System and vehicle

Abstract

A system (1) is provided with a vehicle external system (3) and a vehicle (5). The vehicle comprises: a processing unit (23) that performs processing using one of a first model (25_1) and a second model (25_2); a data generation unit (29) that generates upload data (19) when the used model satisfies a generation condition; an upload unit (31) that uploads the upload data; an observation amount determination unit (33, 35) that determines whether an observation amount exceeds a threshold value (17A, 17B) when the first model is used; and a model change unit (37) that changes the used model from the first model to the second model when the observation amount exceeds the threshold value.

Description

Systems and vehicles CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This international application claims priority based on Japanese Patent Application No. 2024-012934, filed with the Japan Patent Office on January 31, 2024, the entire contents of which are incorporated herein by reference.

[0002] The present disclosure relates to systems and vehicles.

[0003] The vehicle performs various operations using the model stored in the vehicle. A new version of the model may be downloaded to the vehicle and the new version of the model may replace the original version of the model.

[0004] Patent Publication No. 2023-84379

[0005] As a result of detailed investigations by the inventors, the following problem was found: After a new version of a model is downloaded to a vehicle, the new version of the model may behave differently from the model developer's expectations. In this case, it is necessary to revert the model used in the vehicle from the new version of the model to the original version of the model. However, the model used in the vehicle cannot be reverted to the original version of the model until the original version of the model is downloaded again.

[0006] In one aspect of the present disclosure, it is preferable to provide a system and vehicle that can revert the model used in the vehicle to an original version of the model without necessarily having to re-download the original version of the model.

[0007] One aspect of the present disclosure is a system including an off-vehicle system and a vehicle. The vehicle includes: a processing unit configured to perform processing using one of a first model and a second model downloaded to the vehicle before the first model; a data generation unit configured to generate upload data when the model used by the processing unit satisfies a preset generation condition; an upload unit configured to upload the upload data generated by the data generation unit to the off-vehicle system; an observable quantity determination unit configured to determine, when the processing unit is using the first model, whether an observable quantity including at least one of (a) an amount of the upload data uploaded by the upload unit and (b) a usage amount of resources used by the processing unit for processing exceeds a threshold; and a model change unit configured to change the model used by the processing unit from the first model to the second model when the observable quantity determination unit determines that the observable quantity exceeds the threshold. The off-vehicle system includes a data storage unit configured to store the upload data uploaded by the upload unit.

[0008] One aspect of the present disclosure is a system that allows a vehicle to revert from using a first model to using a second model without necessarily having to re-download the second model.

[0009] Another aspect of the present disclosure is a vehicle including: a processing unit configured to perform processing using one of a first model and a second model downloaded before the first model; a data generation unit configured to generate upload data when the model used by the processing unit satisfies a predetermined generation condition; an upload unit configured to upload the upload data generated by the data generation unit to the off-vehicle system; an observable quantity determination unit configured to, when the processing unit is using the first model, determine whether an observable quantity exceeds a threshold, the observable quantity including at least one of (a) the amount of the upload data uploaded by the upload unit and (b) the usage amount of resources used by the processing unit to perform processing; and a model change unit configured to change the model used by the processing unit from the first model to the second model when the observable quantity determination unit determines that the observable quantity exceeds the threshold.

[0010] In another aspect of the present disclosure, a vehicle can revert the model used by the vehicle from a first model to a second model without necessarily having to re-download the second model.

[0011] It is a block diagram showing the configuration of a system in a first embodiment.It is a flowchart showing processing relating to rollback.It is a block diagram showing the configuration of a system in a second embodiment.

[0012] An exemplary embodiment of the present disclosure will be described with reference to the drawings. First Embodiment 1. Configuration of System 1 (1-1) Overall Configuration of System 1 The overall configuration of the system 1 will be described with reference to FIG. 1. The system 1 includes a cloud 3 and a vehicle 5. The vehicle 5 may be a single vehicle or multiple vehicles.

