Development method of autonomous vehicle

Abstract

The automatic driving system applies a machine learning model to at least recognition processing and planning processing among recognition processing for recognizing an external environment, planning processing for performing route planning, and vehicle control processing for performing automatic driving control. A method for developing an autonomous vehicle includes: a first step of training a machine learning model by matching at least a recognition process and a planning process using a first autonomous vehicle having an external sensor of a first sensor specification; and a second step of applying the trained machine learning model performed in the first step to a second autonomous vehicle provided with external sensors of a second sensor specification, at least one of the number, type, and detection performance of which is less than the first sensor specification.

Description

Technical Field

[0001] The present disclosure relates to a method for developing an autonomous vehicle. Background Art

[0002] The detection system described in Patent Document 1 has an analysis module in the form of a convolutional neural network. Images from a camera and sensor data from a sensor device are used as input data of the analysis module and analyzed in real time in the detection system. As a result of the analysis, for example, the safety level of the second car is indicated based on the movement of the body of a first car traveling in front of the second car, and at least one of an optical alarm and an audible alarm is output on the user interface of the second car according to the safety level.

[0003] Prior art literature

[0004] Patent Document 1: Japanese Patent Application Publication No. 2022-082510 Summary of the invention

[0005] Problem that the invention aims to solve

[0006] In an autonomous driving vehicle that recognizes the external environment based on the detection results of external sensors to perform path planning and performs autonomous driving control based on the planned path, the input data to the path plan varies, for example, depending on the number, type, and detection performance of the external sensors. Therefore, in the past, autonomous driving vehicles with different external sensor configurations could not simply use the matching results of the path plan, and needed to re-match the path plan in coordination with different input data. As a result, a huge amount of development work was required.

[0007] The purpose of the present disclosure is to reduce the effort of separately re-matching a second autonomous driving vehicle with a second sensor specification having at least one of quantity, type, and detection performance inferior to the first sensor specification from the first autonomous driving vehicle.

[0008] Means used to solve problems

[0009] A technical solution disclosed herein is a method for developing an autonomous driving vehicle, wherein the autonomous driving vehicle has external sensors whose specifications include quantity, type, and detection performance, and the autonomous driving vehicle has an autonomous driving system to which a machine learning model is applied. The method comprises: using a first autonomous driving vehicle having an external sensor with a first sensor specification to perform at least a matching process of the recognition process and the planning process for performing a path planning based on the result of the recognition process, and outputting a path by taking the detection result of the external sensor as input and outputting a path, and the method comprises: using a first autonomous driving vehicle having an external sensor with a first sensor specification to perform at least a matching process of the recognition process and the planning process, thereby performing a first step of training a machine learning model, and applying a second step of completing the machine learning model trained in the first step to a second autonomous driving vehicle having an external sensor with a second sensor specification whose quantity, type, and detection performance are at least inferior to the first sensor specification.

[0010] In a method for developing an autonomous driving vehicle of a technical solution of the present disclosure, in a first step, a first autonomous driving vehicle having a sensor specification superior to that of a second autonomous driving vehicle is used to train a machine learning model. In a second step, the machine learning model trained in the first autonomous driving vehicle is applied to the second autonomous driving vehicle. As a result, in the second autonomous driving vehicle, the trained machine learning model can be used to output a path equivalent to that of the first autonomous driving vehicle. Therefore, the effort of re-matching the second autonomous driving vehicle having a second sensor specification that is inferior to the first sensor specification in at least one of quantity, type, and detection performance can be reduced separately from the first autonomous driving vehicle.

[0011] Alternatively, in a method for developing an autonomous driving vehicle of one embodiment, a machine learning model is applied to the autonomous driving system for identification processing, planning processing, and vehicle control processing, and outputs the control amount of the autonomous driving control by using the detection results of external sensors as input. In a first step, the machine learning model is trained by using a first autonomous driving vehicle and matching the identification processing, planning processing, and vehicle control processing. In this case, the machine learning model including the vehicle control processing is trained in the first autonomous driving vehicle and applied to the second autonomous driving vehicle. As a result, the trained machine learning model can be used in the second autonomous driving vehicle to output the control amount of the autonomous driving control equivalent to that of the first autonomous driving vehicle.

