Method and system for managing automatic driving function in vehicle
By sending update data to vehicles and collecting verification data through the management server, the autonomous driving control function of the updated trained model is verified. This solves the problem of difficulty in verifying the appropriateness of the model after the update in the existing technology and improves the reliability of the autonomous driving function.
Patent Information
- Application Number
- CN202410567616.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-05-23
- Filing Date
- 2024-05-09
- Publication Date
- 2024-11-26
AI Technical Summary
Existing technologies have failed to effectively verify the appropriateness of using updated, fully trained models in autonomous driving control, making it difficult to guarantee the reliability of autonomous driving functions.
The system sends update data to at least two vehicles via a management server, collects verification data, and verifies the autonomous driving control function of the updated, trained model based on this data, ensuring that the model conforms to the update intent.
This enabled the verification of the autonomous driving function of the updated trained model, improved the reliability of the autonomous driving function, and ensured that the model could still be used appropriately after the update.
Smart Images

Figure CN119025404A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to methods and systems for managing autonomous driving functions in a vehicle. Background Art
[0002] Japanese Patent Publication No. 2015-135552 discloses a system for updating parameters of image recognition processing respectively performed in multiple vehicles. The existing system includes a server that communicates with multiple vehicles. The server collects image data respectively taken by multiple vehicles as learning data, and uses the learning data to perform machine learning processing. In the machine learning processing, data for updating the parameters of the image recognition processing is generated. The server also provides the data generated by the machine learning processing to multiple vehicles as the collection source of the learning data. Multiple vehicles use the data provided from the server to update the parameters of the image recognition processing.
[0003] As documents showing the technical level of the technical field related to the present disclosure, in addition to Japanese Patent Application Publication No. 2015-135552, Japanese Patent Application Publication No. 2022-002118, Japanese Patent Application Publication No. 2021-532487, Japanese Patent Application Publication No. 2018-195184, and Japanese Patent Application Publication No. 2019-156171 can be exemplified. Prior art literature Patent Literature
[0004] Patent Document 1: Japanese Patent Application Publication No. 2015-135552 Patent Document 2: Japanese Patent Application Publication No. 2022-002118 Patent Document 3: Japanese Patent Application Publication No. 2021-532487 Patent Document 4: Japanese Patent Application Publication No. 2018-195184 Patent Document 5: Japanese Patent Application Publication No. 2019-156171 Summary of the invention Technical problem to be solved by the invention
[0005] Image recognition processing performed in a vehicle is useful for autonomous driving control of the vehicle. In addition to image recognition processing, processing for "path planning" and "action planning" for the vehicle are also exemplified as processing useful for autonomous driving control. In recent years, models generated by machine learning (hereinafter also referred to as "trained models") are sometimes used in such processing.
[0006] Consider a case where a model building element such as a trained model or a parameter of a trained model has been updated. In this case, it is desirable to verify whether processing using the updated trained model is properly performed in autonomous driving control. For example, it is desirable to perform the verification immediately after the trained model is updated. Alternatively, it is desirable to perform the verification periodically after the trained model is updated.
[0007] However, existing documents, such as Japanese Patent Application Publication No. 2015-135552, do not focus on the verification of the trained model. Therefore, it can be said that there is still room for development of technology that can verify whether computer processing using an updated trained model is properly performed in autonomous driving control.
[0008] One object of the present disclosure is to provide a technology that can verify when a model generated by machine learning has been updated in a vehicle that uses an autonomous driving control model. Means for solving technical problems
[0009] A first aspect of the present disclosure is a method of managing an autonomous driving function in a vehicle having the following features. The method comprises: The management server transmits update data indicating data for updating the trained model to at least two vehicles on which autonomous driving control is performed, the autonomous driving control using the trained model indicating a model generated by machine learning; The management server collects verification data from each of the at least two vehicles to which the update data is sent, the verification data indicating data acquired or generated in association with execution of the automatic driving control using the updated trained model; and The management server uses the verification data collected from the at least two vehicles to verify the function of the automatic driving control using the updated trained model.