[0013] (1-2) Configuration of Cloud 3 The cloud 3 includes a threshold notification unit 11, a data accumulation unit 13, and a version management unit 15. The threshold notification unit 11 generates thresholds 17A and 17B and provides them to the vehicle 5. The threshold 17A is a threshold for the amount of upload data 19. The upload data 19 will be described later. The threshold 17B is a threshold for the amount of usage of resources 27. The resources 27 will be described later.

[0014] The data storage unit 13 stores upload data 19. The upload data 19 is uploaded from the vehicle 5 to the cloud 3. When a rollback is performed in the vehicle 5, the version management unit 15 receives a notification from the vehicle 5 and stores the notification. Rollback will be described later.

[0015] (1-3) Configuration of Vehicle 5 The vehicle 5 is equipped with a production application 21. Sensor data is input to the production application 21 from sensors equipped in the vehicle 5. The production application 21 performs processing based on the input and produces output. The output of the production application 21 is used to realize the functions of the vehicle 5. Examples of the functions of the vehicle 5 include collecting data using sensors, controlling the operation of components equipped in the vehicle 5, and electrical control.

[0016] The vehicle 5 has a shadow mode 23. The shadow mode 23 performs processing using a model 25. The models 25 include a first model 25_1 and a second model 25_2. The first model 25_1 and the second model 25_2 are each downloaded from the cloud 3 to the vehicle 5 and stored in the vehicle 5. The second model 25_2 is a model 25 that was downloaded to the vehicle 5 earlier than the first model 25_1.

[0017] The shadow mode 23 uses either the first model 25_1 or the second model 25_2. After the first model 25_1 is downloaded, the shadow mode 23 uses the first model 25_1 until a rollback, which will be described later, is performed. The shadow mode 23 used the second model 25_2 until the first model 25_1 was downloaded.

[0018] If the latest model 25 is downloaded after the first model 25_1, the latest model 25 becomes the first model 25_1. The previous first model 25_1 becomes the second model 25_2.

[0019] Sensor data is input to the shadow mode 23. The sensor data input to the shadow mode 23 is the same as that input to the production application 21. The shadow mode 23 performs processing based on the input and produces output. The output of the shadow mode 23 is not activated. In other words, the output of the shadow mode 23 is not used to realize the functions of the vehicle 5. The shadow mode 23 corresponds to a processing unit.

[0020] The shadow mode 23 includes resources 27. When performing processing, the shadow mode 23 uses the resources 27. The resources 27 include, for example, a CPU, memory, storage, and the like.

[0021] The shadow mode 23 can be used to develop a model 25. The model 25 can be developed by updating the model 25 so as to reduce the amount of upload data and the amount of resources 27 used, which will be described later.

[0022] The vehicle 5 includes a data generation unit 29. The data generation unit 29 detects a difference between the output of the production application 21 and the output of the shadow mode 23. The data generation unit 29 compares the detected difference with a reference value. If the difference is greater than the reference value, the data generation unit 29 generates upload data 19. If the difference is equal to or less than the reference value, the data generation unit 29 does not generate upload data 19. Note that a difference greater than the reference value corresponds to a preset generation condition.

[0023] The vehicle 5 includes an upload unit 31. The upload unit 31 uploads the upload data 19 generated by the data generation unit 29 to the cloud 3. The cloud 3 stores the uploaded upload data 19 in the data storage unit 13.

[0024] The vehicle 5 includes a data amount determination unit 33. The data amount determination unit 33 can set a threshold value 17A. When the shadow mode 23 uses the first model 25_1, the data amount determination unit 33 acquires the amount of upload data 19 uploaded per unit time by the upload unit 31 (hereinafter referred to as the upload data amount).

