Map data generation method, map data generation device, and map data generation program

Abstract

To generate map data with high accuracy in a case where the map data is generated from travel log data.SOLUTION: A plurality of data sets 20A and 20B are generated from a travel log data group 10. Each of the data sets 20A and 20B includes a plurality of pieces of travel log data 12. Pieces of map data 30A and 30B for evaluation are generated from each of data sets 20A and 20B. Abnormal map data 30B is identified by comparing the pieces of map data 30A and 30B for evaluation. Also, abnormal travel log data 12a is identified by comparing the data set 20B used to generate the abnormal map data 30B with the data set 20A used to generate normal map data 30A for evaluation. Then, final map data 30C for evaluation is generated using a data set 20C that does not include the abnormal travel log data 12a.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present disclosure relates to a method, an apparatus, and a program for generating map data from driving log data. [Background technology]

[0002] Patent Document 1 discloses a road information generating device. This road information generating device sets multiple combinations of travel data with the same road ID for multiple travel data (travel log data) stored in a travel data storage unit. Next, for each combination of travel data, the road information generating device generates a road map for the road ID based on the combined travel data. Then, the road information generating device calculates the reliability of the road map for each road map generated for the road ID.

[0003] However, the acquired travel data may contain abnormal travel data. The road information generating device disclosed in Patent Document 1 sets a combination of travel data with the same road ID without judging whether the travel data is abnormal. Therefore, if abnormal travel data is included in a combination of travel data with the same road ID, the accuracy of the map generated for that road ID will be affected by the abnormal travel data.

[0004] In addition to the above-mentioned Patent Document 1, the following Patent Documents 2 and 3 can be mentioned as examples of documents showing the state of the art in the technical field related to the present disclosure. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2016-180797 A [Patent Document 2] JP 2012-520491 A [Patent Document 3] JP 2014-160064 A Summary of the Invention [Problem to be solved by the invention]

[0006] The present disclosure has been made in consideration of the above-mentioned problems, and has an object to provide a technique capable of generating map data with high accuracy when creating map data from driving log data. [Means for solving the problem]

[0007] The present disclosure provides a map data generation method, a map data generation device, and a map data generation program as a map data generation technique for achieving the above object.

[0008] The map data generation method of the present disclosure includes the following steps. The first step is to generate a plurality of data sets from a group of driving log data. Each of the plurality of data sets is composed of one or more driving log data. The second step is to generate map data for evaluation from each of the plurality of data sets. The third step is to identify abnormal map data included in the plurality of map data for evaluation by comparing the plurality of map data for evaluation generated from the plurality of data sets with each other. The fourth step is to identify abnormal driving log data by comparing a dataset used to create the abnormal map data with a dataset used to create map data for evaluation other than the abnormal map data. And the fifth step is to generate final map data using a dataset that does not include abnormal driving log data.

[0009] The map data generating device of the present disclosure includes at least one processor and a program memory coupled to the at least one processor. The program memory stores a plurality of executable instructions. The plurality of executable instructions are configured to cause the at least one processor to execute the following processes: A first process is to generate a plurality of data sets from a group of driving log data. Each of the plurality of data sets is composed of one or more driving log data. A second process is to generate map data for evaluation from each of the plurality of data sets. A third process is to identify abnormal map data included in the plurality of map data for evaluation by comparing the plurality of map data for evaluation generated from the plurality of data sets with each other. A fourth process is to identify abnormal driving log data by comparing a dataset used to create the abnormal map data with a dataset used to create map data for evaluation other than the abnormal map data. And a fifth process is to generate final map data using a dataset that does not include abnormal driving log data.

[0010] The map data generation program of the present disclosure is configured to cause a computer to execute the following processes. The first process is to generate a plurality of data sets from a group of driving log data. Each of the plurality of data sets is composed of one or more driving log data. The second process is to generate map data for evaluation from each of the plurality of data sets. The third process is to identify abnormal map data included in the plurality of map data for evaluation by comparing the plurality of map data for evaluation generated from the plurality of data sets with each other. The fourth process is to identify abnormal driving log data by comparing a dataset used to create the abnormal map data with a dataset used to create map data for evaluation other than the abnormal map data. And the fifth process is to generate final map data using a dataset that does not include abnormal driving log data. The map data generation program of the present disclosure may be recorded on a computer-readable recording medium or provided via a network.

[0011] According to the map data generation technique disclosed herein, i.e., the map data generation method, the map data generation device, and the map data generation program, a data set that does not include abnormal driving log data is used to generate map data, making it possible to generate map data with high accuracy.

[0012] In the map data generation technology of the present disclosure, each of the multiple driving log data included in the driving log data group may be defined by multiple parameters. The conditions of at least one of the multiple parameters defining the driving log data may be made uneven among the multiple driving log data included in the multiple data sets. In this way, when an abnormality in the driving log data depends on the parameter conditions, abnormal driving log data can be found out from the driving log data group.

[0013] In the map data generation technology of the present disclosure, a condition that causes an abnormality in the map data among the conditions of a plurality of parameters that define the driving log data may be identified based on the abnormal driving log data. Then, final map data may be generated using a dataset that does not include a condition that causes an abnormality in the map data among the conditions of a plurality of parameters that define the driving log data. In this way, when an abnormality in the driving log data depends on the parameter conditions, the abnormal driving log data can be excluded from the dataset used to generate the map data.

[0014] In the map data generation technology of the present disclosure, a combination of conditions that causes an abnormality in the map data may be identified from among the combinations of conditions of a plurality of parameters that define the driving log data. This can be done by comparing a dataset used to create the abnormal map data with a dataset used to create map data for evaluation other than the abnormal map data. Then, the final map data may be generated using a dataset that does not include a combination of conditions that causes an abnormality in the map data among the combinations of conditions of a plurality of parameters that define the driving log data. In this way, when an abnormality in the driving log data depends on a combination of parameter conditions, it is possible to exclude a combination of driving log data that causes an abnormality in the map data from the dataset used to generate the map data.

[0015] In the map data generation technique of the present disclosure, the map data may be map data of a feature map. In this case, comparing the plurality of evaluation map data with each other may include comparing the variation of the feature between the plurality of evaluation map data. Alternatively, comparing the plurality of evaluation map data with each other may include comparing the success degree of self-location estimation using the feature between the plurality of evaluation map data.

[0016] In the map data generation technique of the present disclosure, the map data may be map data of a road surface shape map. In this case, comparing the plurality of evaluation map data with each other may include comparing the height or road surface gradient between the plurality of evaluation map data.

[0017] In the map data generation technique of the present disclosure, the map data may be map data of a road surface luminance map. In this case, comparing the plurality of map data for evaluation with each other may include comparing the luminance or luminance gradient between the plurality of map data for evaluation.

[0018] In the map data generation technique of the present disclosure, the map data may be map data of a static obstacle map. In this case, comparing the plurality of evaluation map data with each other may include comparing positions of voxels between the plurality of evaluation map data. Effect of the Invention

[0019] As described above, according to the map data generation technique of the present disclosure, map data can be generated with high accuracy by generating map data using a data set that does not include abnormal driving log data. [Brief description of the drawings]

