WO2022137506A1 - 運転評価システム、学習装置、評価結果出力装置、方法およびプログラム - Google Patents
運転評価システム、学習装置、評価結果出力装置、方法およびプログラム Download PDFInfo
- Publication number
- WO2022137506A1 WO2022137506A1 PCT/JP2020/048743 JP2020048743W WO2022137506A1 WO 2022137506 A1 WO2022137506 A1 WO 2022137506A1 JP 2020048743 W JP2020048743 W JP 2020048743W WO 2022137506 A1 WO2022137506 A1 WO 2022137506A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- driving
- cost function
- expert
- learning
- region
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims description 43
- 230000006870 function Effects 0.000 claims abstract description 168
- 238000012549 training Methods 0.000 claims abstract description 50
- 230000002787 reinforcement Effects 0.000 claims abstract description 48
- 238000011156 evaluation Methods 0.000 claims description 168
- 238000013075 data extraction Methods 0.000 claims description 23
- 238000012545 processing Methods 0.000 claims description 12
- 239000000284 extract Substances 0.000 claims description 8
- 238000012854 evaluation process Methods 0.000 claims description 4
- 238000004891 communication Methods 0.000 description 16
- 238000010586 diagram Methods 0.000 description 13
- 230000010365 information processing Effects 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 230000036760 body temperature Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
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
- B60W40/00—Estimation or calculation of non-directly measurable driving parameters for road vehicle drive control systems not related to the control of a particular sub unit, e.g. by using mathematical models
- B60W40/08—Estimation or calculation of non-directly measurable driving parameters for road vehicle drive control systems not related to the control of a particular sub unit, e.g. by using mathematical models related to drivers or passengers
- B60W40/09—Driving style or behaviour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
- G09B—EDUCATIONAL OR DEMONSTRATION APPLIANCES; APPLIANCES FOR TEACHING, OR COMMUNICATING WITH, THE BLIND, DEAF OR MUTE; MODELS; PLANETARIA; GLOBES; MAPS; DIAGRAMS
- G09B9/00—Simulators for teaching or training purposes
- G09B9/02—Simulators for teaching or training purposes for teaching control of vehicles or other craft
- G09B9/04—Simulators for teaching or training purposes for teaching control of vehicles or other craft for teaching control of land vehicles
- G09B9/042—Simulators for teaching or training purposes for teaching control of vehicles or other craft for teaching control of land vehicles providing simulation in a real vehicle
-
- G—PHYSICS
- G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
- G09B—EDUCATIONAL OR DEMONSTRATION APPLIANCES; APPLIANCES FOR TEACHING, OR COMMUNICATING WITH, THE BLIND, DEAF OR MUTE; MODELS; PLANETARIA; GLOBES; MAPS; DIAGRAMS
- G09B9/00—Simulators for teaching or training purposes
- G09B9/02—Simulators for teaching or training purposes for teaching control of vehicles or other craft
- G09B9/04—Simulators for teaching or training purposes for teaching control of vehicles or other craft for teaching control of land vehicles
- G09B9/052—Simulators for teaching or training purposes for teaching control of vehicles or other craft for teaching control of land vehicles characterised by provision for recording or measuring trainee's performance
-
- 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
- B60W2540/00—Input parameters relating to occupants
- B60W2540/30—Driving style
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
Definitions
- the present invention outputs a driving evaluation system for evaluating a subject's driving, a driving evaluation method, a learning device for learning a cost function used for driving evaluation, a learning method and a learning program, and a driving evaluation result.
- the present invention relates to an evaluation result output device, an evaluation result output method, and an evaluation result output program.
- a sudden accelerator, a sudden brake, and a sudden steering wheel can be used as an index for judging whether or not the driving is ecological.
- Patent Document 1 describes a driving skill evaluation system that automatically evaluates a subject's driving skill.
- the system described in Patent Document 1 records sample driving data, and learns a driving pattern of exemplary driving based on the recorded driving data.
- Patent Document 1 automatically evaluates the driving skill of a subject in a driving school or the like by learning by adding the traveling position of the vehicle to the evaluation.
- the model driving pattern of the driver learned by the system described in Patent Document 1 can evaluate the driving according to the actual driving environment.
- the present invention relates to a driving evaluation system and a driving evaluation method capable of evaluating the driving of a subject in consideration of the driving area, an evaluation result output device for outputting a driving evaluation result, an evaluation result output method and an evaluation. It is an object of the present invention to provide a result output program, and a learning device, a learning method, and a learning program for learning a cost function used for evaluating the operation thereof.
- the driving evaluation system is a function input means that accepts input of a cost function represented by a linear sum of terms in which the degree of importance is weighted to each feature amount indicating the driver's driving, and is collected for each region.
- the driving data input means that accepts the input of the information indicating the environment at the time of driving and the user driving data including the position information from which the information is acquired, and the area where the user drives are specified from the position information, and the information thereof is specified.
- Select the cost function corresponding to the region apply the information indicating the environment when the subject is driving to the selected cost function, estimate the driving of the expert in the same environment, and estimate the driving of the expert. It is characterized by being equipped with an evaluation means for outputting an evaluation result comparing driving with the driving of a target person.
- the learning device is a function input means that accepts input of a cost function represented by a linear sum of terms in which the degree of importance is weighted to each feature amount indicating the driver's operation, and a skill collected for each region. It is characterized by having a learning means for learning a cost function for each region by reverse reinforcement learning using expert driving data including information indicating the content of a person's driving as training data.
- the evaluation result output device accepts input of user driving data including information indicating the driving of the target person who evaluates driving, information indicating the environment at the time of driving, and position information from which the information is acquired. Shows the driver's driving learned for each region by reverse reinforcement learning using the driving data input means and the expert driving data including the information showing the contents of the expert's driving collected for each region as training data. From the cost function represented by the linear sum of the terms weighted to each feature amount, the area where the user drives is specified from the position information, and the cost function corresponding to that area is selected and selected. By applying the information indicating the environment when the subject is driving to the calculated cost function, the driving of the expert in the same environment is estimated, and the estimated driving of the expert is compared with the driving of the subject. It is characterized by having an evaluation means for outputting the result.
- the driving evaluation method accepts the input of a cost function represented by the linear sum of the terms in which the degree of importance is weighted to each feature amount indicating the driving of the driver, and the driving of a skilled person collected for each region.
- a cost function represented by the linear sum of the terms in which the degree of importance is weighted to each feature amount indicating the driving of the driver, and the driving of a skilled person collected for each region.
- the cost function for each region is learned, and the information indicating the driving of the target person who evaluates the driving and the environment during driving are shown.
- It accepts the input of information and user driving data including the location information from which the information was acquired, identifies the area where the user drives from the location information, and selects and selects the cost function corresponding to that area.
- the driving of the expert in the same environment is estimated, and the evaluation result comparing the estimated driving of the expert with the driving of the subject. Is characterized by outputting.
- the learning method accepts the input of the cost function represented by the linear sum of the terms in which the degree of importance is weighted to each feature amount indicating the driver's driving, and the expert's driving collected for each region. It is characterized by learning the cost function for each region by reverse reinforcement learning using expert driving data including information showing the contents as training data.