[0010] A second aspect of the present disclosure is a system for managing an autonomous driving function in a vehicle having the following features. The system includes a management server that manages the autonomous driving function. The management server performs the following processing: a process for transmitting update data indicating data for updating a trained model to at least two vehicles on which autonomous driving control is executed, the autonomous driving control using the trained model indicating a model generated by machine learning; a process of collecting verification data from each of the at least two vehicles to which the update data is transmitted, the verification data indicating data acquired or generated in association with execution of the automatic driving control using the updated trained model; and Using the verification data collected from the at least two vehicles, processing of a function of the autonomous driving control using the updated trained model is verified. Effects of the Invention
[0011] According to the present disclosure, verification data is collected from at least two vehicles to which update data is sent. In addition, the verification data is used to verify the function of the autonomous driving control. Therefore, it is possible to determine whether the updated trained model meets the intention of the update. Therefore, for example, when it is determined that the intention of the update is not met, the following countermeasures can be sought: generate data for correcting the updated trained model or model building elements, and update the updated trained model again. In this way, according to the present disclosure, it can be confirmed that the autonomous driving function using the updated trained model performs the function that meets the intention of the update. This helps to improve the reliability of the autonomous driving function using the trained model. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] Figure 1 This is a diagram for explaining the outline of the embodiment. Figure 2 This is a block diagram showing an example of the configuration of an automatic driving system mounted on a vehicle. Figure 3 This is a block diagram showing an example of the functional configuration of a management server. Figure 4 It is a diagram showing an example of instruction data related to the management of log data. Figure 5 This is a diagram for explaining the collection time for extracting and saving log data. Figure 6 This is a flowchart showing the flow of processing performed by the autonomous driving system that is particularly relevant to the embodiment. Figure 7 This is a flowchart showing the flow of processing performed by the management server that is particularly relevant to the embodiment. DETAILED DESCRIPTION
[0013] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. In addition, in each figure, the same or corresponding parts are denoted by the same reference numerals, and their description is simplified or omitted.
[0014] 1. Overview of implementation methods Figure 1 This is a diagram for explaining the outline of this embodiment. Figure 13 vehicles 1 and a management server 200 are depicted. The 3 vehicles 1 are an example of "at least 2 vehicles" in the present disclosure. Each vehicle 1 (hereinafter also referred to as "management object 1") is equipped with an automatic driving system 100 for performing automatic driving control of the management object 1. The so-called automatic driving refers to automatically performing at least one of the steering, acceleration and deceleration of the vehicle regardless of the driving operation of the operator. Automatic driving control is a concept that includes not only fully automatic driving control, but also risk avoidance control, lane keeping assist control, etc. The operator can be a driver who is on the management object 1, or a remote operator who remotely operates the management object 1.
[0015] The autonomous driving system 100 includes one or more processors 110 (hereinafter referred to as "processor 110") and one or more storage devices 120 (hereinafter referred to as "storage device 120"). The processor 110 performs various processes. As the processor 110, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), etc. can be exemplified. The storage device 120 stores various information. As the storage device 120, an HDD (Hard Disk Drive), an SSD (Solid State Drive), a volatile memory, a non-volatile memory, etc. can be exemplified.
[0016] The management server 200 and the autonomous driving system 100 together constitute the management system of the present embodiment. The management server 200 manages the functions of the autonomous driving system 100, that is, the autonomous driving functions. The management server 200 includes one or more processors 210 (hereinafter referred to as "processors 210") and one or more storage devices 220 (hereinafter referred to as "storage devices 220"). The processor 210 performs various processes. Examples of the processor 210 are the same as the examples of the processor 110 described above. The storage device 220 stores various information. Examples of the storage device 220 are the same as the examples of the storage device 120 described above.
[0017] The management server 200 communicates with each autonomous driving system 100. In the communication with the autonomous driving system 100, the management server 200 sends update data UPD and instruction data INS_LOG to each autonomous driving system 100. The update data UPD is data for updating the trained model. The update data UPD includes model building elements such as the trained model itself and the parameters of the trained model. The trained model is included in the function of the autonomous driving control unit possessed by the autonomous driving system 100. The function of the autonomous driving control unit will be described later. The instruction data INS_LOG is data including instructions related to the management of the log data LOG in the autonomous driving system 100. The instruction data INS_LOG will also be described later.
[0018] When the trained model has been updated in the autonomous driving system 100, it is desired to verify whether the processing using the updated trained model is properly performed in the autonomous driving control. Therefore, in the present embodiment, the management server 200 collects the verification data LOG_VRF from each autonomous driving system 100. In addition, the management server 200 performs verification based on the collected verification data LOG_VRF. The verification data LOG_VRF is the log data LOG obtained or generated in the processing related to the autonomous driving control using the updated trained model.
[0019] By using the verification data LOG_VRF for verification, it is possible to determine whether the updated trained model meets the intention of the update. Therefore, when it is determined that the intention of the update is not met, the following countermeasures can be sought: data for correcting the updated trained model or model building elements is generated, and the updated trained model is updated again. In this way, according to this embodiment, it is possible to confirm that the autonomous driving function using the updated trained model performs the function that meets the intention of the update. This helps to improve the reliability of the autonomous driving function using the trained model.
