JP2010198120A - Driving condition determination device and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a technique for determining the driving conditions of a driver with a few data to be acquired in a short period of time. <P>SOLUTION: A driving condition determination device is configured to calculate the featured quantities of the driving characteristics of a driver as the information of the predicted error distribution of a driving operation based on driving operation information, and to acquire a driving characteristic model after learning based on a driving characteristic model comprising the calculated featured quantities of the driving conditions and the information of the predicted error distribution of the driving conditions becoming a preset reference, and to determine whether or not the current driving conditions are unstable driving conditions different from normal driving characteristics based on the acquired driving characteristic model after learning and the featured quantities of the current driving characteristics. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a technique for determining a driving state of a driver for driving support of the driver.

  As a technique for supporting the driving of the driver, there is a technique described in Patent Document 1, for example. In this technology, the traveling pattern stored for each driver is compared with the latest traveling state to determine whether the traveling state is normal or abnormal. When it is determined that there is an abnormality, an alarm is issued. Here, the correction value is updated based on the driver's operation and travel information acquired sequentially. Then, a travel map in which correction values that are sequentially updated are added to the basic map is used as the travel pattern for determination.

Japanese Patent Laying-Open No. 2005-301832 (see paragraphs 0019 and 0020)

In the conventional technology as described above, it is necessary to learn a required time before acquiring a driving pattern or driver characteristics that are more than practical accuracy. This is because short-time data greatly affects the driving environment and becomes information with a lot of noise. For this reason, it takes time until the operating state can be determined.
The present invention has been made paying attention to the above points, and it is an object of the present invention to provide a driving state determination technique capable of determining a driving state of a driver with a small amount of data obtained in a short time. .

  In order to solve the above-described problem, the present invention calculates a feature amount of a driving characteristic of a driver as information on a prediction error distribution of the driving operation based on the driving operation information. A learned driving characteristic model is acquired based on the driving characteristic model including the calculated characteristic value of the driving characteristic and information on the prediction error distribution of the driving operation as a reference set in advance. Then, based on the acquired driving characteristic model after learning and the characteristic amount of the current driving characteristic, it is determined whether or not the current driving state is an unstable driving state different from the normal driving characteristic.

According to the present invention, the information on the prediction error distribution of the driving operation is used as the feature amount of the driving characteristic of the driver. For this reason, learning time can be shortened. In addition, the learning time can be further shortened by providing a driving characteristic model including information on the prediction error distribution of the driving operation as a reference in advance.
From the above, it is possible to determine the driving state of the driver with a small amount of data obtained in a short time.

It is a figure which shows the system configuration | structure which concerns on 1st Embodiment based on this invention. It is a figure which shows the structure of the information presentation controller which concerns on 1st Embodiment based on this invention. It is a figure which shows a driving | running characteristic model. It is a figure which shows a driving | running characteristic model. It is a figure explaining the process of the information presentation controller which concerns on 1st Embodiment based on this invention. It is a figure explaining the calculation process of the unstable index which concerns on 1st Embodiment based on this invention. It is a figure explaining the calculation process of the operation error distribution which concerns on 1st Embodiment based on this invention. It is a figure explaining the calculation process of the driving characteristic feature-value based on 1st Embodiment based on this invention. It is a figure explaining prediction error distribution. It is a figure explaining prediction error distribution. It is a figure which shows frequency distribution of prediction error. It is a figure explaining the process of the information presentation controller which concerns on 2nd Embodiment based on this invention. It is a figure explaining the calculation process of the driving characteristic feature-value based on 3rd Embodiment based on this invention. It is a figure explaining a correction coefficient. It is a figure explaining a correction coefficient.

Next, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a system configuration diagram of a vehicle including an information presentation device including a driving state determination device according to the present embodiment.
In the present embodiment, a case where a steering operation is targeted as a driving operation will be described as an example. Then, the steering angle is used as driving operation information.

