JP2019018609A - Vehicular emergency brake predictor and prediction method - Google Patents

Abstract

An emergency brake prediction device capable of predicting, with high probability, when a driver tries to apply an emergency brake regardless of the amount of depression of an accelerator pedal immediately before. An emergency brake prediction device 4 is created by machine learning using time series data indicating an operation pattern of an accelerator pedal detected by a position sensor 31 when a driver raises his / her foot from the accelerator pedal. A classifier 43 is provided for classifying the pattern into a case where the driver is trying to apply emergency braking and a case where the driver is not applying emergency braking. The prediction device 4 receives time series data indicating the operation pattern of the accelerator pedal detected by the position sensor 11 mounted on the automobile 1 as input, and the driver applies an emergency brake based on the determination result by the classifier 43. When it is predicted that the vehicle is going to be driven, the vehicle control device 12 is instructed to drive the emergency brake. [Selection] Figure 1

Description

  The present invention relates to an automobile emergency brake prediction apparatus and a prediction method for predicting whether or not a driver is about to apply an emergency brake based on time-series data indicating an operation pattern of an accelerator pedal.

  When a sudden change occurs, such as when a passerby jumps out from behind a parked car while driving in a car, the driver applies an emergency brake to stop the car.

  On the other hand, after the driver tries to apply the emergency brake, it takes time to lift the foot from the accelerator pedal and move the foot to the brake pedal and to depress the brake pedal before the brake is actually operated.

  In the meantime, the car runs idle despite the driver's intention to stop the car. This time is 200 ms to 400 ms, and if the car is traveling at 50 km / h, it will move from 2.8 m to 5.6 m. Shortening the idling time is important for improving safety by shortening the stopping distance of the automobile during emergency braking.

  In order to shorten the idling time described above, there has been proposed a device that automatically activates a brake when the return speed of an accelerator pedal becomes a certain value or more (see Patent Document 1).

Japanese Patent Laid-Open No. 10-315938

  However, in the above-mentioned device, due to the characteristics of the spring that returns to the state before the accelerator pedal was depressed, when the amount of depression of the immediately preceding pedal is small, the return speed of the pedal becomes small, so the brake does not operate and the emergency brake is applied. I can not cope.

  The present invention has been made in view of such a conventional problem, and can predict with high probability when a driver tries to apply an emergency brake regardless of the amount of depression of the accelerator pedal immediately before. An object is to provide an emergency brake prediction device and a prediction method.

In order to achieve the above object, an automobile emergency brake prediction apparatus according to the present invention includes:
It is created by performing machine learning using time series data indicating the operation pattern of the accelerator pedal when the driver raises his / her foot from the accelerator pedal, detected by the accelerator pedal position sensor mounted on the drive simulator or automobile, A classifier for classifying the operation pattern into a case where the driver is about to apply emergency braking and a case where the driver is not applying emergency braking;
Predicted that the driver is going to apply emergency braking based on the result of judgment by the classifier using time-series data indicating the operation pattern of the accelerator pedal detected by the accelerator pedal position sensor mounted on the vehicle as input. In this case, the vehicle control device is instructed to drive the emergency brake.

  Here, it is preferable that the time series data is created by sampling the output signal of the accelerator pedal position sensor at regular intervals.

  The automobile emergency brake prediction apparatus according to the present invention includes a data cutout unit that cuts out data from the time series data until the driver returns to the state before the accelerator pedal is depressed from when the driver releases his or her foot. It is preferable to include a feature extraction unit that extracts a plurality of feature amounts from the time-series data extracted by the data extraction unit.

  The learning data storage unit further stores a feature amount of the time series data extracted by the feature extraction unit, and the classifier includes features of a plurality of time series data read from the learning data storage unit. It is preferably created using quantity data.

  The plurality of feature amounts are preferably the position of the accelerator pedal immediately before the driver releases his / her foot, the maximum speed and the average speed when the accelerator pedal returns.

