Valuation data generation device, valuation data generation method and valuation data generation program

Abstract

A data generation device and the like will be provided, which can generate more units of data for evaluating the performance of a driver assistance system with less processing.An evaluation data generation device, which generates evaluation data for evaluating a vehicle's driver assistance function, includes a scenario setting unit that sets a scenario to simulate a journey of the vehicle, a sensor data generation unit that processes the scenario set in the scenario setting unit with a sensor model obtained by modeling a sensor of the vehicle and generates sensor data according to information detected by the sensor model, and a real data conversion unit that links the sensor data generated by the sensor data generation unit with real data according to data acquired by the vehicle's sensor and converts the sensor data based on the linked real data to generate converted real data according to the evaluation data.

Description

Area

[0001] The present disclosure relates to an evaluation data generation apparatus, an evaluation data generation method, and an evaluation data generation program. background

[0002] Driver assistance functions, such as automated driving and collision avoidance, have been developed to assist drivers operating vehicles. Driver assistance functions determine the assistance to be provided based on data collected by various types of sensors installed in the vehicle. Driver assistance functions must respond to the various environments of the moving vehicles.

[0003] In addition, evaluation methods have been proposed to assess the performance of driver assistance functions. According to Patent Document 1, the driving environments of several vehicles are simulated, the simulated information is input to each vehicle, and the driver assistance function is evaluated based on the behavior exhibited. Citation listPatent literature

[0004] Patent Document 1: Japanese Patent Application Laid-Open No. 2017-105453 SummaryTechnical problem

[0005] The device according to Patent Document 1 can perform analysis in various situations by implementing the analysis through simulation. However, since it is a simulation, there is a deviation from the data obtained from an actual vehicle drive. Therefore, there is a limitation in improving reliability with regard to its use in evaluating the driving assistance function.

[0006] The device according to Patent Document 1 can perform the analysis by simulating various conditions and situations. However, with regard to the evaluation data obtained by the simulation, it is difficult to increase the reliability of the data to a level suitable for evaluating the driving assistance performance due to the deviation from the data obtained when the vehicle is actually running.

[0007] However, if the vehicle is driven in real life, simulation data cannot be obtained, but rather real driving data. In this case, it is necessary to conduct tests under various conditions and situations to obtain the reliability suitable for evaluating the driver assistance system's performance. However, there are limitations to conducting driving tests under various conditions and situations when using real vehicles. Given this, in reality, it is difficult to obtain data suitable for evaluating the driver assistance system's performance from real tests in which the vehicle is driven.

[0008] To solve the above-described problem, an object of the present disclosure is to provide an evaluation data generation apparatus, an evaluation data generation method, and an evaluation data generation program that can efficiently generate evaluation data with reliability suitable for evaluating driving assistance performance. Solution to the problem

[0009] The present disclosure provides an evaluation data generation device that generates evaluation data for evaluating at least one of automated driving of a vehicle and a driving assistance function of the vehicle. The evaluation data generation device includes: a scenario setting unit that sets a scenario for simulating a drive of the vehicle; a sensor data generation unit that generates sensor data acquired when the vehicle drives in accordance with the scenario set by the scenario setting unit, using a sensor model obtained by modeling a sensor of the vehicle;and a real data conversion unit that combines the sensor data generated by the sensor data generation unit with real data corresponding to data acquired by the sensor of the vehicle, and converts the sensor data based on the combined real data to generate converted real data corresponding to the evaluation data;

[0010] The present disclosure provides an evaluation data generation method for generating evaluation data for evaluating at least one driving assistance function of automated driving of a vehicle and a driving assistance function of the vehicle, the evaluation data generation method including: a step of setting a scenario for simulating a travel of the vehicle; a step of generating, using a sensor model obtained by modeling a sensor of the vehicle, sensor data acquired when the vehicle travels in accordance with the scenario set by the scenario setting unit; and a step of linking the generated sensor data with real data corresponding to data acquired by the sensor of the vehicle, and converting the sensor data based on the linked real data to generate converted real data corresponding to the evaluation data.

[0011] The present disclosure provides an evaluation data generation program for generating evaluation data for evaluating at least one driving assistance function of automated driving of a vehicle and a driving assistance function of the vehicle, wherein the evaluation data generation program performs a process including: a step of setting a scenario for simulating a travel of the vehicle; a step of generating sensor data acquired when the vehicle travels in accordance with the set scenario using a sensor model obtained by modeling a sensor of the vehicle; and a step of linking the generated sensor data with real data corresponding to data acquired by the sensor of the vehicle and converting the sensor data based on the linked real data to generate converted real data corresponding to the evaluation data. Advantageous effects of the invention

[0012] The above-described design has the effect that more units of evaluation data used to evaluate the performance of a driver assistance system can be generated with less processing. Short description of the drawing Fig. 1 is a block diagram showing an example of an evaluation data generating apparatus. Fig. 2 is a flowchart illustrating an example of a process of the evaluation data generating device. Fig. 3 is a flowchart illustrating an example of the operation of a machine learning unit. Fig. 4 is a flowchart illustrating an example of the process of the evaluation data generating device. Fig. 5 is a flowchart illustrating an example of the process of the evaluation data generating device. Fig. 6 is a flowchart illustrating an example of the process of the evaluation data generating device. Fig. 7 is a flowchart illustrating an example of the process of the evaluation data generating device. Fig. 8 is a block diagram illustrating an example of an evaluation data generating apparatus. Fig. 9 is a flowchart illustrating an example of the operation of the machine learning unit. Fig. 10 is a flowchart illustrating an example of the process of the evaluation data generating device. Description of embodiments

[0013] An embodiment of the present disclosure will be described in detail below with reference to the drawings. Note that the present invention is not limited to this embodiment. Furthermore, the components of the following embodiment include those that can be easily replaced by those skilled in the art or equivalent. Furthermore, the components described below can be combined as needed. Where there is more than one embodiment, the embodiments can be combined. Evaluation data generation device

[0014] Fig. 1 is a block diagram illustrating an example of an evaluation data generation device. In the present embodiment, an evaluation data generation device 10 generates evaluation data used in evaluating driving assistance functions of an electronic control unit (ECU) 6.

