Vehicle control device evaluation device, vehicle control device evaluation system, and vehicle control device evaluation method

Abstract

To efficiently evaluate a vehicle control device in development management.SOLUTION: An evaluation device 100 includes: a loaded vehicle data reading section 103 for taking out loaded vehicle data D1, D2 used for simulating behavior of a vehicle from loaded vehicle data D collected from the vehicle; a map module 104 for storing surrounding map data P at a traveling spot of a vehicle; a vehicle behavior simulation section 105 for reproducing a virtual vehicle and a surrounding map of a traveling spot of the vehicle in a simulation environment on the basis of the loaded vehicle data D1 and the surrounding map data P, and for simulating behavior of the vehicle according to a vehicle control logic of an evaluation object in the simulation environment; a surrounding thing reproduction section 106 for reproducing surrounding things in a simulation environment on the basis of the loaded vehicle data D2; and an evaluation section 107 for evaluating a vehicle control device on the basis of a vehicle behavior simulation result S.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] This application relates to an evaluation device for vehicle control devices, an evaluation system for vehicle control devices, and an evaluation method for vehicle control devices. [Background technology]

[0002] In the development management of vehicle control devices, it is required to eliminate the causes of malfunctions during the pre-debugging process to prevent devices in the development stage from malfunctioning during actual vehicle testing. In this regard, some methods involve reproducing the values ​​of the RAM (Random Access Memory) within the ECU (Electronic Control Unit) of the vehicle control device under analysis in a simulation environment, based on data measured in the actual vehicle, in order to identify the cause of malfunctions that occurred in the actual vehicle (see, for example, Patent Document 1). [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2004-19508 [Overview of the project] [Problems that the invention aims to solve]

[0004] However, when the vehicle control logic of an ECU is modified during development management, and it is anticipated that the vehicle's behavior will change accordingly, it is not immediately clear whether the same problem will occur or whether other problems will occur. It is necessary to re-examine the impact of the modified part of the vehicle control logic. For this reason, there is a problem that conventional technology may not be able to efficiently evaluate the modified ECU, i.e., the vehicle control device.

[0005] This application discloses technology to solve the above-mentioned problems, and aims to provide an evaluation device for vehicle control devices, an evaluation system for vehicle control devices, and an evaluation method for vehicle control devices that enable efficient evaluation of vehicle control devices. [Means for solving the problem]

[0006] The vehicle control device evaluation device disclosed in this application is Based on the vehicle control system before modification. Real-world vehicle data, including at least location information during vehicle operation, is collected from the vehicle, and the evaluation target is used with this real-world vehicle data. After renovation A vehicle control device evaluation device for evaluating vehicle control devices, comprising: a vehicle data extraction unit that extracts vehicle data used to simulate vehicle behavior from collected vehicle data; a map data storage unit that stores surrounding map data showing the surrounding area of ​​the vehicle's travel location; and a device that reproduces a virtual vehicle and the surrounding map of the vehicle's travel location in a simulation environment based on the vehicle data and surrounding map data, and in the simulation environment, After renovation A vehicle behavior simulation unit simulates the behavior of a vehicle according to the vehicle control logic of the vehicle control device, a surrounding object reproduction unit reproduces the surrounding objects of the driving location in the simulation environment based on actual vehicle data, and based on the results of the simulation by the vehicle behavior simulation unit, After renovation The system includes an evaluation unit for evaluating the vehicle control device and an output unit for outputting the results of the evaluation performed by the evaluation unit.

[0007] Furthermore, the evaluation method of the vehicle evaluation device disclosed in this application is Based on the vehicle control system before modification. Real-world vehicle data, including at least location information during vehicle operation, is acquired from the vehicle, and the evaluation target is evaluated using this real-world vehicle data. After renovationA method for evaluating a vehicle control device, comprising the steps of: collecting real vehicle data from a vehicle; extracting real vehicle data from the collected real vehicle data to be used to simulate the behavior of the vehicle; acquiring surrounding map data showing a map of the area where the vehicle is traveling; reproducing a virtual vehicle and a map of the area where the vehicle is traveling in a simulation environment based on the real vehicle data and the surrounding map data; reproducing objects around the travel location in the simulation environment based on the real vehicle data; and in the simulation environment, After renovation The process involves simulating the vehicle's behavior according to the vehicle control logic of the vehicle control device, and then, based on the results of simulating the vehicle's behavior, After renovation The system includes the steps of evaluating a vehicle control device and outputting the results of the evaluation of the vehicle control device. [Effects of the Invention]

[0008] According to the vehicle control device evaluation apparatus, vehicle control device evaluation system, or vehicle control device evaluation method disclosed herein, it is possible to efficiently evaluate a vehicle control device. [Brief explanation of the drawing]

[0009] [Figure 1] This diagram shows an overview of the vehicle control device evaluation system in Embodiment 1. [Figure 2] This is a block diagram showing the evaluation system for the vehicle control device in Embodiment 1. [Figure 3] This is a block diagram showing the evaluation apparatus for the vehicle control device in Embodiment 1. [Figure 4] This figure shows an example of the hardware configuration of the vehicle control device evaluation device in Embodiment 1. [Figure 5] This is a flowchart illustrating the operation of the vehicle control device evaluation system in Embodiment 1. [Figure 6] This is a flowchart showing the operation of the vehicle control device evaluation device in Embodiment 1. [Figure 7]It is a flowchart showing the operation of the evaluation device of the vehicle control device in Embodiment 2. [Figure 8] It is a block diagram showing the evaluation device of the vehicle control device in Embodiment 3. [Figure 9] It is a flowchart showing the operation of the evaluation device of the vehicle control device in Embodiment 3. [Figure 10] It is a block diagram showing the evaluation device of the vehicle control device in Embodiment 4. [Figure 11] It is a block diagram showing the obstacle simulation model generation unit according to Embodiment 4. [Figure 12] It is a flowchart showing the operation of the evaluation device of the vehicle control device in Embodiment 4. [Figure 13] It is a flowchart showing the flow of machine learning by the obstacle simulation model learning unit according to Embodiment 4. [Figure 14] It is a block diagram showing the evaluation device of the vehicle control device in Embodiment 5. [Figure 15] It is a block diagram showing the obstacle simulation model generation unit according to Embodiment 5. [Figure 16] It is a block diagram showing the obstacle simulation model learning unit according to Embodiment 5. [Figure 17] It is a flowchart showing the flow of reinforcement learning by the obstacle simulation model generation unit according to Embodiment 5. [Figure 18] It is a block diagram showing the evaluation device of the vehicle control device in Embodiment 6. [Figure 19] It is a flowchart showing the operation of the evaluation device of the vehicle control device in Embodiment 6. [Figure 20] It is a block diagram showing the scenario change unit according to Embodiment 7. [Figure 21] It is a flowchart showing the operation of the scenario change unit according to Embodiment 7. [Figure 22] It is a block diagram showing the scenario change unit according to another form of Embodiment 7. [Figure 23] It is a block diagram showing the scenario change unit according to Embodiment 8. [Figure 24] This is a block diagram showing the evaluation apparatus for the vehicle control device in Embodiment 9. [Figure 25] This is a flowchart showing the operation of the vehicle control device evaluation device in Embodiment 9. [Figure 26] This diagram shows an overview of the vehicle control device evaluation system in Embodiment 10. [Figure 27] This is a block diagram showing the evaluation system for the vehicle control device in Embodiment 10. [Modes for carrying out the invention]

[0010] Embodiment 1. Embodiment 1 will be described with reference to Figures 1 to 6. Figure 1 is a diagram showing an overview of the vehicle control device evaluation system in Embodiment 1, and Figure 2 is a block diagram of the vehicle control device evaluation system in Embodiment 1. Figure 3 is a block diagram of the vehicle control device evaluation device in Embodiment 1. The evaluation system 1000, that is, the vehicle control device evaluation system, comprises a vehicle data acquisition unit 19 (not shown in Figure 1) mounted on vehicles 10A, 10B, and 10C respectively, which acquires actual vehicle data D described later, and an evaluation device 100, that is, a vehicle control device evaluation device, which receives the actual vehicle data D from vehicles 10A, 10B, and 10C and evaluates the vehicle control device to be evaluated using the actual vehicle data D.

[0011] Vehicles 10A, 10B, and 10C can send and receive their respective vehicle data D via vehicle-to-vehicle communication or communication via a network such as the cloud. However, in Embodiment 1, sending and receiving vehicle data D between vehicles is not mandatory.

[0012] As shown in Figure 2, vehicles 10A, 10B, and 10C are equipped with the above-mentioned communication unit 11, a map module 12 that stores map data of the area around the vehicle (surrounding map data), a positioning module 13, i.e., a position information acquisition unit, that acquires the current location information of the vehicle, a surrounding information acquisition unit 14 that acquires information of the area around the vehicle, an ECU 15, and a vehicle data storage unit 16 that stores vehicle data D. The surrounding information acquisition unit 14 is composed of, for example, a camera that takes images of the area around the vehicle, or a sensor that recognizes the position of surrounding objects.

[0013] Vehicles 10A, 10B, and 10C each acquire their current location information using a positioning module 13 while driving, and observe obstacles and surrounding objects around their vehicles using a surrounding information acquisition unit 14 while driving. In this application, "obstacles" refer to traffic participants that exhibit certain behaviors, such as pedestrians or other vehicles. "Obstacles" may slow down or change direction in response to the behavior of the vehicle. The vehicle may also slow down in response to the behavior of the obstacle. Thus, anything that interacts with the vehicle in traffic is considered an "obstacle." "Surrounding objects" are guardrails, buildings, white lines, etc., and unlike the "obstacles" described above, they do not move and do not interact with the vehicle.

[0014] The actual vehicle data D includes at least the vehicle's position information at each time point. The actual vehicle data D may also include the date, driving data such as the vehicle's speed, acceleration, and direction at each time point, data on surrounding objects observed by the surrounding information acquisition unit 14, the aforementioned obstacle data, and data from within the ECU. It may also include weather, temperature, humidity, and vehicle information for each vehicle (make, model year, grade). The format of the actual vehicle data D is not particularly limited and can be text or images. In the case of images, the actual vehicle data D may include not only the image itself but also the analysis data of the image. In Embodiment 1, the positioning module 13, the surrounding information acquisition unit 14, and the ECU 15 constitute the actual vehicle data acquisition unit 19 that acquires the actual vehicle data D. However, as described above, the actual vehicle data D only needs to include the vehicle's position information at each time point, so the positioning module 13 is essential for the actual vehicle data acquisition unit 19, while the surrounding information acquisition unit 14 and ECU 15 can be omitted as appropriate, for example, if equivalent data can be acquired by the evaluation device 100.

[0015] In Embodiment 1, ECUs 15 are installed in vehicles 10A, 10B, and 10C. However, since vehicles 10A, 10B, and 10C only need to acquire actual vehicle data D while driving and transmit the actual vehicle data D to the evaluation device 100, the vehicle control logic of ECU 15 does not need to be the same as the vehicle control logic of the vehicle control device being evaluated by the evaluation device 100.