  The automobile control device preferably causes the wheels to exert braking force by operating a brake actuator and supplying pressure-adjusted brake fluid to the wheels of the automobile when an emergency brake driving instruction is given. .

Moreover, the emergency brake prediction method for automobiles according to the present invention is as follows.
A data collection step for collecting time-series data indicating an operation pattern of the accelerator pedal when the driver lifts his / her foot from the accelerator pedal using an accelerator pedal position sensor mounted on the drive simulator or the vehicle;
Machine learning using the time series data, and a classifier creating step for creating a classifier that classifies the operation pattern into a case where the driver is trying to apply emergency braking and a case where the driver is not applying emergency braking, and
A step of incorporating the classifier created in the classifier creating step into an emergency brake prediction device mounted on a vehicle;
Based on the result of determination by the classifier incorporated in the emergency brake prediction device, time series data indicating the operation pattern of the accelerator pedal detected by the accelerator pedal position sensor mounted on the automobile is input to the emergency brake prediction device. And an emergency brake prediction step of instructing the control device of the automobile to drive the emergency brake when it is predicted that the driver is going to apply the emergency brake.

  According to the automobile emergency brake prediction device and the prediction method according to the present invention, regardless of the amount of depression of the immediately preceding accelerator pedal, when a driver tries to apply an emergency brake, it predicts that with a high probability, The brake can be activated. As a result, the stopping distance of the automobile during emergency braking can be kept to a minimum.

It is a figure which shows the structure of the emergency brake prediction apparatus mounted in the motor vehicle, and the classifier preparation apparatus which produces the classifier integrated in the said prediction apparatus. It is a figure which shows the member relevant to the brake among the members mounted in the motor vehicle. It is a graph which shows a movement of an accelerator pedal and a brake pedal when a driver applies emergency braking. It is a graph which shows distribution of time delay when an emergency brake is applied. It is a figure which shows the relationship between the maximum rate when an accelerator pedal returns, and the maximum rate when a brake pedal is pushed down. It is a figure which shows the relationship between the average rate when an accelerator pedal returns, and the position when a brake pedal is depressed. It is a graph which shows the change of the amount of depression of a brake pedal when a brake actuator is operated according to prediction of an emergency brake prediction device, and when a driver applies emergency brake with his / her foot. The graph which shows the result of having judged the operation pattern of the accelerator pedal about each of the case where a driver is going to apply emergency brake and the case where normal brake is applied using the classifier created in Embodiment 1. is there. It is a figure which shows the structure at the time of mounting a classifier preparation apparatus with an emergency brake prediction apparatus in a motor vehicle.

  Hereinafter, an automobile emergency brake prediction apparatus and prediction method according to embodiments of the present invention will be described with reference to the drawings.

(Embodiment 1)
<Emergency brake prediction device and configuration of a car equipped with it>
FIG. 1 shows a configuration of an emergency brake prediction device mounted on an automobile and a classifier creation device that creates a classifier incorporated in the prediction device. FIG. 2 shows members related to the brake among the members mounted on the automobile.

  The classifier creating apparatus 2 shown in FIG. 1 is based on time-series data indicating the depression amount of an accelerator pedal detected by an accelerator pedal position sensor (hereinafter, abbreviated as “positive sensor”) 31 mounted on the drive simulator 3. The machine learns the operation pattern of the accelerator pedal when the driver raises his / her foot by machine learning, and creates a classifier that classifies the operation pattern when the driver is about to apply an emergency brake and when it is not. is there.

  The classifier program created by the classifier creation device 2 is incorporated into the classifier 43 of the emergency brake prediction device 4 mounted on the automobile 1. Then, time series data indicating the operation pattern of the accelerator pedal detected by the position sensor 11 of the traveling vehicle 1 is input to the classifier 43.

 The emergency brake prediction device 4 sends an electronic control unit (ECU) 12 that controls the movement of the automobile 1 when it is predicted that the driver is about to apply an emergency brake based on the classification result of the operation pattern. Instruct to drive emergency brake. The ECU 12 that has received the signal from the emergency brake prediction device 4 operates the brake actuator 17 to stop the automobile 1.

 Before describing the configurations and operations of the classifier creation device 2 and the emergency brake prediction device 4, with reference to FIG. 3 to FIG. 6, a classification in which the operation pattern of the accelerator pedal when the driver raises the foot is machine-learned Explain the reason for creating the container.

  FIG. 3 is a graph showing the movements of the accelerator pedal and the brake pedal when the driver applies an emergency brake. The horizontal axis indicates elapsed time, and the vertical axis indicates the position of each pedal with the maximum depression amount being 100. It is expressed as a percentage of time. The lower half of the figure shows the speed at which the accelerator pedal returns (here, indicated as a rate) obtained by dividing the accelerator pedal position [%] by time.

  In the figure, the accelerator pedal position is indicated by a solid line, the brake pedal position is indicated by a broken line, and the accelerator pedal return rate (speed) is indicated by a chain line. The same shall apply thereafter.

 FIG. 3A is a graph when the stepping amount immediately before the driver raises his foot from the accelerator pedal is 56%, and FIG. 3B is a graph when the stepping amount immediately before is 38%. .

  As shown in FIG. 3A, the operation when applying the emergency brake is divided into three sections. In the first section A, the foot that was stepping on the accelerator pedal was released from the pedal. In the next section B, the foot was transferred from the accelerator pedal to the brake pedal. In the last section C, the brake pedal was stepped on and close to the maximum position. Press down.

  In FIG. 4, when the emergency brake is applied, the time delay of adding the above three sections A, B, and C, that is, the driver depresses the brake pedal to the maximum position after releasing the accelerator pedal. The distribution of time required is shown. The data in FIG. 4 is obtained by classifying 316 sample data of 11 drivers measured using a drive simulator for each time.

  It varies depending on the driving characteristics of the driver and the driving conditions when the brakes are applied, but in the central sample, it takes 0.2 to 0.4 seconds to release the brake pedal after releasing the accelerator pedal. ing. Therefore, when the automobile is traveling at 50 km / h, the vehicle runs idle between 2.8 m and 5.6 m. If this idling time can be shortened, the possibility of preventing an accident will increase.

  Next, with reference to FIG. 5 and FIG. 6, the relationship between the operation when the accelerator pedal returns and the operation when the brake pedal is depressed will be described based on the experimental results in the drive simulator.

  FIG. 5 shows the relationship between the maximum rate (speed: horizontal axis) when the accelerator pedal returns and the maximum rate (speed: vertical axis) when the brake pedal is depressed. In the figure, ● is a sample of normal braking, and ○ is a sample during emergency braking.

  As can be seen from the figure, there is a correlation between the operation when the accelerator pedal returns and the operation when the brake pedal is depressed during emergency braking. In particular, when the maximum rate (speed) when the accelerator pedal returns is large, the correlation between the accelerator and brake movement is clear.

 On the other hand, when the maximum rate when the accelerator pedal returns is around 500 [% / S], it is difficult to distinguish because the maximum rate of brake pedal depression during normal operation overlaps that during emergency braking.

  FIG. 6 shows the relationship between the average rate (speed: horizontal axis) when the accelerator pedal returns and the position (vertical axis) when the brake pedal is depressed. As can be seen from the figure, the position of the brake pedal during emergency braking is concentrated between 80% and 100%, which is clearly different from that during normal operation. However, some samples are difficult to distinguish because the samples during normal operation overlap with the samples during emergency braking.

  From the results shown in FIG. 5 and FIG. 6, when the amount of depression of the accelerator pedal is large and the rate (speed) at which the accelerator pedal returns is large as shown in FIG. The movement of the accelerator pedal at can be distinguished by setting a threshold value to the maximum rate or the average rate.

 On the other hand, as shown in FIG. 3 (a), when the amount of depression of the accelerator pedal is small and the rate (speed) when the accelerator pedal returns is low, the movement of the accelerator pedal during normal operation and emergency braking is set as a threshold value. Difficult to distinguish by. This is considered to be because when the amount of depression of the accelerator pedal before applying the emergency brake is small, the speed when the pedal returns is reduced due to the characteristics of the spring, and there is no difference from that during normal operation.

 The present invention employs a method that does not use the threshold value of the speed at which the accelerator pedal returns based on the analysis result described above. Specifically, a classifier that learns the operation pattern of the accelerator pedal when the driver raises his foot by machine learning and classifies the operation pattern when the driver is about to apply an emergency brake and when it is not create.

 Then, the classifier program is incorporated into the emergency brake prediction device installed in the car, and the time series data of the operation pattern of the accelerator pedal of the car newly input to the classifier is based on the result of the judgment by the classifier. When the driver predicts that he or she is about to apply an emergency brake, the brake is automatically activated to stop the car.

  Next, the configuration and operation of the emergency brake prediction device according to the present embodiment will be described with reference to FIGS.

  As shown in FIG. 1, in the present embodiment, time series data indicating the pedal operation pattern detected by the accelerator pedal position sensor 31 of the drive simulator 3 is input to the classifier creating apparatus 2, and the operation pattern is We have created a classifier that classifies when the driver is trying to apply emergency braking and when it is not.

  The drive simulator 3 is a device that simulates driving and driving of a car. The drive simulator 3 is realized by using a computer, a monitor that reflects the scenery seen from the driver's seat, a speaker that emits driving and impact sounds, an accelerator pedal, a brake pedal, and a shift lever Etc.

  The position sensor 31 outputs position data corresponding to the depression amount of the accelerator pedal, and data sampled at a predetermined interval (20 ms in the present embodiment) is output. Although not shown, since an A / D converter is connected to the output stage of the position sensor 31, a digital signal is output from the sensor 31.

  In this embodiment, the data required for creating the classifier is collected using the drive simulator 3 because of the ease of data acquisition. However, the vehicle is actually put on the vehicle, brakes are applied in various driving conditions, The output data of the position sensor 31 may be collected.

  The classifier creating apparatus 2 creates a classifier by machine learning, and includes a data cutout unit 21, a feature extraction unit 22, a learning data storage unit 23, and a learning unit 24. The classifier creating apparatus 2 is realized by using a personal computer, and a program for realizing the functions of the above-described units is stored in the hard disk device.

  The data cutout unit 21 is time-series data indicating the operation pattern of the accelerator pedal from the time-series data output from the position sensor 31, that is, the state before the accelerator pedal is depressed from when the driver's foot leaves the accelerator pedal. The time-series data (data in the sections A and B in FIG. 3) when returning is cut out.

 The feature extraction unit 42 extracts feature quantity data necessary for creating a classifier from the time-series data extracted by the data extraction unit 21. In this embodiment, the position (mP) of the accelerator pedal immediately before the driver lifts his / her foot, the maximum speed (mR) and the average speed (aR) when the accelerator pedal returns are extracted as the feature values. A feature amount calculation method will be described with reference to FIG.

  In FIG. 3A, the position (mP) of the accelerator pedal immediately before the driver lifts his / her foot is the height of the point where the time series data P1 of the accelerator pedal intersects the straight line indicating the start of the region A and the graph. .

  Next, the maximum speed (mR) when the accelerator pedal returns indicates the maximum value among the speeds (ΔP / Δt) of the time series data P1 of the accelerator pedal shown in the area A of FIG. Here, Δt is the sampling interval of the time series data P1, and ΔP is the position data of the accelerator pedal that fluctuates during the period of Δt.

 A graph P2 shown in the lower half of FIG. 3A shows the speed data (ΔP / Δt) of the time series data P1 of the accelerator pedal. Among these data, the maximum absolute value is mR. .