[0020] [Figure 1] FIG. 4 is a conceptual diagram for explaining a first problem associated with generating map data using driving log data. [Diagram 2] FIG. 2 is a conceptual diagram for explaining a means for solving the first problem, and is a conceptual diagram for explaining an overview of a map data generating method according to an embodiment of the present disclosure. [Diagram 3] FIG. 11 is a conceptual diagram for explaining a second problem associated with generating map data using driving log data. [Figure 4] FIG. 11 is a conceptual diagram for explaining a means for solving the second problem, and is a conceptual diagram for explaining an overview of a map data generating method according to an embodiment of the present disclosure. [Diagram 5]FIG. 2 is a block diagram illustrating functions of a map data generating device according to an embodiment of the present disclosure. [Figure 6] FIG. 2 is a block diagram showing an example of a hardware configuration of a map data generating device according to an embodiment of the present disclosure. [Figure 7] FIG. 2 is a conceptual diagram for explaining a first example of a map data generating method implemented by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram showing an example of the configuration of a driving log data database in the first example. [Figure 8] FIG. 2 is a conceptual diagram for explaining a first example of a map data generating method implemented by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining a log data combination generation process in the first example. [Figure 9] FIG. 2 is a conceptual diagram for explaining a first example of a map data generation method implemented by a map data generation device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining a map data generation process in the first example. [Figure 10] FIG. 2 is a conceptual diagram for explaining a first example of a map data generating method implemented by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining a map data evaluation process in the first example. [Figure 11] FIG. 2 is a conceptual diagram for explaining a first example of a map data generating method implemented by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining an abnormal log data determination process in the first example. [Figure 12] FIG. 2 is a diagram for explaining a first example of a map data generation method implemented by a map data generation device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining a final map data generation process in the first example. [Figure 13] 4 is a conceptual diagram for explaining a format of driving log data stored in a driving log data database. FIG. [Figure 14] 4 is a conceptual diagram for explaining a format of driving log data stored in a driving log data database. FIG. [Figure 15]11 is a conceptual diagram for explaining a format of a combination of driving log data in a log data combination generation process. FIG. [Figure 16] FIG. 11 is a conceptual diagram for explaining a second example of a map data generating method implemented by a map data generating device according to an embodiment of the present disclosure, illustrating a configuration example of a driving log data database in the second example. [Figure 17] FIG. 11 is a conceptual diagram for explaining a second example of a map data generating method implemented by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining a log data combination generation process in the second example. [Figure 18] FIG. 11 is a conceptual diagram for explaining a second example of a map data generation method implemented by a map data generation device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining a map data generation process in the second example. [Figure 19] FIG. 11 is a conceptual diagram for explaining a second example of a map data generating method implemented by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining a map data evaluation process in the second example. [Figure 20] FIG. 11 is a conceptual diagram for explaining a second example of a map data generating method implemented by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining an abnormal log data determination process in the second example. [Figure 21] FIG. 11 is a conceptual diagram for explaining a second example of a map data generating method implemented by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining an abnormal parameter condition determination process in the second example. [Figure 22] FIG. 11 is a conceptual diagram for explaining a second example of a map data generation method implemented by a map data generation device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining a final map data generation process in the second example. [Figure 23]FIG. 11 is a conceptual diagram for explaining a third example of a map data generating method implemented by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram showing an example of the configuration of a driving log data database in the third example. [Figure 24] FIG. 13 is a conceptual diagram for explaining a third example of a map data generation method implemented by a map data generation device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining a log data combination generation process in the third example. [Diagram 25] FIG. 11 is a conceptual diagram for explaining a third example of a map data generation method implemented by a map data generation device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining a map data generation process in the third example. [Figure 26] FIG. 11 is a conceptual diagram for explaining a third example of a map data generating method implemented by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining a map data evaluation process in the third example. [Figure 27] FIG. 13 is a conceptual diagram for explaining a third example of a map data generating method implemented by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining an abnormal log data determination process in the third example. [Figure 28] FIG. 13 is a conceptual diagram for explaining a third example of a map data generating method implemented by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining an abnormal combination determination process in the third example. [Figure 29] FIG. 11 is a conceptual diagram for explaining a third example of a map data generation method implemented by a map data generation device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining a final map data generation process in the third example. [Diagram 30] 10A to 10C are diagrams for explaining a specific example of a map data evaluation process performed by a map data generating device according to an embodiment of the present disclosure, and are diagrams showing an example of feature amount detection. [Diagram 31] FIG. 11 is a diagram for explaining a specific example of a map data evaluation process performed by the map data generating device according to an embodiment of the present disclosure, and is a diagram showing examples of types of feature amounts. [Diagram 32] FIG. 1 is a diagram for explaining a specific example of a map data evaluation process performed by a map data generation device according to an embodiment of the present disclosure, and is a diagram showing an example of generation of a feature amount map by SLAM. [Diagram 33] 10A and 10B are diagrams for explaining a specific example of a map data evaluation process performed by a map data generating device according to an embodiment of the present disclosure, and are diagrams showing examples of variations for each feature amount. [Diagram 34] 10 is a diagram for explaining a specific example of a map data evaluation process performed by the map data generating device according to an embodiment of the present disclosure, and is a diagram showing an example of statistics for each map data. FIG. [Diagram 35] FIG. 13 is a diagram for explaining a specific example of map data evaluation processing by the map data generating device according to an embodiment of the present disclosure, and is a diagram showing an example of evaluation using an F-test. [Diagram 36] FIG. 13 is a conceptual diagram for explaining a specific example of a map data evaluation process performed by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining an example of self-position estimation in a pose using SLAM. [Figure 37] FIG. 13 is a conceptual diagram for explaining a specific example of a map data evaluation process performed by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining a successful example of self-position estimation in a pose using SLAM. [Figure 38] FIG. 13 is a conceptual diagram for explaining a specific example of a map data evaluation process performed by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining an example of a failure of self-position estimation in a pose using SLAM. [Figure 39] FIG. 13 is a diagram for explaining a specific example of a map data evaluation process performed by a map data generation device according to an embodiment of the present disclosure, and is a diagram showing an example in which the success rate of self-location estimation by SLAM in each pose is visualized on a feature map. [Diagram 40] FIG. 13 is a diagram for explaining a specific example of map data evaluation processing by the map data generating device according to an embodiment of the present disclosure, and is a diagram showing an example of correlation of changes in p-values ​​between map data. [Diagram 41]FIG. 13 is a diagram for explaining a specific example of the map data evaluation process by the map data generating device according to the embodiment of the present disclosure, and is a diagram showing another example of the correlation of the fluctuation of p-values ​​between map data. [Diagram 42] FIG. 42 is a diagram for explaining a specific example of the map data evaluation process by the map data generation device according to an embodiment of the present disclosure, and is a diagram showing an example of the average value of p-value for each map data in the predetermined section shown in FIG. 41. [Diagram 43] FIG. 13 is a diagram for explaining a specific example of a map data evaluation process performed by the map data generating device according to an embodiment of the present disclosure, and is a diagram showing an example of a correlation coefficient of p-values ​​between map data. [Diagram 44] FIG. 11 is a conceptual diagram for explaining a specific example of a map data evaluation process by a map data generating device according to an embodiment of the present disclosure, and is a conceptual diagram for explaining an example of map data evaluation using a trajectory after SLAM. [Diagram 45] FIG. 1 is a conceptual diagram for explaining a specific example of a map data evaluation process performed by a map data generating device according to an embodiment of the present disclosure, and is a diagram showing an example of a road surface shape map. [Diagram 46] FIG. 13 is a diagram for explaining a specific example of map data evaluation processing by the map data generating device according to the embodiment of the present disclosure, showing an example in which SLAM was successful and an example in which it was unsuccessful on a road surface shape map. [Figure 47] FIG. 13 is a diagram for explaining a specific example of a map data evaluation process by the map data generating device according to an embodiment of the present disclosure, and is a diagram showing an example of correlation of variation in variance in the height direction between map data. [Figure 48] FIG. 13 is a diagram for explaining a specific example of map data evaluation processing by the map data generating device according to an embodiment of the present disclosure, and is a diagram showing an example of correlation of fluctuations in the number of LiDAR point clouds between map data. [Figure 49] 1A and 1B are diagrams for explaining a specific example of map data evaluation processing by a map data generating device according to an embodiment of the present disclosure, and are diagrams showing examples of images of road surface shapes on a road surface shape map. [Figure 50]FIG. 10 is a diagram for explaining a specific example of a map data evaluation process performed by the map data generating device according to an embodiment of the present disclosure, and is a diagram showing an example of correlation of changes in road surface gradient between map data. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0021] Hereinafter, the embodiments of the present disclosure will be described with reference to the drawings. However, when the number, quantity, amount, range, etc. of each element is mentioned in the embodiments shown below, the technical idea of ​​the present disclosure is not limited to the mentioned number unless it is specifically stated or clearly specified in principle. Furthermore, the structures, etc. described in the embodiments shown below are not necessarily essential to the technical idea of ​​the present disclosure unless it is specifically stated or clearly specified in principle.

[0022] 1. Overview of map data generation method In this embodiment, map data refers to data constituting a map for autonomous driving by an autonomous vehicle, such as a feature map, a terrain map, a road surface intensity map, and a stationary obstacle map (background knowledge). The feature map is typically used for self-location estimation of an autonomous vehicle. The road surface shape map is a map in which the shape (height) of the road surface of the surrounding area where the autonomous vehicle is traveling is recorded in cells. The road surface intensity map is a map in which the brightness of the road surface of the surrounding area where the autonomous vehicle is traveling is recorded in cells. The stationary obstacle map is a map in which stationary obstacles such as road structures are recorded in voxels.

[0023] A large amount of driving log data obtained by actually driving a vehicle equipped with sensors is used to create the map data. The entire set of driving log data used to create the map data is called a driving log data group. Driving log data may also be abbreviated to log data. There are several issues involved in creating map data using driving log data. Figure 1 is a conceptual diagram for explaining a first issue related to the generation of map data using log data.

[0024] FIG. 1 shows two examples in which map data is generated using a driving log data group 10. In the example on the left, a data set 20A, which is a collection of multiple log data 12, is generated from the driving log data group 10. Then, map data 30A is generated using the data set 20A. In the example on the right, a data set 20B, which is a collection of multiple log data 12, is generated from the driving log data group 10. Then, map data 30B is generated using the data set 20B. The log data 12 constituting each of the data sets 20A, 20B may be arbitrarily extracted from the driving log data group 10 or may be extracted according to a predetermined selection condition.

[0025] Here, assume that all of the log data 12 included in the data set 20A are normal, whereas the log data 12 included in the data set 20B includes abnormal log data 12a. In this case, the map data 30A generated from the data set 20A is normal, whereas the map data 30B generated from the data set 20B is abnormal.

[0026] One method for determining whether map data is normal or abnormal is a statistical determination. In FIG. 1, only two pieces of map data 30A, 30B are generated, but in reality, a large number of map data can be generated from the driving log data group 10. If there is a group of map data in which the degree of variation in element data constituting the map data is roughly the same (which can also be said to be a group of map data in which the variation in element data falls within a certain range), the map data that belongs to that group is considered to be statistically normal, and map data that deviates from that group is considered to be abnormal. In the example shown in FIG. 1, map data 30A is map data that belongs to the group of normal map data, and map data 30B is map data that deviates from the group of normal map data.

[0027] The abnormal log data 12a varies greatly from the other log data 12, and is log data that causes the map data 30B to deviate from the group of normal map data. Therefore, in order to generate highly accurate map data, it is desirable to eliminate the abnormal log data 12a in advance. However, since the number of log data included in the driving log data group 10 is enormous, it is not easy to extract the abnormal log data 12a from the enormous amount of log data. In addition, whether or not the log data is abnormal depends on whether or not the map data becomes abnormal when the map data is generated using the log data. Therefore, it is not easy to set a judgment criterion for judging in advance whether or not the log data is abnormal.

[0028] The map data generating method according to this embodiment provides a means for solving the first problem related to the creation of map data. Fig. 2 is a conceptual diagram for explaining an overview of the map data generating method according to this embodiment.

[0029] According to the map data generation method of this embodiment, a plurality of data sets 20A, 20B are generated from the travel log data group 10. The log data constituting each of the data sets 20A, 20B may be arbitrarily extracted from the travel log data group 10 or may be extracted according to a predetermined selection condition. Although only two sets of data sets 20A, 20B are generated in FIG. 2, a larger number of data sets are actually generated. The number of log data constituting each data set is preferably multiple, but may be one. Also, the number of log data may be the same or different between the data sets.