[0012] Effects of the Invention

[0013] According to the present disclosure, it is possible to reduce the effort of re-matching a second autonomous driving vehicle separately from a first autonomous driving vehicle, wherein the second autonomous driving vehicle has a second sensor specification that is inferior to the first sensor specification in terms of at least one of quantity, type, and detection performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a block diagram of an autonomous driving vehicle used for training a machine learning model in a method for developing an autonomous driving vehicle in an embodiment.

[0015] Figure 2 This is a block diagram of an autonomous driving vehicle that illustrates an autonomous driving vehicle development method in an embodiment in which a trained machine learning model is applied.

[0016] Figure 3 Flowchart illustrating the steps of a method for developing an autonomous driving vehicle according to an exemplary embodiment. DETAILED DESCRIPTION

[0017] Hereinafter, exemplary embodiments will be described with reference to the accompanying drawings. In the following description, the same or corresponding elements may be denoted by the same reference numerals, and repeated descriptions may be omitted.

[0018] Figure 1 1 is a block diagram illustrating an autonomous driving vehicle used for training a machine learning model in a method for developing an autonomous driving vehicle according to an embodiment. Figure 1 As shown, the autonomous driving vehicle 10 is, for example, a vehicle such as a passenger car. The autonomous driving vehicle 10 is provided with an autonomous driving system 10a for driving the autonomous driving vehicle 10 in an autonomous driving manner. The autonomous driving system 10a may be mounted on the autonomous driving vehicle 10, or may be a technical solution in which a part of the functions is executed by a server capable of communicating with the autonomous driving vehicle 10.

[0019] The autonomous driving system 10 a of the autonomous driving vehicle 10 includes a first external sensor (external sensor) 11 , an internal sensor 12 , a map database 13 , a GNSS receiving unit 14 , an actuator 15 , and an autonomous driving ECU [Electronic Control Unit] 100 .

[0020] The first external sensor 11 is a detection device that detects the surrounding conditions (external environment) of the autonomous driving vehicle 10. The sensor specification (first sensor specification) of the first external sensor 11 is, for example, the highest level sensor specification of the manufacturer of the autonomous driving vehicle 10. The sensor specification includes the number, type, and detection performance of the external sensor.

[0021] The first external sensor 11 includes an external camera group 11a that captures at least the front of the autonomous driving vehicle 10. The first external sensor 11 may also include a radar sensor. The first external sensor 11 may also be configured to reconstruct many properties of the external environment of the autonomous driving vehicle 10 (the position of the vehicle, the relative distance to other vehicles, the relative speed to other vehicles, the direction of other vehicles, the shape of the lane, the lighting status of the traffic light, etc.).

[0022] The external camera group 11a is a photographing device for photographing the external environment of the autonomous driving vehicle 10. The external camera group 11a includes a plurality of cameras of different types and detection performances. Types of the external camera group 11a include, for example, a telephoto camera, a wide-angle camera, a monocular camera, a stereo camera, etc. Detection performance of the external camera group 11a includes, for example, resolution, night image quality, etc. The stereo camera has two photographing units configured to reproduce binocular parallax. The photographing information of the stereo camera also includes information in the depth direction. The external camera group 11a sends the photographing information related to the external environment of the autonomous driving vehicle 10 to the autonomous driving ECU 100.

[0023] The external camera group 11a includes, for example, a front camera installed on the inner side of the front window glass of the autonomous driving vehicle 10. The external camera group 11a may also include a side camera installed on a door mirror of the autonomous driving vehicle 10. The external camera group 11a may also include a rear camera installed on the rear window glass of the autonomous driving vehicle 10. In addition, the external camera group 11a may also include a camera for photographing the external environment of the autonomous driving vehicle 10 for autonomous driving.

[0024] The radar sensor is a detection device that uses radio waves (e.g., millimeter waves) or light to detect objects around the autonomous driving vehicle 10. The radar sensor sends radio waves or light to the surroundings of the autonomous driving vehicle 10, and detects the object by receiving the radio waves or light reflected by the object. The radar sensor sends the detected object information to the autonomous driving ECU 100.