- the evaluation result output method accepts input of user driving data including information indicating the driving of a subject who evaluates driving, information indicating the environment at the time of driving, and position information from which the information was acquired. , Emphasis is placed on each feature amount that indicates the driver's driving, which was learned for each region by reverse reinforcement learning using the expert driving data including information showing the contents of the driver's driving collected for each region as training data. From the cost function represented by the linear sum of the terms weighted by each degree, the area where the user drives is specified from the position information, the cost function corresponding to that area is selected, and the selected cost function is used. Applying information indicating the environment when the subject is driving, the driving of the expert in the same environment is estimated, and the evaluation result comparing the estimated driving of the expert with the driving of the subject is output. It is characterized by.
- the learning program according to the present invention has a function input process that accepts an input of a cost function represented by a linear sum of terms in which the degree of importance is weighted to each feature amount indicating the driver's driving, and a function input process for each region. It is characterized in that a learning process for learning a cost function for each region is executed by reverse reinforcement learning using the collected expert driving data including information indicating the contents of the expert's driving as training data.
- the evaluation result output program is a computer for user driving data including information indicating the driving of a subject who evaluates driving, information indicating the environment during driving, and location information from which the information is acquired.
- a driver learned for each region by driving data input processing that accepts input and reverse reinforcement learning using expert driving data including information showing the contents of the expert's driving collected for each region as training data. From the cost function represented by the linear sum of the terms in which the degree of importance is weighted to each feature amount indicating the operation of, the area where the user drives is specified from the position information and the cost function corresponding to that area is obtained.
- the expert's driving in the same environment is estimated, and the estimated expert's driving and the subject's driving are combined. It is characterized by executing an evaluation process that outputs an evaluation result comparing the above.
- FIG. 1 is a block diagram showing a configuration example of an embodiment of the operation evaluation system according to the present invention.
- the operation evaluation system 1 of the present embodiment includes an evaluation result output device 300 and a learning device 400.
- the evaluation result output device 300 is connected to the vehicle 100 equipped with the on-board unit 101 and the smartphone 200, and is also connected to the learning device 400.
- the vehicle 100 (more specifically, the in-vehicle device 101) and the smartphone 200 are assumed to move in the same manner, and various information is input, instructions are given to the in-vehicle device 101, and the position information of the vehicle 100 is obtained.
- the smartphone 200 is used for the acquisition of.
- the handy smartphone 200 it becomes possible to simplify the input of movement information to the vehicle 100 (more specifically, the on-board unit 101) and the processing at the time of function expansion.
- the vehicle 100 itself may be connected, and the on-board unit 101 may be configured to integrally realize the functions of the smartphone 200.
- the evaluation result output device 300 is provided separately from the vehicle 100 (more specifically, the on-board unit 101) is illustrated.
- the evaluation result output device 300 may be configured to be integrated with the vehicle-mounted device 101.
- the learning device 400 is connected to a storage device 500 that stores various information used for learning and a display device 600 that displays the learning result.
- the storage device 500 is realized by, for example, an external storage server.
- the display device 600 is realized by, for example, a display device.
- the learning device 400 may be configured to include one or both of the storage device 500 and the display device 600.
- the storage device 500 stores data representing the driving performance of the vehicle (hereinafter referred to as driving data) as various information used for learning.
- the driving data includes information indicating the driver's driving (for example, operation information for operating the vehicle), information indicating the environment at the time of the driving, and location information from which these information are acquired (that is, driving). Location information indicating the location) and. It should be noted that these pieces of information can be said to be feature quantities indicating features during driving.
- the information indicating the environment may include the driver's own attributes as well as the situation outside the vehicle.
- the storage device 500 may store only the driving data of the driver defined as a skilled person, or may store the driving data including the general driver. The definition of an expert will be described later.
- FIG. 2 is an explanatory diagram showing an example of operation data.
- the driving data illustrated in FIG. 2 includes items that are roughly classified into four categories (information about the vehicle (information inside the vehicle), information outside the vehicle, time information, and weather information).
- An example of information indicating the driver's driving is the operation information (accelerator opening, brake operation, steering wheel operation, etc.) illustrated in FIG. 2, the vehicle speed in the engine information, and the position information indicating the driving location is GPS.
- the location information acquired in is the location information acquired in.
- operation data illustrated in FIG. 2 is an example, and the operation data may include all the items illustrated in FIG. 2 or may include some items. Further, the operation data may include items other than those illustrated in FIG.
- the vehicle 100 exemplified in this embodiment includes the on-board unit 101.
- Various sensors such as an external camera 140, a vehicle information sensor 150, a biological sensor 160, and an in-vehicle camera 170 are connected to the on-board unit 101.
- the vehicle-mounted device 101 has a control unit 110 including a CPU (Central Processing Unit) 111 and a memory 112, a communication unit 120, and a storage unit 130.
- the communication unit 120 performs various communications with the evaluation result output device 300.
- the storage unit 130 stores various information used for processing by the control unit 110.
- the outside camera 140 is a camera that photographs the outside of the vehicle 100.
- the out-of-vehicle camera 140 may capture, for example, other vehicles, pedestrians, motorcycles, bicycles, and the like existing outside the vehicle. Further, the outside camera 140 may take a picture together with the state of the road on which the vehicle 100 is traveling (road shape, congestion information, signal information, etc.).
- the control unit 110 may perform object recognition processing such as a vehicle or a pedestrian from the captured image, for example.
- the vehicle information sensor 150 detects various states of the vehicle 100.
- the vehicle information sensor 150 may detect information such as the engine speed and the accelerator opening degree based on, for example, CAN (Controller Area Network).
- the biosensor 160 detects various biometric information of the driver.
- the biological sensor 160 may be, for example, a sensor capable of detecting the driver's pulse, heart rate, and body temperature. Further, the biological sensor 160 may detect not only the biological information of the driver but also the biological information of the passenger.
- the in-car camera 170 is a camera that photographs the inside of the car.
- the in-vehicle camera 170 may, for example, capture the presence or absence of a passenger.
- the sensor shown in FIG. 1 is an example, and some or all of these sensors may be connected to the on-board unit 101, or other sensors may be connected to the vehicle-mounted device 101.
- the information detected by these sensors is stored in the storage unit 130 and transmitted to the evaluation result output device 300 via the communication unit 120.
- the smartphone 200 includes a control unit 210 including a CPU 211 and a memory 212, a communication unit 220, a storage unit 230, an input unit 240, and a mobile information database (hereinafter, DB) 250.
- a control unit 210 including a CPU 211 and a memory 212, a communication unit 220, a storage unit 230, an input unit 240, and a mobile information database (hereinafter, DB) 250.
- DB mobile information database
- the control unit 210 controls various processes performed by the smartphone 200.
- the communication unit 220 performs various communications with the evaluation result output device 300.
- the storage unit 230 stores various information used for processing by the control unit 210.
- the input unit 240 receives various inputs to the smartphone 200 as well as control inputs to the vehicle-mounted device 101 from the user.