[0030] Next, evaluation map data 30A, 30B are generated from the generated data sets 20A, 20B, respectively. Then, the evaluation map data 30A, 30B are compared with each other to identify abnormal map data 30B contained in the evaluation map data 30A, 30B. Note that, although two evaluation map data 30A, 30B are generated in FIG. 2, a large number of evaluation map data are actually generated. As described above, abnormal map data is map data that falls outside the group of map data in which the degree of variation of element data constituting the map data is roughly consistent. As a result of the comparison, it is determined that all of the evaluation map data is normal or that some of it is abnormal.

[0031] If abnormal map data 30B is identified, the data set 20B used to create the abnormal map data 30B is compared with the data set 20A used to create the evaluation map data 30A that is determined to be normal. If a large number of evaluation map data have been created, the data set used to create the abnormal map data is compared with the data set used to create evaluation map data other than the abnormal map data.

[0032] By comparing the data set 20A with the data set 20B, the abnormal log data 12a included in the data set 20B is identified. In other words, by using the log data 12 included in the data set 20A as a reference, the log data 12a different from the log data 12 included in the data set 20A is identified as abnormal log data, that is, log data causing an abnormality in the map data 30B. It is not easy to extract abnormal log data in advance from the vast amount of log data included in the driving log data group 10. However, if a method is used that compares a data set used to create the abnormal map data with a data set used to create map data for evaluation other than the abnormal map data, the abnormal log data can be easily identified.

[0033] If abnormal log data 12a is identified, the abnormal log data 12a is removed from the dataset 20B. Then, a new dataset 20C is generated by combining the dataset 20A with the dataset 20B from which the abnormal log data 12a has been removed, and final map data 30C is generated from the dataset 20C. If a large number of map data for evaluation have been created, the abnormal log data is removed from each of the datasets used to create the abnormal map data. Then, a final dataset that does not include abnormal log data is generated, and final map data is generated from the final dataset. By generating map data using a dataset that does not include abnormal log data, it is possible to generate map data with high accuracy.

[0034] As described above, according to the map data generation method of this embodiment, abnormal log data that causes a decrease in accuracy of map data can be removed from the data set used to generate map data. However, abnormal log data is not the only cause of a decrease in accuracy of map data. This is the second problem associated with generating map data using log data. Figure 3 is a conceptual diagram for explaining the second problem associated with generating map data using log data.

[0035] FIG. 3 shows three examples of map data being generated using the driving log data group 10. In the example on the left, a data set 20D, which is a collection of multiple log data 14, is generated from the driving log data group 10. Then, map data 30D is generated using the data set 20D. In the example in the center, a data set 20E, which is a collection of multiple log data 16, is generated from the driving log data group 10. Then, map data 30E is generated using the data set 20E. In the example on the right, a data set 20F, which is a collection of multiple log data 14, 16, is generated from the driving log data group 10. Then, map data 30F is generated using the data set 20F. The log data is defined by multiple parameters. There is a difference in the condition of at least one parameter between the log data 14 and the log data 16.

[0036] Differences in parameter conditions may cause significant differences in the characteristics and trends of log data. For example, when parameters are weather, time of day, the presence or absence of a specific external sensor, differences in parameter conditions significantly affect the characteristics and trends of log data. If there is a significant difference in characteristics and trends between the log data 14 and the log data 16, the map data 30D generated from the log data 14 and the map data 30E generated from the log data 16 belong to different map data groups. Here, it is assumed that the log data 14 and 16 included in each data set 20D and 20E are all normal. The map data 30D generated from normal log data 14 is normal, and the map data 30E generated from normal log data 16 is also normal.

[0037] However, even if both the log data 14 and the log data 16 included in the data set 20F are normal, the map data 30F generated from the log data 14 and the log data 16 may be abnormal. If there is a significant difference in characteristics or trends between the log data 14 and the log data 16, it can be considered that the log data 16 is abnormal log data from the perspective of the log data 14, and that the log data 14 is abnormal log data from the perspective of the log data 16. Therefore, the map data 30F generated from the log data 14 and the log data 16, which have different parameter conditions, may be abnormal map data that does not belong to any group of map data.

[0038] In order to generate highly accurate map data, it is desirable to eliminate in advance any combination of parameter conditions that may cause anomalies in the map data. However, since there are an enormous number of combinations of parameter conditions that define the log data, it is not easy to determine in advance which combinations of parameter conditions will cause anomalies in the map data.

[0039] The map data generating method according to this embodiment provides a means for solving the second problem related to the creation of map data. Fig. 4 is a conceptual diagram for explaining an overview of the map data generating method according to this embodiment.

[0040] According to the map data generating method of this embodiment, a plurality of data sets 20D, 20E, and 20F are generated from the travel log data group 10. The log data 14 constituting the data set 20D and the log data 16 constituting the data set 20E have a difference in at least one parameter condition. The data set 20F is composed of the log data 14 and the log data 16 having a difference in the parameter condition. Note that only three sets of data sets 20D, 20E, and 20F are generated in FIG. 4, but in reality, a larger number of data sets are generated. The number of log data constituting each data set is preferably multiple, but may be one. Also, the number of log data may be the same or different between the data sets.

[0041] Next, evaluation map data 30D, 30E, 30F are generated from the generated data sets 20D, 20E, 20F, respectively. Then, the evaluation map data 30D, 30E, 30F are compared with each other to identify abnormal map data 30F contained in the evaluation map data 30D, 30E, 30F. Note that although the evaluation map data generated in FIG. 4 is three pieces of map data 30D, 30E, 30F, in reality, a large number of evaluation map data are generated. The map data 30F determined to be abnormal is map data that is outside the group of map data to which the map data 30D belongs and the group of map data to which the map data 30E belongs.

[0042] If abnormal map data 30F is identified, the data set 20F used to create the abnormal map data 30F is compared with the data sets 20D, 20E used to create the evaluation map data 30D, 30E that are determined to be normal. If a large number of evaluation map data have been created, the data set used to create the abnormal map data is compared with the data set used to create evaluation map data other than the abnormal map data.

[0043] By comparing the data sets 20D and 20E with the data set 20F, log data having different parameter conditions included in the data set 20F, i.e., the log data 14 and the log data 16, are identified. Then, by comparing the log data 14 with the log data 16, a combination of conditions that causes an abnormality in the map data 30E is identified among the combinations of the conditions of the multiple parameters that define the log data. It is not easy to determine in advance what combination of parameter conditions causes an abnormality in the map data. However, if a method is used that compares a data set used to create the abnormal map data with a data set used to create evaluation map data other than the abnormal map data, it is possible to easily identify the combination of parameter conditions that causes an abnormality in the map data.

[0044] When a combination of parameter conditions that causes an abnormality in the map data is identified, the final map data is generated using a data set that does not include such a combination. In the example shown in Fig. 4, the map data 30D generated using the data set 20D made up of the log data 14 and the map data 30E generated using the data set 20E made up of the log data 16 are generated as the final map data. The map data 30D and the map data 30E are used differently depending on the difference in the parameter conditions. This makes it possible to generate map data with high accuracy by eliminating the influence of the combination of parameter conditions that define the log data.

[0045] 2.Map data generator Next, a map data generating device for implementing the map data generating method according to this embodiment will be described below. Fig. 5 is a block diagram showing the functions of the map data generating device 100 according to this embodiment.

[0046] The map data generation device 100 includes a database and a processing unit. The databases constituting the map data generation device 100 are a driving log data database (hereinafter referred to as driving log data DB) 110 and a map data database (hereinafter referred to as map data DB) 160. The processing units constituting the map data generation device 100 are a log data combination generation unit 120, a map data generation unit 130, a map data evaluation unit 140, and an abnormality log data determination unit 150.

[0047] The driving log data DB110 is a database in which a large amount of log data is stored. Log data acquired by actually driving a vehicle is accumulated in the driving log data DB110 to form a group of driving log data. The vehicle used to acquire the log data may be an autonomous vehicle or a vehicle driven by a driver. However, it is preferable that the type, model, number, and installation position of the external sensors (LiDAR, camera, depth sensor, etc.) used to acquire the log data are the same as those of the autonomous vehicle in which the map data is used in the actual autonomous driving.

[0048] The log data includes data from various sensors and output data from each process of autonomous driving. The various sensors include external sensors as well as GPS and IMU (Inertial Measurement Unit). The autonomous driving process includes, for example, self-location estimation results, object detection results, and route planning results.

[0049] Log data is specified by multiple parameters. Examples of parameters that specify log data and their conditions are as follows. One or more of these parameters are used to specify the log data. Note that sensor-related parameters are set for each sensor. Sensors of the same type but with different model numbers or installed in different locations are treated as different sensors. Car model name: "Prius, e-Palette, Aqua, etc." "8A2B, 300C, 405D, etc." Vehicle name: "Car No. 1, Car No. 2, Car No. 3, etc." "Alice, Belle, Cindy, etc." Date and time: “August 12, 2021 12:00, August 12, 2021 14:00, etc.” Total distance traveled: "10km, 5km, etc." Time: "1 hour 21 minutes 14 seconds, 2 hours 3 minutes 43 seconds, etc." Weather: "Sunny, rainy, cloudy, snowy, foggy, etc." Temperature: "30°C, 20°C, etc." "86°F, 68°F, etc." Sunshine amount: "0MJ / m 2 , 1.0MJ / m 2 , etc." Rainfall: "0mm / h, 1mm / h, etc." "0mm, 2mm, etc." Snowfall amount: "0mm / h, 1mm / h, etc." "0mm, 2mm, etc." Operator name: "Taro Tanaka, Hanako Yamada, etc." "No. 101, No. 203, etc." Number of passengers: "0 people, 1 person, 5 people, etc." "Empty, available" Vehicle speed: "Maximum vehicle speed 20km / h, 40km / h, etc." "Average vehicle speed 10km / h, 20km / h, etc." Sensor presence: "Present, Absent, Fault" Sensor version: "Ver.1, Ver.2, etc." "Prototype, quality assured, mass-produced" Autonomous driving software version: "Ver.1,Ver.2,etc." "master, perception_test, planner_test,etc." Operation method: "Manual, automatic, partially automatic"

[0050] The log data combination generating unit 120 generates a combination of multiple log data, that is, a data set, from the log data stored in the driving log data DB 110. The combination of log data constituting a data set is different for each data set. However, the same log data may be included in multiple data sets. In addition, the data range of the log data read from the driving log data DB 110 can be limited by the above parameters. Examples of the data range that limits the log data include space, section, time, weather, temperature, vehicle, sensor type, etc. For example, the data range can be limited to "data collected by sensor D of vehicle A during the day on a sunny day" and a data set can be generated by combining log data in such a data range.

