Evaluation scenario generation device, evaluation scenario generation method, and autonomous vehicle

Abstract

Provided is an evaluation scenario generation device for efficiently and automatically generating an evaluation scenario that can be executed by a simulator for reproducing a complicated traffic disturbance scenario.　This evaluation scenario generation device comprises: a storage unit that saves a plurality of evaluation scenario elements related to a travel environment of a host vehicle, a plurality of parameters related to the evaluation scenario elements, and a group definition file defining content related to evaluation scenario classifications; an evaluation scenario generation unit that generates a plurality of evaluation scenarios and tag information on the basis of the plurality of evaluation scenario elements and the plurality of parameters; and an evaluation scenario extraction unit that, on the basis of the result of a scene determination acquired by executing a simulation of the evaluation scenario on the basis of the group definition file and the tag information, determines whether the evaluation scenario corresponds to the group definition file.

Description

Evaluation scenario generation device, evaluation scenario generation method, and autonomous driving vehicle

[0001] The present invention relates to an evaluation scenario generation device and an evaluation scenario generation method that automatically generate evaluation scenarios to be used when evaluating and verifying an autonomous driving system, and an autonomous driving vehicle that automatically drives using the generated evaluation scenarios.

[0002] When developing an automated driving system for automobiles, it is necessary to evaluate and verify the safety of the automated driving system at each stage of planning, design, and evaluation. The "Automated Driving Safety Assessment Framework" created by the Japan Automobile Manufacturers Association is known as a guideline that summarizes this evaluation and verification methodology.

[0003] The guidelines propose a variety of traffic disturbance scenarios that incorporate the positional relationships between traffic participants and road shapes, as well as the behavior of traffic participants. Simulating these various traffic disturbance scenarios with a simulator makes it possible to appropriately evaluate and verify the safety of the software and hardware of automated driving systems under development. To simulate a traffic disturbance scenario, a file (hereafter referred to as an "evaluation scenario") that defines the traffic disturbance scenario in a format that can be executed by the simulator (computer) must be prepared in advance.

[0004] Here, a scenario generation device disclosed in Patent Document 1 is known as a conventional technique for automatically generating a large number of evaluation scenarios. For example, the abstract of this document discloses a scenario generation device that, as "a scenario generation device capable of increasing the comprehensiveness of scenarios occurring in a real environment while suppressing the number of scenarios," includes "a real environment scene acquisition unit that acquires real environment scenes, which are scenes that occur in a real environment, from a travel database that stores travel data of real vehicles; a filter generation unit that generates a filter that filters candidate scenarios based on the results of a frequency analysis of analysis target data that includes travel data that indicates the real environment scenes; and an evaluation scenario determination unit that uses the filter generated by the filter generation unit to filter candidate scenarios that are comprehensively generated based on a mathematical model, and determines an evaluation scenario."

[0005] JP 2023-50387 A

[0006] The above guidelines also provide many examples of traffic disturbance scenarios in which the positional relationship between traffic participants and road shapes and the actions of traffic participants are intricately intertwined.

[0007] For example, the blind spot scenario, which is an example of a complex scenario, is a scenario in which three vehicles are traveling: the vehicle itself, a surrounding vehicle, and a vehicle in the blind spot.The scenario is designed to verify whether the autonomous driving system of the vehicle can respond appropriately to a vehicle in the blind spot when, from the driver's perspective or the external sensor of the vehicle, the surrounding vehicle hides the vehicle in the blind spot (blind spot relationship), and the vehicle continues to operate, such as cutting in.

[0008] Another example of a complex scenario is the avoidance operation scenario, which is a scenario in which three vehicles, the subject vehicle, a preceding vehicle, and a surrounding vehicle, are traveling, and when the subject vehicle moves to a rearward area to avoid the surrounding vehicle due to a sudden deceleration of the preceding vehicle, the scenario is intended to verify whether the subject vehicle's autonomous driving system can take secondary avoidance measures against the surrounding vehicle even if the surrounding vehicle moves to the same area.

[0009] As described above, an evaluation scenario is a file that defines a traffic disturbance scenario in a format that can be executed by the simulator, and more specifically, it is a file that defines in detail the road shape, features, topography, the position and movement of the vehicle at each step, and the position and movement of surrounding vehicles at each step.

[0010] In this way, because the evaluation scenario defines the position and behavior of each vehicle to be simulated for each step, the position and behavior of each vehicle at the start of each step can be easily understood at a glance in the evaluation scenario, but it was difficult to confirm the position and behavior of each vehicle during the execution of each step unless the simulation was actually executed.In other words, even if a large number of evaluation scenarios were automatically generated, it was not easy to determine which evaluation scenarios corresponded to complex traffic disturbance scenarios such as blind spot scenarios or avoidance behavior scenarios.

[0011] This means that it has been difficult to automatically generate complex traffic disturbance scenarios, such as blind spot scenarios and avoidance maneuver scenarios, which are key areas of evaluation and verification to ensure the safety of an automated driving system, or to extract only complex traffic disturbance scenarios from a large number of automatically generated evaluation scenarios.As a result, with conventional technology, it has been difficult to fully evaluate and verify the safety of the software and hardware of an automated driving system under development under a variety of traffic disturbance scenarios.

[0012] Therefore, the present invention aims to provide an evaluation scenario generation device and an evaluation scenario generation method that can efficiently and automatically generate a variety of evaluation scenarios that reproduce complex traffic disturbance scenarios, and an autonomous vehicle that runs using the generated evaluation scenarios.

[0013] A representative example of the invention disclosed in the present invention for solving the above-mentioned problems is as follows: That is, the evaluation scenario generation device of the present invention is an evaluation scenario generation device comprising: a storage unit that stores a plurality of evaluation scenario elements related to the driving environment of a vehicle, a plurality of parameters related to the evaluation scenario elements, and a group definition file that defines content related to the classification of evaluation scenarios, an evaluation scenario generation unit that generates a plurality of evaluation scenarios and tag information based on the plurality of evaluation scenario elements and the plurality of parameters, and an evaluation scenario extraction unit that determines whether the evaluation scenario corresponds to the group definition file based on a result of scene determination obtained by executing a simulation of the evaluation scenario based on the group definition file and the tag information.