[0016] The evaluation device 100 will now be described. In the following explanation, we will use the example of a case where vehicle 10A is the own vehicle, and vehicles 10B and 10C are other vehicles, and the evaluation is performed using the actual vehicle data D of vehicle 10A. As shown in Figures 2 and 3, the evaluation device 100 includes a receiving unit 101 that receives actual vehicle data D, a vehicle log data storage unit 102 that stores the actual vehicle data D as vehicle log data D*, an actual vehicle data retrieval unit 103 that reads the vehicle log data D* from the vehicle log data storage unit 102 according to instructions, a map module 104, i.e., a map data storage unit that stores map data common to the vehicle, including surrounding map data P that shows the surrounding map of the vehicle's driving location (the driving location indicated by the driving data included in the actual vehicle data D), a simulator that generates a simulation environment, a vehicle behavior simulation unit 105 that reproduces a virtual vehicle (vehicle 10A) and a surrounding map of the vehicle's driving location in the simulation environment, and simulates the behavior of the vehicle in accordance with the vehicle control logic of the vehicle control device (not shown) under evaluation in the simulation environment, and a surrounding object reproduction unit 106 that generates surrounding object reproduction data T that reproduces surrounding objects of the vehicle's driving location in the simulation environment based on the surrounding object data included in the actual vehicle data D.

[0017] The vehicle log data storage unit 102 collects the actual vehicle data D received by the receiving unit 101 as needed and stores it as vehicle log data D*. The vehicle log data D* is composed of the actual vehicle data D collected from each vehicle, and the necessary actual vehicle data D is retrieved from the vehicle log data D*. In Embodiment 1, it is basically assumed that a large number of actual vehicle data D will be collected, but the vehicle log data D* may be composed of a single actual vehicle data D. In addition, the vehicle log data D* may include not only the actual vehicle data D but also additional information related to the actual vehicle data D. Furthermore, in Embodiment 1, the communication unit 11 on the vehicle side and the receiving unit 101 on the evaluation device 100 side are used to collect the actual vehicle data D via a network, but the method of collecting the actual vehicle data D may be by other means.

[0018] The vehicle data retrieval unit 103 accesses the vehicle log data storage unit 102 in response to instructions from the vehicle behavior simulation unit 105 and reads vehicle log data D* from the vehicle log data storage unit 102. The vehicle data retrieval unit 103 also retrieves vehicle data D1 requested by the vehicle behavior simulation unit 105 from the retrieved vehicle log data D*. The vehicle data retrieval unit 103 outputs the retrieved vehicle data D1 to the vehicle behavior simulation unit 105. The vehicle data retrieval unit 103 also retrieves vehicle data D2 requested by the surrounding object reproduction unit 106 from the vehicle log data D* and outputs the retrieved vehicle data D2 to the surrounding object reproduction unit 106. Vehicle data D1 is data used to simulate the behavior of the vehicle itself and includes the vehicle's driving data. Vehicle data D2 includes at least data indicating the vehicle's location and is data used to reproduce surrounding objects. The actual vehicle data D2 in Embodiment 1 includes data on objects in the vicinity of the vehicle's location, acquired by the vehicle's surrounding information acquisition unit 14.

[0019] The map module 104 outputs surrounding map data P, which shows a map of the area where the vehicle is traveling, to the vehicle behavior simulation unit 105 in response to instructions from the vehicle behavior simulation unit 105. The surrounding map data P includes information about the terrain and roads. The vehicle behavior simulation unit 105 can use the surrounding map data P to reproduce the terrain and roads of the area where the vehicle is traveling in the simulation environment.

[0020] The vehicle behavior simulation unit 105 acquires the actual vehicle data D1 output by the actual vehicle data extraction unit 103 and the surrounding map data P output by the map module 104. Based on the actual vehicle data D1 and the surrounding map data P, it reproduces a virtual vehicle and a map of its surrounding area in the simulation environment, and simulates the behavior of the virtual vehicle in the simulation environment. When simulating the behavior of the virtual vehicle, the vehicle behavior simulation unit 105 simulates the behavior of the vehicle in accordance with the vehicle control logic of the vehicle control device under evaluation.

[0021] The surrounding object reproduction unit 106 acquires actual vehicle data D2 from the actual vehicle data extraction unit 103 and generates surrounding object reproduction data T based on the actual vehicle data D2. The surrounding object reproduction unit 106 outputs the surrounding object reproduction data T to the vehicle behavior simulation unit 105, which reproduces the surrounding objects of the vehicle's driving location in the simulation environment (the simulation environment generated by the simulator of the vehicle behavior simulation unit 105). This simulates how the vehicle would behave, including the influence of surrounding objects, assuming that the vehicle control device under evaluation was installed on the vehicle.

[0022] The vehicle behavior simulation unit 105 outputs the results of the above-described simulation as the vehicle behavior simulation result S. The evaluation unit 107 evaluates the vehicle control device under evaluation based on the vehicle behavior simulation result S. The evaluation unit 107 evaluates based on criteria such as whether the vehicle does not collide with other vehicles or surrounding objects such as guardrails, whether it does not get too close even if it does not collide, and whether it achieves predetermined targets (for example, changing lanes within a specified time). The evaluation unit 107 outputs the results of the evaluation of the vehicle control device as the evaluation result X. The output unit 108 outputs the evaluation result X to the outside. The output unit 108 may be a display unit such as a display that shows the evaluation result X, or an audio output unit such as a speaker that outputs the evaluation result X as sound, or it may convert the evaluation result X into image data or audio data and output it to an external display device or audio output device.

[0023] In Embodiment 1, all components of the evaluation device 100 are housed within the same device. However, some components may be placed in separate devices, and necessary data transmission and reception may be performed between these devices. For example, the vehicle log data D* may be stored in an external storage device, and the vehicle log data storage unit 102 may be omitted. The actual vehicle data retrieval unit 103 reads the necessary actual vehicle data D from the external storage device.

[0024] Next, the hardware configuration for realizing the evaluation device 100 will be described. Figure 4 is a diagram showing an example of the hardware configuration of the vehicle control device evaluation device in Embodiment 1. The evaluation device 100 mainly consists of a processor 91 and a memory 92 and an auxiliary storage device 93 as main memory. The processor 91 is composed of, for example, a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), a DSP (Digital Signal Processor), or an FPGA (Field Programmable Gate Array). The memory 92 is composed of a volatile storage device such as random access memory, and the auxiliary storage device 93 is composed of a non-volatile storage device such as flash memory or a hard disk. The auxiliary storage device 93 stores a predetermined program to be executed by the processor 91, and the processor 91 reads and executes this program as appropriate and performs various calculations. At this time, the predetermined program is temporarily stored in the memory 92 from the auxiliary storage device 93, and the processor 91 reads the program from the memory 92. The calculations performed by each functional unit shown in Figure 1 are realized by the processor 91 executing the predetermined program as described above. The results of the calculations performed by the processor 91 are temporarily stored in the memory 92, and then stored in the auxiliary storage device 93 according to the purpose of the calculations performed.

[0025] Furthermore, the evaluation device 100 includes an input circuit 94 for receiving various inputs from the outside, an output circuit 95 for outputting evaluation results X and other data to the outside, and a communication circuit 96 for sending and receiving various data such as actual vehicle data D.

[0026] Next, the operation will be described. Figure 5 is a flowchart showing the operation of the vehicle control device evaluation system in Embodiment 1, illustrating the general flow of the entire system. Figure 6 is a flowchart showing the operation of the vehicle control device evaluation device in Embodiment 1, illustrating the flow of evaluation of the vehicle control device by the evaluation device 100. First, the actual vehicle data D of the own vehicle (vehicle 10A) is collected and stored as vehicle log data D*. The own vehicle acquires actual vehicle data D while driving and transmits the acquired actual vehicle data D to the evaluation device 100. The evaluation device 100 collects the actual vehicle data D sent from the own vehicle as needed and stores it in the vehicle log data storage unit 102 as vehicle log data D* (step ST001). At this time, actual vehicle data D of other vehicles (vehicles 10B, 10C) may also be collected and included in the vehicle log data D*. For example, if the actual vehicle data D of other vehicles contains information about the own vehicle or its surroundings (including information about surrounding objects), the actual vehicle data D of other vehicles can be used to simulate the behavior of the own vehicle.

[0027] Next, the vehicle control system is evaluated by simulating the behavior of the vehicle under evaluation in a simulation environment using the collected real-world vehicle data D (step ST002). The detailed flow of step ST002 is shown in Figure 6. First, the vehicle log data D* is read, and the real-world vehicle data D necessary for simulating the behavior of the vehicle is extracted (step ST101). The necessary real-world vehicle data D includes the real-world vehicle data D1 mentioned above and real-world vehicle data D2 necessary for reproducing surrounding objects, such as the vehicle's driving data and data on objects surrounding the vehicle's location.

[0028] Next, the surrounding map data P of the vehicle's location is read (step ST102).

[0029] Next, the vehicle and the map of the vehicle's location are reproduced in the simulation environment (step ST103). A virtual vehicle is reproduced in the simulation environment based on the actual vehicle data D1, and the surrounding map of the vehicle's location is reproduced based on the surrounding map data P. More specifically, in the simulation environment, the virtual vehicle is placed in a space that is reproduced as the same location as the actual vehicle when it is driving.

[0030] Next, the surrounding objects of the driving location are reproduced in the simulation environment (step ST104).

[0031] Next, the vehicle behavior is simulated in the simulation environment according to the vehicle control logic of the vehicle control device under evaluation (step ST105). In the steps up to step ST104, a virtual vehicle is placed in the simulation environment in a space that is reproduced as the actual location where the vehicle is driving, and surrounding objects are also reproduced. From this situation, by simulating the behavior of the virtual vehicle according to the vehicle control logic of the vehicle control device under evaluation, it is possible to simulate how the vehicle would behave if the vehicle control device under evaluation were installed on the actual vehicle (vehicle 10A) while it was driving, taking into account the influence of surrounding objects.

[0032] Next, based on the vehicle behavior simulation result S, which is the result of simulating the behavior of the vehicle itself, the vehicle control device to be evaluated is evaluated (step ST106). The evaluation result X is output (step ST107). As described above, the evaluation of the vehicle control device is performed based on whether the vehicle itself collides with obstacles such as other vehicles or surrounding objects such as guardrails.

[0033] According to Embodiment 1, the vehicle control device can be evaluated efficiently. More specifically, the evaluation device includes: a vehicle data extraction unit that collects vehicle data from a vehicle that includes at least position information during vehicle operation and extracts vehicle data from the collected vehicle data to be used to simulate the vehicle's behavior; a map module that stores surrounding map data showing a map of the area where the vehicle is driving; a vehicle behavior simulation unit that reproduces a virtual vehicle and a map of the area where the vehicle is driving in a simulation environment based on the vehicle data and surrounding map data, and simulates the vehicle's behavior in the simulation environment according to the vehicle control logic of the vehicle control device under evaluation; a surrounding object reproduction unit that reproduces surrounding objects of the driving location in the simulation environment based on the vehicle data; an evaluation unit that evaluates the vehicle control device based on the results of the simulation by the vehicle behavior simulation unit; and an output unit that outputs the results of the evaluation by the evaluation unit. As a result, vehicle data acquired by a vehicle equipped with the vehicle control device before modification is collected, and surrounding objects of the driving location are reproduced in the simulation environment using the collected vehicle data, while the vehicle's behavior is simulated according to the vehicle control logic of the modified vehicle control device. Therefore, it is possible to efficiently verify whether the same problems that occurred before the modification occur again in the vehicle whose behavior has changed due to the modification of the vehicle control logic, and to efficiently evaluate the vehicle control system.

