WO2021205616A1 - Moving body control device, moving body control method, and learning device - Google Patents

Moving body control device, moving body control method, and learning device Download PDF

Info

Publication number
WO2021205616A1
WO2021205616A1 PCT/JP2020/016019 JP2020016019W WO2021205616A1 WO 2021205616 A1 WO2021205616 A1 WO 2021205616A1 JP 2020016019 W JP2020016019 W JP 2020016019W WO 2021205616 A1 WO2021205616 A1 WO 2021205616A1
Authority
WO
WIPO (PCT)
Prior art keywords
feature amount
calculation unit
moving body
sensor data
unit
Prior art date
Application number
PCT/JP2020/016019
Other languages
French (fr)
Japanese (ja)
Inventor
貴之 井對
Original Assignee
三菱電機株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 三菱電機株式会社 filed Critical 三菱電機株式会社
Priority to PCT/JP2020/016019 priority Critical patent/WO2021205616A1/en
Publication of WO2021205616A1 publication Critical patent/WO2021205616A1/en

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W60/00—Drive control systems specially adapted for autonomous road vehicles

Definitions

  • the present disclosure relates to a mobile body control device for calculating a control amount of a mobile body, a mobile body control method, and a learning device.
  • Patent Document 1 a technique for learning the control amount of a vehicle for each driving situation is known (for example, Patent Document 1).
  • Hyperparameters are weights of evaluation functions and the like.
  • the present disclosure has been made to solve the above-mentioned problems, and provides a mobile body control device that can set hyperparameters according to a driving situation, which is used in a mobile body control technology, without human intervention.
  • the purpose is to do.
  • the moving body control device is a moving body control device that calculates a control amount of a moving body from sensor data indicating the surrounding environment of the moving body, and is an actual sensor data acquisition unit that acquires sensor data and an actual sensor. Used when calculating the control amount of a moving body by inputting the feature amount calculation unit that calculates the feature amount from the sensor data acquired by the data acquisition unit and the feature amount calculated by the feature amount calculation unit into the machine learning model. It is equipped with a hyper parameter calculation unit that calculates hyper parameters.
  • hyperparameters according to the driving situation used in the mobile control technology can be set without human intervention.
  • FIG. 1 It is a figure which shows the structural example of the mobile body control apparatus which concerns on Embodiment 1.
  • FIG. It is a flowchart for demonstrating operation of the mobile body control apparatus which concerns on Embodiment 1.
  • FIG. In the first embodiment a configuration example of the mobile control system in the case where the learning device is provided outside the mobile control device and the mobile control device and the learning device constitute a mobile control system.
  • FIG. 4A and 4B are diagrams showing an example of the hardware configuration of the mobile control device according to the first embodiment.
  • FIG. 1 is a diagram showing a configuration example of the mobile control device 1 according to the first embodiment.
  • the moving body is assumed to be a vehicle. Further, it is assumed that the mobile control device 1 according to the first embodiment is mounted on a vehicle (not shown).
  • the moving body control device 1 calculates the control amount of the moving body according to the traveling situation in which the moving body is traveling.
  • the traveling condition means various shapes of a road on which a moving body travels, such as a straight road, a curve, an uphill, a downhill, or an intersection.
  • the traveling speed when the moving body travels on the road of various shapes is also included in the traveling situation.
  • the control amount of the moving body is a control amount for controlling the operation of the moving body.
  • the mobile body control device 1 calculates the control amount of the mobile body by using a known mobile body control technique.
  • the mobile body control device 1 calculates the control amount of the mobile body by using a known model predictive control technique. At that time, the moving body control device 1 calculates the hyperparameters used when calculating the control amount of the moving body according to the traveling situation in which the moving body is traveling. In model prediction control, a model that predicts future behavior based on vehicle dynamics is generated in advance, and what kind of input is optimal based on the model based on the evaluation function and constraints. Is calculated.
  • the hyperparameter is the weight of the evaluation function in the model prediction control or the threshold value in the constraint condition.
  • the mobile body control device 1 calculates the optimum control amount of the moving body to be given to the moving body based on the model prediction control.
  • the hyperparameters are proportional gain, integral gain, and differential gain.
  • the moving body control device 1 calculates hyperparameters based on a trained model in machine learning (hereinafter referred to as "machine learning model").
  • the machine learning model is generated by the mobile control device 1.
  • the control amount calculated by the moving body control device 1 is used for automatic driving control in a moving body, in other words, a vehicle.
  • a vehicle In the first embodiment, it is assumed that the vehicle has an automatic driving function. Even when the vehicle has an automatic driving function, the driver can drive the vehicle by himself / herself without executing the automatic driving function.
  • the moving body control device 1 includes a learning device 2, an actual sensor data acquisition unit 14, a data conversion unit 15, a feature amount calculation unit 16, a hyperparameter calculation unit 17, and a control amount calculation unit 18. Be prepared.
  • the learning device 2 generates a machine learning model 13 by learning using the teacher data.
  • the learning device 2 includes a teacher data acquisition unit 11, a learning unit 12, and a machine learning model 13.
  • the teacher data acquisition unit 11 acquires teacher data.
  • the teacher data is data in which hyperparameters are added to the feature amount calculated from the sensor data indicating the surrounding environment of the moving body.
  • the feature amount calculated from the sensor data indicating the surrounding environment of the moving body is the feature amount according to the traveling condition of the moving body.
  • the teacher data acquisition unit 11 outputs the acquired teacher data to the learning unit 12.
  • the learning unit 12 generates a machine learning model 13 by learning using the teacher data acquired by the teacher data acquisition unit 11.
  • the machine learning model 13 is a machine learning model that inputs a feature amount calculated from sensor data indicating the surrounding environment of a moving body and outputs hyperparameters.
  • the machine learning model 13 is provided in the moving body control device 1, but this is only an example.
  • the machine learning model 13 is a moving body control device. It may be provided in a place outside the device 1 where the mobile control device 1 can be referred to.
  • the actual sensor data acquisition unit 14 acquires sensor data (hereinafter referred to as “actual sensor data”) indicating the surrounding environment of the moving body when the moving body actually travels.
  • the actual sensor data is, for example, an image.
  • the actual sensor data acquisition unit 14 acquires, for example, an image captured by a camera (not shown) that captures the front of the moving body.
  • the camera is provided on the moving body.
  • the actual sensor data will be described below assuming that the actual sensor data is an image captured by the camera (hereinafter referred to as “camera image”).
  • the actual sensor data acquisition unit 14 outputs the camera image to the data conversion unit 15.
  • the actual sensor data may be numerical data such as LiDAR data.
  • the data conversion unit 15 performs data conversion on the data elements included in the actual sensor data acquired by the actual sensor data acquisition unit 14. For example, the data conversion unit 15 performs the above data conversion using a known semantic segmentation technique. To give a specific example, for example, among the pixels included in the camera image, the pixel indicating a car is blue, the pixel indicating a road is pink, or the pixel indicating a street tree is green. Data conversion is performed to color-code the pixels of the camera image. When the actual sensor data is numerical data, the data conversion unit 15 performs data conversion for removing noise, for example, so that the numerical data approaches the simulation data indicating the surrounding environment of the moving body.
  • the features included in the teacher data are the features calculated from the simulation data.
  • the feature amount included in the teacher data is a feature amount calculated from an image (hereinafter referred to as “simulation image”) reproduced by an automatic driving simulator (not shown).
  • the teacher data is, for example, a hyper that calculates a feature amount calculated from a simulation image and an ideal control amount when calculating a control amount of a moving body based on the feature amount in an automatic driving simulator.
  • the parameter is the associated data.
  • the automatic driving simulator is a so-called automatic driving simulator using a general simulation technique. In the automatic driving simulator, a simulation image showing the surrounding environment of the moving body is reproduced.
  • the simulation image is, for example, a CG (Computer Graphics) image.
  • the feature amount to be calculated as the same feature amount may not be calculated as the same feature amount in the case where the feature amount is calculated from the camera image and the case where the feature amount is calculated from the simulation image.
  • the feature amount calculated from the camera image is required when the control amount calculation unit 18 calculates the control amount of the moving body in the moving body control device 1.
  • the control amount calculation unit 18 uses the hyperparameters calculated by the hyperparameter calculation unit 17 based on the machine learning model 13.
  • the feature amount calculation unit 16 calculates the feature amount based on the camera image. Details of the control amount calculation unit 18, the feature amount calculation unit 16, and the hyperparameter calculation unit 17 will be described later.
  • the hyperparameters based on the machine learning model 13 generated based on the teacher data including the feature amount calculated from the simulation image If is used, an appropriate control amount may not be calculated.
  • the data conversion unit 15 performs data conversion on the camera image, so that the feature amount to be calculated as the same feature amount is generated depending on whether the feature amount is calculated from the camera image or the simulation image. Absorb the difference.
  • the hyperparameters used when calculating the control amount of the moving body are based on the feature amount different from the feature amount when calculating the control amount of the moving body. Therefore, it is possible to reduce the possibility that the control amount is not calculated appropriately.
  • the same data conversion as the data conversion performed by the data conversion unit 15 on the camera image needs to be performed on the simulation image before the feature amount is calculated.
  • the data conversion unit 15 outputs the data-converted camera image (hereinafter referred to as “converted camera image”) to the feature amount calculation unit 16.
  • the feature amount calculation unit 16 calculates the feature amount according to the traveling state of the moving body from the converted camera image after the data conversion unit 15 has converted.
  • the feature amount calculation unit 16 calculates the feature amount by using a known technique such as image processing or machine learning.
  • the feature amount calculation unit 16 outputs the calculated feature amount to the hyperparameter calculation unit 17.
  • the feature amount calculation unit 16 outputs the calculated feature amount in association with, for example, the converted camera image.
  • the hyperparameter calculation unit 17 calculates the hyperparameters used when calculating the control amount of the moving body by inputting the feature amount calculated by the feature amount calculation unit 16 into the machine learning model 13.
  • the hyperparameter calculation unit 17 outputs the calculated hyperparameters to the control amount calculation unit 18.
