Additional learning method and additional learning system

Abstract

In the present invention, imaging data obtained by imaging the external environments of autonomous vehicles (2) are input into a training model (M1), a base station (1) acquires control values of the autonomous vehicles (2) on the basis of an output from the training model (M1), approval operations by users of the autonomous vehicles (2) are received for the use of unapproved data in additional training of the training model (M1), the unapproved data including image data based on the imaging data and the control values acquired on the basis of the imaging data, and training data, in which are used approved data obtained by adding information indicating the approvals being received to the unapproved data for which the approvals of the users were obtained, is used for additional training of the training model (M1).

Description

Additional Learning Method and Additional Learning System

[0001] The present disclosure relates to an additional learning method and an additional learning system.

[0002] Patent Document 1 discloses an in-vehicle camera device that, when it is determined that misidentification has occurred when a manual driving operation contrary to automatic driving control is performed, stores the misidentified image and performs additional learning using the stored misidentified image.

[0003] Japanese Unexamined Patent Application Publication No. 2023-106029

[0004] An additional learning method according to an aspect of the present disclosure includes: inputting, by a base station provided to be communicable with an autonomous driving vehicle, imaging data obtained by imaging an external environment of the autonomous driving vehicle into a learning model; acquiring, by the base station, a control value for the autonomous driving vehicle based on an output of the learning model; receiving an approval operation by a user of the autonomous driving vehicle for using unapproved data including image data based on the imaging data and the control value acquired based on the imaging data for additional learning of the learning model; and additionally learning learning data using approved data obtained by adding information indicating approval to the unapproved data for which the user's approval has been obtained, to update the learning model.

[0005] It is a diagram showing one configuration of an additional learning system according to an embodiment of the present disclosure. It is a block diagram showing an example of functional configurations of a base station, a server, and a learning device in the additional learning system according to an embodiment of the present disclosure.

[0006] Hereinafter, an embodiment of the present disclosure will be described in detail with reference to the drawings. For ease of understanding, the background and problems of the present disclosure will be described first, and then the details of the present disclosure will be described.

[0007] <Conventional Autonomous Driving System and Problems Thereof> Conventionally, an autonomous driving system that recognizes the external environment of a vehicle (other vehicles, pedestrians, stationary obstacles, signs, etc.) from output data of a camera mounted on the vehicle and autonomously drives the vehicle based on the recognition result is known (for example, Patent Document 1). For recognition of the external environment, a learning model obtained by machine learning of learning image data to which correct labels are attached (hereinafter referred to as learning data) is used.

[0008] In recent years, there has been a growing demand for transparency in the training data used for model learning. To ensure transparency in training data, it is necessary to consider the privacy rights of individuals whose faces or other images are included in the training data, and to properly manage the providers (sources) of the training data. In training models for controlling autonomous vehicles, the training data used for additional training is likely to include images of people such as pedestrians. Therefore, securing transparent training data to improve the estimation accuracy of training models for controlling autonomous vehicles is not easy.

[0009] "Images of appearance, etc." include images of the appearance (face) and physical form (physical appearance) of a person with a right to privacy, etc., images of the license plate of a vehicle driven by a person with a right to privacy, etc., and other images that can be used to identify a person with a right to privacy, etc.

[0010] <Overview of Additional Learning Method> In the additional learning method described herein, the first learning model M1 is further trained by the additional learning system 100 shown in Figure 1. The additional learning system 100 shown in Figure 1 comprises a plurality of base stations 1, a server 3 that is communicatively connected to each base station 1, and a learning device 5 that is communicatively connected to the server 3. The additional learning system 100 is a technology that utilizes the computing infrastructure of the base stations 1 to provide highly immediate services to users and devices around the base stations 1 with low latency. By deploying AI (Artificial Intelligence) and machine learning technology applications at the network edge via RAN (Radio Access Network), it promotes the creation of new industries and solutions that leverage low latency and confidentiality.

[0011] Multiple base stations 1 constitute a mobile communication network such as 3G, 4G, 5G, or 6G. Multiple base stations 1 perform wireless communication with terminals located inside the cell 110. Multiple base stations 1 perform wireless communication with at least an autonomous vehicle 2 traveling inside the cell 110 of each base station 1.

[0012] Each autonomous vehicle 2 captures images of the external environment within the cell 110 it is traveling in and transmits the captured data to the base station 1 corresponding to that cell 110. The captured data may be moving image data or still image data captured at predetermined intervals.

