Vehicle control device and program

Abstract

A vehicle control device (10) installed in a vehicle (101) comprises a scene estimating unit (14) that inputs sensor information acquired by a sensor (20) into an inference model and outputs, for each of a plurality of predetermined specific scenes, a scene reliability indicating the degree of similarity between a scene in which the vehicle (101) is traveling and each of the specific scenes, a determining unit (15) that, if any of the scene reliabilities output by the scene estimating unit is equal to or greater than a predetermined reliability threshold, determines that an acquisition condition has been met, a data acquiring unit (16) that, when it is determined that the acquisition condition has been met, acquires data relating to the sensor information, and a transmitting unit (17) that transmits, to a server (102) via a communication network (103), the data acquired by the data acquisition unit, wherein the inference model is machine-trained in advance using training data including information relating to the plurality of specific scenes.

Description

Vehicle control device and program CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application is based on Japanese Application No. 2024-069042 filed on April 22, 2024, the contents of which are incorporated herein by reference.

[0002] The present disclosure relates to a vehicle control device and a program.

[0003] In recent years, a method has been adopted in which various data actually acquired from a sold vehicle while it is in operation is collected in a server via a communication network, and the collected data is analyzed to be used in future vehicle development or to update the vehicle's control program. Such a technique is described, for example, in Patent Document 1.

[0004] Japanese Patent Application Laid-Open No. 2023-53031

[0005] When collecting data, it is desirable to be able to collect data for each scene in which a vehicle is traveling in order to analyze the data. However, the timing for acquiring data is determined in advance by a rule, such as when the vehicle speed exceeds a threshold, and it is therefore not possible to properly identify a scene and collect data. This has led to problems such as insufficient collection of data for a specific scene or, conversely, excessive collection of data for a specific scene.

[0006] The present disclosure has been made in consideration of the above circumstances, and has as its main object to provide a vehicle control device and a program that can efficiently collect necessary data.

[0007] A vehicle control device that solves the above problem is a vehicle control device mounted on a vehicle, and includes: a scene estimation unit that inputs sensor information acquired by a sensor mounted on the vehicle into an inference model and outputs a scene reliability indicating the degree to which the scene in which the vehicle is traveling is similar to each of a plurality of predetermined specific scenes; a determination unit that determines that an acquisition condition is met if any of the scene reliability output by the scene estimation unit is equal to or greater than a predetermined reliability threshold; a data acquisition unit that acquires data related to the sensor information when the determination unit determines that the acquisition condition is met; and a transmission unit that transmits the data acquired by the data acquisition unit to a server via a communication network, wherein the inference model has been machine-learned in advance using training data containing information about the plurality of specific scenes.

[0008] With the above configuration, compared to when vehicle driving scenes are estimated based on rules, the use of an inference model makes it possible to estimate scenes more appropriately. Also, it is possible to eliminate the effort of creating rules, thereby reducing the burden on design.

[0009] The program for solving the above problem is a program executed by a vehicle control device mounted on a vehicle, and causes the vehicle control device to execute the following steps: a scene estimation step in which sensor information acquired by a sensor mounted on the vehicle is input into an inference model and a scene reliability indicating the degree to which the scene in which the vehicle is traveling is similar to each of a plurality of predetermined specific scenes; a determination step in which, if any of the scene reliability output by the scene estimation step is equal to or greater than a predetermined reliability threshold, an acquisition condition is determined to be met; a data acquisition step in which, if the determination step determines that the acquisition condition is met, data related to the sensor information is acquired; and a transmission step in which the data acquired by the data acquisition unit is transmitted to a server via a communication network; and the inference model is machine-learned in advance using training data containing information about the plurality of specific scenes.

[0010] With the above configuration, compared to when vehicle driving scenes are estimated based on rules, the use of an inference model makes it possible to estimate scenes more appropriately. Also, it is possible to eliminate the effort of creating rules, thereby reducing the burden on design.

[0011] The above and other objects, features, and advantages of the present disclosure will become more apparent from the following detailed description taken in conjunction with the accompanying drawings, in which Fig. 1 is a schematic diagram of a data collection system, Fig. 2 is a block diagram showing functions performed by a verification processing device, Fig. 3 is a diagram for explaining output of scene reliability, Fig. 4 is a diagram for explaining acquisition conditions, Fig. 5 is a flowchart of data acquisition processing, Fig. 6 is a diagram for explaining a method for changing acquisition conditions, Fig. 7 is a block diagram showing functions performed by a data collection system according to a third embodiment, Fig. 8 is a block diagram showing functions performed by a data collection system according to a fourth embodiment, Fig. 9 is a block diagram showing functions performed by a verification processing device according to a fifth embodiment, Fig. 10 is a flowchart of data acquisition processing according to the fifth embodiment, Fig. 11 is a flowchart of data acquisition processing according to a modified example, and Fig. 12 is a block diagram showing functions performed by a verification processing device according to the modified example.

[0012] Hereinafter, embodiments of a vehicle control device and a program according to the present disclosure will be described in detail with reference to the drawings. Note that, between the embodiments and modifications, the same or corresponding parts in the drawings are designated by the same reference numerals, and their description will not be repeated in principle.

[0013] 1 shows a data collection system 100 to which a vehicle control device 10 according to this embodiment is applied. The vehicle control device 10 is mounted on a vehicle 101, and controls the vehicle 101 and performs driving assistance.

[0014] As shown in Fig. 1, a data collection system 100 includes a server 102 and is capable of communicating with one or more vehicles 101 via a communication network 103 such as the Internet. Fig. 1 shows only one vehicle 101. The vehicle 101 includes a vehicle control device 10, a sensor 20, an actuator 30, and the like.

[0015] The sensor 20 includes various sensors for measuring the traveling conditions of the vehicle 101, such as a vehicle speed sensor 21, an acceleration sensor 22, and a yaw rate sensor 23. The sensors for measuring the traveling conditions of the vehicle 101 may include other sensors, or any of the vehicle speed sensor 21, the acceleration sensor 22, and the yaw rate sensor 23 may not be included.

[0016] The sensor 20 also includes various sensors for detecting other vehicles and obstacles, such as a camera 24 and a millimeter-wave radar 25. The sensor for detecting other vehicles and obstacles is not limited to the millimeter-wave radar 25, but may also be a laser radar (LiDAR), an ultrasonic sensor, or a combination of these. The camera 24 may also be a monocular camera or a compound camera. The camera 24 may also capture either still images or videos. The number, position, and type of the cameras 24 may be changed as desired. For example, the vehicle may include a front camera that captures the area in front of the vehicle, a right-side camera that captures the area on the right side of the vehicle, a left-side camera that captures the area on the left side of the vehicle, and a rear camera that captures the area behind the vehicle.

[0017] The sensor 20 also includes various sensors for detecting various amounts of operation by the driver, such as an accelerator sensor 26 that detects the amount of accelerator operation by the driver, a brake sensor 27 that detects the amount of brake operation, and a steering angle sensor 28 that detects the amount of steering (steering angle) of the steering wheel by the driver.

[0018] These sensors 20 are connected to the vehicle control device 10 wirelessly or by wire, and the measurement results (or detection results) of these sensors 20 are input to the vehicle control device 10 or the like as sensor information.