(Constitution)
The vehicle according to the present embodiment controls the vehicle behavior in a state corresponding to the driver's operation, like the ordinary vehicle. That is, a drive source for driving drive wheels such as an engine and a motor is controlled so as to generate a driving force corresponding to the accelerator opening of the accelerator pedal operated by the driver. Further, the braking device for each wheel is controlled so that a braking force according to the brake opening degree of the brake pedal operated by the driver is generated. Further, the steering device steers the steered wheels according to the steering angle of the steering wheel operated by the driver.

An information presentation device including a driving state determination device is provided for the vehicle as described above. Hereinafter, the information presentation apparatus will be described.
The steering angle sensor 1 detects the steering angle of the steering wheel. The steering angle sensor 1 outputs the detected steering angle to the information presentation controller 2.
The information presentation controller 2 learns the driver's normal driving state based on the operation error distribution information based on the steering angle. When the learning is completed, it is determined whether or not the current driving state is an unstable driving state based on the learned normal driving state. And if the information presentation controller 2 determines with the present driving | running state being an unstable driving | running state, it will alert | report that to a driver | operator. That is, the information presentation controller 2 estimates the driver's (ordinary) driving characteristics from information on driving behavior in a short time. And by making this estimation result into the initial value after learning, an unstable state is determined early and information is presented. Further, sequential learning is performed after the identification of model parameters (after learning is completed).

The information presentation controller 2 includes a driving state determination device 2A and information transmission means 2B. The driving state determination device 2A includes a driving characteristic model 2Aa serving as a reference, a driving characteristic feature amount calculation unit 2Ab, a model acquisition unit 2Ac, and an unstable driving state determination unit 2Ad.
The reference driving information model includes information on a prediction error distribution of a driving operation that is set in advance as a reference. This information may be created based on driving operation information of a plurality of persons acquired through experiments or the like. Examples thereof are shown in FIGS.

The driving characteristic feature amount calculation means 2Ab calculates the characteristic amount of the driving characteristic of the driver as information on the prediction error distribution of the driving operation based on the driving operation information acquired by the means for acquiring the driving operation information.
The model acquisition unit 2Ac acquires the learned driving characteristic model M based on the driving characteristic feature amount calculated by the driving characteristic feature amount calculating unit 2Ab and the preset driving characteristic model 2Aa.

The unstable driving state determination means 2Ad is based on the learned driving characteristic model M acquired by the model acquisition means 2Ac and the current characteristic amount calculated by the driving characteristic feature amount calculation means 2Ab. It is determined whether or not the unstable operation state is different from the operation characteristic of.
The information transmission unit 2B outputs the driving state information corresponding to the determination of the unstable driving state determination unit 2Ad to the information presentation unit 3.
Then, the information presenting means 3 presents information corresponding to the driving state information from the information transmitting means 2B to the driver. The information presenting means 3 is configured by, for example, a display unit such as a navigation, a sound generation device, a vibration device that vibrates the seat, or the like.

Next, the process of the information presentation controller 2 will be described with reference to FIG.
This process operates every predetermined sampling period.
First, in step S100, driving operation information is acquired. In the present embodiment, the steering angle is acquired based on the input from the steering angle sensor 1.
In step S200, the acquired steering angle is recorded as a driving action. This record is used to estimate the current driving behavior / driving state from the statistical values / distribution of past data and the latest driving behavior in order to accurately estimate the driver's state.
In step S300, an unstable operation index is calculated. In this embodiment, an unstable driving index is calculated using a steering entropy method. The calculation of the unstable operation index will be described later. As the unstable driving index, a normal unstable driving index for learning and a current (at this time) unstable driving index are calculated separately.

Here, the steering entropy method will be described.
In general, in a state where the driver's attention is not concentrated on driving, the time during which steering is not performed becomes longer than in normal driving where the driving is concentrated, and a large steering angle error is accumulated. Therefore, the corrected steering amount when the driver's attention returns to driving increases. The steering entropy method focuses on this characteristic, and uses an α value as a characteristic value and a steering angle entropy Hp calculated based on the α value. By comparing the reference steering angle entropy with the steering angle entropy calculated based on the measured steering angle, an unstable state of the driving operation with respect to the reference is detected.