[0025] The radar sensor includes multiple radar sensors of different types and detection performances. As types of radar sensors, for example, LiDAR [Light Detection And Ranging] and millimeter wave radars can be listed. As detection performance of radar sensors, for example, resolution and measurable distance can be listed. Here, the radar sensor includes, for example, LiDAR11b and millimeter wave radar 11c.

[0026] The internal sensor 12 is a detection device for detecting the driving state of the autonomous driving vehicle 10. The internal sensor 12 includes a vehicle speed sensor. The vehicle speed sensor is a detector for detecting the speed of the autonomous driving vehicle 10. As the vehicle speed sensor, for example, a well-known wheel speed sensor can be used. The vehicle speed sensor sends the detected vehicle speed information to the autonomous driving ECU 100.

[0027] The internal sensor 12 may also include a known acceleration sensor and a yaw rate sensor. The acceleration sensor is a detector that detects the acceleration of the autonomous driving vehicle 10. The acceleration sensor sends the acceleration information of the autonomous driving vehicle 10 to the autonomous driving ECU 100. The yaw rate sensor is a detector (e.g., a wheel speed sensor) that detects the yaw rate (rotational angular velocity) of the autonomous driving vehicle 10 about the vertical axis of the center of gravity. The yaw rate sensor sends the detected yaw rate information of the vehicle to the autonomous driving ECU 100.

[0028] The map database 13 is a storage device for storing map information. The map database 13 is formed in a storage medium such as a HDD [Hard Disk Drive] mounted on the autonomous driving vehicle 10. The map information includes information about the location and shape of roads, intersections, and branch points (for example, the location and shape of intersections, the width of lanes, lane lines, and the location of crosswalks). In addition, the map information may also include information about structures (the location and type of traffic lights installed at intersections), various traffic rules information (one-way driving, prohibition of left and right turns, prohibition of turning information, etc.). In addition, a part of the map information contained in the map database 13 may also be stored in a storage device different from the HDD. The map database 13 may also be formed in a computer of a facility such as a management center that can communicate with the autonomous driving vehicle 10.

[0029] The GNSS receiving unit 14 receives signals from positioning satellites to measure the position of the vehicle (for example, the latitude and longitude of the vehicle). The GNSS receiving unit 14 transmits the measured position information of the vehicle to the automatic driving ECU 100 .

[0030] The actuator 15 is a device for automatic driving control of the automatic driving vehicle 10. The actuator 15 includes at least a driving actuator, a braking actuator, and a steering actuator. The driving actuator controls the amount of air supplied to the engine (throttle opening) according to, for example, a control signal from the automatic driving ECU 100, thereby controlling the driving force of the automatic driving vehicle 10. In addition, in the case where the automatic driving vehicle 10 is a hybrid vehicle, in addition to the amount of air supplied to the engine, a control signal from the automatic driving ECU 100 is input to the motor serving as a power source to control the driving force. In the case where the automatic driving vehicle 10 is an electric vehicle, a control signal from the automatic driving ECU 100 is input to the motor serving as a power source to control the driving force. The motor serving as a power source in these cases constitutes the actuator 15.

[0031] The brake actuator controls the brake system according to the control signal from the automatic driving ECU 100, and controls the braking force applied to the wheels of the automatic driving vehicle 10. As the brake system, for example, a hydraulic brake system can be used. The steering actuator controls the drive of the auxiliary motor that controls the steering torque in the electric power steering system according to the control signal from the automatic driving ECU 100. Thus, the steering actuator controls the steering torque according to the control signal from the automatic driving ECU 100.

[0032] The autonomous driving ECU 100 is a main component of the autonomous driving system 10a. The autonomous driving ECU 100 performs autonomous driving control of the autonomous driving vehicle 10. The autonomous driving ECU 100 is an electronic control unit having a CPU [Central Processing Unit], a ROM [Read Only Memory], a RAM [Random Access Memory], a CAN [Controller Area Network] communication circuit, and the like. The autonomous driving ECU 100 is connected to a communication network using, for example, a CAN communication circuit, and is connected to the components of the autonomous driving vehicle 10 in a communicative manner. For example, based on a signal output by the CPU, the autonomous driving ECU 100 operates the CAN communication circuit to input and output data, stores data in the RAM, imports the program stored in the ROM into the RAM, and executes the program imported into the RAM by the CPU, thereby realizing the autonomous driving function. The autonomous driving ECU 100 can also be composed of multiple electronic units.

