Scenario generation device and scenario generation method

Abstract

To provide a scenario generation device which attains high comprehensiveness of scenarios arising in an actual environment, while restricting a number of scenarios.SOLUTION: A scenario generation device includes: an actual environment scene acquisition section 22 for acquiring actual environment scenes that are scenes arising in an actual environment, from a travel database in which travel data of actual vehicles has been recorded; a filter generation section 23 for generating a filter 25 for filtering candidate scenarios on the basis of a frequency analysis result of analysis object data containing travel data indicating actual environment scenes; and an evaluation scenario determination section 24 for determining an evaluation scenario by filtering the candidate scenarios that have been generated comprehensively based on a mathematical model using the filter 25 generated by the filter generation section 23.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a scenario generation device and a scenario generation method for generating scenarios for evaluating vehicle control functions, and in particular to a technique for effectively reducing the number of scenarios. [Background technology]

[0002] There are known devices and methods for generating scenarios for evaluating vehicle control functions. In Patent Document 1, a traffic scenario is enhanced by using a critical event that occurred in a real environment as a condition and adapting a traffic log obtained in the real environment to the event driving environment of the critical event. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2020-123351 Summary of the Invention [Problem to be solved by the invention]

[0004] The technology disclosed in Patent Document 1 can increase the number of scenarios. However, the method of increasing traffic logs obtained in a real environment by using critical events as a condition may result in biased scenarios. Therefore, the method disclosed in Patent Document 1 may not be able to cover all the scenarios to be evaluated. Comprehensiveness can be improved by generating scenarios based on a mathematical model rather than obtaining scenarios from a real environment. However, there is a risk that the number of scenarios will become enormous.

[0005] The present disclosure has been made based on this situation, and its purpose is to provide a scenario generation device and a scenario generation method that can increase the comprehensiveness of scenarios that occur in real environments while reducing the number of scenarios. [Means for solving the problem]

[0006] The above object is achieved by the combination of features recited in the independent claims, and the subclaims define further advantageous specific examples. The reference numerals in parentheses in the claims correspond to specific aspects described in the following embodiments as one aspect, and do not limit the technical scope of the disclosure.

[0007] One disclosure relating to a scenario generation device for achieving the above object is: a real environment scene acquisition unit (22) that acquires a real environment scene, which is a scene that occurs in a real environment, from a travel database that stores travel data of a real vehicle; a filter generation unit (23, 223) that generates a filter for filtering candidate scenarios based on a frequency analysis result of analysis target data including driving data showing a real environment scene; The scenario generation device includes an evaluation scenario determination unit (24) that uses the filter generated by the filter generation unit to filter candidate scenarios that are comprehensively generated based on a mathematical model to determine an evaluation scenario.

[0008] Because candidate scenarios are generated based on a mathematical model, they can be comprehensive scenarios. However, because scenarios are created comprehensively based on a mathematical model, the number of candidate scenarios becomes enormous. Therefore, the scenario generation device determines evaluation scenarios by filtering the candidate scenarios. This reduces the number of scenarios.

[0009] The filter generation unit generates the filter used for filtering based on the results of frequency analysis of the analysis target data, including driving data showing scenes in the real environment. By filtering using this filter, it is possible to exclude scenarios related to scenes that do not occur in the real environment or that are unlikely to occur in the real environment from the candidate scenarios. Therefore, it is possible to increase the comprehensiveness of scenarios that occur in the real environment while suppressing the number of scenarios.