  • the control amount calculation unit 18 calculates the control amount of the moving body based on the feature amount calculated by the feature amount calculation unit 16 and the hyperparameters calculated by the hyperparameter calculation unit 17.
  • the control amount calculation unit 18 calculates the control amount of the moving body by the known model prediction control.
  • the control amount calculation unit 18 outputs the calculated control amount of the moving body to an external device (not shown).
  • the external device is, for example, an automatic driving control device (not shown) that controls the automatic driving of the vehicle.
  • the automatic driving control device automatically drives the vehicle based on the control amount output from the control amount calculation unit 18.
  • the hyperparameter calculation unit 17 may output the calculated hyperparameters to the learning unit 12. At this time, the hyperparameter calculation unit 17 also outputs the feature amount when the hyperparameter is calculated. Then, the learning unit 12 may cause the machine learning model 13 to learn based on the hyperparameters and the feature amount output from the hyperparameter calculation unit 17.
  • FIG. 2 is a flowchart for explaining the operation of the mobile control device 1 according to the first embodiment.
  • the teacher data acquisition unit 11 acquires teacher data (step ST201).
  • the teacher data acquisition unit 11 outputs the acquired teacher data to the learning unit 12.
  • the learning unit 12 generates a machine learning model 13 by learning using the teacher data acquired by the teacher data acquisition unit 11 in step ST201 (step ST202).
  • the actual sensor data acquisition unit 14 acquires the actual sensor data (step ST203). Specifically, the actual sensor data acquisition unit 14 acquires, for example, a camera image. The actual sensor data acquisition unit 14 outputs the camera image to the data conversion unit 15.
  • the data conversion unit 15 converts the data elements included in the actual sensor data acquired by the actual sensor data acquisition unit 14 in step ST203 for each set of data elements forming a characteristic category (step). ST204).
  • the data conversion unit 15 outputs the converted camera image to the feature amount calculation unit 16.
  • the feature amount calculation unit 16 calculates the feature amount according to the traveling state of the moving body from the converted camera image after the data conversion unit 15 has converted in step ST204 (step ST205).
  • the feature amount calculation unit 16 outputs the calculated feature amount to the hyperparameter calculation unit 17.
  • the feature amount calculation unit 16 outputs the calculated feature amount in association with, for example, the converted camera image.
  • the hyperparameter calculation unit 17 calculates the control amount of the moving body by inputting the feature amount calculated by the feature amount calculation unit 16 in step ST205 into the machine learning model 13 generated by the learning unit 12 in step ST202. The hyperparameters used in the above are calculated (step ST206). The hyperparameter calculation unit 17 outputs the calculated hyperparameters to the control amount calculation unit 18.
  • the control amount calculation unit 18 calculates the control amount of the moving body based on the feature amount calculated by the feature amount calculation unit 16 in step ST205 and the hyperparameters calculated by the hyperparameter calculation unit 17 in step ST206 (). Step ST207).
  • step ST201 and step ST202 may be performed before the operation of step ST206 is performed.
  • the hyperparameter calculation unit 17 may output the calculated hyperparameters to the learning unit 12. At this time, the hyperparameter calculation unit 17 also outputs the feature amount when the hyperparameter is calculated. After that, the learning unit 12 may cause the machine learning model 13 to learn based on the hyperparameters and the feature amount output from the hyperparameter calculation unit 17.
  • the mobile control device 1 calculates hyperparameters using machine learning. Specifically, the mobile control device 1 calculates hyperparameters by inputting the feature amount calculated from the actual sensor data into the machine learning model 13. Therefore, the mobile body control device 1 can set hyperparameters according to the traveling situation, which are used in the mobile body control technology, without human intervention.
  • the moving body control device 1 calculates the control amount of the moving body using the hyperparameters calculated based on the machine learning model 13. As a result, the moving body control device 1 can obtain a controlled amount of the moving body according to the traveling state of the moving body.
  • the moving body control device 1 generates a machine learning model 13 by learning using teacher data in which hyperparameters are added to feature quantities. Therefore, the mobile body control device 1 can set hyperparameters according to the traveling situation, which are used in the mobile body control technology, without human intervention.
  • the mobile control device 1 is provided with the data conversion unit 15, but the mobile control device 1 is not required to include the data conversion unit 15.
  • the feature amount calculation unit 16 may calculate the feature amount according to the traveling situation from the actual sensor data acquired by the actual sensor data acquisition unit 14.
  • the learning device 2 is provided in the mobile control device 1, but this is only an example.
  • the learning device 2 may be provided outside the mobile control device 1.
  • FIG. 3 shows, in the first embodiment, when the learning device 2 is provided outside the moving body control device 1 and the moving body control device 1a and the learning device 2 constitute a moving body control system. It is a figure which shows the configuration example of the moving body control system. In FIG. 3, the same reference numerals are given to the same configurations as those described with reference to FIG. 1, and duplicate description will be omitted.
  • the moving body control device 1a includes an actual sensor data acquisition unit 14, a data conversion unit 15, a feature amount calculation unit 16, a hyperparameter calculation unit 17, and a control amount calculation unit 18.
  • the mobile control device 1a shown in FIG. 3 is not required to include the data conversion unit 15.
  • the learning device 2 is provided in a server, for example, and is connected to the mobile control device 1a via a network.
  • the learning device 2 may be, for example, an in-vehicle device.
  • the moving body control devices 1 and 1a are in-vehicle devices mounted on the vehicle, and the actual sensor data acquisition unit 14, the data conversion unit 15, the feature quantity calculation unit 16, and the hyper. It is assumed that the parameter calculation unit 17 and the control amount calculation unit 18 are provided in the moving body control devices 1, 1a. Not limited to this, a part of the actual sensor data acquisition unit 14, the data conversion unit 15, the feature amount calculation unit 16, the hyperparameter calculation unit 17, and the control amount calculation unit 18 is mounted on the in-vehicle device of the vehicle.
  • the mobile control system may be configured by the in-vehicle device and the server, assuming that the other is provided in the server connected to the in-vehicle device via the network.
  • the feature amount calculation unit 16, the hyperparameter calculation unit 17, and the control amount calculation unit 18 are provided in the server, and the actual sensor data acquisition unit 14 and the data conversion unit 15 are provided in the in-vehicle device. May be.
  • the feature amount calculation unit 16 acquires the actual sensor data after data conversion from the in-vehicle device.
  • the control amount calculation unit 18 outputs the calculated control amount to the in-vehicle device.
  • FIG. 4A and 4B are diagrams showing an example of the hardware configuration of the mobile control devices 1 and 1a according to the first embodiment.
  • the functions of 18 are realized by the processing circuit 401. That is, the mobile body control device 1 includes a processing circuit 401 for performing control for calculating the control amount of the mobile body using machine learning.
  • the processing circuit 401 may be dedicated hardware as shown in FIG. 4A, or may be a CPU (Central Processing Unit) 405 that executes a program stored in the memory 406 as shown in FIG. 4B.
  • CPU Central Processing Unit
  • the processing circuit 401 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), or an FPGA (Field-Programmable). Gate Array) or a combination of these is applicable.
  • the processing circuit 401 is the CPU 405, the teacher data acquisition unit 11, the learning unit 12, the actual sensor data acquisition unit 14, the data conversion unit 15, the feature amount calculation unit 16, the hyperparameter calculation unit 17, and the control amount.
  • the function of the calculation unit 18 is realized by software, firmware, or a combination of software and firmware. That is, the teacher data acquisition unit 11, the learning unit 12, the actual sensor data acquisition unit 14, the data conversion unit 15, the feature amount calculation unit 16, the hyperparameter calculation unit 17, and the control amount calculation unit 18 are HDDs. It is realized by a processing circuit such as (Hard Disk Drive) 402, a CPU 405 that executes a program stored in a memory 406, and a system LSI (Data-Scale Integration).
  • the programs stored in the HDD 402, the memory 406, etc. are the teacher data acquisition unit 11, the learning unit 12, the actual sensor data acquisition unit 14, the data conversion unit 15, the feature amount calculation unit 16, and the hyperparameter calculation. It can also be said that the procedure or method of the unit 17 and the control amount calculation unit 18 is executed by the computer.
  • the memory 406 is, for example, a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable Read Only Memory), an EEPROM (Electrically Memory), etc.
  • a sexual or volatile semiconductor memory, a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, a DVD (Digital Versaille Disc), or the like is applicable.
  • the teacher data acquisition unit 11 and the actual sensor data acquisition unit 14 are realized by the processing circuit 401 as dedicated hardware, and the learning unit 12, the data conversion unit 15, the feature amount calculation unit 16, and the feature amount calculation unit 16
  • the functions of the hyperparameter calculation unit 17 and the control amount calculation unit 18 can be realized by the processing circuit 401 reading and executing the program stored in the memory 406.
  • the mobile body control device 1 includes a device such as an automatic operation control device, and an input interface device 403 and an output interface device 404 that perform wired communication or wireless communication.
  • the moving body control devices 1 and 1a calculate the control amount of the moving body from the sensor data (actual sensor data) indicating the surrounding environment of the moving body.
  • the feature amount calculation unit that calculates the feature amount from the actual sensor data acquisition unit 14 that acquires the sensor data (actual sensor data) and the sensor data (actual sensor data) acquired by the actual sensor data acquisition unit 14.
  • 16 and a hyper parameter calculation unit 17 for calculating a hyper parameter used when calculating a control amount of a moving body by inputting the feature amount calculated by the feature amount calculation unit 16 into the machine learning model 13 are provided. Configured. Therefore, the mobile body control devices 1, 1a can set hyperparameters according to the traveling situation, which are used in the mobile body control technology, without human intervention.
  • hyperparameters used when calculating the control amount of the moving body are added to the feature amount calculated from the sensor data indicating the surrounding environment of the moving body.
  • the moving body is a vehicle, but this is only an example.
  • the mobile body control devices 1 and 1a according to the first embodiment can be used as a device for calculating the control amount of the mobile body in various mobile bodies capable of automatic control.
  • the mobile body control device is configured so that hyperparameters according to the traveling situation used in the mobile body control technology can be set without human intervention, the movement for calculating the control amount of the mobile body is calculated. It can be applied to body control devices.
  • 1,1a Mobile control device 2 Learning device, 11 Teacher data acquisition unit, 12 Learning unit, 13 Machine learning model, 14 Actual sensor data acquisition unit, 15 Data conversion unit, 16 Feature amount calculation unit, 17 Hyperparameter calculation unit , 18 control amount calculation unit, 401 processing circuit, 402 HDD, 403 input interface device, 404 output interface device, 405 CPU, 406 memory.