[0034] Embodiment 2. Next, Embodiment 2 will be described with reference to Figure 7. Components identical or corresponding to those in Figures 1 to 6 are denoted by the same reference numerals, and their descriptions are omitted. Embodiment 2 simulates the behavior of a vehicle using actual vehicle data from another vehicle that has driven in the same driving environment as the vehicle in question. Since the components of the evaluation device 100 are the same as in Embodiment 1, only its operation will be described. Figure 7 is a flowchart showing the operation of the vehicle control device evaluation device in Embodiment 2. Furthermore, "same driving environment" could mean "same driving location," or it could mean that even if the driving locations are different, the driving environment is considered the same based on factors such as road shape, weather, temperature, humidity, and road surface conditions during driving.

[0035] First, similar to Embodiment 1, the vehicle log data D* is read, and the actual vehicle data necessary for simulating the behavior of the vehicle is obtained (step ST201).

[0036] Next, check the vehicle log data D* for any actual vehicle data D from other vehicles that have driven in the same driving environment as your own vehicle (step ST202). When you extract the actual vehicle data necessary to simulate the behavior of your own vehicle in step ST201, you also obtain information about the driving environment, including the location of the drive. Therefore, you can search the contents of the vehicle log data D* using words indicating the driving environment as keys.

[0037] If there is actual vehicle data D for another vehicle that is in the same condition, that is, if there is actual vehicle data D for another vehicle that has been driven in the same driving environment as the current vehicle (hereinafter referred to as "same environment driving data"), proceed to step ST204; otherwise, proceed to step ST205 (step ST203).

[0038] If identical environmental driving data exists, other vehicles that acquired identical environmental driving data during driving, i.e., other vehicles that drove in the same driving environment as the own vehicle, are treated as the own vehicle for subsequent processing (step ST204). In other words, the actual vehicle data extraction unit 103 extracts the necessary data from the identical environmental driving data, i.e., the data corresponding to the actual vehicle data D1 and D2 described above.

[0039] Next, the process from step ST102 to step ST107 (from reading the surrounding map data to outputting the evaluation result), as shown in Figure 6, is performed (step ST205). The vehicle behavior simulation unit 105 simulates the vehicle's behavior using the same environmental driving data if the same environmental driving data exists, and if it does not exist, it simulates the vehicle's behavior using the actual vehicle data D of its own vehicle, as in Embodiment 1. Other aspects are the same as in Embodiment 1.

[0040] In Embodiment 2, since the vehicle log data D* is required to include the actual vehicle data D of other vehicles, the actual vehicle data D of other vehicles, vehicles 10B and 10C, also needs to be transmitted to the evaluation device 100. The actual vehicle data D of vehicles 10B and 10C may be transmitted directly from vehicles 10B and 10C to the evaluation device 100, or vehicle 10A may receive the actual vehicle data D of vehicles 10B and 10C via vehicle-to-vehicle communication, and then vehicle 10A may transmit the actual vehicle data D of vehicles 10B and 10C to the evaluation device 100.

[0041] According to Embodiment 2, the same effects as in Embodiment 1 can be obtained. Furthermore, the evaluation of vehicle control systems can be performed more efficiently. More specifically, if the same environmental driving data is included in the vehicle log data, the actual vehicle data extraction unit extracts the same environmental driving data from the vehicle log data, and the vehicle behavior simulation unit uses the same environmental driving data to simulate the vehicle's behavior. As a result, the vehicle behavior is simulated using more actual vehicle data than when only the actual vehicle data of the own vehicle is used for the same driving environment, and the variety of vehicle behavior simulation results also increases. By evaluating the vehicle control system using a wide variety of vehicle behavior simulation results, the evaluation of the vehicle control system can be performed more efficiently.

[0042] Embodiment 3. Next, Embodiment 3 will be described with reference to Figures 8 and 9. Components identical or corresponding to those in Figures 1 to 7 are denoted by the same reference numerals, and their descriptions are omitted. Embodiment 3 differs in that, in simulating the behavior of the vehicle itself, it also simulates the behavior of other vehicles or obstacles such as pedestrians. Figure 8 is a block diagram showing the evaluation device for the vehicle control device in Embodiment 3. The evaluation device 300 includes an obstacle simulation model storage unit 309 that stores the obstacle simulation model M.

[0043] The vehicle data retrieval unit 303 accesses the vehicle log data storage unit 102 in response to instructions from the vehicle behavior simulation unit 305 and reads vehicle log data D* from the vehicle log data storage unit 102. The vehicle data retrieval unit 303 also retrieves vehicle data D1 and vehicle data D3 from the read vehicle log data D* as requested by the vehicle behavior simulation unit 305. The vehicle data retrieval unit 303 outputs the retrieved vehicle data D1 and vehicle data D3 to the vehicle behavior simulation unit 305. The vehicle data retrieval unit 303 also retrieves vehicle data D2 as requested by the surrounding object reproduction unit 106 from the vehicle log data D* and outputs the retrieved vehicle data D2 to the surrounding object reproduction unit 106. Vehicle data D1 and vehicle data D2 are the same as in Embodiment 1. The actual vehicle data D3 is actual vehicle data that includes data on obstacles around the vehicle, and is data on obstacles observed while the vehicle is in motion. However, if data on obstacles observed by vehicles other than the vehicle is collected, that data may also be used. The actual vehicle data D3 includes information such as what kinds of obstacles exist around the vehicle, the number of obstacles, size, position, speed, and direction of movement. In other words, the actual vehicle data D in Embodiment 3 includes data on obstacles around the vehicle. As described above, "obstacles" in this application refer to other vehicles and pedestrians, etc., that interact with the vehicle.

[0044] The vehicle behavior simulation unit 305, similar to the vehicle behavior simulation unit 105 in Embodiment 1, reproduces a virtual vehicle and a map of its surrounding area in the simulation environment based on actual vehicle data D1 and surrounding map data P, and simulates the behavior of the virtual vehicle in the simulation environment. When simulating the behavior of the virtual vehicle, the vehicle behavior simulation unit 305 simulates the behavior of the vehicle in accordance with the vehicle control logic of the vehicle control device under evaluation. Furthermore, the vehicle behavior simulation unit 305 reads the obstacle simulation model M from the obstacle simulation model storage unit 309, and when simulating the behavior of the vehicle, it also simulates the behavior of obstacles while referring to the actual vehicle data D3 and the obstacle simulation model M. As described above, the vehicle interacts with obstacles. The vehicle behavior simulation unit 305 also simulates this interaction in the simulation environment and obtains a vehicle behavior simulation result S that takes into account the effect of interaction with obstacles.

[0045] The obstacle simulation model M is a simulation model of an obstacle, and is a predetermined driver model or pedestrian model. The obstacle simulation model M may also be a driver model or pedestrian model that has been trained by an external learning device. The obstacle simulation model M only needs to simulate the behavior of obstacles around the vehicle and the interaction between the vehicle and these obstacles in the simulation environment of the vehicle behavior simulation unit 305, in combination with the obstacle information contained in the actual vehicle data D3.

[0046] Next, the operation will be described. Figure 9 is a flowchart showing the operation of the vehicle control device evaluation device in Embodiment 3. First, the process from step ST101 to step ST104 shown in Figure 6 (from extracting necessary actual vehicle data to reproducing the vehicle itself, the map of the driving location, and surrounding objects in a simulation environment) is performed (step ST301).

[0047] Next, the obstacle simulation model M is loaded (step ST302), and in the simulation environment, the behavior of the vehicle is simulated according to the vehicle control logic under evaluation, while the behavior of obstacles is simulated using the obstacle simulation model M. At this time, the interaction between the vehicle and obstacles is also simulated. By simulating the interaction, it is possible to simulate how the behavior of surrounding obstacles changes and how the behavior of the vehicle is affected by changes in the behavior of obstacles, assuming that the vehicle control device under evaluation is installed on the vehicle under evaluation while it is in motion.

[0048] Next, based on the vehicle behavior simulation result S, which is the result of simulating vehicle behavior, the vehicle control device to be evaluated is evaluated (step ST304). The evaluation result X is output (step ST305). The method of evaluating the vehicle control device is the same as in Embodiment 1. Other aspects are the same as in Embodiment 1.

[0049] According to Embodiment 3, the same effects as in Embodiment 1 can be obtained. Furthermore, the development efficiency of vehicle control logic can be improved. More specifically, the system includes an obstacle simulation model storage unit that stores data showing the behavior of obstacles interacting with the vehicle, including actual vehicle data, and an obstacle simulation model, which is a simulation model of the obstacle. The vehicle behavior simulation unit simulates the behavior of the vehicle while simulating the behavior of obstacles using the obstacle simulation model in a simulation environment. This allows for efficient realization of how the behavior of surrounding obstacles is affected by interaction with the vehicle, assuming the vehicle is controlled according to the modified vehicle control logic during operation, and also allows for efficient verification of how the behavior of the vehicle is affected by changes in the behavior of obstacles. In this way, the development efficiency of vehicle control logic can be improved by efficiently verifying the impact of modifications to the vehicle control logic on obstacles.

[0050] Embodiment 4. Next, Embodiment 4 will be described based on Figures 10 to 13. Note that components identical or corresponding to those in Figures 1 to 9 are denoted by the same reference numerals, and their descriptions are omitted. Embodiment 4 differs from Embodiment 3 in that a trained obstacle simulation model is generated in the evaluation device. Figure 10 is a block diagram showing the evaluation device for the vehicle control device in Embodiment 4. The evaluation device 400 includes an obstacle simulation model generation unit 410 that generates a trained obstacle simulation model M using actual vehicle data D. The obstacle simulation model generation unit 410 outputs the generated obstacle simulation model M to the obstacle simulation model storage unit 309, storing the trained obstacle simulation model M in the obstacle simulation model storage unit 309. In Embodiment 4, the vehicle behavior simulation unit 305 simulates the behavior of obstacles using the trained obstacle simulation model M generated by the obstacle simulation model generation unit 410.

[0051] The actual vehicle data extraction unit 403 is the same as the actual vehicle data extraction unit 303 in Embodiment 3, except that it outputs the actual vehicle data D to the obstacle simulation model generation unit 410. As will be described in detail later, Embodiment 4 uses the actual vehicle data D to perform machine learning on the obstacle simulation model M, and the actual vehicle data D in Embodiment 4 includes information indicating the position, speed, acceleration, and direction of the vehicle and obstacles, as well as data indicating the positions of surrounding objects. It may also include data indicating the size and shape of the vehicle, obstacles, and surrounding objects.

[0052] The obstacle simulation model generation unit 410 will now be described. Figure 11 is a block diagram showing the obstacle simulation model generation unit according to Embodiment 4. The obstacle simulation model generation unit 410 includes a learning data storage unit 411 that stores data used for learning the obstacle simulation model, namely data indicating the position, speed, acceleration, and direction of the vehicle and obstacles as described above, and data indicating the position of surrounding objects, from the actual vehicle data D as learning data L, and an obstacle simulation model learning unit 412 that uses the learning data L to train the obstacle simulation model M on the behavior of target obstacles (obstacles whose behavior is simulated by the obstacle simulation model M) using, for example, an unsupervised learning method, and generates a trained obstacle simulation model M.