[0025] The data amount determination unit 33 determines whether the amount of uploaded data exceeds the threshold 17A. If it is determined that the amount of uploaded data exceeds the threshold 17A, the data amount determination unit 33 notifies the model change unit 37, which will be described later. The amount of uploaded data corresponds to the observation amount. The data amount determination unit 33 also corresponds to the observation amount determination unit.

[0026] The vehicle 5 includes a resource determination unit 35. The resource determination unit 35 can set a threshold 17B. The resource determination unit 35 acquires the usage amount of the resource 27 when the shadow mode 23 is performing processing using the first model 25_1.

[0027] The usage of the resource 27 can be classified into metrics, traces, and logs. Examples of the usage of the resource 27 classified into metrics include CPU usage, memory usage, and storage usage. Examples of the usage of the resource 27 classified into traces include the number of component calls and the length of latency. Examples of the usage of the resource 27 classified into logs include the number or frequency of individual event information. The usage of the resource 27 may be a ratio to some standard usage. The usage of the resource 27 corresponds to the observation amount.

[0028] The resource determination unit 35 determines whether the usage amount of the resource 27 exceeds the threshold 17B. If it is determined that the usage amount of the resource 27 exceeds the threshold 17B, the resource determination unit 35 notifies the model modification unit 37, which will be described later. The resource determination unit 35 corresponds to the observation amount determination unit.

[0029] The vehicle 5 includes a model change unit 37. When the model change unit 37 receives a notification from the data amount determination unit 33 or the resource determination unit 35, the model change unit 37 instructs the shadow mode 23 to change the model 25 used in the shadow mode 23 from the first model 25_1 to the second model 25_2. Hereinafter, changing the model 25 used in the shadow mode 23 from the first model 25_1 to the second model 25_2 will be referred to as a rollback.

[0030] In addition, when a notification is received from the data amount determination unit 33, it is when the shadow mode 23 is using the first model 25_1 and the data amount determination unit 33 determines that the amount of uploaded data exceeds the threshold value 17A.

[0031] In addition, when a notification is received from the resource determination unit 35, the resource determination unit 35 determines that the usage of the resource 27 exceeds the threshold 17B while the shadow mode 23 is using the first model 25_1.

[0032] When the model modification unit 37 performs a rollback, it notifies the version management unit 15 .

[0033] 2. Process for Setting Thresholds 17A and 17B The threshold notification unit 11 generates thresholds 17A and 17B and notifies the vehicle 5. In the vehicle 5, the data amount determination unit 33 sets the notified threshold 17A. After this setting, the data amount determination unit 33 uses the set threshold 17A when determining whether the amount of uploaded data exceeds the threshold 17A.

[0034] Furthermore, in the vehicle 5, the resource determination unit 35 sets the notified threshold value 17B. After this setting, the resource determination unit 35 uses the set threshold value 17B when determining whether or not the usage amount of the resource 27 exceeds the threshold value 17B.

[0035] The threshold 17A is set based on, for example, the actual amount of uploaded data when the second model 25_2 was used. For example, the threshold 17A may be obtained by subtracting a predetermined value from the average amount of uploaded data when the second model 25_2 was used, or the threshold 17A may be obtained by subtracting a predetermined value from the average amount of uploaded data when the second model 25_2 was used and adding a predetermined margin regarding variations in the amount of uploaded data over time, etc. As a result, the threshold 17A may be larger than the average amount of uploaded data when the second model 25_2 was used.

[0036] Using the threshold value 17A, it is possible to realize a mechanism for quickly detecting an increase in the amount of uploaded data on the vehicle 5 side. The threshold value 17A is a threshold value for detecting, for example, that the amount of uploaded data when the first model 25_1 is used has become larger than the amount of uploaded data when the second model 25_2 is used.

[0037] 3. Process Related to Rollback When the shadow mode 23 uses the first model 25_1, the vehicle 5 repeatedly executes the process shown in FIG. 2 at predetermined time intervals.

[0038] In step 1, the amount of observation is acquired. Specifically, the data amount determination unit 33 acquires the amount of uploaded data. Also, the resource determination unit 35 acquires the amount of resource 27 used.