Note that the α value as the characteristic value is a 90% tile value (90% of the steering error) by calculating a steering error within a predetermined time based on the time-series data of the steering angle and measuring the distribution (variation) of the steering error. The range of the distribution in which is included. The steering error is a difference between an estimated value of the steering angle when it is assumed that the steering is operated smoothly and an actual steering angle.
The Hp value as the steering entropy value represents the ambiguity (uncertainty) of the steering error distribution. Similar to the α value, the Hp value decreases when the steering operation is smooth and stable, and increases when it is unstable and unstable. By correcting the Hp value by the α value, it can be used as a driver instability that is not easily affected by the skill or habit of the driver.

Since the steering error is affected by the road alignment and the load state of the driver, it is necessary to measure a reference steering error distribution when the driver is in an unloaded state. Therefore, a steering error distribution is obtained based on the steering angle data measured for a long time.
Next, in step S400, the driving characteristic feature amount is calculated based on the usual unstable driving index for learning. The calculation of the driving characteristic feature amount will be described later.

  Next, in step S500, it is determined whether learning is completed. That is, it is determined whether or not parameter identification has been completed. In the driver characteristic estimation process, if the model is adapted and an initial value is entered, the learning is completed and the process proceeds to step S1410. On the other hand, if it does not match, the process is terminated and the process returns. Whether or not the learning is completed is, for example, counting how many times the calculation process of the driving characteristic feature amount is performed, and when the count exceeds a predetermined value, it is determined that the learning is completed. If the learning time for “ordinary” is 2160 seconds and the sampling number is 5 Hz, the number of data K is 10800.

In step S600, the relative entropy Hp is calculated.
Based on the learned prediction error distribution (pl _i ) and the current prediction error distribution (ps _i ), the relative entropy R_Hpls is calculated by the following equation. Here, the detection of the unstable state is a comparison at this time (most recent) with respect to the usual.

In step S700, it is determined whether the state is unstable. Here, if the relative Hp (R−Hp) exceeds a predetermined threshold (for example, threshold = 0.5), the process proceeds to step S800. On the other hand, when it does not exceed, the process is terminated and the process returns.
In step S800, unstable state information is presented. For information presentation, information presentation means 3 is used to present information from vision, voice, or the like.
Next, the unstable operation index calculation process performed in step S300 will be described with reference to FIG.
First, in step S310, the prediction error PE is calculated.
The prediction error PE is obtained as a difference between the predicted steering angle θp and the actual steering angle θ as in the following equation.
PE = θ−θp

The predicted steering angle θp is a steering angle that is estimated to be obtained by smooth steering based on the past steering angle. For example, when the current time point is n, secondary Taylor expansion around the n-1 time point using the past three steering angles of the last three points (n-3, n-2, n-1). calculate. Specifically, the predicted steering angle θp is calculated from the steering angles θa1, θa2, and θa3 every 150 ms (150 ms before, 300 ms before, and 450 ms before) by the following formula.
θp = θa1 + (θa1-θa2)
+ (1/2) (((θa1-θa2)-(θa2-θa3))
Next, in step S320, two prediction error distributions are updated based on the calculated predicted steering angle. The two prediction error distributions are a normal prediction error distribution and a current (at this time) prediction error distribution. Here, the prediction error distribution is represented by a frequency distribution.

Each distribution ratio pl i of the two prediction error distribution _for updating ps _i, be described with reference to FIG.
Here, the two prediction error distributions are preset as follows.
Each time window for updating the two prediction error distributions is set. The usual time window is set to 2160 seconds (predetermined value) as Nl. “The current time window is set to 60 seconds (predetermined value) as Ns. Further, the value α for determining the bin interval of the prediction error distribution is set to the predetermined value 0.64.

In step S321, the width (B) of each bin of the two prediction error distributions is set based on α. The bin width is calculated by multiplying bin1 to bin9 by α as in the following equation. bin1 to bin9 are set to [-5, -2.5, -1, -0.5, 0.5, 1, 2.5, 5], respectively.
B1 = α × bin1
...
B9 = α × bin9

Next, in step S322, it is determined in which bin range the current prediction error PE calculated in step S310 falls. For example, when α is 0.64, if the currently calculated steering angle prediction error PE is 0.2, the bin of p5 (−0.32 to 0.32) is determined to be within the range.
If it is determined in step S322 that it is out of range, the process proceeds to step S323. On the other hand, if it is determined to be within the range, the process proceeds to step 324. Then, in the process of subsequent steps S323, S324, the distribution ratio of the current steering angle prediction error distribution _ps i {i = 1~9}, the distribution ratio of the usual steering angle prediction error distribution _pl i {i = _1~9 } Is updated respectively. ps _i and p _{i i} indicate the bin ratio of the prediction error distribution.