[0020] 2. Example of management system configuration 2-1. Example of autonomous driving system configuration Hereinafter, this embodiment will be described in more detail. Figure 2 It shows that they are respectively installed on Figure 1 FIG. 1 is a block diagram of an example configuration of an automatic driving system 100 for a vehicle 1 shown in FIG. Figure 2 In the illustrated example, the automatic driving system 100 includes a sensor group 10 , a recognition unit 20 , a planning unit 30 , a control amount calculation unit 40 , and a driving device 50 .
[0021] The sensor group 10 includes an identification sensor 11 for identifying the conditions around the management object 1. Examples of the identification sensor 11 include a camera, LIDAR (Laser Imaging Detection and Ranging), and a radar. The sensor group 10 may also include a state sensor 12 for detecting the state of the management object 1, a position sensor 13 for detecting the position of the management object 1, and the like. Examples of the state sensor 12 include a velocity sensor, an acceleration sensor, a yaw rate sensor, and a steering angle sensor. Examples of the position sensor 13 include a GNSS (Global Navigation Satellite System) sensor.
[0022] The sensor detection information SEN is information obtained by the sensor group 10. For example, the sensor detection information SEN includes an image captured by a camera. As another example, the sensor detection information SEN may include point cloud information obtained by LIDAR. The sensor detection information SEN may also include vehicle status information showing the status of the management object 1. The sensor detection information SEN may also include location information showing the location of the management object 1.
[0023] The recognition unit 20 receives the sensor detection information SEN. The recognition unit 20 recognizes the surrounding conditions of the management object 1 based on the information obtained by the recognition sensor 11. For example, the recognition unit 20 recognizes objects around the management object 1. Examples of objects include pedestrians, other vehicles (preceding vehicles, parked vehicles, etc.), white lines, road structures (e.g., guardrails, curbs), fallen objects, traffic lights, intersections, signs, etc. The recognition result information RES shows the recognition result of the recognition unit 20. For example, the recognition result information RES includes object information showing the relative position and relative speed of the object relative to the management object 1.
[0024] The planner 30 receives the recognition result information RES from the recognition unit 20. In addition, the planner 30 may also receive vehicle status information, position information, and pre-generated map information. The map information may be high-precision three-dimensional map information. The planner 30 generates a driving plan for the management object 1 based on the received information. The driving plan can be used to reach a preset destination or to avoid risks. Examples of driving plans include maintaining the current driving lane, changing lanes, overtaking, turning left or right, steering, accelerating, decelerating, and stopping. In addition, the planner 30 generates a target trajectory TRJ required for the management object 1 to travel according to the driving plan. The target trajectory TRJ includes a target position and a target speed.
[0025] The control amount calculation unit 40 receives the target trajectory TRJ from the planning unit 30. The control amount calculation unit 40 calculates the control amount CON required to make the management object 1 follow the target trajectory TRJ. The control amount CON can also be said to be a control amount requested to reduce the deviation between the management object 1 and the target trajectory TRJ. The control amount CON includes at least one of a steering control amount, a drive control amount, and a braking control amount. As the steering control amount, a target steering angle, a target torque, a target motor angle, a target motor drive current, etc. can be exemplified. As the drive control amount, a target speed, a target acceleration, etc. can be exemplified. As the braking control amount, a target speed, a target deceleration, etc. can be exemplified.
[0026] The travel device 50 includes a steering device 51, a drive device 52 and a brake device 53. The steering device 51 steers the wheels. For example, the steering device 51 includes an electric power steering (EPS: Electric Power Steering) device. The drive device 52 is a power source that generates a driving force. As the drive device 52, an engine, an electric motor, a hub motor, etc. can be exemplified. The brake device 53 generates a braking force. The travel device 50 receives the control quantity CON from the control quantity calculation unit 40. The travel device 50 respectively activates the steering device 51, the drive device 52 and the brake device 53 according to the steering control quantity, the drive control quantity and the brake control quantity. As a result, the management object 1 travels in a manner of following the target trajectory TRJ.
[0027] The recognition unit 20 includes at least one of a rule-based model and a trained model. The rule-based model performs recognition processing based on a predetermined rule group. As a trained model, NN (Neural Network), SVM (Support Vector Machine), regression model, decision tree model, etc. can be exemplified. NN can be CNN (Convolutional Neural Network), RNN (Recurrent Neural Network) or a combination thereof. The type, number of layers, and number of nodes of each layer in NN are arbitrary. A model generated or updated in advance by machine learning is applied to the trained model. The recognition unit 20 performs recognition processing by inputting sensor detection information SEN into the model. Recognition result information RES is output from the model or generated based on the output from the model.