[0039] In step 2, it is determined whether the observation amount acquired in step 1 exceeds a threshold. Specifically, the data amount determination unit 33 determines whether the upload data amount acquired in step 1 exceeds a threshold 17A. In addition, the resource determination unit 35 determines whether the usage amount of the resource 27 acquired in step 1 exceeds a threshold 17B.

[0040] If it is determined that the amount of uploaded data exceeds the threshold 17A, or if it is determined that the amount of resource 27 usage exceeds the threshold 17B, the process proceeds to step 3. If it is determined that the amount of uploaded data does not exceed the threshold 17A and that the amount of resource 27 usage does not exceed the threshold 17B, the process ends. In this case, the shadow mode 23 continues to use the first model 25_1.

[0041] The first model 25_1 is a model 25 that has been updated for improvement (hereinafter referred to as the updated model), and the second model 25_2 is a model 25 before the update (hereinafter referred to as the pre-updated model).

[0042] One of the reasons why the amount of uploaded data when the first model 25_1 is used is larger than the amount of uploaded data when the second model 25_2 is used is a failure in model improvement.

[0043] For example, let us consider an example in which the production application 21 is an application that receives the output of a rain sensor and outputs a rainfall amount. A rain sensorless rainfall estimation model (hereinafter referred to as a rainfall estimation model) that receives an image captured by an in-vehicle camera and can output a rainfall amount that is identical to the rainfall amount of the production application 21 is developed using the shadow mode 23. The model 25 corresponds to the rainfall estimation model.

[0044] Assume that development in this example is proceeding as follows:

[0045] A1. The cloud 3 distributes the first version of the rainfall estimation model to the vehicle 5. The vehicle 5 operates the rainfall estimation model in shadow mode 23. When a difference occurs between the output of the production application 21 and the output of the rainfall estimation model, the vehicle 5 transmits various data to the cloud 3 as upload data 19. The various data are designated in advance by the cloud 3 as data that will contribute to improving the rainfall estimation model.

[0046] After that, we will continue to improve the rainfall estimation model by repeating the cycle of A2 and A3 below.

[0047] A2. The cloud 3 uses the upload data 19 (i.e., various data transmitted when a difference in output occurs between the production application 21 and the rainfall estimation model) to update the rainfall estimation model with the aim of improving it (i.e., to enable it to produce an output that is closer to the output of the production application 21), thereby generating an updated model, and distributes the updated model to the vehicle 5.

[0048] A3. The vehicle 5 operates the updated model in shadow mode 23. When a difference occurs between the output of the production application 21 and the output of the updated model, the vehicle 5 transmits various data to the cloud 3 as upload data 19. The various data are specified in advance by the cloud 3.

[0049] As the cycle between A2 and A3 continues, there may be cases where the amount of data uploaded per unit time by the updated model becomes larger than the amount of data uploaded by the model 25 before the update (i.e., the pre-update model). In this case, the improvement aimed at by the update (i.e., reducing the difference between the output of the production application 21 and the output of the rainfall estimation model) has failed.

[0050] In step 3, the model change unit 37 instructs the shadow mode 23 to perform a rollback. After the rollback, the shadow mode 23 performs processing using the second model 25_2.

[0051] The rollback mechanism is a mechanism in which the second model 25_2 is kept on the vehicle 5 side even after the first model 25_1 has been distributed to the vehicle 5, and in response to detecting that the amount of uploaded data when using the first model 25_1 has exceeded the threshold 17A, the model 25 operating in the shadow mode 23 is switched to the second model 25_2.

[0052] This mechanism makes it possible to prevent the first model 25_1, which has not achieved the intended improvement (i.e., the improvement of reducing the difference between the output of the production application 21 and the output of the model 25), from continuing to send large amounts of upload data 19, contrary to the intended purpose.