[0019] The actuators 30 include, for example, actuators for driving the vehicle 101, such as a motor that serves as the main engine of the vehicle 101. The actuators 30 may also include actuators for controlling the behavior of the vehicle 101, such as an actuator for operating a steering wheel, an actuator for operating a brake pedal, or an actuator for operating an accelerator pedal. The actuators 30 may also include devices for operating accessories of the vehicle 101, such as a display, a speaker, an indicator, and a headlight. The driving and operation of the vehicle 101 are controlled by the actuators 30.

[0020] The vehicle control device 10 is primarily composed of a microcomputer including a processing unit 10a such as a CPU and a storage unit 10b such as various types of memory. The functions provided by the microcomputer can be provided by software recorded in a physical memory device and a computer executing the software, software alone, hardware alone, or a combination thereof. For example, when the microcomputer is provided by electronic circuits, which are hardware, the functions can be provided by digital circuits including numerous logic circuits or analog circuits. For example, the processing unit 10a of the microcomputer executes programs stored in a non-transitory tangible storage medium (non-transitory tangible storage medium) that serves as the storage unit 10b. The programs include, for example, programs that realize functions shown in FIG. 2 . Execution of the programs results in the execution of methods corresponding to the programs. The storage unit 10b is, for example, a non-volatile memory. The programs stored in the storage unit 10b can be downloaded and updated via a communication network 103, such as the Internet, via over-the-air (OTA) or other means.

[0021] The vehicle control device 10 has various functions (application programs) for assisting the driving of the vehicle 101, and these functions control the actuator 30 based on sensor information input from the sensor 20.

[0022] Typical functions for assisting the driving of the vehicle 101 include, for example, an adaptive cruise control system (ACC), a forward collision warning (FCW), an advanced emergency braking system (AEBS), a night vision / pedestrian detection (NV / PD), a traffic sign recognition (TSR), a lane departure warning (LDW), a lane keeping assist system (LKAS), a rear cross traffic alert (RCTA), an adaptive front lighting system (AFS), and an advanced parking assist (APA). The vehicle 101 may be provided with all or some of these functions. Furthermore, the vehicle 101 may also be provided with other driving assistance functions. Furthermore, the vehicle 101 may also be provided with a function for realizing autonomous driving of the vehicle 101. These functions are realized by the arithmetic processing unit 10a executing a driving assistance control program stored in the storage unit 10b.

[0023] Furthermore, in order to verify the performance and safety of the driving assistance control program, the vehicle control device 10 has a function called a shadow mode in which a driving assistance control program to be verified (hereinafter referred to as a verification target program) is run while the vehicle is running after sales, that is, in an actual use case, and output values ​​are checked. In verification using the shadow mode, for example, the verification target program is run in the background, that is, in a manner not related to vehicle control, and data related to the output values ​​is stored, and the server 102 collects the data related to the output values ​​via the communication network 103. Note that the shadow mode is a type of data collection mode for collecting data.

[0024] The configuration and functions of the shadow mode according to this embodiment will be described below. The timing of switching to the shadow mode may be any timing. For example, the switching may be performed by an operation by the driver or by an instruction from the server 102. Alternatively, the switching to the shadow mode may be performed when the ignition switch is turned on.

[0025] In the following, the driving assistance control program that actually operates the vehicle 101 will be referred to as an "implemented program" to contrast it with the program to be verified. It will also be referred to as an "implemented program" in FIG. 1 . The vehicle control device 10 is equipped with a verification processing device 11 for executing the program to be verified, taking into account the processing load when executing the implemented program. The verification processing device 11 may have performance equivalent to or different from that of the processing device 10a, as long as it has the performance to execute the program to be verified. The program to be verified is stored in the storage unit 10b. A storage device dedicated to the shadow mode may also be provided.

[0026] Execution of the program to be verified executes various functions in the background to assist the driving of the vehicle 101. The functions implemented by execution of the program to be verified may be the same as the functions actually realized by the arithmetic processing device 10a, or some of the functions may be omitted. Furthermore, functions other than the functions actually realized by the arithmetic processing device 10a may also be implemented.

[0027] When the verification target program is running in shadow mode, the verification processing unit 11, like the processing unit 10a, receives sensor information from the various sensors 20. Based on the received sensor information, the verification processing unit 11 then performs various functions and outputs various control signals for operating the actuators. These control signals (output results) are not actually input to the actuators 30, but are stored in the storage unit 10b as data related to the output results. At this time, the verification processing unit 11 also stores data related to the sensor information, which is an input value, in association with the data.

[0028] The verification processing unit 11 also receives the control signal output from the processing unit 10a after inputting sensor information as an input value and processing it based on the implementation program, and stores the control signal in the storage unit 10b as data related to the output result. That is, the verification processing unit 11 also stores the control signal output from the processing unit 10a after processing it based on the implementation program in order to compare and verify the output result of the implementation program with the output result of the program to be verified.

[0029] Furthermore, the verification processing device 11 may input sensor information (vehicle speed, yaw rate, acceleration amount, accelerator pedal operation amount, brake pedal operation amount, steering angle, etc.) detected when a control signal processed and output based on the implementation program is input to the actuator 30 and the vehicle 101 moves as a result, and store the information in the storage unit 10b. In other words, the sensor information related to the actual operation of the vehicle 101 for each scene, that is, the sensor information necessary to verify the correct operation, may also be stored.

[0030] The vehicle control device 10 uploads the data thus stored in the storage unit 10b to the server 102 via the communication network 103 at a predetermined transmission timing. The predetermined transmission timing may be any timing, for example, a timing when an upload instruction is issued from the server 102. Alternatively, the predetermined transmission timing may be, for example, a timing when the vehicle 101 is charging, when the vehicle is parked or stopped, or when the ignition switch is turned off.

[0031] However, the storage capacity of the storage unit 10b is insufficient to store all data while the vehicle 101 is traveling. For this reason, data is acquired when a predetermined acquisition condition (also referred to as a trigger condition) is met. Several acquisition conditions are set for each scene (hereinafter simply referred to as a scene) in which the vehicle 101 is traveling, allowing data corresponding to the scene to be acquired. However, if the acquisition conditions are not set appropriately, the acquisition conditions may not be met, and data for a particular scene may not be collected as expected. Therefore, the vehicle control device 10 of this embodiment is configured as follows so that appropriate acquisition conditions can be set. This will be explained in detail below.

[0032] 2, the verification processing device 11 of the vehicle control device 10 has a function as an input unit 12, a function as a processing unit 13, a function as a scene estimation unit 14, a function as a determination unit 15, a function as a data acquisition unit 16, a function as a transmission unit 17, and a function as a change unit 18. These functions are realized by the verification processing device 11 by executing an in-vehicle program stored in the storage unit 10b.

[0033] The input unit 12 inputs sensor information from the sensor 20. Then, the input unit 12 inputs part or all of the input sensor information to the processing unit 13 as input values ​​for the program to be verified. The input unit 12 also inputs part or all of the input sensor information to the scene estimation unit 14 and the data acquisition unit 16.