[0051] The map data generation unit 130 generates map data for evaluation using the data sets generated by the log data combination generation unit 120. The map data generation unit 130 generates one piece of map data for evaluation for one data set. In this way, the map data generation unit 130 generates a plurality of map data for evaluation, the number of which is the same as the number of data sets generated by the log data combination generation unit 120.

[0052] The map data evaluation unit 140 compares and evaluates multiple map data generated by the map data generation unit 130. Because a map basically has a one-to-one relationship with the real environment, the map data should converge uniquely regardless of the combination of log data. However, if a data set contains abnormal log data, the map data itself generated from that data set will also contain abnormal values. Therefore, the map data evaluation unit 140 detects abnormal map data that is different from the others by comparing multiple map data generated from different data sets.

[0053] The abnormal log data determination unit 150 determines abnormal log data based on the evaluation result of the map data evaluation unit 140. In particular, a data set used to create the abnormal map data is compared with a data set used to create map data for evaluation other than the abnormal map data. This comparison identifies abnormal driving log data from the data set used to create the abnormal map data. In addition to identifying abnormal log data, the abnormal log data determination unit 150 can also identify conditions (e.g., weather, time of day, sensor failure, etc.) that cause abnormality in the map data among the conditions of each of the multiple parameters that define the log data.

[0054] The determination result of the abnormal log data determination unit 150 is sent to the log data combination generation unit 120. The log data combination generation unit 120 removes the abnormal log data identified by the abnormal log data determination unit 150 from the multiple data sets initially generated, and combines the data sets from which the abnormal log data has been removed to generate a final data set.

[0055] The map data generation unit 130 generates final map data using the final data set generated by the log data combination generation unit 120, i.e., a data set that does not include abnormal log data. The map data generation unit 130 stores the generated final map data in the map data DB 160.

[0056] Next, an example of the hardware configuration of the map data generating device 100 for realizing the above-mentioned functions will be described with reference to Fig. 6. Fig. 6 is a block diagram showing an example of the hardware configuration of the map data generating device 100 according to this embodiment.

[0057] 6, the map data generating device 100 includes a processor 101, a memory 102, a storage 104, a communication module 105, and a user interface 106. These elements constituting the map data generating device 100 are coupled by a bus 107.

[0058] The processor 101 performs calculations for generating map data. The map data generating device 100 may be provided with one or more processors 101 (here, it is assumed that one processor is provided).

[0059] The memory 102 is a non-transitory program memory that stores a program 103. The program 103 is a computer-executable program (map data generation program) for implementing the map data generation method according to this embodiment. The program 103 includes a plurality of instructions for causing the processor 101 to execute processes. When executed by the processor 101, the program 103 causes the processor 101 to function as a log data combination generation unit 120, a map data generation unit 130, a map data evaluation unit 140, and an abnormal log data determination unit 150.

[0060] The storage 104 is, for example, an SSD or an HDD. A driving log data DB 110 and a map data DB 160 are constructed in the storage 104. The storage in which the driving log data DB 110 is provided and the storage in which the map data DB 160 is provided may be the same or different. The driving log data DB 110 may be provided outside the map data generating device 100. The map data DB 160 may also be provided outside the map data generating device 100.

[0061] The communication module 105 is provided for communication with an external device. The communication method by the communication module 105 may be wireless communication or wired communication. The storage of log data in the driving log data DB 110 and the reading of map data from the map data DB 160 are performed using the communication module 105. The user interface 106 is provided for inputting operations by the operator of the map data generating device 100 and outputting information to the operator.

[0062] 3. Specific examples of map data generation methods 3-1. First Example Next, a specific example of a map data generation method implemented by the map data generation device 100 according to this embodiment will be described. Figures 7 to 12 are conceptual diagrams for explaining a first example of the map data generation method. Note that the first example described below corresponds to the above-mentioned first problem and its solving means.

[0063] 7 is a conceptual diagram showing an example of the configuration of the driving log data DB 110 in the first example. In the first example, the parameters that define the log data are the vehicle name, the date and time, and four pieces of log data, log data 1 to log data 4, are stored in the driving log data DB 110. The contents of the log data differ depending on the type of map data that is ultimately generated. The map data generation method according to this embodiment is widely applicable to map data generated from log data, so the contents of the log data are not limited here.

[0064] 8 is a conceptual diagram for explaining the log data combination generation process performed by the log data combination generation unit 120 in the first example. The log data combination generation unit 120 generates a data set by combining four log data stored in the driving log data DB 110. If two different log data are combined, six data sets from data set 1 to data set 6 are generated.

[0065] FIG. 9 is a conceptual diagram for explaining the map data generation process performed by the map data generation unit 130 in the first example. The map data generation unit 130 generates map data from each of the six data sets generated by the log data combination generation unit 120. Map data 1 is generated from log data 1 and log data 2. Map data 2 is generated from log data 1 and log data 3. Map data 3 is generated from log data 1 and log data 4. Map data 4 is generated from log data 2 and log data 3. Map data 5 is generated from log data 2 and log data 4. Then, map data 6 is generated from log data 3 and log data 4. These six map data are map data for evaluation.

[0066] 10 is a conceptual diagram for explaining the map data evaluation process performed by the map data evaluation unit 140 in the first example. The map data evaluation unit 140 calculates the degree of variation of element data constituting the map data for each piece of map data. For example, if the map data is feature map data, the variation of the feature values ​​of the entire map is calculated. The degree of variation is expressed, for example, by standard deviation or probability distribution.

[0067] 10, the degree of variation in element data in each of map data from map data 1 to map data 6 is expressed as a probability distribution. Among these, the probability distributions of map data 1, map data 2, and map data 4 indicate normal distributions, and the degrees of convergence to the average value are roughly the same. In contrast, the degrees of convergence of map data 3, map data 5, and map data 6 are different from the degree of convergence of map data 1 etc., and are also different from each other. The map data evaluation unit 140 evaluates map data 1, map data 2, and map data 4, which form the main group, as normal map data, and evaluates map data 3, map data 5, and map data 6, which are outside the group, as abnormal map data.

[0068] FIG. 11 is a conceptual diagram for explaining the abnormal log data judgment process performed by the abnormal log data judgment unit 150 in the first example. The abnormal log data judgment unit 150 acquires the evaluation result of each map data by the map data evaluation unit 140 and information on the log data used to create each map data. The map data evaluated as abnormal by the map data evaluation unit 140 are map data 3, map data 5, and map data 6. The log data common to these map data is log data 4. And, the log data 4 was not used in the creation of the map data 1, map data 2, and map data 4 evaluated as normal. Based on such analysis, the abnormal log data judgment unit 150 judges that the log data 4 is log data that causes an abnormality in the map data, that is, abnormal log data.

[0069] The abnormal log data determination unit 150 passes the determination result to the log data combination generation unit 120. The log data combination generation unit 120 removes the abnormal log data 4 from the previously generated data set, and combines the log data 1, log data 2, and log data 3 to generate a final data set.

[0070] 12 is a conceptual diagram for explaining the final map data generation process performed by the map data generation unit 130 in the first example. The map data generation unit 130 generates final map data using the final data set generated by the log data combination generation unit 120. The final map data is highly accurate map data in which no abnormal log data 4 is used. The map data generation unit 130 stores the final map data in the map data DB 160.

[0071] Here, a description will be given of the format of the log data stored in the driving log data DB 110. Figures 13 and 14 are conceptual diagrams for explaining the format of the log data stored in the driving log data DB 110.

[0072] 13, parameters that define the log data are the vehicle name, date and time, time, and the presence or absence of each sensor. Six log data, from log data 1 to log data 6, are stored in the traveling log data DB 110. Log data 1, log data 2, log data 3, and log data 4 are log data acquired using sensors A, B, and C. Log data 5 and log data 6 are log data acquired using only sensor D.

[0073] In the example shown in FIG. 13, the data from sensors A, B, and C are grouped together to form one log data. However, the data acquired by sensors A, B, and C is independent. There is no limitation on the type of sensor that acquires the log data. For example, sensor A used to acquire log data 1 to 4 may be a camera, and sensor D used to acquire log data 5 and 6 may be a LiDAR. Even log data acquired by different sensors can be mapped on one map.

[0074] In the example shown in FIG. 14, parameters that define the log data are the vehicle name, date and time, time, and the presence or absence of each sensor. Fourteen pieces of log data are stored in the traveling log data DB 110. Log data 1a and log data 1b are partial data of log data 1. Log data 1 includes data of sensors A, B, and C, whereas log data 1a includes data of sensors A and C, and log data 1b includes only data of sensor B. Similarly, log data 2a and log data 2b are partial data of log data 2. Log data 3a and log data 3b are partial data of log data 3. And log data 4a and log data 4b are partial data of log data 4.

[0075] Just as log data 1a and 1b are generated from log data 1, log data can be decomposed for each parameter, and the partial data obtained by the decomposition can be made into independent log data. Log data can be decomposed for each parameter in any way. For example, in the example shown in FIG. 14, daytime log data and nighttime log data can be generated from each log data. Also, if there is a change in the weather while acquiring log data, the log data can be decomposed for each weather to generate independent log data for each weather.

[0076] Next, the format of the combination of log data in the log data combination generation process will be described. Fig. 15 is a conceptual diagram for explaining the format of the combination of log data. However, the log data that can be combined, that is, the log data stored in the driving log data DB 110, is assumed to be four, namely, log data 1 to log data 4.