[0053] In step 4, the model modification unit 37 sends a rollback notification to the version management unit 15, indicating that a rollback has been performed. By sending the rollback notification to the cloud 3, the developer of the model 25 can understand that the improvement made to the first model 25_1 (i.e., the improvement to reduce the difference from the output of the production application 21) may have been a failure. As a result, the developer can be prompted to reconsider the improvement.

[0054] 4. Processing Regarding Failure in Starting Up the First Model 25_1 When the shadow mode 23 is using the first model 25_1, the model change unit 37 performs the following processing.

[0055] The model change unit 37 monitors whether the first model 25_1 is successfully started in the shadow mode 23. Every time the first model 25_1 fails to start, the model change unit 37 increments the count of the number of consecutive failures by 1. When the first model 25_1 is successfully started, the model change unit 37 resets the count of the number of consecutive failures to 0. The initial value of the count of the number of consecutive failures is 0.

[0056] When the count of the number of consecutive failures reaches a preset number, the model change unit 37 performs a rollback. When the count of the number of consecutive failures reaches a preset number, this means that the first model 25_1 has failed to start up that many times in a row. The preset number is, for example, 2 to 10 times, and preferably 5 times.

[0057] 5. Advantages of the System 1 and the Vehicle 5 (1A) The vehicle 5 can perform a rollback without re-downloading the second model 25_2 to the vehicle 5. Furthermore, the vehicle 5 can autonomously perform a rollback without receiving a rollback instruction from the cloud 3.

[0058] (1B) When the shadow mode 23 is using the first model 25_1, a state in which the observable quantity exceeds the thresholds 17A and 17B means that the first model 25_1 is behaving differently from what the developer expected. When the shadow mode 23 is using the first model 25_1 and the observable quantity exceeds the thresholds 17A and 17B, the vehicle 5 performs a rollback.

[0059] Therefore, the vehicle 5 can perform a rollback when the first model 25_1 behaves differently from the developer's expectations.

[0060] (1C) The threshold notification unit 11 generates thresholds 17A and 17B and notifies the vehicle 5. The data amount determination unit 33 uses the threshold 17A notified by the threshold notification unit 11. The resource determination unit 35 uses the threshold 17B notified by the threshold notification unit 11. Therefore, the vehicle 5 does not need to generate the thresholds 17A and 17B.

[0061] (1D) When a rollback is performed, the model change unit 37 notifies the cloud 3. Therefore, the cloud 3 can recognize that a rollback has been performed.

[0062] (1E) The model change unit 37 performs a rollback when the startup of the first model 25_1 fails a preset number of times in succession. The state in which the startup of the first model 25_1 fails a preset number of times in succession means that the first model 25_1 is behaving differently from the developer's expectations.

[0063] Therefore, the vehicle 5 can perform rollback when the first model 25_1 exhibits behavior different from what the developer expects. Second Embodiment 1. Differences from the First Embodiment The second embodiment has the same basic configuration as the first embodiment, and therefore differences will be described below. Note that the same reference numerals as those in the first embodiment indicate the same configuration, and reference is made to the preceding description.

[0064] In the first embodiment described above, all of the thresholds 17A and 17B are generated by the threshold notification unit 11 and provided to the vehicle 5. In contrast, in the second embodiment, as shown in Fig. 3 , the vehicle 5 is different from the first embodiment in that it further includes a threshold generation unit 41.

[0065] The threshold generation unit 41 generates a threshold 17B. The threshold 17B generated by the threshold generation unit 41 is notified to the resource determination unit 35. The resource determination unit 35 sets the threshold 17B notified by the threshold generation unit 41. After this setting, the resource determination unit 35 uses the set threshold 17B when determining whether or not the usage amount of the resource 27 exceeds the threshold 17B.

[0066] The resource determination unit 35 selectively uses, depending on the situation, the threshold 17B generated by the threshold notification unit 11 and the threshold 17B generated by the threshold generation unit 41. Alternatively, the resource determination unit 35 always uses the threshold 17B generated by the threshold generation unit 41.