Here, in the following process, in fact distribution ratio _ps i, is performed individually update processing for pl _i, since the same process, a representative as a distribution ratio _{pi (: ps i, pl i} ) will be described as .
In step S323, the distribution ratio pi is obtained by the following equation. That is, the bin value is divided by (1 + 1 / N). That is, the probability density decreases. Thereafter, the process proceeds to step S325.
p _i = p _i / (1 + 1 / N)
Here, N is the sample equivalent number (Ns, Nl) corresponding to the time window of the prediction error distribution. p _i is a distribution ratio (ps _i {i = 1 to 9}, pl _i {i = 1 to 9}) corresponding to the time window of the prediction error distribution.

In step S324, the distribution ratio pi is obtained by the following equation. That is, (1 / N) is added to the bin value and divided by (1 + 1 / N). That is, the probability density increases. Thereafter, the process proceeds to step S325.
p _i = (p _i + 1 / N) / (1 + 1 / N)
Here, N is the sample equivalent number (Ns, Nl) corresponding to the time window of the prediction error distribution. p _i is a distribution ratio (ps _i {i = 1 to 9}, pl _i {i = 1 to 9}) corresponding to the time window of the prediction error distribution.

In step S325, the bin is advanced by one. That is, i is incremented.
In step S326, it is determined whether all nine bins have been updated (i = 10).
If all nine bins have been updated (i = 10), the process ends and returns. On the other hand, when all the bins have not been updated, the process returns to step S322.
This completes the process of the unstable index calculation unit.

Next, driving characteristic feature amount calculation processing will be described with reference to FIG.
First, in step S410, the size of the prediction error PE (hereinafter referred to as PEdr) within a predetermined range in the normal prediction error distribution is obtained.
The predetermined range is within a certain part (percentile (% tile value) or within a predetermined value of a normalized range. For example, 90% tile is adopted.
Next, in step S420, it is determined whether the data amount for estimating the driver characteristics is satisfied. The amount of data refers to the amount of data for which driver characteristics can be estimated within a short time. If the amount of data is not satisfied, it is not necessary to identify the parameter, so the driver characteristic estimation process is terminated and the process returns. On the other hand, when the data amount is satisfied, the process proceeds to step S430 in order to identify the parameter.

Next, in step S430, the driving characteristic model is updated.
Here, the basic operation characteristic model used as a reference is set in advance as described above. The basic operating characteristic model used as a reference is set in the state of the prediction error distribution, and the degree of the prediction error distribution is set through experiments or the like. Specifically, as information on the prediction error distribution of the basic operating characteristic model, the PE size within the predetermined range (hereinafter referred to as PE basic) and the horizontal axis of the prediction error distribution (BIN _i {i = 1 to 9}) And the ratio of the vertical axis (DIST _i {i = 1 to 9}). Here, the number of bins is set to nine here, but is not limited to this value.
In step S440, a ratio between the basic operating characteristic model PEbasic and the PEdr calculated in step S410 is obtained. Further, by multiplying the horizontal axis BIN prediction error distribution in the ratio, as shown in Equation obtains a prediction error distribution of the learned driving characteristics model _{pl i = BIN_MODEL i {i =} 1~9}.
That is,
BIN_MODE _i = (PEdr / PEbasic) × BIN _i
Through this step, the predicted error distribution pl _i of the learned driving characteristic model is
[X, y] = [BIN_MODE _i , DIST _i ]
Can be expressed.

(Operation / Action)
A technique such as the prior art requires a long learning period (for example, 4 to 5 days), and information can be presented after the learning period ends. On the other hand, in the learning of the driving state in the present embodiment, it is possible to collect driving characteristics with a predetermined accuracy in a short period of time by using the error distribution of the driving operation. That is, information can be presented in a short time by estimating the driver characteristics and identifying parameters close to the driver characteristics based on the estimation results.

