the environmental scene is a natural environment (a cluttered environment). Fig. 18 is an explanatory diagram showing an example of cluster features when the environmental scene is a natural environment (a bush, etc.). Fig. 19 is an explanatory diagram showing an example of cluster features when the environmental scene is an indoor environment (an artificial environment).

[0079] As shown in Figures 17 to 19, when environmental data is clustered, the geometric characteristics (cluster features) of each generated cluster are similar to those of neighboring clusters, resulting in different characteristics for each environment. Figures 17 to 19 show the consistency of normal directions with neighboring clusters and the variance of the normal directions of clusters. In the above-mentioned specific examples 2 and 3, these features are used to represent the environmental geometric feature α, making it possible to numerically represent the environment.

[0080] 17, in a natural environment (a noisy environment), the normal directions of the clusters are noisy and vary widely. The thicknesses of the clusters often have a large amount of dispersion.

[0081] 18, in a natural environment (such as a bush), the normal directions of the clusters tend to be more or less in the same direction, and the thickness of each cluster tends to have a large amount of variance.

[0082] As shown in Figure 19, in an indoor environment (artificial environment), the normal direction of each cluster faces the same direction on the same plane. The thickness of each cluster often has a small amount of variance (the amount of variance increases at the boundaries with the floor and walls).

[0083] (Exemplary Expressions of Environmental Geometric Feature α by Specific Examples 2 and 3 of Calculation) FIG. 20 is an explanatory diagram showing an example of a secondary expression of the environmental geometric feature α.

[0084] In the expression (Equation (3)) representing the environmental geometric feature α in the above specific examples 2 and 3, the environmental geometric feature α is expressed as a numerical value between 0 and 1. Note that Equation (3) above is an example of an expression for the environmental geometric feature α, and the expression method is not limited to Equation (3). The value representing the environmental geometric feature α depends on how the environmental features are expressed using the geometric features of the clusters. For this reason, the meaning of the environmental geometric feature α is not always the same, but it is generally used to determine the degree to which an environment is natural or artificial (e.g., floors, walls). A natural environment is an environment in which the directions and variances of the normal vectors of each cluster are mixed together in a complex manner. An artificial environment is an environment in which planes in which the normal vectors of each cluster face in the same direction or clusters with small variance are found. Naturally, secondary and tertiary uses of the environmental geometric feature α are also conceivable, such as further expressing the probability of human presence Pr=f(α) from the environmental geometric feature α. In this way, the representation of the environmental geometric feature α using the local geometric information of the cluster can be expanded not only as a geometric numerical representation but also as a secondary representation derived therefrom. In the example of Fig. 20, the secondary representation of the environmental geometric feature α represents the probability of a person's presence Pr = f(α) and, for example, the level of caution regarding collisions with people in the environment in which the robot 20 operates.

[0085] 21 is an explanatory diagram showing a modified example of the method of expressing the environmental geometric feature α. In Fig. 21, the vertical axis represents the statistic σ of the variance value of each cluster, and the horizontal axis represents the consistency ε of the normal direction of each cluster with respect to neighboring clusters.

[0086] The environmental geometric feature α may be finally expressed in a discrete expression in human language as shown in FIG. 21 . While the above-described specific example 3 presented an equation (Equation (1)′) that takes into account the variance value in the normal direction consistency ε of neighboring clusters, the variance can also be expressed as a separate parameter. By defining the variance statistics σ as a separate dimensional index, as shown in FIG. 21 , the environmental geometric feature α can be expressed as two-dimensional data. In FIG. 21 , the variance of all clusters is expressed as some statistical numerical value, such as the mean value or variance. It is believed that there is a correlation between the variance statistics σ and the normal direction consistency ε of each cluster. Considering these values ​​is believed to improve the accuracy of environmental scene determination. The environmental scene determination unit 16 uses the environmental geometric feature α to perform a final environmental scene determination. This allows the environmental geometric feature α, expressed numerically, to be used in various applications, such as verbalizing the environment to ultimately define the environment in which the robot 20 will operate, or as a discrete representation of the environment to determine the robot system's behavior plan.

[0087] (Determination Process for Implementing Feedback Control) FIG. 22 is a flowchart showing an example of the process performed by the feedback processing unit 14 to determine whether or not it is necessary to implement feedback control.