[0067] The threshold generation unit 41 generates the threshold 17B, for example, as follows: The threshold generation unit 41 acquires the operation log 43 from the shadow mode 23. The operation log 43 is data representing the usage amount of the resource 27 when the shadow mode 23 used the second model 25_2 in the past. The threshold generation unit 41 generates the threshold 17B, for example, by multiplying the operation log 43 by a certain coefficient. The coefficient is a positive value. The coefficient may be smaller than 1, may be 1, or may be greater than 1.

[0068] 2. Effects of the System 1 and the Vehicle 5 According to the second embodiment described above in detail, in addition to the effects (1A) to (1B) and (1D) to (1E) of the first embodiment described above, the system 1 and the vehicle 5 further have the following effects.

[0069] (2A) Even when the vehicle 5 is unable to acquire the threshold value 17B from the cloud 3, the vehicle 5 can generate and use the threshold value 17B using the threshold value generation unit 41. Furthermore, the threshold value generation unit 41 can generate a more appropriate threshold value 17B by using the operation log 43. <Other Embodiments> Although the embodiments of the present disclosure have been described above, the present disclosure is not limited to the above-described embodiments and can be implemented in various modifications.

[0070] (1) The vehicle 5 may include a processing unit different from that in the shadow mode 23. The generation condition for generating the upload data 19 may be a condition different from the condition that the difference between the output of the production application 21 and the output of the shadow mode 23 is greater than a reference value. For example, the generation condition may be a condition that the first model 25_1 exhibits behavior different from the developer's expectations for some reason. For example, an allowable range may be set in advance for the behavior of the first model 25_1, and the generation condition may be that the behavior of the first model 25_1 falls outside the allowable range.

[0071] (2) The observation amount may be only the amount of uploaded data or only the amount of resource 27 used.

[0072] (3) The determination method in step 2 may be another method. For example, if it is determined that the amount of uploaded data exceeds threshold 17A and the amount of resource 27 used exceeds threshold 17B, the process may proceed to step 3, and otherwise the process may be terminated.

[0073] (4) In the first and second embodiments, the system 1 includes the cloud 3 as an off-vehicle system. The system 1 may include an on-premise system outside the vehicle 5 instead of the cloud 3. In this case, the on-premise system outside the vehicle 5 has the same functions as the cloud 3 in the first or second embodiment. The system 1 may also include a combination of the cloud 3 and the on-premise system outside the vehicle 5. In this case, the combination of the cloud 3 and the on-premise system outside the vehicle 5 has the same functions as the cloud 3 in the first or second embodiment.

[0074] (5) The production application 21, shadow mode 23, upload unit 31, data amount determination unit 33, resource determination unit 35, model modification unit 37, threshold generation unit 41, and the methods described in the present disclosure may be realized by a dedicated computer provided by configuring a processor and memory programmed to perform one or more functions embodied in a computer program.

[0075] Alternatively, the production application 21, shadow mode 23, upload unit 31, data amount determination unit 33, resource determination unit 35, model modification unit 37, threshold generation unit 41, and the methods described in the present disclosure may be realized by a dedicated computer provided by configuring a processor with one or more dedicated hardware logic circuits.

[0076] Alternatively, the production application 21, the shadow mode 23, the upload unit 31, the data amount determination unit 33, the resource determination unit 35, the model modification unit 37, the threshold generation unit 41, and the methods described herein may be implemented by one or more dedicated computers configured by combining a processor and memory programmed to perform one or more functions with a processor configured by one or more hardware logic circuits. Furthermore, the computer program may be stored on a computer-readable non-transitory tangible recording medium as instructions to be executed by the computer.

[0077] The method for realizing the functions of each part included in the production application 21, shadow mode 23, upload unit 31, data volume determination unit 33, resource determination unit 35, model modification unit 37, and threshold generation unit 41 does not necessarily need to include software, and all of the functions may be realized using one or more pieces of hardware.