[0033] Next, an example of the functional configuration of the automatic driving ECU 100 will be described. The automatic driving ECU 100 includes a recognition processing unit 110 , a planning processing unit 120 , and a vehicle control processing unit 130 .

[0034] The autonomous driving system 10a is configured to be able to perform recognition processing, planning processing, and vehicle control processing. Recognition processing is a process of recognizing the external environment based on the detection results of the external sensors mounted on the autonomous driving vehicle 10. Planning processing is a process of performing path planning using the results of recognition processing. Vehicle control processing is a process of performing autonomous driving control based on the planned path.

[0035] In the automatic driving system 10a, as an example, a machine learning model is applied to each of the recognition processing, the planning processing, and the vehicle control processing. The machine learning model here includes: a first recognition model 111 of the recognition processing unit 110, a first planning model 121 of the planning processing unit 120, and a first vehicle control model 131 of the vehicle control processing unit 130.

[0036] The recognition processing unit 110 uses the first recognition model 111 to perform recognition processing based on the detection result of the first external sensor 11. The first recognition model 111 is, for example, a machine learning model obtained by performing learning based on deep learning in a manner of recognizing objects (such as other vehicles and traffic lights, etc.) based on the captured images of the external camera group 11a, and a machine learning model obtained by performing machine learning (such as deep learning) in a manner of recognizing objects (such as lane lines, etc.) based on the detection results of the radar sensor. The neural network constituting the first recognition model 111 is, for example, a convolutional neural network [CNN: Convolutional neural network] including multiple layers such as a plurality of convolutional layers and a pooling layer. The neural network can also be configured as a recurrent neural network [RNN: Recurrent neural network]. The recognition processing unit 110 obtains the recognition result of the object output from the first recognition model 111 by inputting the detection result of the first external sensor 11 into the first recognition model 111.

[0037] The recognition processing unit 110 recognizes the external environment of the autonomous driving vehicle 10 based on the recognition result of the object output from the first recognition model 111. The external environment includes the relative position of the surrounding objects with respect to the autonomous driving vehicle 10. The external environment may also include the relative speed and movement direction of the surrounding objects with respect to the autonomous driving vehicle 10.

[0038] The recognition processing unit 110 recognizes the driving state of the autonomous driving vehicle 10 based on the detection results of the internal sensor 12. The driving state includes, for example, the speed of the autonomous driving vehicle 10, the acceleration of the autonomous driving vehicle 10, and the yaw rate of the autonomous driving vehicle 10. The recognition processing unit 110 obtains the road environment (for example, road shape, intersection shape, type of signal, etc.) around the autonomous driving vehicle 10 based on map information, for example.

[0039] The recognition processing unit 110 obtains the current vehicle position of the autonomous driving vehicle 10 based on the position information from the GNSS receiving unit 14, for example. The recognition processing unit 110 may also use SLAM technology to obtain the current vehicle position of the autonomous driving vehicle 10. In addition, the recognition processing unit 110 may also obtain the current vehicle position of the autonomous driving vehicle 10 by a known method.

[0040] The planning processing unit 120 generates a path plan using the first planning model 121. The first planning model 121 is a machine learning model obtained by performing machine learning (e.g., deep learning) in a manner such that a path plan is output based on the recognition result of the object output from the first recognition model 111, the target route, the current vehicle position of the autonomous driving vehicle 10, and the driving state of the autonomous driving vehicle 10. The neural network constituting the first planning model 121 can be either a convolutional neural network or a recursive neural network. The neural network constituting the first planning model 121 can be either the same type as the first recognition model 111 or a different type of neural network from the first recognition model 111.