- the movement information DB 250 stores the movement information of the vehicle 100. Specifically, the movement information DB 250 stores the position information of the vehicle 100 acquired from the GPS (Global Positioning System) by the control unit 210 in chronological order. This makes it possible to associate the position information of the vehicle 100 (that is, the position information indicating the place where the vehicle was driven) with the driving data.
- GPS Global Positioning System
- the learning device 400 includes a control unit 410 including a CPU 411 and a memory 412, a communication unit 420, an input unit 430, a storage unit 440, and a learning unit 450.
- the control unit 410 controls the processing of the learning unit 450, which will be described later.
- the communication unit 420 performs various communications with the evaluation result output device 300.
- the storage unit 440 stores various information used for processing by the control unit 410 and the learning unit 450. Further, the storage unit 440 may store the operation data for which the input unit 430, which will be described later, has received the input.
- the storage unit 440 is realized by, for example, a magnetic disk or the like.
- the input unit 430 receives input of operation data from the storage device 500.
- the input unit 430 may acquire operation data from the storage device 500 in response to an explicit instruction to the learning device 400, or may acquire operation data in response to a notification from the storage device 500. Further, the input unit 430 may store the acquired operation data in the storage unit 440. Since the received driving data is data used for learning by the reverse reinforcement learning unit 453, which will be described later, the driving data may be referred to as expert driving data or training data.
- FIG. 3 is a block diagram showing a configuration example of the learning unit 450.
- the learning unit 450 of the present embodiment includes a cost function input unit 451, a data extraction unit 452, a reverse reinforcement learning unit 453, and a learning result output unit 454.
- the cost function input unit 451 accepts the input of the cost function used for learning by the inverse reinforcement learning unit 453, which will be described later. Specifically, the cost function input unit 451 accepts an input of a cost function represented by a linear sum of terms in which the degree of emphasis is weighted on each feature amount indicating the driver's operation as illustrated in FIG. .. It can be said that this degree of importance represents the intention at the time of evaluation. Therefore, the value calculated by the cost function can be said to be an evaluation index used for the evaluation of operation.
- the cost function input unit 451 inputs not only the feature amount indicating the driver's driving but also the input of the cost function including a term in which the degree of importance is weighted to each feature amount indicating the driving environment as a linear sum. You may accept it.
- the feature quantities indicating the driver's driving are, for example, the speed, the distance to the vehicle in front, and the amount of depression of the accelerator pedal. Further, the feature amount indicating the environment at the time of driving is, for example, road shape, congestion information, and the like.
- the cost function input unit 451 may accept the input of the constraint condition to be satisfied together with the cost function.
- the cost function and constraints are predetermined by the analyst and the like. That is, the candidate of the feature amount to be considered when evaluating the operation is selected in advance by an analyst or the like and defined as a cost function.
- Equation 1 when evaluating driving, when speed, distance to a vehicle in front, and accelerator pedal depression amount are candidates for feature quantities, the cost function is represented by Equation 1 illustrated below.
- the data extraction unit 452 extracts training data for each region from the operation data received by the input unit 430. Specifically, the data extraction unit 452 extracts training data for each region based on the position information from which the driving data (training data) has been acquired. The data extraction unit 452 may, for example, determine the area from the latitude and longitude acquired from GPS and extract training data.
- the data extraction unit 452 performs a process of converting an item included in the operation data into a feature amount (calculation, conversion to a binary value, etc.), a data integration process, and data cleansing in order to match the feature amount included in the cost function. And so on.
- the data extraction unit 452 extracts the operation data of a skilled person from the operation data as a candidate based on a predetermined standard.
- the method of extracting the operation data of a skilled person is arbitrary and is predetermined by an analyst or the like.
- the data extraction unit 452 may use a driver having a long total driving time and a driver having a small accident history or a violation history as a skilled person, and extract the driving data of the driver as the driving data of the skilled person.
- the data extraction unit 452 may preferentially select the driving data of the driver related to the corresponding area from the driving data of the skilled worker as the more appropriate driving data of the skilled worker. For example, a driver residing in the area is considered to be more aware of the situation in that area.
- the data extraction unit 452 may determine the driver's relevant area from the license plate, for example.
- the reverse reinforcement learning unit 453 learns the cost function for each region by reverse reinforcement learning using the training data for each region extracted by the data extraction unit 452. Specifically, the reverse reinforcement learning unit 453 learns the cost function for each region by reverse reinforcement learning using the driving data of the expert collected for each region as training data. That is, this training data includes information representing the content of the driving of a skilled person. In addition, the training data may include information indicating the environment during the driving.
- the method by which the reverse reinforcement learning unit 453 performs reverse reinforcement learning is arbitrary.
- the inverse reinforcement learning unit 453 executes, for example, a mathematical optimization process that generates driving data of a skilled worker based on an input cost function and constraints, and a difference between the generated driving data of the skilled worker and training data.
- the cost function may be learned by repeating the estimation process of the cost function that updates the parameter (degree of importance) of the cost function so as to reduce the value.
- the learning result output unit 454 outputs the learned cost function. Specifically, the learning result output unit 454 outputs the feature amount included in the cost function learned for each region in association with the weight of the feature amount.
- the learning result output unit 454 may display the contents of the cost function on the display device 600, or may store the contents in the storage unit 440. By displaying the contents of the cost function on the display device 600, it becomes possible to visually recognize the items to be emphasized for each region.
- Cost function 100 x speed + 50 x distance to the vehicle ahead + 10 x accelerator pedal depression amount (Equation 2)
- the learning result output unit 454 may output the evaluation weight of [speed, distance to the vehicle ahead] as [100, 50, 10].
- the learning result output unit 454 may output a predetermined number in order from the feature amount having the highest degree of importance as the weight of the evaluation. By doing so, it becomes possible to grasp the feature amount that more reflects the intention of the expert.
- the learning unit 450 (more specifically, the cost function input unit 451 and the data extraction unit 452, the inverse reinforcement learning unit 453, and the learning result output unit 454) is driven by a computer processor that operates according to the program (learning program). It will be realized.
- the program is stored in the storage unit 440 of the learning device 400, the processor reads the program, and according to the program, the learning unit 450 (more specifically, the cost function input unit 451 and the data extraction unit 452). It may operate as the inverse reinforcement learning unit 453 and the learning result output unit 454). Further, each function of the learning unit 450 (more specifically, the cost function input unit 451 and the data extraction unit 452, the inverse reinforcement learning unit 453, and the learning result output unit 454) is in the SaaS (Software as a Service) format. May be provided at.
- SaaS Software as a Service
- the cost function input unit 451, the data extraction unit 452, the reverse reinforcement learning unit 453, and the learning result output unit 454 may each be realized by dedicated hardware. Further, a part or all of each component of each device may be realized by a general-purpose or dedicated circuit (circuitry), a processor, or a combination thereof. These may be composed of a single chip or may be composed of a plurality of chips connected via a bus. A part or all of each component of each device may be realized by the combination of the circuit or the like and the program described above.
- each component of the learning unit 450 (more specifically, the cost function input unit 451, the data extraction unit 452, the inverse reinforcement learning unit 453, and the learning result output unit 454) is plural.