[0088] The following describes the determination by the feedback processing unit 14 using the environmental geometric feature α as to whether or not feedback control is necessary. The determination criteria and other factors vary depending on the usage scenario and usage pattern; therefore, the determination example shown in FIG. 22 is merely an example. If the calculation result of the environmental geometric feature α is highly likely, even if this is the first feedback control, it is possible to determine that feedback control is unnecessary and terminate the control without updating the clustering parameters. The example in FIG. 22 illustrates a case in which, if the change in the environmental geometric feature α during the previous feedback control is below a certain level, the clustering parameters are deemed to have converged to an optimal value, and the repetition of the calculation for feedback control is stopped. In the example in FIG. 22 , since there is no difference between the first feedback control and the previous environmental geometric feature α, it is determined that feedback control is necessary. From the second feedback control onward, if the change in the environmental geometric feature α compared with the previous calculation result is below a threshold, it is determined that the accuracy of clustering determined by the clustering parameters has converged to an optimal value, and feedback control is not performed, and the clustering parameters are not updated.

[0089] First, the feedback processing unit 14 determines whether or not the feedback control is being performed for the second time or later (step S201). If it is determined that the feedback control is not being performed for the second time or later (step S201; N), the feedback processing unit 14 determines that the feedback control needs to be performed (step S204), and ends the process of determining whether or not to perform the feedback control.

[0090] On the other hand, if it is determined that the feedback control is being performed for the second time or later (step S201; Y), the feedback processing unit 14 then determines whether the amount of change in the environmental geometric feature α from the time of the previous control execution is within a certain range (step S202). If it is determined that the amount of change is not within a certain range (step S202; N), the feedback processing unit 14 determines that feedback control execution is necessary (step S204) and terminates the feedback control execution determination process. On the other hand, if it is determined that the amount of change is within a certain range (step S202; Y), the feedback processing unit 14 determines that feedback control execution is not necessary (step S203) and terminates the feedback control execution determination process.

[0091] (Specific Example of Updating Clustering Parameters by Feedback Control) (Specific Example of Updating 1) FIG. 23 is an explanatory diagram showing a specific example of updating clustering parameters by feedback control.

[0092] For example, suppose the initial clustering parameters are set so that the environmental geometric feature α is 0.5 in an environment exactly halfway between a natural environment and an artificial environment. Then, suppose the calculation result of the environmental geometric feature α in the first clustering (previous result) is a result closer to an artificial environment, where the environmental geometric feature α is greater than 0.5, as shown in FIG. 23 . Next, suppose that it is desired to increase the number of data points in a cluster (i.e., reduce the number of clusters) without significantly changing (decreasing) the environmental geometric feature α, and change the clustering parameters (update the cluster size in a direction to increase it). As a result, as shown in FIG. 23 , the calculation result of the environmental geometric feature α (current result) changes by Δdiff, which is the difference from the previous result. If the feedback processing unit 14 determines in its feedback control execution determination process that the difference Δdiff exceeds the threshold Thr and thus determines that feedback control is necessary, the cluster parameter determination unit 15 can update the clustering parameters using, for example, the intermediate value of the clustering sizes from the previous and previous clustering times as the update parameters.

[0093] Furthermore, when the equation constituting the environmental geometric feature α is expressed as an equation with the clustering parameter Pcnum as a variable, the difference Δdiff can be expressed as a differential equation with the clustering parameter Pcnum as a variable, as follows, with the difference Δdiff as the error. This allows the cluster parameter determination unit 15 to update the clustering parameters. This is a mechanism equivalent to backpropagation in deep learning. Δdiff / ΔPcnum=(Δdiff / Δα)*(Δα / ΔΦ)*(ΔΦ / Δφ)...

[0094] The update formula for the clustering parameter Pcnum is expressed as follows, for example: Pcnum=Pcnum-μ*(Δdiff / ΔPcnum), where μ represents an update weight.

[0095] In this way, in the feedback of the environmental geometric feature α, the cluster parameter determination unit 15 may update the clustering parameters on a rule-based basis from the value or change amount of the environmental geometric feature α, or may calculate and determine the update amount using machine learning as in the above formula.

[0096] (Specific Example 2 of Updating) (Specific Example 1 of Updating When There Are Two or More Clustering Parameters) FIG. 24 is an explanatory diagram showing a specific example 2 of updating clustering parameters by feedback control.