Landscapes

  • Engineering & Computer Science (AREA)
  • Automation & Control Theory (AREA)
  • Human Computer Interaction (AREA)
  • Transportation (AREA)
  • Mechanical Engineering (AREA)
  • Traffic Control Systems (AREA)
  • Control Of Driving Devices And Active Controlling Of Vehicle (AREA)

Abstract

A moving body control device comprises an actual sensor data acquisition unit (14) that acquires sensor data, a feature amount calculation unit (16) that calculates a feature amount from the sensor data acquired by the actual sensor data acquisition unit (14), and a hyperparameter calculation unit (17) that inputs the feature amount calculated by the feature amount calculation unit (16) to a machine learning model (13) to thereby calculate a hyperparameter used when calculating a control amount of a moving body.

Description

移動体制御装置、移動体制御方法、および、学習装置Mobile control device, mobile control method, and learning device
　本開示は、移動体の制御量を算出する移動体制御装置、移動体制御方法、および、学習装置に関するものである。 The present disclosure relates to a mobile body control device for calculating a control amount of a mobile body, a mobile body control method, and a learning device.
　従来、移動体の自動運転の分野において、走行状況毎に車両の制御量を学習する技術が知られている（例えば、特許文献１）。 Conventionally, in the field of automatic driving of a moving body, a technique for learning the control amount of a vehicle for each driving situation is known (for example, Patent Document 1).
特開２０１９－１０９６７号公報JP-A-2019-10967
　モデル予測制御またはＰＩＤ制御等の移動体制御技術において、走行状況に応じた制御量を得るためには、走行状況に応じたハイパーパラメータを、人手で設定しなければならないという課題があった。ハイパーパラメータとは、評価関数の重み等である。 In mobile control technology such as model prediction control or PID control, there is a problem that hyperparameters according to the driving situation must be manually set in order to obtain a control amount according to the driving situation. Hyperparameters are weights of evaluation functions and the like.
　本開示は上記のような課題を解決するためになされたもので、移動体制御技術において用いられる、走行状況に応じたハイパーパラメータを、人手を介することなく設定可能とした移動体制御装置を提供することを目的とする。 The present disclosure has been made to solve the above-mentioned problems, and provides a mobile body control device that can set hyperparameters according to a driving situation, which is used in a mobile body control technology, without human intervention. The purpose is to do.
　本開示に係る移動体制御装置は、移動体の周辺環境を示すセンサデータから移動体の制御量を算出する移動体制御装置であって、センサデータを取得する実センサデータ取得部と、実センサデータ取得部が取得したセンサデータから特徴量を算出する特徴量算出部と、特徴量算出部が算出した特徴量を機械学習モデルに入力することにより、移動体の制御量を算出する際に用いるハイパーパラメータを算出するハイパーパラメータ算出部とを備えたものである。 The moving body control device according to the present disclosure is a moving body control device that calculates a control amount of a moving body from sensor data indicating the surrounding environment of the moving body, and is an actual sensor data acquisition unit that acquires sensor data and an actual sensor. Used when calculating the control amount of a moving body by inputting the feature amount calculation unit that calculates the feature amount from the sensor data acquired by the data acquisition unit and the feature amount calculated by the feature amount calculation unit into the machine learning model. It is equipped with a hyper parameter calculation unit that calculates hyper parameters.
　本開示によれば、移動体制御技術において用いられる、走行状況に応じたハイパーパラメータを、人手を介することなく設定することができる。 According to the present disclosure, hyperparameters according to the driving situation used in the mobile control technology can be set without human intervention.
実施の形態１に係る移動体制御装置の構成例を示す図である。It is a figure which shows the structural example of the mobile body control apparatus which concerns on Embodiment 1. FIG. 実施の形態１に係る移動体制御装置の動作を説明するためのフローチャートである。It is a flowchart for demonstrating operation of the mobile body control apparatus which concerns on Embodiment 1. FIG. 実施の形態１において、学習装置が、移動体制御装置の外部に備えられ、移動体制御装置と学習装置とで移動体制御システムを構成するものとした場合の、当該移動体制御システムの構成例を示す図である。In the first embodiment, a configuration example of the mobile control system in the case where the learning device is provided outside the mobile control device and the mobile control device and the learning device constitute a mobile control system. It is a figure which shows. 図４Ａ，図４Ｂは、実施の形態１に係る移動体制御装置のハードウェア構成の一例を示す図である。4A and 4B are diagrams showing an example of the hardware configuration of the mobile control device according to the first embodiment.
　以下、本開示の実施の形態について、図面を参照しながら詳細に説明する。
実施の形態１．
　図１は、実施の形態１に係る移動体制御装置１の構成例を示す図である。
　実施の形態１において、移動体とは、車両を想定している。また、実施の形態１に係る移動体制御装置１は、車両（図示省略）に搭載されることを想定している。
　移動体制御装置１は、移動体が走行している走行状況に応じて、当該移動体の制御量を算出する。実施の形態１において、走行状況とは、直線道路、カーブ、上り坂、下り坂、または、交差点等、移動体が走行する道路の、種々の形状を意味する。実施の形態１では、移動体が種々の形状の道路を走行する際の走行速度も、走行状況に含まれるものとする。また、実施の形態１において、移動体の制御量とは、移動体の運転制御を行うための制御量である。移動体制御装置１は、既知の移動体制御技術を用いて、移動体の制御量を算出する。 Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings.
Embodiment 1.
FIG. 1 is a diagram showing a configuration example of the mobile control device 1 according to the first embodiment.
In the first embodiment, the moving body is assumed to be a vehicle. Further, it is assumed that the mobile control device 1 according to the first embodiment is mounted on a vehicle (not shown).
The moving body control device 1 calculates the control amount of the moving body according to the traveling situation in which the moving body is traveling. In the first embodiment, the traveling condition means various shapes of a road on which a moving body travels, such as a straight road, a curve, an uphill, a downhill, or an intersection. In the first embodiment, the traveling speed when the moving body travels on the road of various shapes is also included in the traveling situation. Further, in the first embodiment, the control amount of the moving body is a control amount for controlling the operation of the moving body. The mobile body control device 1 calculates the control amount of the mobile body by using a known mobile body control technique.
　実施の形態１において、移動体制御装置１は、既知のモデル予測制御の技術を用いて、移動体の制御量を算出するものとする。その際、移動体制御装置１は、移動体が走行している走行状況に応じて、移動体の制御量を算出する際に用いるハイパーパラメータを算出する。
　モデル予測制御では、車両ダイナミクスに基づき未来の挙動を予測するモデルが予め生成されており、評価関数および制約条件のもと、当該モデルに基づいて、どのような入力を与えることが最適であるかが算出される。ハイパーパラメータとは、モデル予測制御における評価関数の重みまたは制約条件における閾値である。移動体制御装置１は、モデル予測制御に基づいて、移動体に対して与える、最適な移動体の制御量を算出する。なお、ＰＩＤ制御においては、ハイパーパラメータとは、比例ゲイン、積分ゲイン、微分ゲインである。
　移動体制御装置１は、ハイパーパラメータを、機械学習における学習済みのモデル（以下「機械学習モデル」という。）に基づいて算出する。機械学習モデルは、移動体制御装置１が生成する。
　移動体制御装置１が算出した制御量は、移動体、言い換えれば、車両において、自動運転制御に用いられる。実施の形態１において、車両は、自動運転機能を有することを前提とする。なお、車両が自動運転機能を有する場合であっても、運転者が、当該自動運転機能を実行せず、自ら車両を運転することができる。 In the first embodiment, the mobile body control device 1 calculates the control amount of the mobile body by using a known model predictive control technique. At that time, the moving body control device 1 calculates the hyperparameters used when calculating the control amount of the moving body according to the traveling situation in which the moving body is traveling.
In model prediction control, a model that predicts future behavior based on vehicle dynamics is generated in advance, and what kind of input is optimal based on the model based on the evaluation function and constraints. Is calculated. The hyperparameter is the weight of the evaluation function in the model prediction control or the threshold value in the constraint condition. The mobile body control device 1 calculates the optimum control amount of the moving body to be given to the moving body based on the model prediction control. In PID control, the hyperparameters are proportional gain, integral gain, and differential gain.
The moving body control device 1 calculates hyperparameters based on a trained model in machine learning (hereinafter referred to as "machine learning model"). The machine learning model is generated by the mobile control device 1.
The control amount calculated by the moving body control device 1 is used for automatic driving control in a moving body, in other words, a vehicle. In the first embodiment, it is assumed that the vehicle has an automatic driving function. Even when the vehicle has an automatic driving function, the driver can drive the vehicle by himself / herself without executing the automatic driving function.
　移動体制御装置１は、図１に示すように、学習装置２、実センサデータ取得部１４、データ変換部１５、特徴量算出部１６、ハイパーパラメータ算出部１７、および、制御量算出部１８を備える。 As shown in FIG. 1, the moving body control device 1 includes a learning device 2, an actual sensor data acquisition unit 14, a data conversion unit 15, a feature amount calculation unit 16, a hyperparameter calculation unit 17, and a control amount calculation unit 18. Be prepared.
　学習装置２は、教師データを用いた学習により、機械学習モデル１３を生成する。学習装置２は、教師データ取得部１１、学習部１２、および、機械学習モデル１３を備える。
　教師データ取得部１１は、教師データを取得する。
　実施の形態１において、教師データは、移動体の周辺環境を示すセンサデータから算出された特徴量にハイパーパラメータが付与されたデータである。移動体の周辺環境を示すセンサデータから算出された特徴量は、移動体の走行状況に応じた特徴量である。
　教師データ取得部１１は、取得した教師データを、学習部１２に出力する。 The learning device 2 generates a machine learning model 13 by learning using the teacher data. The learning device 2 includes a teacher data acquisition unit 11, a learning unit 12, and a machine learning model 13.
The teacher data acquisition unit 11 acquires teacher data.
In the first embodiment, the teacher data is data in which hyperparameters are added to the feature amount calculated from the sensor data indicating the surrounding environment of the moving body. The feature amount calculated from the sensor data indicating the surrounding environment of the moving body is the feature amount according to the traveling condition of the moving body.
The teacher data acquisition unit 11 outputs the acquired teacher data to the learning unit 12.
　学習部１２は、教師データ取得部１１が取得した教師データを用いた学習により、機械学習モデル１３を生成する。
　機械学習モデル１３は、移動体の周辺環境を示すセンサデータから算出された特徴量を入力とし、ハイパーパラメータを出力する機械学習モデルである。
　なお、実施の形態１では、図１に示すように、機械学習モデル１３は、移動体制御装置１に備えられるものとするが、これは一例に過ぎない、機械学習モデル１３は、移動体制御装置１の外部の、移動体制御装置１が参照可能な場所に備えられるようにしてもよい。 The learning unit 12 generates a machine learning model 13 by learning using the teacher data acquired by the teacher data acquisition unit 11.