[0077] 15, dataset 2 is a combination of two pieces of log data, and dataset N is a combination of four pieces of log data. Thus, the number of pieces of log data included in one dataset does not have to be the same between datasets. Also, there may be a dataset that includes only one piece of log data, such as dataset 1, and there may be a dataset that includes all the log data, such as dataset N.

[0078] In the log data combination generation process, it is not necessary to cover all possible combinations of log data. On the other hand, there may be a combination of overlapping log data, such as data set 2 and data set N-2. Also, one data set may contain multiple pieces of the same log data, such as data set 4 containing two pieces of log data 1.

[0079] 3-2. Second Example 16 to 22 are conceptual diagrams for explaining a second example of a map data generating method. The second example described below corresponds to the first problem and its solving means, similar to the first example.

[0080] FIG. 16 is a conceptual diagram showing an example of the configuration of the driving log data DB 110 in the second example. In the second example, the log data is specified by five parameters, namely, the vehicle name, the date and time, the presence or absence of data of sensor A, and the presence or absence of data of sensor B. Log data 1a includes only data of sensor A, and log data 1b includes only data of sensor B. These are independent log data generated by dividing log data 1 including data of both sensors A and B. Similarly, log data including only data of sensor A and log data including only data of sensor B are generated from log data 2, log data 3, and log data 4. Both sensors A and B are external sensors. Sensors A and B may be different types of sensors, such as a camera and LiDAR, or may be sensors of the same type but different models, or may be sensors of the same model but different mounting positions on the vehicle.

[0081] FIG. 17 is a conceptual diagram for explaining the log data combination generation process performed by the log data combination generation unit 120 in the second example. The log data combination generation unit 120 generates N data sets by combining 12 log data stored in the driving log data DB 110. The log data 1a and log data 1b generated from the log data 1 are also combined with other log data as independent log data. Although not shown in FIG. 17, log data in a parent-child relationship such as the log data 1 and the log data 1a may be combined. The number (N) of data sets generated by the log data combination generation unit 120 depends on the method of combining the log data. Here, two log data are combined to generate a data set, but any number of log data can be combined as described with reference to FIG. 15.

[0082] 18 is a conceptual diagram for explaining the map data generation process performed by the map data generation unit 130 in the second example. The map data generation unit 130 generates map data from each of the N data sets generated by the log data combination generation unit 120. The total of N map data generated from each data set is map data for evaluation.

[0083] FIG. 19 is a conceptual diagram for explaining the map data evaluation process performed by the map data evaluation unit 140 in the second example. The map data evaluation unit 140 calculates the degree of variation of element data constituting the map data for each piece of map data. In the example shown in FIG. 19, the degree of variation of element data in each piece of map data from map data 1 to map data N is expressed by a probability distribution. Here, it is assumed that at least map data 9 and map data N belong to a main group in which the convergence degrees are roughly the same, and each piece of map data from map data 1 to map data 8 is outside that group. It is also assumed that the convergence degrees of each piece of map data from map data 1 to map data 8 vary. In this case, the map data evaluation unit 140 evaluates at least each piece of map data from map data 1 to map data 8 as abnormal map data.

[0084] 20 is a conceptual diagram for explaining the abnormal log data judgment process performed by the abnormal log data judgment unit 150 in the second example. The abnormal log data judgment unit 150 acquires the evaluation result of each map data by the map data evaluation unit 140 and information on the log data used to create each map data. The map data evaluated as abnormal by the map data evaluation unit 140 are at least map data 1 to map data 8. By comparing the log data used to create these abnormal map data with the log data used to create map data 9 evaluated as normal, it is found that log data 1, log data 1a, log data 2, and log data 2a are abnormal log data.

[0085] FIG. 21 is a conceptual diagram for explaining the abnormal parameter condition judgment process performed by the abnormal log data judgment unit 150 in the second example. The abnormal log data judgment unit 150 compares the parameter conditions that define the log data 1, log data 1a, log data 2, and log data 2a that are judged to be abnormal log data with the parameter conditions that define the other log data. Then, the abnormal log data judgment unit 150 identifies the parameter conditions that are common only to the abnormal log data among the five parameter conditions that define the log data 2a. The identified parameter conditions are the conditions that cause an abnormality in the map data. In this example, the presence of data from sensor A is the condition that causes an abnormality in the map data. Therefore, the abnormal log data judgment unit 150 judges the log data 3, log data 3a, log data 4, and log data 4a that have data from sensor A as log data that is highly likely to be abnormal.

[0086] The abnormal log data determination unit 150 passes the determination result to the log data combination generation unit 120. The log data combination generation unit 120 removes the log data including the data of sensor A from the previously generated data set, and generates a final data set by combining only log data 1b, log data 2b, log data 3b, and log data 4b that do not include the data of sensor A. In this case, the administrator may be notified that some abnormality has occurred in sensor A. The some abnormality may be a sensor failure, a deviation in the sensor calibration, a lack of accuracy of the sensor in the first place, etc.

[0087] 22 is a conceptual diagram for explaining the final map data generation process performed by the map data generation unit 130 in the second example. The map data generation unit 130 generates final map data using the final data set generated by the log data combination generation unit 120. The final map data is highly accurate map data in which no log data including data from sensor A is used. The map data generation unit 130 stores the final map data in the map data DB 160.

[0088] The log data can be divided according to the conditions of the various parameters listed above, not limited to the presence or absence of sensor data. For example, one log data may be divided into log data acquired during the day and log data acquired at night. One log data may also be divided into log data acquired when the vehicle speed is equal to or greater than a predetermined speed and log data acquired when the vehicle speed is less than the predetermined speed. One log data may also be divided into log data acquired while traveling in a predetermined section and log data acquired outside the predetermined section. One log data may also be divided into log data acquired during automatic driving and log data acquired during manual driving.

[0089] 3-3.Third Example 23 to 29 are conceptual diagrams for explaining a third example of a map data generating method. The third example explained below corresponds to the second problem and the means for solving the problem.

[0090] 23 is a conceptual diagram showing an example of the configuration of the driving log data DB 110 in the third example. In the third example, the log data is defined by four parameters, namely, the vehicle name, the date and time, the time, and the weather. Here, the parameter "weather" has two conditions: sunny and rainy.

[0091] Fig. 24 is a conceptual diagram for explaining the log data combination generation process performed by the log data combination generation unit 120 in the third example. The log data combination generation unit 120 generates N data sets by combining seven pieces of log data stored in the driving log data DB 110. The number (N) of data sets generated by the log data combination generation unit 120 depends on how the log data are combined. Here, two pieces of log data are combined to generate a data set, but any number of pieces of log data can be combined as described with reference to Fig. 15.

[0092] FIG. 25 is a conceptual diagram for explaining the map data generation process performed by the map data generation unit 130 in the third example. The map data generation unit 130 generates map data from each of the N data sets generated by the log data combination generation unit 120. For example, map data 1 is generated from log data 1 and log data 2, and map data 2 is generated from log data 1 and log data 3. Map data i-1 is generated from log data 3 and log data 4, and map data i is generated from log data 3 and log data 5. Map data N-1 is generated from log data 5 and log data 7, and map data N is generated from log data 6 and log data 7. A total of N map data generated from each data set is map data for evaluation.

[0093] FIG. 26 is a conceptual diagram for explaining the map data evaluation process performed by the map data evaluation unit 140 in the third example. The map data evaluation unit 140 calculates the degree of variation of element data constituting the map data for each map data. In the example shown in FIG. 26, there are two groups of map data whose convergence degrees are roughly the same. Map data 1 and map data 2 belong to the first group. Map data N-1 and map data N belong to the second group. Both the map data belonging to the first group and the map data belonging to the second group show normal distribution, but there is a clear difference in the degree of convergence to the average value. On the other hand, at least map data i-1 and map data i do not belong to either the first group or the second group, and the degrees of convergence do not match each other. The map data evaluation unit 140 evaluates map data that does not belong to either the first group or the second group, including map data i-1 and map data i, as abnormal map data.

[0094] FIG. 27 is a conceptual diagram for explaining the abnormal log data judgment process performed by the abnormal log data judgment unit 150 in the third example. The abnormal log data judgment unit 150 acquires the evaluation result of each map data by the map data evaluation unit 140 and information on the log data used to create each map data. Then, the abnormal log data judgment unit 150 compares the log data used to create the map data evaluated as abnormal by the map data evaluation unit 140 with the log data used to create the map data evaluated as normal. In the example shown in FIG. 27, it is found from the comparison result that there is no specific log data causing an abnormality in the map data. In this case, it is assumed that the cause of the abnormality in the map data is a combination of log data, more specifically, a combination of parameter conditions that define the log data.

[0095] FIG. 28 is a conceptual diagram for explaining the abnormal combination judgment process performed by the abnormal log data judgment unit 150 in the third example. The abnormal log data judgment unit 150 compares the parameter conditions between the log data used to create each map data evaluated as abnormal. For the map data i-1, the parameter conditions are compared between the log data 3 and the log data 4. For the map data i, the parameter conditions are compared between the log data 3 and the log data 5. Based on the comparison results, the abnormal log data judgment unit 150 judges that the combination of the log data in which the parameter "weather" condition is sunny and the log data in which the parameter "weather" condition is rainy is the cause of the abnormality in the map data.

[0096] That is, even if the log data acquired on a sunny day and the log data acquired on a rainy day are normal log data, they may not be combined. This will be described by taking a white line on a feature map as an example. When the external sensor that acquires the feature is a camera, the camera can clearly capture the white line on a sunny day, whereas it is difficult to capture the white line on a rainy day. For this reason, the variance of the feature acquired on a sunny day is small and the convergence is high, whereas the variance of the feature acquired on a rainy day tends to be large. Therefore, although the feature acquired on a sunny day and the feature acquired on a rainy day can be matched, the feature acquired on a sunny day and the feature acquired on a rainy day cannot be matched, and an anomaly occurs in the map data due to the inconsistency of the feature.