[0013] Each base station 1 generates control values ​​for the autonomous vehicle 2 based on the imaging data received from the autonomous vehicle 2 and transmits these control values ​​to the autonomous vehicle 2. The control values ​​for the autonomous vehicle 2 include, for example, the speed of the autonomous vehicle 2, the direction of travel, the braking force applied to the wheels of the autonomous vehicle 2, the accelerator pedal opening, the amount the brake pedal is depressed, and the steering wheel angle.

[0014] Each base station 1 uses a first learning model M1 to generate control values ​​for the autonomous vehicle 2. The first learning model M1 receives image data from the autonomous vehicle 2 as input. The output of the first learning model M1 is, for example, the image recognition result of objects included in the external environment of the autonomous vehicle 2, which is the source of the image data. The image recognition result of objects included in the external environment includes images of other vehicles, pedestrians, traffic lights, signs, stationary obstacles, and their positions, speeds, and directions of movement, as well as images of the external environment captured by the autonomous vehicle 2.

[0015] Each base station 1 acquires control values ​​for the autonomous vehicle 2 based on the results of the autonomous vehicle 2's recognition of its external environment. For example, if a speed limit sign is recognized, the base station 1 acquires control values ​​related to the speed of the autonomous vehicle 2 so that the autonomous vehicle 2's speed is within the speed limit indicated by the sign. The base station 1 transmits these control values ​​to the autonomous vehicle 2.

[0016] In the additional learning system 100, the base station 1 outputs control values ​​for the autonomous vehicle 2 using the first learning model M1, and acquires data usable for further learning of the first learning model M1 based on the image data received from the autonomous vehicle 2. The image data received from the autonomous vehicle 2 may include the appearances of pedestrians passing around the autonomous vehicle 2 and drivers of oncoming vehicles. In recent years, there has been a growing demand for transparency regarding the learning data used for training, and disclosure of the learning data is sometimes required. Therefore, it is not appropriate to include image data containing images of the appearances of people passing around the autonomous vehicle 2 as learning data as is. In the additional learning system 100, when acquiring data usable for further learning of the first learning model M1, the base station 1 performs image processing to convert images of the appearances of people included in the image data into images in which people cannot be identified. Images in which people cannot be identified include, for example, images to which mosaic processing has been applied or images to which a mask image has been superimposed. This type of image processing allows for consideration of the privacy rights of individuals whose images, such as facial features, were included in the captured data. Hereinafter, the captured data that has undergone image processing to transform it into an image in which individuals cannot be identified will be referred to as processed data.

[0017] In the additional learning system 100, the base station 1 transmits unauthorized data, including the processed data and the control values ​​of the autonomous vehicle 2 output based on the image data, to the server 3. The server 3 stores the unauthorized data received from the base station 1. In order to ensure the transparency of the learning data for additional learning of the first learning model M1, the server 3 obtains user approval (permission, authorization, consent) for using the unauthorized data for additional learning.

[0018] For example, Server 3 functions as a web server for a portal site accessible to the user of the autonomous vehicle 2, and displays the portal site on the information terminal 4 held by the user of the autonomous vehicle 2. When the user of the autonomous vehicle 2 logs into the portal site, Server 3 presents the user with unauthorized data based on image data of the external environment of the user's autonomous vehicle 2, selected from the unauthorized data stored in Server 3. For example, Server 3 displays a list of unauthorized data on the information terminal 4 held by the user. When the user uses the information terminal 4 to authorize the use of the unauthorized data for additional training of the first learning model M1, a tag or metadata indicating that it has been approved is added to the unauthorized data, and it becomes approved data. The tag or metadata indicating that it has been approved includes, for example, information indicating the user (data provider) who gave the approval.

[0019] Users are paid a fee based on the volume and number of unauthorized data they have approved for use in the additional training of the first learning model M1. For example, if a user is using a mobile communication network configured by base station 1, the usage fee may be discounted based on the volume and number of approved unauthorized data.

[0020] The developer of the first learning model M1 connects to the server 3 via the learning device 5 and obtains approved data. The developer of the first learning model M1 performs annotation work on this approved data to generate additional training data for further training of the first learning model M1. The learning device 5 updates the first learning model M1 by further training with this training data. The updated first learning model M1 is transmitted to each base station 1.

[0021] The additional learning system 100 ensures transparency of the learning data by generating learning data for additional training of the first learning model M1 using approved data that has been tagged or has metadata indicating that it has been approved.