[0014] According to the present invention, it is possible to efficiently automatically generate a variety of evaluation scenarios that reproduce complex traffic disturbance scenarios. Problems, configurations, and effects other than those described above will become apparent from the following description of the embodiments.

[0015] 1. A configuration diagram of an evaluation scenario generation device according to a first embodiment. A flowchart of an evaluation scenario extraction unit according to the first embodiment. A diagram showing the contents of an evaluation scenario according to the first embodiment. A diagram expressing the contents of an evaluation scenario according to the first embodiment. A diagram showing tag information of an evaluation scenario according to the first embodiment. A diagram showing the contents of a group definition file according to the first embodiment. A diagram showing the result of the processing of step S4 according to the first embodiment. A diagram showing a method for determining a blind spot scenario according to the first embodiment. A diagram showing a method for determining an avoidance operation scenario according to the first embodiment. A diagram showing the results of the processing of steps S8 and S9 according to the first embodiment. A diagram showing the result of the processing of step S10 according to the first embodiment. A diagram showing the result of the processing of an evaluation scenario extraction unit according to the first embodiment. A configuration diagram in which an evaluation scenario generation device according to a second embodiment is mounted on an autonomously driven vehicle. A diagram showing a scene in which the evaluation scenario generation device according to the second embodiment is utilized for vehicle control. A diagram showing the result of scene determination according to the second embodiment. A diagram showing avoidance area candidates according to the second embodiment. A diagram showing the avoidance operation of an autonomous driving control device according to the second embodiment. A configuration diagram in which a server according to a third embodiment and an autonomously driven vehicle are connected by communication.

[0016] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0017] Hereinafter, an evaluation scenario generation device 1 according to a first embodiment will be described with reference to FIGS. 1 to 6C.

[0018] 1 is a configuration diagram of an evaluation scenario generation device 1 according to this embodiment. As shown in the figure, the evaluation scenario generation device 1 includes evaluation scenario elements 2, parameters 3, a group definition file 4, an evaluation scenario generation unit 5, an evaluation scenario extraction unit 6, and an evaluation scenario group 7. Specifically, the evaluation scenario generation device 1 is a computer equipped with hardware such as a calculation unit such as a CPU, a main storage device such as a semiconductor memory, an auxiliary storage device such as a hard disk, and a communication device. The calculation unit executes a predetermined program to realize each functional unit, such as the evaluation scenario generation unit 5 and the evaluation scenario extraction unit 6. Below, the details of each unit will be explained in order, omitting techniques well known in the computer field as appropriate.

[0019] <Various Data Stored in the Storage Unit> As shown in the figure, the storage unit (auxiliary storage device) of the evaluation scenario generating device 1 stores evaluation scenario elements 2, parameters 3, and a group definition file 4.

[0020] The evaluation scenario elements 2 are elements that make up the evaluation scenario and are data pre-registered in the storage unit. These include various elements related to the driving environment of the vehicle, such as road shape (straight, ramp, merging, branching, intersection, etc.), traffic participants (cars, trucks, motorcycles, pedestrians, etc.), traffic participant behavior (gait, acceleration, deceleration, lane changes, following the preceding vehicle, turning right or left, etc.), features (walls, guardrails, houses, buildings, signs, etc.), and topography.

[0021] The parameters 3 are parameters that specifically define the evaluation scenario elements 2, and are data that are registered in advance in the storage unit. These include numerical values, ranges, and states for defining each element.

[0022] The group definition file 4 is a file that defines the evaluation scenario group 7 and is data that is pre-registered in the storage unit. This file contains standards, specifications, guidelines, etc. for classifying the evaluation scenarios generated by the evaluation scenario generation unit 5.

[0023] <Evaluation scenario generation unit 5> The evaluation scenario generation unit 5 generates multiple evaluation scenarios and tag information (described later) based on the input evaluation scenario elements 2 and parameters 3. If possible, the evaluation scenario generation unit 5 generates all evaluation scenarios and tag information that can be generated from combinations of the evaluation scenario elements 2 and parameters 3. However, if generating all of them is not realistic considering the time required for simulation, etc., the evaluation scenario generation unit 5 generates a randomly combined number (a limited number) of evaluation scenarios and tag information while maintaining comprehensiveness. In any case, the evaluation scenario generation unit 5 generates a large number of evaluation scenarios and tag information (hereinafter referred to as large-scale evaluation scenarios and large-scale tag information) that are sufficient to fully evaluate and verify the safety of the autonomous driving system.

[0024] <Evaluation Scenario Extraction Unit 6> The evaluation scenario extraction unit 6 receives a large number of evaluation scenarios, a large amount of tag information, and the group definition file 4. The evaluation scenario extraction unit 6 then determines whether each of the evaluation scenarios generated by the evaluation scenario generation unit 5 is a desired evaluation scenario (e.g., blind spot scenario) defined in the group definition file 4, and outputs the corresponding group of evaluation scenarios as an evaluation scenario group 7.

[0025] 1, the evaluation scenario extraction unit 6 includes an evaluation scenario acquisition unit 6a, an evaluation scenario execution unit 6b, a scene determination unit 6c, a tag information acquisition unit 6d, a group definition file acquisition unit 6e, an evaluation scenario analysis unit 6f, and an evaluation scenario composition unit 6g. Each of these will be described in detail later, but an overview is as follows.

[0026] The evaluation scenario acquisition unit 6a acquires and manages a large number of evaluation scenarios and tag information from the evaluation scenario generation unit 5.

[0027] The evaluation scenario execution unit 6b includes a simulator that simulates the evaluation scenario of the analysis target.