[0028] The planning unit 30 also includes at least one of a rule-based model and a trained model. The planning unit 30 performs planning processing by inputting recognition result information RES to the model. The target trajectory TRJ is output from the model or generated based on the output from the model.
[0029] The control amount calculation unit 40 also includes at least one of a rule-based model and a trained model. The control amount calculation unit 40 performs control amount calculation processing by inputting the target trajectory TRJ to the model. The control amount CON is output from the model or generated based on the output from the model.
[0030] Two or more of the recognition unit 20, the planning unit 30, and the control amount calculation unit 40 may also be configured as an integral unit. The recognition unit 20, the planning unit 30, and the control amount calculation unit 40 may all be configured as an integral unit (End-to-End structure). For example, the recognition unit 20 and the planning unit 30 may be configured as an integral unit by a NN that outputs a target trajectory TRJ from the sensor detection information SEN. Even in the case of an integral unit, an intermediate product such as the recognition result information RES and the target trajectory TRJ may be output. For example, in the case where the recognition unit 20 and the planning unit 30 are configured as an integral unit by a NN, the recognition result information RES may be the output of an intermediate layer of the NN.
[0031] The recognition unit 20, the planning unit 30, and the control amount calculation unit 40 constitute an "autonomous driving control unit" that controls the autonomous driving of the management object 1. In this embodiment, a trained model is used in at least a part of the autonomous driving control unit. That is, at least one of the recognition unit 20, the planning unit 30, and the control amount calculation unit 40 includes a trained model. The autonomous driving control unit uses the trained model to perform at least a part of the autonomous driving control of the management object 1.
[0032] 2-2. Example of Management Server Configuration Figure 3 It is shown Figure 1 FIG. 2 is a block diagram showing an example of a functional configuration of the management server 200 shown in FIG. Figure 3 In the example shown, the management server 200 includes a data receiving unit 230, a model generating unit 240, a data transmitting unit 250, a function verifying unit 260, and an instruction setting unit 270. The functions of the model generating unit 240, the function verifying unit 260, etc. are provided by, for example, Figure 1 This is achieved by the processor 210 executing a predetermined program stored in the storage device 220 .
[0033] The data receiving unit 230 has a function as an interface for receiving various data from outside the management server 200. The various data received by the data receiving unit 230 include the verification data LOG_VRF. The verification data LOG_VRF can be exemplified as: Figure 2 The sensor detection information SEN, control amount CON, recognition result information RES, target trajectory TRJ, etc. are shown. The verification data LOG_VRF may also include Figure 2The verification data LOG_VRF may also include the reason for the judgment in the recognition process of the recognition unit 20 shown. Figure 2 The reason for the judgment in the planning process of the planning unit 30 is shown. The verification data LOG_VRF may include whether the operator intervenes in the automatic driving control.
[0034] The model generation unit 240 generates data for updating the trained model currently being applied to the autonomous driving system 100 (i.e., update data UPD). The update data UPD includes model construction elements such as the trained model itself and the parameters of the trained model. As examples of machine learning models for updating, the same models as the examples of machine learning models already applied to the autonomous driving system 100 can be cited. In the case where a trained model is generated, the trained model currently being applied to the autonomous driving system 100 is updated by being replaced with the generated trained model. In the case where a model construction element is generated, the trained model currently being applied to the autonomous driving system 100 is updated by correcting the model construction element.
[0035] The data transmission unit 250 has a function as an interface for transmitting various data from the management server 200 to the outside. The various data transmitted by the data transmission unit 250 include the update data UPD received from the model generation unit 240. The update data UPD includes the trained model and model building elements generated by the model generation unit 240. The various data transmitted by the data transmission unit 250 also include the indication data INS_LOG received from the indication setting unit 270.
[0036] The function verification unit 260 uses the verification data LOG_VRF to verify whether the processing related to the automatic driving control is properly performed using the updated trained model. Figure 2 The trained model is used in at least a part of the automatic driving control unit described in . Therefore, the automatic driving function using the updated trained model becomes the object of verification by the function verification unit 260.
[0037] The verification of the automatic driving function is performed after a prescribed period (e.g., 1 week, 1 month) has passed after the update data UPD is sent. Alternatively, the verification of the automatic driving function is performed after the driving distance of the management object 1 reaches a prescribed distance (e.g., 5 to 10 km) after the update data UPD is sent. In the verification of the automatic driving function, a pre-set verification index is used. As a verification index, the number of times the operator intervenes in the automatic driving control can be exemplified.