  Next, the average speed (aR) when the accelerator pedal returns is a value given by mP / t0 in FIG. Here, t0 indicates the time from when the driver removes his / her foot from the accelerator pedal to when the driver returns to the state before the accelerator pedal is depressed, and mP indicates the distance the accelerator pedal has moved during that time.

  Returning to the description of FIG. 1, the learning data storage unit 23 stores data of three feature amounts of the individual time-series data P <b> 1 calculated by the feature extraction unit 22.

  The learning unit 24 performs machine learning by reading feature amount data of many operation patterns stored in the learning data storage unit 23. Although a support vector machine or a neural network can be used as a machine learning method, in this embodiment, a classifier is constructed using XGBoost (eXtreme Gradient Boosting) and hyper parameters are adjusted. For this purpose, genetic algorithms were adopted.

  The learning unit 24 extracts the features extracted by the feature extraction unit 22 with respect to the time-series data indicating the operation pattern of the accelerator pedal during emergency braking and the time-series data indicating the operation pattern of the accelerator pedal during normal braking acquired by the position sensor 31. Machine learning is performed using quantity data. Then, the operation pattern when the driver is about to apply the emergency brake is classified as “1”, the operation pattern when the driver is not applied is classified as “0”, and intermediate values are assigned to the other operation patterns.

 The classifier program created in this way is incorporated into the emergency brake prediction device 4 installed in the automobile 1. The emergency brake prediction device 4 includes a CPU that performs arithmetic processing for realizing the functions of each unit of the device, a ROM that stores a program for the arithmetic processing, and a RAM that temporarily stores data necessary for the arithmetic processing. It has a well-known configuration.

 As shown in FIG. 1, the emergency brake prediction device 4 includes a data cutout unit 41, a feature extraction unit 42, a classification unit 43, and a prediction unit 44. As described above, the classifier 43 of the emergency brake prediction device 4 stores the classifier program created by the classifier creation device 2.

  The position sensor 11 mounted on the automobile 1, the data extraction unit 41 and the feature extraction unit 42 of the emergency brake prediction device 4 have the same functions as the position sensor 31, the data extraction unit 21 and the feature extraction unit 22, respectively. Therefore, detailed description is omitted.

 Based on the data of the three feature quantities mP, mR, and aR calculated by the feature extraction unit 42, the classifier 43 has an operation pattern of time series data detected by the positive sensor 11 of class 1 (the emergency brake is about to be applied). Case) or class 0 (when emergency braking is not being applied) is determined, and an intermediate value between 0 and 1 is output.

 The prediction unit 44 instructs the ECU 12 to drive the emergency brake based on the determination result of the classifier 43. A threshold value is set in the prediction unit 44 in advance. For example, when the output of the classifier 43 is 0.5 or more, it is predicted that the driver is going to apply an emergency brake, and the emergency brake is driven to the ECU 12. Instruct. On the other hand, when the output of the classifier 43 is less than 0.5, nothing is instructed to the ECU 12.

  FIG. 2 clearly shows members related to braking among the components mounted on the automobile 1. In addition to the position sensor 11 and the emergency brake prediction device 4 described above, a vehicle speed sensor 13, a shift position sensor 14, a slot valve sensor 15, a slot valve motor 16 and a brake actuator 17 are connected to the ECU 12.

 The ECU 12 performs control related to driving and braking of the automobile. Although not shown, the ECU 12 performs a CPU for performing various arithmetic processes, a ROM that stores a program necessary for the arithmetic processes, and an arithmetic process, similar to the emergency brake prediction device 4. A known configuration such as a RAM for temporarily storing necessary data is provided.

  Various sensors connected to the ECU 12 will be described. The vehicle speed sensor 13 outputs a signal corresponding to the traveling speed of the automobile 1. The shift position sensor 14 outputs a signal indicating the shift position in the transmission. The throttle valve sensor 15 is attached to the intake pipe of the engine and outputs a signal corresponding to the opening of the throttle valve that adjusts the amount of air supplied to the engine.