- a plurality of information processing devices, circuits, or the like may be centrally arranged or distributed.
- the information processing device, the circuit, and the like may be realized as a form in which each is connected via a communication network, such as a client-server system and a cloud computing system.
- the learning unit 450 may be included in the control unit 410 itself.
- the control unit 410 may read the program (learning program) stored in the memory 412 by the CPU 411 and operate as the learning unit 450 according to the program.
- the evaluation result output device 300 includes a control unit 310 including a CPU 311 and a memory 312, a communication unit 320, an input unit 330, an operation record DB 340, a user DB 350, a display unit 360, and an evaluation unit 370. ..
- the control unit 310 controls the processing of the evaluation unit 370, which will be described later.
- the communication unit 320 performs various communications with the vehicle 100 (more specifically, the on-board unit 101), the smartphone 200, the learning device 400, and the like.
- the operation record DB 340 stores operation data generated based on various information sent from the on-board unit 101 and the smartphone 200.
- the user DB 350 stores various information (for example, age, gender, past driving history, self-history, total driving time, etc.) of the user to be evaluated for driving.
- the operation record DB 340 and the user DB 350 are realized by, for example, a magnetic disk or the like.
- the input unit 330 receives the input of the user's operation data received via the communication unit 320. Specifically, the input unit 330 receives input of driving data including information indicating the driving of the target person who evaluates driving, information indicating the environment at the time of driving, and position information from which these information are acquired. .. In addition, in order to distinguish from the above-mentioned expert driving data (expert driving data), the user's driving input here may be referred to as user driving data.
- the evaluation unit 370 outputs an evaluation result comparing the operation of the expert and the operation of the target person. Specifically, the evaluation unit 370 specifies the area where the user drives from the location information, and selects the cost function corresponding to the area. Next, the evaluation unit 370 applies the information of the environment in which the subject is driven to the selected cost function, and estimates the driving of the expert in the same environment. Then, the evaluation unit 370 outputs an evaluation result comparing the estimated driving of the expert and the driving of the target person. The evaluation unit 370 may display the evaluation result on the display unit 360.
- the evaluation unit 370 may evaluate the user's operation collectively (that is, collectively the period from the start of operation to the end of operation), or may sequentially evaluate the user's operation at the time of operation.
- the display unit 360 is a display device that outputs the evaluation result by the evaluation unit 370.
- the evaluation result output device 300 may transmit and display the content to be displayed on the display unit 360 to the on-board unit 101 or the smartphone 200.
- the first specific example is an example of visualizing the difference between the driving of the subject and the driving of the expert.
- FIG. 4 is an explanatory diagram illustrating a first specific example (an example of a process for visualizing a difference during operation).
- condition an environment (condition) of "Minato-ku, Tokyo, gentle curve, traffic jam” is assumed.
- T a certain time T at "60 km / h, accelerator pedal depression amount 20%”.
- the evaluation unit 370 estimates the operation of the expert estimated in the same environment (condition) based on the learned cost function.
- an expert for example, a "driving school instructor in Minato-ku, Tokyo" who is a driver related to the relevant area can be mentioned.
- the evaluation unit 370 calculates the difference between the driving of the subject and the driving of the expert in chronological order. The evaluation unit 370 may visualize this calculation result as illustrated in FIG.
- the condition for notifying the driver is defined as "when the speed difference exceeds ⁇ 5 km".
- the speed difference from the driving of a skilled person at time T is "+5 km”. Therefore, the evaluation unit 370 may notify the driver by voice, display, or the like, "Let's step on the accelerator strongly" in order to increase the speed. As described above, when the calculated difference satisfies the predetermined notification condition, the evaluation unit 370 may notify the content indicating the difference.
- the notification conditions may be determined based on the learned cost function.
- the notification condition may be determined based on the feature amount of the cost function having a high degree of importance. In this way, by using the learned cost function, it is possible to define the notification conditions by focusing on the evaluation items that should be paid more attention to.
- FIG. 5 is an explanatory diagram illustrating a second specific example (an example of a process of scoring and outputting a difference in operation).
- the evaluation unit 370 shall similarly estimate the operation of the expert. ..
- a method of notifying when the speed difference exceeds a predetermined threshold value is illustrated.
- the evaluation unit 370 cumulatively adds points in chronological order based on a predetermined method of scoring according to the difference between the driving of a skilled person and the driving of a target person, and calculates the added result. indicate.
- the method of scoring is defined as "points are deducted when the difference from the expert is 5 km / h or more, and points are added when the difference from the expert is less than 5 km / h for 10 seconds".
- the evaluation unit 370 cumulatively adds points based on the defined method. For example, as illustrated in FIG. 5, the evaluation unit 370 may superimpose the result of scoring on a graph showing the difference between the operation of the expert and the operation of the target person. In the example shown in FIG. 5, the driving of the driver A has a large difference from the skilled person throughout the running, so that there are many origins, and the driving of the driver B has almost no difference from the skilled person, so that points are added. ..
- the evaluation unit 370 may not only score one running record of an individual, but also calculate the cumulative value of a plurality of running records. This makes it possible, for example, to score points for a predetermined period (monthly) or to score points for each region.
- the evaluation unit 370 may aggregate the magnitude of the difference for each feature amount and output the evaluation corresponding to the feature amount having a large difference. For example, when the difference in the feature amount indicating the acceleration at the time of starting is large, the evaluation unit 370 may output a message such as "Please suppress the acceleration at the time of starting".
- the evaluation unit 370 is realized by a computer processor that operates according to a program (evaluation result output program). Further, the evaluation unit 370 may be included in the control unit 310 itself.
- FIG. 6 is a flowchart showing an operation example of the learning device 400 of the present embodiment.
- the input unit 430 accepts the input of the cost function (step S11).
- the learning unit 450 learns the cost function for each region by reverse reinforcement learning using the expert driving data including the information indicating the contents of the expert's driving collected for each region as training data (step S12).
- FIG. 7 is a flowchart showing an operation example of the evaluation result output device 300 of the present embodiment.
- the input unit 330 receives input of user driving data including information indicating the driving of the target person for evaluating driving, information indicating the environment at the time of driving, and the position information from which the information was acquired (step S21). ..
- the evaluation unit 370 identifies the area where the user drives from the location information, and selects the cost function corresponding to that area (step S22).
- the evaluation unit 370 applies information indicating the environment in which the subject is operating to the selected cost function, and estimates the operation of the expert in the same environment (step S23). Then, the evaluation unit 370 outputs an evaluation result comparing the estimated driving of the expert and the driving of the target person (step S24).
- the input unit 430 accepts the input of the cost function
- the learning unit 450 uses the expert driving data collected for each region as training data by reverse reinforcement learning for each region. Learn the cost function.
- the input unit 330 accepts the input of the user driving data
- the evaluation unit 370 identifies the area where the user drives from the position information, selects the corresponding cost function, and selects the target person for the selected cost function. Estimate the driving of a skilled person in the same environment by applying the information indicating the environment when driving. Then, the evaluation unit 370 outputs an evaluation result comparing the estimated driving of the expert and the driving of the target person. Therefore, it is possible to evaluate the driving of the target person in consideration of the driving area.