[0038] As already explained, the verification data LOG_VRF may include data on the operator's intervention in the automatic driving control. If the number of interventions in the automatic driving control is large within a specified period (or during driving of a specified distance), the operator may feel uncomfortable with the automatic driving function using the updated trained model. For example, when the number of interventions after this update is significantly increased compared to the number of interventions after the previous update, it can be expected that the operator will feel a strong discomfort with the trained model after this update.
[0039] The above-mentioned examples of verification indicators are examples of the case where one management object 1 (one operator) is considered. When more than two management objects 1 (more than two operators) are considered, the number of interventions (i.e., the average number of times) for each management object 1 can be used as a verification indicator. However, it is taken into account that the frequency of intervention in automatic driving control varies according to the operator's driving proficiency. Therefore, when more than two management objects 1 are considered, it is preferable to calculate the verification indicator by weighting corresponding to the driving proficiency. For example, assuming that the higher the proficiency of manual driving, the higher the frequency of intervention, consider an example of increasing the weight coefficient of the operator with low proficiency. Alternatively, assuming that the higher the proficiency of automatic driving, the lower the frequency of intervention, consider an example of reducing the weight coefficient of the operator with high proficiency.
[0040] The indication setting unit 270 sets the indication data INS_LOG. In the indication data INS_LOG, the predetermined event PE, etc., which is the collection target of the verification data LOG_VRF, among the events related to the behavior of the management object 1, is determined. The indication data INS_LOG is provided from the management server 200 to the management object 1 at a predetermined timing in a day (for example, the timing of the ignition ON of the management object 1). The original data of the indication data INS_LOG is stored in the storage device 220. The content of the original data is updated by the management server 200.
[0041] Figure 4 FIG. 2 is a diagram showing an example of the instruction data INS_LOG. Figure 4 In the example shown, the indication data INS_LOG includes items such as group ID, event ID, event name EN, priority PL, and event determination On / Off. In addition, the indication data INS_LOG may also include other items in addition to these items. As other items, a mode for acquiring the verification data LOG_VRF, and a list of location conditions (latitude and longitude), time conditions (morning, afternoon, late night), and weather conditions (sunny, cloudy, rainy) for acquiring the verification data LOG_VRF may be exemplified. The location, time, and weather conditions may set the conditions for collectively collecting the verification data LOG_VRF.
[0042] The group ID is an item for identifying the group to which the specified event PE belongs. Figure 4 In the example shown, GI1, GI2, and GIm (m is a natural number greater than or equal to 3) are set as group IDs. GI1 is, for example, the ID of a group related to the behavior control of the vehicle 1. GI2 is, for example, the ID of a group related to the external identification performed by the management object 1. GIm is, for example, the ID of a group related to the occurrence of abnormalities in various systems installed in the management object 1.
[0043] Event ID is an item that identifies a specified event PE. Figure 4 In the example shown, EI1 to E15 are set as event IDs of GI1. EI1 is, for example, an ID indicating detection of emergency acceleration or emergency deceleration of management object 1. EI2 is, for example, an ID indicating detection of a driving change request (TD: Transition Demand, takeover request) by the operator of management object 1. EI3 is, for example, an ID indicating detection of an emergency turn of management object 1. EI4 is, for example, an ID indicating detection of a behavior of management object 1 that is different from that of surrounding vehicles. EI5 is, for example, an ID indicating detection of a short distance between the preceding vehicle and the following vehicle.
[0044] exist Figure 4 In the example shown, EI6 to E18 are also set as event IDs of GI2. EI6 is, for example, an ID indicating misrecognition. EI7 is, for example, an ID indicating detection of clouds. EI8 is, for example, an ID indicating detection of raindrops. EIn (n is a natural number greater than or equal to 3) is set as the event ID of GIm. EIn is, for example, an ID indicating detection of an abnormal signal.
[0045] The event name EN is an item that directly indicates the content of the specified event PE. Figure 4 In the example shown, EN1 to ENn are set as event names EN. EN1 to ENn correspond to EI1 to EIn, respectively.
[0046] The priority level PL is an item indicating the priority of collecting the verification data LOG_VRF to the storage device 120 when a predetermined event PE occurs. The priority level PL is also an item indicating the priority of sending the verification data LOG_VRF to the management server 200. Figure 4 In the example shown, the priority levels PL are divided into five levels, 1 to 5. The larger the number of the priority level PL, the higher the priority of collecting the verification data LOG_VRF of the predetermined event PE. Also, the larger the number of the priority level PL, the higher the priority of sending the verification data LOG_VRF.