[0010] One disclosure relating to a scenario generation method for achieving the above object is: At least one processor or circuitry Acquire a real environment scene, which is a scene that occurs in a real environment, from a travel database that stores travel data of a real vehicle; A filter is generated to filter candidate scenarios based on the frequency analysis results of the analysis target data including driving data showing real-world scenes; This is a scenario generation method that uses a filter to determine an evaluation scenario by filtering candidate scenarios that are comprehensively generated based on a mathematical model. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 is a diagram showing the configuration of a scenario generation device 10 according to a first embodiment. [Figure 2] A diagram showing the parameters of the mathematical model that defines traffic disturbances. [Figure 3] A diagram illustrating the coordinate system of the parameters shown in Figure 2. [Figure 4] FIG. 10 is a diagram showing an equation for deriving a mathematical model showing the deceleration of another vehicle. [Figure 5] FIG. 10 is a diagram showing a frequency distribution of the speed of the host vehicle extracted from driving data showing a real-world scene. [Figure 6] FIG. 10 is a diagram showing a frequency distribution of the speeds of other vehicles extracted from driving data showing a real-world scene. [Figure 7] FIG. 2 is a diagram illustrating filtering performed by an evaluation scenario determination unit 24. [Figure 8] FIG. 10 is a diagram showing the configuration of a scenario generation device 200 according to a second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0012] First Embodiment Hereinafter, embodiments will be described with reference to the drawings. FIG. 1 is a diagram showing the configuration of a scenario generation device 10 according to a first embodiment. The scenario generation device 10 is a device that generates evaluation scenarios for evaluating the performance of a vehicle control function that controls the driving of a vehicle. Hereinafter, the vehicle whose vehicle control function is to be evaluated will be referred to as the subject vehicle.

[0013] The vehicle control function is not limited to the function of automatically driving the host vehicle, but also includes the function of controlling the driving of the host vehicle to assist the driver. A scenario refers to a series of behaviors of the host vehicle from start to finish. The start and end of the series of behaviors can be set arbitrarily. For example, a scenario may be one in which another vehicle cuts in front of the host vehicle. In this case, the start of the scenario can be the point at which the other vehicle starts to move laterally, and the end of the scenario can be the point at which the other vehicle finishes moving laterally. An evaluation scenario specifies specific numerical values ​​that express the above series of behaviors. Since the evaluation scenario specifies specific numerical values, it can be evaluated using the virtual environment evaluation device 50 or a vehicle in the real world.

[0014] The scenario generation device 10 includes a control unit 20 and an evaluation scenario storage unit 30. The control unit 20 can be realized by a configuration including at least one processor. For example, the control unit 20 can be realized by a computer including a processor, nonvolatile memory, RAM, I / O, and a bus line connecting these components. The nonvolatile memory stores a scenario generation program for operating a general-purpose computer as the control unit 20. The processor executes the program stored in the nonvolatile memory while utilizing the temporary storage function of the RAM, causing the control unit 20 to operate as a candidate scenario generation unit 21, a real environment scene acquisition unit 22, a filter generation unit 23, and an evaluation scenario determination unit 24. Execution of these operations means that a scenario generation method corresponding to the scenario generation program is being executed.

[0015] The candidate scenario generator 21 generates candidate scenarios to be used as evaluation scenarios. The candidate scenarios are generated to cover as many theoretically possible scenarios as possible. To increase the coverage, candidate scenarios are generated that take into account either recognition performance or vehicle dynamics performance or both for traffic disturbance scenarios that affect basic vehicle behavior.

[0016] The basic vehicle behaviors are lane keeping, lane changing, etc. The evaluation scenarios are scenarios that evaluate safety when a traffic disturbance occurs to the basic vehicle behaviors. A traffic disturbance refers to a potentially dangerous traffic situation that arises as a combination of road structure, the behavior of the vehicle itself, and the positions and movements of surrounding vehicles.

[0017] Recognition performance is the performance related to recognition when vehicle control is divided into three elements: recognition, judgment, and operation. Recognition performance is the performance of the recognition-related sensor system installed on the vehicle. A sensor system is equipped with one or more sensors such as a camera, millimeter-wave radar, or Lidar. Recognition performance will be explained in detail. For example, for millimeter-wave radar, this includes the maximum detection distance, detection range, and resolution. The maximum detection distance, detection range, and resolution are basic or static performance. Recognition performance changes dynamically due to recognition disturbances.

[0018] Recognition disturbance refers to a disturbance in the recognition process. Recognition disturbance refers to a state in which the sensor system is unable to recognize information that it should recognize. An example of recognition disturbance is a state in which a camera, millimeter wave, Lidar, etc. is blocked by a surrounding vehicle and is unable to recognize anything further away than the surrounding vehicle. Another example of recognition disturbance is a decrease in recognition due to sensor dirt, etc. Recognition disturbance also includes sensor blind spots and communication disturbances.