[0034] The processing unit 13 runs the program to be verified and performs processing based on various functions (application programs) realized by the program to be verified, based on input values ​​(sensor information) input from the input unit 12. The processing unit 13 then inputs control signals (control signals for the actuator 30) as the processing results to the data acquisition unit 16. Note that, as described above, these control signals are not input to the actuator 30.

[0035] As shown in Figure 3, the scene estimation unit 14 inputs the sensor information received from the input unit 12 into an inference model and outputs a scene reliability (score value) for each specific scene indicating how similar the scene in which the vehicle 101 is traveling is to each of a number of predetermined specific scenes, thereby estimating the scene in which the vehicle 101 is traveling.

[0036] The scenes refer to various scenes that are expected when the vehicle 101 is traveling, such as a scene of overtaking a vehicle ahead, a scene of slipping through vehicles, a scene of following a vehicle ahead, a scene of a pedestrian crossing in front of the vehicle 101 at night, a scene of the vehicle 101 merging from an acceleration lane onto a main lane on a motorway, a scene of parking and stopping, a scene of waiting at a traffic light, etc. A specific scene refers to a scene selected from these various scenes by the data acquirer (developer).

[0037] The scene reliability indicating the degree of similarity of the scene in which the vehicle 101 is traveling to each specific scene is output (estimated) based on sensor information. Specifically, sensor information such as camera images is input to a machine-learned inference model such as a deep neural network, and the scene reliability indicating the degree of similarity of each specific scene is output for each specific scene, thereby estimating which specific scene the scene is most similar to.

[0038] The inference model of this embodiment extracts features from input camera images and compares them with the features of each specific scene to output a scene reliability for each specific scene. Specific scenes are typical scenes that are expected when the vehicle 101 travels, such as a scene where the vehicle overtakes a vehicle ahead or a scene where the vehicle passes between vehicles, and are scenes from which data collection is desired. These specific scenes are expected in advance when machine learning is performed and are included in the training data.

[0039] The determination unit 15 determines whether the data acquisition condition is met. As shown in FIG. 4 , in this embodiment, a plurality of acquisition conditions are provided for each scene (specific scene) from which data is acquired, from a first acquisition condition to an Xth acquisition condition (X is an arbitrary number). A reliability threshold value for the scene reliability is provided for each acquisition condition, i.e., for each specific scene. For example, if the scene reliability of a scene in which a vehicle overtakes a preceding vehicle is "80," it is determined that the first acquisition condition is met.

[0040] The reliability threshold is set for each specific scene (i.e., for each acquisition condition). The reliability threshold may be the same value regardless of each specific scene, or a different value may be set for each specific scene.

[0041] When the determination unit 15 determines that the acquisition condition is met, the data acquisition unit 16 stores data regarding the input values ​​and output results in the storage unit 10b. As described above, the input values ​​include, for example, sensor information input to the verification processing device 11. The output results include, for example, control signals processed and output based on the program to be verified, and control signals processed and output based on the implemented program. The output results may also include sensor information related to the actual operation of the vehicle 101 for each scene. The output results may also include the difference between the output results based on the implemented program and the output results of the program to be verified.

[0042] The data to be stored (acquired) may be changed depending on the acquisition condition that is met. For example, when an acquisition condition (first acquisition condition) for a scene in which the vehicle is overtaking a preceding vehicle is met, data related to the camera image from the front camera may be acquired, whereas when an acquisition condition (second acquisition condition) for a scene in which the vehicle is passing between vehicles is met, data related to the camera image from the side camera may be acquired in addition to data related to the camera image from the front camera.

[0043] The transmitter 17 transmits the data acquired by the data acquisition unit 16 and stored in the storage unit 10b to the server 102 via the communication network 103 at a predetermined transmission timing. The predetermined transmission timing is as described above.

[0044] The change unit 18 changes the reliability threshold based on the quantity of data acquired by the data acquisition unit 16. Specifically, the change unit 18 stores the number of times data has been acquired for each acquisition condition, that is, for each specific scene, i.e., the number of times each acquisition condition has been met, and if there is an acquisition condition for which the number of times it has been met is zero (i.e., not met) or is equal to or less than a specified lower limit even after a predetermined period has elapsed since the start of the shadow mode, the change unit 18 changes the reliability threshold for that acquisition condition. In this case, the change unit 19 lowers the reliability threshold for the unmet acquisition condition so as to relax the acquisition condition.

[0045] On the other hand, if the number of times that the acquisition condition is met is equal to or greater than the specified upper limit, the change unit 18 strengthens the acquisition condition by increasing the reliability threshold for the acquisition condition that has been met the specified upper limit. How to change the reliability threshold may be stored in advance in the storage unit 10b. Alternatively, the information may be downloaded from an external device such as the server 102 via the communication network 103.

[0046] The flow of data acquisition in this embodiment will be described below with reference to FIG. 5. The flow of data acquisition shown below is the flow of data acquisition processing performed by the verification processing unit 11. These processes are performed after the shadow mode is set. The timing at which the shadow mode is set is arbitrary, but may be, for example, the timing at which the ignition switch is turned on.

[0047] First, the input unit 12 of the verification processing device 11 inputs sensor information from the sensor 20 (step S101). Step S101 corresponds to an input step. Next, the processing unit 13 runs the program to be verified and performs processing based on various functions realized by the program to be verified, based on the input values ​​(sensor information) input from the input unit 12 (step S102). In step S102, the processing unit 13 inputs a control signal as a result of the processing to the data acquisition unit 16.

[0048] The determination unit 15 inputs the sensor information received from the input unit 12 into an inference model to output a scene reliability, and determines whether the acquisition conditions are met based on the scene reliability (step S103). Step S103 corresponds to a scene estimation step and a determination step. The acquisition conditions in step S103 are the acquisition conditions initially set when the shadow mode starts or the acquisition conditions changed by the change unit 18.

[0049] If the determination result is positive (if the acquisition condition is met), the data acquisition unit 16 stores (acquires) data regarding the input values ​​and output results in the storage unit 10b (step S104). Step S104 corresponds to the data acquisition step. When storing the data, the data acquisition unit 16 may also store the met acquisition condition in association with the data. The data acquisition unit 16 also adds 1 to the number of times the met acquisition condition has been met. The data acquisition process then ends. On the other hand, if the determination result is negative, the verification processing unit 11 simply ends the data acquisition process. The data acquisition process is executed at regular intervals after the shadow mode is set.

[0050] The transmitting unit 17 transmits the data acquired by the data acquiring unit 16 and stored in the storage unit 10b to the server 102 via the communication network 103 at a predetermined transmission timing. At this time, the transmitting unit 17 may also transmit information related to the satisfied acquisition condition associated with the data (such as the number of times the condition was satisfied). In this way, the transmitting unit 17 performs the transmission step.

[0051] When the server 102 receives the data transmitted from the transmitter 17 of each vehicle 101, the server 102 stores the data in the storage device of the server 102. Note that information regarding the acquisition conditions that were met when the data was acquired may also be stored in association with the data. This makes it easier to identify the acquired scene.

[0052] According to the first embodiment, the following effects are achieved.