[0097] The abnormal log data determination unit 150 passes the determination result to the log data combination generation unit 120. In response to the determination result, the log data combination generation unit 120 generates a data set combining only log data acquired on sunny days and a data set combining only log data acquired on rainy days.

[0098] FIG. 29 is a conceptual diagram for explaining the final map data generation process performed by the map data generation unit 130 in the third example. The map data generation unit 130 generates final map data using the final data set generated by the log data combination generation unit 120. The final map data generated by the map data generation unit 130 is two types of map data: map data for fine weather and map data for rainy weather. The map data for fine weather is generated using only log data 1, log data 2, and log data 3 acquired on a fine day. On the other hand, the map data for rainy weather is generated using only log data 4, log data 5, log data 6, and log data 7 acquired on a rainy day. The map data generation unit 130 stores the map data for fine weather and the map data for rainy weather separately in the map data DB 160.

[0099] The map data for sunny weather stored in the map data DB 160 is used for autonomous driving on sunny days. On the other hand, the map data for rainy weather is used for autonomous driving on rainy days. When the external sensor is a camera, it is difficult to recognize white lines on rainy days, so in the map data for rainy weather, the weight of the feature amount of curbs, which are relatively easy to recognize even on rainy days, is increased, and the weight of the feature amount of white lines is decreased. As a result, the map data for sunny weather can achieve high accuracy in autonomous driving on sunny days, while the map data for rainy weather can achieve high accuracy in autonomous driving on rainy days.

[0100] The parameter "weather" given here is an example of a parameter whose combination of conditions may cause anomalies in the map. In addition to weather, parameters such as vehicle type, time period (daytime or nighttime), sensor type (LiDAR or camera), and map generation algorithm may cause anomalies in the map data depending on the combination of conditions. In addition, even if the sensor type is the same, if there is a difference in performance, map data may be generated for each sensor. For example, a LiDAR that is easy to detect curbs and a LiDAR that is difficult to detect curbs have different weights for curbs as features in the map data. For this reason, it is preferable to treat the log data acquired by these two LiDARs as separate categories and generate map data separately using the respective log data.

[0101] 4. Specific examples of map data evaluation processing 4-1.Map data evaluation process for feature maps 4-1-1. Overview of feature map In the above explanation, the degree of variation in element data constituting map data (degree of convergence of map data) is visually expressed by a probability distribution. However, this is only one example of an evaluation method for identifying map data with an abnormality, and it is not necessary to evaluate only by looking at the degree of convergence of certain statistical values. Below, an overview of maps actually used in autonomous driving systems is given, and specific examples of map data evaluation processing corresponding to each map are explained. Figures 30 to 50 are diagrams for explaining specific examples of map data evaluation processing by the map data generation device 100 according to this embodiment.

[0102] First, a feature map will be described. A feature map is map data that an autonomous vehicle uses to estimate its own position. A feature map is generated by superimposing unique features acquired from sensor data of LiDAR or a camera, and performing optimization calculations, specifically SLAM (Simultaneous Localization and Mapping). Since distance accuracy is required in autonomous driving, LiDAR is an example of a preferred external sensor. In the example described below, features are acquired by LiDAR.

[0103] Figure 30 is a diagram showing an example of feature detection for a feature map. The image shown on the left side of Figure 30 is a point cloud image obtained by mapping a three-dimensional point cloud obtained from a LiDAR onto a plane. The image located in the center of the image is a vehicle equipped with a LiDAR. This point cloud image is processed by a feature detector to detect features, resulting in the feature image shown on the right side of Figure 30. In the feature image, white lines, curbs, and building corners are detected as features around the vehicle located in the center of the image.

[0104] 31A, 31B, and 31C are diagrams showing examples of types of feature amounts. In the feature amount image shown in FIG. 31A, a white line and a curb (step) are detected as feature amounts. The white line is detected as a straight line using reflection intensity information of the LiDAR point cloud. The curb is detected as a straight line using distance information of the LiDAR point cloud.

[0105] In the feature image shown in Fig. 31B, the road sign and the pole supporting it are detected as features. For the road sign, flat surfaces of a certain size or more and less than a certain size are detected as flat surfaces using the distance information of the LiDAR point cloud. For the pole, cylinders of a certain thickness or more and less than a certain length or more and less than a certain length are detected as straight lines extending in the Z direction using the distance information of the LiDAR point cloud.

[0106] In the feature image shown in Fig. 31C, the convex and concave angles of the building are detected as features. Regarding the corners of the building, corners that intersect with a plane at 90 degrees and have a certain length or more and less than a certain length are detected as straight lines using distance information of the LiDAR point cloud.

[0107] It should be noted that not all features are detected after identifying the object, but only the feature-likeness defined by the algorithm is detected from the LiDAR point cloud. Therefore, for example, if a plane of a certain size or less is detected from the LiDAR point cloud, it may be detected as a road sign even though it is actually a building wall. However, in self-localization using a feature map, it is sufficient to have repeatability in the detection of features. Therefore, even if what is detected as a road sign is actually a building wall, it can be used for self-localization without any problems. This also applies to other features.

[0108] FIG. 32 is a diagram showing an example of generating a feature map by SLAM. The image on the left side of FIG. 32 is an image obtained by superimposing the feature values ​​obtained from each log data with the output of the locator (GPS and IMU). The output of the locator is coordinates (x, y, z) and attitude (roll, pitch, yaw). In this image, the feature values ​​seem to match, but in reality, the feature values ​​vary across the entire image. In other words, if the feature values ​​of all the log data are simply superimposed with the output of the locator, the feature values ​​will remain varied. Therefore, SLAM, that is, optimization calculation is performed on the superimposed feature values ​​of all the log data. The image on the right side of FIG. 32 is a feature map obtained by performing optimization calculation on the image on the left side. The created feature map is stored in the map data DB 160 and is used for self-location estimation when the autonomous vehicle actually runs.

[0109] 4-1-2. Evaluation of map data based on statistics of the entire map As described above, optimization calculations are performed when generating a feature map. Although optimization calculations increase the degree of match of features, the features do not necessarily match perfectly between all log data.

[0110] 33A, 33B, and 33C are diagrams showing examples of variation for each feature in a feature map. FIG. 33A shows examples of feature amounts for a road sign, a pole, a building corner (vertical), and a building corner (horizontal) registered in the feature map. FIG. 33B shows an example of a feature amount for a white line registered in the feature map. All of these features are obtained from the same object. However, variation occurs in the position of the feature after optimization calculation due to a number of reasons, including the detection accuracy of the LiDAR itself, the accuracy of the LiDAR calibration (mounting position and orientation), the accuracy of the feature detection algorithm, and the accuracy of the optimization calculation.

[0111] In Fig. 33C, the variation of each feature in a feature map is shown as a standard deviation. However, the variation of the feature is not a statistic focusing on a specific object, but a statistic for the entire map data. In the example shown in Fig. 33C, each feature has a standard deviation of about 0.05 m.

[0112] As described above, for each feature map, the statistics when the feature is viewed as an entire map, that is, the variation of the feature, can be obtained. Therefore, whether the map data can be said to match can be determined by comparing the statistics between the map data. In other words, abnormal map data can be identified by comparing the statistics. FIG. 34 is a diagram showing an example of the statistics for each map data. Here, for simplicity of explanation, three map data are used, and the only feature is the white line.

[0113] The F-test is known as a method for determining whether two sets have equal variances. In the F-test, the probability that the variances of two sets are equal can be found. For example, a probability of 5% or more can be determined as "equal variances", and a probability of less than 5% can be determined as "not equal variances". Each of the three map data shown in FIG. 34 is tested according to this. FIG. 35 is a diagram showing an example of evaluation using the F-test. When data with equal variances are grouped together, map data 1 and map data 2 have equal variances, but only map data 3 does not have equal variance with the other map data, and its standard deviation is also larger than the other map data. Therefore, the test result of the F-test can determine that map data 3 is abnormal map data.

[0114] Another method for determining abnormal map data based on the statistics of the entire map is based on the difference from the map data with the smallest standard deviation. Simply put, the map data with the smallest standard deviation of the feature is considered to be the most accurate map data. A threshold may be set based on the standard deviation of that map data, and map data with a standard deviation equal to or greater than that threshold may be determined to be abnormal map data.

[0115] 34, the map data with the smallest standard deviation is map data 1. If map data having a standard deviation that is 20% or more greater than the standard deviation of map data 1 is determined to be abnormal, map data 3 is determined to be abnormal map data. If map data having a standard deviation that is 0.001 m or more greater than the standard deviation of map data 1 is determined to be abnormal, map data 2 and map data 3 are determined to be abnormal map data.

[0116] 4-1-3.Evaluation of map data based on the success rate of self-location estimation Poses that indicate the vehicle's own position and orientation are used when generating a feature map. In Fig. 32, the centers of the circles that are continuously placed on the road represent the vehicle's pose at each time, and the lines connecting them represent the route that the vehicle has traveled. In SLAM, the features are optimized, and at the same time, each pose is also optimized using the features detected at that pose. The degree of success of self-location estimation at a pose using SLAM can be found from the statistical values ​​explained below.