[0053] The learning data storage unit 411 acquires actual vehicle data D from the actual vehicle data extraction unit 303, extracts data to be used as learning data L, and stores the extracted data.

[0054] The obstacle simulation model learning unit 412 acquires training data L from the training data storage unit 411 and trains the obstacle simulation model M by extracting features from the training data L that indicate the behavior of other target vehicles, including interactions with the vehicle itself. For feature extraction, a learning algorithm such as deep learning or clustering using the k-means method can be used. The obstacle simulation model learning unit 412 stores the trained obstacle simulation model M in the obstacle simulation model storage unit 309.

[0055] Next, the operation will be described. Figure 12 is a flowchart showing the operation of the vehicle control device evaluation device in Embodiment 4. First, the obstacle simulation model generation unit 410 generates a learned obstacle simulation model M (step ST401). The subsequent operation is the same as in Embodiment 3, and the processing from step ST301 to step ST305 (from extracting necessary actual vehicle data to outputting evaluation results) shown in Figure 9 is performed (step ST402).

[0056] The following describes the machine learning process using the obstacle simulation model learning unit 412. Figure 13 is a flowchart showing the machine learning process using the obstacle simulation model learning unit according to Embodiment 4. As a prerequisite, it is assumed that the storage of training data L in the training data storage unit 411 is complete. First, the training data L is obtained from the training data storage unit 411 (step ST42).

[0057] Next, features representing the behavior of the target obstacle are extracted from the training data L, and the obstacle simulation model M is trained (step ST42). The method for extracting features is as described above.

[0058] Next, the trained obstacle simulation model M is stored in the obstacle simulation model storage unit 309 (step ST43).

[0059] According to Embodiment 4, the same effects as in Embodiment 3 can be obtained. Furthermore, since the behavior of obstacles is simulated using an obstacle simulation model trained through machine learning, the behavior of obstacles can be simulated more accurately.

[0060] Embodiment 5. Next, Embodiment 5 will be described with reference to Figures 14 to 17. Components identical or corresponding to those in Figures 1 to 13 are denoted by the same reference numerals, and their descriptions are omitted. Embodiment 5, like Embodiment 4, uses machine learning to train an obstacle simulation model, but differs from Embodiment 4 in that it uses reinforcement learning. Figure 14 is a block diagram showing the evaluation device for the vehicle control device in Embodiment 5. As shown in Figure 14, in the evaluation device 500, the actual vehicle data extraction unit 303 is the same as in Embodiment 3, and does not output actual vehicle data D to the obstacle simulation model generation unit 510.

[0061] The obstacle simulation model generation unit 510 will now be described. Figure 15 is a block diagram showing the obstacle simulation model generation unit according to Embodiment 5. The obstacle simulation model generation unit 510 comprises a simulation playback unit 511 and an obstacle simulation model learning unit 512. The simulation playback unit 511 has a simulator (not shown) that generates a simulation environment in which various situations of the target obstacle (an obstacle whose behavior is simulated by the obstacle simulation model M) and its surroundings can be reproduced, and obstacle behavior data L1 and obstacle surrounding data L2 can be acquired. The obstacle simulation model learning unit 512 trains the obstacle simulation model M and outputs the trained obstacle simulation model M to the obstacle simulation model storage unit 309. The obstacle behavior data L1 includes data indicating the position, velocity, acceleration, and direction of the target obstacle. The obstacle behavior data L1 may also include information such as the size and shape of the target obstacle. Obstacle surrounding data L2 includes data indicating the position, speed, acceleration, and direction of objects surrounding the target obstacle (e.g., other obstacles, the vehicle itself, and surrounding objects). Obstacle surrounding data L2 may also include information such as the size and shape of objects surrounding the target obstacle.

[0062] The simulation playback unit 511 outputs obstacle behavior data L1 and obstacle surrounding data L2 in the simulation environment generated by the simulator of the simulation playback unit 511, along with a reward R determined based on predetermined criteria, to the obstacle simulation model learning unit 512. The obstacle simulation model learning unit 512 outputs action instruction A to the simulation playback unit 511. The simulation playback unit 511 plays back the situation when the target obstacle acts based on action instruction A.

[0063] The simulation playback unit 511 uses a simulator capable of reproducing various situations to simulate and reproduce the situation when a target obstacle in the simulation environment acts based on a given action instruction A. Specific examples of action instruction A include accelerating, decelerating, moving to the left lane, moving to the right lane, and maintaining the current position. After issuing the action instruction, the simulation playback unit 511 advances time by one predetermined time step. During this time, the simulation environment changes due to the action of the target obstacle. The simulation playback unit 511 outputs obstacle behavior data L1 and obstacle surrounding data L2 for the next time step to the obstacle simulation model learning unit 512.

[0064] The obstacle simulation model generation unit 510 trains the obstacle simulation model M using reinforcement learning. Reinforcement learning deals with the problem of an agent in a given environment learning from its interactions with the environment to achieve a goal. The agent continuously performs actions in response to the environment, and in response to those actions, the environment gives a new state and a reward. The agent repeatedly performs actions and receives a new state and reward, learning the action strategy that will yield the most reward. In the simulation environment of the simulation playback unit 511, the agent is the "target obstacle" described above, and the agent's actions in the simulation environment are determined by the action instruction A of the obstacle simulation model learning unit 512.

[0065] The obstacle simulation model learning unit 512 will be described further. Figure 16 is a block diagram showing the obstacle simulation model learning unit according to Embodiment 5. The obstacle simulation model learning unit 512 includes a function update unit 5121 that updates the obstacle simulation model M based on obstacle behavior data L1, obstacle surrounding data L2, and reward R using a reinforcement learning method, and outputs the updated obstacle simulation model M to the obstacle simulation model storage unit 309, and an action instruction unit 5122 that outputs action instruction A to the simulation playback unit 511.

[0066] As representative methods of reinforcement learning, Q-learning and TD-learning are known. The function update unit 5121 in Embodiment 5 uses the Q-learning method. Therefore, Q-learning will be described below. However, the TD-learning method may also be used. In the case of Q-learning, the general update formula for the action value function Q(s, a) is represented by the following formula (1).

Equation

[0067] The function update unit 5121 updates the action value function Q according to Equation (1) for the obstacle simulation model M during learning. As a result, the obstacle simulation model M is also updated. The function update unit 5121 stores the updated obstacle simulation model M in the obstacle simulation model storage unit 309. The action instruction unit 5122 determines the next action instruction A based on the updated action value function Q and outputs it to the simulation playback unit 511.

[0068] The update formula, represented by equation (1), increases the action value Q of action a that maximizes the Q value at time t+1 if that action value Q is greater than the action value Q of action a performed at time t, and decreases the action value Q if the opposite is true. In other words, it updates the action value function Q(s,a) so that the action value Q of action a at time t approaches the best action value at time t+1. This causes the best action value in a given environment to propagate sequentially to the action values ​​in previous environments.

[0069] The reward R is set to increase as the probability of the target obstacle colliding with other obstacles, the vehicle itself, or surrounding objects decreases. Alternatively, the reward R may be set to decrease when the distance between the target obstacle and other obstacles is smaller than a predetermined value. It is also conceivable that the reward R may decrease when the target obstacle deviates from its normal path or comes to a sudden stop. Conversely, it is conceivable that the reward R may increase when the target obstacle travels along its normal path at a predetermined speed (such as the legal speed limit if the target obstacle is another vehicle). In this way, the elements of the reward R determined for each condition are added together to calculate the overall reward R.

[0070] Next, we will explain the operation. Since the operation of the entire evaluation device is the same as in Embodiment 4, we will explain the flow of reinforcement learning by the obstacle simulation model learning unit. Figure 17 is a flowchart showing reinforcement learning by the obstacle simulation model generation unit according to Embodiment 5. First, the action instruction unit 5122 of the obstacle simulation model learning unit 512 outputs action instruction A to the simulation playback unit 511. This action instruction A is given to the agent (i.e., the target obstacle) in the simulation environment (step ST51). Specific examples of action instruction A are as described above.

[0071] Next, the simulation playback unit 511 makes the agent act based on the action instruction A and calculates the reward R (step ST52). More specifically, the simulation playback unit 511 performs one step of simulation in response to the action instruction A and calculates the reward R in the simulation environment after the one step has elapsed. The method for calculating the reward R is as described above. The simulation playback unit 511 outputs the reward R to the function update unit 5121 of the obstacle simulation model learning unit 512.

[0072] Next, the simulation playback unit 511 outputs the obstacle behavior data L1 and the obstacle surrounding data L2 for the next time step, i.e., after one step has elapsed, to the function update unit 5121 (step ST53).

[0073] Next, the function update unit 5121 updates the obstacle simulation model M and the action value function Q based on the obstacle behavior data L1, the obstacle surrounding data L2, and the reward R (step ST54). The function update unit 5121 stores the updated obstacle simulation model M in the obstacle simulation model storage unit 309.

[0074] The obstacle simulation model learning unit 512 determines whether or not to terminate reinforcement learning (step ST55), and if it does, it terminates the process. If it does not terminate reinforcement learning, it returns to step ST51, and the action instruction unit 5122 outputs the current action instruction A to the simulation playback unit 511.

[0075] The decision of whether or not to terminate reinforcement learning can be made, for example, by determining whether the number of times the process from step ST51 to step ST55 has been performed is equal to or greater than a predetermined number.

[0076] According to Embodiment 5, the same effects as in Embodiment 4 can be obtained. Furthermore, since the obstacle simulation model is trained using reinforcement learning, there is no need to generate training data from actual vehicle data.

[0077] Embodiment 6. Next, Embodiment 6 will be described with reference to Figures 18 and 19. Components identical or corresponding to those in Figures 1 to 17 are denoted by the same reference numerals, and their descriptions are omitted. Embodiment 6 differs from Embodiments 1 to 5 in that, when simulating the behavior of the vehicle, the initial state in the simulation environment is changed from that shown by the actual vehicle data before simulating the vehicle's behavior. Here, the initial state refers to the initial state of the vehicle, surrounding objects, and obstacles, including their position, velocity, acceleration, and direction at the initial point in time. Changing the initial state in this way also changes the results of subsequent time evolution, thus changing the simulation scenario (the time change in the state of the vehicle, surrounding objects, and obstacles in the simulation environment). Figure 18 is a block diagram showing the evaluation device for the vehicle control device in Embodiment 6. The evaluation device 600 includes a scenario changing unit 620.

[0078] The vehicle data extraction unit 603 outputs vehicle data D1 and D3 to the vehicle behavior simulation unit 605, outputs vehicle data D2 to the surrounding object reproduction unit 106, outputs vehicle data D used for machine learning to the obstacle simulation model generation unit 410, and also outputs vehicle data D4 to the scenario change unit 620. Vehicle data D4 is data that includes data indicating the initial state of the vehicle, surrounding objects, and obstacles, and includes information such as the initial position, initial speed, initial acceleration of the vehicle and obstacles, and the position of surrounding objects.