[0028] The scene determination unit 6c performs scene determination on the simulation results of the evaluation scenario execution unit 6b.

[0029] The tag information acquisition unit 6d acquires tag information of the analysis target from the evaluation scenario acquisition unit 6a, and also edits the contents of the tag information as necessary.

[0030] The group definition file acquisition unit 6e acquires the group definition file 4 and adds information to the group definition file as necessary.

[0031] The evaluation scenario analysis unit 6f refers to the tag information acquired by the tag information acquisition unit 6d and the group definition file 4 acquired by the group definition file acquisition unit 6e to determine whether the evaluation scenario to be analyzed is the desired evaluation scenario (e.g., blind spot scenario) defined in the group definition file 4.

[0032] The evaluation scenario composition unit 6g searches for group information in the large amount of tag information managed by the evaluation scenario acquisition unit 6a, extracts evaluation scenarios that match the group information defined in the group definition file, and generates an evaluation scenario group 7.

[0033] <Evaluation Scenario Group 7> The evaluation scenario group 7 is a group of evaluation scenarios output by the evaluation scenario configuration unit 6g, and is composed of only the desired evaluation scenarios (for example, blind spot scenarios) defined in the group definition file 4. Furthermore, the evaluation scenario group 7 is input into a separately provided simulator (not shown) for the autonomous driving system, and traffic disturbance scenarios are reproduced in a simulation, for use in evaluating and verifying the safety of the autonomous driving system.

[0034] <Flowchart of Evaluation Scenario Extraction Processing> Fig. 2 is a flowchart of the processing executed by the evaluation scenario extraction unit 6. Each step of the flowchart will be described below.

[0035] <<Step S1>> In this step, the evaluation scenario acquisition unit 6a acquires a large number of evaluation scenarios and a large amount of tag information generated by the evaluation scenario generation unit 5.

[0036] FIG. 3A is a diagram showing an example of an evaluation scenario generated by the evaluation scenario generation unit 5. This evaluation scenario is an example in which the road shape is a ramp (curve) and there are three traffic participants: the subject vehicle V0 and nearby vehicles V1 and V2, and the positions and movements of each vehicle are defined for each step. For convenience, FIG. 3A only briefly describes the movements of the subject vehicle V0 and nearby vehicles V1 and V2, and omits the others. FIG. 3B is a conceptual diagram that expresses the evaluation scenario of FIG. 3A in a plan view.

[0037] The evaluation scenario in FIG. 3A defines the behavior of three vehicles step by step, and the simulation time is 60 seconds. The behavior of each vehicle is defined by parameters for longitudinal behavior, lateral behavior, and duration. The behavior of each vehicle will be explained using FIGS. 3A and 3B. First, the host vehicle V0 travels in the second lane in the center of the ramp at 100 km / h for 60 seconds. The peripheral vehicle V1 is located in front of the host vehicle V0, travels in the second lane at 80 km / h for 30 seconds, changes lanes to the left in 5 seconds, and then travels in the third lane on the far left at 80 km / h for 25 seconds. The peripheral vehicle V2 is located in front of the peripheral vehicle V1, travels in the first lane on the far right at 80 km / h for 28 seconds, changes lanes to the left in 5 seconds, and then travels in the second lane at 80 km / h for 27 seconds.

[0038] The aim of creating the evaluation scenario of Figure 3A is to recreate a scene in which, as the host vehicle V0 approaches the surrounding vehicle V1 and the distance between the vehicles becomes shorter, the surrounding vehicle V2 is hidden by the surrounding vehicle V1, creating a blind spot position relationship, and the surrounding vehicles V1 and V2 then change lanes to the left, causing the surrounding vehicle V2 to suddenly appear in front of the host vehicle V0, i.e., a blind spot scenario. However, as mentioned above, simply looking at the contents of the evaluation scenario of Figure 3A does not reveal the movements of each vehicle during step execution, making it impossible to determine whether complex vehicle movements such as those in a blind spot scenario will occur.

[0039] Figure 3C shows an example of tag information for the evaluation scenario in Figure 3A. The tag information is information created at the same time as the evaluation scenario is generated by the evaluation scenario generation unit 5, and the evaluation scenario is associated with the tag information using a generation number (e.g., "00001") issued by the evaluation scenario generation unit 5. As shown in Figure 3C, the content of the tag information includes the content and parameter values ​​of the evaluation scenario elements used to generate the evaluation scenario, so the content of the evaluation scenario can be roughly determined from the tag information. Note that the "group" information not registered in Figure 3C is information that will be added after the scene determination process described below.

[0040] The above has explained an example of an evaluation scenario and tag information, but the large number of evaluation scenarios and large amount of tag information are a large number of evaluation scenarios and a large amount of tag information related to them, each of which differs in road shape, vehicle position, longitudinal behavior, lateral behavior, measurement time, number of steps, number of vehicles, etc., and are generated in large quantities by the evaluation scenario generation unit 5 and stored in the evaluation scenario acquisition unit 6a, without it being clear whether they are blind spot scenarios or avoidance operation scenarios.

[0041] <<Step S2>> In this step, the group definition file acquisition unit 6e acquires the group definition file 4.

[0042] 4A is a diagram showing an example of a group definition file, in which a blind spot scenario when the road shape is "ramp" is defined as a desired evaluation scenario. Note that for convenience, the group definition file is shown in a simple table in FIG. 4A, but the format of the group definition file is not limited to the example shown in FIG. 4A.

[0043] In Figure 4A, column items are checkpoints that determine the content of the evaluation scenario, such as "road," "traffic," and "vehicle behavior," and row items are checkpoints that determine whether or not the evaluation scenario is desired, such as "ramp," "lane keeping," and "cut-in." After marking the checkpoints in the column items as applicable or inapplicable, by hitting the "group" that corresponds to the checkpoint in the row items, it is possible to determine whether or not the evaluation scenario to be analyzed is the desired evaluation scenario defined in the group definition file. Note that in Figure 4A, the content that corresponds to the "group" in the row items is left blank, and the content that does not apply is shaded.