- the learning unit 450 defines a skilled person (for example, a skilled driver, a taxi driver, a driving school instructor, a patrol car driver) who is supposed to drive well, and the driver of the skilled person is driven. By performing machine learning from the data, the characteristics of the driver are extracted. This makes it possible to extract the feature amount for evaluating the operation.
- a skilled driver for example, a skilled driver, a taxi driver, a driving school instructor, a patrol car driver
- the learning unit 450 (more specifically, the learning result output unit 454) visualizes the evaluation weight, it becomes possible to extract items to be improved.
- a driving school can be considered.
- the driving evaluation system of the present embodiment it is possible to embody the guidance given to the students by the instructor, improve the skills of the graduates, and improve the quality of the instructors. ..
- the operation evaluation system of the present embodiment it is possible to review the speed limit and the like according to the region. In addition, if accidents in the area can be reduced, it will be possible to appeal that the area is safe to the outside world. Specifically, the speed limit can be reviewed by comparing the difference between the running speed of the expert and the legal speed. In addition, it will be possible to grasp driving trends in areas where there are many accidents, to directly call attention to them, and to develop infrastructure that reduces accidents.
- FIG. 8 is a block diagram showing an outline of the operation evaluation system according to the present invention.
- the operation evaluation system 70 (for example, the operation evaluation system 1) according to the present invention is a function input that accepts an input of a cost function represented by a linear sum of terms in which the degree of importance is weighted to each feature amount indicating the driver's operation.
- Cost function for each region by means 71 (for example, cost function input unit 451) and inverse reinforcement learning using expert driving data including information indicating the contents of the expert's driving collected for each region as training data.
- the learning means 72 (for example, the reverse reinforcement learning unit 453) for learning the above, information indicating the driving of the target person for evaluating the driving, information indicating the environment at the time of driving, and the position information from which the information was acquired are included.
- the driving data input means 73 (for example, the input unit 330) that accepts the input of the user driving data and the area where the user drives are specified from the position information, and the cost function corresponding to the area is selected and selected. By applying the information indicating the environment when the subject is driving to the cost function, the driving of the expert in the same environment is estimated, and the evaluation result comparing the estimated driving of the expert with the driving of the subject is obtained. It is provided with an evaluation means 74 (for example, an evaluation unit 370) for outputting.
- the operation evaluation system 70 may include a learning result output means (for example, a learning result output unit 454) that outputs the feature amount included in the cost function in association with the weight of the feature amount.
- a learning result output means for example, a learning result output unit 454.
- the operation evaluation system 70 may be provided with a data extraction means (for example, a data extraction unit 452) for extracting training data for each region. Then, the learning means 72 may learn the cost function for each region by using the extracted training data for each region.
- a data extraction means for example, a data extraction unit 452 for extracting training data for each region.
- the learning means 72 may learn the cost function for each region by using the extracted training data for each region.
- the data extraction means may extract the training data of an expert from the training data as a candidate based on a predetermined standard.
- the evaluation means 74 calculates the difference between the driving of the expert and the driving of the target person in chronological order, and when the calculated difference satisfies the predetermined notification condition, the evaluation means 74 notifies the content indicating the difference. May be good.
- the evaluation means 74 cumulatively adds points in time series based on a predetermined method of scoring according to the difference between the driving of the expert and the driving of the target person, and displays the added result. May be good.
- the function input means 71 may accept input of a cost function including a term in which the degree of emphasis is weighted to each feature amount indicating the environment at the time of operation as a linear sum.
- FIG. 9 is a block diagram showing an outline of the learning device according to the present invention.
- the learning device 80 (for example, the learning device 400) according to the present invention includes a function input means 81 and a learning means 82.
- the contents of the function input means 81 and the learning means 82 are the same as those of the function input means 71 and the learning means 72 illustrated in FIG.
- FIG. 10 is a block diagram showing an outline of the evaluation result output device according to the present invention.
- the evaluation result output device 90 (for example, the evaluation result output device 300) according to the present invention includes an operation data input means 91 and an evaluation means 92.
- the contents of the operation data input means 91 and the evaluation means 92 are the same as those of the operation data input means 73 and the evaluation means 74 illustrated in FIG.
- a function input means that accepts input of a cost function represented by a linear sum of terms in which the degree of importance is weighted to each feature amount indicating the driver's operation.
- a learning means for learning the cost function for each region by reverse reinforcement learning using expert driving data including information indicating the contents of the expert's driving collected for each region as training data.
- a driving evaluation system characterized in that it is equipped with an evaluation means that estimates the driving of a skilled person in the environment of the above and outputs an evaluation result comparing the estimated driving of the skilled person with the driving of the subject.
- Appendix 2 The operation evaluation system according to Appendix 1 provided with a learning result output means for outputting the feature amount included in the cost function in association with the weight of the feature amount.
- Appendix 3 Equipped with a data extraction means to extract training data for each region
- the learning means is the operation evaluation system according to Appendix 1 or Appendix 2, which learns the cost function for each region using the extracted training data for each region.
- the data extraction means is the operation evaluation system according to Appendix 3, which extracts training data of an expert from candidate training data based on a predetermined standard.
- the evaluation means calculates the difference between the driving of the expert and the driving of the target person in chronological order, and when the calculated difference satisfies the predetermined notification condition, the evaluation means notifies the content indicating the difference.
- the operation evaluation system according to any one of Supplementary note 1 to Supplementary note 4.
- the evaluation means cumulatively adds points in chronological order based on a predetermined method of scoring according to the difference between the driving of a skilled person and the driving of a target person, and displays the added result.
- the operation evaluation system according to any one of Supplementary note 1 to Supplementary note 5.
- the function input means accepts the input of a cost function including a term in which the degree of emphasis is weighted to each feature amount indicating the environment during operation as a linear sum.
- the operation evaluation system described in one.
- a function input means that accepts input of a cost function represented by a linear sum of terms in which the degree of importance is weighted to each feature amount indicating the driver's operation. It is provided with a learning means for learning the cost function for each region by reverse reinforcement learning using expert driving data including information indicating the contents of the expert's driving collected for each region as training data. Characterized learning device.
- An operation data input means that accepts input of user operation data including information indicating the operation of the target person who evaluates the operation, information indicating the environment at the time of operation, and the position information from which the information was acquired.
- Emphasis is placed on each feature amount that indicates the driver's driving, which was learned for each region by reverse reinforcement learning using the expert driving data including information showing the contents of the driver's driving collected for each region as training data.
- the cost functions represented by the linear sum of the terms weighted by each degree From the cost functions represented by the linear sum of the terms weighted by each degree, the area where the user drives is specified from the position information, the cost function corresponding to the area is selected, and the selected cost function is selected.
- Appendix 11 The operation evaluation method according to Appendix 10, wherein the feature amount included in the cost function and the weight of the feature amount are associated and output.
- a learning method characterized by learning the cost function for each region by reverse reinforcement learning using expert driving data including information indicating the contents of the expert's driving collected for each region as training data.
- a program storage medium that stores the evaluation result output program for executing the evaluation process to be output.