[0041] The path plan is a plan related to the path of the autonomous driving vehicle 10. The path plan is used in the autonomous driving of the autonomous driving vehicle 10. The path plan may also be continuous coordinate data on a map. The path plan may also be data containing the target lateral position of the driving lane (the position of the control target in the width direction of the driving lane) relative to the target longitudinal position (the position of the control target in the extension direction of the driving lane). The path plan may also be data containing the target steering angle of the driving lane relative to the target longitudinal position. The path plan may also be time series data of the target lateral position relative to the lane and / or time series data of the target steering angle. In the path plan, the target steering torque may also be used instead of the target steering angle.

[0042] The vehicle control processing unit 130 may also use the first vehicle control model 131 to perform vehicle control. The first vehicle control model 131 is, for example, a machine learning model obtained by performing machine learning (e.g., deep learning) in a manner that outputs a control amount of autonomous driving control for causing the autonomous driving vehicle 10 to automatically drive according to the path plan based on the path plan output from the first planning model 121 and the driving state of the autonomous driving vehicle 10. The neural network constituting the first vehicle control model 131 may be either a convolutional neural network or a recursive neural network. The neural network constituting the first vehicle control model 131 may also be of the same type as at least one of the first recognition model 111 and the first planning model 121. The neural network constituting the first vehicle control model 131 may also be a neural network of a different type from the first recognition model 111 and the first planning model 121. The control amount of autonomous driving control can be set, for example, as data (steering angle plan and vehicle speed plan) of the steering angle and vehicle speed of the autonomous driving vehicle 10 set at predetermined intervals on the target route. The vehicle control processing unit 130 controls the actuator 15 in accordance with the path plan output from the first planning model 121 , and performs autonomous driving of the autonomous driving vehicle 10 .

[0043] Generally speaking, the lower the support level of the autonomous driving control, the easier it is to use cheap external sensors. On the contrary, the higher the support level, the more expensive external sensors are used, and the more the number of external sensors increases. As the composition of external sensors is different, it may not be possible to use a system with a high support level in the past. Therefore, even if an autonomous driving system with a high support level is developed first, it is sometimes necessary to develop an autonomous driving system with a low support level separately. As a result, development costs may increase.

[0044] Therefore, in the method for developing an autonomous driving vehicle of the present embodiment, the first recognition model 111, the first planning model 121, and the first vehicle control model 131 can be applied to the autonomous driving vehicle 20 having the second external sensor 21. The first recognition model 111, the first planning model 121, and the first vehicle control model 131 are trained using the autonomous driving vehicle 10 having the first external sensor 11.

[0045] Figure 2 1 is a block diagram illustrating an autonomous driving vehicle to which a trained machine learning model is applied in a method for developing an autonomous driving vehicle according to an embodiment. Figure 2 As shown, the autonomous driving vehicle 20 has a configuration similar to that of the autonomous driving vehicle 10. The autonomous driving vehicle 20 is a vehicle such as a passenger car, and includes an autonomous driving system 20a that causes the autonomous driving vehicle 20 to travel in an autonomous manner.

[0046] The autonomous driving system 20 a of the autonomous driving vehicle 20 includes a second external sensor (external sensor) 21 , an internal sensor 22 , a map database 23 , a GNSS receiving unit 24 , an actuator 25 , and an autonomous driving ECU 200 .

[0047] The sensor specification (second sensor specification) of the automatic driving system 20a is inferior to the first sensor specification in at least one of the quantity, type, and detection performance. "Inferior" in terms of quantity means a small number. "Inferior" in terms of type means a simpler structure among external sensors with different structures. "Inferior" in terms of detection performance means, for example, that the detection performance is further suppressed in external sensors with a common structure. "Inferior" in terms of detection performance may also mean that the detection performance is further suppressed in external sensors with different structures.

[0048] The second external sensor 21 includes, for example, an external camera 21a and a millimeter wave radar 21C. The external camera 21a may be, for example, a sensor specification that is smaller than the number of cameras in the external camera group 11a by omitting a portion of the cameras in the external camera group 11a. In this case, the external camera 21a may also be a sensor specification that is the same as that of a portion of the cameras included in the external camera group 11a.

[0049] The LiDAR is omitted from the radar sensor in the second external sensor 21. The millimeter wave radar 21C may be the same in number, type, and detection performance as the millimeter wave radar 11C.

[0050] The automatic driving system 20 a may also have basically the same configuration except for the second external sensor 21 (the internal sensor 22 , the map database 23 , the GNSS receiving unit 24 , and the actuator 25 ).