[0097] When a cluster is generated using two (or more) clustering parameters P1, P2 (, P3, ...), if the environmental geometric feature α is differentiable with respect to P1, P2 (, P3, ...), the cluster parameter determination unit 15 can update the clustering parameters P1, P2 (, P3, ...) from the difference Δdiff in the environmental geometric feature α using the same method as above.

[0098] For example, if cluster Cl is composed of P1, P2 (, P3...), cluster Cl can be expressed as Cl(P1, P2 (, P3...)). Since the environmental geometric feature α is determined by this cluster, the difference Δdiff in the environmental geometric feature α can be expressed by the clustering parameters P1, P2 (, P3...). In other words, the difference Δdiff can be differentiated by the clustering parameters P1, P2 (, P3...). Δdiff / ΔP1=(Δdiff / Δα)*(Δα / ΔΦ)*(ΔΦ / Δφ)... Δdiff / ΔP2=(Δdiff / Δα)*(Δα / ΔΦ)*(ΔΦ / Δφ)...

[0099] Using the above formula, the cluster parameter determination unit 15 can update the clustering parameters P1, P2 (, P3, ...) as follows. The update formula for the clustering parameters P1, P2 (, P3, ...) is expressed, for example, as follows: μ indicates an update weight. Pk=Pk-μk*(Δdiff / ΔPk) k=1, 2 (, 3, ...)

[0100] (Specific example 3 of updating) (Specific example 2 of updating when there are two or more clustering parameters) Fig. 25 is an explanatory diagram showing specific example 3 of updating of clustering parameters by feedback control. In Fig. 25, the vertical axis represents the statistic σ of the variance values ​​of each cluster, and the horizontal axis represents the consistency ε of the normal directions of each cluster with neighboring clusters.

[0101] When the environmental geometric feature α is data of two or more dimensions, the cluster parameter determination unit 15 may update the clustering parameters constituting each dimension in response to the amount of change in each dimension. In Fig. 25, the environmental scene is determined using two environmental geometric feature α: the statistical quantity σ of the variance values ​​of each cluster and the consistency ε of the normal directions of each cluster with neighboring clusters. The differences between the previous calculation results for the values ​​of σ and ε and the current calculation results can be expressed as Δσ and Δε, respectively.

[0102] If the variance statistics σ shifts toward the natural environment, as shown in FIG. 25 , updating the maximum variance of the clustering parameters to a higher value improves the expressiveness of the natural environment. Furthermore, in the example of FIG. 25 , the normal direction consistency ε also shifts toward the natural environment, so for example, by updating the clustering parameters that determine the cluster size (maximum cluster size or maximum number of points in a cluster) in the direction of decreasing the cluster size, the expressiveness of the uneven natural environment (cluster resolution) changes to a more optimal one. In this way, the cluster parameter determination unit 15 can update multiple clustering parameters.

[0103] In the process of determining whether or not feedback control needs to be performed by the feedback processing unit 14, various determination conditions such as (i), (ii), and (iii) below can be set, for example: (i) Δσ is equal to or greater than a threshold Σ, or Δε is equal to or greater than a threshold E; (ii) Δσ is equal to or greater than a threshold Σ, and Δε is equal to or greater than a threshold E; or (iii) α*Δσ+β*Δε is equal to or greater than a threshold μ (α and β are weighting coefficients).

[0104] (Specific Example 4 of Updating) FIGS. 26 and 27 are explanatory diagrams showing a specific example 4 of updating clustering parameters by feedback control. FIGS. 26 and 27 schematically show an example of the state of the clusters before and after updating the clustering parameters. FIG. 26 schematically shows an example of the environmental scene when it is an artificial environment, and FIG. 27 schematically shows an example of the environmental scene when it is a natural environment. Normally, the entire environment is clustered, but FIGS. 26 and 27 schematically show an example in which only specific locations are displayed as clusters for ease of understanding. For example, FIG. 26 schematically shows an example in which only the floor surface is displayed as a cluster.

[0105] For example, if the environmental geometric feature α indicates a value close to that of an artificial environment, it can be assumed that the environment contains artificial objects such as floors and walls. Since an artificial environment is expected to have planes of the same geometric shape occupying much of the space, the cluster parameter determination unit 15 may set larger clusters as shown in Fig. 26 by increasing the cluster size of the clustering parameters or by increasing the number of point groups in each cluster. This makes it possible to represent the environment with a smaller amount of data.