The machine learning model 13 is a machine learning model that inputs a feature amount calculated from sensor data indicating the surrounding environment of a moving body and outputs hyperparameters.
In the first embodiment, as shown in FIG. 1, the machine learning model 13 is provided in the moving body control device 1, but this is only an example. The machine learning model 13 is a moving body control device. It may be provided in a place outside the device 1 where the mobile control device 1 can be referred to.
　実センサデータ取得部１４は、移動体が実際に走行する際の、移動体の周辺環境を示すセンサデータ（以下「実センサデータ」という。）を取得する。
　実施の形態１において、実センサデータは、例えば、画像である。実センサデータ取得部１４は、例えば、当該移動体の前方を撮像するカメラ（図示省略）が撮像した画像を、取得する。カメラは、移動体に設けられている。実施の形態１において、実センサデータは、カメラが撮像した画像（以下「カメラ画像」という。）であるものとして、以下説明する。実センサデータ取得部１４は、カメラ画像を、データ変換部１５に出力する。
　尚、実センサデータは、ＬｉＤＡＲデータ等の数値データであってもよい。 The actual sensor data acquisition unit 14 acquires sensor data (hereinafter referred to as “actual sensor data”) indicating the surrounding environment of the moving body when the moving body actually travels.
In the first embodiment, the actual sensor data is, for example, an image. The actual sensor data acquisition unit 14 acquires, for example, an image captured by a camera (not shown) that captures the front of the moving body. The camera is provided on the moving body. In the first embodiment, the actual sensor data will be described below assuming that the actual sensor data is an image captured by the camera (hereinafter referred to as “camera image”). The actual sensor data acquisition unit 14 outputs the camera image to the data conversion unit 15.
The actual sensor data may be numerical data such as LiDAR data.
　データ変換部１５は、実センサデータ取得部１４が取得した実センサデータに含まれているデータ要素に対してデータ変換を行う。例えば、データ変換部１５は、既知のセマンティックセグメンテーション技術を用いて、上記データ変換を行う。具体例を挙げると、例えば、データ変換部１５は、カメラ画像に含まれている画素のうち、車を示す画素を青、道路を示す画素をピンク、または、街路樹を示す画素を緑というように、カメラ画像の画素を色分けするデータ変換を行う。また、実センサデータが数値データの場合には、データ変換部１５は、例えば、当該数値データについて、移動体の周辺環境を示すシミュレーションデータに近付けるよう、ノイズを除去するデータ変換を行う。 The data conversion unit 15 performs data conversion on the data elements included in the actual sensor data acquired by the actual sensor data acquisition unit 14. For example, the data conversion unit 15 performs the above data conversion using a known semantic segmentation technique. To give a specific example, for example, among the pixels included in the camera image, the pixel indicating a car is blue, the pixel indicating a road is pink, or the pixel indicating a street tree is green. Data conversion is performed to color-code the pixels of the camera image. When the actual sensor data is numerical data, the data conversion unit 15 performs data conversion for removing noise, for example, so that the numerical data approaches the simulation data indicating the surrounding environment of the moving body.
　教師データに含まれている特徴量は、シミュレーションデータから算出された特徴量である。例えば、教師データに含まれている特徴量は、自動運転シミュレータ（図示省略）で再生された画像（以下「シミュレーション画像」という。）から算出された特徴量である。具体的には、教師データは、例えば、シミュレーション画像から算出された特徴量と、自動運転シミュレータにおいて当該特徴量に基づく移動体の制御量を算出する場合に理想的な制御量が算出されるハイパーパラメータとが、対応付けられたデータである。
　なお、自動運転シミュレータは、一般的なシミュレーション技術を用いたいわゆる自動運転シミュレータである。自動運転シミュレータにおいて、移動体の周辺環境を示すシミュレーション画像が再生される。当該シミュレーション画像は、例えば、ＣＧ（Ｃｏｍｐｕｔｅｒ　Ｇｒａｐｈｉｃｓ）画像である。
　ここで、カメラ画像から特徴量を算出した場合と、シミュレーション画像から特徴量を算出した場合とでは、同じ特徴量として算出されるべき特徴量が、同じ特徴量として算出されない可能性がある。なお、カメラ画像から算出された特徴量は、移動体制御装置１において、制御量算出部１８が移動体の制御量を算出する際に必要となる。その際、制御量算出部１８は、ハイパーパラメータ算出部１７が機械学習モデル１３に基づいて算出したハイパーパラメータを用いる。カメラ画像に基づく特徴量の算出は、特徴量算出部１６が行う。制御量算出部１８、特徴量算出部１６、および、ハイパーパラメータ算出部１７の詳細については、後述する。
　そうすると、カメラ画像から算出された特徴量に基づいて移動体の制御量を算出する際に、シミュレーション画像から算出された特徴量を含む教師データに基づいて生成された機械学習モデル１３に基づくハイパーパラメータを用いると、適切な制御量が算出されない可能性がある。 The features included in the teacher data are the features calculated from the simulation data. For example, the feature amount included in the teacher data is a feature amount calculated from an image (hereinafter referred to as “simulation image”) reproduced by an automatic driving simulator (not shown). Specifically, the teacher data is, for example, a hyper that calculates a feature amount calculated from a simulation image and an ideal control amount when calculating a control amount of a moving body based on the feature amount in an automatic driving simulator. The parameter is the associated data.
The automatic driving simulator is a so-called automatic driving simulator using a general simulation technique. In the automatic driving simulator, a simulation image showing the surrounding environment of the moving body is reproduced. The simulation image is, for example, a CG (Computer Graphics) image.
Here, there is a possibility that the feature amount to be calculated as the same feature amount may not be calculated as the same feature amount in the case where the feature amount is calculated from the camera image and the case where the feature amount is calculated from the simulation image. The feature amount calculated from the camera image is required when the control amount calculation unit 18 calculates the control amount of the moving body in the moving body control device 1. At that time, the control amount calculation unit 18 uses the hyperparameters calculated by the hyperparameter calculation unit 17 based on the machine learning model 13. The feature amount calculation unit 16 calculates the feature amount based on the camera image. Details of the control amount calculation unit 18, the feature amount calculation unit 16, and the hyperparameter calculation unit 17 will be described later.
Then, when calculating the control amount of the moving body based on the feature amount calculated from the camera image, the hyperparameters based on the machine learning model 13 generated based on the teacher data including the feature amount calculated from the simulation image. If is used, an appropriate control amount may not be calculated.
　そこで、データ変換部１５は、カメラ画像に対してデータ変換を行うことで、同じ特徴量として算出されるべき特徴量について、カメラ画像から算出された場合とシミュレーション画像から算出された場合とで生じる差異を、吸収する。これにより、移動体制御装置１は、移動体の制御量を算出する際に用いられるハイパーパラメータが、当該移動体の制御量を算出する際の特徴量とは異なる特徴量に基づくものであることにより、適切に制御量が算出されない可能性を低減することができる。
　なお、データ変換部１５がカメラ画像に対して行うデータ変換と同様のデータ変換は、特徴量が算出される前のシミュレーション画像に対しても行われる必要がある。
　データ変換部１５は、データ変換後のカメラ画像（以下「変換後カメラ画像」という。）を、特徴量算出部１６に出力する。 Therefore, the data conversion unit 15 performs data conversion on the camera image, so that the feature amount to be calculated as the same feature amount is generated depending on whether the feature amount is calculated from the camera image or the simulation image. Absorb the difference. As a result, in the moving body control device 1, the hyperparameters used when calculating the control amount of the moving body are based on the feature amount different from the feature amount when calculating the control amount of the moving body. Therefore, it is possible to reduce the possibility that the control amount is not calculated appropriately.
It should be noted that the same data conversion as the data conversion performed by the data conversion unit 15 on the camera image needs to be performed on the simulation image before the feature amount is calculated.
The data conversion unit 15 outputs the data-converted camera image (hereinafter referred to as “converted camera image”) to the feature amount calculation unit 16.
　特徴量算出部１６は、データ変換部１５が変換した後の変換後カメラ画像から、移動体の走行状況に応じた特徴量を算出する。
　特徴量算出部１６は、例えば、画像処理または機械学習といった既知の技術を用いて、特徴量を算出する。
　特徴量算出部１６は、算出した特徴量を、ハイパーパラメータ算出部１７に出力する。特徴量算出部１６は、算出した特徴量を、例えば、変換後カメラ画像と対応付けて、出力する。 The feature amount calculation unit 16 calculates the feature amount according to the traveling state of the moving body from the converted camera image after the data conversion unit 15 has converted.
The feature amount calculation unit 16 calculates the feature amount by using a known technique such as image processing or machine learning.
The feature amount calculation unit 16 outputs the calculated feature amount to the hyperparameter calculation unit 17. The feature amount calculation unit 16 outputs the calculated feature amount in association with, for example, the converted camera image.
　ハイパーパラメータ算出部１７は、特徴量算出部１６が算出した特徴量を機械学習モデル１３に入力することにより、移動体の制御量を算出する際に用いるハイパーパラメータを算出する。
　ハイパーパラメータ算出部１７は、算出したハイパーパラメータを、制御量算出部１８に出力する。 The hyperparameter calculation unit 17 calculates the hyperparameters used when calculating the control amount of the moving body by inputting the feature amount calculated by the feature amount calculation unit 16 into the machine learning model 13.
The hyperparameter calculation unit 17 outputs the calculated hyperparameters to the control amount calculation unit 18.
　制御量算出部１８は、特徴量算出部１６が算出した特徴量と、ハイパーパラメータ算出部１７が算出したハイパーパラメータとに基づき、移動体の制御量を算出する。制御量算出部１８は、既知のモデル予測制御によって、移動体の制御量を算出する。
　制御量算出部１８は、算出した、移動体の制御量を、外部装置（図示省略）に出力する。外部装置とは、例えば、車両の自動運転を制御する自動運転制御装置（図示省略）である。自動運転制御装置は、制御量算出部１８から出力された制御量に基づいて、車両を自動運転させる。 The control amount calculation unit 18 calculates the control amount of the moving body based on the feature amount calculated by the feature amount calculation unit 16 and the hyperparameters calculated by the hyperparameter calculation unit 17. The control amount calculation unit 18 calculates the control amount of the moving body by the known model prediction control.
The control amount calculation unit 18 outputs the calculated control amount of the moving body to an external device (not shown). The external device is, for example, an automatic driving control device (not shown) that controls the automatic driving of the vehicle. The automatic driving control device automatically drives the vehicle based on the control amount output from the control amount calculation unit 18.
　なお、実施の形態１において、ハイパーパラメータ算出部１７は、算出したハイパーパラメータを学習部１２に出力するようにしてもよい。このとき、ハイパーパラメータ算出部１７は、ハイパーパラメータを算出した際の特徴量を、あわせて出力するようにする。そして、学習部１２は、ハイパーパラメータ算出部１７から出力されたハイパーパラメータと特徴量とに基づき、機械学習モデル１３に学習させるようにしてもよい。 In the first embodiment, the hyperparameter calculation unit 17 may output the calculated hyperparameters to the learning unit 12. At this time, the hyperparameter calculation unit 17 also outputs the feature amount when the hyperparameter is calculated. Then, the learning unit 12 may cause the machine learning model 13 to learn based on the hyperparameters and the feature amount output from the hyperparameter calculation unit 17.