[0079] The scenario modification unit 620 refers to the actual vehicle data D4 and modifies the initial state of at least one of the following: the own vehicle, surrounding objects, and obstacles. Modifications to the initial state may include, for example, shifting the initial positions of the own vehicle and other vehicles by a few centimeters in the forward and backward directions, moving a specific surrounding object several meters away from the path of the own vehicle, or increasing (or decreasing) the initial speed of a specific obstacle by several kilometers per hour. The scenario modification unit 620 outputs the change from the initial state shown in the actual vehicle data D4 as a scenario modification parameter V to the vehicle behavior simulation unit 605. The scenario modification parameter V is a parameter that determines the simulation scenario after the modification.

[0080] The vehicle behavior simulation unit 605 simulates the behavior of its own vehicle and obstacles, reflecting the changes in the initial state due to the scenario change parameter V. Other aspects are the same as those of the vehicle behavior simulation unit 305 in Embodiment 3. As a result, the vehicle behavior simulation result S is obtained based on the simulation scenario modified according to the scenario change parameter V.

[0081] The amount of change in the initial position of the vehicle, etc., included in the scenario change parameter V, may be determined randomly. Alternatively, if a change in a specific initial state is assumed, that content may be reflected. However, in Embodiment 6, since the simulation scenario assumes different conditions at the same driving location, the surrounding map data P used to simulate the vehicle's behavior is not changed. Furthermore, the above-mentioned amount of change is also limited to a range where the surrounding map data P does not need to be changed.

[0082] Although the evaluation device 600 includes an obstacle simulation model storage unit 309 and an obstacle simulation model generation unit 410, the feature of Embodiment 6 is the scenario change unit 620, so the obstacle simulation model storage unit 309 and the obstacle simulation model generation unit 410 may be omitted.

[0083] Next, the operation will be described. Figure 19 is a flowchart showing the operation of the vehicle control device evaluation device in Embodiment 6. First, the processes from step ST101 to step ST104 shown in Figure 6 (from retrieving necessary actual vehicle data to reproducing the vehicle itself, the map of the driving location, and surrounding objects in a simulation environment) are performed (step ST601). Also, the obstacle simulation model M is read out (step ST602).

[0084] Next, the scenario change unit 620 changes the initial state of at least one of the following: the vehicle itself, surrounding objects, and obstacles (step ST603). Details of the initial state change are as described above.

[0085] Next, based on the initial state changed in step ST603, from step ST303 Step ST305 is executed (from simulating the behavior of the vehicle and obstacles to outputting the evaluation results) (Step ST604).

[0086] Note that the flow shown in Figure 19 shows the operation when simulating the behavior of obstacles, similar to Embodiment 3, but step ST602 can be omitted if the behavior of obstacles is not to be simulated. Also, in the explanation of step ST604, "steps ST303 to ST305" should be "steps ST105 to ST107".

[0087] According to Embodiment 6, the same effects as in Embodiment 1 can be obtained. Furthermore, it is possible to efficiently simulate vehicle behavior when the initial state changes within a range where the driving location does not change, such as when the position of surrounding objects has changed since the acquisition of actual vehicle data. This allows for efficient evaluation of vehicle control devices for simulation scenarios similar to those based on actual vehicle data.

[0088] Embodiment 7. Next, Embodiment 7 will be described with reference to Figures 20 to 22. Components that are the same as or equivalent to those in Figures 1 to 19 are denoted by the same reference numerals, and their descriptions are omitted. Embodiment 7 differs from Embodiment 6 in its scenario modification unit. Figure 20 is a block diagram showing the scenario modification unit according to Embodiment 7. The scenario modification unit 720 includes a scenario group generation unit 721 that generates a simulation scenario group N consisting of multiple simulation scenarios from actual vehicle data D4, a scenario evaluation unit 722 that calculates an evaluation value W for each simulation scenario included in the simulation scenario group N according to predetermined evaluation criteria, a determination unit 723 that determines whether the evaluation value W for each simulation scenario is equal to or greater than a target value K, and a target value holding unit 724 that holds the target value K for the evaluation value W.

[0089] The scenario group generation unit 721 acquires actual vehicle data D4 from the actual vehicle data extraction unit 603, generates multiple simulation scenarios using the initial state indicated by the actual vehicle data D4 to generate a simulation scenario group N, and outputs the generated simulation scenario group N to the scenario evaluation unit 722. The scenario group generation unit 721 generates the simulation scenario group N using a genetic algorithm. That is, first, it generates an initial population consisting of multiple simulation scenarios by randomly changing the initial state indicated by the actual vehicle data D4, and selects a replica of the parent population from this initial population. Next, the scenario group generation unit 721 generates a child population from the parent population by crossover (for example, real-valued crossover), performs survival selection on the next population (the population generated by the above-mentioned parent population and child population), and makes the selected population the simulation scenario group N.

[0090] The scenario evaluation unit 722 calculates an evaluation value W for each simulation scenario included in the simulation scenario group N, and outputs the calculated evaluation value W and the simulation scenario group N to the determination unit 723. The evaluation value W indicates the degree of adaptation to the environment in the genetic algorithm described above. In embodiment 7, it is calculated based, for example, on whether the distance between the vehicle and surrounding objects (surrounding objects and obstacles) is less than or equal to a predetermined distance, or the number of times this occurs (the number of times the distance between the vehicle and surrounding objects is less than or equal to a predetermined distance). In other words, a simulation scenario in which the distance between the vehicle and surrounding objects is less than or equal to a predetermined distance many times, and therefore collisions are more likely to occur, will have a large evaluation value W, indicating a high degree of adaptation to the environment.

[0091] The determination unit 723 receives the simulation scenario group N and the evaluation value W for each simulation scenario from the scenario evaluation unit 722, and determines whether the evaluation value W for each simulation scenario is equal to or greater than the target value K. The determination unit 723 adopts the simulation scenarios in which the evaluation value W is equal to or greater than the target value K as the adopted scenarios. The determination unit 723 also stores the scenario change parameters V corresponding to the adopted scenarios in the storage unit (not shown). As a result, the scenario change parameters V for changing the initial simulation scenarios (unchanged simulation scenarios) to adopted scenarios, i.e., simulation scenarios in which the evaluation value W is equal to or greater than the target value K, are sequentially stored. In Embodiment 7, a number of adopted scenarios greater than or equal to a preset value is generated. If a number of simulation scenarios greater than or equal to the preset value is adopted (if a number of adopted scenarios greater than or equal to the preset value is generated), the determination unit 723 outputs all of the scenario change parameters V. The number of scenario change parameters V output is the same as the number of adopted scenarios, and by applying each scenario change parameter V to change the initial state, the vehicle behavior is simulated for all adopted scenarios. Furthermore, if the number of adopted scenarios generated by the determination unit 723 is less than a set value, it sends a notification to the scenario group generation unit 721, causing it to generate the simulation scenario group N again.

[0092] The target value holding unit 724 holds the target value K of the evaluation value W. The target value K is the threshold for the evaluation value W determined by the determination unit 723, and as described above, simulation scenarios with an evaluation value W greater than or equal to the target value K are selected as the adopted scenarios.

[0093] In Embodiment 7, simulation scenarios are also assumed for situations where the conditions differ at the same driving location. Therefore, as in Embodiment 6, the amount of change in the initial position of the vehicle, etc., included in the scenario change parameter V, is set to a range where it is not necessary to change the surrounding map data P.

[0094] Next, the operation will be described. Since the operation of the entire evaluation device is the same as in Embodiment 6, Figure 21 is a flowchart showing the operation of the scenario change unit according to Embodiment 7. First, a group of simulation scenarios N is generated using a genetic algorithm (step ST71). The generation of the group of simulation scenarios N using a genetic algorithm is as described above.

[0095] Next, the evaluation value W for each simulation scenario included in the simulation scenario group N is calculated (step ST72).

[0096] Next, for each simulation scenario that makes up the simulation scenario group N, it is determined whether the evaluation value W is equal to or greater than the target value K. Simulation scenarios for which the evaluation value W is equal to or greater than the target value K are adopted as adopted scenarios, and the scenario change parameter V corresponding to the adopted scenario is stored (step ST73). After the above determination has been made for all simulation scenarios that make up the simulation scenario group N, it is determined whether the number of adopted scenarios is equal to or greater than a set value (step ST74).

[0097] If the number of adopted scenarios exceeds the set value, all stored scenario change parameters V are output (step ST75).

[0098] If the number of adopted scenarios is less than the set value, the process returns to step ST71. This repeats steps ST71 through ST74 until the number of adopted scenarios is equal to or greater than the set value.

[0099] The vehicle behavior simulation unit 605 simulates the behavior of its own vehicle and obstacles based on the modified simulation scenario obtained through the above operations, thereby simulating the behavior of its own vehicle and obstacles according to the simulation scenario in which the evaluation value W is high, that is, the simulation scenario in which collisions are more likely to occur.

[0100] The scenario modification unit 720 may be as shown in Figure 22. In the scenario modification unit 7201, which is a scenario modification unit according to another embodiment of Embodiment 7 shown in Figure 22, the obstacle simulation model M is also input to the scenario group generation unit 721. When generating the simulation scenario group N, the scenario group generation unit 721 incorporates the output of the obstacle simulation model M regarding the behavior of obstacles.

[0101] According to Embodiment 7, the same effects as in Embodiment 6 can be obtained. Furthermore, evaluation values ​​are set based on certain evaluation criteria, and among simulation scenarios similar to those based on actual vehicle data, only those simulation scenarios in which the above evaluation values ​​are equal to or greater than the target value are adopted. This makes it possible to simulate the behavior of vehicles and obstacles in environments that rarely occur in normal driving, that is, environments that are difficult to reproduce using only actual vehicle data. As a result, it is possible to efficiently evaluate vehicle control systems that take into account a wider variety of environments.

[0102] Embodiment 8. Next, Embodiment 8 will be described with reference to Figure 23. Note that components identical or corresponding to those in Figures 1 to 22 are denoted by the same reference numerals, and their descriptions are omitted. Embodiment 8 differs from Embodiment 7 in that the generation of simulation scenarios that fall under predetermined exclusion conditions is prevented. Figure 23 is a block diagram showing the scenario modification unit according to Embodiment 8. The scenario modification unit 820 includes an exclusion condition setting unit 825 that sets exclusion condition B to the scenario group generation unit 821.

[0103] When the scenario group generation unit 821, which has exclusion condition B set, generates simulation scenarios for the generation of the simulation scenario group N, if the generated simulation scenario falls under exclusion condition B, it discards the generated simulation scenario and generates a different simulation scenario. As a result, the simulation scenario group N consists only of simulation scenarios that do not fall under exclusion condition B.

[0104] Exclusion condition B is when, for example, it is certain that a collision cannot be avoided by controlling the vehicle itself, or when no useful information for evaluating the vehicle control system can be obtained, and therefore there is no value in simulating the vehicle's behavior.

[0105] While the discarding of simulation scenarios based on exclusion condition B may be performed by the determination unit 723, discarding them in the scenario group generation unit 821 avoids unnecessary calculations in the scenario evaluation unit 722.

[0106] Other aspects are the same as in Embodiment 7. Note that the same configuration as in the exclusion condition setting unit 825 may be applied to the scenario changing unit 620 of Embodiment 6.