[0015] The ECU 6 is installed in a vehicle and performs driving assistance functions based on information acquired by sensors installed in the vehicle. The driving assistance to be performed by the ECU 6 is not subject to any specific restrictions. Examples of driving assistance include controlling vehicle speed, controlling brakes, controlling steering, controlling automated driving, controlling driver warnings, and controlling warnings to other vehicles and pedestrians in the surrounding area. In other words, at least one of the automated driving function and the driving assistance function of the vehicle is the target.Sensors mounted in the vehicle relate to various types of information, including information acquired by a distance sensor such as a LiDAR, a temperature sensor, a camera, and the like; information acquired from surrounding vehicles, roadside facilities, and the cloud through communication; operational information about acceleration, braking, steering, and the like input to the vehicle; information about sensors arranged in a driving mechanism; and the like.

[0016] When evaluating the ECU 6, the evaluation data generated by the evaluation data generation device 10 is input to the ECU 6. When input as sensor data, the ECU 6 processes the input evaluation data and outputs the processing results to an evaluation device 8 as instruction information for driving assistance.

[0017] The evaluation device 8 evaluates the driving assistance functions of the ECU 6. The evaluation device 8 is a device having an arithmetic processing function such as a CPU and a storage function such as a ROM and a RAM, and performs an evaluation function with a computer program. The evaluation device 8 acquires the evaluation data input from the evaluation data generation device 10 and the conditions of the evaluation data, as well as the instruction information for driving assistance output after processing the evaluation data in the ECU 6. The evaluation device 8 determines whether the timing and content of the driving assistance performed by the ECU 6 are suitable for the situation of a scenario for the evaluation data. By evaluating the ECU 6 using the evaluation data, the driving assistance performance of the ECU 6 can be evaluated.Using the evaluation data, the ECU 6 is evaluated, and the functions of the ECU 6 are adjusted to improve the performance of the driver assistance system. The evaluation data can also be used as learning data for the ECU 6.

[0018] Next, the evaluation data generation device 10 will be described. The evaluation data generation device 10 simulates the driving state of the vehicle to be evaluated and generates the evaluation data that simulates the data to be input to the ECU 6 when the vehicle to be evaluated is running.

[0019] The evaluation data generation device 10 includes an input unit 12, an output unit 14, a communication unit 15, an arithmetic unit 16, and a storage unit 18. The input unit 12 includes input devices such as a keyboard and a mouse, a touch panel, and a microphone that receives speech from an operator, and outputs signals corresponding to the operations performed by the operator on the input device to the arithmetic unit 16. The output unit 14 includes a display device such as a display and displays a screen containing various types of information such as processing results and images for processing based on display signals output from the arithmetic unit 16. The output unit 14 may also include a recording device that outputs data to a recording medium.The communication unit 15 transmits data using a communication interface. The communication unit 15 communicates with the ECU 6 and the evaluation device 8 to transmit and receive data. The communication unit 15 transmits various types of data and computer programs obtained through communication with external devices to the storage unit 16 and stores the data and computer programs. The communication unit can be connected to the external device via a wired communication line or a wireless communication line.

[0020] The arithmetic unit 16 includes an integrated circuit (processor), such as a central processing unit (CPU) or a graphics processing unit (GPU), and a memory serving as a work area. It executes various processes by executing various computer programs using the hardware resources mentioned above. Specifically, the arithmetic unit 16 reads the computer program stored in the memory unit 18, transfers the computer program to a memory, and causes the processor to execute the instructions contained in the computer program transferred to the memory, thereby executing various processes. The arithmetic unit 16 includes a scenario setting unit 22, a sensor data generation unit 24, a real data conversion unit 26, a link processing unit 28, and a machine learning unit 30.Before describing the respective units in the arithmetic unit 16, the storage unit 18 is described.

[0021] The storage unit 18 includes a non-volatile storage device, such as a magnetic storage device or a semiconductor storage device, and stores various computer programs and data. The storage unit 18 includes a data generation program 32, a learning program 34, a scenario model 36, an environmental object model 37, a weather model 38, a light source model 40, a vehicle model 42, a sensor model 44, a link table 46, real data 48, and a learned model 50.

[0022] The data stored in the storage unit 18 includes the scenario model 36, the environmental object model 37, the weather model 38, the light source model 40, the vehicle model 42, the sensor model 44, the link table 46, the real data 48, and the learned model 50. Several units of the following are provided: the scenario model 36, the environmental object model 37, the weather model 38, the light source model 40, the vehicle model 42, the sensor model 44, the link table 46, the real data 48, and the learned model 50, wherein the models, data, and tables to be used can be selected.