　実施の形態１に係る移動体制御装置１の動作について説明する。
　図２は、実施の形態１に係る移動体制御装置１の動作を説明するためのフローチャートである。
　教師データ取得部１１は、教師データを取得する（ステップＳＴ２０１）。
　教師データ取得部１１は、取得した教師データを、学習部１２に出力する。 The operation of the mobile control device 1 according to the first embodiment will be described.
FIG. 2 is a flowchart for explaining the operation of the mobile control device 1 according to the first embodiment.
The teacher data acquisition unit 11 acquires teacher data (step ST201).
The teacher data acquisition unit 11 outputs the acquired teacher data to the learning unit 12.
　学習部１２は、ステップＳＴ２０１にて教師データ取得部１１が取得した教師データを用いた学習により、機械学習モデル１３を生成する（ステップＳＴ２０２）。 The learning unit 12 generates a machine learning model 13 by learning using the teacher data acquired by the teacher data acquisition unit 11 in step ST201 (step ST202).
　実センサデータ取得部１４は、実センサデータを取得する（ステップＳＴ２０３）。具体的には、実センサデータ取得部１４は、例えば、カメラ画像を取得する。
　実センサデータ取得部１４は、カメラ画像を、データ変換部１５に出力する。 The actual sensor data acquisition unit 14 acquires the actual sensor data (step ST203). Specifically, the actual sensor data acquisition unit 14 acquires, for example, a camera image.
The actual sensor data acquisition unit 14 outputs the camera image to the data conversion unit 15.
　データ変換部１５は、ステップＳＴ２０３にて実センサデータ取得部１４が取得した実センサデータに含まれているデータ要素について、特徴的なカテゴリを形成するデータ要素の集まり毎にデータ変換を行う（ステップＳＴ２０４）。
　データ変換部１５は、変換後カメラ画像を、特徴量算出部１６に出力する。 The data conversion unit 15 converts the data elements included in the actual sensor data acquired by the actual sensor data acquisition unit 14 in step ST203 for each set of data elements forming a characteristic category (step). ST204).
The data conversion unit 15 outputs the converted camera image to the feature amount calculation unit 16.
　特徴量算出部１６は、ステップＳＴ２０４にてデータ変換部１５が変換した後の変換後カメラ画像から、移動体の走行状況に応じた特徴量を算出する（ステップＳＴ２０５）。
　特徴量算出部１６は、算出した特徴量を、ハイパーパラメータ算出部１７に出力する。特徴量算出部１６は、算出した特徴量を、例えば、変換後カメラ画像と対応付けて、出力する。 The feature amount calculation unit 16 calculates the feature amount according to the traveling state of the moving body from the converted camera image after the data conversion unit 15 has converted in step ST204 (step ST205).
The feature amount calculation unit 16 outputs the calculated feature amount to the hyperparameter calculation unit 17. The feature amount calculation unit 16 outputs the calculated feature amount in association with, for example, the converted camera image.
　ハイパーパラメータ算出部１７は、ステップＳＴ２０５にて特徴量算出部１６が算出した特徴量を、ステップＳＴ２０２にて学習部１２が生成した機械学習モデル１３に入力することにより、移動体の制御量を算出する際に用いるハイパーパラメータを算出する（ステップＳＴ２０６）。
　ハイパーパラメータ算出部１７は、算出したハイパーパラメータを、制御量算出部１８に出力する。 The hyperparameter calculation unit 17 calculates the control amount of the moving body by inputting the feature amount calculated by the feature amount calculation unit 16 in step ST205 into the machine learning model 13 generated by the learning unit 12 in step ST202. The hyperparameters used in the above are calculated (step ST206).
The hyperparameter calculation unit 17 outputs the calculated hyperparameters to the control amount calculation unit 18.
　制御量算出部１８は、ステップＳＴ２０５にて特徴量算出部１６が算出した特徴量と、ステップＳＴ２０６にてハイパーパラメータ算出部１７が算出したハイパーパラメータとに基づき、移動体の制御量を算出する（ステップＳＴ２０７）。 The control amount calculation unit 18 calculates the control amount of the moving body based on the feature amount calculated by the feature amount calculation unit 16 in step ST205 and the hyperparameters calculated by the hyperparameter calculation unit 17 in step ST206 (). Step ST207).
　なお、図２を用いて説明した、移動体制御装置１の動作について、ステップＳＴ２０１およびステップＳＴ２０２の動作は、ステップＳＴ２０６の動作が行われるまでに行われていればよい。
　また、ステップＳＴ２０６にて、ハイパーパラメータ算出部１７は、算出したハイパーパラメータを学習部１２に出力するようにしてもよい。このとき、ハイパーパラメータ算出部１７は、ハイパーパラメータを算出した際の特徴量を、あわせて出力するようにする。その後、学習部１２は、ハイパーパラメータ算出部１７から出力されたハイパーパラメータと特徴量とに基づき、機械学習モデル１３に学習させるようにしてもよい。 Regarding the operation of the mobile control device 1 described with reference to FIG. 2, the operations of step ST201 and step ST202 may be performed before the operation of step ST206 is performed.
Further, in step ST206, the hyperparameter calculation unit 17 may output the calculated hyperparameters to the learning unit 12. At this time, the hyperparameter calculation unit 17 also outputs the feature amount when the hyperparameter is calculated. After that, the learning unit 12 may cause the machine learning model 13 to learn based on the hyperparameters and the feature amount output from the hyperparameter calculation unit 17.
　このように、移動体制御装置１は、機械学習を用いて、ハイパーパラメータを算出する。具体的には、移動体制御装置１は、実センサデータから算出した特徴量を、機械学習モデル１３に入力することにより、ハイパーパラメータを算出する。そのため、移動体制御装置１は、移動体制御技術において用いられる、走行状況に応じたハイパーパラメータを、人手を介することなく設定することができる。 In this way, the mobile control device 1 calculates hyperparameters using machine learning. Specifically, the mobile control device 1 calculates hyperparameters by inputting the feature amount calculated from the actual sensor data into the machine learning model 13. Therefore, the mobile body control device 1 can set hyperparameters according to the traveling situation, which are used in the mobile body control technology, without human intervention.
　また、移動体制御装置１は、機械学習モデル１３に基づいて算出したハイパーパラメータを用いて、移動体の制御量を算出する。これにより、移動体制御装置１は、移動体の走行状況に応じた当該移動体の制御量を得ることができる。 Further, the moving body control device 1 calculates the control amount of the moving body using the hyperparameters calculated based on the machine learning model 13. As a result, the moving body control device 1 can obtain a controlled amount of the moving body according to the traveling state of the moving body.
　また、移動体制御装置１は、特徴量にハイパーパラメータが付与された教師データを用いた学習により、機械学習モデル１３を生成する。そのため、移動体制御装置１は、移動体制御技術において用いられる、走行状況に応じたハイパーパラメータを、人手を介することなく設定することができる。 Further, the moving body control device 1 generates a machine learning model 13 by learning using teacher data in which hyperparameters are added to feature quantities. Therefore, the mobile body control device 1 can set hyperparameters according to the traveling situation, which are used in the mobile body control technology, without human intervention.
　以上の実施の形態１では、移動体制御装置１は、データ変換部１５を備えるものとしたが、移動体制御装置１は、データ変換部１５を備えることを必須としない。特徴量算出部１６は、実センサデータ取得部１４が取得した実センサデータから、走行状況に応じた特徴量を算出するようにしてもよい。 In the above-described first embodiment, the mobile control device 1 is provided with the data conversion unit 15, but the mobile control device 1 is not required to include the data conversion unit 15. The feature amount calculation unit 16 may calculate the feature amount according to the traveling situation from the actual sensor data acquired by the actual sensor data acquisition unit 14.
　また、以上の実施の形態１では、学習装置２は、移動体制御装置１に備えられるものとしたが、これは一例に過ぎない。例えば、学習装置２は、移動体制御装置１の外部に備えられるようにしてもよい。
　図３は、実施の形態１において、学習装置２が、移動体制御装置１の外部に備えられ、移動体制御装置１ａと学習装置２とで移動体制御システムを構成するものとした場合の、当該移動体制御システムの構成例を示す図である。
　図３において、図１を用いて説明した構成と同様の構成には、同じ符号を付して重複した説明を省略する。
　移動体制御装置１ａは、実センサデータ取得部１４、データ変換部１５、特徴量算出部１６、ハイパーパラメータ算出部１７、および、制御量算出部１８を備える。なお、図３に示す移動体制御装置１ａは、データ変換部１５を備えることを必須としない。
　図３において、学習装置２は、例えば、サーバに備えられ、移動体制御装置１ａとネットワークを介して接続されるものとしている。学習装置２は、例えば、車載装置であってもよい。 Further, in the above-described first embodiment, the learning device 2 is provided in the mobile control device 1, but this is only an example. For example, the learning device 2 may be provided outside the mobile control device 1.
FIG. 3 shows, in the first embodiment, when the learning device 2 is provided outside the moving body control device 1 and the moving body control device 1a and the learning device 2 constitute a moving body control system. It is a figure which shows the configuration example of the moving body control system.
In FIG. 3, the same reference numerals are given to the same configurations as those described with reference to FIG. 1, and duplicate description will be omitted.
The moving body control device 1a includes an actual sensor data acquisition unit 14, a data conversion unit 15, a feature amount calculation unit 16, a hyperparameter calculation unit 17, and a control amount calculation unit 18. The mobile control device 1a shown in FIG. 3 is not required to include the data conversion unit 15.
In FIG. 3, the learning device 2 is provided in a server, for example, and is connected to the mobile control device 1a via a network. The learning device 2 may be, for example, an in-vehicle device.
　また、以上の実施の形態１において、移動体制御装置１，１ａは、車両に搭載される車載装置とし、実センサデータ取得部１４と、データ変換部１５と、特徴量算出部１６と、ハイパーパラメータ算出部１７と、制御量算出部１８とは、移動体制御装置１，１ａに備えられているものとした。これに限らず、実センサデータ取得部１４と、データ変換部１５と、特徴量算出部１６と、ハイパーパラメータ算出部１７と、制御量算出部１８のうち、一部を車両の車載装置に搭載されるものとし、その他を当該車載装置とネットワークを介して接続されるサーバに備えられるものとして、車載装置とサーバとで移動体制御システムを構成するようにしてもよい。
　例えば、特徴量算出部１６と、ハイパーパラメータ算出部１７と、制御量算出部１８とが、サーバに備えられ、実センサデータ取得部１４と、データ変換部１５とが、車載装置に備えられるものとしてもよい。特徴量算出部１６は、車載装置から、データ変換後の実センサデータを取得する。制御量算出部１８は、算出した制御量を、車載装置に出力する。 Further, in the above-described first embodiment, the moving body control devices 1 and 1a are in-vehicle devices mounted on the vehicle, and the actual sensor data acquisition unit 14, the data conversion unit 15, the feature quantity calculation unit 16, and the hyper. It is assumed that the parameter calculation unit 17 and the control amount calculation unit 18 are provided in the moving body control devices 1, 1a. Not limited to this, a part of the actual sensor data acquisition unit 14, the data conversion unit 15, the feature amount calculation unit 16, the hyperparameter calculation unit 17, and the control amount calculation unit 18 is mounted on the in-vehicle device of the vehicle. The mobile control system may be configured by the in-vehicle device and the server, assuming that the other is provided in the server connected to the in-vehicle device via the network.