[0022] In the annotation process, tags or metadata indicating the correct recognition result of the external environment of the autonomous vehicle 2 are attached to the approved data. These tags are what the first learning model M1 should output when the imaging data that formed the basis of the approved data is input to the first learning model M1. Since the approved data includes the control values ​​of the autonomous vehicle 2, the developer of the first learning model M1 can select approved data for a desired situation. A desired situation might be, for example, when the speed of the autonomous vehicle 2 is rapidly reduced.

[0023] Such an additional learning system 100 can be applied, for example, to the additional learning of a learning model used for the automatic operation of railway vehicles.

[0024] According to the above configuration, when a user of an autonomous vehicle approves of using unapproved data for additional training of the learning model, information indicating approval is added to the unapproved data to make it approved data. Since training data is generated based on this approved data, transparent training data can be easily secured for additional training of the first learning model related to the control of the autonomous vehicle.

[0025] Furthermore, by using image processing to convert images that can be used to identify individuals within the captured data into images that cannot be used to identify individuals, the privacy rights of those individuals can be respected.

[0026] <Additional Learning System 100> The additional learning system 100 will be described below with reference to the drawings. Figure 2 is a block diagram showing an example of the functional configuration of the base station 1, server 3, and learning device 5 in the additional learning system 100 according to one embodiment of the present disclosure. As shown in Figure 2, the additional learning system 100 according to one embodiment of the present disclosure includes a base station 1, an autonomous vehicle 2, a server 3, an information terminal 4, and a learning device 5. The server 3 and the information terminal 4 may be realized as an integrated configuration as a single information processing terminal device, but in the following description, they will be described as separate devices.

[0027] <Base Station 1> Base station 1 comprises a control unit 10, a storage unit 11, an antenna 12, and a communication unit 13. The control unit 10 controls each part of base station 1 by executing instructions written in a program. The control unit 10 has a RIC (RAN Intelligent Controller) that improves the efficiency and performance of the radio access network (RAN) using at least one of artificial intelligence and machine learning technologies. The control unit 10 also includes an arithmetic unit, registers, peripheral circuits, main memory, etc. The storage unit 11 stores data including the first learning model M1 and a program. The storage unit 11 is, for example, flash memory, HDD (Hard Disk Drive), or magneto-optical disk. The antenna 12 is used for wireless communication with the autonomous vehicle 2 inside cell 110 (Figure 1). The communication unit 13 is an interface for communicating with server 3 and learning device 5 via the network.

[0028] <Autonomous Driving Vehicle 2> The autonomous driving vehicle 2 is equipped with an imaging unit 21, an autonomous driving unit 22, and a communication unit 23. The imaging unit 21 is, for example, an on-board camera that captures images of the external environment of the autonomous driving vehicle 2. The autonomous driving unit 22 is, for example, composed of an ECU, and controls each part of the autonomous driving vehicle 2 based on control values ​​of the autonomous driving vehicle 2 input from the base station 1, thereby driving the autonomous driving vehicle 2 automatically. The communication unit 23 is an interface for wireless communication with the base station 1.

[0029] <Server 3> Server 3 is an information processing device that functions, for example, as a web server for a portal site accessible to users of the autonomous vehicle 2. Server 3 comprises a control unit 30, a storage unit 31, and a communication unit 32. The control unit 30 controls each part of Server 3 by executing instructions written in a program. The control unit 30 includes, for example, an arithmetic unit, registers, peripheral circuits, and main memory. Data and programs are stored in the storage unit 31. The storage unit 31 is, for example, flash memory, an HDD (Hard Disk Drive), or a magneto-optical disk. The communication unit 33 is an interface for communicating with the base station 1, information terminal 4, and learning device 5 via a network.

[0030] <Information Terminal 4> Information terminal 4 is an information processing device operated by the user of the autonomous vehicle 2. Information terminal 4 includes a display unit 41, an input unit 42, and a communication unit 43. Information terminal 4 may be, for example, a stationary PC (Personal Computer), a laptop PC, a mobile terminal such as a smartphone or tablet, an HMD (HeadMount Display), or a wearable terminal such as a smartwatch.

[0031] The display unit 41 is, for example, a display device such as a liquid crystal display device or an organic EL display device. The display unit 41 displays, for example, a screen of a portal site accessible to the user of the autonomous vehicle 2. The input unit 42 accepts input operations from the user of the autonomous vehicle 2. The input unit 42 can use a mouse, keyboard, touchpad, etc. The user of the autonomous vehicle 2 can operate the input unit 42 to perform the task of approving unauthorized data. The communication unit 43 is an interface for communicating with the server 3 via a network.