[0053] The scene estimation unit 14 inputs the sensor information into the inference model and outputs a scene reliability (score value) indicating the degree of similarity of the scene in which the vehicle 101 is traveling to each of a plurality of predetermined specific scenes for each specific scene. If the determination unit 15 determines that any of the output scene reliability values ​​is equal to or greater than a predetermined reliability threshold, the data acquisition unit 16 determines that the acquisition condition is met and acquires data related to the sensor information, etc. The inference model is machine-learned in advance using training data containing information related to a plurality of specific scenes.

[0054] This allows the use of an inference model to more appropriately estimate a scene than when a scene of the vehicle 101 traveling is estimated based on rules. Furthermore, it eliminates the need to create rules, reducing the burden on design.

[0055] In addition, the scene reliability for each specific scene is output and compared with the reliability threshold for each specific scene. Therefore, for scenes where the scene reliability tends to be low, the reliability threshold can be set low, and other conditions can be flexibly changed. Conversely, for scenes where the scene reliability tends to be high, the reliability threshold can be set high, and other conditions can be flexibly changed. This makes it possible to prevent problems such as no data being collected or too much data being collected.

[0056] The change unit 18 changes the reliability threshold depending on the number of times the acquisition condition is met, that is, depending on the amount of data for each specific scene. Specifically, if the number of times the condition is met is small, the change unit 18 lowers the reliability threshold to make it easier to acquire data, while if the number of times the condition is met is large, the change unit 18 raises the reliability threshold to make it more difficult to acquire data. This makes it possible to prevent problems such as no data being collected or too much data being collected due to individual circumstances of the vehicle 101.

[0057] Furthermore, since a shadow mode can be set, in an actual use case, the program to be verified can be run in the background to collect data on the results of the program's operation.

[0058] (Variations of the First Embodiment) In the first embodiment, it is not necessary to acquire the control signal (data related to the output result) that is processed and output based on the implementation program. On the other hand, after the acquisition condition is met, sensor information (vehicle speed, yaw rate, acceleration amount, accelerator pedal operation amount, brake pedal operation amount, steering angle, etc.) detected when the vehicle 101 is operating may be acquired until a predetermined time has elapsed.

[0059] In the first embodiment, the driving assistance function does not need to be implemented. In this case, sensor information (vehicle speed, yaw rate, acceleration amount, accelerator pedal operation amount, brake pedal operation amount, steering angle, etc.) detected when the vehicle 101 is operating due to an operation by the driver may be acquired until a predetermined time has elapsed after the acquisition condition is met.

[0060] In the first embodiment, the verification processing unit 11 is provided. However, if the processing unit 10a can run the program to be verified together with the implementation program, the verification processing unit 11 does not have to be provided.

[0061] In the above embodiment, the determination unit 15 determines that the acquisition condition is met if any of the scene reliabilities output by the scene estimation unit 14 is equal to or greater than the reliability threshold. As a variation of this, the determination unit 15 may determine whether the acquisition condition is met by determining whether the most reliable scene reliability among the scene reliabilities output from the inference model is equal to or greater than the reliability threshold. This reduces the processing load.

[0062] In the above embodiment, the acquisition condition is changed depending on the number of times the acquisition condition is met. As a variation of this, the acquisition condition may be changed depending on whether the amount of data for each acquisition condition, i.e., for each specific scene, is greater or less than a specified amount.

[0063] Second Embodiment A second embodiment will be described in which the configuration of the vehicle control device 10 in the first embodiment is partially modified.

[0064] The acquisition condition in the first embodiment was whether the scene reliability was equal to or greater than a reliability threshold. In the second embodiment, a rule-based condition is added to the condition based on the scene reliability.

[0065] The determination unit 15 of the second embodiment will be described. The determination unit 15 of the second embodiment determines that the acquisition condition is met when any of the scene reliabilities estimated (output) by the scene estimation unit 14 is equal to or greater than a predetermined reliability threshold and when parameters related to the traveling conditions of the vehicle 101 satisfy predetermined conditions (rules).

[0066] For example, the determination unit 15 of the second embodiment sets an acquisition condition that the scene reliability is equal to or greater than the reliability threshold, and that one or more parameters included in the sensor information or the like are equal to or less than the threshold (or equal to or greater than the threshold), as in the first embodiment. Note that, hereinafter, a parameter related to the traveling state of the vehicle 101 that is referenced in the acquisition condition may be simply referred to as a reference parameter.

[0067] The reference parameters may be, for example, parameters included in sensor information such as vehicle speed, yaw rate, accelerator operation amount, brake operation amount, or steering amount, or parameters included in output results such as required torque or an output instruction signal to a display. Parameters calculated or estimated from sensor information or output results may also be referenced. For example, the relative distance to an obstacle (such as a preceding vehicle or pedestrian), the number of obstacles, the relative speed to the obstacle, and the time-to-collision (TTC) calculated from the recognition results of a camera image or the detection results of the millimeter-wave radar 25 may be set as reference parameters. The difference between the output results based on the implemented program and the output results of the program to be verified may also be set as a reference parameter. For example, the difference between the brake operation amount output by processing based on the implemented program and the brake operation amount output by processing based on the program to be verified may be set as a reference parameter. Note that the number of reference parameters may be one or more.

[0068] The threshold value is arbitrarily set for each reference parameter. Two threshold values, an upper limit value and a lower limit value, may be set to determine whether or not a reference parameter is within a specified range (or outside the specified range). For example, to determine whether or not the speed is within a range of 30 to 50 km / h, the threshold values ​​may be set so that the lower limit value is 30 km / h and the upper limit value is 50 km / h.

[0069] The configuration of the modification unit 18 in the second embodiment may be modified in accordance with the modification of the determination unit 15. As shown in Fig. 6, several modification methods will be described assuming that the first acquisition condition before modification is that the reliability threshold of the scene reliability of the "scene of overtaking a preceding vehicle" is "80" or higher and the vehicle speed range is "50 km / h to 70 km / h."

[0070] In the first modification method, the reliability threshold and the threshold of the reference parameter are modified. For example, when relaxing the conditions in the first modification method as shown in FIG. 6 , the modification unit 18 modifies the reliability threshold of the scene reliability to "75" or higher and the speed range of the vehicle speed to "40 km / h to 70 km / h." In addition, when strengthening the conditions, the modification unit 18 modifies the reliability threshold of the scene reliability to "85" or higher and the speed range of the vehicle speed to "60 km / h to 70 km / h."

[0071] In the second modification method, the reliability threshold of the scene reliability is not modified, and only the threshold of the reference parameter is modified. For example, as shown in FIG. 6 , when the conditions are relaxed in the second modification method, the modification unit 18 modifies the vehicle speed to "50 km / h or more" (with no upper limit) while keeping the reliability threshold of the scene reliability at "80" or more. In other words, only the vehicle speed threshold is modified. To strengthen the conditions, the vehicle speed threshold can be modified in the same way.

[0072] In the third modification method, a rule is deleted or added to modify the scene. For example, when relaxing the conditions in the third modification method as shown in Fig. 6, the modification unit 18 deletes the condition related to the vehicle speed (threshold value of the reference parameter) and sets the reliability threshold value of the scene reliability to "80" or more.