Here, by comparing the driving behavior indexes of “normal (long time)” and “currently (current)” of each driver, a different driving state is detected and information is presented. Further, the steering entropy method is used to calculate a wobbling driving index from two prediction error (PE) distributions of “normal” and “currently at this time”, and detect the wobbling driving state.
In general, long-term data is required to obtain “normal” driving behavior data. However, it is difficult to present information effectively until long-term data can be acquired. The reason why long time data is required is that long time data is less affected by noise.

Here, as a result of investigation and examination by the present inventors through experiments, etc., it was found that the influence of noise affects the outside of a certain width of the prediction error distribution of driving operation, and that the influence of noise is hardly affected by the inside of the width. It was.
Furthermore, we found that the difference in conditions such as subject differences and driving environment differences in PE distribution with little noise influence (for long-term data) does not change the distribution proportions only by a certain specific width. .
From the above, by normalizing with a certain representative value, almost the same distribution proportion can be derived. Therefore, if the representative value of the distribution is known, the distribution itself can be estimated.

The above explanation will be made that “the influence of noise affects the outside of a certain width and is less susceptible to noise inside the width”.
FIG. 9 and FIG. 10 show the PE distribution of the subject's 10-minute running data and the PE distribution for 5 trips (approximately 450 minutes), respectively. As can be seen from this figure, although the standard deviations that are easily affected by values away from the median value are different, the difference between the two is small when viewed in width (90% tile width: 5 to 95% tile). .
It can be seen that a prediction error distribution with little influence of noise can be obtained by calculating the size of a specific width using this characteristic and constructing a model having the same width.

Next, explanation will be made on the above-mentioned “difference in difference in conditions such as differences in subjects and driving environments is that the proportion of distribution does not change only by a certain specific width”.
FIG. 11 shows the normalized PE frequency distribution and the normal PE frequency distribution. FIG. 11 shows the distribution of PEs of 12 subjects (distributed distributions of a plurality of subjects are displayed). The upper figure shows the frequency distribution of PE normalized by a representative value (90% tile) of a certain distribution, and the lower figure shows the usual frequency distribution. As can be seen from FIG. 5, it can be seen that the normalized frequency distribution has a smaller vertical variation (difference between subjects is smaller). Furthermore, it turns out that there is no big difference in the proportion of each subject.

Using this characteristic, different distributions can be estimated if a specific width is known.
That is, if a specific width is obtained, a distribution with less noise can be estimated.
In this embodiment, the learning period is shortened by using the error distribution of the driving operation, and the learning period can be further shortened by having a distribution model as an initial value in advance.
Here, the steering angle sensor 1 constitutes driving operation information acquisition means. Step S300 constitutes the driving characteristic feature amount calculating means 2Ab. Step S400 constitutes model acquisition means 2Ac. Steps S600 and S700 constitute unstable operation state determination means 2Ad.

(Effect of this embodiment)
(1) The driving operation information acquisition means acquires driving operation information by the driver. The driving characteristic feature amount calculating unit 2Ab calculates the characteristic amount of the driving characteristic of the driver as information on the prediction error distribution of the driving operation based on the driving operation information acquired by the driving operation information acquiring unit. The model acquisition unit 2Ac is based on the driving characteristic model calculated from the driving characteristic feature amount calculated by the driving characteristic feature amount calculating unit 2Ab and the driving error prediction error distribution information serving as a preset reference. Obtain an operating characteristic model. The unstable driving state determination means 2Ad is based on the learned driving characteristic model acquired by the model acquisition means 2Ac and the current characteristic amount calculated by the driving characteristic feature amount calculation means 2Ab. It is determined whether or not an unstable operation state different from the operation characteristics.

Information on the prediction error distribution of the driving operation is used as a feature amount of the driving characteristic of the driver. For this reason, learning time can be shortened. In addition, the learning time can be further shortened by providing a driving characteristic model including information on the prediction error distribution of the driving operation as a reference in advance.
From the above, it is possible to determine the driving state of the driver with a small amount of data obtained in a short time.