[0044] <<Step S3>> In this step, the tag information acquiring unit 6d acquires one tag information to be analyzed from among the multiple pieces of tag information acquired by the evaluation scenario acquiring unit 6a.

[0045] <<Step S4>> In this step, the evaluation scenario analysis unit 6f adds the contents of the tag information acquired by the tag information acquisition unit 6d to the group definition file acquired by the group definition file acquisition unit 6e.

[0046] 4B shows the results of the processing in step S4, and for convenience sake, the illustration is simplified. In FIG. 4B, the circles indicate items where the column items and tag information match.

[0047] <<Step S5>> In this step, the evaluation scenario analysis unit 6f determines whether there are any missing entries in the column items of the group definition file after the processing of Step S4. If there are any missing entries (Step S5: Yes), the process proceeds to Step S6; if there are no missing entries (Step S5: No), the process proceeds to Step S9.

[0048] In the example of FIG. 4B, since there is a lack of input content in the column item "scene" of the group definition file, the process proceeds to step S6.

[0049] <<Step S6>> In this step, the evaluation scenario analysis unit 6f simulates the evaluation scenario of the analysis target using the evaluation scenario execution unit 6b. The evaluation scenario of the analysis target is selected from the evaluation scenario acquisition unit 6a using the generation number of the tag information acquired by the tag information acquisition unit 6d, and input to the evaluation scenario execution unit 6b.

[0050] <<Step S7>> The scene determination unit 6c receives the simulation results from the evaluation scenario execution unit 6b and determines whether the evaluation scenarios to be analyzed include a blind spot scenario and an avoidance operation scenario. The method for determining each scenario will be described below.

[0051] <<<<Method for Determining Blind Spot Scenario>>> Figure 5A is a conceptual diagram illustrating a method for determining a blind spot scenario. In Figure 5A, the host vehicle V0 and a preceding peripheral vehicle V1 are traveling in the center second lane, and a peripheral vehicle V2 preceding the peripheral vehicle V1 is in the middle of changing lanes from the rightmost first lane to the left. At this time, the inter-vehicle distance between the host vehicle V0 and the peripheral vehicle V1 is short, and furthermore, the peripheral vehicle V2 is in a blind spot due to the influence of the peripheral vehicle V1, so that the peripheral vehicle V2 cannot be seen from the host vehicle V0. Therefore, the scene determination unit 6c determines the scenario in Figure 5A to be a blind spot scenario.

[0052] An example of this determination method is as follows. First, a reference point is set on the vehicle V0, and a scan is performed radially from the reference point to detect targets. Here, if two or more targets are detected on the scan line (e.g., scans L2 and L3), it can be determined that a target detected far from the reference point is in a blind spot relationship with a target detected close to the reference point. On the other hand, if one or less targets are detected on the scan line (e.g., scan L1), it can be determined that there is no blind spot relationship. While the above method can be used to determine a blind spot scenario, it is not necessary to be limited to the described method, and other alternative methods may be used if available.

[0053] <<<<Method for Determining Avoidance Operation Scenario>>> Fig. 5B is a conceptual diagram illustrating a method for determining an avoidance operation scenario. In Fig. 5B, the host vehicle V0 and the preceding peripheral vehicle V1 are traveling in the center second lane, and the following peripheral vehicle V2 is traveling in the first lane. Then, at time t1, the peripheral vehicle V1 suddenly decelerates, and the host vehicle V0 attempts to move to an open space behind the peripheral vehicle V2 to avoid a collision. Furthermore, at time t2, the peripheral vehicle V2 changes lanes to the left, and the inter-vehicle distance between the host vehicle V0, which has moved backward, and the peripheral vehicle V2 suddenly becomes shorter. Therefore, the host vehicle V0 needs to take a secondary avoidance action against the peripheral vehicle V2, and the scene determination unit 6c determines the scenario of Fig. 5B as an avoidance operation scenario.

[0054] An example of this determination method is as follows. First, an empty area A is set behind the host vehicle V0, and target detection is performed in area A using an on-board sensor. Here, if a target other than the host vehicle V0 is detected in area A (for example, if a peripheral vehicle V2 is detected at time t2 in FIG. 5B ), it can be determined that an avoidance operation scenario has occurred. On the other hand, if no target other than the host vehicle V0 is detected in area A, it can be determined that an avoidance operation scenario has not occurred. Using the above method, it is possible to determine an avoidance operation scenario, but it is not necessary to be limited to the method described, and any alternative method may be used.

[0055] <<Step S8>> In this step, the evaluation scenario analysis unit 6f adds the contents of the scene determination result acquired from the scene determination unit 6c to the group definition file in the group definition file acquisition unit 6e.

[0056] 6A shows the results of the processing in step S8, and is simplified for convenience. In FIG. 6A, a circle is entered in the "Vehicle in blind spot" column in the "Scene" column. This is because the evaluation scenario in FIG. 3A was simulated by the evaluation scenario execution unit 6b, and after the scene determination unit 6c determined that it was a blind spot scenario, the scene determination result was acquired by the evaluation scenario analysis unit 6f and entered into the group definition file of the group definition file acquisition unit 6e.

[0057] <<Step S9>> In this step, the evaluation scenario analysis unit 6f checks the contents of the group definition file in the group definition file acquisition unit and determines whether it matches any of the "group" line items. If it matches (Step S9: Yes), proceed to Step S10; if it does not match (Step S9: No), return to Step S3.

[0058] 6A also shows the results of the processing in step S9. The group determination method is, for example, to find a group that has a circle in the blank space and no circle in the shaded area. In FIG. 6A, this corresponds to the group "Blind Spot Scenario," "Lamp," and "2." In other words, the evaluation scenario to be analyzed, which was executed in steps S3 to S9 described above, is determined to be the group "Blind Spot Scenario Lamp 2."