[0106] On the other hand, if the environment is determined to be natural (with many bumps and dips), the default cluster size may be too large compared to the degree of bumpiness, and may not be able to represent the bumpiness of the environment. For example, the cluster size (resolution) may not match the size of the bumps and dips, resulting in an averaged representation. In such an environment, the cluster parameter determination unit 15 may set the cluster size or the number of points included in the cluster to a small value, as shown in FIG. 27. This allows the bumps and dips of the environment to be represented with good resolution.

[0107] (Specific example 5 of updating) (Specific example 3 of updating when there are two or more clustering parameters) Fig. 28 is an explanatory diagram showing specific example 5 of updating the clustering parameters by feedback control. In Fig. 28, the solid line indicates the state of the cluster before updating the clustering parameters (previous filter value), and the solid line indicates the state of the cluster after updating the clustering parameters (after strengthening the filter).

[0108] As described above, for example, if the environmental geometric feature α indicates a value close to that of an artificial environment, the cluster parameter determination unit 15 may set the cluster to a larger size by increasing the cluster size of the clustering parameter or the number of point clouds in the cluster. This allows the environment to be represented with a smaller amount of data. In addition, point clouds are unlikely to occur in distant environments in artificial environments, and if point cloud data exists in an outlying location, it can be determined as an outlier (noise). For this reason, the cluster parameter determination unit 15 may update not only the cluster size but also the filter variable of the cluster as a clustering parameter. The feedback processing unit 14 only needs to determine coplanar surfaces, etc., and may determine that outlying data is likely to be noise and is unnecessary even if it is correct data. It is sufficient to obtain data on mass areas such as floors and walls. For this reason, the cluster parameter determination unit 15 may set a stronger filter value for the cluster as a clustering parameter.

[0109] (Correction of Environmental Scene Judgment Results Using Time-Series Environmental Scene Information and Reflection in Action Plan) FIG. 29 shows an outline of a modified example of a mobile object control system to which a data processing method according to one embodiment is applied.

[0110] The mobile object control system according to one embodiment may further include a determination result recording unit 31 in addition to the first configuration example of the mobile object control system shown in FIG.

[0111] The determination result recording unit 31 records the determination result of the environmental scene by the environmental scene determination unit 16 every time a determination result is produced. When a new environmental scene is determined, the environmental scene determination unit 16 reads out the determination result of the past environmental scene recorded in the determination result recording unit 31 and makes corrections for the determination of the new environmental scene.

[0112] 30 and 31 are explanatory diagrams showing an example of a process for correcting the determination result of the environmental scene using time-series environmental scene information.

[0113] The environmental scene determination unit 16 corrects the determination result of the environmental scene based on time-series data indicating past determination results of the environmental scene. For example, when the environmental geometric feature value α is calculated and the environmental scene is determined in a certain frame, the environmental scene determination unit 16 may correct the determination result of the current environmental scene using information about the past environmental scene and the environmental geometric feature value α. Furthermore, the environmental scene determination unit 16 may reflect this in the behavior plan of the behavior plan processing unit 23 when causing the robot 20 to travel autonomously. This makes it possible to perform robot control, etc., that is more optimal for the environmental scene.

[0114] For example, by considering the time series of environmental scenes of the past N scenes, it is possible to obtain information such as whether the current environmental scene is a new environmental scene that has transitioned from the previous environmental scene, and what kind of environment the person was in just before.

[0115] For example, as in the example of Figure 30, if the environmental scene judgment results (e.g., natural environment) of the past N (e.g., N = 4) frames and the current environmental scene judgment result (e.g., natural environment) are all the same, the environmental scene judgment unit 16 does not correct the current environmental scene judgment result.

[0116] On the other hand, as shown in the example of FIG. 31 , if there is a difference between the environmental scene determination results (e.g., artificial environment and natural environment) for the past N (e.g., N = 4) frames and the current environmental scene determination result (e.g., natural environment), the environmental scene determination unit 16 corrects the current environmental scene determination result. For example, if the environmental scene determination result in the past N frames includes "artificial environment" and the current environmental scene determination result is "natural environment," the environmental scene determination unit 16 may correct the current environmental scene determination result from "natural environment" to "natural environment with nearby artificial environment." This allows the robot 20 to be considered to have been in an artificial environment until just before and to be in the current natural environment near the artificial environment, increasing the likelihood, for example, that a person is nearby. Therefore, the behavior plan processing unit 23 can perform control such as correcting the behavior plan from a behavior plan for a completely natural environment (a behavior plan that requires less attention to people) to a behavior plan that is closer to the artificial environment (a behavior plan that requires more attention to people and performs autonomous behavior).