[0078] (6) Multiple functions possessed by one component in the above embodiments may be realized by multiple components, or one function possessed by one component may be realized by multiple components. Also, multiple functions possessed by multiple components may be realized by one component, or one function realized by multiple components may be realized by one component. Also, part of the configuration of the above embodiments may be omitted. Also, at least part of the configuration of the above embodiments may be added to or substituted for the configuration of another of the above embodiments.

[0079] (7) In addition to the above-described system 1, the present disclosure can also be realized in various forms, such as a higher-level system that includes the system 1 as a component, a program for causing a computer to function as the production application 21, the shadow mode 23, the upload unit 31, the data amount determination unit 33, the resource determination unit 35, the model modification unit 37, and the threshold generation unit 41, a non-transient physical recording medium such as a semiconductor memory on which this program is recorded, a method for manufacturing the system 1, a method for manufacturing the vehicle 5, etc. [Technical Ideas Disclosed in the Present Specification] [Item 1] A system (1) including an off-vehicle system (3) and a vehicle (5), wherein the vehicle includes: a processing unit (23) configured to perform processing using one model (25) of a first model (25_1) and a second model (25_2) downloaded to the vehicle prior to the first model; a data generation unit (29) configured to generate upload data (19) when the model used by the processing unit satisfies a preset generation condition; an upload unit (31) configured to upload the upload data generated by the data generation unit to the off-vehicle system; and an observation quantity determination unit (33, 35) configured to determine whether or not an observation quantity including at least one of (a) the amount of the upload data uploaded by the upload unit and (b) the usage amount of a resource (27) used by the processing unit to perform processing exceeds a threshold value (17A, 17B) when the processing unit is using the first model. a model change unit (37) configured to change the model used by the processing unit from the first model to the second model when the observable quantity determination unit determines that the observable quantity exceeds the threshold, and the off-vehicle system comprises a data storage unit (13) configured to store the upload data uploaded by the upload unit. [Item 2] The system according to item 1, wherein the off-vehicle system further comprises a threshold notification unit (11) configured to generate the threshold and notify the vehicle of the threshold, and the observable quantity determination unit is configured to use the threshold notified by the threshold notification unit.[Item 3] The system according to item 1 or 2, wherein the vehicle further includes a threshold generating unit (41) configured to generate the threshold, and the observable quantity determining unit is configured to use the threshold generated by the threshold generating unit. [Item 4] The system according to any one of items 1 to 3, wherein the model changing unit is configured to notify the off-vehicle system when the model is changed. [Item 5] The system according to any one of items 1 to 4, wherein the processing unit is in a shadow mode (23). [Item 6] The system according to any one of items 1 to 5, wherein the model changing unit is configured to change the model used by the processing unit from the first model to the second model when activation of the first model fails a preset number of times consecutively. [Item 7] The system according to Item 1, wherein the vehicle includes a production application (21) and a shadow mode as the processing unit; the first model is an improvement over the second model so as to provide an output closer to an output of the production application (21); the data generation unit (29) is configured to generate the upload data (19) when a difference between an output of the production application (21) and an output of the shadow mode is greater than a reference value; the observation quantity determination unit (33, 35) is configured to determine whether or not an amount of upload data exceeds a threshold when the shadow mode is using the first model; and the model change unit (37) is configured to change the model used by the shadow mode from the first model to the second model and notify the off-vehicle system of the change, when it is determined that the amount of upload data exceeds the threshold; and the threshold is a threshold for detecting that the amount of upload data when the first model is used is larger than the amount of upload data when the second model is used.[Item 8] A vehicle (5) comprising: a processing unit (23) configured to perform processing using one model (25) of a first model (25_1) and a second model (25_2) downloaded before the first model; a data generation unit (29) configured to generate upload data (19) when the model used by the processing unit satisfies a preset generation condition; an upload unit (31) configured to upload the upload data generated by the data generation unit to an off-vehicle system; an observation quantity determination unit (33, 35) configured to determine whether or not an observation quantity including at least one of (a) the amount of the upload data uploaded by the upload unit and (b) the usage amount of a resource (27) used by the processing unit to perform processing exceeds a threshold value (17A, 17B) when the processing unit is using the first model; and a model change unit (37) configured to change the model used by the processing unit from the first model to the second model when the observation quantity determination unit determines that the observation quantity exceeds the threshold value.

Claims

1. A system (1) comprising an off-vehicle system (3) and a vehicle (5), wherein the vehicle comprises: a processing unit (23) configured to perform processing using one model (25) of a first model (25_1) and a second model (25_2) downloaded to the vehicle prior to the first model; a data generation unit (29) configured to generate upload data (19) when the model used by the processing unit satisfies a preset generation condition; an upload unit (31) configured to upload the upload data generated by the data generation unit to the off-vehicle system; and an observation quantity determination unit (33, 35) configured to determine whether an observation quantity including at least one of (a) the amount of upload data uploaded by the upload unit and (b) the usage amount of a resource (27) used by the processing unit to perform processing exceeds a threshold value (17A, 17B) when the processing unit is using the first model. a model change unit (37) configured to change the model used by the processing unit from the first model to the second model when the observation quantity determination unit determines that the observation quantity exceeds the threshold, wherein the off-vehicle system comprises a data storage unit (13) configured to store the upload data uploaded by the upload unit.

2. A system according to claim 1, wherein the off-vehicle system further comprises a threshold notification unit (11) configured to generate the threshold and notify the vehicle of the threshold, and the observable quantity determination unit is configured to use the threshold notified by the threshold notification unit.

3. A system according to claim 1, wherein the vehicle further comprises a threshold generating unit (41) configured to generate the threshold, and the observable quantity determining unit is configured to use the threshold generated by the threshold generating unit.

4. A system according to claim 1 or 2, wherein the model change unit is configured to notify the external vehicle system when the model is changed.

5. A system according to claim 1 or 2, wherein the processing unit is in shadow mode (23).

6. A system according to claim 1 or 2, wherein the model change unit is configured to change the model used by the processing unit from the first model to the second model when the first model fails to start a predetermined number of times in succession.

7. A system according to claim 1, wherein the vehicle is equipped with a production application (21) and a shadow mode as the processing unit, the first model is an improvement over the second model so as to provide an output closer to the output of the production application (21), the data generation unit (29) is configured to generate the upload data (19) when a difference between the output of the production application (21) and the output of the shadow mode is greater than a reference value, the observation quantity determination unit (33, 35) is configured to determine whether or not the amount of upload data exceeds the threshold when the shadow mode is using the first model, and the model change unit (37) is configured to change the model used by the shadow mode from the first model to the second model and notify the off-vehicle system of the change, when it is determined that the amount of upload data exceeds the threshold, and the threshold is a threshold for detecting that the amount of upload data when the first model is used is larger than the amount of upload data when the second model is used.

8. A vehicle (5) comprising: a processing unit (23) configured to perform processing using one model (25) of a first model (25_1) and a second model (25_2) downloaded before the first model; a data generation unit (29) configured to generate upload data (19) when the model used by the processing unit satisfies a preset generation condition; an upload unit (31) configured to upload the upload data generated by the data generation unit to an off-vehicle system; an observation quantity determination unit (33, 35) configured to determine whether an observation quantity including at least one of (a) the amount of upload data uploaded by the upload unit and (b) the usage amount of a resource (27) used by the processing unit to perform processing exceeds a threshold value (17A, 17B) when the processing unit is using the first model; and a model change unit (37) configured to change the model used by the processing unit from the first model to the second model when the observation quantity determination unit determines that the observation quantity exceeds the threshold value.

Images