[0107] According to Embodiment 8, the same results as in Embodiment 7 can be obtained. Furthermore, the scenario group generation unit is configured to set predetermined exclusion conditions, and during the generation of the simulation scenario group, simulation scenarios that fall under these exclusion conditions are discarded. This prevents unnecessary simulation of vehicle and obstacle behavior by setting the exclusion scenarios to exclude simulation scenarios that do not provide useful information for evaluating the vehicle control device, thereby enabling efficient use of the evaluation device's resources.

[0108] Embodiment 9. Next, Embodiment 9 will be described based on Figures 24 and 25. Note that components identical or corresponding to those in Figures 1 to 23 are denoted by the same reference numerals, and their descriptions are omitted. Embodiment 9 differs from Embodiment 6 in that it uses not only actual vehicle data but also surrounding map data to reproduce surrounding objects.

[0109] Figure 24 is a block diagram showing an evaluation device for a vehicle control device in Embodiment 9. The evaluation device 900 differs from the evaluation device 600 in Embodiment 6 in that it includes a map module 904 and a surrounding object reproduction unit 906 instead of the map module 104 and the surrounding object reproduction unit 106.

[0110] The map module 904 outputs surrounding map data P* to the vehicle behavior simulation unit 605 and the surrounding object reproduction unit 906. The surrounding map data P* includes information about terrain and roads, as well as information about objects in the vicinity of the location indicated by the surrounding map data P*.

[0111] The surrounding object reproduction unit 906 acquires actual vehicle data D2 from the actual vehicle data extraction unit 603 and surrounding map data P* from the map module 904, and generates surrounding object reproduction data T* that can reproduce surrounding objects with an arbitrary degree of accuracy based on the actual vehicle data D2 and surrounding map data P*. The surrounding object reproduction unit 906 outputs the surrounding object reproduction data T* to the vehicle behavior simulation unit 605, which reproduces the surrounding objects of the vehicle's driving location in the simulation environment (the simulation environment generated by the simulator of the vehicle behavior simulation unit 605) with an arbitrary degree of accuracy. Here, "degree of accuracy" is an indicator that determines how faithfully the surrounding objects are reproduced.

[0112] Let's further explain the accuracy of the reproduction of surrounding objects. While the actual vehicle data D2 includes data on surrounding objects, this data is acquired by the vehicle's surrounding information acquisition unit 14. Depending on the state of the cameras and sensors that make up the surrounding information acquisition unit 14, objects that actually exist may not be recognized as surrounding objects, or they may be recognized as surrounding objects but incomplete data may be acquired. For example, surrounding objects in the camera's blind spot will not be recognized. Also, data obtained from sensors may be incomplete due to malfunctions, dirt, errors, etc. Furthermore, even if there are no problems with the cameras and sensors, if the target surrounding object is a white line, accurate data may not be acquired while the vehicle is in motion due to fading caused by aging. In these cases, there may be surrounding objects that are not included in the actual vehicle data D2, or the data may be incomplete, so reproducing surrounding objects using only the actual vehicle data D2 may not faithfully reproduce the actual surrounding objects.

[0113] On the other hand, the surrounding object data included in the surrounding map data P* does not have the same problems as the actual vehicle data D2 mentioned above, and as long as it is data included in the surrounding map data P*, it can faithfully reproduce the surrounding objects. For this reason, when generating the surrounding object reproduction data T*, the surrounding object reproduction unit 906 generates surrounding object reproduction data T* that reproduces the surrounding objects of the vehicle's location in the simulation environment with an arbitrary degree of reproduction by combining the surrounding object data included in the actual vehicle data D2 and the surrounding map data P*, respectively. The above degree of reproduction is adjusted by the proportion in which the surrounding object data included in the actual vehicle data D2 and the surrounding map data P* are combined. In other words, generally, the data included in the surrounding map data P* is more faithful to the actual surrounding objects (higher degree of reproduction), so if you want a higher degree of reproduction, increase the proportion of surrounding map data P* used. Other aspects are the same as in Embodiment 6.

[0114] Although the evaluation device 900 includes an obstacle simulation model storage unit 309, an obstacle simulation model generation unit 410, and a scenario modification unit 620, the features of Embodiment 9 are the map module 904 and the surrounding object reproduction unit 906, so the obstacle simulation model storage unit 309, the obstacle simulation model generation unit 410, and the scenario modification unit 620 may be omitted.

[0115] Next, the operation will be described. Figure 25 is a flowchart showing the operation of the vehicle control device evaluation device in Embodiment 9. First, the process from step ST101 to step ST103 (from extracting the necessary actual vehicle data to reproducing the map of the vehicle and driving location in the simulation environment) is performed (step ST901).

[0116] Next, based on the actual vehicle data D2 and the surrounding map data P*, the surrounding objects of the vehicle's location are reproduced (step ST902). Surrounding object reproduction data T*, which is data for reproducing surrounding objects in the simulation environment, is generated from the surrounding object data included in the actual vehicle data D2 and the surrounding map data P*, respectively, and the surrounding object reproduction data T* is applied to the simulation environment. As a result, the surrounding objects of the vehicle's location are reproduced in the simulation environment. When generating the surrounding object reproduction data T*, the reproduction degree is set in advance so that the surrounding objects are reproduced with the set reproduction degree.

[0117] Next, the process from step ST602 to step ST604 (from reading out the obstacle simulation model M and changing its initial state to outputting the evaluation results), as shown in Figure 19, is performed (step ST903).

[0118] According to Embodiment 9, the same effects as in Embodiment 6 can be obtained. Furthermore, surrounding objects can be reproduced in the simulation environment with a higher degree of accuracy, and the vehicle's behavior can be simulated in a situation where surrounding objects are faithfully reproduced, allowing for a more accurate evaluation of the vehicle control system. More specifically, the surrounding map data includes data on objects surrounding the vehicle's location, and the surrounding object reproduction unit is configured to reproduce surrounding objects in the simulation environment using not only actual vehicle data but also surrounding map data. The surrounding object data included in the surrounding map data reproduces surrounding objects more faithfully and includes data on surrounding objects that are difficult to acquire by the vehicle's surrounding information acquisition unit, i.e., data that may be incomplete in actual vehicle data. In Embodiment 9, by combining the surrounding object data included in the actual vehicle data and the surrounding map data, it is possible to reproduce surrounding objects with a higher degree of accuracy. It is also possible to achieve any desired degree of accuracy by adjusting the combination of actual vehicle data and surrounding map data.

[0119] Furthermore, the ability to reproduce surrounding objects with a higher degree of accuracy, as in Embodiment 9, means that evaluation can be performed while sufficiently eliminating the influence of errors and inaccuracies in the reproduction of surrounding objects. This also means that the control loop of the vehicle control logic being evaluated (for example, a fully cruised loop with feedback) can be verified efficiently.

[0120] Embodiment 10. Next, Embodiment 10 will be described with reference to Figures 26 and 27. Components identical or corresponding to those in Figures 1 to 25 are denoted by the same reference numerals, and their descriptions are omitted. Embodiment 10 achieves the same functions as the evaluation device in Embodiment 1 using multiple devices connected to each other on a network. Figure 26 is a diagram showing an overview of the vehicle control device evaluation system in Embodiment 10. The evaluation system 2000 includes a vehicle data acquisition unit 19 (not shown in Figure 26) mounted on vehicles 10A, 10B, and 10C, respectively, which acquires vehicle data D, and servers 2001A to 2001E configured in the cloud 2001. The vehicle data D is transmitted to the cloud 2001. The configurations of vehicles 10A, 10B, and 10C are the same as in Embodiment 1.

[0121] Cloud 2001 is equipped with server devices 2001A to 2001E that are connected to each other via a network. Server devices 2001A to 2001E work together to evaluate the vehicle control device and output the evaluation result X. In other words, the evaluation device 100 of Embodiment 1 is realized by the cooperation of server devices 2001A to 2001E.

[0122] This section describes each server in Cloud 2001. Figure 27 is a block diagram showing the evaluation system for the vehicle control device in Embodiment 10, illustrating the functions of each server device. Note that server devices 2001A to 2001E also include a communication unit for sending and receiving data, but this is omitted from the illustration in Figure 27.

[0123] The server device 2001A includes a vehicle log data storage unit 102 and a real vehicle data retrieval unit 103. The server device 2001A collects real vehicle data D and stores it as vehicle log data. In response to a request from the vehicle behavior simulation unit 105, the server device 2001A retrieves the real vehicle data D necessary for simulating the behavior of its own vehicle, transmits real vehicle data D1, which includes the vehicle's driving data, to the server device 2001D, and transmits real vehicle data D2, which includes data on surrounding objects, to the server device 2001C.

[0124] Server device 2001B includes a map module 104. Server device 2001B transmits surrounding map data P to server device 2001D in response to a request from the vehicle behavior simulation unit 105.

[0125] The server device 2001C includes a surrounding object reproduction unit 106. The server device 2001C generates surrounding object reproduction data T from the actual vehicle data D2 and transmits the generated surrounding object reproduction data T to the server device 2001D.

[0126] Server device 2001D includes a vehicle behavior simulation unit 105. Based on actual vehicle data D1, surrounding map data P, and surrounding object reproduction data T, server device 2001D reproduces a virtual vehicle, a map of its location, and surrounding objects in a simulation environment, and simulates the vehicle's behavior. Server device 2001D transmits the vehicle behavior simulation result S to server device 2001E.

[0127] The server device 2001E comprises an evaluation unit 107 and an output unit 108. The server device 2001E evaluates the vehicle control device based on the vehicle behavior simulation result S and outputs the evaluation result X. Other aspects are the same as in Embodiment 1.

[0128] As described above, in Embodiment 10, the server devices 2001A to 2001E work together to achieve the same functionality as the evaluation device 100 in Embodiment 1. It is sufficient that the server devices 2001A to 2001E work together to achieve the functionality of the evaluation device 100; therefore, the server devices 2001A to 2001E do not necessarily need to be built on the cloud. Furthermore, as long as the server devices 2001A to 2001E as a whole can achieve the functionality of the evaluation device 100, the functional division of each server device is not limited to that shown in Figure 27. For example, the evaluation unit 107 may be moved from server device 2001E to server device 2001D. Also, while "server devices" are used as the hardware to achieve each function, the type of hardware is not limited as long as it can achieve similar functionality.

[0129] Although Embodiment 10 describes a case similar to that of the evaluation device 100, the same functions as those in Embodiments 2 to 9 can also be realized by adding the necessary configurations to Cloud 2001.

[0130] According to Embodiment 10, the same effects as in Embodiment 1 can be obtained.

[0131] Although this application describes exemplary embodiments, the various features, aspects, and functions described in the embodiments are not limited to the application of any particular embodiment, but can be applied individually or in various combinations to the embodiments. Accordingly, countless variations not illustrated are conceivable within the scope of the art disclosed herein. These include, for example, modifications, additions, or omissions of at least one component.