For example, the feature amount calculation unit 16, the hyperparameter calculation unit 17, and the control amount calculation unit 18 are provided in the server, and the actual sensor data acquisition unit 14 and the data conversion unit 15 are provided in the in-vehicle device. May be. The feature amount calculation unit 16 acquires the actual sensor data after data conversion from the in-vehicle device. The control amount calculation unit 18 outputs the calculated control amount to the in-vehicle device.
　図４Ａ，図４Ｂは、実施の形態１に係る移動体制御装置１，１ａのハードウェア構成の一例を示す図である。
　実施の形態１において、教師データ取得部１１と、学習部１２と、実センサデータ取得部１４と、データ変換部１５と、特徴量算出部１６と、ハイパーパラメータ算出部１７と、制御量算出部１８の機能は、処理回路４０１により実現される。すなわち、移動体制御装置１は、機械学習を用いて移動体の制御量を算出する制御を行うための処理回路４０１を備える。
　処理回路４０１は、図４Ａに示すように専用のハードウェアであっても、図４Ｂに示すようにメモリ４０６に格納されるプログラムを実行するＣＰＵ（Ｃｅｎｔｒａｌ　Ｐｒｏｃｅｓｓｉｎｇ　Ｕｎｉｔ）４０５であってもよい。 4A and 4B are diagrams showing an example of the hardware configuration of the mobile control devices 1 and 1a according to the first embodiment.
In the first embodiment, the teacher data acquisition unit 11, the learning unit 12, the actual sensor data acquisition unit 14, the data conversion unit 15, the feature amount calculation unit 16, the hyperparameter calculation unit 17, and the control amount calculation unit. The functions of 18 are realized by the processing circuit 401. That is, the mobile body control device 1 includes a processing circuit 401 for performing control for calculating the control amount of the mobile body using machine learning.
The processing circuit 401 may be dedicated hardware as shown in FIG. 4A, or may be a CPU (Central Processing Unit) 405 that executes a program stored in the memory 406 as shown in FIG. 4B.
　処理回路４０１が専用のハードウェアである場合、処理回路４０１は、例えば、単一回路、複合回路、プログラム化したプロセッサ、並列プログラム化したプロセッサ、ＡＳＩＣ（Ａｐｐｌｉｃａｔｉｏｎ　Ｓｐｅｃｉｆｉｃ　Ｉｎｔｅｇｒａｔｅｄ　Ｃｉｒｃｕｉｔ）、ＦＰＧＡ（Ｆｉｅｌｄ－Ｐｒｏｇｒａｍｍａｂｌｅ　Ｇａｔｅ　Ａｒｒａｙ）、またはこれらを組み合わせたものが該当する。 When the processing circuit 401 is dedicated hardware, the processing circuit 401 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), or an FPGA (Field-Programmable). Gate Array) or a combination of these is applicable.
　処理回路４０１がＣＰＵ４０５の場合、教師データ取得部１１と、学習部１２と、実センサデータ取得部１４と、データ変換部１５と、特徴量算出部１６と、ハイパーパラメータ算出部１７と、制御量算出部１８の機能は、ソフトウェア、ファームウェア、または、ソフトウェアとファームウェアとの組み合わせにより実現される。すなわち、教師データ取得部１１と、学習部１２と、実センサデータ取得部１４と、データ変換部１５と、特徴量算出部１６と、ハイパーパラメータ算出部１７と、制御量算出部１８は、ＨＤＤ（Ｈａｒｄ　Ｄｉｓｋ　Ｄｒｉｖｅ）４０２、メモリ４０６等に記憶されたプログラムを実行するＣＰＵ４０５、システムＬＳＩ（Ｌａｒｇｅ－Ｓｃａｌｅ　Ｉｎｔｅｇｒａｔｉｏｎ）等の処理回路により実現される。また、ＨＤＤ４０２、メモリ４０６等に記憶されたプログラムは、教師データ取得部１１と、学習部１２と、実センサデータ取得部１４と、データ変換部１５と、特徴量算出部１６と、ハイパーパラメータ算出部１７と、制御量算出部１８の手順または方法をコンピュータに実行させるものであるとも言える。ここで、メモリ４０６とは、例えば、ＲＡＭ（Ｒａｎｄｏｍ　Ａｃｃｅｓｓ　Ｍｅｍｏｒｙ）、ＲＯＭ（Ｒｅａｄ　Ｏｎｌｙ　Ｍｅｍｏｒｙ）、フラッシュメモリ、ＥＰＲＯＭ（Ｅｒａｓａｂｌｅ　Ｐｒｏｇｒａｍｍａｂｌｅ　Ｒｅａｄ　Ｏｎｌｙ　Ｍｅｍｏｒｙ）、ＥＥＰＲＯＭ（Ｅｌｅｃｔｒｉｃａｌｌｙ　Ｅｒａｓａｂｌｅ　Ｐｒｏｇｒａｍｍａｂｌｅ　Ｒｅａｄ－Ｏｎｌｙ　Ｍｅｍｏｒｙ）等の、不揮発性もしくは揮発性の半導体メモリ、または、磁気ディスク、フレキシブルディスク、光ディスク、コンパクトディスク、ミニディスク、ＤＶＤ（Ｄｉｇｉｔａｌ　Ｖｅｒｓａｔｉｌｅ　Ｄｉｓｃ）等が該当する。 When the processing circuit 401 is the CPU 405, the teacher data acquisition unit 11, the learning unit 12, the actual sensor data acquisition unit 14, the data conversion unit 15, the feature amount calculation unit 16, the hyperparameter calculation unit 17, and the control amount. The function of the calculation unit 18 is realized by software, firmware, or a combination of software and firmware. That is, the teacher data acquisition unit 11, the learning unit 12, the actual sensor data acquisition unit 14, the data conversion unit 15, the feature amount calculation unit 16, the hyperparameter calculation unit 17, and the control amount calculation unit 18 are HDDs. It is realized by a processing circuit such as (Hard Disk Drive) 402, a CPU 405 that executes a program stored in a memory 406, and a system LSI (Data-Scale Integration). Further, the programs stored in the HDD 402, the memory 406, etc. are the teacher data acquisition unit 11, the learning unit 12, the actual sensor data acquisition unit 14, the data conversion unit 15, the feature amount calculation unit 16, and the hyperparameter calculation. It can also be said that the procedure or method of the unit 17 and the control amount calculation unit 18 is executed by the computer. Here, the memory 406 is, for example, a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable Read Only Memory), an EEPROM (Electrically Memory), etc. A sexual or volatile semiconductor memory, a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, a DVD (Digital Versaille Disc), or the like is applicable.
　なお、教師データ取得部１１と、学習部１２と、実センサデータ取得部１４と、データ変換部１５と、特徴量算出部１６と、ハイパーパラメータ算出部１７と、制御量算出部１８の機能について、一部を専用のハードウェアで実現し、一部をソフトウェアまたはファームウェアで実現するようにしてもよい。例えば、教師データ取得部１１と実センサデータ取得部１４については専用のハードウェアとしての処理回路４０１でその機能を実現し、学習部１２と、データ変換部１５と、特徴量算出部１６と、ハイパーパラメータ算出部１７と、制御量算出部１８については処理回路４０１がメモリ４０６に格納されたプログラムを読み出して実行することによってその機能を実現することが可能である。
　また、移動体制御装置１は、自動運転制御装置等の装置と、有線通信または無線通信を行う入力インタフェース装置４０３および出力インタフェース装置４０４を備える。 Regarding the functions of the teacher data acquisition unit 11, the learning unit 12, the actual sensor data acquisition unit 14, the data conversion unit 15, the feature amount calculation unit 16, the hyperparameter calculation unit 17, and the control amount calculation unit 18. , Some may be realized by dedicated hardware, and some may be realized by software or firmware. For example, the teacher data acquisition unit 11 and the actual sensor data acquisition unit 14 are realized by the processing circuit 401 as dedicated hardware, and the learning unit 12, the data conversion unit 15, the feature amount calculation unit 16, and the feature amount calculation unit 16 The functions of the hyperparameter calculation unit 17 and the control amount calculation unit 18 can be realized by the processing circuit 401 reading and executing the program stored in the memory 406.
Further, the mobile body control device 1 includes a device such as an automatic operation control device, and an input interface device 403 and an output interface device 404 that perform wired communication or wireless communication.
　以上のように、実施の形態１によれば、移動体制御装置１，１ａは、移動体の周辺環境を示すセンサデータ（実センサデータ）から移動体の制御量を算出する移動体制御装置１，１ａであって、センサデータ（実センサデータ）を取得する実センサデータ取得部１４と、実センサデータ取得部１４が取得したセンサデータ（実センサデータ）から特徴量を算出する特徴量算出部１６と、特徴量算出部１６が算出した特徴量を機械学習モデル１３に入力することにより、移動体の制御量を算出する際に用いるハイパーパラメータを算出するハイパーパラメータ算出部１７とを備えるように構成した。そのため、移動体制御装置１，１ａは、移動体制御技術において用いられる、走行状況に応じたハイパーパラメータを、人手を介することなく設定することができる。 As described above, according to the first embodiment, the moving body control devices 1 and 1a calculate the control amount of the moving body from the sensor data (actual sensor data) indicating the surrounding environment of the moving body. , 1a, the feature amount calculation unit that calculates the feature amount from the actual sensor data acquisition unit 14 that acquires the sensor data (actual sensor data) and the sensor data (actual sensor data) acquired by the actual sensor data acquisition unit 14. 16 and a hyper parameter calculation unit 17 for calculating a hyper parameter used when calculating a control amount of a moving body by inputting the feature amount calculated by the feature amount calculation unit 16 into the machine learning model 13 are provided. Configured. Therefore, the mobile body control devices 1, 1a can set hyperparameters according to the traveling situation, which are used in the mobile body control technology, without human intervention.
　また、実施の形態１によれば、学習装置２は、移動体の周辺環境を示すセンサデータから算出された特徴量に、移動体の制御量を算出する際に用いられるハイパーパラメータ、が付与された教師データを取得する教師データ取得部１１と、教師データ取得部１１が取得した教師データを用いた学習により、特徴量を入力としハイパーパラメータを出力する機械学習モデル１３を生成する学習部１２とを備えるように構成した。そのため、移動体制御装置１，１ａは、移動体制御技術において用いられる、走行状況に応じたハイパーパラメータを、人手を介することなく設定することができる。 Further, according to the first embodiment, in the learning device 2, hyperparameters used when calculating the control amount of the moving body are added to the feature amount calculated from the sensor data indicating the surrounding environment of the moving body. A teacher data acquisition unit 11 that acquires the teacher data, and a learning unit 12 that generates a machine learning model 13 that inputs a feature amount and outputs hyperparameters by learning using the teacher data acquired by the teacher data acquisition unit 11. It was configured to be equipped with. Therefore, the mobile body control devices 1, 1a can set hyperparameters according to the traveling situation, which are used in the mobile body control technology, without human intervention.
　以上の実施の形態１では、移動体は車両としたが、これは一例に過ぎない。実施の形態１に係る移動体制御装置１，１ａは、自動制御が可能な種々の移動体において、当該移動体の制御量を算出するための装置として用いることができる。 In the above-described first embodiment, the moving body is a vehicle, but this is only an example. The mobile body control devices 1 and 1a according to the first embodiment can be used as a device for calculating the control amount of the mobile body in various mobile bodies capable of automatic control.
　なお、本開示の範囲内において、実施の形態の任意の構成要素の変形、もしくは実施の形態の任意の構成要素の省略が可能である。 Within the scope of the present disclosure, it is possible to modify any component of the embodiment or omit any component of the embodiment.
　本開示に係る移動体制御装置は、移動体制御技術において用いられる、走行状況に応じたハイパーパラメータを、人手を介することなく設定可能とするように構成したため、移動体の制御量を算出する移動体制御装置に適用することができる。 Since the mobile body control device according to the present disclosure is configured so that hyperparameters according to the traveling situation used in the mobile body control technology can be set without human intervention, the movement for calculating the control amount of the mobile body is calculated. It can be applied to body control devices.
　１，１ａ　移動体制御装置、２　学習装置、１１　教師データ取得部、１２　学習部、１３　機械学習モデル、１４　実センサデータ取得部、１５　データ変換部、１６　特徴量算出部、１７　ハイパーパラメータ算出部、１８　制御量算出部、４０１　処理回路、４０２　ＨＤＤ、４０３　入力インタフェース装置、４０４　出力インタフェース装置、４０５　ＣＰＵ、４０６　メモリ。 1,1a Mobile control device, 2 Learning device, 11 Teacher data acquisition unit, 12 Learning unit, 13 Machine learning model, 14 Actual sensor data acquisition unit, 15 Data conversion unit, 16 Feature amount calculation unit, 17 Hyperparameter calculation unit , 18 control amount calculation unit, 401 processing circuit, 402 HDD, 403 input interface device, 404 output interface device, 405 CPU, 406 memory.