[0059] <<Step S10>> In this step, the evaluation scenario analysis unit 6f adds the group information obtained in step S9 to the tag information of the tag information acquisition unit 6d, and the tag information acquisition unit 6d updates the tag information of the analysis target stored in the evaluation scenario acquisition unit 6a to the latest tag information including the group information.

[0060] 6B shows the result of the processing in step S10. In Fig. 6B, "Blind Spot Scenario Lamp 2," which is the group information determined in step S9, has been added to the "Group" item that was "Unregistered" in Fig. 3C.

[0061] <<Step S11>> In this step, the evaluation scenario analysis unit 6f determines whether all of the large number of evaluation scenarios have been analyzed. If all of the evaluation scenarios have been analyzed (Step S11: Yes), the process proceeds to Step S11. If not all of the evaluation scenarios have been analyzed (Step S11: No), the process returns to Step S3.

[0062] <<Step S12>> In this step, the evaluation scenario composition unit 6g references group information from the large amount of tag information in the evaluation scenario acquisition unit 6a, extracts evaluation scenarios that correspond to each common group name, composes them as evaluation scenario groups, and outputs the evaluation scenario group 7. This makes it possible to compose an evaluation scenario group 7 that collects only blind spot scenarios, or an evaluation scenario group 7 that collects only avoidance operation scenarios, for example.

[0063] The above is an explanation of the flowchart in Fig. 2 executed by the evaluation scenario extraction unit 6. This process makes it possible to efficiently extract specific complex traffic disturbance scenarios (blind spot scenarios and avoidance operation scenarios) from the large number of evaluation scenarios generated by the evaluation scenario generation unit 5, and therefore the evaluation scenario generation device 1 of this embodiment can efficiently automatically generate a group that includes only specific traffic disturbance scenarios.

[0064] <Example of Evaluation Scenario Group Output by Evaluation Scenario Extraction Unit 6> Figure 6C is a conceptual diagram showing the processing results of the evaluation scenario extraction unit 6, illustrating the extraction of "Blind Spot Scenario Lamp 2" used in the explanation of the flowchart in Figure 2. In this example, the vertical axis of the graph represents the distance between the subject vehicle V0 and the peripheral vehicle V1, and the horizontal axis represents the direction between the subject vehicle V0 and the peripheral vehicle V2. Region R1 in the figure indicates a scenario that was determined not to be a blind spot scenario among the large number of evaluation scenarios generated by the evaluation scenario generation unit 5. Region R2 in the figure indicates a scenario that was determined to be a blind spot scenario among the large number of evaluation scenarios generated by the evaluation scenario generation unit 5.

[0065] 2, the evaluation scenario extraction unit 6 extracts evaluation scenarios that fall within region R2 from a large number of evaluation scenarios and outputs them as evaluation scenario group 7. As explained above, the evaluation scenarios included in evaluation scenario group 7 have been subjected to accurate content analysis, such as scene determination through simulation, and therefore evaluation scenarios that reproduce complex traffic disturbances can be automatically generated with high accuracy.

[0066] <Comparison of Scenario Generation Efficiency Between Conventional Technology and the Present Invention> Here, we will explain the scenario generation efficiency of this embodiment. Table 1 shows the values ​​used when estimating the scenario generation efficiency, and for convenience, all values, such as the number and time required for creation, are assumed.

[0067]

[0068] Table 2 shows an example of the results of trial calculations of the time required to create 400 blind spot scenarios for each evaluation scenario creation method. Here, evaluation scenario creation method M1 is a scenario creation method in which an average user of the autonomous driving system simulator software generates 400 blind spot scenarios by manually entering all steps. Evaluation scenario creation method M2 is a scenario creation method in which 1,000 evaluation scenarios are created by the evaluation scenario creation unit 5 of this embodiment, and then the average user manually checks the scenes to extract 400 blind spot scenarios. Evaluation scenario creation method M3 is a scenario creation method in which 1,000 evaluation scenarios are created by the evaluation scenario creation unit 5, and then 400 blind spot scenarios are extracted by the evaluation scenario extraction unit 6.

[0069]

[0070] In Table 2, efficiency indicates the number of evaluation scenarios created per hour. Therefore, in the example of Table 2, the efficiency of evaluation scenario creation method M3 is approximately 177 times that of evaluation scenario creation method M1 and approximately 29 times that of evaluation scenario creation method M2. Since this embodiment corresponds to evaluation scenario creation method M3, it can be seen from Table 2 that the evaluation scenario generation device 1 of this embodiment can efficiently automatically generate evaluation scenarios that reproduce complex traffic disturbance scenarios.

[0071] Next, an automatically driven vehicle according to a second embodiment of the present invention will be described with reference to Fig. 7 and Fig. 8A to Fig. 8D. Note that, in the following, overlapping descriptions of points common to the first embodiment will be omitted.

[0072] 7 is a configuration diagram of an autonomous driving vehicle (hereinafter referred to as "vehicle V0") of this embodiment, which uses the evaluation scenario generation device 1 for autonomous driving control. As shown here, in addition to the evaluation scenario generation device 1, the vehicle V0 is equipped with an external sensor 11, a map unit 12, a communication unit 13, a vehicle sensor 14, an autonomous driving control device 15, and a vehicle drive unit 16. Each of these will be explained in turn below.

[0073] The external sensors 11 are a camera sensor and a radar sensor. The camera sensor acquires information about objects detected from an image of the surroundings of the vehicle. The radar sensor acquires information about objects detected from the distance to a predetermined point around the vehicle.

[0074] The map unit 12 is a unit that receives satellite signals from the Global Navigation Satellite System (GNSS) and has a map database and a locator. The map unit 12 calculates the vehicle position using the locator based on the received GNSS signals, selects a surrounding map from the map database, and outputs the vehicle position and the surrounding map.