[0019] Vehicle dynamics performance is the performance related to operation when vehicle control is divided into three elements: perception, judgment, and operation. Vehicle dynamics performance includes deceleration performance, acceleration performance, and steering performance. Vehicle dynamics performance changes dynamically due to vehicle dynamic disturbances.

[0020] Vehicle motion disturbance refers to a situation in which the vehicle may not be able to be controlled to the desired state. Vehicle motion disturbances include factors inside the vehicle and factors outside the vehicle. Factors inside the vehicle include, for example, the total vehicle weight and weight balance. Factors outside the vehicle include road surface irregularities, road surface inclination, wind, etc.

[0021] The evaluation scenario is a scenario for evaluating whether a vehicle will collide with a nearby traffic participant or structure. To evaluate this, a traffic disturbance scenario is first considered. The evaluation scenario is a scenario that evaluates whether a collision will occur even when a recognition disturbance and a vehicle disturbance are added based on the traffic disturbance scenario.

[0022] Here, we will explain traffic disturbance scenarios. Traffic disturbance scenarios of other vehicles relative to the subject vehicle can be defined by combining three elements: "road shape," "subject vehicle behavior," and "other vehicle behavior." The subject vehicle behavior can be divided into "lane keeping" and "lane change." The other vehicle behavior can be divided into "deceleration," "cut-in," and "cut-out." Combining the elements of the subject vehicle and other vehicles, 3 x 2 = 6 different scenarios can be obtained.

[0023] "Deceleration" refers to the movement of another vehicle in front of your vehicle that is in the same lane as your vehicle approaching you. "Cut-in" refers to the movement of another vehicle in a different lane from your vehicle moving into the same lane as your vehicle. "Cut-out" refers to the movement of another vehicle in front of your vehicle that is traveling in the same lane as your vehicle moving into a different lane.

[0024] The behavior of the subject vehicle and other vehicles can be represented by a mathematical model. In other words, traffic disturbances can be represented by a mathematical model. A mathematical model is an expression using an equation. As an example, a mathematical model that represents the "deceleration" of other vehicles will be explained.

[0025] The "deceleration" of another vehicle can be expressed, for example, by the parameters shown in Figure 2. The coordinate system of the parameters shown in Figure 2 is shown in Figure 3. Using the parameters shown in Figure 2, the "deceleration" of the other vehicle's movement can be expressed based on Equation 1 and Equation 2 shown in Figure 4.

[0026] When another vehicle is decelerating, it can be considered that the following two conditions are met. Condition 1 is that the other vehicle is approaching the own vehicle in the front-rear direction. Condition 2 is that the other vehicle and the own vehicle are traveling on the same lane. When condition 1 is met, t a ≦t≦tb and the relative velocity (V rx =V ox -V ex ) continues with a negative value. Therefore, f(V rx The average value of (i), 0) is 1. From this, we obtain Equation 3 and Equation 4.

[0027] If condition 2 is met, the distance to the left lane marker from the perspective of another vehicle will be a negative value, and the distance to the right lane marker will be a positive value. y (i)-D ll (i) <0 and D y (i)-D lr (i)>0 continues. We define a function cp to calculate the product of these two. The function cp is shown in Equation 5.

[0028] When the function cp is 1, the vehicle and the other vehicle are traveling in the same lane. When the function cp is 0, the vehicle and the other vehicle are traveling in different lanes. a ≦t≦t b If the average of the function cp is 1, then condition 2 is met. This is shown by equations 6 and 7. If the result of ANDing equations 4 and 7 is 1, that is, if equation 8 is met, then the other vehicle is decelerating.

[0029] Although the description of the mathematical model for the behavior of the subject vehicle and the behavior of other vehicles other than the deceleration of other vehicles will be omitted, the behavior of the subject vehicle and the behavior of other vehicles other than the deceleration of other vehicles can also be expressed by a mathematical model. Therefore, traffic disturbances can be expressed by a mathematical model.

[0030] Next, a mathematical model of recognition performance will be specifically described. As described above, recognition performance can be expressed as a parameter using the maximum detection distance, etc. It is preferable to model recognition performance in consideration of recognition disturbances. An example of a recognition disturbance is a decrease in reception sensitivity in millimeter-wave radar. A decrease in reception sensitivity reduces the maximum detection distance. Therefore, the relationship between a decrease in reception sensitivity and the maximum detection distance can be expressed by a mathematical model.