[0073] In addition, when the conditions are strengthened in the third change direction, thresholds of different types of reference parameters such as acceleration, inter-vehicle distance, etc. may be added. For example, while the reliability threshold of the scene reliability is set to "80" or more and the vehicle speed range is set to "50 km / h to 70 km / h," a condition that the inter-vehicle distance be "within 20 m" may be added.

[0074] Furthermore, any of the first to third change methods may be adopted, and the change method may be changed for each acquisition condition. For example, the first acquisition condition may be changed using the first change method, and the second acquisition condition may be changed using the third change method. Furthermore, the change method may be changed depending on the number of changes. For example, the third change method may be used for the first change, and the first change method may be used for the second and subsequent changes.

[0075] According to the second embodiment, the following effects are achieved.

[0076] The determination unit 15 determines whether the scene reliability is equal to or greater than the reliability threshold, and also determines whether the parameters related to the driving situation of the vehicle 101 satisfy predetermined conditions. Specifically, the determination unit 15 determines whether the reference parameters are equal to or greater than the threshold (or equal to or less than the threshold), in addition to whether the scene reliability is equal to or greater than the reliability threshold. This makes it possible to acquire data after determining the driving situation of the vehicle 101, which is difficult to determine from camera images. For example, although it is difficult to determine vehicle speed from camera images, adding vehicle speed as a reference parameter to the conditions makes it possible to suitably acquire information about a scene that takes vehicle speed into consideration.

[0077] Furthermore, by specifying the conditions for the reference parameters as rules, i.e., by specifying the threshold values ​​for the reference parameters, it becomes possible to specify detailed driving conditions, which makes it easier to analyze data in more detail.

[0078] Furthermore, when the change unit 18 changes the acquisition conditions, it is possible to gradually relax or strengthen the acquisition conditions by changing the threshold values ​​of the reference parameters.

[0079] - In the above embodiment, in order to prevent overlearning, it is desirable that the inference model be machine-learned in advance using training data that includes information about the predetermined plurality of specific scenes and information about scenes unrelated to the specific scenes.

[0080] (Variation of the Second Embodiment) The reference parameter in the second embodiment may be information (such as an implementation flag) indicating an implemented function (implemented function) among various functions (application programs) based on the program to be verified. For example, if the scene reliability of a scene in which the vehicle is following a preceding vehicle is equal to or greater than the reliability threshold value of 80 and an implementation flag indicating that the collision damage mitigation braking control function has been implemented is set, it may be determined that the acquisition condition is met.

[0081] Which function has been implemented can be determined based on the output result (control signal) input from the processing unit 13. Furthermore, two or more types of implemented functions may be included as components included in one acquisition condition, and in this case, the implemented functions may be AND conditions or OR conditions. For example, the condition may be that both the forward vehicle approach warning function and the collision damage mitigation braking control function have been implemented, or that either one of them has been implemented.

[0082] Third Embodiment A third embodiment will be described in which the data collection system 100 in the first embodiment is partially modified.

[0083] In the third embodiment, the server 202 is configured to change the reliability threshold. This will be explained in detail below. As shown in Fig. 7, the server 202 has a function as a collection unit 216, a function as a setting change unit 218, and a function as an instruction unit 217. These functions are realized by an arithmetic processing unit 202a, such as a CPU, of the server 202 executing a server program stored in a storage device 202b, such as a storage device.

[0084] The collection unit 216 collects data received from the transmitters 17 of the multiple vehicles 101 connected via the communication network 103 and stores the data in the storage device 202b of the server 202. At this time, the data is classified and stored by the acquisition condition that was met, i.e., by specific scene. This makes it possible to know the number of times each acquisition condition was met by counting the number of data.

[0085] The setting change unit 218 changes the reliability threshold based on the amount of data collected by the collection unit 216 from the multiple vehicles 101. Specifically, as described above, the setting change unit 218 grasps the number of times each acquisition condition is met. Then, if there is an acquisition condition whose number of times of fulfillment is zero (i.e., not met) or is equal to or less than a specified lower limit even after a predetermined period has elapsed since the start of data collection, the setting change unit 218 changes the reliability threshold for that acquisition condition. In this case, the reliability threshold is changed so as to relax that acquisition condition.

[0086] On the other hand, if an acquisition condition exists that has been met a specified upper limit of the number of times after a predetermined period has elapsed since the start of data collection, the setting change unit 218 changes the reliability threshold so as to strengthen the acquisition condition. Note that the relaxation method and the strengthening method are the same as those in the above embodiment, and therefore detailed explanations will be omitted.

[0087] When the setting change unit 218 changes the acquisition conditions, the instruction unit 217 transmits (downloads) download data including the changes to each vehicle 101. The change unit 18 of the vehicle control device 10 changes the acquisition conditions in accordance with the changes included in the download data, that is, in accordance with the instructions of the server 202.

[0088] According to the third embodiment, the following effects are achieved.

[0089] The server 202 is provided with the setting change unit 218, which can reduce the control burden on the vehicle control device 10 of each vehicle 101. Furthermore, the setting change unit 218 in the server 202 keeps track of the number of times each acquisition condition is met based on data collected from multiple vehicles 101. Therefore, the acquisition conditions are changed after understanding the trends in all vehicles 101 connected to the communication network 103, so the amount of data collected by the server 202 can be controlled. This makes it possible to perform control taking into account the communication load and the analysis load on the server 202 side.

[0090] (Variation of the Third Embodiment) In the third embodiment, the number of times that a condition is met is determined from the number of pieces of data collected by the collection unit 216. However, the number of times that each acquisition condition is met may be received from each vehicle 101 and tallied. Furthermore, the acquisition conditions may be changed depending on whether the amount of data for each acquisition condition, that is, for each specific scene, is greater or less than a specified amount.

[0091] Fourth Embodiment A fourth embodiment will be described, in which the data collection system 100 in the first embodiment is partially modified.

[0092] In the fourth embodiment, the inference model used by the scene estimation unit 14 is downloaded and updated at a predetermined timing from the server 302 of the fourth embodiment. This updated inference model is a new inference model generated by re-learning based on data transmitted from multiple vehicles 101 connected to the communication network 103 (data collected by the server 102).

[0093] 8, the server 302 has a function as a collection unit 316, a function as an instruction unit 317, and a function as a learning unit 318. These functions are realized by an arithmetic processing unit 302a such as a CPU of the server 302 executing a server program stored in a storage device 302b such as a storage.

[0094] The collection unit 316 collects data received from the transmission units 17 of the multiple vehicles 101 connected via the communication network 103, and stores the data in the storage device of the server 302. At this time, the data is classified and stored according to the acquisition conditions that have been met, i.e., according to the specific scenes.

[0095] Then, true / false annotation information is assigned to image data included in the collected data classified by specific scene. That is, a true / false determination is made as to whether the image data actually contains the specific scene, i.e., whether the classification is correct or not. The true / false determination may be performed manually by looking at the image data. Alternatively, a true / false determination may be performed based on the image data using a highly accurate inference model generated by machine learning using a larger amount of training data using a processing device with higher performance than the processing device 10a or the like. In this case, the processing device 302a of the server 302 may perform the determination.