[0075] The communication unit 13 transmits and receives wireless communications between vehicles (vehicle-to-vehicle communications) and between vehicles and infrastructure facilities (road-to-vehicle communications).

[0076] The vehicle sensor 14 acquires information relating to the running of the vehicle, such as engine operation information of the vehicle drive device 16, brake operation information, and steering operation information.

[0077] The autonomous driving control device 15 determines whether autonomous driving is necessary and controls the vehicle. An autonomous driving determination unit of the autonomous driving control device 15 inputs object information from sensors, vehicle position information, and surrounding map information, and calculates actions, routes, speeds, etc. required for autonomous driving. A vehicle control unit of the autonomous driving control device 15 inputs the calculation results of the autonomous driving determination unit and controls the vehicle drive device 16.

[0078] The vehicle drive system 16 includes and controls the engine, brakes, and steering.

[0079] The evaluation scenario generation device 1 is shown in a simplified form and includes evaluation scenario elements 2, parameters 3, a group definition file 4, an evaluation scenario generation unit 5, and an evaluation scenario extraction unit 6. Note that the explanation of each unit has been given in Example 1, so the explanation will be omitted here.

[0080] <Operation of the Vehicle V0> The operation of the vehicle V0 in FIG. 7 will be described below with reference to specific examples in FIGS. 8A to 8D.

[0081] 8A shows an example of a scene in which the evaluation scenario generation device 1 is used for vehicle control. In Fig. 8A, the host vehicle V0 and a preceding peripheral vehicle V1 are traveling in lane 2, while peripheral vehicles V2 and V3 are traveling in lane 1. In this situation, if the preceding peripheral vehicle V1 suddenly decelerates, the host vehicle V0 must perform an avoidance operation to avoid a collision with the peripheral vehicle V1.

[0082] Therefore, in order to determine a safe avoidance operation, the automatic driving control device 15 acquires information for identifying the situation in Fig. 8A, such as road shape and the position, speed, direction, type, and shape of the vehicle and surrounding objects, from the external sensor 11, map unit 12, communication unit 13, and vehicle sensor 14, and outputs this information to the evaluation scenario generation device 1. In addition, the automatic driving control device 15 outputs a group definition file 4 that defines the avoidance operation to the evaluation scenario generation device 1.

[0083] Then, based on the input current status information, the evaluation scenario generation device 1 uses the evaluation scenario elements 2, parameters 3, and evaluation scenario generation unit 5 to generate a large number of evaluation scenarios and a large amount of tag information that comprehensively reproduce the behavior of surrounding objects that may occur in the future. However, if the number of evaluation scenarios is too large, it will take a long time to generate them, so they are generated by limiting the number to a practical number. Furthermore, if the simulation time of the evaluation scenario is set longer than necessary, it will take a long time to execute the simulation in the later stages, so they are generated with the minimum simulation time setting required.

[0084] The evaluation scenario extraction unit 6 performs a simulation of the evaluation scenario using the large number of evaluation scenarios and large amount of tag information generated by the evaluation scenario generation unit 5, and the group definition file 4 obtained from the automatic driving control device 15, and uses the scene determination process of the avoidance operation scenario to determine whether the evaluation scenario to be analyzed results in a safe avoidance operation.

[0085] For example, if the analysis scenario matches the major category "Scene" and minor category "No intrusion into avoidance area" in Figure 4A, it can be determined that a safe avoidance action was taken. Note that the simulation here places emphasis on simulation speed, and is performed at ultra-high speed, specializing in the processing required for scene determination.

[0086] 8B is a conceptual diagram showing the results of scene determination by the simulation performed by the evaluation scenario extraction unit 6. It is necessary to extract an evaluation scenario in which the host vehicle V0 is in an avoidance area where it will not collide with any of the surrounding vehicles and can safely perform an avoidance operation.

[0087] In Figure 8B, the avoidance area A1 is an area into which the host vehicle V0 and the peripheral vehicle V2 enter at the same time, so the avoidance operation is evaluated as a failure. The avoidance area A2 is an area into which the host vehicle V0 and the peripheral vehicle V2 enter at the same time, so the avoidance operation is evaluated as a failure. The avoidance area A3 is an area into which only the host vehicle V0 enters, so the avoidance operation is evaluated as a success. The avoidance area A4 is an area into which only the host vehicle V0 enters, so the avoidance operation is evaluated as a success. The avoidance area A5 is an area into which the host vehicle V0 and the peripheral vehicle V3 enter at the same time, so the avoidance operation is evaluated as a failure.

[0088] That is, the evaluation scenario extraction unit 6 in Example 1 outputs only evaluation scenarios in which the host vehicle and surrounding vehicles simultaneously enter an empty area in order to extract only avoidance operation scenarios (specific traffic disturbance scenarios) that are the targets of evaluation and verification of the autonomous driving system from the large number of evaluation scenarios generated by the evaluation scenario generation unit 5, whereas the evaluation scenario extraction unit 6 in Example 2 outputs evaluation scenarios in which only the host vehicle enters an empty area in order to extract only options that can be adopted to maintain safe autonomous driving of the host vehicle V0.

[0089] Therefore, the evaluation scenario extraction unit 6 of this embodiment determines whether the scene determination result matches the avoidance operation (success) defined in the group definition file, and then updates the tag information using the group information. Once the analysis of all evaluation scenarios is complete, the evaluation scenario extraction unit 6 references the group information in the large amount of tag information to extract multiple evaluation scenarios that reproduce safe avoidance operations, and generates an evaluation scenario group (hereinafter, avoidance area candidate) including information such as the position and shape of the avoidance area and the longitudinal behavior, lateral behavior, and duration of the host vehicle V0 when moving to the avoidance area, obtained by the scene determination process using simulation, and outputs this to the automatic driving control device 15. Note that although the phrase "when the analysis of all evaluation scenarios is complete" is used here, it is also possible to immediately generate avoidance area candidate as soon as even one evaluation scenario that reproduces a safe avoidance operation can be extracted based on the urgency level of the host vehicle V0, and output this to the automatic driving control device 15 as soon as possible.