[0031] Next, a mathematical model of vehicle dynamics performance will be described. Vehicle dynamics performance can be modeled because it includes deceleration performance, acceleration performance, steering performance, etc. For example, deceleration performance can be modeled using parameters such as deceleration start speed, vehicle weight, and brake oil pressure. It is preferable that vehicle dynamics performance also be modeled taking into account vehicle dynamics disturbances. An example of a vehicle dynamics disturbance is road friction resistance. The smaller the road friction resistance, the longer the shortest stopping distance. Therefore, the shortest stopping distance can be expressed by a model that includes road friction resistance.

[0032] The candidate scenario generator 21 comprehensively generates candidate scenarios based on a traffic disturbance scenario, taking into consideration recognition performance and vehicle dynamics performance. By varying the specific values ​​entered into parameters in a scenario expressed as a mathematical model, specific candidate scenarios that cover the entire range of parameter variation in that scenario can be generated. The pitch at which the specific values ​​entered into the parameters are varied can be determined appropriately based on the required evaluation accuracy.

[0033] Comprehensiveness can be ensured by generating candidate scenarios by varying the specific values ​​entered into the parameters for each of a large number of mathematical models. However, the number of candidate scenarios becomes enormous. Therefore, this scenario generation device 10 narrows down the candidate scenarios using a filter 25 generated by statistically processing real-world scenes, and determines the evaluation scenario.

[0034] The real-world scene acquisition unit 22 extracts real-world scenes to be statistically processed from the traveling data storage unit 40. A scene is a concept similar to a scenario, and like a scenario, it refers to a series of behaviors of the vehicle and surrounding traffic participants. A scenario is an expected behavior, while a scene can be used primarily to describe behavior observed in a real environment. However, a scene can also be used to mean a part of a scenario. Therefore, when it is clear that the environment is real, it is referred to as a real-world scene. Note that a scene may also be the entire scenario.

[0035] The travel data storage unit 40 stores a travel database that accumulates travel data measured by a measurement vehicle, which is an actual vehicle. The travel data measured by the measurement vehicle can include data that appears outside the measurement vehicle, such as the vehicle speed, acceleration, and position of the measurement vehicle, as well as data that indicates control amounts of actuators within the vehicle, such as accelerator opening, brake oil pressure, and steering angle. Furthermore, the travel database may include travel data observed at observation points in addition to the travel data measured by the measurement vehicle.

[0036] The running data storage unit 40 does not need to be provided in the scenario generation device 10. The scenario generation device 10 only needs to be able to read out running data from the running data storage unit 40. The running data storage unit 40 only needs to be connected to the scenario generation device 10 via a wired or wireless network.

[0037] The real-world scene acquisition unit 22 acquires driving data representing a real-world scene from the driving data storage unit 40 based on a preset extraction logic. A real-world scene refers to a driving scene occurring in a real environment. The extraction logic can be determined for each vehicle control function. Examples of vehicle control functions include adaptive cruise control and collision mitigation braking systems. However, vehicle control functions are not limited to these. The reason for determining the extraction logic for each vehicle control function is that different vehicle control functions require different scenes to be evaluated.

[0038] For example, a collision mitigation braking system needs to extract scenes in which other vehicles cut in and decelerate, but does not need scenes in which other vehicles cut out and accelerate.On the other hand, an adaptive cruise control system also needs to extract scenes in which other vehicles accelerate.

[0039] The extraction logic may be any logic that can extract necessary scenes determined for each vehicle control function. The extracted scenes preferably include scenes in which the vehicle is safely driven by the vehicle control function, in other words, standard scenes, and dangerous scenes.

[0040] Since it is sufficient to extract these scenes, the real environment scene acquisition unit 22 may acquire the scenes manually. Alternatively, extraction logic may be generated by machine learning based on scenes manually acquired by a user. Alternatively, a traffic disturbance scenario required for evaluating the vehicle control function may be used as the extraction logic.