[0047] The event judgment On / Off is an item that shows whether the specified event PE is processed as the "collection target event CTE". The so-called collection target event STE refers to an event in which the ongoing log data LOG in the specified event PE is collected as the verification data LOG_VRF in the storage device 21. The specified event PE with the event judgment "On" is processed as the collection target event CTE. The specified event PE with the event judgment "Off" is not processed as the collection target event CTE. In other words, the ongoing log data LOG of the specified event PE with the event judgment "On" is collected in the storage device 21 as the verification data LOG_VRF, and the log data LOG that is not the case is not collected in the storage device 21.
[0048] Here, refer to Figure 5 The ongoing log data LOG of the specified event PE is described. Figure 5 In the example shown, a predetermined event PE with event ID = EI3 occurs at time T1. Also, a predetermined event PE with event ID = EI1 occurs at time T2. In addition, both the predetermined event PE with event ID = EI1 and event ID = EI3 correspond to predetermined event PEs with event determination of "On" (i.e., collection target events CTE) (see Figure 4 ).
[0049] When a predetermined event PE occurring at a certain time corresponds to the collection target event CTE, log data of a predetermined collection time CT is extracted from the log data LOG of the collection target event CTE. The extracted log data LOG is stored in the storage device 120. The log data LOG corresponds to the verification data LOG_VRF. Figure 5 The upper part of depicts log data LOG including sensor detection information SEN, control amount CON, recognition result information RES, and target trajectory TRJ.
[0050] The collection time CT for extracting and saving the log data LOG is set based on the time when the collection target event CTE is determined to have occurred. Figure 5 In the example shown, it is determined that two types of collection target events CTE occurred at time T1 and T2. The collection time CT1 for the first collection target event CTE (event ID = EI3) includes the time period Z1 before time T1 and the time period Z2 after time T1. In addition, the collection time CT2 for the second collection target event CTE (event ID = EI1) includes the time period Z3 before time T2 and the time period Z4 after time T2.
[0051] The length of each time period Z1, Z2, Z3 and Z4 is, for example, 1 to 10 seconds. However, it sometimes takes time to determine whether the specified event PE is equivalent to the collection object event CTE. Therefore, it can be imagined that the time when the collection object event CTE is determined to have occurred is slightly later than the time when the collection object event CTE has occurred. Therefore, the length of each time period Z1, Z2, Z3 and Z4 (that is, the length of the collection time CT1 and CT2) is preferably set according to the content (event ID) of each collection object event CTE.
[0052] In addition, in the case of an operator intervention event in the automatic driving control, such as an operator's driving replacement request, it is expected that the operator will feel uncomfortable before the reference time. In addition, it is considered that the frequency of intervention in the automatic driving control varies according to the operator's driving proficiency. Therefore, in the case of an operator intervention event in the automatic driving control, it is preferable to set the collection time CT according to the driving proficiency. For example, assuming that the higher the proficiency of manual driving, the shorter the time to decide intervention, consider setting the reference time ( Figure 5 The time period before time T1 or T2) ( Figure 5 Alternatively, assuming that the higher the proficiency of the autonomous driving, the longer the intervention time is, consider an example in which the time period before the reference time is set to be long.
[0053] 3. Example of processing performed by the processor 3-1. Processing Example in the Automatic Driving System 100 Figure 6 1 is a flowchart showing the flow of processing performed by processor 110 that is particularly relevant to the embodiment. Figure 6 The illustrated routine is performed, for example, immediately after the automated driving system 100 is started.
[0054] exist Figure 6 In the routine shown, first, the startup data IGN is sent to the management server 200 (step S11). The startup data IGN is data for notifying the management server 200 of the ignition ON of the management object 1. The startup data IGN includes, for example, identification data of the management object 1, version data of the automatic driving system 100, etc. The version data also includes version data of the trained model being applied to the automatic driving system 100. The startup data IGN may also include verification data LOG_VRF that is not sent to the management server 200.
[0055] After the process of step S11, it is determined whether the confirmation signal SCF is received (step S12). The confirmation signal SCF is a signal notifying the managed object 1 that the management server 200 has confirmed the startup data IGN. The process of step S12 is repeated until an affirmative determination result is obtained.
[0056] If the result of the determination in step S12 is positive, it is determined whether update data UPD is transmitted (step S13). When the trained model currently applied to the automatic driving system 100 is updated, a signal notifying the existence of transmission of update data UPD is added to the confirmation signal SCF received in the process of step S12. Therefore, when the additional signal is recognized, it is determined that update data UPD exists.