[0032] <Learning Device 5> The learning device 5 is an information processing device used by the developer of the first learning model M1. The learning device 5 comprises a control unit 50, a storage unit 51, and a communication unit 52. The control unit 50 controls each part of the learning device 5 by executing instructions written in the program. The control unit 50 includes, for example, an arithmetic unit, registers, peripheral circuits, and main memory. The storage unit 51 stores data and programs, including the first learning model M1. The storage unit 51 is, for example, flash memory, an HDD (Hard Disk Drive), or a magneto-optical disk. The communication unit 53 is an interface for communicating with the base station 1, the information terminal 4, and the learning device 5 via a network using wired or wireless communication standards.

[0033] <Functional Configuration of Base Station 1> The control unit 10 of base station 1 functions as an image acquisition unit 101, an operation control unit 102, an image processing unit 103, and a data transmission unit 104 by executing a program stored in the memory unit 11.

[0034] <Image Acquisition Unit 101> The image acquisition unit 101 acquires image data from the autonomous vehicle 2 via the antenna 12, which captures the external environment of the autonomous vehicle 2. The image data acquired by the image acquisition unit 101 is, for example, image data captured by the imaging unit 21 of the autonomous vehicle 2, which captures the external environment of the autonomous vehicle 2.

[0035] <Driving Control Unit 102> The driving control unit 102 generates control values ​​for the autonomous vehicle 2 based on the image data acquired by the image acquisition unit 101. The driving control unit 102 transmits the generated control values ​​for the autonomous vehicle 2 to the autonomous vehicle 2 via the antenna 12. The autonomous driving unit 22 of the autonomous vehicle 2 performs autonomous driving control of the autonomous vehicle 2 based on the control values ​​for the autonomous vehicle 2 generated by the driving control unit 102.

[0036] The driving control unit 102 inputs the image data acquired by the image acquisition unit 101 into the first learning model M1 stored in the storage unit 11. The driving control unit 102 outputs information from the first learning model M1 indicating the recognition result of the external environment of the autonomous vehicle 2. Based on this information indicating the recognition result of the external environment of the autonomous vehicle 2, the driving control unit 102 generates control values ​​for the autonomous vehicle 2. For the method of generating control values ​​for the autonomous vehicle 2 from the recognition result of the external environment of the autonomous vehicle 2, known autonomous driving technology may be used.

[0037] <Image Processing Unit 103> The image processing unit 103 performs image processing on the image data acquired by the image acquisition unit 101 to convert images containing the features of people into images in which people cannot be identified. For example, the image processing unit 103 inputs the image data acquired by the image acquisition unit 101 to the second learning model M2 stored in the storage unit 11. The second learning model M2 detects images such as the features of people from the input image. For example, the second learning model M2 detects face images from the input image. The image processing unit 103 applies image processing such as mosaic processing or mask image superposition to the face images detected by the second learning model M2.

[0038] <Data transmission unit 104> The data transmission unit 104 transmits to the server 3 via the communication unit 13 unapproved data, which consists of the image processing data obtained by the image processing unit 103 from the image processing data and the control values ​​of the autonomous vehicle 2 output based on that image processing data.

[0039] <Functional Configuration of Server 3> The control unit 30 of Server 3 functions as a data receiving unit 301 and an approval acquisition unit 302 by executing a program stored in the storage unit 31.

[0040] <Data Receiving Unit 301> The data receiving unit 301 receives unauthorized data transmitted by the data transmission unit 104 of the base station 1 via the communication unit 32. The data receiving unit 301 stores the received unauthorized data D1 in the storage unit 31.

[0041] <Approval Acquisition Unit 302> The approval acquisition unit 302 acquires approval from a user of an autonomous driving vehicle 2 for unapproved data D1 stored in a storage unit 31. For example, the approval acquisition unit 302 causes the server 3 to provide web page data of a portal site accessible to the user of the autonomous driving vehicle 2 to the information terminal 4. When the user of the autonomous driving vehicle 2 logs into the portal site, the approval acquisition unit 302 presents the unapproved data D1 related to the logged-in user to the user from among the unapproved data D1 stored in the storage unit 31. For example, the approval acquisition unit 302 causes the display unit 41 of the information terminal 4 owned by the user to display the unapproved data D1.