Claims (6)

  1. 　移動体の周辺環境を示すセンサデータから前記移動体の制御量を算出する移動体制御装置であって、
    　前記センサデータを取得する実センサデータ取得部と、
    　前記実センサデータ取得部が取得した前記センサデータから特徴量を算出する特徴量算出部と、
    　前記特徴量算出部が算出した特徴量を機械学習モデルに入力することにより、前記移動体の制御量を算出する際に用いるハイパーパラメータを算出するハイパーパラメータ算出部
    　とを備えた移動体制御装置。     A mobile control device that calculates the control amount of the mobile from sensor data indicating the surrounding environment of the mobile.
    The actual sensor data acquisition unit that acquires the sensor data, and
    A feature amount calculation unit that calculates a feature amount from the sensor data acquired by the actual sensor data acquisition unit, and a feature amount calculation unit.
    A moving body control device including a hyperparameter calculation unit that calculates hyperparameters used when calculating the control amount of the moving body by inputting the feature amount calculated by the feature amount calculation unit into a machine learning model.
  2. 　前記特徴量算出部が算出した特徴量と、前記ハイパーパラメータ算出部が算出したハイパーパラメータとに基づき、前記移動体の制御量を算出する制御量算出部
    　を備えた請求項１記載の移動体制御装置。     The mobile control according to claim 1, further comprising a control amount calculation unit that calculates a control amount of the moving body based on the feature amount calculated by the feature amount calculation unit and the hyperparameters calculated by the hyperparameter calculation unit. Device.
  3. 　前記実センサデータ取得部が取得した前記センサデータに含まれているデータ要素に対してデータ変換を行うデータ変換部を備え、
    　前記特徴量算出部は、
    　前記データ変換部が変換した後の前記センサデータから特徴量を算出する
    　ことを特徴とする請求項１記載の移動体制御装置。     A data conversion unit that performs data conversion on the data elements included in the sensor data acquired by the actual sensor data acquisition unit is provided.
    The feature amount calculation unit
    The mobile control device according to claim 1, wherein a feature amount is calculated from the sensor data after the data conversion unit has converted the data.
  4. 　前記移動体の周辺環境を示すセンサデータから算出された特徴量に前記ハイパーパラメータが付与された教師データを取得する教師データ取得部と、
    　前記教師データ取得部が取得した教師データを用いた学習により、前記特徴量を入力とし前記ハイパーパラメータを出力する前記機械学習モデルを生成する学習部を備え、
    　前記ハイパーパラメータ算出部は、前記特徴量算出部が算出した特徴量を、前記学習部が生成した前記機械学習モデルに入力することにより、前記ハイパーパラメータを算出する
    　ことを特徴とする請求項１記載の移動体制御装置。     A teacher data acquisition unit that acquires teacher data in which the hyperparameters are added to a feature amount calculated from sensor data indicating the surrounding environment of the moving object, and a teacher data acquisition unit.
    A learning unit for generating the machine learning model that inputs the feature amount and outputs the hyperparameters by learning using the teacher data acquired by the teacher data acquisition unit is provided.
    The hyperparameter calculation unit is characterized in that it calculates the hyperparameters by inputting the feature amount calculated by the feature amount calculation unit into the machine learning model generated by the learning unit. Mobile control device.
  5. 　移動体の周辺環境を示すセンサデータから前記移動体の制御量を算出する移動体制御方法であって、
    　実センサデータ取得部が、前記センサデータを取得するステップと、
    　特徴量算出部が、前記実センサデータ取得部が取得した前記センサデータから特徴量を算出するステップと、
    　ハイパーパラメータ算出部が、前記特徴量算出部が算出した特徴量を機械学習モデルに入力することにより、前記移動体の制御量を算出する際に用いるハイパーパラメータを算出するステップ
    　とを備えた移動体制御方法。     It is a moving body control method that calculates a control amount of the moving body from sensor data indicating the surrounding environment of the moving body.
    The step in which the actual sensor data acquisition unit acquires the sensor data,
    A step in which the feature amount calculation unit calculates the feature amount from the sensor data acquired by the actual sensor data acquisition unit, and
    The hyperparameter calculation unit inputs the feature amount calculated by the feature amount calculation unit into the machine learning model, so that the hyperparameter calculation unit includes a step of calculating the hyperparameters used when calculating the control amount of the moving body. Control method.
  6. 　移動体の周辺環境を示すセンサデータから算出された特徴量に、前記移動体の制御量を算出する際に用いられるハイパーパラメータ、が付与された教師データを取得する教師データ取得部と、
    　前記教師データ取得部が取得した教師データを用いた学習により、前記特徴量を入力とし前記ハイパーパラメータを出力する機械学習モデルを生成する学習部
    　を備えた学習装置。     A teacher data acquisition unit that acquires teacher data in which hyperparameters used when calculating the control amount of the moving body are added to the feature amount calculated from the sensor data indicating the surrounding environment of the moving body.
    A learning device provided with a learning unit that generates a machine learning model that inputs the feature amount and outputs the hyperparameters by learning using the teacher data acquired by the teacher data acquisition unit.
PCT/JP2020/016019 2020-04-09 2020-04-09 Moving body control device, moving body control method, and learning device WO2021205616A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/JP2020/016019 WO2021205616A1 (en) 2020-04-09 2020-04-09 Moving body control device, moving body control method, and learning device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2020/016019 WO2021205616A1 (en) 2020-04-09 2020-04-09 Moving body control device, moving body control method, and learning device