[0057] If the result of the determination in step S13 is positive, the update data UPD is updated (step S14). In addition, when the process of step S14 is performed, it is preferable to notify the operator of the management object 1 through the user interface that part of the automatic driving control of the management object 1 may not be performed before the update of the update data UPD is completed.
[0058] After the processing of step S14, a completion signal SUP is sent to the management server 200, and counting of the verification distance LVRF is started (step S15). The completion signal SUP is a signal that notifies the management server 200 that the update of the update data UPD is completed. That is, the completion signal SUP is generated and sent to the management server 200 when the update of the update data UPD in the management object 1 is completed. The verification distance LVRF is set to collect the verification data LOG_VRF.
[0059] Following the processing of step S15, it is determined whether the verification distance LVRF is less than the threshold value LTH (step S16). The threshold value LTH corresponds to the above-mentioned "prescribed driving distance (for example, 5 to 10 km) of the management object 1 after the transmission of the update data UPD". If the determination result of step S16 is negative, that is, the verification distance LVRF exceeds the threshold value LTH, the collection of the update data UPD is completed.
[0060] If the result of the determination in step S16 is positive, it is determined whether the collection condition is satisfied (step S17). Whether the collection condition is satisfied can be determined based on whether the collection target event CTE has occurred. As already explained, the collection target event CTE is an event in the specified event PE that saves the ongoing log data LOG to the storage device 21. Therefore, it is determined whether the specified event PE has occurred. If it is determined that the specified event PE has occurred, refer to Figure 4Indication data INS_LOG described in . Thereby, it can be determined whether the collection target event CTE has occurred.
[0061] If the result of the determination in step S17 is negative, the process returns to step S16. On the other hand, if the result of the determination in step S17 is positive, the log data LOG is stored in the storage device 120 (step S18). In the process of step S18, first, based on the time when the collection target event CTE is determined to have occurred, the log data of the specified period is extracted from the log data LOG of the collection target event CTE. The concept of the specified period is as follows: Figure 5 As described above. Then, the extracted log data LOG is stored in the storage device 120. The log data LOG corresponds to the verification data LOG_VRF described above. The verification data LOG_VRF stored in the storage device 120 is transmitted to the management server 200 as appropriate.
[0062] After the processing of step S18, it is determined whether the verification distance LVRF exceeds the threshold value LTH (step S19). The determination content of step S19 is opposite to the determination content of step S16. If the determination result of step S19 is negative, that is, if the verification distance LVRF is less than the threshold value LTH, the processing after step S17 is performed. If the determination result of step S19 is positive, the collection of the update data UPD is completed. In this way, when there is an update of the update data UPD, the processing of steps S16 to S19 is repeated to collect the update data UPD.
[0063] 3-2. Example of processing in the management server 200 Figure 7 : is a flowchart showing the flow of processing performed by processor 210 that is particularly relevant to the embodiment. Figure 7 The routine shown is repeatedly executed in a predetermined control cycle.
[0064] exist Figure 7 In the routine shown, first, it is determined whether the startup data IGN is received from the management object 1 (step S21). The startup data IGN is as described above. If the startup data IGN includes the verification data LOG_VRF, the verification data LOG_VRF is appropriately stored in the storage device 220.
[0065] After the processing of step S21, it is determined whether the update data UPD needs to be sent (step S22). In the processing of step S22, for example, based on the startup data IGN (version data of the trained model) received in the processing of step S21, it is determined whether the version upgrade of the trained model is needed.
[0066] If the determination result of step S22 is negative, a confirmation signal SCF is sent to the management object 1 (step S23). The confirmation signal SCF is as described above. On the other hand, if the determination result of step S22 is positive, in addition to sending the confirmation signal SCF, update data UPD and indication data INS_LOG are sent to the management object 1 (step S24). The indication data INS_LOG is as described above.
[0067] After the process of step S24, it is determined whether a completion signal SUP is received from the management object 1 (step S25). The completion signal SUP is as described above. The process of step S25 is repeated until a positive determination result is obtained.
[0068] If the result of determination in step S25 is affirmative, counting of the verification period PVRF is started (step S26). The verification period PVRF is set to collect the verification data LOG_VRF.
[0069] Following the process of step S26, it is determined whether the verification condition is satisfied (step S27). The verification condition includes the following conditions, for example. (i) During the verification period, PVRF is greater than the threshold value PTH. (ii) A valid amount of verification data LOG_VRF is obtained. (iii) Completion of traveling the distance LVRF is verified in at least two managed objects 1 . The threshold value PTH corresponds to the above-mentioned "predetermined period (for example, one week or one month) after the transmission of the update data UPD".