[0117] In this way, by utilizing information on time-series environmental scenes, information on past environmental scenes can be reflected in information on the current environmental scene and used to correct the behavior plan of the robot 20, etc.

[0118] [1.3 Effects] As described above, according to a data processing method according to one embodiment, an environmental geometric feature value α for determining an environmental scene is calculated based on the cluster feature value of each cluster, and feedback control is performed to update the clustering parameters based on the environmental geometric feature value α. This improves the accuracy of determining an environmental scene and enables optimal clustering according to the environmental scene.

[0119] Furthermore, according to the data processing method of one embodiment, it is possible to more accurately determine the environmental scene without using RGB images. Furthermore, according to the data processing method of one embodiment, it is possible to accurately determine the current environmental scene, thereby making it possible to plan the robot 20's behavior optimal for the environmental scene. This makes it possible to plan the robot 20's behavior taking into consideration danger, reduce damage to the robot 20, optimize the running energy, etc.

[0120] The effects described in this specification are merely examples and are not limiting, and other effects may also be achieved. The same applies to the effects of other embodiments described below.

[0121] 2. Other Embodiments The technology according to the present disclosure is not limited to the description of the above embodiment, and various modifications are possible.

[0122] For example, the present technology can be configured as follows. According to the present technology configured as follows, geometric features of the environment for determining the environmental scene are calculated based on the cluster features of each cluster, and feedback control for updating the clustering parameters is performed based on the geometric features of the environment. This improves the accuracy of determining the environmental scene and enables optimal clustering according to the environmental scene.

[0123] (1) A data processing method comprising: clustering environmental data around a mobile object into a plurality of clusters based on clustering parameters; extracting cluster features of each of the clusters; calculating geometric features of the environment based on the cluster features of each of the clusters; performing feedback control to update the clustering parameters based on the geometric features of the environment; and determining an environmental scene based on the geometric features of the environment. (2) The data processing method described in (1) above, further comprising: generating a behavior plan for the mobile object based on the results of the environmental scene determination. (3) The data processing method described in (2) above, further comprising: generating an environmental map based on the environmental data; drawing information of the generated environmental map so that it is visible; and accepting input of control information for controlling the behavior plan, generated based on information of the environmental map. (4) The data processing method described in (3) above, further comprising: accepting input of a setting for a determination condition for whether or not to perform the feedback control. (5) The data processing method according to (3) or (4) above, further including accepting input of update settings for the clustering parameters. (6) The data processing method according to any one of (3) to (5) above, further including changing a drawing method of the environmental map to a method suitable for the environmental scene based on the determination result of the environmental scene. (7) The data processing method according to any one of (3) to (6) above, further including accepting input of control information for controlling the action plan in a specific area in which the mobile object is movable, determined based on information about the environmental map. (8) The data processing method according to any one of (1) to (7) above, further including correcting the determination result of the environmental scene based on time-series data indicating past determination results of the environmental scene.(9) The data processing method according to any one of (1) to (8), wherein at least one piece of information among a normal direction of each cluster, variance information of each cluster, and a size of each cluster is extracted as the cluster feature. (10) The data processing method according to any one of (1) to (9), further comprising: extracting information of neighboring clusters present around any of the clusters from which the cluster feature has been extracted; and, when calculating the environment geometric feature, calculating the environment geometric feature using the cluster feature of each cluster and the information of the neighboring clusters. (11) The data processing method according to any one of (1) to (10), further comprising: expressing the environment geometric feature as a one-dimensional numerical value or two-dimensional data. (12) A mobile object control system comprising: a cluster generation unit that clusters environmental data around a mobile object into a plurality of clusters based on clustering parameters, a cluster feature extraction unit that extracts cluster features of each of the clustered clusters, an environmental geometric feature calculation unit that calculates geometric features of the environment based on the cluster features of each of the clusters, a feedback processing unit that performs feedback control to update the clustering parameters based on the geometric features of the environment, an environmental scene determination unit that determines an environmental scene based on the geometric features of the environment, and a behavior plan processing unit that generates a behavior plan for the mobile object based on a result of the environmental scene determination. (13) The mobile object control system according to (12), further comprising: an environmental map generation unit that generates an environmental map based on the environmental data, an environmental map drawing device that draws information of the generated environmental map so that it is visible, and an input unit that receives input of control information for controlling the behavior plan, the control information being generated based on information of the environmental map.(14) A program that causes a computer to execute a process including: clustering environmental data around a moving object into a plurality of clusters based on clustering parameters; extracting cluster features of each of the clustered clusters; calculating geometric features of the environment based on the cluster features of each of the clusters; performing feedback control to update the clustering parameters based on the geometric features of the environment; and determining an environmental scene based on the geometric features of the environment.