[0132] The various aspects of this disclosure are summarized below as an appendix. (Note 1) A vehicle control device evaluation device that collects actual vehicle data from a vehicle, which includes at least position information of the vehicle while it is in motion, and uses the actual vehicle data to evaluate the vehicle control device to be evaluated, A real vehicle data extraction unit extracts real vehicle data from the collected real vehicle data to be used to simulate the behavior of the vehicle, A map data storage unit that stores surrounding map data showing a map of the area where the vehicle is traveling, Based on the actual vehicle data and the surrounding map data, a vehicle behavior simulation unit reproduces a virtual vehicle and a map of the area where the vehicle is driving in the simulation environment, and simulates the behavior of the vehicle in the simulation environment according to the vehicle control logic of the vehicle control device. Based on the aforementioned actual vehicle data, the surrounding objects reproduction unit reproduces the surrounding objects of the driving location in the simulation environment, An evaluation unit evaluates the vehicle control device based on the results of the simulation performed by the vehicle behavior simulation unit, An evaluation device for a vehicle control device, characterized by comprising an output unit that outputs the results of the evaluation performed by the evaluation unit. (Note 2) When another vehicle is operating in the same driving environment as the aforementioned vehicle, the actual vehicle data of that other vehicle is treated as identical environmental driving data. An evaluation device for a vehicle control device as described in Appendix 1, wherein the actual vehicle data extraction unit extracts the same environmental driving data from the collected actual vehicle data, and the vehicle behavior simulation unit simulates the behavior of the vehicle using the same environmental driving data. (Note 3) The aforementioned actual vehicle data includes data on obstacles that interact with the vehicle, The vehicle control device evaluation device further comprises an obstacle simulation model storage unit that stores an obstacle simulation model, which is a simulation model of the obstacle. The vehicle behavior simulation unit is an evaluation device for a vehicle control device according to Appendix 1 or 2, which simulates the behavior of the vehicle while simulating the behavior of the obstacle in the simulation environment using the obstacle simulation model. (Note 4) The system further comprises an obstacle simulation model generation unit that trains the aforementioned obstacle simulation model using machine learning and generates the trained obstacle simulation model, The vehicle behavior simulation unit is an evaluation device for a vehicle control device as described in Appendix 3, which simulates the behavior of an obstacle using a learned obstacle simulation model. (Note 5) The aforementioned vehicle data includes data indicating the position, speed, and acceleration of the vehicle and the obstacle, and data indicating the position of the surrounding objects. The obstacle simulation model generation unit includes a learning data storage unit that stores the actual vehicle data as learning data, An evaluation device for a vehicle control device as described in Appendix 4, comprising: an obstacle simulation model learning unit that takes the aforementioned learning data as input and causes the obstacle simulation model to learn the behavior of the obstacles to generate a trained obstacle simulation model. (Note 6) The obstacle simulation model generation unit is, A simulator is provided that generates a simulation environment in which an obstacle simulated by the aforementioned obstacle simulation model is placed, and obstacle behavior data including the position, velocity, and acceleration of the obstacle, and obstacle surrounding data including the position, velocity, and acceleration of objects surrounding the obstacle are acquired, and that reproduces the situation when the obstacle acts based on a given action instruction, and a simulation playback unit calculates a reward for the action instruction according to a predetermined standard, and outputs the reward and the obstacle behavior data and obstacle surrounding data updated by the action of the obstacle, respectively. An evaluation device for a vehicle control device as described in Appendix 4, comprising: an obstacle simulation model learning unit that updates a simulated model of the target obstacle by reinforcement learning based on the reward, obstacle behavior data, and obstacle surrounding data input from the simulation playback unit, and generates a trained simulated obstacle model. (Note 7) The system further includes a scenario modification unit that modifies the simulation scenario by changing the initial state of at least one of the vehicle and the surrounding objects when simulating the behavior of the vehicle. The vehicle behavior simulation unit simulates the behavior of the vehicle based on the initial state modified by the scenario modification unit, and is an evaluation device for a vehicle control device according to any one of the appendices 1 to 6. (Note 8) The scenario changing unit is an evaluation device for the vehicle control device described in Appendix 7, which randomly changes the initial state. (Note 9) The scenario modification unit includes a scenario group generation unit that generates a plurality of simulation scenarios using the actual vehicle data, A scenario evaluation unit calculates evaluation values ​​for each of the aforementioned simulation scenarios according to predetermined evaluation criteria, An evaluation device for a vehicle control device as described in Appendix 7, comprising a determination unit that determines whether the evaluation value of each of the simulation scenarios is equal to or greater than a target value, and adopts a simulation scenario in which the evaluation value is equal to or greater than the target value, thereby changing the simulation scenario to the simulation scenario adopted by the determination unit. (Note 10) The vehicle control device described in Appendix 9 is an evaluation device for a vehicle control device, wherein the evaluation value is calculated based on whether or not the distance between the vehicle and objects surrounding the vehicle is less than or equal to a predetermined distance. (Note 11) The aforementioned scenario modification unit is an evaluation device for a vehicle control device as described in any one of the appendices 7 to 10, which discards the simulation scenario if the modified simulation scenario falls under a predetermined exclusion condition. (Note 12) The aforementioned surrounding map data includes data on objects in the vicinity of the vehicle's location. The surrounding object reproduction unit is an evaluation device for a vehicle control device according to any one of the appendices 1 to 11, which reproduces the surrounding objects using the actual vehicle data and the surrounding map data. (Note 13) A vehicle data acquisition unit is mounted on the vehicle and acquires actual vehicle data that includes at least location information of the vehicle while it is in motion. A vehicle control device evaluation system comprising a vehicle control device evaluation device according to any one of the appendices 1 to 12, which collects the aforementioned actual vehicle data and uses the aforementioned actual vehicle data to evaluate the vehicle control device to be evaluated. (Note 14) The vehicle control device evaluation device is the vehicle control device evaluation system described in Appendix 13, which is built on the cloud. (Note 15) A method for evaluating a vehicle control device, comprising acquiring actual vehicle data, which includes at least positional information of the vehicle while it is in motion, and using the actual vehicle data to evaluate the vehicle control device to be evaluated, The steps include: collecting the actual vehicle data from the aforementioned vehicle; The steps include extracting actual vehicle data from the collected actual vehicle data to be used to simulate the behavior of the vehicle, The steps include: acquiring surrounding map data showing a map of the area where the vehicle is traveling; The steps include: recreating a virtual vehicle and a map of the surrounding area of ​​the driving location in a simulation environment based on the actual vehicle data and the surrounding area map data; Based on the aforementioned actual vehicle data, the simulation environment is used to reproduce the surrounding objects of the driving location. The steps include: 1. In the simulation environment, 2. Simulating the behavior of the vehicle in accordance with the vehicle control logic of the vehicle control device; A step of evaluating the vehicle control device based on the results of simulating the behavior of the vehicle, A method for evaluating a vehicle control device, comprising the step of outputting the results of evaluating the vehicle control device. (Note 16) When another vehicle is operating in the same driving environment as the aforementioned vehicle, the actual vehicle data of that other vehicle is treated as identical environmental driving data. The vehicle control device evaluation method according to Appendix 15, wherein in the step of extracting the actual vehicle data, the same environmental driving data is extracted from the collected actual vehicle data, and in the step of simulating the behavior of the vehicle, the behavior of the vehicle is simulated using the same environmental driving data. (Note 17) The aforementioned actual vehicle data includes data on obstacles that interact with the vehicle, The vehicle control device evaluation method according to Appendix 15 or 16, wherein in the step of simulating the behavior of the vehicle, the behavior of the vehicle is simulated in the simulation environment while simulating the behavior of the obstacle using an obstacle simulation model, which is a pre-stored simulation model of the obstacle. (Note 18) The method further comprises the step of training the aforementioned obstacle simulation model using machine learning to generate the trained obstacle simulation model. The vehicle control device evaluation method according to Appendix 17, wherein in the step of simulating the behavior of the vehicle, the behavior of the obstacle is simulated using the learned obstacle simulation model. (Note 19) The actual vehicle data includes data indicating the position, speed, and acceleration of the vehicle and the obstacle, and data indicating the position of surrounding objects, and the step of generating the obstacle simulation model is: The steps include: accumulating the aforementioned actual vehicle data as training data, The method for evaluating a vehicle control device according to Appendix 18, comprising the steps of taking the aforementioned learning data as input, training the obstacle simulation model on the behavior of the obstacle, and generating the trained obstacle simulation model. (Note 20) The step of generating the aforementioned obstacle simulation model is: In a simulation environment where an obstacle simulated by the aforementioned obstacle simulation model is placed, and obstacle behavior data including the position, velocity, and acceleration of the obstacle, and obstacle surrounding data including the position, velocity, and acceleration of objects surrounding the obstacle are acquired, the step of recreating the situation in which the obstacle acts based on a given action instruction, A step of calculating the reward for the aforementioned behavioral instruction according to predetermined criteria, The steps include outputting the reward and the obstacle behavior data and obstacle surrounding data that have been updated by the behavior of the target obstacle, respectively. The evaluation method for a vehicle control device according to Appendix 18, comprising the step of updating a simulated model of the target obstacle by reinforcement learning based on the reward, the obstacle behavior data, and the data surrounding the obstacle. (Note 21) The simulation further includes a step of changing the simulation scenario by changing the initial state of at least one of the vehicle and the surrounding objects when simulating the behavior of the vehicle, An evaluation method for a vehicle control device according to any one of the appendices 15 to 20, wherein in the step of simulating the behavior of the vehicle, the behavior of the vehicle is simulated based on the modified initial state. (Note 22) The vehicle control device evaluation method according to Appendix 21, wherein the initial state is randomly changed in the step of changing the simulation scenario. (Note 23) The step of changing the aforementioned simulation scenario is: The steps include generating a group of simulation scenarios consisting of multiple simulation scenarios using the actual vehicle data, A step of calculating the evaluation value for each of the aforementioned simulation scenarios according to predetermined evaluation criteria, The steps include determining whether the evaluation value of each simulation scenario is equal to or greater than the target value, and adopting the simulation scenario in which the evaluation value is equal to or greater than the target value, The method for evaluating a vehicle control device as described in Appendix 21, further comprising the step of changing the aforementioned simulation scenario to the adopted simulation scenario. (Note 24) The evaluation method for a vehicle control device described in Appendix 23, wherein the evaluation value is calculated based on whether or not the distance between the vehicle and objects surrounding the vehicle is less than or equal to a predetermined distance. (Note 25) The vehicle control device evaluation method according to any one of the appendices 21 to 24, wherein in the step of changing the simulation scenario, if the changed simulation scenario falls under a predetermined exclusion condition, the simulation scenario is discarded. (Note 26) The aforementioned surrounding map data includes data on objects in the vicinity of the vehicle's location. The vehicle control device evaluation method according to any one of Appendix 15 to 25, wherein in the step of reproducing the surrounding objects, the surrounding objects are reproduced using the actual vehicle data and the surrounding map data. [Explanation of symbols]