[0023] The scenario model 36 is a model for setting the conditions for simulating the vehicle's travel. The scenario model 36 is a combination of models selected from the environmental object model 37, the weather model 38, the light source model 40, and the vehicle model 42. The scenario model 36 can be either a model for a specific moment without a time delay for evaluating the processing of the ECU 6 at a specific moment, or a model with a time delay for evaluating the processing of the ECU 6 during the vehicle's travel in a predetermined section.

[0024] The environment object model 37 is a model that contains information about the shapes surrounding the vehicle. The environment object model 37 includes information about the shape of the road the vehicle is traveling on (straight, curved, with an intersection, presence / absence of guardrails, presence / absence of traffic lights), as well as information about surrounding vehicles, surrounding pedestrians, and the like. In other words, the environment object model 37 contains a model of the three-dimensional shape surrounding the vehicle.

[0025] Weather Model 38 is a model that simulates the weather around the vehicle, for example, as sunny, cloudy, rainy, stormy, snowy, and sleet. Weather Model 38 is a model in which the simulated situation changes according to the amount of clouds, rainfall and snowfall, temperature, wind speed, and the like.

[0026] The light source model 40 is a model of the light emitted by objects around the vehicle, such as the sun, light poles, buildings and oncoming vehicles.

[0027] The vehicle model 42 is a model that simulates the vehicle being analyzed and the surrounding vehicles. The vehicle model 42 contains information about the vehicle's shape, driving performance, in-vehicle sensors, and the like.

[0028] Sensor model 44 is a model of the sensors mounted in the vehicle being analyzed. Sensor model 44 is a model in which the type of information output is adjusted according to the surrounding information.

[0029] The link table 46 is a model containing information about the link between sensor data generated during simulation and real data acquired by driving the vehicle in real life in advance. The link table 46 contains information about the link conditions. The link table 46 includes a table of the conditions for linking the real data when the real data is linked based on the scenario conditions or the sorting conditions set in advance based on the acquired sensor data. When the trained model 50 is used for the link, a processing condition for inputting the sensor data to the trained model 50 or the like is linked.

[0030] The real data 48 is data acquired by causing the vehicle to drive in real time. The real data 48 is data detected by the sensor mounted in the vehicle. The real data 48 also includes information about the vehicle driving in real time, information about surrounding objects, information about the weather, information about light sources, and information about the vehicle.

[0031] The trained model 50 is a model created by the machine learning unit 30 through machine learning. Upon input of the sensor data, the trained model 50 outputs the real data corresponding to the sensor data.

[0032] The computer programs to be stored in the storage unit 18 include the data generation program 32 and the learning program 34.

[0033] The data generation program 32 is a computer program for generating the evaluation data (evaluation data generation program). The data generation program 32 is a computer program that implements the functions of the scenario setting unit 22, the sensor data generation unit 24, the real data conversion unit 26, and the link processing unit 28.

[0034] The learning program 34 is a computer program that generates a trained model, which is used in a part of the process of the data generation program 32, through machine learning. The learning program 34 is a program that realizes the functions of the machine learning unit 30. The learning program 34 performs a deep learning process and generates the trained model 50 using teacher data, inputting the sensor data generated in a simulation based on a scenario, and outputting the real data corresponding to the sensor data. A generative adversarial network (GAN) or the like can be used as the deep learning model. The trained model and the machine learning method are not particularly limited as long as the sensor data can be converted into the real data.

[0035] In the storage unit 18, the data generation program 32 and the learning program 34 can be installed by reading the data generation program 32 and the learning program 34 recorded on the recording medium, or the data generation program 32 and the learning program 34 can be installed by reading the data generation program 32 and the learning program 34 provided via a network.

[0036] The functions of the respective units of the arithmetic unit 16 are described below. The respective units of the arithmetic unit 16 can be executed by executing the computer programs stored in the storage unit 18. The scenario setting unit 22 acquires the condition of the simulation for which the analysis is to be performed and generates a scenario model, which is a simulation model, by selecting a model based on the condition from among the environmental object model 37, the weather model 38, the light source model 40, and the vehicle model 42. The condition of the simulation for which the analysis is to be performed is set based on the result of the user's input. If the scenario model 36 corresponding to the condition of the simulation for which the analysis is to be performed is stored in the storage unit 18, the scenario setting unit 22 reads the stored scenario model 36.

[0037] The sensor data generation unit 24 generates the sensor data when the vehicle is traveling according to the scenario set by the scenario setting unit 22, using the sensor model that models the sensor in the vehicle. The sensor data generation unit 24 performs a simulation based on the scenario model generated by the scenario setting unit 22 and, based on the scenario model, generates the sensor data that is the data detected by the sensor when the vehicle is traveling.

[0038] The linking processing unit 28 links the sensor data generated in the sensor data generation unit 24, and the real data conversion unit 26 converts the data into the real data using the linking result.

[0039] The link processing unit 28 links the real data with the sensor data output from the real data conversion unit 26. The link processing unit 28 specifies the real data corresponding to the sensor data using the link table 46. In the case of using the trained model 50, the link processing unit 28 inputs the sensor data to the machine learning unit 30 and acquires the real data that is output.

[0040] The machine learning unit 30 performs a deep learning process and generates the trained model 50 using the teacher data. Inputting the sensor data generated during a simulation based on the scenario, the machine learning unit 30 outputs the real data corresponding to the sensor data. The machine learning unit 30 further performs the process using the trained model 50. Specifically, the machine learning unit 30 inputs the input sensor data to the trained model 50 and outputs the corresponding real data. Proceedings

[0041] Next, an evaluation data generation procedure is developed based on Fig. 2 described. Fig. Figure 2 is a flowchart illustrating an example of a process of the evaluation data generation device. The process shown in Fig. 2 is a process that does not use machine learning in the linking process.

[0042] The arithmetic unit 16 causes the scenario setting unit 22 to determine the scenario for the simulation to be performed (step S12). The scenario setting unit 22 determines the scenario to be performed based on the condition input by the user, a preset condition, or the like.

[0043] The arithmetic unit 16 causes the scenario setting unit 22 to determine the environmental object model, the weather model, the light source model, and the vehicle model to be used based on the scenario and generate a scenario model (step S14).

[0044] The arithmetic unit 16 inputs the determined scenario model into the sensor model and generates the sensor data (step S16). The arithmetic unit 16 causes the sensor data generation unit 24 to generate the sensor model based on the vehicle model 42. The sensor data generation unit 24 inputs the scenario model into the sensor model to perform the simulation and generates the sensor data detected by the sensor model, that is, the data to be detected by the sensor when driving according to the scenario model.

[0045] The arithmetic unit 16 determines the real data corresponding to the sensor data (step S18). The arithmetic unit 16 causes the real data conversion unit 26 and the link processing unit 28 to process the sensor data using the link table 46 and determines the real data to be linked to the sensor data. In the present embodiment, based on the sensor data and the scenario model, the real data with the closest condition is selected from the real data linked in the link table 46.

[0046] The arithmetic unit 16 performs a real data conversion process (step S20). The arithmetic unit 16 causes the real data conversion unit 26 to convert the sensor data into real data based on the relationship determined in the relationship processing unit 28.

[0047] The arithmetic unit 16 stores the generated real data (step S22). The real data is used as evaluation data.

[0048] In the present embodiment, as described above, the evaluation data can be made the real data by specifying the real data for the sensor data generated by the simulation and converting the sensor data into the real data.

[0049] Here, for example, the scenario model is created by defining the shape of an object as a polygon, setting the properties of the object (such as reflectivity, transmittance, and surface texture) as materials, and combining them to obtain a model with a three-dimensional shape. In the present embodiment, if the scenario model setting is created using a simple model, for example, using only the information set during initial setting in the model creation software, the real data can be used as the evaluation data to be supplied to the ECU 6 by converting the sensor data into the real data.

[0050] This allows evaluation data close to the data in the case of real driving to be generated without requiring an accurate model to output values ​​close to the real data from the sensor model. Furthermore, since evaluation data capable of highly accurate evaluation can be generated using a simple model, such as the scenario model, it is possible to reduce the time required to generate evaluation data and simplify the process. By linking the real data, it is also possible to counteract any deviation of the evaluation data from the real sensor detection.

[0051] In the above-described embodiment, the link processing unit 28 determines a unit of the real data 48 according to the sensor data, using the determined real data as evaluation table data; however, the present disclosure is not limited to this. The evaluation data generation device 10 may generate the trained model in the machine learning unit and convert the sensor data into the real data using the trained model. Generation method for a trained model

[0052] Fig. 3 is a flowchart illustrating an example of the operation of the machine learning unit. Fig. The process illustrated in Figure 3 is performed by the machine learning unit. The machine learning unit 30 acquires the real data (step S32). The machine learning unit 30 can acquire either the real data stored in the storage unit 18 or the real data via the communication unit 15.

[0053] The machine learning unit 30 generates the sensor data (step S34). The machine learning unit 30 executes the process in the scenario setting unit 22 and the sensor data generation unit 24 and generates the sensor data according to the real data. The machine learning unit 30 detects, for example, the driving condition in the real data and generates the sensor data using the generated condition.

[0054] The machine learning unit 30 performs the linking between the sensor data and the real data (step S36). In other words, the machine learning unit 30 links the real data acquired in step S32 with the sensor data generated in step S34.

[0055] The machine learning unit 30 determines whether the generation of data for learning is complete (step S38). The criterion of the machine learning unit 30 is, for example, whether the number of combinations between the sensor data and the real data is greater than or equal to a preset number.

[0056] If the machine learning unit 30 determines that the generation of the learning data is not complete (NO at step S38), the process returns to step S32, and another combination of the sensor data and the real data is generated.

[0057] If the machine learning unit 30 determines that the generation of the learning data is complete (YES at step S38), the machine learning unit 30 performs learning of the trained model by inputting the sensor data and outputting the corresponding real data (step S40). For example, the machine learning unit 30 uses part of the combination of the sensor data and the real data serving as learning data as learning data and the remainder as verification data, and performs learning with the learning data and verification with the verification data.

[0058] In the above-described process, the machine learning unit 30 generates a trained model that outputs the corresponding real data upon input of the sensor data. The method for generating the trained model is only an example and is not limited thereto. Although the above-described embodiment describes supervised learning, it is possible to create a trained model and real data using unsupervised learning and specify the real data for the sensor data. Further examples of the evaluation data generation process

[0059] Fig. 4 is a flowchart illustrating an example of a process of the evaluation data generation device. Fig. The process illustrated in Figure 4 is the same as in Fig. 2 are designated by the same reference numerals, and a detailed description thereof is omitted.

[0060] The arithmetic unit 16 causes the scenario setting unit 22 to determine the scenario for the simulation to be performed (step S12). The arithmetic unit 16 causes the scenario setting unit 22 to determine the environmental object model, the weather model, the light source model, and the vehicle model to be used based on the scenario and generate a scenario model (step S14). The arithmetic unit 16 inputs the determined scenario model into the sensor model and generates the sensor data (step S16).

[0061] The arithmetic unit 16 inputs the sensor data into the trained model and converts the sensor data into the real data (step S52).

[0062] The arithmetic unit 16 stores the generated real data (step S22). The real data is used as evaluation data.

[0063] As in Fig. 4, the evaluation data generation device 10 converts the sensor data into the real data using the trained model generated by machine learning (deep learning), thereby enabling, for example, the data conversion without setting various conditions.

[0064] Here, the evaluation data generation device 10 can use multiple trained models. When using the trained models, the classification conditions are set, and the trained data is generated in which the combination of the sensor data and the real data satisfies each classification condition. The machine learning unit 30 performs machine learning for each unit of the trained data, thereby generating the trained model for each classification condition.

[0065] Fig. 5 is a flowchart illustrating an example of a process of the evaluation data generation device. In the Fig. The process illustrated in Figure 5 is the same as in Fig. 4 are designated by the same reference numerals, and a detailed description thereof is omitted. In the case of Fig. 5 the trained model is generated for each of the weather and light source conditions.

[0066] The arithmetic unit 16 causes the scenario setting unit 22 to determine the scenario for the simulation to be performed (step S12). The arithmetic unit 16 causes the scenario setting unit 22 to determine the environmental object model, the weather model, the light source model, and the vehicle model to be used based on the scenario and generate a scenario model (step S14). The arithmetic unit 16 inputs the determined scenario model into the sensor model and generates the sensor data (step S16).

[0067] Using the real data conversion unit 26, the arithmetic unit 16 determines the trained model to be used based on the weather model and the light source model (step S54). Based on the information about the scenario model, the real data conversion unit 26 acquires the weather and light source conditions from the sensor data and determines the trained model according to the acquired conditions. The arithmetic unit 16 inputs the sensor data into the trained model and converts the sensor data into real data (step S56).

[0068] The arithmetic unit 16 stores the generated real data (step S22). The real data is used as evaluation data.

[0069] By providing the learning data for each condition, the accuracy of specifying the real data can be increased. It is possible to link the real data that is close to the real data obtained when the real drive is performed under the condition in the scenario model.

[0070] By setting the weather and light source as classification conditions, as described in the present embodiment, it is possible to identify conditions that are difficult to determine with the sensor data generated from the scenario model. For example, in the case of an image, it is possible to counteract mixed learning results depending on whether it is raining or snowing, or depending on whether it is dark due to clouds, dark due to evening, or dark due to being indoors, and the like. Furthermore, the accuracy of conversion to real data based on the sensor data can be improved.

[0071] The evaluation data generating device can determine the learning data to be used based on the input of a user. Fig. 6 is a flowchart illustrating an example of a process of the evaluation data generation device. Fig. The process illustrated in Figure 6 is the same as in Fig. 5 are designated by the same reference numerals, and a detailed description thereof is omitted.

[0072] The arithmetic unit 16 causes the scenario setting unit 22 to determine the scenario for the simulation to be performed (step S12). The arithmetic unit 16 causes the scenario setting unit 22 to determine the environmental object model, the weather model, the light source model, and the vehicle model to be used based on the scenario and generate a scenario model (step S14). The arithmetic unit 16 inputs the determined scenario model into the sensor model and generates the sensor data (step S16).

[0073] The arithmetic unit 16 determines the trained model to be used based on the input using the real data conversion unit 26 (step S62). The real data conversion unit 26 determines the corresponding trained model based on the weather and light source information input to the input unit 12. The arithmetic unit 16 inputs the sensor data into the trained model and converts the sensor data into the real data (step S56).

[0074] The arithmetic unit 16 stores the generated real data (step S22). The real data is used as evaluation data.

[0075] The evaluation data generation device 10 can thereby convert the sensor data into real data corresponding to the weather and light source conditions required by the user. Although the information is input by the user in the present embodiment, the information may also be input from another database.

[0076] The evaluation data generation device 10 can use the weather and light source in the scenario to be determined in step S12 as the default settings. Therefore, the model generation effort can be reduced. Furthermore, since this is the process of selecting the trained model based on an optional setting by the user, it is possible to convert the data into the real data corresponding to the weather and light source conditions required by the user as the weather and light source of the real data to be used as evaluation data.

[0077] The evaluation data generating device can classify the trained models by conditions of the environment object model. Fig. 7 is a flowchart illustrating an example of a process of the evaluation data generation device. In the Fig. The process illustrated in Figure 7 is the same as in Fig. 5 are designated by the same reference numerals, and a detailed description thereof is omitted.

[0078] The arithmetic unit 16 causes the scenario setting unit 22 to determine the scenario for the simulation to be performed (step S12). The arithmetic unit 16 causes the scenario setting unit 22 to determine the environmental object model, the weather model, the light source model, and the vehicle model to be used based on the scenario and generate a scenario model (step S14). The arithmetic unit 16 inputs the determined scenario model into the sensor model and generates the sensor data (step S16).

[0079] The arithmetic unit 16 determines the trained model to be used based on the environmental object model using the real data conversion unit 26 (step S64). Based on the information about the scenario model, the real data conversion unit 26 acquires the weather and light source conditions from the sensor data and determines the trained model according to the acquired conditions. The arithmetic unit 16 inputs the sensor data into the trained model and converts the sensor data into real data (step S56).

[0080] The arithmetic unit 16 stores the generated real data (step S22). The real data is used as evaluation data.

[0081] The evaluation data generation device 10 classifies the environmental object models according to the condition and performs learning under conditions where people, buildings, trees, and the like are similar, thereby increasing the accuracy of the trained models. In addition, the divergence of calculations during learning can be counteracted and the model generation overhead can be reduced.

[0082] It may be preferable for the evaluation data generation device 10 to also input the environmental object model as input information during learning. Even in a case where the detection state changes depending on the weather or a light source (daytime and nighttime), linking is possible while processing the physical environment model. Accordingly, the detection accuracy can be further increased. Another example of an evaluation data generation device

[0083] Fig. Fig. 8 is a block diagram illustrating an example of an evaluation data generating device. Fig. The evaluation data generating device 10a shown in Fig. 8 includes a weather and light source conversion unit 29 and a second trained model 52 in addition to the configuration of the evaluation data generating device 10. The trained model of the evaluation data generating device 10 corresponds to the first trained model 50. The evaluation data generating device 10 performs the data conversion process twice.

[0084] The evaluation data generation device 10a includes two trained models, namely the first trained model 50 and the second trained model 52. The first trained model 50 is a trained model used in the process of converting the sensor data into the real data as described above. The first trained model 50 is generated by performing learning based on the sensor model and the real data using the weather and light source conditions set to correspond to the standard conditions.

[0085] The second trained model 52 is a trained model used in the process of converting the real data into real data with different weather and light source conditions. A method for generating the second trained model 52 is described below.

[0086] The weather and light source conversion unit 29 uses the second trained model 52 to generate second real data whose weather and light source conditions are changed compared to the real data (first real data) generated by the real data conversion unit 26. The weather and light source conversion unit 29 changes the weather and light source conditions based on the conditions input by the input unit 12.

[0087] Fig. 9 is a flowchart illustrating an example of the operation of the machine learning unit. As shown in Fig. As shown in Figure 9, the operation is performed in the machine learning unit. The machine learning unit 30 acquires the real data with the light source and weather under standard conditions (step S72). The machine learning unit 30 can acquire either the real data stored in the storage unit 18 or the real data via the communication unit 15.

[0088] The machine learning unit 30 acquires the real data whose light source and weather conditions differ from the standard conditions (step S74). The machine learning unit 30 can acquire either the real data stored in the storage unit 18 or the real data via the communication unit 15.

[0089] The machine learning unit 30 associates the real data with the same conditions except for the light source and weather (step S76). In other words, the machine learning unit 30 associates the real data acquired in step S72 with the sensor data generated in step S74.

[0090] The machine learning unit 30 determines whether the generation of the data for learning is complete (step S78). The criterion of the machine learning unit 30 is, for example, whether the number of combinations of the real data is greater than or equal to a set number.

[0091] If the machine learning unit 30 determines that the generation of the learning data is not complete (NO at step S78), the process returns to step S72, and another combination of the sensor data and the real data is generated.

[0092] If the machine learning unit 30 determines that the generation of the learning data is complete (YES at step S78), the machine learning unit 30 executes the learning of the second trained model such that the real data is input at the standard condition and the corresponding real data is output (step S80).

[0093] Through the above-described process, the machine learning unit 30 generates the second trained model, which converts the real data under the standard condition into the real data under other weather and light source conditions.

[0094] Fig.10 is a flowchart illustrating an example of a process of the evaluation data generation device. The arithmetic unit 16 causes the scenario setting unit 22 to determine the scenario for the simulation to be performed (step S12). The arithmetic unit 16 causes the scenario setting unit 22 to determine the environmental object model and the vehicle model to be used based on the scenario and generate a scenario model (step S90). The weather model and the light source model are the models under standard conditions. The arithmetic unit 16 inputs the determined scenario model into the sensor model and generates the sensor data (step S16).

[0095] The arithmetic unit 16 determines the first trained model (step S92). The real data conversion unit 26 determines the trained model according to the detected condition based on the information about the scenario model and the input. The first trained model may be a single trained model. The arithmetic unit 16 inputs the sensor data to the first trained model and converts the sensor data into the first real data (step S94).

[0096] The arithmetic unit 16 determines the second trained model to be used based on the input using the weather and light source conversion unit 29 (step S96). The weather and light source conversion unit 29 determines the corresponding second trained model based on the weather and light source information input to the input unit 12. The arithmetic unit 16 inputs the first real data to the second trained model and converts the data into the second real data (step S98).

[0097] The arithmetic unit 16 stores the generated second real data (step S99). The second real data is used as evaluation data.

[0098] By performing the conversion process twice, the evaluation data generation device 10a can use the weather and light source data under the standard condition in the scenario of the sensor data generated by the sensor data generation unit 24. This can reduce the overhead of generating the sensor data. In addition, the overhead of generating the first trained model can also be reduced.

[0099] The evaluation data generating device 10a causes the weather and light source converting unit 29 to change the weather and light source conditions so that the first real data converted from the sensor data can be converted into the second real data having the desired weather and light source conditions.

[0100] The evaluation data generation device 10a can perform machine learning through the process of selecting the corresponding real data as the second trained model. However, it may be preferable if the machine learning is performed during the learning process of converting the first real data into the second real data based on the teacher data. By using the second trained model 52 that enables conversion from the standard state to the desired weather and light source, the evaluation data generation device 10a can convert and correct the influence of the weather and light source with higher accuracy. Accordingly, a plurality of second real data with changed weather and light source can be generated, and more units of evaluation data can be generated.

[0101] In the above-described embodiment, the real data undergoes a process of changing the weather and light source; however, the changing process is not limited to the weather and light source. Based on various conditions set in the scenario, the real data can be changed, that is, the real data can be subjected to a correction process. Therefore, real data that is closer to the scenario can be obtained. Since the correction process can be performed additionally, more units of evaluation data can be generated under different or different conditions.

[0102] The present disclosure discloses the following inventions. The present disclosure is not limited to the following.

[0103] (1) An evaluation data generation device that generates evaluation data for evaluating at least one of automated driving of a vehicle and a driving assistance function of the vehicle, the evaluation data generation device including: a scenario setting unit that sets a scenario for simulating a drive of the vehicle; a sensor data generation unit that generates, using a sensor model obtained by modeling a sensor of the vehicle, sensor data acquired when the vehicle drives in accordance with the scenario set by the scenario setting unit;and a real data conversion unit that combines the sensor data generated by the sensor data generation unit with real data corresponding to data detected by the sensor of the vehicle, and converts the sensor data based on the combined real data to generate the converted real data corresponding to the evaluation data;

[0104] This allows evaluation data close to the data in the case of a real drive to be generated from the sensor model without creating an accurate model for outputting values ​​close to the real data. Furthermore, since evaluation data capable of highly accurate evaluation can be generated while using a simple model as the scenario model, it is possible to reduce the time required to generate the evaluation data and simplify the process.

[0105] (2) The evaluation data generation device according to (1), further comprising: a storage unit that stores the real data therein; and a link processing unit that compares the sensor data acquired by the real data conversion unit with the real data stored in the storage unit to determine the real data to be linked to the sensor data. By linking the real data, it is possible to counteract the deviation of the evaluation data from the real sensor detection.

[0106] (3) The evaluation data generation device according to (1), further comprising a linking processing unit that processes the sensor data with a trained model to determine the real data to be linked to the sensor data. The trained model has performed learning such that the sensor data is input and the corresponding real data is output using the learning data obtained by linking the sensor data with the real data. By using the trained model, the overhead of the linking process can be reduced.

[0107] (4) The evaluation data generation device according to (3), wherein the link processing unit includes a plurality of trained models classified according to a condition of the scenario, and determines the trained model to be used based on the scenario set by the scenario setting unit. Accordingly, the accuracy of the trained model can be further increased. In addition, the divergence of calculations during learning can be counteracted, and the model generation overhead can also be reduced.

[0108] (5) The evaluation data generation device according to (4), wherein the link processing unit determines the learned model to be used based on at least one of weather and an ambient light source in the scenario. By setting the weather and the light source as classification conditions, it is possible to identify conditions that are difficult to determine with the sensor data generated from the scenario model. For example, in the case of an image, it is possible to counteract mixed learning results depending on whether it is raining or snowing, or depending on whether it is dark due to a cloud, dark due to the evening, or dark due to being indoors, and the like. Furthermore, the accuracy of conversion to the real data based on the sensor data can be improved.

[0109] (6) The evaluation data generation device according to (4) or (5), further comprising an input unit that detects an input, wherein the link processing unit determines the trained model to be used based on the input detected by the input unit. Thus, the data can be converted into real data according to the condition required by the user.

[0110] (7) The evaluation data generation device according to any one of (4) to (6), wherein the link processing unit determines the learned model to be used based on environmental conditions of the vehicle. Accordingly, the accuracy of the learned model can be further increased. In addition, the divergence of calculations during learning can be counteracted, and the model generation overhead can also be reduced.

[0111] (8) The evaluation data generation device according to any one of (1) to (7), further comprising a weather and light source conversion unit that performs a process of converting at least one of the weather and the ambient light source from the scenario on the converted real data generated by the real data conversion unit to generate second converted real data. This allows the generation burden of the sensor data to be reduced, and more units of evaluation data can be generated.

[0112] (9) The evaluation data generation device according to (8), wherein the weather and light source conversion unit converts the converted real data into the second converted real data using a trained model in association with the real data that differs in at least one of the weather and an ambient light source, the trained model having performed the learning such that the real data of the weather and the ambient light source in the scenario generated by the scenario generation unit is input and the linked real data that differs in at least one of the weather and the ambient light source is output. Thus, real data that is closer to the scenario can be obtained. In addition, since the rectification process can be performed, more units of evaluation data with different conditions can be generated.

[0113] (10) An evaluation data generation method for generating evaluation data for evaluating at least one automated driving function of a vehicle and a driving assistance function of the vehicle, the evaluation data generation method including: a step of setting a scenario for simulating a travel of the vehicle; a step of generating, using a sensor model obtained by modeling a sensor of the vehicle, sensor data acquired when the vehicle travels in accordance with the scenario set by the scenario setting unit; and a step of linking the generated sensor data with real data corresponding to data acquired by the sensor of the vehicle, and converting the sensor data based on the linked real data to generate converted real data corresponding to the evaluation data.