  On the other hand, a throttle valve motor 16 and a brake actuator 17 driven by a drive signal from the ECU 12 are connected to the ECU 12 as drive members.

 The throttle valve motor 16 adjusts the opening degree by rotationally driving a throttle valve attached to the intake pipe of the engine. The ECU 12 adjusts the opening degree of the throttle valve by driving the throttle valve motor 16 so that the opening degree of the throttle valve becomes an opening degree corresponding to the depression amount of the accelerator pedal by the driver. However, when the emergency brake is applied, the ECU 12 controls the opening degree of the throttle valve so as to close the throttle valve.

 The brake actuator 17 is used to apply an emergency brake, and includes, for example, a pump that discharges a high-pressure brake fluid, and an electromagnetic valve that supplies the wheel cylinder of each wheel while adjusting the pressure of the brake fluid. The brake actuator 17 operates in accordance with a drive signal from the ECU 12, and supplies brake fluid adjusted to the wheel cylinder of each wheel, thereby generating braking force on the wheel.

<Operation of emergency brake prediction device>
Next, the operation of the emergency brake prediction device 4 will be described with reference to FIGS. 1 and 2.

 When the driver is driving the automobile 1, data indicating the amount of depression of the accelerator pedal is input to the emergency brake prediction device 4 from the position sensor 11 at a constant interval (20 ms).

  The data extraction unit 41 of the emergency brake predicting device 4 determines that the accelerator pedal is not released from the time when the driver releases the accelerator pedal when the value of the input time-series data rapidly decreases as shown in FIG. Cut out the time-series data up to the time when it returned to the state before it was stepped on.

  For the time series data indicating the accelerator pedal operation pattern output from the data cutout unit 41, the feature extraction unit 42 calculates the above-described three feature amounts mP, mR, and aR in real time.

 The data of the three feature quantities mP, mR, and aR calculated by the feature extraction unit 42 are input to the classifier 43, and based on the learning result, the operation pattern is class 1 (when an emergency brake is applied) or class It is determined whether it is close to 0 (when emergency braking is not being applied).

  The output of the classifier 43 is input to the prediction unit 44. As described above, a threshold is set in the prediction unit 44 in advance, and when the output of the classifier 43 is 0.5 or more, the ECU 12 is instructed to drive the emergency brake.

  In FIG. 2, when the prediction unit 44 instructs the ECU 12 to drive the emergency brake, the ECU 12 checks the depression amount of the accelerator pedal output from the position sensor 11 in order to confirm the vehicle state when the emergency brake is operated. The data and throttle valve opening data output from the throttle valve sensor 13 are stored. In addition, the ECU 12 drives the brake actuator 17 to operate the emergency brake.

  Thereafter, the ECU 12 determines whether or not the automobile has stopped based on the output of the vehicle speed sensor 13. When it is determined that the automobile has stopped, it is determined based on the output of the shift position sensor 14 whether or not the shift position of the transmission has been switched to parking.

 When the shift position is switched to parking, it is considered that the vehicle is completely stopped and safety is secured, and the control of the throttle valve motor 16 and the brake actuator 17 is switched to the normal operation mode.

 FIG. 7A shows a change in the amount of depression of the brake pedal when the ECU 12 operates the brake actuator 17 in accordance with the determination of the emergency brake prediction device 4. For comparison, FIG. 7B shows a change in the depression amount of the brake pedal when the driver applies an emergency brake with his / her foot. In each graph, the operation pattern of the accelerator pedal is the same.

  As shown in FIG. 7 (a), when the brake actuator 17 is operated, the brake actuator 17 is operated and the brake pedal is depressed to the maximum position almost simultaneously with the return to the state before the accelerator pedal is depressed. .

 On the other hand, as shown in FIG. 7B, when the driver applies an emergency brake with his / her foot, it takes time to switch from the accelerator pedal to the brake pedal and to depress the brake pedal to the maximum position. Therefore, a delay of about 0.4 seconds occurs compared to the case where the brake actuator 17 is operated.