[0041] The filter generation unit 23 performs a frequency analysis on the travel data showing the real environment scene acquired by the real environment scene acquisition unit 22 as the data to be analyzed. Examples of frequency analysis results are shown in Figures 5 and 6. Figure 5 is a diagram showing a frequency distribution of the speed of the subject vehicle extracted from the travel data showing the real environment scene extracted by the real environment scene acquisition unit 22. Figure 6 is a diagram showing a frequency distribution of the speed of other vehicles extracted from the same travel data as Figure 5. The horizontal axis of Figures 5 and 6 represents speed (km / h), and the vertical axis represents frequency.

[0042] In Figures 5 and 6, the dashed lines indicate a preset extraction threshold. The extraction threshold is preferably set to a small value so as not to miss any scenes that may occur in the real world. The extraction threshold may be the same regardless of the parameter, or may be different for each parameter. In Figure 5, the frequency continuously exceeds the extraction threshold in the range of 40 km / h to 60 km / h. In Figure 6, the frequency continuously exceeds the extraction threshold in the range of 40 km / h to 60 km / h. An example of a filter 25 generated based on the frequency analysis results shown in Figures 5 and 6 is a filter 25 that extracts the speeds of both the host vehicle and other vehicles in the range of 40 km / h to 60 km / h.

[0043] The filter 25 generated by the filter generating unit 23 may define the extraction range for three or more types of parameters, or may define the extraction range for only one type of parameter.

[0044] The evaluation scenario determination unit 24 filters the comprehensive candidate scenarios generated by the candidate scenario generation unit 21 using the filter 25 generated by the filter generation unit 23. The evaluation scenario determination unit 24 stores the filtered scenarios in the evaluation scenario storage unit 30 as evaluation scenarios.

[0045] The filtering performed by the evaluation scenario determination unit 24 will be specifically described using Figure 7. Note that Figure 7 is a diagram conceptually explaining filtering, and easy-to-understand numerical values ​​are used for the parameters. In Figure 7, region R1 is the region where candidate scenarios exist. Boundaries L1 and L2 are lines determined by the mathematical model used to generate the candidate scenarios. Region R1 is the region surrounded by boundaries L1 and L2, a line indicating that the other vehicle's speed is 0 km / h, and a line indicating that the host vehicle's speed is 100 km / h. Candidate scenarios exist evenly in region R1. Note that, as in Patent Document 1, in scenarios obtained in a real environment, region R1 has parts with many scenarios and parts with few scenarios. Region R2 is the region extracted by filter 25.

[0046] The evaluation scenario determination unit 24 uses a filter 25 to extract candidate scenarios in region R2 from the candidate scenarios generated by the candidate scenario generation unit 21, that is, candidate scenarios in region R1.

[0047] 7, region R2 is completely contained within region R1. However, various extraction logics can be used by the real environment scene acquisition unit 22 to extract the real environment scene. Depending on the specific extraction logic, part of region R2 may be outside region R1.

[0048] The evaluation scenario storage unit 30 is a writable storage unit that stores the evaluation scenario determined by the evaluation scenario determination unit 24.

[0049] The virtual environment evaluation device 50 is a device that evaluates the evaluation scenarios stored in the evaluation scenario storage unit 30 in a virtual environment. The virtual environment evaluation device 50 is a simulation device that runs a virtual vehicle in a virtual environment, and is realized by a computer running a simulation program. The virtual vehicle virtually includes sensors, actuators that control the vehicle, a vehicle control application, and the like. Weather, day / night, traffic participants, and the like can be virtually set in the virtual environment. The virtual environment evaluation device 50 evaluates each evaluation scenario.

[0050] When an evaluation scenario is evaluated by the virtual environment evaluation device 50, a simulation result may be obtained, such as an accident occurring in the evaluation scenario. The evaluation result can be fed back to modify one or both of the algorithms and parameters of the vehicle control application.

[0051] Summary of the first embodiment In the scenario generation device 10 of the first embodiment described above, the candidate scenario generation unit 21 generates candidate scenarios based on a mathematical model. This allows for comprehensive generation of candidate scenarios. Furthermore, the evaluation scenario determination unit 24 determines evaluation scenarios by filtering the candidate scenarios. This allows for a reduction in the number of evaluation scenarios.