[0133] 10A, 10B, 10C Vehicle, 11 Communication Unit, 12 Map Module, 13 Positioning Module, 14 Surrounding Information Acquisition Unit, 15 ECU, 19 Actual Vehicle Data Acquisition Unit, 100, 300, 400, 500, 600, 900 Evaluation Device, 103, 303, 403, 603 Actual Vehicle Data Extraction Unit, 104, 904 Map Module, 105, 305, 605 Vehicle Behavior Simulation Unit, 106, 906 Surrounding Object Reproduction Unit, 107 Evaluation Unit, 108 Output Unit, 309 Obstacle Simulation Model Storage Unit, 410, 510 Obstacle Simulation Model Generation Unit, 411 Learning Data Storage Unit, 412, 512 Obstacle Simulation Model Learning Unit, 511 Simulation Playback Unit, 620, 720, 7201, 820 Scenario modification unit, 721, 821 Scenario group generation unit, 722 Scenario evaluation unit, 723 Judgment unit, 825 Exclusion condition setting unit, 1000, 2000 Evaluation system, 2001 Cloud, A Action instructions, D, D1, D2, D3, D4 Actual vehicle data, L Training data, L1 Obstacle behavior data, L2 Obstacle surroundings data, M Obstacle simulation model, N Simulation scenario group, P, P* Surroundings map data, R Reward, S Vehicle behavior simulation results, T, T* Surroundings object reproduction data, X Evaluation results

Claims

1. An evaluation device for a vehicle control device that collects actual vehicle data from a vehicle, including at least position information of the vehicle while it is running based on the vehicle control device before modification, and uses the actual vehicle data to evaluate the modified vehicle control device to be evaluated, A real vehicle data extraction unit extracts real vehicle data from the collected real vehicle data to be used to simulate the behavior of the vehicle, A map data storage unit that stores surrounding map data showing a map of the area where the vehicle is traveling, Based on the actual vehicle data and the surrounding map data, a vehicle behavior simulation unit reproduces a virtual vehicle and a map of the area where the vehicle is driving in the simulation environment, and simulates the behavior of the vehicle in the simulation environment according to the vehicle control logic of the modified vehicle control device. Based on the aforementioned actual vehicle data, the surrounding objects reproduction unit reproduces the surrounding objects of the driving location in the simulation environment, An evaluation unit evaluates the modified vehicle control device based on the results of the simulation by the vehicle behavior simulation unit, An evaluation device for a vehicle control device, characterized by comprising an output unit that outputs the results of the evaluation performed by the evaluation unit.

2. When another vehicle is operating in the same driving environment as the aforementioned vehicle, the actual vehicle data of that other vehicle is treated as identical environmental driving data. An evaluation device for a vehicle control device according to claim 1, wherein the actual vehicle data extraction unit extracts the same environmental driving data from the collected actual vehicle data, and the vehicle behavior simulation unit simulates the behavior of the vehicle using the same environmental driving data.

3. The aforementioned actual vehicle data includes data on obstacles that interact with the vehicle, The vehicle control device evaluation device further comprises an obstacle simulation model storage unit that stores an obstacle simulation model, which is a simulation model of the obstacle. The vehicle behavior simulation unit simulates the behavior of the vehicle while simulating the behavior of the obstacle in the simulation environment using the obstacle simulation model, as an evaluation device for a vehicle control device according to claim 1 or 2.

4. The system further comprises an obstacle simulation model generation unit that trains the aforementioned obstacle simulation model using machine learning and generates the trained obstacle simulation model, The vehicle behavior simulation unit simulates the behavior of the obstacle using the learned obstacle simulation model, as described in claim 3, for the vehicle control device evaluation device.

5. The aforementioned vehicle data includes data indicating the position, speed, and acceleration of the vehicle and the obstacle, and data indicating the position of the surrounding objects. The obstacle simulation model generation unit includes a learning data storage unit that stores the actual vehicle data as learning data, An evaluation device for a vehicle control device according to claim 4, comprising: an obstacle simulation model learning unit that takes the aforementioned learning data as input and causes the obstacle simulation model to learn the behavior of the obstacles to generate a trained obstacle simulation model.

6. The obstacle simulation model generation unit is, A simulator is provided that generates a simulation environment in which an obstacle simulated by the aforementioned obstacle simulation model is placed, and obstacle behavior data including the position, velocity, and acceleration of the obstacle, and obstacle surrounding data including the position, velocity, and acceleration of objects surrounding the obstacle are acquired, and that reproduces the situation when the obstacle acts based on a given action instruction, and a simulation playback unit calculates a reward for the action instruction according to a predetermined standard, and outputs the reward and the obstacle behavior data and obstacle surrounding data updated by the action of the obstacle, respectively. An evaluation device for a vehicle control device according to claim 4, comprising: an obstacle simulation model learning unit that updates a simulated model of the target obstacle by reinforcement learning based on the reward, obstacle behavior data, and obstacle surrounding data input from the simulation playback unit, and generates a trained simulated obstacle model.

7. The system further includes a scenario modification unit that modifies the simulation scenario by changing the initial state of at least one of the vehicle and the surrounding objects when simulating the behavior of the vehicle. The vehicle control device evaluation device according to claim 1, wherein the vehicle behavior simulation unit simulates the behavior of the vehicle based on the initial state changed by the scenario change unit.

8. The vehicle control device evaluation device according to claim 7, wherein the scenario changing unit randomly changes the initial state.

9. The scenario modification unit includes a scenario group generation unit that generates a plurality of simulation scenarios using the actual vehicle data, A scenario evaluation unit calculates evaluation values ​​for each of the aforementioned simulation scenarios according to predetermined evaluation criteria, An evaluation device for a vehicle control device according to claim 7, comprising a determination unit that determines whether the evaluation value of each of the simulation scenarios is equal to or greater than a target value, and adopts a simulation scenario in which the evaluation value is equal to or greater than the target value, and the simulation scenario is changed to the simulation scenario adopted by the determination unit.

10. The vehicle control device evaluation device according to claim 9, wherein the evaluation value is calculated based on whether or not the distance between the vehicle and an object surrounding the vehicle is less than or equal to a predetermined distance.

11. The vehicle control device evaluation device according to any one of claims 7 to 10, wherein the scenario modification unit discards the simulation scenario if the modified simulation scenario falls under predetermined exclusion conditions.

12. The aforementioned surrounding map data includes data on objects in the vicinity of the vehicle's location. The vehicle control device evaluation device according to claim 1, wherein the surrounding object reproduction unit reproduces the surrounding objects using the actual vehicle data and the surrounding map data.

13. A vehicle data acquisition unit is mounted on the vehicle and acquires actual vehicle data that includes at least location information of the vehicle while it is in motion. A vehicle control device evaluation system comprising: an evaluation device for a vehicle control device according to claim 1, which collects the aforementioned actual vehicle data and uses the aforementioned actual vehicle data to evaluate the vehicle control device to be evaluated.

14. The vehicle control device evaluation device is built on the cloud, as described in claim 13.

15. A method for evaluating a vehicle control device, comprising: acquiring actual vehicle data in the vehicle that includes at least position information of the vehicle while it is running based on the vehicle control device before modification; and evaluating the modified vehicle control device to be evaluated using the actual vehicle data, The steps include: collecting the actual vehicle data from the aforementioned vehicle; The steps include extracting actual vehicle data from the collected actual vehicle data to be used to simulate the behavior of the vehicle, The steps include: acquiring surrounding map data showing a map of the area where the vehicle is traveling; The steps include: recreating a virtual vehicle and a map of the surrounding area of ​​the driving location in a simulation environment based on the actual vehicle data and the surrounding area map data; Based on the aforementioned actual vehicle data, the simulation environment is used to reproduce the surrounding objects of the driving location. The steps include: simulating the behavior of the vehicle in the aforementioned simulation environment according to the vehicle control logic of the modified vehicle control device; A step of evaluating the modified vehicle control device based on the results of simulating the behavior of the vehicle, A method for evaluating a vehicle control device, comprising the step of outputting the results of evaluating the vehicle control device.

16. When another vehicle is operating in the same driving environment as the aforementioned vehicle, the actual vehicle data of that other vehicle is treated as identical environmental driving data. The vehicle control device evaluation method according to claim 15, wherein in the step of extracting the actual vehicle data, the same environmental driving data is extracted from the collected actual vehicle data, and in the step of simulating the behavior of the vehicle, the behavior of the vehicle is simulated using the same environmental driving data.

17. The aforementioned actual vehicle data includes data on obstacles that interact with the vehicle, The vehicle control device evaluation method according to claim 15 or 16, wherein in the step of simulating the behavior of the vehicle, the behavior of the vehicle is simulated in the simulation environment while simulating the behavior of the obstacle using an obstacle simulation model which is a pre-stored simulation model of the obstacle.

18. The method further comprises the step of training the aforementioned obstacle simulation model using machine learning to generate the trained obstacle simulation model. The vehicle control device evaluation method according to claim 17, wherein in the step of simulating the behavior of the vehicle, the behavior of the obstacle is simulated using the learned obstacle simulation model.

19. The actual vehicle data includes data indicating the position, speed, and acceleration of the vehicle and the obstacle, and data indicating the position of surrounding objects, and the step of generating the obstacle simulation model is: The steps include: accumulating the aforementioned actual vehicle data as training data, The method for evaluating a vehicle control device according to claim 18, further comprising the steps of: taking the aforementioned learning data as input, and having the obstacle simulation model learn the behavior of the obstacles to generate the trained obstacle simulation model.

20. The step of generating the aforementioned obstacle simulation model is: In a simulation environment where an obstacle simulated by the aforementioned obstacle simulation model is placed, and obstacle behavior data including the position, velocity, and acceleration of the obstacle, and obstacle surrounding data including the position, velocity, and acceleration of objects surrounding the obstacle are acquired, the step of recreating the situation in which the obstacle acts based on a given action instruction, A step of calculating the reward for the aforementioned behavioral instruction according to predetermined criteria, The steps include outputting the reward and the obstacle behavior data and obstacle surrounding data that have been updated by the behavior of the target obstacle, respectively. A method for evaluating a vehicle control device according to claim 18, comprising the step of updating a simulated model of the target obstacle by reinforcement learning based on the reward, the obstacle behavior data, and the data surrounding the obstacle.

21. The simulation further includes a step of changing the simulation scenario by changing the initial state of at least one of the vehicle and the surrounding objects when simulating the behavior of the vehicle, The vehicle control device evaluation method according to claim 15, wherein in the step of simulating the behavior of the vehicle, the behavior of the vehicle is simulated based on the changed initial state.

22. The vehicle control device evaluation method according to claim 21, wherein the initial state is randomly changed in the step of changing the simulation scenario.

23. The step of changing the aforementioned simulation scenario is: The steps include generating a group of simulation scenarios consisting of multiple simulation scenarios using the actual vehicle data, A step of calculating the evaluation value for each of the aforementioned simulation scenarios according to predetermined evaluation criteria, The steps include determining whether the evaluation value of each simulation scenario is equal to or greater than the target value, and adopting the simulation scenario in which the evaluation value is equal to or greater than the target value, The method for evaluating a vehicle control device according to claim 21, further comprising the step of changing the aforementioned simulation scenario to an adopted simulation scenario.

24. The evaluation method for a vehicle control device according to claim 23, wherein the evaluation value is calculated based on whether or not the distance between the vehicle and an object surrounding the vehicle is less than or equal to a predetermined distance.

25. A method for evaluating a vehicle control device according to any one of claims 21 to 24, wherein, in the step of changing the simulation scenario, the simulation scenario is discarded if the changed simulation scenario falls under a predetermined exclusion condition.

26. The aforementioned surrounding map data includes data on objects in the vicinity of the vehicle's location. The method for evaluating a vehicle control device according to claim 15, wherein in the step of reproducing the surrounding objects, the surrounding objects are reproduced using the actual vehicle data and the surrounding map data.

Images