[0124] This application claims priority based on Japanese Patent Application No. 2024-090312, filed on June 3, 2024, in the Japan Patent Office, the entire contents of which are incorporated herein by reference.

[0125] Those skilled in the art will recognize that various modifications, combinations, subcombinations, and variations may occur depending on design requirements and other factors, and are intended to be within the scope of the appended claims and their equivalents.

Claims

1. A data processing method comprising: clustering environmental data around a moving object into a plurality of clusters based on clustering parameters; extracting cluster features of each of the clustered clusters; calculating geometric features of the environment based on the cluster features of each of the clusters; performing feedback control to update the clustering parameters based on the geometric features of the environment; and determining an environmental scene based on the geometric features of the environment.

2. The data processing method according to claim 1, further comprising generating a behavior plan for the moving object based on the determination result of the environmental scene.

3. The data processing method according to claim 2, further comprising: generating an environmental map based on the environmental data; drawing the information of the generated environmental map so that it is visible; and accepting input of control information for controlling the action plan, the control information being generated based on the information of the environmental map.

4. The data processing method according to claim 3, further comprising: receiving an input of a setting of a determination condition for whether or not to perform the feedback control.

5. The data processing method according to claim 3, further comprising: receiving an input of update settings for the clustering parameters.

6. The data processing method according to claim 3, further comprising: changing a method for drawing the environmental map to a method suitable for the environmental scene based on the determination result of the environmental scene.

7. The data processing method according to claim 3, further comprising: accepting input of control information for controlling the action plan in a specific area in which the mobile body can move, determined based on information on the environmental map.

8. The data processing method according to claim 1, further comprising: correcting the environmental scene determination result based on time-series data indicating past environmental scene determination results.

9. The data processing method according to claim 1, wherein at least one of information on the normal direction of each of the clusters, information on the variance of each of the clusters, and information on the size of each of the clusters is extracted as the cluster feature.

10. The data processing method according to claim 1, further comprising: extracting information on neighboring clusters existing around any of the clusters from which the cluster features have been extracted; and when calculating the environmental geometric features, calculating the environmental geometric features using the cluster features of each of the clusters and the information on the neighboring clusters.

11. The data processing method according to claim 1, further comprising expressing the environmental geometric features as one-dimensional numerical values ​​or two-dimensional data.

12. A mobile object control system comprising: a cluster generation unit that clusters environmental data around a mobile object into a plurality of clusters based on clustering parameters; a cluster feature extraction unit that extracts cluster features of each of the clustered clusters; an environmental geometric feature calculation unit that calculates geometric features of the environment based on the cluster features of each of the clusters; a feedback processing unit that performs feedback control to update the clustering parameters based on the geometric features of the environment; an environmental scene determination unit that determines an environmental scene based on the geometric features of the environment; and a behavior plan processing unit that generates a behavior plan for the mobile object based on the results of the environmental scene determination.

13. The mobile object control system of claim 12, further comprising: an environmental map generation unit that generates an environmental map based on the environmental data; an environmental map drawing device that draws the information of the generated environmental map so that it can be seen; and an input unit that receives input of control information for controlling the action plan, generated based on the information of the environmental map.

14. A program that causes a computer to execute processes including: clustering environmental data around a moving object into a plurality of clusters based on clustering parameters; extracting cluster features of each of the clustered clusters; calculating geometric features of the environment based on the cluster features of each of the clusters; performing feedback control to update the clustering parameters based on the geometric features of the environment; and determining the environmental scene based on the geometric features of the environment.

Images