 When the vehicle is driving at 50 km / h, the braking distance is about 5.6 m when the driver applies the emergency brake with his / her foot compared to the case where the brake actuator 17 is operated according to the instruction of the emergency brake prediction device 4. Because of the difference, there is a big difference in ensuring safety.

 In addition, as shown to Fig.7 (a), when a brake pedal is stepped on suddenly, a tire may lock | rock and slip on a road surface. Once the tires lock and begin to slip, the steering wheel operation becomes ineffective and control becomes impossible, as well as being exposed to the risk of serious accidents such as rear-end collisions, side slips, and rollovers.

 In order to prevent this, it is necessary to add an anti-lock brake system (ABS) to the ECU 12 to prevent the tire from locking. However, since ABS is a technology that is generally adopted, the explanation will be given. Omitted.

  FIG. 8 shows the result of determining the operation pattern of the accelerator pedal for each of the case where the driver is going to apply emergency braking and the case where normal braking is applied using the classifier created in the present embodiment. Indicates.

 In the figure, the horizontal axis is a number indicating the determination result of the classifier, class 1 when the driver determines that the emergency brake is about to be applied, and class 0 when the driver determines that the emergency brake is not about to be applied, About the sample data of 269, the possibility of belonging to class 1 is indicated by an intermediate value.

  As for sample data, it is preferable to collect time-series data when emergency braking is applied during actual driving, but since there is danger in applying emergency braking during driving, a drive simulator is used in the experiment. Time series data was collected by emergency braking in various driving situations.

  The performance of recent drive simulators is very high, and it is possible to realize a state close to the actual driving state, so even with time series data showing the operation pattern of the accelerator pedal when trying to apply an emergency brake, the car is actually driven You can get data that is inferior to what you are doing.

  In FIG. 8, for the time series data when the threshold is set to 0.5 and the driver is about to apply emergency braking, the probability that the prediction unit 44 predicted that the emergency braking will be applied is 90%. It was. This value can sufficiently exhibit the required performance from the viewpoint of ensuring safety even if the classifier program created in the present embodiment is incorporated in the emergency brake prediction device 4 of an automobile and operated in actual driving. It is.

(Embodiment 2)
In the first embodiment, since it is easy to obtain the learning data, the time series data indicating the operation pattern of the accelerator pedal is collected using the position sensor mounted on the drive simulator. On the other hand, in the present embodiment, a classifier creating apparatus is mounted on a vehicle in order to create a classifier corresponding to the driver's own operating characteristics.

  FIG. 9 shows a configuration when the classification device creation device is mounted on an automobile together with the emergency brake prediction device. In the present embodiment, the data extraction unit and the feature extraction unit of the classifier creation device 2a are shared by the data extraction unit 41 and the feature extraction unit 42 of the emergency brake prediction device 4.

  Therefore, when collecting time-series data for calculating the three feature values mP, mR, and aR, which are learning data, the driver actually drives the car in various driving situations while considering safety. Repeat normal brake operation and emergency brake operation.

  The feature amount data calculated from the time-series data collected in this manner is stored in the learning data storage unit 23, and the learning unit 24 uses the data, as described in the first embodiment. Machine learning is performed in the form of, and a classifier is created. Then, the created classifier program is incorporated into the emergency brake prediction device 4 and used to predict whether or not the emergency brake is to be applied.

  As described above, when the driver actually drives the vehicle and collects time-series data indicating the operation pattern of the accelerator pedal, sufficient consideration is necessary for ensuring safety, and data collection takes time. On the other hand, since it is possible to create a classifier that reflects the individual driving characteristics of the driver, it is possible to improve the prediction accuracy of whether or not the driver is about to apply an emergency brake.

  In each of the above-described embodiments, the spring force is used to return to the state before the accelerator pedal is depressed. However, the motor may be used to return to the state before the accelerator pedal is depressed. Good. In this case, the motor generates a force that restores the state before the driver stepped on, with a weaker force than the driver's stepping on the accelerator pedal.