[0090] 8C is a conceptual diagram showing avoidance area candidates generated by the evaluation scenario extraction unit 6. As shown here, the evaluation scenario extraction unit 6 extracts only the evaluation scenarios for avoidance areas A3 and A4 where safe avoidance operations are possible.

[0091] The automatic driving control device 15 inputs the avoidance area candidates, determines the current situation of the host vehicle V0, the preceding surrounding vehicle V1, surrounding vehicle V2, surrounding vehicle V3, and other surrounding vehicles, and selects the optimal avoidance area A4 from the avoidance area candidates.The automatic driving control device 15 then refers to information contained in the avoidance area candidates, such as the position and shape of the avoidance area A4, and the longitudinal behavior, lateral behavior, and duration of the host vehicle V0 when moving to the avoidance area A4, to optimize the area, and executes the avoidance operation by actually controlling the engine, brakes, and steering of the vehicle drive device 16 from the vehicle control unit.

[0092] 8D shows a situation in which the automatic driving control device 15 refers to the avoidance area candidate and controls the vehicle drive device 16 to perform an avoidance operation. In this scene, the preceding surrounding vehicle V1 suddenly decelerates while traveling in the second lane, causing the distance between the preceding surrounding vehicle V1 and the following vehicle to suddenly decrease. The host vehicle V0 changes lanes to the third lane before colliding with the preceding surrounding vehicle V1. Therefore, the host vehicle V0 can safely continue automatic driving.

[0093] According to the present embodiment described above, in a scene where the host vehicle needs to take evasive action due to a sudden deceleration of a preceding vehicle, a safe avoidance area candidate is generated, and the host vehicle is controlled by inputting the avoidance area candidate, so that the host vehicle can take evasive action and prevent an accident from occurring.

[0094] Next, an automatic driving system according to a third embodiment of the present invention will be described with reference to Fig. 9. Note that a duplicated description of points common to the second embodiment will be omitted.

[0095] In Example 2, the subject vehicle V0 is equipped with an evaluation scenario generation device 1, but in order to calculate candidate avoidance areas to be used for autonomous driving in real time, it is necessary to provide the evaluation scenario generation device 1 with high computing performance. Therefore, the subject vehicle V0 in Example 2 tends to be expensive due to the installation of a high-performance evaluation scenario generation device 1.

[0096] Therefore, in this embodiment, an evaluation scenario generation device 1 with high computing performance is configured as a server 20 on the cloud, and the evaluation scenario generation device 1 is removed from the autonomously driving vehicle (host vehicle V0), thereby making it possible to achieve autonomous driving equivalent to that in embodiment 2 with a relatively inexpensive vehicle configuration. Needless to say, the server 20 (evaluation scenario generation device 1) in this embodiment can communicate with multiple autonomously driving vehicles (host vehicle V0) in parallel, and there is no need to provide a server 20 for each autonomously driving vehicle.

[0097] 9 is a configuration diagram of the autonomous driving system of this embodiment, in which the server 20 and the host vehicle V0 are connected. Hereinafter, the server 20 and the host vehicle V0 will be described in detail with reference to FIG.

[0098] <Server 20> The server 20 of this embodiment includes an evaluation scenario generation device 1 and a communication unit 21. As in the second embodiment, the evaluation scenario generation device 1 includes evaluation scenario elements 2, parameters 3, a group definition file 4, an evaluation scenario generation unit 5, and an evaluation scenario extraction unit 6. The communication unit 21 is a unit for wireless communication with the communication unit 13 of the host vehicle V0, and transmits current status information received from the host vehicle V0 to the evaluation scenario generation device 1, and also transmits avoidance area candidates received from the evaluation scenario generation device 1 to the host vehicle V0.

[0099] <Own vehicle V0> The own vehicle V0 of this embodiment is basically configured such that the evaluation scenario generation device 1 is excluded from the own vehicle V0 of embodiment 2, but the communication unit 13 of this embodiment differs from the communication unit 13 of embodiment 2 in the following respects. That is, the communication unit 13 of this embodiment transmits current status information, which is the output of external sensors 11, etc., to the communication unit 21 of the server 20, and receives avoidance area candidates calculated by the evaluation scenario generation device 1 of the server 20 from the communication unit 21. Note that the methods of generating and using the current status information and avoidance area candidates are the same as in embodiment 2, so a duplicated explanation will be omitted in this embodiment.

[0100] According to the present embodiment described above, even if the vehicle V0 does not have the evaluation scenario generation device 1 installed, it can achieve automatic driving control equivalent to that of Example 2 by working in conjunction with the evaluation scenario generation device 1 in the server 20.

[0101] The present invention is not limited to the above-described embodiments, and includes various modifications and equivalent configurations within the spirit of the claims. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to configurations including all of the described configurations. Furthermore, part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Furthermore, the configuration of another embodiment may be added to the configuration of one embodiment. Furthermore, part of the configuration of each embodiment may be added, deleted, or replaced with other configurations.

[0102] Furthermore, the aforementioned configurations, functions, processing units, processing means, etc. may be realized in part or in whole in hardware, for example by designing them as integrated circuits, or may be realized in software by a processor interpreting and executing a program that realizes each function.

[0103] Information such as programs, tables, and files that realize each function can be stored in a storage device such as a memory, hard disk, or SSD (Solid State Drive), or in a recording medium such as an IC card, SD card, or DVD.

[0104] In addition, the control lines and information lines shown are those that are considered necessary for the explanation, and do not necessarily show all the control lines and information lines that are necessary for implementation. In reality, it can be considered that almost all components are interconnected.