[0051] In the automatic driving system 20a, the automatic driving ECU 200 has an identification processing unit 210, a planning processing unit 220, and a vehicle control processing unit 230. In the automatic driving system 20a, for example, the second identification model 211, the second planning model 221, and the second vehicle control model 231 can also be understood as a unit of an integrated machine learning model that takes the detection result of the second external sensor 21 as an input, outputs a path, and performs vehicle control.

[0052] As the second recognition model 211 of the recognition processing unit 210, the first recognition model 111 trained in the automatic driving system 10a is applied. In this case, the external camera 21a can be used as the input of the first recognition model 111 trained in the automatic driving system 10a instead of the camera with the same specification as the external camera 21a among the cameras included in the external camera group 11a. In addition, as the input of the first recognition model 111 trained in the automatic driving system 10a, it is also possible to set it as no input from LiDAR.

[0053] The recognition processing unit 210 inputs the detection result of the second external sensor 21 into the second recognition model 211, thereby obtaining the recognition result of the object output from the second recognition model 211. The obtained result is the same as the result obtained by inputting the detection result of the second external sensor 21 into the first recognition model 111, thus saving the effort of matching in the second recognition model 211.

[0054] The second planning model 221 as the planning processing unit 220 is adapted to the first planning model 121 trained in the autonomous driving system 10a. The difference in the specification of the second external sensor 21 is adjusted when input to the second recognition model 211. Therefore, the second planning model 221 can output the path plan of the autonomous driving vehicle 20 based on the recognition result of the object output from the second recognition model 211, the target route, the current vehicle position of the autonomous driving vehicle 20, and the driving state of the autonomous driving vehicle 20.

[0055] The vehicle control processing unit 230 may also use the second vehicle control model 231 to control the vehicle. The second vehicle control model 231 may also be the first vehicle control model 131 trained in the autonomous driving system 10a. The vehicle control processing unit 230, for example, based on the path plan output from the second planning model 221 and the driving state of the autonomous driving vehicle 20, causes the autonomous driving vehicle 20 to automatically drive according to the path plan of the autonomous driving vehicle 20.

[0056] Next, refer to Figure 3 Explain the steps of the development method of an autonomous vehicle. Figure 3 It is a flowchart of the steps of the method for developing an autonomous driving vehicle in an implementation manner. Figure 3 The flowchart shown can be executed, for example, when developing the automated driving vehicle 10 in advance, in parallel with the development, or after the development of the automated driving vehicle 20 .

[0057] like Figure 3As shown, in the development method of the autonomous driving vehicle of the embodiment, as S1 (first step), a first autonomous driving vehicle having an external sensor with a first sensor specification is used to train a machine learning model by performing at least recognition processing and matching of planning processing. In this embodiment, for example, an autonomous driving vehicle 10 having a first external sensor 11 is used to perform recognition processing and matching of planning processing, thereby training a first recognition model 111 and a first planning model 121, wherein the first external sensor 11 has an external camera group 11a, a LiDAR 11b, and a millimeter wave radar 11C.

[0058] As S2 (second step), the machine learning model trained in the first step is applied to the second autonomous driving vehicle with an external sensor of the second sensor specification. In this embodiment, for example, the first recognition model 111 and the first planning model 121 trained in step S1 are applied to the autonomous driving vehicle 20 with the second external sensor 21, and the second external sensor 21 has an external camera 21a and a millimeter wave radar 21C. After that, the training is completed. Figure 3 steps.

[0059] As described above, in the method for developing an autonomous driving vehicle of the embodiment, in step S1, at least the first recognition model 111 and the first planning model 121 are trained using the autonomous driving vehicle 10 having a sensor specification superior to that of the autonomous driving vehicle 20. In step S2, the first recognition model 111 and the first planning model 121 trained in the autonomous driving vehicle 10 are applied to the autonomous driving vehicle 20. As a result, in the autonomous driving vehicle 20, the trained first recognition model 111 and the first planning model 121 can be used to output a path equivalent to that of the autonomous driving vehicle 10. Therefore, the effort of re-matching the autonomous driving vehicle 20 having the second sensor specification whose number, type, and detection performance are inferior to the first sensor specification can be reduced separately from the autonomous driving vehicle 10.