DESCRIPTION OF SYMBOLS 1 Car 2, 2a Classifier preparation apparatus 3 Drive simulator 4 Emergency brake prediction apparatus 11, 31 Position sensor 12 ECU
DESCRIPTION OF SYMBOLS 13 Vehicle speed sensor 14 Shift position sensor 15 Throttle valve sensor 16 Throttle valve motor 17 Brake actuator 21, 41 Data extraction part 22, 42 Feature extraction part 23 Learning data storage part 24 Learning part 43 Classifier 44 Prediction part

Claims (9)

It is created by performing machine learning using time series data indicating the operation pattern of the accelerator pedal when the driver raises his / her foot from the accelerator pedal, detected by the accelerator pedal position sensor mounted on the drive simulator or automobile, A classifier for classifying the operation pattern into a case where the driver is about to apply emergency braking and a case where the driver is not applying emergency braking;
Predicted that the driver is going to apply emergency braking based on the result of judgment by the classifier using time-series data indicating the operation pattern of the accelerator pedal detected by the accelerator pedal position sensor mounted on the vehicle as input. Then, an emergency brake prediction device for an automobile, which instructs the control device of the automobile to drive an emergency brake.   The emergency brake prediction device according to claim 1, wherein the time series data is created by sampling an output signal of the accelerator pedal position sensor at a constant interval.   A data cut-out unit that cuts out data from the time series data until the driver returns to the state before the accelerator pedal is depressed from the time when the driver releases the accelerator pedal, and the time-series data cut out by the data cut-out unit The automobile emergency brake prediction device according to claim 1, further comprising a feature extraction unit that extracts a plurality of feature amounts from the feature extraction unit.   The learning data storage unit further stores the feature amount of the time series data extracted by the feature extraction unit, and the classifier includes feature amounts of a plurality of time series data read from the learning data storage unit. The emergency brake prediction apparatus for automobiles according to claim 3, which is created using the data.   The automobile emergency brake prediction device according to claim 3 or 4, wherein the plurality of feature amounts are a position of an accelerator pedal immediately before the driver releases his / her foot, a maximum speed and an average speed when the accelerator pedal returns.   The vehicle control device, when instructed to perform emergency braking, operates a brake actuator to supply a brake fluid pressure-adjusted to the wheel of the vehicle, thereby causing the wheel to exert a braking force. The automobile emergency brake prediction apparatus according to any one of 1 to 5. A data collection step for collecting time-series data indicating an operation pattern of the accelerator pedal when the driver lifts his / her foot from the accelerator pedal using an accelerator pedal position sensor mounted on the drive simulator or the vehicle;
Machine learning using the time series data, and a classifier creating step for creating a classifier that classifies the operation pattern into a case where the driver is trying to apply emergency braking and a case where the driver is not applying emergency braking, and
A step of incorporating the classifier created in the classifier creating step into an emergency brake prediction device mounted on a vehicle;
Based on the result of determination by the classifier incorporated in the emergency brake prediction device, time series data indicating the operation pattern of the accelerator pedal detected by the accelerator pedal position sensor mounted on the automobile is input to the emergency brake prediction device. An emergency brake prediction method for an automobile, comprising: an emergency brake prediction step for instructing the controller of the automobile to drive an emergency brake when the driver predicts that an emergency brake is to be applied.   8. The automobile emergency brake prediction method according to claim 7, wherein the time-series data is created by sampling an output signal of the accelerator pedal position sensor at a constant interval.   The classifier creation step includes a data extraction step of extracting data from the time series data until the driver returns to the state before the accelerator pedal is depressed from when the foot is released from the accelerator pedal; and the data extraction step The automobile emergency brake prediction method according to claim 7 or 8, further comprising a feature extraction step of extracting a plurality of feature amounts serving as learning data from the time-series data cut out in step (b).

Images