[0117] FIG. 36 is a conceptual diagram for explaining an example of self-location estimation in a pose by SLAM. i Vehicle pose in X i The feature values ​​F1, F2, and F3 observed at that pose are shown. The feature value F1 is a white line, the feature value F2 is a road sign, and the feature value F3 is a pole. The measurement distances d1, d2, and d3 of the features F1, F2, and F3 measured by the LiDAR include predetermined measurement errors σ1, σ2, and σ3. In SLAM, the pose X is calculated based on the position, measurement distance, and measurement error of the features F1, F2, and F3. i A self-localization is performed for the

[0118] 37 and 38 are conceptual diagrams for explaining the results of self-location estimation in a pose using SLAM. Band-shaped regions B1, B2, and B3 in Figs. 37 and 38 are defined by the positions, measurement distances, and measurement errors of the feature amounts F1, F2, and F3. More specifically, band-shaped regions B1, B2, and B3 are separated from the positions of the feature amounts F1, F2, and F3 by measurement distances d1 to d3, and have widths of 2σ1 to 2σ3. Region E where the three regions B1, B2, and B3 overlap is i is the estimated area of ​​the self-location. Estimated area of ​​the self-location E i and Pose X i The region R around i The larger the overlapping area, the greater the pose X i It can be determined that the degree of success of self-location estimation in the self-location estimation area E i and Pose X i The region R around i Since and overlap, this is pose X.i On the other hand, in Fig. 38, the self-location estimation area E i and Pose X i The region R around i Since it does not overlap with pose X, i This shows an example of self-location estimation failure.

[0119] A certain pose X i In general, the consistency of the observed features can be determined by calculating the chi-squared value. The chi-squared value is calculated, for example, using the expected value of the measured distance of the feature and the observed value. i The statistical probability of the chi-squared value occurring in can be expressed as a p-value. The larger the chi-squared value, the smaller the p-value, and the smaller the chi-squared value, the larger the p-value. i The degree of success of self-localization in is expressed by the p value as a probability ranging from 0 (high probability of failure) to 1 (high probability of success).

[0120] By using the p-value, it is possible to visualize the degree of success of self-localization by SLAM for each pose on a feature map. Fig. 39 is a diagram showing an example of visualization of the degree of success of self-localization by SLAM for each pose on a feature map. On the feature map shown in Fig. 39, the reciprocal of the p-value is visualized as a line extending from the pose in the z direction. The lower the p-value at a pose, the longer the line extending from that pose in the z direction.

[0121] Ideally, the p-value should be high across the entire map. However, the p-value is fundamentally location-dependent, and in sections where features cannot be sufficiently acquired or where there is a bias in the features that can be acquired, such as the area circled on the right, many poses with low p-values ​​appear. The occurrence of many poses with low p-values ​​means that it is highly likely that SLAM has failed. On the other hand, in sections where features can be sufficiently detected, such as the area circled on the left, the p-value is high for each pose. The few poses with low p-values ​​mean that it is highly likely that SLAM was successful.

[0122] If the p-value depends on the location, then when multiple map data are generated, there should be a strong correlation in the fluctuation of the p-value between the map data within the common section between those map data. A common method to measure the strength of the correlation is to compare the average p-values ​​within the common section between map data.

[0123] For example, suppose that three map data, namely map data 1, map data 2, and map data 3, are created using log data acquired in a common section from point A to point B shown in FIG. 40A. By plotting the p-values ​​of each map data on the same graph with the horizontal axis representing the distance from point A, a graph showing the correlation of the fluctuation in p-values ​​between map data, as shown in FIG. 40B, can be obtained. Assuming that all map data are expressed in N poses from point A to point B, and the p-value of the n-th pose of the i-th map data is expressed as p i,n In this case, the average value of p in the common section of map data i is expressed by the following formula.

number

[0124] 40B, the average p-values ​​of map data 1 and map data 2 roughly match, whereas the average p-value of map data 3 is far smaller than them. Therefore, map data 3 in the common section from point A to point B is determined to be abnormal map data in comparison with map data 1 and map data 2.

[0125] Fig. 41 is a diagram showing another example of the correlation of the fluctuation of p-values ​​between map data. In the example shown in Fig. 41, the p-value of map data 3 is significantly lower than the p-values ​​of map data 1 and map data 2 only in the section within window W. However, if the average p-value is calculated for the entire section from point A to point B, the influence of the section within window W is weakened. In other words, there is a possibility that there will be no difference in the average p-values ​​between map data 1 and map data 2 and map data 3.

[0126] Therefore, considering that differences in p-values ​​may occur between map data in some sections as in the example shown in Fig. 41, the average value of p-values ​​may be calculated for each certain section. For example, the average value of p-values ​​may be calculated for the sections within window W while moving window W from point A to point B. Fig. 42 is a diagram showing an example of the average value of p-values ​​for each map data in the sections within window W shown in Fig. 41.

[0127] In the example shown in FIG. 42, the map data with the highest average p-value is map data 1. The average p-value of map data 1 is set as the reference, and a value 0.1 smaller than this is set as the threshold. In this case, map data 2 with an average p-value of 0.45 is determined to match map data 1. However, map data 3 with an average p-value of 0.2 does not match map data 1, that is, it is determined to be abnormal. Note that when map data 3 is determined to be abnormal in the section of window W, it is also possible to cut out only that section from map data 3 and treat the remaining section as normal map data.

[0128] In addition to the method of using the average value of the p-values ​​to evaluate the match / mismatch of map data, there is also a method of using the correlation coefficient of the p-values ​​of each map data. FIG. 43 is a diagram showing an example of the correlation coefficient of the p-values ​​between map data. The threshold value of the correlation coefficient at which two map data are considered to match is assumed to be, for example, 0.8. In that case, if the correlation coefficient of the p-values ​​between map data 1, map data 2, and map data 3 is expressed as shown in FIG. 43, it is determined that map data 1 and map data 2 match. However, since map data 3 does not match either map data 1 or map data 2, map data 3 is determined to be abnormal map data.

[0129] In the above example, it is assumed that at least one of the multiple map data is valid. However, since the p-value is a numerical value that expresses the success rate of SLAM at a pose as a probability, map data with a low p-value is highly likely to be unsuitable for autonomous driving without even needing to be compared with other map data. Therefore, an absolute reference value may be set for the p-value, and map data with a p-value lower than the absolute reference value may be determined to be abnormal. For example, if the average p-value of the entire map data is determined to be abnormal if it is 0.4 or less, only map data 3 is determined to be abnormal in the example shown in FIG. 41. However, if the lowest p-value in all sections is determined to be abnormal if it is 0.4 or less, all map data is determined to be abnormal in the example shown in FIG. 41.

[0130] In addition to the p-value at each pose, the following values ​​can also be used as a method for comparing map data. Number of features of detected white lines Number of features of detected curbs - Number of features of detected road signs - Number of features of detected building corners Total number of detected features Pose altitude (height in z-direction)

[0131] Generally, the p-value will be low when there are few detectable features. However, for example, if a vehicle is equipped with multiple sensors and only one sensor is faulty, the p-value may not drop much because the other sensors can detect the features. In such cases, it is possible to detect sensor failure by comparing the number of detectable features between map data.

[0132] 4-1-4.Evaluation of map data based on trajectories after SLAM Next, a method of evaluating map data using a trajectory after SLAM of map data, that is, a chronological arrangement of poses, is described. Fig. 44 is a conceptual diagram for explaining an example of map data evaluation using a trajectory after SLAM.

[0133] FIG. 44A shows four map data, map data 1 to map data 4. All map data were generated along the same route in the section from point A to point B. The route from point A to point B in each map data corresponds to the trajectory after SLAM. In order to compare the route shapes of these four map data, point A in each map data is moved to the origin in the xyz space, and the direction of the pose at point A in each map data (vehicle traveling direction) is rotated so that it matches the x-axis.

[0134] Figure 44B is obtained by the above processing. Figure 44B shows the mapping of each map data onto the xy plane. In the xy plane shown in Figure 44B, the arrow lines extending from the origin show the trajectory of each map data after SLAM as seen from point A. The tip of each arrow line is point B. However, because the SLAM results are different, the relative position of point B as seen from point A differs for each map data.

[0135] Here, point B of each map data is clustered using coordinates (x, y, z) and attitude (roll, pitch, yaw). Clustering here means grouping a plurality of objects close to each other, and a general method such as the shortest distance method can be applied. For example, point B of each map data can be clustered using the shortest distance method with a distance threshold of 3 m. In the example shown in FIG. 44B, map data 1 and map data 2 are clustered into one group. Since a group with the largest number of map data items belonging thereto can be said to have the highest reliability, the group to which map data 1 and map data 2 belong is determined to have the highest reliability. Map data 3 and map data 4 that do not belong to that group are determined to be abnormal map data.

[0136] 4-2. Map data evaluation process for road surface shape map and road surface brightness map 4-2-1. Overview of road surface shape map and road surface brightness map Fig. 45 is a diagram showing an example of image data of a road surface shape map. In the road surface shape map, image data as shown in Fig. 45 is decomposed in the x and y directions at a preset resolution, and information is stored in each cell obtained by decomposing the image data. Similarly, in the road surface luminance map, information is stored for each cell obtained by decomposing the image data. The cell resolution is, for example, 0.1 m in both the x and y directions.

[0137] Each cell of the road surface shape map stores the number of LiDAR point clouds that have reached the cell, as well as the average height and variance. The average height and variance are calculated using the point cloud values ​​(height values ​​in the z-direction) of all LiDARs that have reached the cell. Each cell of the road surface brightness map stores the number of LiDAR point clouds that have reached the cell, as well as the average road surface brightness and variance. The average road surface brightness and variance are calculated using the point cloud values ​​(reflection intensity) of all LiDARs that have reached the cell.

[0138] The following mainly describes the road surface shape map, but the following description can also be applied to the road surface brightness map by replacing the height information stored in each cell with brightness information as necessary.

[0139] Due to the nature of road surface shape maps, which express road surface shapes in cells with a certain area, in areas with overgrown grass or steps, the heights of the LiDAR points that hit these areas will not be the same, so even if SLAM is successful, the variance in heights will be large. On the other hand, on roads where vehicles travel and sidewalks where pedestrians walk, the range of the cells is often flat, so if SLAM is successful, the variance in heights is expected to be small.