[0052] The filter generation unit 23 generates the filter 25 used for filtering based on the results of frequency analysis of real-world scenes. By filtering using this filter 25, it is possible to exclude scenarios related to scenes that do not occur in the real world or that are unlikely to occur in the real world from the candidate scenarios. This makes it possible to increase the comprehensiveness of scenarios that can occur in the real world while suppressing the number of scenarios.

[0053] The candidate scenarios are determined based on the recognition performance, traffic disturbances, and vehicle dynamics performance expressed by a mathematical model, thereby increasing the comprehensiveness of the candidate scenarios.

[0054] The real environment scene acquisition unit 22 extracts the real environment scene using extraction logic determined for each vehicle control function. This allows the filter 25 generated by the filter generation unit 23 to be suitable for specific vehicle control, thereby increasing the validity of the evaluation scenario extracted by the filter 25.

[0055] Second Embodiment Next, a second embodiment will be described. In the following description of the second embodiment, elements having the same reference numerals as those used previously are the same as those in the previous embodiments unless otherwise specified. Furthermore, when only a portion of the configuration is described, the previously described embodiment can be applied to the other portions of the configuration.

[0056] 8 shows a scenario generation device 200 of the second embodiment. The scenario generation device 200 differs from the control unit 20 of the first embodiment in the processing executed by the control unit 220. The control unit 220 differs from the control unit 20 of the first embodiment in that the processing of the filter generation unit 223 is partially different from the processing of the filter generation unit 23, and in that the control unit 220 is equipped with a scene generation unit 226.

[0057] The scene generation unit 226 generates a driving scene using the driving data stored in the driving data storage unit 40 as input. The scene generation unit 226 includes a generator trained by machine learning, and inputs driving data representing at least one driving scene stored in the driving data storage unit 40 to this generator.

[0058] The generator is trained using a part or all of the real environment scenes extracted by the real environment scene acquisition unit 22. The part of the real environment scenes is, for example, a dangerous scene that is set in advance for each vehicle control function.

[0059] A technology is known that learns a driving scene and predicts the entire driving scene from driving data that is shorter than the learned driving scene. This technology is used for hazard prediction, etc. In this embodiment, a generator trained as described above is used to generate a driving scene using driving data that was not used in the learning as input. The driving scene generated in this manner represents a driving scene similar to the real-world scene used to learn the generator.

[0060] The scene generation unit 226 preferably selects one or more parameters included in the driving data used for learning, and performs a frequency analysis of the driving data used for learning for the selected parameters. Based on the results of this frequency analysis, driving data similar to the driving data used for learning is selected as the driving data to be input. For example, suppose the result of the frequency analysis shows that the parameter α is α1 to α2 (α1<α2). In this case, driving data in which the parameter α is in the range of α1-Δβ to α1 or α2+Δβ is selected as the driving data to be input. In this way, it is possible to prevent the generated driving scene from differing significantly from the real environment scene used for learning.

[0061] The filter generation unit 223 adds to the frequency analysis target data not only the driving data indicating the real environment scene acquired by the real environment scene acquisition unit 22, but also the driving data generated by the scene generation unit 226. Except for this point, the processing executed by the filter generation unit 223 is the same as that executed by the filter generation unit 23.

[0062] Summary of the second embodiment The scenario generation device 200 of the second embodiment includes a scene generation unit 226. The scene generation unit 226 includes a generator that learns using some or all of the real environment scenes extracted by the real environment scene acquisition unit 22. This scene generation unit 226 can generate driving data that shows driving scenes similar to the real environment scenes extracted by the real environment scene acquisition unit 22 from driving data that was not extracted by the real environment scene acquisition unit 22.

[0063] Then, the filter generation unit 223 performs statistical processing not only on the driving data representing the real environment scene but also on the driving data generated by the scene generation unit 226. In this way, even if the amount of driving data stored in the driving data storage unit 40 is small, the number of driving scenes that the filter generation unit 223 performs frequency analysis on can be increased. As a result, even if the amount of driving data stored in the driving data storage unit 40 is small, the quality of the filter 25 generated based on the frequency analysis result can be improved.

[0064] Although the embodiments have been described above, the disclosed technology is not limited to the above-described embodiments, and the following modifications are also included in the scope of the disclosure. Furthermore, various modifications other than those described below can be made without departing from the spirit of the invention.