[0096] The learning unit 318 then performs re-learning using the image data to which annotation information has been added as new training data to generate a new inference model. Note that in this embodiment, the learning unit 318 of the arithmetic processing device 202a of the server 302 is caused to perform the learning, but the re-learning may also be performed by an arithmetic processing device provided outside the server 302.

[0097] When a new inference model is generated, the instruction unit 317 transmits (downloads) download data including the inference model to each vehicle 101.

[0098] The scene estimation unit 14 of the vehicle control device 10 downloads the inference model included in the download data, that is, updates the inference model in accordance with instructions from the server 202.

[0099] According to the fourth embodiment, the following effects are achieved.

[0100] The inference model is updated by re-learning, allowing for more accurate scene estimation and for efficient and accurate collection of scene-specific data. Furthermore, the inference model is re-learned using data collected by the server 302. Therefore, image data from actual use cases is used as training data, further improving accuracy.

[0101] Fifth Embodiment A fifth embodiment will be described, in which the vehicle control device 10 in the first embodiment is partially modified.

[0102] The vehicle control device 10 in the fifth embodiment is configured to acquire and collect only data related to sensor information and data related to output results based on the installed program, without executing the program to be verified. In other words, in the fifth embodiment, a data collection mode is set in which the program to be verified is not executed.

[0103] The configuration and functions relating to the data collection mode according to the fifth embodiment will be described below. In the fifth embodiment, the verification processing unit 11 is not provided, and the processing unit 10a performs the processing relating to the data collection mode. As in the first embodiment, the verification processing unit 11 may be provided and may perform the processing in place of the processing unit 10a. Furthermore, the verification target program is not stored.

[0104] 9, in the fifth embodiment, the arithmetic processing device 10a of the vehicle control device 10 has a function as an input unit 112, a function as a processing unit 113, a function as a scene estimation unit 114, a function as a determination unit 115, a function as a data acquisition unit 116, a function as a transmission unit 117, and a function as a change unit 118. These functions are realized by the arithmetic processing device 10a as the arithmetic processing device 10a executes an in-vehicle program stored in the storage unit 10b.

[0105] The input unit 112 inputs sensor information from the sensor 20. The input unit 112 inputs part or all of the input sensor information to the processing unit 113 as input values ​​for the implementation program. The input unit 112 also inputs part or all of the input sensor information to the scene estimation unit 114 and the data acquisition unit 116.

[0106] The processing unit 113 runs the implementation program and performs processing based on various functions (application programs) realized by the implementation program, based on input values ​​(sensor information) input from the input unit 112. The processing unit 113 then inputs control signals (control signals for the actuator 30) as the processing results to the data acquisition unit 116. These control signals are also input to the actuator 30.

[0107] The scene estimation unit 114, the determination unit 115, the transmission unit 117, and the modification unit 118 have the same functions as the scene estimation unit 14, the determination unit 15, the transmission unit 17, and the modification unit 18 in the first embodiment, respectively, and therefore detailed explanations are omitted.

[0108] When the acquisition condition is satisfied, the data acquisition unit 116 stores data relating to the input values ​​and output results in the storage unit 10b. As described above, the input values ​​include, for example, sensor information input to the arithmetic processing device 10a. The output results include, for example, control signals processed and output based on the implementation program.

[0109] Furthermore, the data acquisition unit 116 of the arithmetic processing device 10a may input sensor information (vehicle speed, yaw rate, acceleration amount, accelerator pedal operation amount, brake pedal operation amount, steering angle, etc.) detected when a control signal processed and output based on the implementation program is input to the actuator 30 and the vehicle 101 operates, and store the information in the storage unit 10b. In other words, the sensor information related to the actual operation of the vehicle 101 for each scene, that is, the sensor information necessary to verify the correct operation, may also be stored.

[0110] The flow of data acquisition in this embodiment will be described below with reference to FIG. 10. The flow of data acquisition shown below is the flow of data acquisition processing performed by the arithmetic processing device 10a. These processes are performed after the data collection mode is set. The timing at which the data collection mode is set is the same as the timing at which the shadow mode is set in the first embodiment.

[0111] First, the input unit 112 of the arithmetic processing device 10a inputs sensor information from the sensor 20 (step S201). Next, the processing unit 113 runs an implementation program and performs processing based on various functions realized by the implementation program, based on the input values ​​(sensor information) input from the input unit 112 (step S202). The processing unit 113 inputs control signals as the processing results to the actuator 30 and the data acquisition unit 116. The explanations for steps S203 and S204 are substantially the same as those for steps S103 and S104 in the first embodiment, and therefore detailed explanations will be omitted.

[0112] According to the fifth embodiment, the following effects are achieved.

[0113] Since the program to be verified is not executed, the processing load can be reduced. Accordingly, the verification processing unit 11 does not need to be provided, and the configuration can be simplified.

[0114] (Variation of Fifth Embodiment) In the fifth embodiment, the data acquisition unit 116 acquires data related to the output results (control signals) based on the operation of the implementation program. However, it is not necessary to acquire data related to the output results based on the operation of the implementation program. In this case, only sensor information that serves as an input value may be acquired. Furthermore, a control signal processed and output based on the implementation program may be input to the actuator 30, and sensor information (vehicle speed, yaw rate, acceleration amount, accelerator pedal operation amount, brake pedal operation amount, steering angle, etc.) detected when the vehicle 101 operates may be acquired.

[0115] In the fifth embodiment, the implementation program is executed to perform various driving assistance functions during the data collection mode, but the various driving assistance functions do not have to be executed during the data collection mode. In other words, during the data collection mode, sensor information (vehicle speed, yaw rate, acceleration amount, accelerator pedal operation amount, brake pedal operation amount, steering angle, etc.) detected when the vehicle 101 is operating may be acquired based on operations by the driver.

[0116] More specifically, the input unit 112 receives sensor information from the sensor 20 and inputs part or all of the received sensor information to the scene estimation unit 114 and the data acquisition unit 116 .

[0117] When the acquisition condition is satisfied, the data acquisition unit 116 stores data related to the input values ​​in the storage unit 10b. After the acquisition condition is satisfied, the data acquisition unit 116 of the arithmetic processing device 10a inputs sensor information detected when the vehicle 101 is operating based on an operation by the driver via the input unit 112 or the like, and stores data related to the sensor information in the storage unit 10b. The sensor information detected when the vehicle 101 is operating includes vehicle speed, yaw rate, acceleration amount, accelerator pedal operation amount, brake pedal operation amount, steering angle, etc. After the acquisition condition is satisfied, the data acquisition unit 116 inputs and stores the sensor information until a predetermined time has elapsed. The predetermined time may be changed depending on the type of acquisition condition that has been satisfied. The scene estimation unit 114, the determination unit 115, the change unit 118, and the transmission unit 117 are the same as those in the fourth embodiment, and therefore, description thereof will be omitted.

[0118] The flow of data acquisition in this modified example will be described below with reference to Fig. 11. The flow of data acquisition shown below is the flow of data acquisition processing performed by the arithmetic processing device 10a. This processing is performed after the data collection mode is set.