[0140] FIG. 46A shows a specific example where SLAM was successful on a road surface shape map. On the other hand, FIG. 46B shows a specific example where SLAM failed on a road surface shape map. The black and white shading on the road surface in each figure represents the standard deviation of height for each cell in a heat map. Areas that appear white are areas with a small standard deviation of height, and areas that appear black are areas with a large standard deviation of height. If SLAM is successful, the variance of the road surface shape will be low, so the road surface will be displayed white as shown in FIG. 46A. The area in the center that appears particularly white is an area with an accuracy of about 0.01 m standard deviation. On the other hand, if SLAM fails, the variance of the road surface shape will be large, so the road surface will be displayed black as shown in FIG. 46B. The area in the center that appears particularly black is an area with an accuracy of about 0.15 m standard deviation.

[0141] 4-2-2.Evaluation of map data based on road surface shape The method of evaluating map data based on road surface shape is explained below, but the following explanation can also be applied to the evaluation of map data based on road surface luminance by replacing the height information stored in each cell with luminance information as necessary.

[0142] In the case of a road surface shape map, if there is grass growing within a cell or if there is a step such as a curb, the variance in the height direction will be large even if SLAM is successful. However, in the area where the vehicle has passed, as long as it is a paved road, the inside of the cell is flat, so the variance in the height direction should be small. In other words, if the variance in the height direction of the area where the vehicle has passed is large, there is a high possibility that the map data is not normal. In other words, by focusing on the variance in the height direction of the road surface shape of the area where the vehicle has passed, it is possible to determine whether there is an abnormality in the map data of the road surface shape map.

[0143] For example, assume that three map data, that is, map data 1, map data 2, and map data 3, are created using log data acquired in a common section from point A to point B. The variance σ of the road surface shape in the height direction of each map data is hBy plotting these on the same graph with the horizontal axis being the distance from point A, the vertical variance σ between the map data as shown in Figure 47 can be obtained. h A graph showing the correlation of the variation of the road surface shape in the height direction σ h may be the variance in the height direction of the cells below the pose, i.e., directly below the vehicle, or may be the average of the variance in the height direction of multiple cells around the vehicle.

[0144] The abnormality of the map data of the road surface shape map is the vertical dispersion σ h The method can be applied to the road surface shape map. The methods for determining abnormalities in the map data of the feature map can be applied to the road surface shape map. There are three methods for determining abnormalities in the map data of the feature map: a method for determining abnormalities by comparing the average values ​​of p-values, a method for determining abnormalities by comparing the correlation coefficients of p-values, and a method for determining abnormalities based on the minimum p-value. In these methods, the p-value is calculated by multiplying the variance σ h By replacing the p-value with the variance of brightness, it can be used as a method for determining anomalies in map data of a road surface shape map. Similarly, by replacing the p-value with the variance of brightness, it can be used as a method for determining anomalies in map data of a road surface brightness map.

[0145] In addition, by plotting the number of LiDAR point clouds that reach each map data cell on the same graph with the distance from point A on the horizontal axis, the number of LiDAR point clouds N between map data as shown in Figure 48 can be calculated. points A graph showing the correlation of the fluctuations in the LiDAR point cloud number N points may be the number of LiDAR point clouds that reach the cell under the pose, i.e., directly under the vehicle, or may be the average number of LiDAR point clouds that reach multiple cells around the vehicle. points By replacing it with, it becomes possible to use it as a method for determining an abnormality in the map data of the road surface shape map. It can also be used as a method for determining an abnormality in the map data of the road surface luminance map.

[0146] As a method for evaluating the match / mismatch of map data based on road surface shape, in addition to using the variance of the road surface shape in the height direction, there is also a method that uses the road surface gradient. A road surface shape map is a map that represents the shape of the road surface, that is, the height. The gradient of the road surface can be calculated from the height. Since the road gradient is fixed in an area where the vehicle has passed, it can be expected that the road surface shape map will represent the gradient within the range of "road gradient + allowable error".

[0147] FIG. 49 is a diagram showing an example of an image of a road surface shape in a road surface shape map. More specifically, the example shown in FIG. 49 is an image of the road surface shape obtained from the road surface shape information viewed from the y direction when the resolution of the road surface shape information in the x and y directions is 0.1 m. In the example shown in FIG. 49, the height increases by 0.01 m for every 0.1 m, so the road gradient is rising steadily by about 10%. However, only one cell near the center is higher by 0.05 m or more. Considering the road structure, a gradient of 0.05 m or more (a gradient of 50%) is impossible. Therefore, in the example shown in FIG. 49, it can be determined that there is an abnormality in the generated road surface shape.

[0148] Here, the average height of the cells at coordinates (i,j) on the xy plane is h i,j Then, the road surface gradient Δx in the x direction at coordinate (i,j) i,j and the road surface gradient in the y direction Δy i,j can be calculated using the following formulas, respectively. Δx i,j =h i,j -h i-1,j Δy i,j =h i,j -h i,j-1

[0149] An abnormality in the map data of the road surface shape map can be determined from the correlation of the variation of the road surface gradient Δx and Δy between the map data. For example, assume that three map data, namely, map data 1, map data 2, and map data 3, are created using log data acquired in a common section from point A to point B. By plotting the road surface gradient Δx in the x direction of each map data on the same graph with the distance from point A on the horizontal axis, a graph showing the correlation of the variation of the road surface gradient Δx in the x direction between the map data as shown in FIG. 50 is obtained. The road surface gradient Δx in the x direction may be the road surface gradient in the x direction of the cell below the pose, i.e., directly below the vehicle, or may be the average value of the road surface gradients in the x direction of multiple cells around the vehicle. A graph showing the correlation of the variation of the road surface gradient Δy in the y direction between the map data can also be obtained in a similar manner. In the above-mentioned method for determining an abnormality in the map data of the feature map, by replacing the p value with the road surface gradient Δx in the x direction or the road surface gradient Δy in the y direction, it becomes possible to use it as a method for determining an abnormality in the map data of the road surface shape map.

[0150] Furthermore, in the method using the road gradient, it is not necessary to plot the road gradient along the route from point A to point B. If the map data includes even one cell where the road gradient exceeds the allowable range, the map data may be determined to be abnormal.

[0151] 5.Other The map data evaluation process by the map data generating device 100 can also be applied to static obstacle maps. Abnormal map data can be determined by comparing the positions of voxels on the static obstacle map between map data. For example, in the method for determining an abnormality in map data of a feature map described above, by replacing the p-value with the position of the voxel, it becomes possible to use the method as a method for determining an abnormality in map data of a static obstacle map. [Explanation of symbols]

[0152] 10. Driving log data set 12,14,16 Driving log data 12a Abnormal driving log data 20A, 20B, 20D, 20E, 20F Dataset 20C Final Data Set 30A, 30B, 30D, 30E, 30F Map data for evaluation 30C Final map data 100 Map data generating device 101 Processor 102 Memory 103 Map data generation program 104 Storage 105 Communication Module 106 User Interface 107 Bus 110 Driving log data database 120 Log data combination generation unit 130 Map data generator 140 Map Data Evaluation Unit 150 Abnormal log data determination unit 160 Map Data Database

Claims

1. Generate multiple data sets from the driving log data, Each of the plurality of data sets is composed of one or more pieces of driving log data acquired in a common section, generating map data for evaluation in the common section from each of the plurality of data sets; comparing a plurality of evaluation map data generated from the plurality of data sets with each other to identify abnormal map data included in the plurality of evaluation map data; Identifying abnormal driving log data by comparing a data set used to create the abnormal map data with a data set used to create evaluation map data other than the abnormal map data; Generate final map data using the data set that does not include the abnormal driving log data A map data generating method comprising:

2. 2. The map data generating method according to claim 1, Each of the travel log data included in the travel log data group is defined by a plurality of parameters; The condition of at least one of the plurality of parameters is made non-uniform among the plurality of driving log data included in the plurality of data sets. A map data generating method comprising:

3. 3. The map data generating method according to claim 2, Identifying a condition that causes an abnormality in the map data among the conditions of each of the plurality of parameters based on the abnormal driving log data; The final map data is generated using a data set in which the conditions of the plurality of parameters do not include a condition that causes an abnormality in the map data. A map data generating method comprising:

4. The map data generating method according to claim 2, identifying a combination of conditions that causes an abnormality in the map data from among the combinations of conditions of the plurality of parameters by comparing a data set used to create the abnormal map data with a data set used to create evaluation map data other than the abnormal map data; The final map data is generated using a data set that does not include a combination of conditions that cause an abnormality in the map data among the combinations of conditions of the plurality of parameters. A map data generating method comprising:

5. 5. The map data generating method according to claim 1, further comprising: The map data is map data of a feature map, Comparing the plurality of map data for evaluation with each other includes comparing the variation of the feature amount between the plurality of map data for evaluation. A map data generating method comprising:

6. 5. The map data generating method according to claim 1, further comprising: The map data is map data of a feature map, Comparing the plurality of evaluation map data with each other includes comparing the degree of success of self-location estimation using the feature amount between the plurality of evaluation map data. A map data generating method comprising:

7. At least one processor; a program memory coupled to the at least one processor and storing a plurality of executable instructions; The plurality of executable instructions may be configured to cause the at least one processor to: Generate multiple data sets from the driving log data, Each of the plurality of data sets is composed of one or more pieces of driving log data acquired in a common section, generating map data for evaluation in the common section from each of the plurality of data sets; comparing a plurality of evaluation map data generated from the plurality of data sets with each other to identify abnormal map data included in the plurality of evaluation map data; Identifying abnormal driving log data by comparing a data set used to create the abnormal map data with a data set used to create evaluation map data other than the abnormal map data; and generating final map data using a data set that does not include the abnormal driving log data. A map data generating device comprising:

8. Generate multiple data sets from the driving log data, Each of the plurality of data sets is composed of one or more pieces of driving log data acquired in a common section, generating map data for evaluation in the common section from each of the plurality of data sets; comparing a plurality of evaluation map data generated from the plurality of data sets with each other to identify abnormal map data included in the plurality of evaluation map data; Identifying abnormal driving log data by comparing a data set used to create the abnormal map data with a data set used to create evaluation map data other than the abnormal map data; generating final map data using a dataset that does not include the anomalous driving log data. A map data generating program comprising:

Images