[0065] <Variation 1> The scenario generation device 10 in the embodiment was equipped with a candidate scenario generation unit 21 that generates candidate scenarios. However, candidate scenarios are scenarios that can be created regardless of the actual environment. Therefore, candidate scenarios only need to be created once and do not need to be created each time. Therefore, if the scenario generation device 10 acquires candidate scenarios from an external source or stores them in advance, it does not need to be equipped with a candidate scenario generation unit 21.

[0066] <Variation 2> The scene generation unit 226 may generate driving scenes using driving data included in the driving scenes used for learning as input data. Even if driving data included in the driving scenes used for learning is used as input data, it is not necessarily possible to generate driving scenes that match the driving scenes used for learning. Therefore, this method can also increase the number of driving scenes.

[0067] Furthermore, the multiple driving scenes generated by the scene generation unit 226 may be narrowed down using the same conditions as those used by the real environment scene acquisition unit 22 to extract driving data from the driving data storage unit 40, and used as data to be analyzed.

[0068] <Variation 3> In the embodiment, the evaluation scenario is evaluated by the virtual environment evaluation device 50. However, the evaluation scenario may be evaluated by an actual vehicle. Also, the evaluation scenario may be evaluated by a Hardware In the Loop Simulator / Hardware In the Loop System (HILS), which is an evaluation device that combines hardware and software.

[0069] <Variation 4> The controller 20, 220 and the method described herein may be implemented by a special-purpose computer comprising a processor programmed to perform one or more functions embodied in a computer program. Alternatively, the controller 20, 220 and the method described herein may be implemented by a dedicated hardware logic circuit. Alternatively, the controller 20, 220 and the method described herein may be implemented by one or more special-purpose computers configured by a combination of a processor executing a computer program and one or more hardware logic circuits. The hardware logic circuit may be, for example, an ASIC or FPGA.

[0070] Furthermore, the storage medium for storing the computer program is not limited to a ROM, and the program may be stored in any computer-readable, non-transitory storage medium as instructions to be executed by a computer. For example, the program may be stored in a flash memory. [Explanation of symbols]

[0071] 10: Scenario generation device 20: Control unit 21: Candidate scenario generation unit 22: Real environment scene acquisition unit 23: Filter generation unit 24: Evaluation scenario determination unit 25: Filter 30: Evaluation scenario storage unit 40: Traveling data storage unit 50: Virtual environment evaluation device 200: Scenario generation device 220: Control unit 223: Filter generation unit 226: Scene generation unit

Claims

1. a real environment scene acquisition unit (22) that acquires a real environment scene, which is a scene that occurs in a real environment, from a travel database that stores travel data of a real vehicle; a filter generation unit (23, 223) that generates a filter for filtering candidate scenarios based on a frequency analysis result of analysis target data including the driving data representing the real environment scene; and an evaluation scenario determination unit (24) that uses the filter generated by the filter generation unit to filter the candidate scenarios that are comprehensively generated based on a mathematical model to determine an evaluation scenario.

2. The scenario generation device according to claim 1 , wherein the candidate scenario is determined based on recognition performance, traffic disturbances, and vehicle dynamics performance expressed by a mathematical model.

3. 3. The scenario generation device according to claim 1, The real environment scene acquisition unit extracts the real environment scene from the driving database using extraction logic defined for each vehicle control function.

4. The scenario generation device according to any one of claims 1 to 3, a scene generation unit (226) that receives the driving data stored in the driving database as an input and generates driving data representing a driving scene by a generator that has been trained using the real environment scene acquired by the real environment scene acquisition unit; A scenario generation device, wherein the filter generation unit (223) takes the driving data generated by the scene generation unit as the analysis target data in addition to the driving data representing the real environment scene.

5. At least one processor or circuit: Acquire a real environment scene, which is a scene that occurs in a real environment, from a travel database that stores travel data of a real vehicle; generating a filter for filtering candidate scenarios based on a frequency analysis result of the analysis target data including the driving data representing the real environment scene; A scenario generation method that uses the filter to filter the candidate scenarios that are exhaustively generated based on a mathematical model to determine an evaluation scenario.

Images