[0070] For example, when conditions (i) and (ii) are met, it is determined that the verification condition is met. In another example, when all conditions (i) to (iii) are met, it is determined that the verification condition is met. When the result of the determination in step S27 is affirmative, the verification of the autonomous driving function is performed (step S28). The verification of the autonomous driving function is performed using a verification indicator set in advance. Examples of verification and examples of verification indicators are as described above. [Explanation of Reference Numerals]
[0071] 1: vehicle (management object), 10: sensor group, 20: identification unit, 30: planning unit, 40: control amount calculation unit, 50: driving device, 100: automatic driving system, 110, 210: processor, 120, 220: storage device, 200: management server, CON: control amount, INS_LOG: instruction, LOG: log data, RES: identification result information, SEN: sensor detection information, TRJ: target trajectory, UPD: update data, LOG_VRF: verification data, CT1, CT2: collection time.
Claims
1. A method for managing an automatic driving function, which manages an automatic driving function in a vehicle, characterized in that: include: The management server transmits update data indicating data for updating the trained model to at least two vehicles on which autonomous driving control is performed, the autonomous driving control using the trained model indicating a model generated by machine learning; The management server collects verification data from each of the at least two vehicles to which the update data is sent, the verification data indicating data acquired or generated in association with execution of the automatic driving control using the updated trained model; and The management server uses the verification data collected from the at least two vehicles to verify the function of the automatic driving control using the updated trained model.
2. The method for managing the automatic driving function according to claim 1, characterized in that: Also includes: A step of collecting, when the autonomous driving control using the updated trained model is in progress and a prescribed collection condition is satisfied, data acquired or generated by the management object in association with the execution of the autonomous driving control within a prescribed collection time based on a timing at which the collection condition is satisfied, as the verification data related to the management object, by a processor that executes the autonomous driving control of one management object included in the at least two vehicles; and The management object sends the verification data related to the management object to the management server.
3. The method for managing the automatic driving function according to claim 2, characterized in that: The collection conditions include: the existence of an operator of the management object intervening in the automatic driving control, The method for managing an automatic driving function further includes a step in which the management object variably sets the length of the collection time based on a timing at which the collection condition is satisfied, according to the driving proficiency of an operator of the management object.
4. The method for managing the automatic driving function according to claim 2 or 3, characterized in that: The collection conditions include: the driving environment conditions of the management object are consistent with the pre-set conditions.
5. The method for managing the automatic driving function according to claim 2, characterized in that: The collection conditions include: the existence of an operator of the management object intervening in the automatic driving control, In the step of verifying the autonomous driving function, the number of interventions in the autonomous driving control obtained during a prescribed verification period after the update data is sent, or during a period from when the driving distance of the management object after the update data is sent to when the driving distance reaches a prescribed verification distance, is used as a verification indicator.
6. The method for managing the automatic driving function according to claim 5, characterized in that: In a case where intervention of the operators of the management objects in the automatic driving control exists in at least two of the management objects, in the step of verifying the automatic driving function, the number of interventions in the automatic driving control as the verification indicator is calculated by weighting corresponding to the driving proficiency of the operators of the at least two management objects.
7. A management system for an automatic driving function, which manages the automatic driving function in a vehicle, characterized in that: including a management server for managing the autonomous driving function, The management server performs the following processing: a process for transmitting update data indicating data for updating a trained model to at least two vehicles on which autonomous driving control is executed, the autonomous driving control using the trained model indicating a model generated by machine learning; a process of collecting verification data from each of the at least two vehicles to which the update data is transmitted, the verification data indicating data acquired or generated in association with execution of the automatic driving control using the updated trained model; and Using the verification data collected from the at least two vehicles, processing of a function of the autonomous driving control using the updated trained model is verified.
8. The management system for the automatic driving function according to claim 7, characterized in that: further comprising a processor for executing the autonomous driving control of one of the at least two vehicles to be managed, The processor performs the following processing: When the autonomous driving control using the updated trained model is in progress and a predetermined collection condition is satisfied, data acquired or generated by the management object in association with the execution of the autonomous driving control within a predetermined collection time based on a timing at which the collection condition is satisfied is collected as the verification data related to the management object; as well as The process of sending the verification data related to the management object to the management server.
Citation Information
Patent Citations
Learning system, on-vehicle device, and server
JP2015135552A
Vehicle device, safe driving support system, safe driving support method
JP2018195184A
Distracted driving prediction system
JP2021532487A
Information processor, server device, information processing method, and program
JP2022002118A