- Operation evaluation system 100 Vehicle 101 On-board unit 110, 210, 310, 410 Control unit 111, 211, 311, 411 CPU 112,212,312,412 Memory 120,220,320,420 Communication unit 130,230,440 Storage unit 140 External camera 150 Vehicle information sensor 160 Biosensor 170 In-vehicle camera 200 Smartphone 240,330,430 Input unit 250 Movement information DB 300 Evaluation result output device 340 Operation record DB 350 User DB 360 Display unit 370 Evaluation unit 400 Learning device 450 Learning unit 451 Cost function input unit 452 Data extraction unit 453 Reverse reinforcement learning unit 454 Learning result output unit 500 Storage device 600 Display device
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Educational Administration (AREA)
- Aviation & Aerospace Engineering (AREA)
- Educational Technology (AREA)
- Business, Economics & Management (AREA)
- Software Systems (AREA)
- Mechanical Engineering (AREA)
- Transportation (AREA)
- Automation & Control Theory (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Traffic Control Systems (AREA)
Abstract
Description
コスト関数=100×速度+50×前方車との距離+10×アクセルペダル踏み込み量 (式2)
この場合、学習結果出力部454は、[速度,前方車との距離]の評価の重みを[100,50,10]と出力してもよい。
第一の例として、OEM(original equipment manufacturer)への適用が考えられる。本実施形態の運転評価システムを用いることで、車両の実際の使用傾向が把握できるため、対象とする国や地域専用の車両(例えば、寒冷地専用車、名古屋用専用車、など)を開発することが可能になり、社会価値の高い車両作りが可能になる。
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により、当該地域ごとの前記コスト関数を学習する学習手段と、
運転を評価する対象者の運転を示す情報、運転の際の環境を示す情報、および、当該情報が取得された位置情報を含む利用者運転データの入力を受け付ける運転データ入力手段と、
前記利用者が運転する地域を位置情報から特定して、当該地域に対応するコスト関数を選択し、選択されたコスト関数に前記対象者の運転の際の環境を示す情報を適用して、同一の環境における熟練者の運転を推定し、推定された熟練者の運転と前記対象者の運転とを比較した評価結果を出力する評価手段とを備えた
ことを特徴とする運転評価システム。
付記1記載の運転評価システム。
学習手段は、抽出された地域ごとのトレーニングデータを用いて、当該地域ごとのコスト関数を学習する
付記1または付記2記載の運転評価システム。
付記3記載の運転評価システム。
付記1から付記4のうちのいずれか1つに記載の運転評価システム。
付記1から付記5のうちのいずれか1つに記載の運転評価システム。
付記1から付記6のうちのいずれか1つに記載の運転評価システム。
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により、当該地域ごとの前記コスト関数を学習する学習手段とを備えた
ことを特徴とする学習装置。
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により当該地域ごとに学習された、運転者の運転を示す各特徴量に重視度合いがそれぞれ重み付けされた項の線形和で表されたコスト関数の中から、前記利用者が運転する地域を位置情報から特定して当該地域に対応するコスト関数を選択し、選択されたコスト関数に前記対象者の運転の際の環境を示す情報を適用して、同一の環境における熟練者の運転を推定し、推定された熟練者の運転と前記対象者の運転とを比較した評価結果を出力する評価手段とを備えた
ことを特徴とする評価結果出力装置。
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により、当該地域ごとの前記コスト関数を学習し、
運転を評価する対象者の運転を示す情報、運転の際の環境を示す情報、および、当該情報が取得された位置情報を含む利用者運転データの入力を受け付け、
前記利用者が運転する地域を位置情報から特定して、当該地域に対応するコスト関数を選択し、
選択されたコスト関数に前記対象者の運転の際の環境を示す情報を適用して、同一の環境における熟練者の運転を推定し、
推定された熟練者の運転と前記対象者の運転とを比較した評価結果を出力する
ことを特徴とする運転評価方法。
付記10記載の運転評価方法。
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により、当該地域ごとの前記コスト関数を学習する
ことを特徴とする学習方法。
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により当該地域ごとに学習された、運転者の運転を示す各特徴量に重視度合いがそれぞれ重み付けされた項の線形和で表されたコスト関数の中から、前記利用者が運転する地域を位置情報から特定して当該地域に対応するコスト関数を選択し、
選択されたコスト関数に前記対象者の運転の際の環境を示す情報を適用して、同一の環境における熟練者の運転を推定し、
推定された熟練者の運転と前記対象者の運転とを比較した評価結果を出力する
ことを特徴とする評価結果出力方法。
運転者の運転を示す各特徴量に重視度合いがそれぞれ重み付けされた項の線形和で表されたコスト関数の入力を受け付ける関数入力処理、および、
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により、当該地域ごとの前記コスト関数を学習する学習処理
を実行させるための学習プログラムを記憶するプログラム記憶媒体。
運転を評価する対象者の運転を示す情報、運転の際の環境を示す情報、および、当該情報が取得された位置情報を含む利用者運転データの入力を受け付ける運転データ入力処理、および、
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により当該地域ごとに学習された、運転者の運転を示す各特徴量に重視度合いがそれぞれ重み付けされた項の線形和で表されたコスト関数の中から、前記利用者が運転する地域を位置情報から特定して当該地域に対応するコスト関数を選択し、選択されたコスト関数に前記対象者の運転の際の環境を示す情報を適用して、同一の環境における熟練者の運転を推定し、推定された熟練者の運転と前記対象者の運転とを比較した評価結果を出力する評価処理
を実行させるための評価結果出力プログラムを記憶するプログラム記憶媒体。
運転者の運転を示す各特徴量に重視度合いがそれぞれ重み付けされた項の線形和で表されたコスト関数の入力を受け付ける関数入力処理、および、
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により、当該地域ごとの前記コスト関数を学習する学習処理
を実行させるための学習プログラム。
運転を評価する対象者の運転を示す情報、運転の際の環境を示す情報、および、当該情報が取得された位置情報を含む利用者運転データの入力を受け付ける運転データ入力処理、および、
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により当該地域ごとに学習された、運転者の運転を示す各特徴量に重視度合いがそれぞれ重み付けされた項の線形和で表されたコスト関数の中から、前記利用者が運転する地域を位置情報から特定して当該地域に対応するコスト関数を選択し、選択されたコスト関数に前記対象者の運転の際の環境を示す情報を適用して、同一の環境における熟練者の運転を推定し、推定された熟練者の運転と前記対象者の運転とを比較した評価結果を出力する評価処理
を実行させるための評価結果出力プログラム。
100 車両
101 車載器
110,210,310,410 制御部
111,211,311,411 CPU
112,212,312,412 メモリ
120,220,320,420 通信部
130,230,440 記憶部
140 車外カメラ
150 車両情報用センサ
160 生体センサ
170 車内カメラ
200 スマートフォン
240,330,430 入力部
250 移動情報DB
300 評価結果出力装置
340 運転実績DB
350 利用者DB
360 表示部
370 評価部
400 学習装置
450 学習部
451 コスト関数入力部
452 データ抽出部
453 逆強化学習部
454 学習結果出力部
500 記憶装置
600 表示装置
Claims (15)
- 運転者の運転を示す各特徴量に重視度合いがそれぞれ重み付けされた項の線形和で表されたコスト関数の入力を受け付ける関数入力手段と、
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により、当該地域ごとの前記コスト関数を学習する学習手段と、