[0105] REFERENCE SIGNS LIST 1 Evaluation scenario generation device 2 Evaluation scenario element 3 Parameters 4 Group definition file 5 Evaluation scenario generation unit 6 Evaluation scenario extraction unit 6a Evaluation scenario acquisition unit 6b Evaluation scenario execution unit 6c Scene determination unit 6d Tag information acquisition unit 6e Group definition file acquisition unit 6f Evaluation scenario analysis unit 6g Evaluation scenario configuration unit 7 Evaluation scenario group V0 Vehicle 11 External sensor 12 Map unit 13 Communication unit 14 Vehicle sensor 15 Automatic driving control device 16 Vehicle drive device 20 Server 21 Communication unit V1, V2, V3 Peripheral vehicles

Claims

1. An evaluation scenario generation device comprising: a memory unit that stores a plurality of evaluation scenario elements related to the driving environment of a vehicle, a plurality of parameters related to the evaluation scenario elements, and a group definition file that defines contents related to the classification of evaluation scenarios; an evaluation scenario generation unit that generates a plurality of evaluation scenarios and tag information based on the plurality of evaluation scenario elements and the plurality of parameters; and an evaluation scenario extraction unit that determines whether the evaluation scenario is an evaluation scenario corresponding to the group definition file based on a result of scene determination obtained by executing a simulation of the evaluation scenario based on the group definition file and the tag information.

2. In the evaluation scenario generating device as described in claim 1, the evaluation scenario extraction unit comprises: a scene determination unit which inputs the tag information into the group definition file and performs the scene determination if there is a deficiency in the input contents of the group definition file; and an evaluation scenario analysis unit which inputs the results of the scene determination into the missing parts of the group definition file and determines whether the evaluation scenario is an evaluation scenario corresponding to the group definition file.

3. An evaluation scenario generating device as described in claim 2, characterized in that when the evaluation scenario analysis unit determines that the evaluation scenario is an evaluation scenario corresponding to the group definition file, it adds group information of the evaluation scenario corresponding to the group definition file to the tag information.

4. In the evaluation scenario generating device described in claim 3, the evaluation scenario extraction unit is characterized in that it has an evaluation scenario construction unit that extracts evaluation scenarios containing the desired group information from the multiple tag information, constructs them as one evaluation scenario group, and outputs them.

5. In the evaluation scenario generating device described in claim 3, the scene determination unit determines, as the scene determination, whether the host vehicle and a second surrounding vehicle are in a blind spot relationship based on the relationship between the host vehicle and a first surrounding vehicle, and the evaluation scenario analysis unit adds information indicating a blind spot scenario to the tag information as the group information if the scene determination results in the blind spot relationship.

6. In the evaluation scenario generating device described in claim 3, the scene determination unit determines, as the scene determination, based on the relationship between the host vehicle and a first surrounding vehicle, whether the host vehicle moves into an avoidance area and a second surrounding vehicle also moves to the avoidance area at the same time, and the evaluation scenario analysis unit adds, as the group information, information indicating that the scenario is an avoidance operation scenario to the tag information when the scene determination results in the host vehicle and the second surrounding vehicle moving into the avoidance area at the same time.

7. An evaluation scenario generating device as claimed in claim 1, characterized in that the evaluation scenario elements include elements relating to roads, traffic, the behaviour of the vehicle, traffic behaviour, features and topography.

8. An evaluation scenario generating apparatus according to claim 1, wherein the parameters include values, ranges, and states associated with the evaluation scenario elements.

9. An evaluation scenario generating device according to claim 1, wherein the group definition file is created based on information including standards, specifications, and guidelines.

10. An autonomous vehicle comprising an evaluation scenario generation device and an autonomous driving control device, wherein the evaluation scenario generation device comprises: a memory unit storing a group definition file defining a plurality of evaluation scenario elements related to the driving environment of the vehicle, a plurality of parameters related to the evaluation scenario elements, and contents related to the classification of evaluation scenarios; an evaluation scenario generation unit generating a plurality of evaluation scenarios and tag information based on the plurality of evaluation scenario elements and the plurality of parameters; and an evaluation scenario extraction unit determining whether the evaluation scenario is an evaluation scenario corresponding to the group definition file based on a result of a scene determination obtained by executing a simulation of the evaluation scenario based on the group definition file and the tag information, wherein the evaluation scenario extraction unit has an evaluation scenario configuration unit extracting an evaluation scenario including group information of a desired evaluation scenario from the plurality of tag information, configuring it as one evaluation scenario group and outputting it; and the autonomous driving control device controlling the autonomous driving of the vehicle using the output result from the evaluation scenario configuration unit.

11. An autonomous driving system comprising a server and an autonomous vehicle, wherein the server comprises: a communication unit for communicating with the autonomous vehicle; a memory unit for storing a group definition file that defines a plurality of evaluation scenario elements related to the driving environment of the vehicle, a plurality of parameters related to the evaluation scenario elements, and content related to the classification of evaluation scenarios; an evaluation scenario generation unit for generating a plurality of evaluation scenarios and tag information based on the plurality of evaluation scenario elements and the plurality of parameters; and an evaluation scenario extraction unit for determining whether the evaluation scenario is an evaluation scenario corresponding to the group definition file based on a result of a scene determination obtained by executing a simulation of the evaluation scenario based on the group definition file and the tag information, wherein the evaluation scenario extraction unit has an evaluation scenario configuration unit that extracts an evaluation scenario including group information of a desired evaluation scenario from the plurality of tag information, configures it as one evaluation scenario group, and outputs it; and the autonomous vehicle performs autonomous driving control using the output result of the evaluation scenario configuration unit obtained via the communication unit of the server.

12. An evaluation scenario generating method executed by a computer, comprising: a step of generating a plurality of evaluation scenarios and tag information based on a plurality of evaluation scenario elements and a plurality of parameters related to the evaluation scenario elements; a step of simulating the evaluation scenarios and determining a scene; and a step of determining whether or not the evaluation scenario is a desired evaluation scenario based on the results of the scene determination.

Images