[0042] The approval acquisition unit 302 receives an approval operation for selecting unapproved data D1 that is approved to be used for additional learning of the first learning model M1 from among the unapproved data D1 displayed on the display unit 41 of the information terminal 4. The approval acquisition unit 302 adds a tag or metadata indicating that the unapproved data D1 has been approved by the user to the unapproved data D1 for which the user of the autonomous driving vehicle 2 has performed the approval operation, and generates approved data D2. The approval acquisition unit 302 causes the storage unit 31 to store the generated approved data D2. The approval acquisition unit 302 may further receive an operation of refusing to use the unapproved data D1 for additional learning of the first learning model M1 or an operation of canceling the approval operation. The unapproved data D1 that the user of the autonomous driving vehicle 2 has refused to use for additional learning of the first learning model M1 is deleted from the storage unit 31 of the server 3.

[0043] The approved data D2 is read from the storage unit 31 of the server 3 via the learning device 5 when a developer of the first learning model M1 generates learning data. The developer of the first learning model M1 performs annotation work on the read approved data D2 to generate learning data D3. The generated learning data D3 is stored in a storage unit 51 of the learning device 5.

[0044] <Functional Configuration of Learning Device 5> A control unit 50 of the learning device 5 functions as an additional learning unit 501 by executing a program stored in a storage unit 51.

[0045] <Additional Learning Unit 501> The additional learning unit 501 performs additional learning of the first learning model M1 using the learning data D3 stored in the storage unit 51. The first learning model M1 updated by the additional learning performed by the additional learning unit 501 may be transmitted to the base station 1 via the server 3, for example.

[0046] <Modified Example> In the above embodiment, when the imaging data received from the autonomous driving vehicle 2 is input to the first learning model M1, the first learning model M1 outputs a recognition result of the external environment of the autonomous driving vehicle 2 that is the transmission source of the imaging data. However, the output result of the first learning model M1 is not limited to only the recognition result of the external environment of the autonomous driving vehicle 2 that is the transmission source of the imaging data. For example, the first learning model M1 may output a control value for the autonomous driving vehicle 2 that is the transmission source of the imaging data.

[0047] In the above embodiment, the additional learning of the first learning model M1 is performed by the learning device 5. The additional learning of the first learning model M1 may be performed by an information processing device other than the learning device 5. For example, the functions of the server 3 and the learning device 5 may be integrated into a single information processing device, and the additional learning may be performed by an information processing device having a control unit that functions as the data receiving unit 301, the approval acquisition unit 302, and the additional learning unit 501.

[0048] In the above embodiment, the developer of the first learning model M1 performs an annotation operation on the approved data D2 to generate learning data. However, the annotation operation for the approved data D2 may be performed by the control unit 50 of the learning device 5 based on the control value of the autonomous driving vehicle 2 included in the approved data D2 or the like.

[0049] In the above embodiment, the image processing unit 103 performs, on the imaging data acquired by the image acquisition unit 101, image processing that converts images such as a person's appearance included in the imaging data into images in which the person cannot be identified. However, the target of the image processing that makes the image unidentifiable by the image processing unit 103 is not limited to a person's appearance or the like. For example, the image processing unit 103 may further perform image processing that converts an image of a copyrighted work included in the imaging data acquired by the image acquisition unit 101 into an image in which the copyrighted work cannot be identified.

[0050] <Example of implementation by software> The functions of each device (hereinafter referred to as "device") that constitutes the additional learning system 100 are programs that cause a computer to function as the device, and these can be implemented by programs that cause a computer to function as each control block of the device (particularly each part included in the control units 10, 30, and 50).

[0051] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., memory) as hardware for executing the program. By executing the program using this control device and storage device, the functions described in each of the embodiments are realized.

[0052] The above program may be recorded on one or more computer-readable recording media, not temporary ones. These recording media may or may not be provided by the above device. In the latter case, the program may be supplied to the above device via any wired or wireless transmission medium.

[0053] Furthermore, some or all of the functions of each of the above control blocks can also be implemented by logic circuits. For example, an integrated circuit in which logic circuits functioning as each of the above control blocks are formed is also included in the scope of this disclosure. In addition, it is also possible to implement the functions of each of the above control blocks by, for example, a quantum computer.

[0054] Furthermore, each process described in the above embodiments may be performed by AI (Artificial Intelligence). In this case, the AI ​​may operate on the control device described above, or it may operate on another device (for example, an edge computer or a cloud server).