[0119] The input unit 112 of the arithmetic processing device 10a inputs sensor information from the sensor 20 (step S301). The determination unit 115 determines whether or not an acquisition condition is met (step S302).

[0120] If the determination result is positive (if the acquisition condition is met), the data acquisition unit 116 stores data related to the input value in the storage unit 10b (step S303). The data acquisition unit 116 also inputs sensor information detected when the vehicle 101 is operating until a predetermined period of time has elapsed, and stores the data related to the sensor information in the storage unit 10b. When storing the data, the data acquisition unit 116 also stores the data in association with the met acquisition condition. Then, the processing ends. On the other hand, if the determination result is negative, the arithmetic processing device 10a ends the processing. The following description is omitted as it is the same as in the fifth embodiment.

[0121] The controller and methods described herein may be implemented by a special-purpose computer configured with a processor and memory programmed to perform one or more functions embodied in a computer program. Alternatively, the controller and methods described herein may be implemented by a special-purpose computer configured with a processor configured with one or more dedicated hardware logic circuits. Alternatively, the controller and methods described herein may be implemented by one or more special-purpose computers configured with a processor and memory programmed to perform one or more functions in combination with a processor configured with one or more hardware logic circuits. Furthermore, the computer program may be stored as instructions executed by a computer on a computer-readable non-transitory storage medium.

[0122] (Modifications of the Above-described Embodiments) In the above-described embodiments, the modification unit 18, 118 may modify the reliability threshold for a specific scene according to the occurrence probability of the specific scene. The occurrence probability of a specific scene is determined, for example, by the ratio of the number of times the specific scene occurs per predetermined driving time (or imaging time). The occurrence probability of a specific scene may be an actual measurement value or may be determined by experiment or simulation. In this case, it is desirable to set the reliability threshold for a specific scene to a smaller value when the occurrence probability of the specific scene is low compared to when the occurrence probability is high.

[0123] In each of the above embodiments, the change unit 18, 118 may change the reliability threshold for a specific scene depending on the amount of data related to the specific scene transmitted by the transmission unit 17, 117 to the server 102 or the amount of data related to the specific scene stored in the server 102. The amount of data refers to the number of times or amount of data transmitted or stored. The amount of data related to the specific scene stored in the server 102 may be received from the server 102. In this case, when the amount of data transmitted by the transmission unit 17, 117 or the amount of data stored in the server 102 is small, it is desirable to set the reliability threshold for the specific scene to a lower value than when the amount of data is large, thereby making it easier to acquire the data. When the amount of data transmitted by the transmission unit 17, 117 or the amount of data stored in the server 102 increases, it is desirable to set the reliability threshold for the specific scene to a higher value, thereby making it more difficult to acquire the data.

[0124] In each of the above embodiments, as shown in FIG. 12 , a driving skill determination unit 801 may be provided that determines the driving skill of the driver of the vehicle 101. The driving skill determination unit 801 may make a determination based on the frequency of sudden steering, sudden braking, sudden acceleration, and the like, based on sensor information. The driving skill determination unit 801 may also make a determination based on the frequency with which various functions for assisting the driving of the vehicle 101 are implemented. For example, if the number of warnings issued by the forward vehicle approach warning function or the lane departure warning function is high, the driving skill may be determined to be poorer than if the number of warnings issued is low. For example, if the vehicle 101 is controlled frequently by the lane departure prevention support function, the driving skill may be determined to be poorer than if the number of warnings issued is low. In this modified example, the driving skill determination unit 801 determines that the driver has a predetermined driving skill if the frequency of these warnings is equal to or less than a predetermined value.

[0125] Then, when any of the scene reliabilities output by the scene estimation unit 14, 114 is equal to or greater than a predetermined reliability threshold, the determination unit 15, 115 may determine whether the acquisition condition is met in accordance with the determination result of the driving skill determination unit 801. For example, the determination unit 15, 115 may determine that the acquisition condition is met when any of the scene reliabilities output by the scene estimation unit 14, 114 is equal to or greater than a predetermined reliability threshold and the driving skill determination unit 801 determines that the driver has the predetermined driving skill. Conversely, the determination unit 15, 115 may determine that the acquisition condition is met when any of the scene reliabilities output by the scene estimation unit 14, 114 is equal to or greater than a predetermined reliability threshold and the driving skill determination unit 801 determines that the driver does not have the predetermined driving skill.

[0126] The following describes technical ideas that can be derived from the above-described embodiment and modifications.

[0127] [Configuration 1] A vehicle control device (10) mounted on a vehicle (101), comprising: a scene estimation unit (14, 114) that inputs sensor information acquired by a sensor (20) mounted on the vehicle into an inference model and outputs, for each specific scene, a scene reliability indicating how similar a scene in which the vehicle is traveling is to each of a plurality of predetermined specific scenes; a determination unit (15, 115) that determines that an acquisition condition is met when any of the scene reliability output by the scene estimation unit is equal to or greater than a predetermined reliability threshold; a data acquisition unit (16, 116) that acquires data related to the sensor information when the determination unit determines that the acquisition condition is met; and a transmission unit (17, 117) that transmits the data acquired by the data acquisition unit to a server (102) via a communication network (103), wherein the inference model is machine-learned in advance using training data including information about the plurality of specific scenes.

[0128] [Configuration 2] The vehicle control device according to Configuration 1, wherein the reliability threshold value for each of the specific scenes is determined, thereby setting the acquisition condition for each of the specific scenes, and further comprising a change unit (18) that changes the reliability threshold value based on the number of times each of the acquisition conditions is met.

[0129] [Configuration 3] The vehicle control device according to Configuration 1, wherein the inference model is machine-learned in advance using the training data that includes information about the plurality of predetermined specific scenes and information about scenes unrelated to the specific scenes.

[0130] [Configuration 4] The vehicle control device according to Configuration 1, wherein the determination unit determines that an acquisition condition is met when any of the scene reliabilities estimated by the scene estimation unit is equal to or greater than a predetermined reliability threshold and when a parameter related to a vehicle driving situation satisfies a predetermined condition.

[0131] [Configuration 5] The vehicle control device according to any one of configurations 1 to 4, wherein the reliability threshold for each specific scene is determined, thereby setting the acquisition condition for each specific scene, and further comprising a change unit that changes the reliability threshold for the specific scene depending on the occurrence probability of the specific scene.

[0132] [Configuration 6] A vehicle control device according to any one of configurations 1 to 5, wherein the reliability threshold for each specific scene is determined, thereby setting the acquisition condition for each specific scene, and the reliability threshold for the specific scene is set to a smaller value when the probability of occurrence of the specific scene is low compared to when the probability of occurrence is high.

[0133] [Configuration 7] The vehicle control device according to any one of configurations 1 to 4, wherein the reliability threshold for each specific scene is determined, thereby setting the acquisition condition for each specific scene, and further comprising a change unit that changes the reliability threshold for the specific scene depending on the amount of data related to the specific scene that the transmission unit has transmitted to the server or the amount of data related to the specific scene that is stored in the server.

[0134] [Configuration 8] A vehicle control device according to any one of configurations 1 to 4, wherein the reliability threshold for each specific scene is determined, thereby setting the acquisition condition for each specific scene, and the reliability threshold for the specific scene is set to a lower value when the amount of data relating to the specific scene transmitted by the transmitting unit to the server or the amount of data relating to the specific scene stored in the server is small, compared to when the amount is large.