運転を評価する対象者の運転を示す情報、運転の際の環境を示す情報、および、当該情報が取得された位置情報を含む利用者運転データの入力を受け付ける運転データ入力手段と、
前記利用者が運転する地域を位置情報から特定して、当該地域に対応するコスト関数を選択し、選択されたコスト関数に前記対象者の運転の際の環境を示す情報を適用して、同一の環境における熟練者の運転を推定し、推定された熟練者の運転と前記対象者の運転とを比較した評価結果を出力する評価手段とを備えた
ことを特徴とする運転評価システム。 - コスト関数に含まれる特徴量と、当該特徴量の重みとを対応付けて出力する学習結果出力手段を備えた
請求項1記載の運転評価システム。 - 地域ごとにトレーニングデータを抽出するデータ抽出手段を備え、
学習手段は、抽出された地域ごとのトレーニングデータを用いて、当該地域ごとのコスト関数を学習する
請求項1または請求項2記載の運転評価システム。 - データ抽出手段は、予め定めた基準に基づいて、候補とするトレーニングデータから熟練者のトレーニングデータを抽出する
請求項3記載の運転評価システム。 - 評価手段は、熟練者の運転と対象者の運転との差異を時系列に算出し、算出した差異が予め定めた通知条件を満たした場合に、当該差異を示す内容を通知する
請求項1から請求項4のうちのいずれか1項に記載の運転評価システム。 - 評価手段は、熟練者の運転と対象者の運転との差異に応じて得点化する予め定めた方法に基づき、時系列に得点を累積的に加算し、加算した結果を表示する
請求項1から請求項5のうちのいずれか1項に記載の運転評価システム。 - 関数入力手段は、運転の際の環境を示す各特徴量に重視度合いがそれぞれ重み付けされた項を線形和として含むコスト関数の入力を受け付ける
請求項1から請求項6のうちのいずれか1項に記載の運転評価システム。 - 運転者の運転を示す各特徴量に重視度合いがそれぞれ重み付けされた項の線形和で表されたコスト関数の入力を受け付ける関数入力手段と、
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により、当該地域ごとの前記コスト関数を学習する学習手段とを備えた
ことを特徴とする学習装置。 - 運転を評価する対象者の運転を示す情報、運転の際の環境を示す情報、および、当該情報が取得された位置情報を含む利用者運転データの入力を受け付ける運転データ入力手段と、
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により当該地域ごとに学習された、運転者の運転を示す各特徴量に重視度合いがそれぞれ重み付けされた項の線形和で表されたコスト関数の中から、前記利用者が運転する地域を位置情報から特定して当該地域に対応するコスト関数を選択し、選択されたコスト関数に前記対象者の運転の際の環境を示す情報を適用して、同一の環境における熟練者の運転を推定し、推定された熟練者の運転と前記対象者の運転とを比較した評価結果を出力する評価手段とを備えた
ことを特徴とする評価結果出力装置。 - 運転者の運転を示す各特徴量に重視度合いがそれぞれ重み付けされた項の線形和で表されたコスト関数の入力を受け付け、
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により、当該地域ごとの前記コスト関数を学習し、
運転を評価する対象者の運転を示す情報、運転の際の環境を示す情報、および、当該情報が取得された位置情報を含む利用者運転データの入力を受け付け、
前記利用者が運転する地域を位置情報から特定して、当該地域に対応するコスト関数を選択し、
選択されたコスト関数に前記対象者の運転の際の環境を示す情報を適用して、同一の環境における熟練者の運転を推定し、
推定された熟練者の運転と前記対象者の運転とを比較した評価結果を出力する
ことを特徴とする運転評価方法。 - コスト関数に含まれる特徴量と、当該特徴量の重みとを対応付けて出力する
請求項10記載の運転評価方法。 - 運転者の運転を示す各特徴量に重視度合いがそれぞれ重み付けされた項の線形和で表されたコスト関数の入力を受け付け、
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により、当該地域ごとの前記コスト関数を学習する
ことを特徴とする学習方法。 - 運転を評価する対象者の運転を示す情報、運転の際の環境を示す情報、および、当該情報が取得された位置情報を含む利用者運転データの入力を受け付け、
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により当該地域ごとに学習された、運転者の運転を示す各特徴量に重視度合いがそれぞれ重み付けされた項の線形和で表されたコスト関数の中から、前記利用者が運転する地域を位置情報から特定して当該地域に対応するコスト関数を選択し、
選択されたコスト関数に前記対象者の運転の際の環境を示す情報を適用して、同一の環境における熟練者の運転を推定し、
推定された熟練者の運転と前記対象者の運転とを比較した評価結果を出力する
ことを特徴とする評価結果出力方法。 - コンピュータに、
運転者の運転を示す各特徴量に重視度合いがそれぞれ重み付けされた項の線形和で表されたコスト関数の入力を受け付ける関数入力処理、および、
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により、当該地域ごとの前記コスト関数を学習する学習処理
を実行させるための学習プログラムを記憶するプログラム記憶媒体。 - コンピュータに、
運転を評価する対象者の運転を示す情報、運転の際の環境を示す情報、および、当該情報が取得された位置情報を含む利用者運転データの入力を受け付ける運転データ入力処理、および、
地域ごとに収集された熟練者の運転の内容を表わす情報を含む熟練者運転データをトレーニングデータとして用いた逆強化学習により当該地域ごとに学習された、運転者の運転を示す各特徴量に重視度合いがそれぞれ重み付けされた項の線形和で表されたコスト関数の中から、前記利用者が運転する地域を位置情報から特定して当該地域に対応するコスト関数を選択し、選択されたコスト関数に前記対象者の運転の際の環境を示す情報を適用して、同一の環境における熟練者の運転を推定し、推定された熟練者の運転と前記対象者の運転とを比較した評価結果を出力する評価処理
を実行させるための評価結果出力プログラムを記憶するプログラム記憶媒体。
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2020/048743 WO2022137506A1 (ja) | 2020-12-25 | 2020-12-25 | 運転評価システム、学習装置、評価結果出力装置、方法およびプログラム |
EP20966992.8A EP4250272A4 (en) | 2020-12-25 | 2020-12-25 | DRIVING EVALUATION SYSTEM, LEARNING DEVICE, EVALUATION RESULTS GENERATION DEVICE, METHOD AND PROGRAM |
US18/269,443 US20240083441A1 (en) | 2020-12-25 | 2020-12-25 | Driving evaluation system, learning device, evaluation result output device, method, and program |
JP2022570949A JP7552727B2 (ja) | 2020-12-25 | 2020-12-25 | 運転評価システム、学習装置、評価結果出力装置、方法およびプログラム |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2020/048743 WO2022137506A1 (ja) | 2020-12-25 | 2020-12-25 | 運転評価システム、学習装置、評価結果出力装置、方法およびプログラム |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022137506A1 true WO2022137506A1 (ja) | 2022-06-30 |
Family
ID=82157589
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2020/048743 WO2022137506A1 (ja) | 2020-12-25 | 2020-12-25 | 運転評価システム、学習装置、評価結果出力装置、方法およびプログラム |
Country Status (4)
Country | Link |
---|---|
US (1) | US20240083441A1 (ja) |
EP (1) | EP4250272A4 (ja) |
JP (1) | JP7552727B2 (ja) |
WO (1) | WO2022137506A1 (ja) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115545118A (zh) * | 2022-11-16 | 2022-12-30 | 北京集度科技有限公司 | 车辆驾驶评价及其模型的训练方法、装置、设备及介质 |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190072959A1 (en) * | 2017-09-06 | 2019-03-07 | GM Global Technology Operations LLC | Unsupervised learning agents for autonomous driving applications |
WO2019150452A1 (ja) * | 2018-01-30 | 2019-08-08 | 日本電気株式会社 | 情報処理装置、制御方法、及びプログラム |
WO2019167457A1 (ja) * | 2018-02-28 | 2019-09-06 | ソニー株式会社 | 情報処理装置、情報処理方法、プログラム、及び移動体 |