[0060] In the development method of the autonomous driving vehicle of the embodiment, the autonomous driving system 10a is applied with a machine learning model that outputs the control amount of the autonomous driving control by taking the detection results of the external sensor as input for the recognition processing, the planning processing, and the vehicle control processing. In step S1, the first recognition model 111, the first planning model 121, and the first vehicle control model 131 are trained by matching the recognition processing, the planning processing, and the vehicle control processing using the autonomous driving vehicle 10. According to this configuration, the machine learning model that also includes the vehicle control processing is trained in the autonomous driving vehicle 10 and applied to the autonomous driving vehicle 20. As a result, in the autonomous driving vehicle 20, the first recognition model 111, the first planning model 121, and the first vehicle control model 131, which are the trained machine learning models, can be used to output the control amount of the autonomous driving control equivalent to that of the autonomous driving vehicle 10.

[0061] In addition, according to the method for developing an autonomous driving vehicle of the implementation mode, not only the development cost of the autonomous driving vehicle 20 of the cheaper autonomous driving system 20a is suppressed, but also the performance can be further improved by learning the output of the expensive autonomous driving system 10a compared to the case of developing the autonomous driving system 20a alone.

[0062] [Modifications]

[0063] Various exemplary embodiments have been described above, but the present invention is not limited to the exemplary embodiments described above, and various omissions, substitutions, and changes may be made.

[0064] In the above embodiment, the first vehicle control model 131 trained in the autonomous driving vehicle 10 is applied to the autonomous driving vehicle 20, but the present invention is not limited to this example. The machine learning model can also be trained by matching at least the recognition process and the planning process. In this case, in the autonomous driving system, at least the machine learning model for the recognition process and the machine learning model for the planning process can also be understood as a unit of an integrated machine learning model that takes the detection results of the external sensor as input and outputs a path.

[0065] In the above embodiment, the second external sensor 21 having an external camera 21a and a millimeter-wave radar 21C is illustrated, but "the second sensor specification in which at least one of the quantity, type, and detection performance is inferior to the first sensor specification" is not limited to this example. For example, in a case where the resolution of the external camera is different from the resolution of all cameras included in the external camera group 11a, the resolution of the external camera can be converted to match the resolution of the cameras included in the external camera group 11a, and used as the input of the machine learning model trained in the automatic driving system 10a. Alternatively, in a case where the image quality of the external camera is different from the image quality of all cameras included in the external camera group 11a, noise is given to the image captured by the external camera 21a in a manner that matches the image quality of the cameras included in the external camera group 11a, so that it can be used as the input of the machine learning model trained in the automatic driving system 10a. The same applies to the resolution and noise of the lidar.

Claims

1. A method for developing an autonomous driving vehicle, wherein the autonomous driving vehicle has external sensors including quantity, type, and detection performance as sensor specifications, The autonomous driving vehicle includes an autonomous driving system to which a machine learning model is applied, wherein the machine learning model uses the detection result of the external sensor as input and outputs the path for at least the recognition processing and the planning processing among recognition processing for recognizing an external environment based on the detection result of the external sensor, planning processing for performing path planning using the result of the recognition processing, and vehicle control processing for performing autonomous driving control based on the planned path, The development method comprises: In a first step, the first autonomous driving vehicle having the external sensor of the first sensor specification is used to perform at least the recognition processing and the matching of the planning processing, thereby training the machine learning model; and In the second step, the machine learning model trained in the first step is applied to the second autonomous driving vehicle having the external sensor of the second sensor specification, wherein the second sensor specification is inferior to the first sensor specification in at least one of quantity, type, and detection performance.

2. The method for developing an autonomous driving vehicle according to claim 1, wherein the method is applied to the autonomous driving system: the machine learning model that uses the detection result of the external sensor as input and outputs the control amount of the autonomous driving control for the recognition processing, the planning processing, and the vehicle control processing, In the first step, the first autonomous driving vehicle is used to perform matching of the recognition processing, the planning processing, and the vehicle control processing, thereby training the machine learning model.

Images