Publications (1)

Publication Number Publication Date
WO2021205616A1 true WO2021205616A1 (en) 2021-10-14

Family

ID=78023265

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2020/016019 WO2021205616A1 (en) 2020-04-09 2020-04-09 Moving body control device, moving body control method, and learning device

Country Status (1)

Country Link
WO (1) WO2021205616A1 (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2014106685A (en) * 2012-11-27 2014-06-09 Osaka Univ Vehicle periphery monitoring device
JP2019008460A (en) * 2017-06-22 2019-01-17 株式会社東芝 Object detection device and object detection method and program
JP2019028939A (en) * 2017-08-03 2019-02-21 日本電信電話株式会社 Estimation method and estimation device

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2014106685A (en) * 2012-11-27 2014-06-09 Osaka Univ Vehicle periphery monitoring device
JP2019008460A (en) * 2017-06-22 2019-01-17 株式会社東芝 Object detection device and object detection method and program
JP2019028939A (en) * 2017-08-03 2019-02-21 日本電信電話株式会社 Estimation method and estimation device

Similar Documents

Publication Publication Date Title
JP7262503B2 (en) Method and apparatus, electronic device, computer readable storage medium and computer program for detecting small targets
US20210142146A1 (en) Intelligent image sensor stack
US11769059B2 (en) Systems and methods for distributed training of deep learning models
JP6605642B2 (en) Vehicle and system for managing and controlling vehicle
US20210129852A1 (en) Configuration of a Vehicle Based on Collected User Data
JP2020083308A (en) Real time prediction of object behavior
CN113033029A (en) Automatic driving simulation method and device, electronic equipment and storage medium
US11459028B2 (en) Adjusting vehicle sensitivity
WO2018168539A1 (en) Learning method and program
CN112784885B (en) Automatic driving method, device, equipment, medium and vehicle based on artificial intelligence
AU2018202380A1 (en) System and method for predictive condition modeling of asset fleets under partial information
US20220032932A1 (en) Image sensor for processing sensor data to reduce data traffic to host system
WO2021205616A1 (en) Moving body control device, moving body control method, and learning device
CN110574041A (en) Collaborative activation for deep learning domains
US20230196619A1 (en) Validation of virtual camera models
WO2021205615A1 (en) Teacher data generation apparatus and teacher data generation method
US20210374598A1 (en) A System and Method for Using Knowledge Gathered by a Vehicle
CN108960160B (en) Method and device for predicting structured state quantity based on unstructured prediction model
CN110719487B (en) Video prediction method and device, electronic equipment and vehicle
WO2021171398A1 (en) Inference device, driving assistance device, inference method, and server
JP6947091B2 (en) Driving support device, driving support method, driving support program, motion control device, motion control method, and motion control program
KR20220143326A (en) APPARATUS AND SYSTE FOR Real-time Pixel-wise Semantic Segmentation
CN114581876A (en) Method for constructing lane detection model under complex scene and method for detecting lane line
US11636339B2 (en) In-memory content classification and control
KR102631367B1 (en) Enforcement apparatus automatic registration of enforcement areas, initialization of zoom magnification data and real-time update for crackdown on illegal parking and thereof operation method

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20929995

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20929995

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: JP