[0114] This allows evaluation data close to the data in the case of a real drive to be generated from the sensor model without creating an accurate model for outputting values ​​close to the real data. Furthermore, since evaluation data capable of highly accurate evaluation can be generated while using a simple model as the scenario model, it is possible to reduce the time required to generate the evaluation data and simplify the process.

[0115] (11) An evaluation data generation program for generating evaluation data for evaluating at least one automated driving function of a vehicle and a driving assistance function of the vehicle, wherein the evaluation data generation program performs a process including: a step of setting a scenario for simulating a drive of the vehicle; a step of generating sensor data acquired when the vehicle drives in accordance with the set scenario using a sensor model obtained by modeling a sensor of the vehicle; and a step of linking the generated sensor data with real data corresponding to data acquired by the sensor of the vehicle, and converting the sensor data based on the linked real data to generate converted real data corresponding to the evaluation data.

[0116] This allows evaluation data close to the data in the case of a real drive to be generated from the sensor model without creating an accurate model for outputting values ​​close to the real data. Furthermore, since evaluation data capable of highly accurate evaluation can be generated while using a simple model as the scenario model, it is possible to reduce the time required to generate the evaluation data and simplify the process. List of reference symbols 6 ECU 8 Evaluation device 10 Evaluation data generation device 12 Input unit 14 Output unit 15 Communication unit 16 arithmetic units 18 storage unit 22 Scenario setting unit 24 Sensor data generation unit 26 Real data conversion unit 28 Link processing unit 30 Machine Learning Unit 32 Data generation program 34 Learning program 36 Scenario model 37 Environment object model 38 weather model 40 light source model 42 vehicle model 44 Sensor model 46 Linking table 48 real data 50 Trained model QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited patent literature

[0000] JP 2017-105453

[0004]

Claims

[1] An evaluation data generating device that generates evaluation data for evaluating at least one of automated driving of a vehicle and a driving assistance function of the vehicle, the evaluation data generating device comprising: a scenario setting unit that sets a scenario for simulating a trip of the vehicle; a sensor data generation unit that generates, using a sensor model obtained by modeling a sensor of the vehicle, sensor data acquired when the vehicle travels in accordance with the scenario set by the scenario setting unit; and a real data conversion unit that links the sensor data generated by the sensor data generation unit with real data corresponding to data detected by the sensor of the vehicle, and converts the sensor data based on the linked real data to generate converted real data corresponding to the evaluation data. [2] The evaluation data generating apparatus according to claim 1, further comprising: a storage unit that stores the real data therein; and a link processing unit that compares the sensor data acquired by the real data conversion unit with the real data stored in the storage unit to determine the real data to be linked to the sensor data. [3] The evaluation data generating apparatus according to claim 1, further comprising a link processing unit that processes the sensor data with a trained model to determine the real data to be linked with the sensor data, wherein the trained model has performed the learning such that the sensor data is input and the corresponding real data is output using the learning data obtained by linking the sensor data with the real data. [4] The evaluation data generating device according to claim 3, wherein the link processing unit includes a plurality of trained models classified according to a condition of the scenario, and determines the trained model to be used based on the scenario set by the scenario setting unit. [5] The evaluation data generating apparatus according to claim 4, wherein the link processing unit determines the trained model to be used based on at least one of weather and an ambient light source in the scenario. [6] The evaluation data generating apparatus according to claim 4, further comprising an input unit that detects an input, wherein the link processing unit determines the trained model to be used based on the input detected by the input unit. [7] The evaluation data generating apparatus according to claim 4, wherein the link processing unit determines the learned model to be used based on environmental conditions of the vehicle. [8] The evaluation data generating device according to any one of claims 1 to 7, further comprising a weather and light source converting unit that performs a process of converting at least one of weather and the ambient light source from the scenario on the converted real data generated by the real data converting unit to generate second converted real data. [9] The evaluation data generating device according to claim 8, wherein the weather and light source converting unit converts the converted real data into the second converted real data using a learned model in association with the real data different in at least one of weather and an ambient light source, wherein the learned model has performed the learning such that the real data of the weather and the ambient light source in the scenario generated by the scenario generating unit is input and the linked real data different in at least one of the weather and the ambient light source is output. [10] Evaluation data generation method for generating evaluation data for evaluating at least one driver assistance function of automated driving of a vehicle and a driver assistance function of the vehicle, the evaluation data generation method comprising: a step of setting a scenario to simulate a journey of the vehicle; a step of generating sensor data using a sensor model obtained by modeling a sensor of the vehicle, which are detected when the vehicle travels in accordance with the scenario set by the scenario setting unit; and a step of linking the generated sensor data with real data corresponding to data acquired by the sensor of the vehicle, and converting the sensor data based on the linked real data to generate converted real data corresponding to the evaluation data. [11] Evaluation data generation program for generating evaluation data for evaluating at least one driver assistance function of automated driving of a vehicle and a driver assistance function of the vehicle, wherein the evaluation data generation program performs a process comprising: a step of setting a scenario to simulate a journey of the vehicle; a step of generating sensor data using a sensor model obtained by modeling a sensor of the vehicle, which are recorded when the vehicle drives in accordance with the set scenario; and a step of linking the generated sensor data with real data corresponding to data acquired by the sensor of the vehicle, and converting the sensor data based on the linked real data to generate converted real data corresponding to the evaluation data.

Images