[0135] [Configuration 9] A vehicle control device described in any of configurations 1 to 8, comprising a driving skill judgment unit that judges whether or not a predetermined driving skill is possessed, and the judgment unit judges whether or not the acquisition condition is met depending on the judgment result of the driving skill judgment unit when any of the scene reliabilities output by the scene estimation unit is equal to or greater than a predetermined reliability threshold.

[0136] [Configuration 10] The vehicle control device according to Configuration 1, wherein the reliability threshold is specified by the server, and a value corresponding to the quantity of the data transmitted from the vehicle or the number of times the acquisition condition is satisfied is specified. [Configuration 11] The vehicle control device according to any of Configurations 1 to 10, wherein the vehicle control device is configured to be able to implement a shadow mode in which the program to be verified runs in a manner independent of control of the vehicle, and the data acquisition unit acquires data related to an output result of the program to be verified together with the data related to the sensor information, or acquires data related to the output result instead of the data related to the sensor information, using sensor information acquired by various sensors mounted on the vehicle as input values.

[0137] [Configuration 12] A vehicle control device according to any one of configurations 1 to 10, wherein the inference model is updated at a predetermined timing by downloading a new inference model generated by re-learning based on the data transmitted from the vehicle from the server.

[0138] [Configuration 13] A program executed by a vehicle control device (10) mounted on a vehicle (101), the program causing the vehicle control device to perform the following steps: a scene estimation step of inputting sensor information acquired by a sensor (20) mounted on the vehicle into an inference model, and outputting, for each specific scene, a scene reliability indicating how similar the scene in which the vehicle is traveling is to each of a plurality of predetermined specific scenes; a determination step of determining that an acquisition condition is met if any of the scene reliability output by the scene estimation step is equal to or greater than a predetermined reliability threshold; a data acquisition step of acquiring data related to the sensor information if it is determined by the determination step that the acquisition condition is met; and a transmission step of transmitting the data acquired by the data acquisition step to a server (102) via a communication network (103), wherein the inference model is machine-learned in advance using training data including information about the plurality of specific scenes.

[0139] Although the present disclosure has been described with reference to the embodiments, it is understood that the present disclosure is not limited to the embodiments or structures. The present disclosure also encompasses various modifications and equivalent modifications. In addition, various combinations and forms, including only one element, more than one element, or less than one element, are also within the scope and spirit of the present disclosure.

Claims

1. A vehicle control device (10) mounted on a vehicle (101), comprising: a scene estimation unit (14, 114) that inputs sensor information acquired by a sensor (20) mounted on the vehicle into an inference model and outputs a scene reliability indicating the degree to which the scene in which the vehicle is traveling is similar to each of a plurality of predetermined specific scenes; a determination unit (15, 115) that determines that an acquisition condition is met if any of the scene reliability output by the scene estimation unit is equal to or greater than a predetermined reliability threshold; a data acquisition unit (16, 116) that acquires data related to the sensor information when the determination unit determines that the acquisition condition is met; and a transmission unit (17, 117) that transmits the data acquired by the data acquisition unit to a server (102) via a communication network (103), wherein the inference model is machine-learned in advance using training data including information about the plurality of specific scenes.

2. A vehicle control device as described in claim 1, wherein the reliability threshold value for each specific scene is determined, thereby setting the acquisition condition for each specific scene, and the vehicle control device is provided with a change unit (18) that changes the reliability threshold value based on the number of times each acquisition condition is met.

3. A vehicle control device as described in claim 1, wherein the inference model is machine-learned in advance using training data that includes information about the plurality of predetermined specific scenes and information about scenes unrelated to the specific scenes.

4. A vehicle control device as described in claim 1, wherein the determination unit determines that the acquisition condition is met when any of the scene reliabilities estimated by the scene estimation unit is equal to or greater than a predetermined reliability threshold and when parameters related to the vehicle's driving conditions satisfy predetermined conditions.

5. A vehicle control device as described in claim 1, wherein the reliability threshold for each specific scene is determined, thereby setting the acquisition condition for each specific scene, and the vehicle control device is provided with a modification unit that modifies the reliability threshold for the specific scene depending on the occurrence probability of the specific scene.

6. A vehicle control device as described in claim 1, wherein the reliability threshold for each specific scene is determined, thereby setting the acquisition condition for each specific scene, and the reliability threshold for the specific scene is set to a smaller value when the probability of occurrence of the specific scene is low compared to when the probability is high.

7. A vehicle control device as described in claim 1, wherein the reliability threshold for each specific scene is determined, thereby setting the acquisition conditions for each specific scene, and the vehicle control device is provided with a modification unit that modifies the reliability threshold for the specific scene depending on the amount of data related to the specific scene transmitted to the server by the transmission unit or the amount of data related to the specific scene stored in the server.

8. A vehicle control device as described in claim 1, wherein the reliability threshold for each specific scene is determined, thereby setting the acquisition conditions for each specific scene, and the reliability threshold for the specific scene is set to a lower value when the amount of data related to the specific scene transmitted by the transmitting unit to the server or the amount of data related to the specific scene stored in the server is small compared to when the amount is large.

9. A vehicle control device as described in claim 1, comprising a driving skill judgment unit that judges whether or not a predetermined driving skill is possessed, and the judgment unit judges whether or not the acquisition condition is met depending on the judgment result of the driving skill judgment unit when any of the scene reliabilities output by the scene estimation unit is equal to or greater than a predetermined reliability threshold.

10. A vehicle control device according to claim 1, wherein the reliability threshold is specified by the server and a value is specified according to the quantity of data transmitted from the vehicle or the number of times the acquisition condition is met.

11. The vehicle control device is configured to be able to implement a shadow mode in which the program to be verified operates in a manner not related to the control of the vehicle, and the data acquisition unit acquires data on the output results of the program to be verified together with data on the sensor information when the program to be verified is executed using sensor information acquired by various sensors mounted on the vehicle as input values, or acquires data on the output results instead of data on the sensor information. A vehicle control device as described in any one of claims 1 to 10.

12. A vehicle control device as described in any one of claims 1 to 10, wherein the inference model is updated at a predetermined timing by downloading a new inference model generated by re-learning based on the data transmitted from the vehicle from the server.

13. A program executed by a vehicle control device (10) mounted on a vehicle (101), which causes the vehicle control device to perform the following steps: a scene estimation step of inputting sensor information acquired by a sensor (20) mounted on the vehicle into an inference model and outputting, for each specific scene, a scene reliability indicating the degree to which the scene in which the vehicle is traveling is similar to each of a plurality of predetermined specific scenes; a determination step of determining that an acquisition condition is met if any of the scene reliability output by the scene estimation step is equal to or greater than a predetermined reliability threshold; a data acquisition step of acquiring data related to the sensor information if it is determined by the determination step that the acquisition condition is met; and a transmission step of transmitting the data acquired by the data acquisition step to a server (102) via a communication network (103), wherein the inference model is machine-learned in advance using training data including information about the plurality of specific scenes.

Images