WO2020049737A1 (ja) | 2018-09-07 | 2020-03-12 | 株式会社オファサポート | 運転技能評価システム、方法及びプログラム |
WO2020115904A1 (ja) * | 2018-12-07 | 2020-06-11 | 日本電気株式会社 | 学習装置、学習方法、および学習プログラム |
WO2020183609A1 (ja) * | 2019-03-12 | 2020-09-17 | 三菱電機株式会社 | 移動体制御装置および移動体制御方法 |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9090255B2 (en) * | 2012-07-12 | 2015-07-28 | Honda Motor Co., Ltd. | Hybrid vehicle fuel efficiency using inverse reinforcement learning |
CN108427985B (zh) * | 2018-01-02 | 2020-05-19 | 北京理工大学 | 一种基于深度强化学习的插电式混合动力车辆能量管理方法 |
-
2020
- 2020-12-25 WO PCT/JP2020/048743 patent/WO2022137506A1/ja active Application Filing
- 2020-12-25 JP JP2022570949A patent/JP7552727B2/ja active Active
- 2020-12-25 EP EP20966992.8A patent/EP4250272A4/en active Pending
- 2020-12-25 US US18/269,443 patent/US20240083441A1/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190072959A1 (en) * | 2017-09-06 | 2019-03-07 | GM Global Technology Operations LLC | Unsupervised learning agents for autonomous driving applications |
WO2019150452A1 (ja) * | 2018-01-30 | 2019-08-08 | 日本電気株式会社 | 情報処理装置、制御方法、及びプログラム |
WO2019167457A1 (ja) * | 2018-02-28 | 2019-09-06 | ソニー株式会社 | 情報処理装置、情報処理方法、プログラム、及び移動体 |
WO2020049737A1 (ja) | 2018-09-07 | 2020-03-12 | 株式会社オファサポート | 運転技能評価システム、方法及びプログラム |
WO2020115904A1 (ja) * | 2018-12-07 | 2020-06-11 | 日本電気株式会社 | 学習装置、学習方法、および学習プログラム |
WO2020183609A1 (ja) * | 2019-03-12 | 2020-09-17 | 三菱電機株式会社 | 移動体制御装置および移動体制御方法 |
Non-Patent Citations (3)
Title |
---|
"ENCYCLOPEDIA OF ARTIFICIAL INTELLIGENCE CHAPTER 6", 1 July 2017, article ANONYMOUS: " Machine learning and data mining [6-10] Reinforcement Learning, Reinforcement Learning", pages: 1 - 1, XP055951354 * |
ANONYMOUS: "Developed by NEC, an AI that can make decisions at the same level as experts Learns ``intentions'' with ``reverse reinforcement learning''-ITmedia NEWS", ITMEDIA, 17 July 2019 (2019-07-17), pages 1 - 4, XP055951357, Retrieved from the Internet <URL:https://www.itmedia.co.jp/news/articles/1907/17/news095.html> [retrieved on 20220815] * |
See also references of EP4250272A4 |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115545118A (zh) * | 2022-11-16 | 2022-12-30 | 北京集度科技有限公司 | 车辆驾驶评价及其模型的训练方法、装置、设备及介质 |
Also Published As
Publication number | Publication date |
---|---|
EP4250272A4 (en) | 2024-01-17 |
JP7552727B2 (ja) | 2024-09-18 |
JPWO2022137506A1 (ja) | 2022-06-30 |
EP4250272A1 (en) | 2023-09-27 |
US20240083441A1 (en) | 2024-03-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111164660B (zh) | 信息处理装置、信息处理方法和程序 | |
CN106874597B (zh) | 一种应用于自动驾驶车辆的高速公路超车行为决策方法 | |
JP5434912B2 (ja) | 運転状態判定方法、運転状態判定システム及びプログラム | |
US10915964B1 (en) | System and method for providing vehicle services based on driving behaviors | |
KR101405184B1 (ko) | 운전의 적절성 및 테스트 대상의 성과를 평가하고 기록하는 장치 및 방법 | |
JP5375805B2 (ja) | 運転支援システム及び運転支援管理センター | |
US8738523B1 (en) | Systems and methods to identify and profile a vehicle operator | |
US12005922B2 (en) | Toward simulation of driver behavior in driving automation | |
CN111325230B (zh) | 车辆换道决策模型的在线学习方法和在线学习装置 | |
JP6703465B2 (ja) | 運転支援装置、センタ装置 | |
US20180370537A1 (en) | System providing remaining driving information of vehicle based on user behavior and method thereof | |
WO2011033840A1 (ja) | 運転評価システム、車載機及び情報処理センター | |
US20160196613A1 (en) | Automated Driver Training And Premium Discounts | |
JPWO2017038166A1 (ja) | 情報処理装置、情報処理方法及びプログラム | |
CN112749740B (zh) | 确定车辆目的地的方法、装置、电子设备及介质 | |
US10475350B1 (en) | System and method for a driving simulator on a mobile device | |
CN111856969A (zh) | 自动驾驶仿真测试方法及装置 | |
CA2805439C (en) | Systems and methods using a mobile device to collect data for insurance premiums | |
WO2022137506A1 (ja) | 運転評価システム、学習装置、評価結果出力装置、方法およびプログラム | |
JP2010152497A (ja) | 運転特性取得装置及び交通シミュレーションシステム | |
JP6619316B2 (ja) | 駐車位置探索方法、駐車位置探索装置、駐車位置探索プログラム及び移動体 | |
CN114126943A (zh) | 应用于自主车辆中的方法和设备 | |
Toledo | Topics in traffic simulation and travel behavior modeling | |
KR102572305B1 (ko) | 자율주행차량의 튜토리얼 서비스 시스템 및 이를 제공하는 방법 | |
US20160167674A1 (en) | Telematics enabled drivers education |
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: 20966992 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2022570949 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 18269443 Country of ref document: US |
|
ENP | Entry into the national phase |
Ref document number: 2020966992 Country of ref document: EP Effective date: 20230621 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |