Abnormality detection method, abnormality detection device, and program

Abstract

The abnormality detection method, which is executed by a computer, for detecting an abnormality in an autonomous traveling vehicle that autonomously travels from a departure point to a destination, includes: an acquisition step of acquiring a travel path of the autonomous traveling vehicle from the departure point to an arbitrary point in time; and a detection step (S7300) in which information relating to a plurality of routes on which the autonomous traveling vehicle may travel from the departure point to the destination is read, one of the plurality of routes is estimated as an estimated traveling route from the departure point to the destination of the autonomous traveling vehicle on the basis of the read information relating to the plurality of routes, the degree of abnormality is calculated, and the degree of abnormality is calculated. When the calculated degree of abnormality exceeds a predetermined threshold value, it is detected that an abnormality has occurred in the autonomous traveling vehicle, the degree of abnormality indicating the degree to which the travel route up to an arbitrary point in time deviates from the estimated travel route.

Description

Technical Field

[0001] This disclosure relates to anomaly detection methods for detecting anomalies related to autonomous vehicles. Background Technology

[0002] In recent years, various efforts have been made toward the practical application of autonomous vehicles. For example, practical verification experiments have been conducted on services that utilize autonomous vehicles, such as indoor and outdoor goods delivery, people mobility, cleaning robots, and security robots.

[0003] To provide safe services for autonomous vehicles, communication with remote monitoring locations is necessary for status monitoring and emergency operations. However, this communication exposes them to the risk of cyberattacks. Therefore, to mitigate this risk, proposed methods for detecting anomalies in cyberattacks against autonomous vehicles have been developed.

[0004] For example, Patent Document 1 discloses a method for detecting vehicle anomalies that utilizes the characteristics of a location, such as the vehicle's current location information.

[0005] Existing technical documents

[0006] Patent documents

[0007] Patent Document 1: International Publication No. 2021 / 149340 Summary of the Invention

[0008] The technical problem that the invention aims to solve

[0009] According to the anomaly detection method disclosed in Patent Document 1, based on the vehicle's current location information, the best anomaly detection model that matches the characteristics of the location (such as the difference between general roads and highways) is selected, and the selected anomaly detection model is used to detect vehicle anomalies, provided that the vehicle's current location information is correct.

[0010] However, if the vehicle's current location information is incorrect, such as when it has been tampered with by an attacker in a cyberattack, it may be impossible to select the optimal anomaly detection model and thus fail to detect anomalies. Furthermore, even if the vehicle's current location information is correct and has not been tampered with, it can be difficult to determine whether the vehicle is in an abnormal position if it is illegally operated by an attacker and drives to a location different from its original destination. Therefore, there are inadequate aspects in the anomaly detection of autonomous vehicles.

[0011] Therefore, this disclosure provides anomaly detection methods, etc., that can perform anomaly detection more appropriately.

[0012] Technical solutions for solving the problem

[0013] One technical solution disclosed herein relates to an anomaly detection method, a computer-executed anomaly detection method for detecting anomalies in an autonomous vehicle autonomously traveling from a departure point to a destination, comprising: an acquisition step, acquiring the travel path of the autonomous vehicle from the departure point to any point in time; and a detection step, reading information about multiple possible paths that the autonomous vehicle may travel from the departure point to the destination, based on the read information about the multiple paths, presuming one of the multiple paths as the presumed travel path of the autonomous vehicle from the departure point to the destination, calculating an anomaly degree, and detecting an anomaly in the autonomous vehicle if the calculated anomaly degree exceeds a predetermined threshold, wherein the anomaly degree indicates how much the travel path up to the arbitrary point in time deviates from the presumed travel path.

[0014] Furthermore, one technical solution disclosed herein relates to an anomaly detection device for detecting anomalies in autonomous vehicles traveling from a departure point to a destination. The device comprises: an acquisition unit that acquires the travel path of the autonomous vehicle from the departure point to any point in time; a storage unit that stores information about multiple possible paths the autonomous vehicle may travel from the departure point to the destination; and an anomaly detection unit that, based on the information about the multiple paths read from the storage unit, presumes one of the multiple paths as the presumed travel path of the autonomous vehicle from the departure point to the destination, calculates an anomaly degree, and detects an anomaly in the autonomous vehicle if the calculated anomaly degree exceeds a predetermined threshold. The anomaly degree indicates the extent to which the travel path up to the arbitrary point in time deviates from the presumed travel path.

[0015] In addition, one technical solution disclosed herein relates to a program for executing the aforementioned anomaly detection method by the computer.

[0016] Invention Effects

[0017] Based on the above technical solution, it is possible to more appropriately detect anomalies in autonomous vehicles. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating an example of autonomous vehicle route selection.

[0019] Figure 2 This is a schematic diagram illustrating an example of autonomous vehicle route selection.

[0020] Figure 3 This is a configuration diagram illustrating an example of the overall structure of the anomaly detection system in the implementation method.

[0021] Figure 4 This is a schematic diagram illustrating an example of map information in an implementation method.

[0022] Figure 5 This is a configuration diagram illustrating an example of the configuration of the anomaly detection unit in the implementation method.

[0023] Figure 6 This is a schematic diagram illustrating an example of a data structure for vehicle information in an implementation method.

[0024] Figure 7 This is a schematic diagram illustrating an example of a data structure representing information related to the transported object in an implementation method.

[0025] Figure 8 This is a schematic diagram illustrating an example of a data structure for weather information in an implementation method.

[0026] Figure 9 This is a schematic diagram illustrating an example of a data structure for road traffic information in an implementation method.

[0027] Figure 10 This is a schematic diagram illustrating an example of a data structure that utilizes management information in an implementation method.

[0028] Figure 11 This is a flowchart illustrating an example of anomaly detection processing in an implementation method.

[0029] Figure 12 This is a schematic diagram illustrating an example of a combination of driving path determination parameters in an implementation method.

[0030] Figure 13A This is a schematic diagram illustrating an example of the data structure of the presumed model in the implementation method.

[0031] Figure 13B This is a schematic diagram illustrating an example of the data structure of the presumed model in the implementation method.

[0032] Figure 14A This is a schematic diagram illustrating an example of the data structure of the anomaly detection model in the implementation method.

[0033] Figure 14B This is a schematic diagram illustrating an example of the data structure of the anomaly detection model in the implementation method.

[0034] Figure 14CThis is a schematic diagram illustrating an example of the data structure of the anomaly detection model in the implementation method.

[0035] Figure 15 This is a flowchart illustrating an example of the anomaly detection model selection and processing action in the implementation method.

[0036] Figure 16 This is a flowchart illustrating an example of the exception handling action in the implementation method.

[0037] Figure 17A This is a schematic diagram illustrating an example of the anomaly detection and handling process in the implementation method.

[0038] Figure 17B This is a schematic diagram illustrating an example of the anomaly detection and handling process in the implementation method.

[0039] Figure 18 This is a schematic diagram illustrating an example of a data structure representing the anomaly detection results in the implementation method.

[0040] Figure 19 This is a configuration diagram illustrating an example of the configuration of the model learning unit in the implementation method.

[0041] Figure 20 This is a flowchart illustrating an example of the model learning process in the implementation method.

[0042] Figure 21A This is a schematic diagram illustrating an example of information related to the driving path in the past during the implementation of the method.

[0043] Figure 21B This is a schematic diagram illustrating an example of information related to the driving path in the past during the implementation of the method.

[0044] Figure 22 This is a flowchart illustrating an example of the action of grouping the driving path determination parameters in the implementation method.

[0045] Figure 23A This is an example of a histogram representing the frequency of use of each driving path in the past during the implementation of the method.

[0046] Figure 23B This is an example of a histogram representing the frequency of use of each driving path in the past during the implementation of the method.

[0047] Figure 23C This is an example of a histogram representing the frequency of use of each driving path in the past during the implementation of the method.

[0048] Figure 23DThis is an example of a histogram representing the frequency of use of each driving path in the past during the implementation of the method.

[0049] Figure 24 This is a schematic diagram illustrating an example of the process of grouping driving path determination parameters in the implementation method.

[0050] Figure 25 This is a schematic diagram illustrating an example of the process of grouping driving path determination parameters in the implementation method.

[0051] Figure 26 This is a flowchart illustrating an example of the action of calculating the importance of the driving path determination parameters in the implementation method.

[0052] Figure 27A This is a schematic diagram illustrating an example of the process for calculating the importance of the driving path determination parameters in the implementation method.

[0053] Figure 27B This is a schematic diagram illustrating an example of the process for calculating the importance of the driving path determination parameters in the implementation method.

[0054] Figure 28 This is a flowchart illustrating an example of the learning and processing actions of the anomaly detection model in the implementation method.

[0055] Figure 29A This is a schematic diagram illustrating an example of the learning and processing process of the anomaly detection model in the implementation method.

[0056] Figure 29B This is a schematic diagram illustrating an example of the learning and processing process of the anomaly detection model in the implementation method. Detailed Implementation

[0057] (The process of obtaining a technical solution disclosed herein)

[0058] As a solution to labor shortages, services utilizing autonomous vehicles are expanding. However, IoT-enabled autonomous vehicles are vulnerable to unauthorized external intrusion into their control systems due to network connectivity or direct device connections. Once the control system is compromised, simple commands can trigger replay attacks, making them high-risk targets for cyberattacks. To address this, numerous anomaly detection methods and devices for remotely monitoring and monitoring autonomous vehicles have been proposed.

[0059] However, in anomaly detection of remotely operated autonomous vehicles, the excessive over-detection caused by the diversity of autonomous vehicle driving modes is considered a problem. Given the large number of anomaly targets, suppressing over-detection has become a challenge. Generally, in the area and location targeted, autonomous vehicles determine their own driving path based on their destination. Using information from LiDAR (Light Detection and Ranging) and / or cameras, they identify and avoid obstacles, and perform actions such as speed changes, turning, and illuminating turn signals. Therefore, methods that rely solely on threshold judgments of speed information, where the driving mode is difficult to determine, are prone to over-detection.

[0060] Therefore, as disclosed in Patent Document 1, an anomaly detection method was developed as follows: based on the current vehicle location information, the anomaly detection model is modified to perform anomaly detection that better matches the characteristics of the location, thereby suppressing over-detection. However, in the aforementioned anomaly detection method, in cases where the vehicle's location information has been tampered with by an attacker, or where, although the attacker has illegally operated the vehicle, the driving at that location is not significantly different from normal operation, the detection may be evaded and the anomaly cannot be determined.

[0061] In view of the above-mentioned issues, the inventors of this application conceived of the need to determine whether the current driving path and position of an autonomous vehicle traveling toward a pre-set destination are abnormal.

[0062] On the other hand, to determine whether an autonomous vehicle's current driving path is normal or abnormal, what logic is needed to select and decide the actual path from a large number of possible candidates? As mentioned above, autonomous vehicles automatically determine the optimal driving path from the origin to the destination based on map information such as the set destination, the user's region, and the location. However, the logic involved in determining the driving path is generally a black box, only used by the autonomous vehicle supplier, and not disclosed. This presents a problem where practitioners conducting remote vehicle monitoring and anomaly detection have difficulty obtaining this information.

[0063] Therefore, the inventors of this application considered whether it was possible to deduce the logic by which the autonomous vehicle selects and determines its driving path, and conducted actual verification experiments on the autonomous vehicle's driving. Furthermore, during the actual verification experiments, the inventors discovered that even with the same combination of origin and destination, the autonomous vehicle's driving path is not necessarily the same every time.

[0064] As one of the actual verification experiments, in Figure 1 , Figure 2 The diagram shows an example of route selection when an autonomous vehicle used in a goods transport service travels toward its destination.

[0065] Figure 1 Two paths (Path 1 and Path 2) from the starting point to the destination are shown, along with their branching points. Path 1, the upper path, is shorter to the destination but includes bridges over rivers and unpaved roads. Path 2, the lower path, is longer to the destination but is mostly paved. In actual verification experiments, even with the same combination of start and destination, the autonomous vehicle demonstrated its ability to... Figure 1 The autonomous vehicle travels on both the first and second paths. For example, on a sunny day, when transporting goods including fresh produce, the autonomous vehicle travels on the first path; on a rainy day, when transporting goods including fragile items, the autonomous vehicle travels on the second path.

[0066] Here, since the first route is the shortest distance to the destination, the autonomous vehicle's route is chosen based on factors such as the short distance or the short time required to reach the destination. Conversely, if the second route is chosen, the autonomous vehicle's route is chosen based on factors such as whether the route includes dangerous areas prone to river flooding in inclement weather, for example, avoiding the first route which includes bridges over rivers. Alternatively, the autonomous vehicle's route is chosen based on factors such as minimizing vibrations caused by road conditions, for example, avoiding the first route which includes unpaved roads.

[0067] in addition, Figure 2 The diagram also shows two paths (Path 3 and Path 4) from the starting point to the destination, and their branches. Path 3, the upper path, is shorter to the destination but includes a school route, while Path 4, the lower path, is longer but does not include a school route. Figure 2 In the example, also with Figure 1 Similarly, even with the same combination of origin and destination, actual verification experiments have confirmed that autonomous vehicles... Figure 2 The autonomous vehicle travels on both the third and fourth routes. For example, when driving at night and transporting goods including fresh produce, the autonomous vehicle travels on the third route; when driving in the morning or evening and transporting goods including fragile items, the autonomous vehicle travels on the fourth route.

[0068] Here, route 3 is the shortest distance to the destination. Therefore, when route 3 is chosen, the autonomous vehicle's route is determined based on the short distance to the destination or the short time required to reach the destination. In contrast, when route 4 is chosen, the autonomous vehicle's route is determined based on the low probability of a collision or the low probability of delays in reaching the destination due to road congestion. For example, route 3, which includes the school route, would not be chosen during school commuting hours.

[0069] Thus, it is believed that the driving route of an autonomous vehicle is not simply determined based on the distance to the destination, but rather, in addition to information such as the congestion and / or accident risk of the route at different times, the location of the danger, the presence and / or paving conditions of road construction, and the impact of weather on the route, when the autonomous vehicle is used for freight transport services, information such as the type of goods to be transported and / or the transport conditions can also be considered to determine the driving route.

[0070] Therefore, the inventors of this application conducted repeated and in-depth research, and as a result, conceived of a method to infer the driving path chosen and determined by an autonomous vehicle by presuming the information or factors considered when deciding on its driving path, as described above—in other words, the driving path determination parameters. Furthermore, an anomaly detection method was found, in which the optimal anomaly detection model is selected based on the inferred driving path to perform anomaly detection for the autonomous vehicle.

[0071] The first technical solution of this disclosure relates to an anomaly detection method, which is a computer-executed anomaly detection method for detecting anomalies in autonomous vehicles autonomously driving from a departure point to a destination. The method includes: an acquisition step, acquiring the driving path of the autonomous vehicle from the departure point to any point in time; and a detection step, reading information about multiple possible paths the autonomous vehicle may have taken from the departure point to the destination, based on the read information about the multiple paths, presuming one of the multiple paths as the presumed driving path of the autonomous vehicle from the departure point to the destination, calculating an anomaly degree, and detecting an anomaly in the autonomous vehicle if the calculated anomaly degree exceeds a predetermined threshold. The anomaly degree indicates the extent to which the driving path up to any point in time deviates from the presumed driving path.

[0072] Therefore, because of the presumed driving path of the autonomous vehicle, anomaly detection can be performed even if the location information of the autonomous vehicle is tampered with or if illegal operations are performed. That is, anomaly can be detected if the path traveled by the autonomous vehicle up to any point in time deviates from the presumed driving path.

[0073] Furthermore, the anomaly detection method of the second technical solution is further described in the anomaly detection method of the first technical solution, wherein when the travel path up to any point in time is obtained, or whenever a predetermined period has elapsed, information about multiple paths is obtained, and the information about multiple paths read in order to make an estimation of the estimated travel path includes information about multiple paths obtained in the past.

[0074] Therefore, it is possible to update information about multiple routes and to estimate the driving route corresponding to the condition of the route that may change over time.

[0075] In addition, the anomaly detection method involved in the third technical solution, in the anomaly detection method described in the first or second technical solution, further includes the following steps: when an anomaly of the autonomous vehicle is detected in the detection step, a detection result indicating that an anomaly has occurred in the autonomous vehicle and predetermined map information are displayed, wherein the predetermined map information plots the driving path from the starting point to any point in time and the estimated driving path.

[0076] Therefore, it is possible to output information related to the driving path of autonomous vehicles and information related to anomaly detection results.

[0077] Furthermore, in the anomaly detection method described in any of the first to third technical solutions, the information about multiple paths is a driving path determination parameter. The driving path determination parameter is a parameter composed of a combination of the values ​​of various events that indicate the presence or absence of multiple events. These multiple events are events that may affect the driving path determined by the autonomous driving vehicle from the origin to the destination. In the detection step, the estimated driving path is estimated based on the driving path determination parameter.

[0078] Therefore, as information about multiple paths, the driving path can be estimated based on the presence or absence of each of the multiple events that may affect the driving path determined by the autonomous vehicle.

[0079] Furthermore, in the anomaly detection method described in the fourth technical solution, the information regarding multiple paths includes at least one of the distance from the origin to the destination for each of the multiple paths and the time required for the autonomous vehicle to travel from the origin to the destination when it travels on each of the multiple paths.

[0080] Therefore, as information about multiple routes, it is possible to estimate the travel route based on information about distance or required time.

[0081] Furthermore, in the anomaly detection method described in the fourth or fifth technical solutions, the information regarding multiple paths in the sixth technical solution includes information representing events that may become obstacles when the autonomous vehicle is traveling on each of the multiple paths at any given time.

[0082] Therefore, it is possible to estimate the driving path based on information about events that may become obstacles to the autonomous vehicle's driving on each path.

[0083] Furthermore, in the anomaly detection method described in the sixth technical solution, the information representing events that may become obstacles when an autonomous vehicle is traveling on each of the multiple paths includes information indicating whether the autonomous vehicle is capable of traveling on each of the multiple paths.

[0084] Therefore, it is possible to estimate the driving path based on information indicating paths that autonomous vehicles cannot travel among multiple paths.

[0085] Furthermore, in the anomaly detection method of the eighth technical solution, in any of the anomaly detection methods described in the fourth to seventh technical solutions, the autonomous vehicle is a vehicle used for transporting goods. In the acquisition step, information about the goods being transported by the autonomous vehicle is further acquired. The information about the goods being transported includes at least one of the following: a travel path determination parameter representing the attributes of the goods being transported and at least one of the following: a set of information representing the transport conditions of the goods being transported. In the detection step, an estimated travel path is estimated based on the information about multiple paths and the information about the goods being transported.

[0086] Therefore, when the autonomous vehicle is used to transport goods, in addition to information about the route, the estimated travel route can also be estimated based on information about the goods being transported.

[0087] Furthermore, the anomaly detection method involved in the 9th technical solution, in any of the anomaly detection methods described in the 4th to 8th technical solutions, further includes a reading step of selecting and reading out one of the multiple anomaly detection models corresponding to each of the multiple driving path determination parameters. The multiple anomaly detection models respectively determine one of the multiple paths. In the reading step, the anomaly detection model corresponding to the driving path determination parameters is selected and read out. In the detection step, the path determined by the read out anomaly detection model is estimated as the estimated driving path.

[0088] Therefore, a path determined by the anomaly detection model corresponding to the driving path determination parameters can be estimated as the estimated driving path.

[0089] Furthermore, in the anomaly detection method described in the 9th technical solution, a path determined by multiple anomaly detection models is updated to one of the multiple paths at predetermined intervals.

[0090] Furthermore, in the anomaly detection method described in the 10th technical solution, the anomaly detection method of the 11th technical solution involves multiple anomaly detection models that determine multiple paths and path weights. The path weights represent the degree to which the autonomous vehicle travels on each of the multiple paths. The path weights are updated every predetermined period. Based on the path weights of the multiple paths, one of the multiple paths is determined as the updated path of each of the multiple anomaly detection models.

[0091] Based on these, a path that is updated every predetermined period can be estimated as a presumed travel path.

[0092] Furthermore, the anomaly detection method involved in the 12th technical solution, in the anomaly detection method described in the 10th or 11th technical solutions, further includes a learning step of learning multiple anomaly detection models respectively. In the learning step, at each predetermined period, past driving paths are obtained. These past driving paths are the driving paths of multiple past driving paths of an autonomous vehicle that has the same combination of values ​​of driving path determination parameters from the origin to the destination. Based on each of the obtained past driving paths, the frequency of autonomous driving vehicles driving on each of the multiple paths is calculated. Based on the calculated frequency, one of the multiple paths is determined as a path in the anomaly detection model corresponding to the driving path determination parameters.

[0093] Therefore, a path determined by the anomaly detection model can be updated based on multiple past driving records of autonomous vehicles. Thus, it is possible to make an inference of the presumed driving path that reflects the tendency of autonomous vehicles to choose driving paths in the past.

[0094] Furthermore, in the anomaly detection method described in any of the technical solutions 9 to 12, the multiple paths are each composed of a combination of one or more intervals connecting two locations arbitrarily determined on the multiple paths. In each of the multiple anomaly detection models, the weights of the intervals are determined according to one or more intervals. The weights of the intervals represent the degree to which the autonomous vehicle travels in each of the more than one intervals. In the detection step, the anomaly degree of the travel path up to the arbitrary time point is calculated based on the ratio of the sum of the weights of the intervals determined by the intervals of the more than one intervals included in the travel path from the starting point to any time point to the weights of the intervals determined by the intervals of the more than one intervals included in the estimated travel path. If the calculated anomaly degree exceeds a preset threshold, an anomaly is detected in the autonomous vehicle.

[0095] Therefore, it is possible to calculate the anomaly degree of the driving path according to each interval included in the driving path of the autonomous vehicle.

[0096] Furthermore, in the anomaly detection method described in the 13th technical solution, the 14th technical solution involves a detection step in which, if an event is determined that causes the autonomous vehicle to decide on an exceptional driving path that differs from the norm, and the driving path decision parameter is a parameter that includes a value indicating the existence of the event, a correction is made to reduce the degree of anomaly. If the corrected degree of anomaly exceeds a preset threshold, an anomaly is detected in the autonomous vehicle.

[0097] Therefore, it is possible to reduce the degree of abnormality when an autonomous vehicle performs an exceptional application that is different from the usual one, thus suppressing the occurrence of false detections.

[0098] Furthermore, the anomaly detection method involved in the 15th technical solution, in any of the anomaly detection methods described in the 9th to 14th technical solutions, is provided with multiple parameter groups for classifying arbitrary driving path determination parameters into multiple parameter groups, each of the multiple parameter groups determines one or more driving path determination parameters, each of the multiple anomaly detection models corresponds to each of the multiple parameter groups, in the detection step, it is determined which parameter group among the multiple parameter groups determines the driving path determination parameter, the anomaly detection model corresponding to the parameter group that determines the driving path determination parameter is selected and read out, and a path determined by the read out anomaly detection model is estimated as the estimated driving path.

[0099] Therefore, the parameters determining the driving path can be classified into multiple parameter groups, and a path determined by the anomaly detection model corresponding to each parameter group can be inferred as the inferred driving path. This reduces the number of anomaly detection models required.

[0100] Furthermore, in the anomaly detection method described in the 15th technical solution, the 16th technical solution involves determining the importance of a tendency to represent the value of one or more determined driving path determination parameters by multiple parameter groups. The importance is composed of a combination of multiple values ​​corresponding to the combination of values ​​represented by the driving path determination parameters. In the detection step, if no driving path determination parameter is determined in any of the multiple parameter groups, the anomaly detection model corresponding to the parameter group most similar to the importance and driving path determination parameter in the multiple parameter groups is selected and read out.

[0101] Therefore, the importance can be determined for each parameter group, and even when the combination of values ​​of the driving path determination parameters is a new combination, the driving path determination parameters can be classified into parameter groups.

[0102] Furthermore, the anomaly detection method of the 17th technical solution, in the anomaly detection method described in the 15th or 16th technical solutions, further includes a learning step of learning multiple anomaly detection models respectively. In the learning step, multiple driving path determination parameters and multiple past driving paths obtained in each of the multiple past drivings of the autonomous vehicle from the origin to the destination are obtained. The multiple past driving paths are the driving paths of the autonomous vehicle in the multiple past drivings. Multiple parameter groups are created by grouping the multiple driving path determination parameters determined in one or more of the multiple past driving paths that are similar to each other into the same parameter group.

[0103] Therefore, it is possible to classify multiple past driving path determination parameters that have become similar driving paths into the same parameter group, thus enabling more accurate classification of driving path determination parameters into each parameter group.

[0104] Furthermore, the anomaly detection method of the 18th technical solution, in the anomaly detection method described in the 16th technical solution, further includes a learning step of learning multiple anomaly detection models respectively. In the learning step, in each parameter group of multiple parameter groups, when the value of an event representing the presence or absence of one or more events included in the driving path determination parameters characterizes the parameter group, the value of importance corresponding to the value representing the presence or absence of that event is determined to be higher than the value in other cases. When the value of an event representing the presence or absence of one or more events included in the driving path determination parameters does not characterize the parameter group, the value of importance corresponding to the value representing the presence or absence of that event is determined to be lower than the value in other cases.

[0105] Therefore, the characteristics or tendencies of more than one path determination parameter included in each parameter group can be reflected in the importance, thus enabling more accurate classification of path determination parameters into each parameter group.

[0106] Furthermore, in the anomaly detection method described in the 17th technical solution, the anomaly detection method of the 19th technical solution, in the learning step, calculates the frequency of autonomous driving vehicles driving on each of the multiple past driving paths that are similar to a grouped parameter group, and determines the weight of each path based on the calculated frequency, thereby further creating an anomaly detection model.

[0107] Therefore, it is possible to create anomaly detection models based on multiple past driving records of autonomous vehicles, thus enabling the estimation of driving paths that reflect historical information.

[0108] Furthermore, the anomaly detection device disclosed in the 20th technical solution is an anomaly detection device for detecting anomalies in autonomous vehicles autonomously traveling from a departure point to a destination. It comprises: an acquisition unit that acquires the travel path of the autonomous vehicle from the departure point to any point in time; a storage unit that stores information about multiple possible paths that the autonomous vehicle may travel from the departure point to the destination; and an anomaly detection unit that, based on the information about the multiple paths read from the storage unit, presumes one of the multiple paths as the presumed travel path of the autonomous vehicle from the departure point to the destination, calculates an anomaly degree, and detects an anomaly in the autonomous vehicle if the calculated anomaly degree exceeds a predetermined threshold. The anomaly degree indicates how much the travel path up to any point in time deviates from the presumed travel path.

[0109] Therefore, because of the presumed driving path of the autonomous vehicle, anomaly detection can be performed even if the location information of the autonomous vehicle is tampered with or if illegal operations are performed. That is, anomaly can be detected if the path traveled by the autonomous vehicle up to any point in time deviates from the presumed driving path.

[0110] Furthermore, the 21st technical solution of this disclosure relates to a program for causing a computer to execute the anomaly detection method described in any of the technical solutions 1 to 19.

[0111] Therefore, because of the presumed driving path of the autonomous vehicle, anomaly detection can be performed even if the location information of the autonomous vehicle is tampered with or if illegal operations are performed. That is, anomaly can be detected if the path traveled by the autonomous vehicle up to any point in time deviates from the presumed driving path.

[0112] The embodiments described below are all specific examples of this disclosure. The numerical values, shapes, constituent elements, configurations of constituent elements, steps, and order of steps shown in the following embodiments are merely examples and are not intended to limit this disclosure. Furthermore, constituent elements in the following embodiments that are not described in the independent claims are described as arbitrary constituent elements. Additionally, some or all of the respective contents may be combined in all embodiments.

[0113] (Implementation Method)

[0114] The anomaly detection device involved in the implementation method will be described below.

[0115] In the anomaly detection of the driving path implemented by the anomaly detection device, the driving path determination parameters, which are external factors considered when determining the driving path, such as past driving conditions, current driving conditions, and current weather and / or road traffic information, are used as input to determine whether the current driving position is strange due to an attacker or due to system anomalies, and the information is prompted to the monitor.

[0116] <Composition of the Anomaly Detection System>

[0117] Figure 3 This is a diagram illustrating the configuration of the anomaly detection system in the implementation method. Figure 3 In this system, the anomaly detection system consists of 1000 monitored objects (autonomous vehicles), 5000 monitoring centers, 3000 external information sources, and 4000 networks.

[0118] The monitoring center 5000 is an on-premise analysis environment including an anomaly detection device 2000 and a remote monitoring device 5100. The monitoring center 5000 acquires information exchanged via a network 4000 through an autonomous vehicle network 1100 located remotely within the monitored object 1000, performs remote monitoring of the monitored object 1000, and analyzes and / or stores information based on anomaly detection. Furthermore, a monitor 5200 may also be present in the monitoring center 5000 to perform control of the monitored object 1000 based on the results of remote monitoring and / or anomaly detection. Additionally, at least a portion of the anomaly detection device 2000 and the remote monitoring device 5100 included in the monitoring center 5000 may be located in the cloud. Furthermore, the anomaly detection device 2000 may also be located outside the monitoring center 5000.

[0119] The anomaly detection device 2000 obtains information related to the driving of the monitored object 1000 from the autonomous vehicle network 1100 located remotely within the monitored object 1000 via the network 4000 and the monitoring center 5000, and performs analysis such as anomaly detection on the obtained information and / or saves the information. Alternatively, the anomaly detection device 2000 can also directly obtain information related to the driving of the monitored object 1000 from the monitored object 1000 via the network 4000.

[0120] The remote monitoring device 5100 acquires information related to the operation of the monitored object 1000 exchanged within the autonomous vehicle network 1100 within the monitored object 1000 via the network 4000 and the monitoring center 5000, and monitors the status of the monitored object 1000 based on the acquired information. Furthermore, the remote monitoring device 5100 can also transmit information related to the control of the monitored object 1000 to the monitored object 1000 via the network 4000. That is, the remote monitoring device 5100 can also remotely control the monitored object 1000.

[0121] The monitored object 1000 is an autonomous vehicle, such as an autonomous driving vehicle and / or a cargo transport vehicle, or a cleaning robot and / or a security robot. The monitored object 1000 uses map information about the object's location and surrounding area to determine a travel route from its initial point of origin to a predetermined destination, and then autonomously travels along that route. In this embodiment, the monitored object 1000 is described as an autonomous cargo transport vehicle used in a cargo transport service.

[0122] Figure 4 This is an example of a map information used by the monitored object 1000 when determining its driving path, represented in the form of a graph. Figure 4 In order to facilitate the processing of map information in the anomaly detection device 2000, circular nodes and solid lines are used for representation. Figure 4 In this diagram, the circular node represented by S1 indicates the starting point of the monitored object 1000, the circular node represented by G1 indicates the destination of the monitored object 1000, and the circular nodes represented by W1 to W4 indicate intermediate points corresponding to various locations predetermined in the map information. Intermediate points can be set at intersections in the map information, or at any location on the map accessible to autonomous vehicles. Furthermore, the starting point and destination can be either one of the intermediate points or, for example, any address different from an intermediate point. Additionally, the destination is not limited to one; multiple destinations may be determined. The solid lines represented by E1 to E9 represent intervals, each interval being a path that the monitored object 1000 can actually travel on, connecting the intermediate points, the starting point, and the destination. Furthermore, the edges are not limited to straight lines and may include curves in a manner corresponding to the actual map information. In this embodiment, the number of edges between nodes is one or less. That is, there is only one path connecting the intermediate points, the starting point, and the destination.

[0123] exist Figure 4In this context, the monitored object 1000, for example, decides on a driving path such as starting from the origin S1, traveling along path E1, passing through intermediate point W2, and further traveling along path E8 to reach the destination G1, and then autonomously drives. Furthermore, in... Figure 4 In this map, intermediate locations, starting points, and destinations are represented by circular nodes, and each path is represented by a solid line. However, the map information used by the monitored object 1000 when determining its travel path is not limited to this; it can be represented by other methods, or it can be an actual map and / or aerial photograph. Furthermore, the map information may also include, in addition to these, the distances of each path and / or the average travel time required for the monitored object 1000 to travel on each path. The map information used in the following embodiments is... Figure 4 The map information shown in the implementation is configured such that all locations and paths are set to "yes". Figure 4 The map information includes all locations and routes. Additionally, Figure 4 The map information shown is stored in the data storage unit 2300, but it can also be accessed through the external information retrieval unit 2120 (hereinafter referred to as the external information retrieval unit). Figure 5 (Description) Obtained from external information source 3000.

[0124] The monitored object 1000 has an autonomous vehicle network 1100, and transmits information related to the movement of the monitored object 1000, exchanged within the autonomous vehicle network 1100, to the anomaly detection device 2000 via the network 4000. Furthermore, the monitored object 1000 is not limited to a single autonomous vehicle, but can also consist of multiple autonomous vehicles. In the presence of multiple monitored objects 1000, each primary vehicle transmits its own information to the anomaly detection device 2000 via the network 4000.

[0125] Network 4000, for example, uses a Virtual Private Network (VPN) communication over a regular Internet connection to securely transmit information about the monitored object 1000 to the anomaly detection device 2000. Furthermore, network 4000 can also consist of multiple networks.

[0126] External information source 3000 provides various information related to vehicle operation, such as weather information, delivery information, and road traffic information, to anomaly detection device 2000 via network 4000. External information source 3000 may include, for example, a weather information server 3200 that processes weather information, a road traffic information server 3300 that processes road traffic information, a delivery management server 3100 that manages information on goods transported by freight vehicles, and an application management server 3400 that manages the use of autonomous vehicles. Furthermore, depending on the usage and service model of the autonomous vehicle, external information source 3000 may include servers or information sources that manage or provide information other than those mentioned above. For example, if the monitored object 1000 is a security robot used for guarding specific facilities or locations, external information source 3000 may include servers that manage or provide information related to the facilities, information related to human density, etc.

[0127] <Composition of the components of the anomaly detection device 2000>

[0128] The anomaly detection device 2000 consists of an anomaly detection unit 2100, an output unit 2200, a data storage unit 2300, an estimated model storage unit 2400, an anomaly detection model storage unit 2500, and a model learning unit 2600.

[0129] The anomaly detection unit 2100 performs anomaly detection based on information obtained from the monitored object 1000.

[0130] The output unit 2200 includes, for example, a display unit such as a UI (User Interface) to display to the monitor 5200 information visualized by mapping the anomaly detection results of the anomaly detection unit 2100 onto map information. Furthermore, if an anomaly is detected in the travel path of the monitored object 1000, the output unit 2200 can also output control instructions or notifications for the monitored object 1000 to the remote monitoring device 5100. For example, it can output control instructions to the remote monitoring device 5100 to cause the monitored object 1000 to stop urgently. In this embodiment, the output unit 2200 functions as a display unit.

[0131] The data storage unit 2300 stores and accumulates data related to the movement of the monitored object 1000 acquired by the anomaly detection unit 2100. Furthermore, the data storage unit 2300 may also store map information, and additionally, it may store information acquired by the anomaly detection unit 2100 related to the determination of the monitored object 1000's driving path, including information related to the status of each path in the map information. In this embodiment, the data storage unit 2300 is equivalent to a storage unit.

[0132] The estimation model storage unit 2400 stores the estimation model and the parameter information included in each driving path determination parameter group within the estimation model. This parameter information determines the types of driving path determination parameters that constitute the combination of driving path determination parameters. The estimation model, driving path determination parameter groups, and driving path determination parameters will be described later.

[0133] The anomaly detection model storage unit 2500 stores anomaly detection models corresponding to the driving path determination parameter groups. The anomaly detection models will be described later. Furthermore, in this embodiment, the estimation model storage unit 2400 and the anomaly detection model storage unit 2500 also function as storage units.

[0134] The model learning unit 2600 performs model learning processing based on data related to the driving of the monitored object 1000 stored in the data storage unit 2300, and learns the inference model and the anomaly detection model. The structure of the model learning unit 2600 and the model learning processing will be described later.

[0135] <Composition of the Anomaly Detection Unit 2100>

[0136] Figure 5 This diagram illustrates the configuration of the anomaly detection unit 2100 included in the anomaly detection device 2000 according to the embodiment. Figure 5 In this unit, the anomaly detection unit 2100 consists of a vehicle data acquisition unit 2110, an external information acquisition unit 2120, a driving path determination parameter estimation unit 2130, an anomaly detection model selection unit 2140, an anomaly determination unit 2150, and a display control unit 2160.

[0137] The vehicle data acquisition unit 2110 receives vehicle information related to the current driving of the monitored object 1000 from the monitored object 1000 via the network 4000. This vehicle information is generated based on vehicle control signals exchanged through the autonomous driving vehicle network 1100 of the monitored object 1000. The vehicle control signals referred to here are signals used to control the vehicle's actions and behaviors, such as engine, brakes, acceleration, and steering. The received vehicle information is also transmitted and stored in the data storage unit 2300.

[0138] Figure 6 This is a diagram showing an example of a data structure representing vehicle information of the monitored object 1000 received by the vehicle data acquisition unit 2110.

[0139] like Figure 6As shown, the generated vehicle information consists of a timestamp indicating the date and time, vehicle ID, driving speed and angular velocity, location information of monitored object 1000, intermediate points passed through, and the origin and destination of monitored object 1000. The vehicle ID is the identifier of monitored object 1000, the driving speed and angular velocity are information indicating the driving status of monitored object 1000, and the intermediate points passed through indicate intermediate points encountered during the journey up to this point. For example, for monitored object 1000 with vehicle ID "V001", the origin is shown to be "S1" and the destination to be "G1". It shows that on the date and time represented by the timestamp "1612304523", the vehicle traveled at a speed of "3.423…", an angular velocity of "0.23…", at a latitude of "32.123…", and a longitude of "131.532…", passing through intermediate point W2 during the journey up to this point. Furthermore, the vehicle information is not limited to… Figure 6 The information shown may also include information other than those mentioned above. For example, vehicle information may include information about the use of turn signals and hazard warning lights, as it relates to the operation of equipment while driving.

[0140] Furthermore, the vehicle data acquisition unit 2110 can also generate vehicle information related to driving by receiving vehicle control signals from the monitored object 1000 and parsing the received vehicle control signals. That is, the vehicle data acquisition unit 2110 receives vehicle control signals in data formats such as CAN (Controller Area Network), FlexRay, or ROS, and generates vehicle information through parsing and data processing. In this case, vehicle information is generated whenever the vehicle data acquisition unit 2110 receives and parses a vehicle control signal.

[0141] The external information acquisition unit 2120 acquires external information related to the determination of the driving route from the external information source 3000. For example, in addition to information related to the transported goods, the external information includes weather information, road traffic information, information related to the status of each route in the map information, and operational management information of the autonomous vehicle. The received external information is also transmitted and stored in the data storage unit 2300. Furthermore, in this embodiment, the vehicle data acquisition unit 2110 and the external information acquisition unit 2120 are equivalent to acquisition units.

[0142] Figures 7-9 This is a diagram showing an example of external information obtained by the external information acquisition unit 2120 from the external information source 3000 that is related to the determination of the driving path.

[0143] Figure 7This diagram illustrates an example of information related to the goods being transported by the monitored object 1000, obtained by the external information acquisition unit 2120 when the monitored object 1000 is an autonomous vehicle used for goods transport. Information related to the transported goods, such as that provided by the goods transport service provider or the system at the goods' shipping location, is stored in the distribution management server 3100, which is one of the external information sources 3000. The external information acquisition unit 2120 obtains this information from the distribution management server 3100.

[0144] exist Figure 7 In the example shown, the information related to the transported item consists of the item ID, information indicating whether the item is perishable or fragile (representing the type and conditions of the item), information indicating whether the conditions are time-bound, the weight of the item, the shipping location, and the destination (delivery address). The item ID is an identifier assigned to each item. The perishable, fragile, and time-bound items can take values ​​of 0 or 1. That is, the perishable and fragile items are valued as 1 if the item is perishable or fragile, and as 0 if it is not perishable or fragile. The time-bound item is valued as 1 if the conditions are time-bound, and as 0 if the conditions are not time-bound. Figure 7 For example, for a transported item with an ID of "001", the system indicates that the item type is perishable goods, that strict time-bound transport conditions are set, that the weight is "3kg", that the shipping location is "S1", and that the destination as the delivery address is "G1". Furthermore, information related to the transported item is not limited to... Figure 7 The information shown may also include information other than those mentioned above. For example, it may include information indicating that the type of goods being transported is not moisture-resistant, refrigerated, or frozen. In addition, a specified delivery time may be set as a transport condition.

[0145] Figure 8 This diagram illustrates an example of weather information acquired by the external information acquisition unit 2120. The external information acquisition unit 2120 acquires weather information, for example, from a weather information server 3200, which provides weather information as one of the external information sources 3000.

[0146] exist Figure 8 In the example shown, the weather information is based on weather forecasts and consists of date and time, weather conditions, expected precipitation, and location information. Figure 8For example, the weather forecast for the location "Area A" on the date and time "2022-08-01-AM" is "Sunny" with an expected precipitation of "0mm". Furthermore, the external information acquisition unit 2120 typically acquires weather information about the area where the monitored object 1000 is located, but it may also acquire weather information about the surrounding areas of the area where the monitored object 1000 is located. The weather information is not limited to... Figure 8 The information shown may also include information other than those mentioned above. For example, it may include information such as wind speed, snowfall, temperature, and humidity based on weather forecasts.

[0147] Figure 9 This diagram illustrates an example of road traffic information acquired by the external information acquisition unit 2120. The external information acquisition unit 2120 acquires road traffic information, for example, from a road traffic information server 3300, which provides road traffic information services and is one of the external information sources 3000.

[0148] exist Figure 9 In the example shown, road traffic information consists of the content of an event that occurs on the road and affects traffic, the date and time of the event, and information about the location or area where the event occurred. Figure 9 For example, it shows that a "construction" event occurred in the interval "E2, E4" starting from the date and time "2022-08-01". The external information acquisition unit 2120 typically acquires road traffic information about the intervals where the monitored object 1000 may travel, but it may also acquire road traffic information about surrounding intervals. Furthermore, the road traffic information typically includes only information about ongoing events and does not include information about eliminated events, but it is not limited to this and may also include information about all events that have occurred. In addition, it may also include, for example, information indicating whether travel is possible in the interval where the event occurred, information indicating the duration of the event, or information about the location where the event occurred.

[0149] Figure 10 This diagram illustrates an example of application management information acquired by the external information acquisition unit 2120, which manages the daily usage status of autonomous vehicles. The application management information is stored in an application management server 3400, which is, for example, one of the external information sources 3000 used by a monitor 5200 or an operator performing remote monitoring and anomaly detection of autonomous vehicles. The external information acquisition unit 2120 obtains the application information from the application management server 3400.

[0150] exist Figure 10In the example shown, the management information consists of information about the following intervals: the registration date and time of the interval, interval details indicating the type of event in the registered interval, and information about the time period during which congestion occurs in the interval due to the event. These intervals are registered by monitor 5200 or practitioners performing remote monitoring and anomaly detection of the monitored object 1000 as indicating the presence of events that may become obstacles to the movement of the monitored object 1000. Figure 10 For example, the section details for "E6, E7" are shown as "School Route," registered on the date and time "2022-08-01," with congestion times for this section being "8-10" and 15-17." Furthermore, the use of management information is not limited to... Figure 10 The information shown may also include information other than those mentioned above. For example, it may include path information about the characteristics of the sections that the monitored object 1000 may travel through, information about dangerous locations, and information indicating the registration or duration of events in the sections. Additionally, it may include information indicating whether each section is traversable within a specific time period.

[0151] In addition, the external information acquisition unit 2120 can acquire information from external information source 3000, for example... Figure 4 The map information shown can also provide information such as the distance of each path in the map information and the average time required for the monitored object 1000 to travel on each path. Alternatively, external information can be directly obtained in the form of a combination of the travel path determination parameters described later.

[0152] The above is an example of external information related to determining the travel path when the monitored object 1000 is a goods transport vehicle, as described in this embodiment. However, if the monitored object 1000 is an autonomous vehicle used for services other than goods transport vehicles, the external information related to determining the travel path may differ from the above. For example, if the monitored object 1000 is a security robot used for security of specific facilities or locations, the external information may not include information related to the transported goods, but may include information such as the rank of the VIP to be visited, the room to be used, or the density of people cooperating with the indoor LPS (Local Positioning System) to determine the priority of patrol routes. Thus, the external information related to determining the travel path may vary depending on the purpose of the monitored object 1000.

[0153] Explanation of the structure of the anomaly detection unit 2100.

[0154] The driving path determination parameter estimation unit 2130 obtains vehicle information of the monitored object 1000 from the vehicle data acquisition unit 2110, and also obtains various external information related to the determination of the driving path from the external information acquisition unit 2120. Furthermore, it obtains grouping information of the combinations of driving path determination parameters involved in the driving path determination from the estimation model storage unit 2400. Based on the obtained vehicle information and various external information, it determines the combination of driving path determination parameters for the monitored object 1000 in this driving session. The driving path determination parameters and their combinations will be described later. In this embodiment, "this driving session" refers to a series of driving sessions from a predetermined starting point to reaching all predetermined destinations, as will be explained later.

[0155] The anomaly detection model selection unit 2140 obtains the combination of driving path determination parameters that affect the current driving path determination from the driving path determination parameter estimation unit 2130. Additionally, it obtains an estimation model from the estimation model storage unit 2400. Then, based on the combination of driving path determination parameters, it performs the anomaly detection model selection process (described later) to select an appropriate anomaly detection model for the current driving path and retrieves it from the anomaly detection model storage unit 2500.

[0156] The anomaly determination unit 2150 uses the anomaly detection model selected and acquired by the anomaly detection model selection unit 2140 to perform anomaly determination processing on the current vehicle information of the monitored object 1000 acquired by the vehicle data acquisition unit 2110. In the anomaly determination processing, information on frequently used driving paths (i.e., estimated driving paths) associated with combinations of driving path determination parameters included in the anomaly detection model is compared with the current driving situation to determine anomalies. Details of the anomaly determination processing will be described later.

[0157] If the anomaly determination unit 2150 determines that there is an anomaly in the anomaly determination process, the display control unit 2160 transmits the determination result to the output unit 2200 and displays it in order to notify the monitor 5200 that the monitored object 1000 may be traveling along an abnormal path. Map information may also be transmitted along with the determination result.

[0158] <Actions for anomaly detection and handling>

[0159] Next, the anomaly detection process performed by the anomaly detection device 2000 will be explained. Figure 11 This is a flowchart illustrating an example of anomaly detection processing in an implementation method.

[0160] In the anomaly detection process, firstly, the driving path determination parameter estimation unit 2130 obtains vehicle information of the monitored object 1000 from the vehicle data acquisition unit 2110, and obtains various external information related to the determination of the driving path from the external information acquisition unit 2120. Additionally, parameter information is obtained from the estimation model storage unit 2400. The parameter information determines the types of driving path determination parameters included in the combination of driving path determination parameters. Furthermore, based on the obtained parameter information, the combination of driving path determination parameters for the monitored object 1000 during this driving phase, which is the object of anomaly detection, is determined according to the vehicle information and various external information (S7100).

[0161] Here, the driving path determination parameter is an indicator representing the presence or absence of various external information elements that may affect the determination of the driving path of the monitored object 1000. It can take a value of 0 or 1. 0 indicates the absence of the element, and 1 indicates the presence of the element. Elements that may affect the determination of the driving path of the monitored object 1000 include, for example, the transported goods being fragile, the weather being rainy, and anticipated traffic congestion in the area.

[0162] Furthermore, the combination of driving route determination parameters includes a set of individual driving route determination parameters and information about the origin and destination included in the vehicle information of the monitored object 1000. An example of the combination of driving route determination parameters determined by the driving route determination parameter estimation unit 2130 in step S7100 is shown below. Figure 12 .

[0163] Figure 12 It shows Figure 6 The vehicle ID shown, "V001", indicates that monitored object 1000 was moved on the date and time "2022-08-01-AM". Figure 7 The combination of travel path determination parameters is shown for the case where the transport item ID is represented by "001". Furthermore, in... Figure 12 In the example shown, the departure location of the transported goods is the departure point of the monitored object 1000, and the destination of the transported goods is the destination of the monitored object 1000. Here, the information that the monitored object 1000, whose vehicle ID is "V001", transports the transported goods, whose transported goods ID is "001", is input to the external information source 3000 or the monitored object 1000 by, for example, the monitor 5200 or the worker who loads the transported goods onto the monitored object 1000. The travel path determination parameter estimation unit 2130 determines the combination of travel path determination parameters based on the information of the transported goods associated with each monitored object 1000.

[0164] exist Figure 12In this context, the combination of route determination parameters consists of vehicle information and route determination parameters assigned based on information related to the transported goods, weather information, road traffic information, and application management information. As for vehicle information, it is shown that... Figure 6 The origin and destination of monitored object 1000, whose vehicle ID is represented by "V001," are shown. Information related to the transported goods is shown... Figure 7 The values ​​for items with transport item IDs represented by "001" (perishable goods, fragile goods, and time-sensitive goods) are shown as driving route determination parameters. Additionally, for items indicating whether or not the item is loaded, values ​​indicating whether the monitored object 1000 is currently transporting the item are shown as driving route determination parameters. Here, the driving route determination parameter for items indicating whether or not the item is loaded is 1 if the monitored object 1000 is currently transporting the item, and 0 if or not the item is not being transported. For example, when returning from a delivered item, moving from the garage to the collection point (shipping location), the monitored object 1000 is in a state without a loaded item; therefore, the driving route determination parameter is 0. Weather information is shown... Figure 8 The weather corresponding to the travel time of monitored object 1000 is shown as a travel route determination parameter. Here, the travel route determination parameter is 1 when the weather is "sunny". In the road traffic information and operation management information, a method based on... Figure 9 and Figure 10 The information shown regarding each section indicates whether each section is a congested section as expected, and is used as a driving route determination parameter. Here, the driving route determination parameter is 1 if the monitored object 1000 travels on each route on the date and time, and if each route is a congested section; and 0 if it is not a congested section. Furthermore, the method for determining the driving route determination parameter and the combination of driving route determination parameters are not limited to the above. For example, it may include parameters indicating sections where the monitored object 1000 is temporarily unable to travel, parameters indicating that the section is a school route, parameters based on precipitation, etc. Additionally, when the external information acquisition unit 2120 obtains a combination of driving route determination parameters as external information, the obtained combination of driving route determination parameters is determined as the combination of driving route determination parameters for the monitored object 1000's current travel.

[0165] Refer again Figure 11Next, in step S7100, the anomaly detection model selection unit 2140 obtains the combination of driving path determination parameters determined by the driving path determination parameter estimation unit 2130. Additionally, the anomaly detection model selection unit 2140 obtains an estimated model from the estimated model storage unit 2400. Then, the anomaly detection model selection unit 2140 performs anomaly detection model selection processing, selecting a group of driving path determination parameters from the estimated model that corresponds to the combination of driving path determination parameters determined in step S7100, and obtaining the anomaly detection model corresponding to the selected group of driving path determination parameters from the anomaly detection model storage unit 2500 (S7200).

[0166] Here, the presumed models stored in the presumed model storage unit 2400 will be explained. Figure 13A and Figure 13B This is a diagram representing an example of a presumptive model. In Figure 13A and Figure 13B In the model, the parameter group A is determined by the driving path. Figure 13A ), Driving route determination parameter group B ( Figure 13B This consists of multiple groups of driving path determination parameters. Additionally, the inferred model includes a third and subsequent group of driving path determination parameters, such as group C (not shown). A driving path determination parameter group includes a combination of more than one driving path determination parameter and the importance of the driving path determination parameters.

[0167] For a combination of more than one route determination parameter contained in a route determination parameter group, at least the origin and destination are the same, but the individual route determination parameters within the group are different. The combinations of route determination parameters are grouped into any route determination parameter group through the route determination parameter grouping process described later. When the monitored object 1000 determines its route based on combinations of route determination parameters grouped into the same route determination parameter group, these are considered similar routes. Figure 13A and Figure 13B The model only shows a portion of the route determination parameter group with origin S1 and destination G1, but the inferred model may include route determination parameter groups with other origins or other destinations.

[0168] The importance of each driving path decision parameter represents the degree to which each parameter influences the driving path decision made by the monitored object 1000. A driving path parameter with a high importance indicates that it will significantly influence the driving path decision within that group. The calculation process for parameter importance will be described later, but it is calculated in a way that driving path decision parameters with a large number of values ​​of "1" in multiple combinations within the same group have high importance. Furthermore, for driving path decision parameters with a high proportion of values ​​of "1" in other groups, their importance is calculated as lower than that of driving path decision parameters with a low proportion of values ​​of "1" in other groups. Conversely, for driving path decision parameters with a low proportion of values ​​of "1" in other groups, their importance is calculated as higher than that of driving path decision parameters with a high proportion of values ​​of "1" in other groups. Figure 13A In route determination parameter group A, the importance of the route determination parameter "Cargo or Not" is 0.5, "Fresh Produce" is 0.9, and "Strict Timekeeping" is 0.8. These three route determination parameters are more important than the other route determination parameters within the group. This is because there are more values ​​of "1" for these route determination parameters within the group. Furthermore, the route determination parameters "Fresh Produce" and "Strict Timekeeping" are more important than "Cargo or Not" because the proportion of values ​​of "1" is low in other groups. In other words, the importance within a group can be considered a representative value indicating a tendency or characteristic to be easily combined among the multiple route determination parameters included in that route determination parameter group.

[0169] In addition, the anomaly detection model stored in the anomaly detection model storage unit 2500 will be explained. Figures 14A-14C This is a diagram representing an example of an anomaly detection model. Figures 14A-14C In this study, an anomaly detection model is associated with each group of driving route decision parameters. The anomaly detection model consists of a departure point, a destination, and weights for each path. The weights of each path are based on their probability, which is the probability that each path will be selected when the driving route is determined by the combination of driving route decision parameters within the group for the monitored object 1000. Paths with a high probability of selection have lower weights, while paths with a low probability of selection have higher weights.

[0170] The weights of each path are determined through learning and processing by the anomaly detection model, as described later. Figure 14AIn the anomaly detection model corresponding to the driving path determination parameter group A, the weight of path "E1" is 2, and the weight of path "E2" is 6. That is, this means that path E1 may be selected with a higher probability than E2 when determining the driving path.

[0171] In addition, the anomaly detection model includes average models for each combination of origin and destination. The average model is an anomaly detection model composed of the average of all anomaly detection models, correlated with all groups of driving path determination parameters identical to the combinations of origin and destination. Figure 14C The figure shows the average model, which is composed of the average of all anomaly detection models with origin S1 and destination G1.

[0172] Next, the selection and processing of anomaly detection models will be explained.

[0173] The anomaly detection model selection process is performed by the anomaly detection model selection unit 2140. It involves selecting an anomaly detection model from the anomaly detection models stored in the anomaly detection model storage unit 2500 that is suitable for the combination of driving path determination parameters determined by the driving path determination parameter estimation unit 2130, i.e., the combination of driving path determination parameters related to the path determination in this driving scenario. Figure 15 The flowchart shows an example of an anomaly detection model selecting and processing actions.

[0174] In the anomaly detection model selection process, it is first determined whether any one of the multiple driving path determination parameter groups included in the estimated model stored in the estimated model storage unit 2400 includes a combination of driving path determination parameters that is the same as the combination of driving path determination parameters used in this instance (S7210). For example, in this embodiment, the combination of driving path determination parameters used in this instance is... Figure 12 In the case of the combination shown, determine whether the combination exists in any group of driving path decision parameters. If it is determined that the same combination exists (step S7210: yes), proceed to step S7220; if it is determined that the same combination does not exist (step S7210: no), proceed to step S7230.

[0175] The anomaly detection model selection unit 2140 retrieves the anomaly detection model associated with the driving path determination parameter group, which includes the same combination of driving path determination parameters as the current driving path determination parameters, from the anomaly detection model storage unit 2500 (S7220). Then, the anomaly detection model selection process ends.

[0176] Alternatively, to determine which group of driving path decision parameters the current combination of driving path decision parameters is similar to in terms of features, the anomaly detection model selection unit 2140 calculates the similarity between the current combination of driving path decision parameters and the importance of driving path decision parameters in each group of driving path decision parameters included in the inferred model (S7230). Furthermore, the similarity between driving path decision parameter groups that share the same origin and destination combinations as the current combination of driving path decision parameters is calculated. For example, a method utilizing cosine similarity can be used to calculate the similarity. In the following explanation of the similarity calculation method, for simplicity, only the combinations and importance of driving path decision parameters for "whether there is cargo," "fresh goods," "fragile goods," and "strict adherence to time" are used to calculate the similarity between group A and group B. However, in practice, the similarity is calculated for all groups based on the combinations and importance of all driving path decision parameters.

[0177] For example, in Figure 13A and Figure 13B In the driving route determination parameter group A, the importance of the four driving route determination parameters "loaded or unloaded", "fresh goods", "fragile goods", and "strict adherence to time" is [0.5, 0.9, 0.1, 0.8], while the importance of the four driving route determination parameters in driving route determination parameter group B is [0.5, 0.1, 0.9, 0.1]. In this case, when the combination of the four driving route determination parameters is as follows... Figure 12 When the given values ​​are [1, 1, 0, 1], the similarity between the current route decision parameter set and groups A and B is calculated using cosine similarity. The similarity for group A is 0.97, which is greater than the similarity for group B, which is 0.39. Therefore, in the above example, it can be determined that the current route decision parameter set is more similar to the route decision parameter group A, which has a higher similarity.

[0178] Next, a determination is made (S7240) as to whether the similarity among the groups of driving path determination parameters calculated in step S7230, which share the same combination of origin and destination, exceeds a pre-set threshold. If the similarity is determined to exceed the threshold (step S7240: Yes), the process proceeds to step S7250; otherwise, it proceeds to step S7260. The threshold is determined to be any value between 0 and 1, for example, 0.7. The threshold is determined, for example, by the monitor 5200.

[0179] The anomaly detection model selection unit 2140 retrieves the anomaly detection model associated with the group of driving path decision parameters that has the highest similarity to the combination of driving path decision parameters in this case from the anomaly detection model storage unit 2500 (S7250). For example, if the similarity to group A is the highest, the anomaly detection model associated with group A is retrieved. Then, the anomaly detection model selection process ends.

[0180] Alternatively, since there is no group in the group of driving path determination parameters stored in the anomaly detection model storage unit 2500 that is significantly similar to the combination of driving path determination parameters for the current time in a manner exceeding a threshold, the anomaly detection model selection unit 2140 obtains from the anomaly detection model storage unit 2500 the average model corresponding to the combination of origin and destination included in the combination of driving path determination parameters for the current time (S7260). Then, the anomaly detection model selection process ends.

[0181] Return to the instructions for anomaly detection and handling.

[0182] The anomaly determination unit 2150 uses the anomaly detection model selected and acquired by the anomaly detection model selection unit 2140 to perform anomaly determination processing on the current vehicle information of the monitored object 1000 acquired by the vehicle data acquisition unit 2110 (S7300). Further, the anomaly determination result and map information are transmitted to the output unit 2200 and output via the display control unit 2160. Then, the anomaly detection processing ends.

[0183] The anomaly detection processing will be explained. Anomaly detection processing is performed by the anomaly detection unit 2150, and it determines the degree of anomaly (abnormality) of the current driving of the monitored object 1000 based on the weights of each path included in the anomaly detection model. In this embodiment, for example, the anomaly detection model obtained by the anomaly detection model selection unit 2140 is the anomaly detection model corresponding to the driving path determination parameter group A, and the vehicle information obtained by the vehicle data acquisition unit 2110 is... Figure 6 The vehicle ID shown is explained by the vehicle information of monitored object 1000, represented by "V001". Figure 16 This is a flowchart illustrating an example of an exception handling procedure. Additionally, an example of the exception handling process is shown below. Figure 17A and Figure 17B .

[0184] The anomaly determination unit 2150 acquires the current vehicle information of the monitored object 1000 acquired by the vehicle data acquisition unit 2110. Next, based on the information including intermediate points passed through the vehicle information, it determines the path traveled by the monitored object 1000 up to the current point, starting from its origin. Furthermore, for the determined path, it calculates the sum of the weights of each path traveled up to the current point using the weights of each path included in the anomaly detection model (S7310). For example, as... Figure 17B As shown, the time stamp "1612304523" indicates the date and time, and the intermediate location for monitored object 1000 is "W2". Since the departure point is "S1", therefore, as in... Figure 17A As indicated by the dashed arrow, the path traveled up to the present is determined to be only "E1". Therefore, based on the weight value of path E1, the total weight of all paths traveled up to this date and time is calculated to be "2". Similarly, since the monitored object 1000's intermediate location for the date and time represented by the timestamp "1612304527" is "W2→W3→W1", therefore, as in Figure 17A As shown by the arrow pointing to the straight line, the path traveled up to the present is determined to be "E1→E7→E4". Therefore, the total weight of each path traveled up to this date and time is calculated as "12" because 2+3+7=12.

[0185] The anomaly determination unit 2150 uses the weights of each path included in the anomaly detection model to determine a presumed driving path that the monitored object 1000 is expected to travel. Furthermore, it calculates the sum of the weights in the determined presumed driving path (S7320). The presumed driving path is, for example, a combination of paths from the starting point to the destination, where the sum of the weights of each path is the minimum value. Figure 17A In this system, there are multiple combinations of paths from the starting point to the destination, but the combination of paths with the minimum total weight is the path from S1 to G1 via "E1→E8", as shown by the dashed arrow. In this embodiment, this path is the estimated driving path, and the total weight is calculated as "3" because 2+1=3. The estimated driving path is determined based on the anomaly detection model, but the anomaly detection model is selected based on the combination of driving path parameters determined in step S7100. Therefore, in other words, the estimated driving path is determined based on the combination of driving path parameters.

[0186] Next, the anomaly determination unit 2150 compares the total weight of the paths traveled so far, calculated in step S7310, with the total weight of the estimated travel path, calculated in step S7320, and calculates the anomaly degree (S7330). The anomaly degree indicates the extent to which the paths traveled by the monitored object 1000 so far are abnormal. The anomaly degree is calculated as the ratio of the total weight of the paths traveled so far to the total weight of the estimated travel path. Figure 17B In the above calculation, since the total weight of the paths traveled up to the date and time represented by the timestamp "1612304523" is 2, the anomaly score becomes 2 / 3 = 0.666..., and is calculated as "0.67". Similarly, since the total weight of the paths traveled up to the date and time represented by the timestamp "1612304527" is 12, the anomaly score becomes 12 / 3 = 4, and is calculated as "4". The calculation of the anomaly score is not limited to the above. For example, if the total weight of the paths traveled up to the present is greater than the total weight of the estimated traveled paths, the difference between the total weights of the two paths can be simply calculated as the anomaly score.

[0187] The anomaly determination unit 2150 determines whether to correct the anomaly degree calculated in step S7330 (S7340). For example, when a driving path determination parameter is determined to be a factor that causes the monitored object 1000 to make an exceptional driving path selection that is different from the usual one, if there is a driving path determination parameter that is the factor among the driving path determination parameters related to the current driving path selection of the monitored object 1000, it is determined that correction should be made. Factors that cause an exceptional driving path selection include, for example, driving in maintenance mode during operation, driving in an area where manual driving must always be performed, etc. In such cases, a driving path selection may be made according to the maintenance content, a driving path selection may be based on the experience of the driver performing manual driving, or a driving path that is different from the usual one. As a result, a deviation may occur from the estimated driving path determined by the learned anomaly detection model, and the reliability of the calculated anomaly degree may become lower. Therefore, by making corrections in such cases to reduce the anomaly degree, over-detection can be suppressed. In addition, the anomaly determination unit 2150 may also determine to perform correction if the anomaly detection model used is of a specific category and if the correction value β for correcting the anomaly degree is preset in the anomaly determination unit 2150. If the anomaly degree is determined to be corrected because there is a parameter that should be corrected (step S7340: Yes), the process proceeds to step S7350; if the anomaly degree is determined not to be corrected because there is no parameter that should be corrected (step S7340: No), the process proceeds to step S7360.

[0188] After the anomaly determination unit 2150 corrects the anomaly degree calculated in step S7330 using the correction value β (S7350), it proceeds to step S7360. For example, it performs corrections such as dividing the anomaly degree calculated in step S7330 by the correction value β or subtracting the correction value β from the anomaly degree calculated in step S7330 in order to reduce the anomaly degree.

[0189] The anomaly determination unit 2150 determines whether the anomaly degree calculated in step S7330 or the anomaly degree corrected in step S7350 exceeds a preset threshold θ (S7360). For example, ... Figure 17B As shown, when the threshold θ is set to "θ=3", if the anomaly degree is greater than "3", the driving path of monitored object 1000 up to the present is determined to be abnormal. Figure 17B In this example, the path traveled up to the date and time represented by the timestamp "1612304523" has an anomaly score of 0.67, therefore it is considered normal. However, the path traveled up to the date and time represented by the timestamp "1612304527" has an anomaly score of 4, exceeding the threshold θ, therefore it is considered abnormal. Furthermore, in this embodiment, the threshold θ is set to "θ = 3", but the threshold θ can also be any value other than "θ = 3". Additionally, the threshold θ can be set according to each anomaly detection model, or a unified value can be set across all anomaly detection models.

[0190] Furthermore, the anomaly determination unit 2150 transmits the anomaly determination result and map information to the output unit 2200 via the display control unit 2160 and outputs them. Then, the anomaly determination process ends. Alternatively, the anomaly determination unit 2150 may only transmit the anomaly determination result to the output unit 2200 and output it if the path's anomaly degree exceeds a threshold θ. Therefore, if the path's anomaly degree is below the threshold θ, the output unit 2200 may not output anything.

[0191] The output transmitted from the anomaly detection unit 2150 to the output unit 2200 via the display control unit 2160 includes, for example, combinations of... Figure 17A and Figure 17B The anomaly detection results and map information shown are displayed. Alternatively, for example, it can be drawn on an actual map. Figure 17AThe output unit 2200 outputs information such as locations, paths, and the weights of each path from the map information. This provides information that allows for immediate visual understanding of driving paths that differ from the norm, helping the monitor 5200 and others to identify anomalies in the current movement of the monitored object 1000, facilitating subsequent analysis. Alternatively, the output unit 2200 can also output information after plotting the estimated driving path determined in step S7320 onto the map information.

[0192] In addition to the anomaly determination result, the current vehicle information of the monitored object 1000 acquired by the vehicle data acquisition unit 2110 in step S7310 can also be output. For example, the following can also be output: Figure 18 The data is obtained by merging the result of the anomaly determination and the vehicle information obtained by the vehicle data acquisition unit 2110, as shown. Figure 18 The data shown includes Figure 6 The current vehicle information of the monitored object 1000 shown includes various data and the anomaly determination result. The anomaly determination result includes, for example, the anomaly degree calculated in step S7330 or corrected in step S7350, the anomaly determination result based on the anomaly degree in step S7360, and a data sequence ID representing a continuous sequence of anomaly determinations. For the data sequence ID, if it is an anomaly determination result based on a continuous series of moving vehicle information of the monitored object 1000 from its origin to its destination, the data sequence ID is assigned the same value. For example, in... Figure 18 In this sequence, timestamps from "1612304523" to "1612304527" represent the same monitored object 1000 moving from the same origin towards the same destination, thus constituting the same data sequence, and are assigned the same value as the data sequence ID (S001). On the other hand, timestamp "1612304530" represents the same monitored object 1000 moving from the same origin towards a different destination, thus constituting a different data sequence, and are assigned a different value as the data sequence ID (S002). Furthermore, the data sequence ID is assigned to the anomaly determination result and output when anomaly detection processing is performed.

[0193] <Composition of Model Learning Unit 2600>

[0194] Next, the structure of the model learning unit 2600 will be explained.

[0195] Figure 19 This diagram illustrates the configuration of the model learning unit 2600 included in the anomaly detection device 2000. Figure 19In this model learning unit 2600, there are a data acquisition unit 2610, a parameter combination and grouping unit 2620, a driving path determination parameter importance calculation unit 2630, and an anomaly detection model learning unit 2640.

[0196] The data acquisition unit 2610 acquires, from the data storage unit 2300, vehicle information of the monitored object 1000 in the past and combinations of driving path determination parameters involved in the driving path determination for each past driving period, based on information from an arbitrarily set learning period.

[0197] The parameter combination grouping unit 2620 acquires data from the data acquisition unit 2610. Next, based on the acquired vehicle information from past driving and the combinations of driving path determination parameters involved in each past driving path determination, the parameter combination grouping unit 2620 groups the combinations of driving path determination parameters for vehicles that have performed similar driving actions. Furthermore, the parameter combination grouping unit 2620 saves the grouped results as an estimation model in the estimation model storage unit 2400.

[0198] The driving path determination parameter importance calculation unit 2630 obtains the grouping information of driving path determination parameter combinations from the parameter combination grouping unit 2620. Furthermore, the driving path determination parameter importance calculation unit 2630 calculates the importance of each driving path determination parameter under each driving path determination parameter group as a measure of its contribution to the driving path determination of that group, and appends the calculation results to the estimation model stored in the estimation model storage unit 2400.

[0199] The anomaly detection model learning unit 2640 obtains past driving data and driving path determination parameters involved in driving path determination at each driving time from the data storage unit 2300. Additionally, the anomaly detection model learning unit 2640 obtains driving path determination parameter groups and information representing the frequency of use of driving paths corresponding to the driving path determination parameter groups from the parameter combination grouping unit 2620. Furthermore, the anomaly detection model learning unit 2640 learns the anomaly detection model according to each group and saves the learned anomaly detection model in the anomaly detection model storage unit 2500.

[0200] <Actions processed by model learning>

[0201] Next, the model learning process will be explained.

[0202] Figure 20 This is a flowchart representing an example of the actions the model learns and processes.

[0203] The anomaly detection device 2000 defines driving path determination parameters (S6100) from a plurality of parameters that may be included in information obtained from the external information source 3000, which are considered to be related to driving path determination. Specifically, as parameters related to driving path determination, the anomaly detection device 2000 defines parameters preset by the monitor 5200 as driving path determination parameters. Information about the defined driving path determination parameters is stored in the data storage unit 2300, the anomaly detection unit 2100, the model learning unit 2600, etc.

[0204] The parameters related to route determination are set by the monitor 5200 after studying the service category and content of the autonomous vehicle used as the monitored object 1000. For example, if the monitored object 1000 is an autonomous vehicle used for cargo transportation, parameters representing information about the type of cargo being transported and information about the weather can be set as parameters related to route determination. In addition, parameters representing road conditions and road traffic information can also be set as parameters related to route determination. Furthermore, generally, in various services using autonomous vehicles, the service providers of the autonomous vehicles implement the optimal route determination algorithm that does not cause obstacles to operation; therefore, various parameters can be set as parameters related to route determination depending on the utilization of the autonomous vehicle.

[0205] The data acquisition unit 2610 acquires information about the travel routes included in the past travel information of the monitored object 1000 from the data storage unit 2300 (S6200). The information about the travel routes includes at least the origin, the routes taken by the monitored object 1000, and the destination. Furthermore, the information about the travel routes may also include information about intermediate points passed by the monitored object 1000. An example of the acquired past travel route information is shown below. Figure 21A .exist Figure 21A In the diagram, the information about the travel route consists of a travel number, a date and time indicating the start or end date and time of the trip, the origin, the destination, and information indicating whether or not the trip occurred on each route. The travel number is an identifier assigned to the monitored object 1000 for each trip from its origin to its destination. The information indicating whether or not the trip occurred on each route takes a value of 0 or 1, indicating that the monitored object 1000 traveled on the routes assigned a value of 1 and did not travel on the routes assigned a value of 0. For example, in a trip where the travel number is "1", the monitored object 1000's origin is shown to be "S1", it traveled on routes "E3, E4, E5, E9", and arrived at its destination "G1".

[0206] The data acquisition unit 2610 acquires from the data storage unit 2300 a combination of driving path determination parameters related to the past driving path determination, corresponding to the driving path information obtained in step S6200 (S6300). Figure 21A The combination of driving path determination parameters shown in the previous driving times is expressed as follows: Figure 21B .exist Figure 21B In this diagram, the combination of route determination parameters consists of the vehicle number corresponding to the vehicle number in the route information and each route determination parameter. Furthermore, the route determination parameters for each congested section (shown as vehicle numbers 1 to 4 in the diagram) will be the same for the following explanation.

[0207] The parameter combination grouping unit 2620 acquires the combination of past driving path information and corresponding driving path determination parameters acquired by the data acquisition unit 2610 in steps S6100 and S6200. Furthermore, based on the past driving path information, the parameter combination grouping unit 2620 performs driving path determination parameter grouping processing, grouping combinations of driving path determination parameters (S6400). Thus, driving path determination parameter groups are created, and each group is further associated with the past driving path information.

[0208] The grouping process for determining the driving route is explained. Figure 22 This is a flowchart illustrating an example of the driving path determination parameter grouping process performed by the parameter grouping unit 2620.

[0209] The parameter combination grouping unit 2620 generates a histogram related to the frequency of use of the driving path based on information from past driving paths (S6410). Regarding... Figure 21A The histograms of each past journey shown are, for example, Figures 23A to 23D The vertical axis shown represents a histogram of usage frequency. Furthermore, in Figures 23A to 23D For simplicity, the combination of driving route determination parameters corresponding to past driving route information is only represented for the items "whether there is cargo", "fresh produce", "fragile goods", "strict adherence to time", and "weather".

[0210] Next, the parameter combination grouping unit 2620 groups histograms that show similar usage frequency of each past driving path based on the histograms (S6420). Then, the grouped histograms are summarized into a single histogram. Furthermore, the parameter combination grouping unit 2620 groups the combinations of driving path determination parameters corresponding to each grouped histogram as driving path determination parameter groups, thus ending the driving path determination parameter grouping process.

[0211] As a method for calculating the similarity of the tendencies of each histogram, for example, the Earth Mover's Distance (EMD) algorithm can be used to calculate the distance between two histograms. Then, by using agglomerative clustering, histograms that are close in EMD distance, i.e., similar, are grouped together one by one, and the groups are successively summarized, thereby summarizing them into multiple groups in which histograms with similar tendencies are grouped together. The summarized results are then stored in the inference model storage unit 2400 and transmitted to the anomaly detection model learning unit 2640.

[0212] exist Figure 24 and Figure 25 Showing the Figures 23A to 23D This is an example of the process of grouping and summarizing the four histograms shown. First, the parameter combination grouping unit 2620 calculates the parameters in all combinations. Figures 23A to 23D The similarity between two histograms in the four histograms shown is then calculated. Next, the parameter combination grouping unit 2620 groups the histograms with the highest similarity (above a threshold) and summarizes the histograms. In this embodiment, since... Figure 24 As shown, the histograms representing driving numbers 1 and 4 have the highest similarity, therefore, they are grouped together, and the histograms are grouped into group 1. Simultaneously, combinations of driving path determination parameters are also grouped, resulting in combinations containing two driving path determination parameters within that group.

[0213] Next, the parameter combination grouping unit 2620 calculates the similarity between two histograms in all combinations: the histogram grouped into group 1, the histogram representing the ungrouped travel number (represented by 2), and the histogram representing the travel number (represented by 3). Furthermore, when multiple histogram groups exist, the similarity between groups can be calculated; if the similarity between groups is above a threshold, they are grouped together. The parameter combination grouping unit 2620 groups the groups with the highest similarity (above the threshold) and summarizes the histograms. In this embodiment, as... Figure 25As shown, since the histogram representing driving number 2 has the highest similarity to the histogram of group 1, they are grouped together. That is, the histogram representing driving number 2 is grouped into group 1. At the same time, the combinations of driving path determination parameters are also grouped, becoming combinations with three driving path determination parameters in this group.

[0214] The parameter combination grouping unit 2620 performs the same grouping process sequentially until the similarity between histograms becomes less than a threshold. Furthermore, if the calculated similarity is less than the threshold, the parameter combination grouping unit 2620 ends the grouping process after treating the ungrouped histograms as independent groups. In this embodiment, since... Figure 25 As shown, the similarity between the histogram of group 1 and the histogram of ungrouped vehicle No. 3 is less than the threshold. Therefore, the histogram of vehicle No. 3 is not grouped into group 1, but becomes an independent group 2.

[0215] Thus, each group, which contains one or more driving path determination parameters, is stored as a driving path determination parameter group in the estimation model storage unit 2400. Furthermore, the summarized histograms corresponding to each driving path determination parameter group are transmitted to the anomaly detection model learning unit 2640.

[0216] In this way, by grouping the driving path decision parameters, combinations of driving path decision parameters using similar driving paths can be aggregated. Therefore, even if the number of combinations of driving path decision parameters increases, the increase in the number of anomaly detection models can be suppressed. Thus, compared to simply selecting an anomaly detection model based on each combination of driving path decision parameters, problems such as insufficient data for learning and the potential for overlearning can be suppressed.

[0217] Return to the description of the model learning process.

[0218] The driving path determination parameter importance calculation unit 2630 obtains the grouping information of the driving path determination parameter combinations after grouping in step S6400 from the parameter combination grouping unit 2620. Next, the driving path determination parameter importance calculation unit 2630 performs driving path determination parameter importance calculation processing, calculating the importance of driving path determination parameters related to the driving path determination of each group in each driving path determination parameter group (S6500). Furthermore, the driving path determination parameter importance calculation unit 2630 adds the calculated driving path determination parameter importance to the driving path determination parameter group, thereby saving it as an estimation model in the estimation model storage unit 2400.

[0219] The calculation and processing of the importance of the driving path determination parameters are explained.

[0220] Figure 26 This is a flowchart illustrating an example of the actions performed by the driving path determination parameter importance calculation unit 2630, specifically the calculation of the driving path determination parameter importance. Additionally, [further details will be provided]. Figure 25 An example of the process for calculating the importance of driving path determination parameters in the driving path determination parameter group, represented by group 1, which includes a combination of three driving path determination parameters, is shown in [the figure]. Figure 27A and Figure 27B .

[0221] First, the driving route determination parameter importance calculation unit 2630 calculates the frequency of each driving route determination parameter with a value of 1 in each group after grouping in step S6400 (S6510). Furthermore, the driving route determination parameter importance calculation unit 2630 determines the driving route determination parameters with a high frequency of a value of 1. For example, in... Figure 27A Among the route determination parameter groups shown, the route determination parameters with a value of 1 that are most frequent are identified as "Cargo or Not", "Fragile Goods", and "(Congested Area) E2".

[0222] Next, for each route determination parameter, the proportion of route determination parameters with a high frequency of being identified as having a value of 1 in other groups is calculated. Furthermore, based on the calculated proportion and the frequency of having a value of 1 calculated in step S6510, the importance of the route determination parameter in each group is calculated, for example using the TF-IDF (Team Frequency-Inverse Document Frequency) method (S6520). That is, a route determination parameter that has a high frequency of having a value of 1 within its group but a low proportion of having a high frequency of having a value of 1 in other groups can be considered a group-specific route determination parameter, and therefore, its importance is calculated as high. Conversely, a route determination parameter that has a high frequency of having a value of 1 within its group but a high proportion of having a high frequency of having a value of 1 in other groups has its importance calculated as lower than the above. Furthermore, a route determination parameter that has a low frequency of having a value of 1 within its group has its importance calculated as even lower than the above. Furthermore, the importance score can range from 0 to 1. Additionally, there can be multiple path determination parameters specific to a group, with the latter being assigned a high importance score.

[0223] For example, in Figure 27AIn the shown group of route determination parameters, the route determination parameters "Loaded or Unloaded", "Fragile Goods", and "(Congested Section) E2", which have a high frequency of 1, are calculated to be of high importance. Furthermore, regarding the route determination parameter for "Fragile Goods", since the proportion of route determination parameters with a high frequency of 1 in other groups is low, this route determination parameter can be considered a group-specific route determination parameter. Therefore, as... Figure 27B As shown, the importance was calculated as a higher 0.9.

[0224] In this way, by assigning importance to each path determination parameter, even when a new combination of path determination parameters that has not been observed in the past appears during the anomaly detection model selection process, the similarity to which group this new combination of path determination parameters is similar can be calculated. Therefore, even with a new combination of path determination parameters, an appropriate group of path determination parameters can be selected.

[0225] Return to the description of the model learning process.

[0226] The anomaly detection model learning unit 2640 obtains the driving path determination parameter groups grouped in step S6400 from the driving path determination parameter combination grouping unit 2620. Additionally, the anomaly detection model learning unit 2640 obtains a summarized histogram representing the frequency of use of driving paths corresponding to each driving path determination parameter group. Then, it performs anomaly detection model learning processing (S6600), saving the driving path determination parameter groups including the learned anomaly detection model in the anomaly detection model storage unit 2500, and ends the model learning processing.

[0227] The learning process of the anomaly detection model is explained.

[0228] Figure 28 This is a flowchart illustrating an example of the anomaly detection model learning process performed by the anomaly detection model learning unit 2640. In this process, the anomaly detection model is learned to determine the legitimacy or anomalousness of a driving path based on a group of parameters determined by the driving path.

[0229] An anomaly detection model is associated with each group of driving path decision parameters. This model determines the frequency of use of each driving path across multiple driving paths based on the driving path decision parameters within each group. By applying the anomaly detection model in the anomaly detection process, the legitimacy or anomalousness of the driving path selected during the driving of monitored object 1000 can be determined. For example... Figures 14A-14CAs shown, the anomaly detection model comprises a origin, a destination, and weights, where the weights are set based on the frequency of use of each path from the origin to the destination. An example of the learning process for the anomaly detection model is illustrated below. Figure 29A and Figure 29B Figure 29 illustrates the learning process of the anomaly detection model corresponding to the driving path determination parameter group that has been grouped into group 1 via step S6420.

[0230] The anomaly detection model learning unit 2640 obtains the histogram of driving paths (S6610) summarized in step S6400, which determines the parameter groups for each driving path. For example, it obtains... Figure 29A The histogram shown represents the frequency of use for each travel route.

[0231] Next, the anomaly detection model learning unit 2640 calculates the utilization probability of each path based on the histogram of the driving path obtained in step S6610 (S6620). For example, in Figure 29B In the calculation, the utilization probability of paths E3 and E9 is calculated as 1, the utilization probability of paths E4 and E5 is calculated as 0.67, and the utilization probability of path E6 is calculated as 0.33. Furthermore, the anomaly detection model learning unit 2640, when there are paths that are never used and are calculated as having a utilization probability of 0, adds a small value α to all paths for smoothing. Figure 29B As shown, α is, for example, 0.1.

[0232] The anomaly detection model learning unit 2640 obtains the number of instances where the combination of origin and destination represented by each travel path decision parameter group matches the combination of origin and destination for each trip during the past travel of the monitored object 1000. That is, the number of matching instances indicates the number of times the monitored object 1000 has traveled with the combination of origin and destination represented by each travel path decision parameter group. Furthermore, the anomaly detection model learning unit 2640 determines whether the number of trips for each travel path decision parameter group that is the object of the anomaly detection model learning process is below a threshold (S6630). If it is determined that it is not below the threshold (step S6630: No), it proceeds to step S6640; if it is determined that it is below the threshold (step S6630: Yes), it proceeds to step S6650.

[0233] The anomaly detection model learning unit 2640 calculates the cost of utilizing each path based on the utilization probability calculated in step S6620, and sets the weight of each path based on the utilization cost (S6640). The cost of utilizing each path is calculated, for example, by taking the negative log-likelihood of the utilization probability of each path. Furthermore, the weight of each path is set, for example, based on the ratio of the costs of utilizing each path. Figure 29B In this example, the weight of path E1 is set to 9, the weight of path E2 is set to 9, the weight of path E3 is set to 3, and so on. Furthermore, the weight values ​​for each path can be set after appropriate scaling, as long as the ratio remains unchanged. For example, by scaling the weight values ​​based on the maximum sum of the weights of the paths traveled from a starting point to a destination, the threshold setting for anomaly detection in step S7360 can be easily performed.

[0234] Alternatively, the anomaly detection model learning unit 2640 calculates the cost of utilizing each path and sets the weight of each path based on the value obtained by adding the utilization probability of each path calculated in step S6620 to the average utilization probability of each path in the group of all driving paths with the same origin and destination. This allows for setting the weight of each path based on the average utilization probability across all data, even when the number of trips from origin to destination represented by the driving path determination parameter group is below a threshold and the learning may be insufficient. Therefore, it can suppress the deterioration of anomaly detection accuracy and the increase in over-detection caused by insufficient learning.

[0235] Industrial availability

[0236] This invention can be widely used as an anomaly detection device for detecting anomalies related to the driving of autonomous vehicles.

[0237] Explanation of reference numerals in the attached figures

[0238] 1000 monitored objects

[0239] 1100 Autonomous Vehicle Network

[0240] 2000 Anomaly Detection Device

[0241] 2100 Anomaly Detection Department

[0242] 2110 Vehicle Data Acquisition Department

[0243] 2120 External Information Acquisition Department

[0244] 2130 Driving Path Determination Parameter Estimation Department

[0245] 2140 Anomaly Detection Model Selection Department

[0246] 2150 Anomaly Detection Department

[0247] 2160 Display Control Unit

[0248] 2200 Output Section

[0249] 2300 Data Storage Department

[0250] 2400 Estimated Model Storage Unit

[0251] 2500 Anomaly Detection Model Storage Unit

[0252] 2600 Model Learning Department

[0253] 2610 Data Acquisition Department

[0254] 2620 Parameter Combination Grouping Department

[0255] 2630 Driving Path Determination Parameter Importance Calculation Department

[0256] 2640 Anomaly Detection Model Learning Department

[0257] 3000 external information sources

[0258] 3100 Delivery Management Server

[0259] 3200 Weather Information Server

[0260] 3300 Road Traffic Information Server

[0261] 3400 uses a management server

[0262] 4000 Network

[0263] 5000 Monitoring Center

[0264] 5100 Remote Monitoring Device

[0265] 5200 Watchers

Claims

1. An anomaly detection method, a computer-executed anomaly detection method for detecting anomalies in autonomous vehicles autonomously driving from a starting point to a destination, comprising: The step is to obtain the driving path of the autonomous vehicle from the starting point to any point in time. and The detection step involves reading information about multiple possible routes that the autonomous vehicle may travel from the origin to the destination. Based on the read information about the multiple routes, one of the multiple routes is presumed as the presumed travel route of the autonomous vehicle from the origin to the destination. Anomaly is calculated, and if the calculated anomaly exceeds a predetermined threshold, an anomaly is detected in the autonomous vehicle. The anomaly indicates how much the travel route up to the arbitrary time point deviates from the presumed travel route.

2. The anomaly detection method according to claim 1, Furthermore, when the travel route up to the arbitrary point in time is obtained, or whenever a predetermined period has elapsed, information about the plurality of routes is obtained. The information about the plurality of routes read in order to estimate the estimated travel route includes information about the plurality of routes obtained in the past.

3. The anomaly detection method according to claim 1, It also includes the following display steps: if an anomaly of the autonomous vehicle is detected in the detection step, the detection result of the anomaly of the autonomous vehicle and the predetermined map information are displayed, wherein the predetermined map information plots the driving path from the starting point to the arbitrary time point and the estimated driving path.

4. The anomaly detection method according to claim 1, Information about the multiple paths is a parameter for determining the driving route. The driving route determination parameter is a combination of values ​​representing the presence or absence of multiple events, which are events that may influence the autonomous vehicle's determination of the driving route from the origin to the destination. In the detection step, the estimated driving path is estimated based on the driving path determination parameters.

5. The anomaly detection method according to claim 4, Information about the plurality of paths includes at least one of the following: the distance from the origin to the destination for each of the plurality of paths, and the time required for the autonomous vehicle to travel from the origin to the destination if the autonomous vehicle has traveled on each of the plurality of paths.

6. The anomaly detection method according to claim 4, Information about the multiple paths includes information representing events that might become obstacles if the autonomous vehicle is traveling on any of the multiple paths at any given time.

7. The anomaly detection method according to claim 6, Information representing events that could become obstacles when the autonomous vehicle is traveling on each of the plurality of paths includes information indicating whether the autonomous vehicle is able to travel on each of the plurality of paths.

8. The anomaly detection method according to claim 4, The autonomous vehicle is a vehicle used for transporting and moving goods. In the acquisition step, further information about the transported items carried by the autonomous vehicle is obtained. The information about the transported object includes at least one of the following: the travel path determination parameters, which are information representing the attributes of the transported object and information representing the transport conditions of the transported object. In the detection step, the estimated travel path is estimated based on information about the plurality of paths and information about the transported object.

9. The anomaly detection method according to claim 4, It also includes a step of selecting and reading out an anomaly detection model from among multiple anomaly detection models corresponding to each of the multiple driving path determination parameters. One path is determined from the multiple paths by the multiple anomaly detection models. In the readout step, the anomaly detection model corresponding to the driving path determination parameters is selected and read out. In the detection step, a path determined by the read anomaly detection model is presumed as the presumed driving path.

10. The anomaly detection method according to claim 9, A path determined by the plurality of anomaly detection models is updated to one of the plurality of paths at predetermined intervals.

11. The anomaly detection method according to claim 10, The multiple anomaly detection models determine the multiple paths and their weights, where each path weight represents the probability that the autonomous vehicle will travel on each of the multiple paths. The weights of the paths are updated at predetermined intervals. Based on the weights of each of the multiple paths, one of the multiple paths is determined as the updated path of each of the multiple anomaly detection models.

12. The anomaly detection method according to claim 10, It also includes learning steps for learning the multiple anomaly detection models respectively. In the learning steps, according to each predetermined period, Obtain past driving paths, which are the driving paths of multiple past trips of the autonomous vehicle from the origin to the destination, where the values ​​of the driving path determination parameters are the same. Based on each of the multiple past driving paths obtained, the frequency of the autonomous vehicle driving on each of the multiple paths is calculated. Based on the calculated frequency, one of the multiple paths is determined as a path in the anomaly detection model corresponding to the driving path determination parameters.

13. The anomaly detection method according to claim 9, Each of the multiple paths is formed by combining one or more intervals connecting two locations from a plurality of arbitrarily determined locations on the multiple paths. In each of the plurality of anomaly detection models, the weight of each interval is determined according to one or more intervals, wherein the weight of each interval represents the probability of the autonomous vehicle driving in each of the one or more intervals. In the detection step, The anomaly degree of the travel path up to the arbitrary time point is calculated based on the ratio of the sum of the weights of the intervals determined by each interval of the more than one interval included in the travel path from the origin to the arbitrary time point, to the weights of the intervals determined by each interval of the more than one interval included in the estimated travel path. If the calculated anomaly exceeds a preset threshold, an anomaly is detected in the autonomous vehicle.

14. The anomaly detection method according to claim 13, In the detection step, If an event is identified that causes the autonomous vehicle to decide on an exceptional driving path that differs from the norm, and the driving path decision parameters include parameters in the combination that indicate the presence of such an event, a correction is made to reduce the degree of anomaly. If the corrected anomaly level exceeds a preset threshold, an anomaly is detected in the autonomous vehicle.

15. The anomaly detection method according to claim 9, The system includes multiple parameter groups for classifying any given driving path determination parameters into these groups. One or more driving path determination parameters are determined by each of the multiple parameter groups. Each of the multiple anomaly detection models corresponds to a different parameter group among the multiple parameter groups. In the detection step, Determine which of the multiple parameter groups the driving path determination parameter is determined by. Select and read out the anomaly detection model corresponding to the parameter group that has determined the driving path determination parameters. The path determined by the anomaly detection model is presumed as the presumed driving path.

16. The anomaly detection method according to claim 15, The importance of each of the multiple parameter groups is determined by expressing the tendency to represent the value of one or more of the determined driving path determination parameters. The importance is composed of a combination of multiple values ​​corresponding to the combination of values ​​represented by the driving path determination parameters. In the detection step, if none of the parameter groups in the plurality of parameter groups determines the driving path determination parameter, the anomaly detection model corresponding to the parameter group in the plurality of parameter groups that is most similar to the importance and the driving path determination parameter is selected and read out.

17. The anomaly detection method according to claim 15, It also includes learning steps for learning the multiple anomaly detection models respectively. In the learning steps, The system obtains multiple driving path determination parameters and multiple past driving paths obtained from various past driving journeys of the autonomous vehicle from the origin to the destination, wherein the multiple past driving paths are the driving paths of the autonomous vehicle during the multiple past driving journeys. The plurality of parameter groups are created by grouping the plurality of driving path determination parameters determined in one or more of the plurality of past driving paths that are similar to each other into the same parameter group.

18. The anomaly detection method according to claim 16, It also includes learning steps for learning the multiple anomaly detection models respectively. In the learning step, within each of the plurality of parameter groups, When the value of one of the multiple events included in the driving path determination parameters represents the group of parameters, the importance value corresponding to the value indicating the presence or absence of that event is determined to be higher than the value in other cases. When the value of an event representing the absence of one or more of the events included in the driving path determination parameters characterizes the parameter group, the value of importance corresponding to the value representing the presence or absence of that event is determined to be lower than the value in other cases.

19. The anomaly detection method according to claim 17, In the learning step, the frequency of the autonomous vehicle driving on each of the multiple past driving paths corresponding to a grouped parameter group is calculated. Based on the calculated frequency, the weight of each path is determined for each of the multiple paths, thereby further creating the anomaly detection model.

20. An anomaly detection device for detecting anomalies in an autonomous vehicle traveling from a starting point to a destination, comprising: The acquisition unit acquires the driving path of the autonomous vehicle from the starting point to any point in time. A storage unit that stores information about multiple routes the autonomous vehicle may take from the origin to the destination; and An anomaly detection unit, based on information about the plurality of paths read from the storage unit, presumes one of the plurality of paths as the presumed driving path of the autonomous vehicle from the origin to the destination, calculates an anomaly degree, and detects an anomaly in the autonomous vehicle if the calculated anomaly degree exceeds a predetermined threshold. The anomaly degree indicates how much the driving path up to the arbitrary point in time deviates from the presumed driving path.

21. A program for causing the anomaly detection method according to any one of claims 1 to 19 to be executed by the computer.

Images