[0055] <Summary> This disclosure contains at least the following aspects:

[0056] The additional learning method according to Embodiment 1 of this disclosure involves a base station, which is configured to communicate with the autonomous vehicle, inputting image data of the external environment of the autonomous vehicle into a learning model; the base station acquiring control values ​​of the autonomous vehicle based on the output of the learning model; accepting an approval operation from the user of the autonomous vehicle regarding the use of unapproved data, which includes image data based on the image data and the control values ​​acquired based on the image data, for additional learning of the learning model; and updating the learning model by additionally learning with approved data, which is obtained by adding information indicating that the unapproved data has been approved by the user. According to the above configuration, the additional learning method according to Embodiment 1 of this disclosure can easily ensure the transparency of the learning data used for additional learning of the learning model related to the control of the autonomous vehicle. As a result, it is possible to secure sufficient learning data with little bias, and thus the estimation accuracy of the learning model can be improved.

[0057] The additional learning method according to aspect 2 of this disclosure is characterized in that, in aspect 1, the base station performs image processing to convert the image of a person included in the imaging data into an image in which the person cannot be identified, and the unauthorized data includes the data obtained by applying the image processing to the imaging data and the control values ​​obtained based on the imaging data. With this configuration, the additional learning method according to aspect 2 of this disclosure can take into consideration the privacy rights of the person by converting an image that can be used to identify a person included in the imaging data into an image that cannot be used to identify a person through image processing.

[0058] The additional learning method according to aspect 3 of the present disclosure, in aspect 1 or 2, wherein the unapproved data is stored in an information processing device that can communicate with the base station, the information processing device presents to the user the unapproved data based on imaging data of the external environment of the user's autonomous vehicle from among the unapproved data stored in the information processing device, and accepts the approval operation for the unapproved data presented to the user. According to the above configuration, the additional learning method according to aspect 3 of the present disclosure makes it possible to obtain the user's approval for use in additional learning of the learning model in an objectively explainable manner.

[0059] The additional learning system according to aspect 4 of this disclosure is an additional learning system comprising the base station, the autonomous vehicle, and the information processing device according to aspect 3, wherein the base station constitutes a mobile communication network. According to the above configuration, the additional learning system according to aspect 4 of this disclosure has a base station that constitutes a mobile communication network. Base stations that constitute a mobile communication network have been deployed in various locations for some time. By using base stations for mobile communication networks that have been deployed for some time to collect data for additional learning of the learning model, learning data can be collected efficiently and transparently with little capital investment. Furthermore, by controlling autonomous vehicles based on base stations that constitute a mobile communication network that have been deployed for some time, transportation infrastructure for autonomous vehicles can be deployed efficiently with little capital investment.

[0060] Each aspect of this disclosure contributes to the construction of resilient transportation infrastructure and thus contributes to achieving Sustainable Development Goal (SDG) 9, "Build resilient infrastructure, promote inclusive and sustainable industrialization and foster innovation."

[0061] (Additional Notes) This disclosure is not limited to the embodiments described above, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of this disclosure.

[0062] 1 Base station 2 Autonomous vehicle 3 Server 4 Information terminal 5 Learning device 100 Additional learning system 101 Image acquisition unit 102 Driving control unit 103 Image processing unit 104 Data transmission unit 301 Data reception unit 302 Approval acquisition unit 501 Additional learning unit D1 Unapproved data D2 Approved data D3 Learning data M1 First learning model

Claims

1. An additional learning method comprising: inputting image data of the external environment of an autonomous vehicle into a learning model at a base station that is communicable with the autonomous vehicle; acquiring control values ​​of the autonomous vehicle based on the output of the learning model at the base station; receiving an approval operation from the user of the autonomous vehicle to use unapproved data, which includes image data based on the image data and the control values ​​acquired based on the image data, for additional learning of the learning model; and updating the learning model by additionally learning with approved data, which is obtained by adding information indicating that the unapproved data has been approved by the user.

2. The additional learning method according to claim 1, wherein the base station performs image processing to convert the image of a person included in the imaging data into an image in which the person cannot be identified, and the unauthorized data includes the data obtained by applying the image processing to the imaging data and the control value obtained based on the imaging data.

3. The additional learning method according to claim 1, wherein the unapproved data is stored in an information processing device that can communicate with the base station, the information processing device presents to the user the unapproved data based on imaging data of the external environment of the user's autonomous vehicle from among the unapproved data stored in the information processing device, and accepts the approval operation for the unapproved data presented to the user.

4. An additional learning system comprising the base station, the autonomous vehicle, and the information processing device according to claim 3, wherein the base station constitutes a mobile communication network.

Images