(2) The characteristic amount of the driving characteristic is information related to the steering angle prediction error distribution of the driving operation.
It becomes possible to determine the driving state by the steering operation of the driver.
(3) The model acquisition unit 2Ac acquires a learned driving characteristic model using a prediction error having a predetermined cumulative frequency in the prediction error distribution of the driving operation as a characteristic amount of the driving characteristic.
By using a prediction error having a predetermined cumulative frequency as a feature amount of driving characteristics, learning time can be performed in a shorter time.

(Modification)
(1) In the above embodiment, when calculating the characteristic value of the driving characteristic in step S400, the size of the PE within the predetermined range is calculated (step S410). Instead of this, the cumulative frequency in a predetermined PE may be calculated when calculating the driving characteristic feature amount. In this case, the cumulative frequency of the predetermined PE may be calculated in step S410, and the driving characteristic feature amount may be calculated based on the cumulative frequency.
That is, the model acquisition unit 2Ac acquires the learned driving characteristic model using the cumulative frequency of the predetermined prediction error value in the driving operation prediction error distribution as the characteristic amount of the driving characteristic.
The effect is the same as in the above embodiment, and the operation state can be determined in a short time.
(2) The driving operation may be operation information of an operator such as an accelerator pedal or a brake pedal.

(Second Embodiment)
Next, a second embodiment will be described with reference to the drawings. In addition, the same code | symbol is attached | subjected and demonstrated about the structure similar to the said 1st Embodiment.
The basic configuration of the present invention is the same as that of the above embodiment.
However, the conditions for performing the driver estimation are limited and aiming for more accurate driving characteristic feature amount calculation.
The vehicle speed sensor 430 detects the vehicle speed of the host vehicle by measuring the number of rotations of the wheels and the number of rotations on the output side of the transmission, and outputs the detected host vehicle speed to the information presentation controller 2.

Processing of the information presentation controller 2 of the present embodiment will be described with reference to FIG.
In the present embodiment, step S1000 is added between the process of step S300 and the process of step S400. Since other processes other than step S1000 are the same as those in the first embodiment, description thereof will be omitted.
In step S1000, it is determined whether or not a feature amount calculation condition is satisfied. If the feature amount calculation condition is satisfied, the process proceeds to step S400. If the feature amount calculation condition is not satisfied, the process ends and returns.

The feature amount calculation condition is a condition that allows the driver characteristics to be estimated with a predetermined accuracy or higher. The feature amount calculation condition is, for example, a case where the vehicle speed is 40 km / h or more and the curve ratio (described later) of the traveling road is 10% or less. The curve ratio is obtained, for example, by determining that the vehicle is traveling on a curve when the steering angle is a predetermined value (for example, 12 degrees) or more and integrating the time ratio.
In addition, what is necessary is just to implement the data to be used with reference to the driving action data record recorded in step S200.

(Operation / Action)
Since the steering error is affected by the road alignment and the load state of the driver, it is preferable to measure the steering error distribution for learning while the driver is in no load state. Therefore, in order to make it less susceptible to these effects, the steering error distribution of the model is updated and learned based on the steering angle data measured based on the steering angle data during road linear traveling that can be regarded as a straight road.
Here, the vehicle speed sensor 4 constitutes a travel information acquisition unit.

(Effect of this embodiment)
(1) The travel information acquisition means acquires the travel information of the vehicle. The model acquisition means 2Ac is the driving characteristic when the driving characteristic characteristic amount calculated by the driving characteristic characteristic calculation means 2Ab is determined to be a predetermined driving state based on the vehicle driving information acquired by the driving information acquisition means. A feature model is used to obtain a driving characteristic model after learning.
When learning the model, the accuracy of the model is improved by using data that is not easily affected by the road alignment and the load state of the driver.

(Third embodiment)
Next, a third embodiment will be described with reference to the drawings. In addition, about the structure similar to said each embodiment, the same code | symbol is attached | subjected and demonstrated.
The basic configuration of the present invention is the same as that of the above embodiment.
However, a part of the calculation process of the driving characteristic feature amount is different.
That is, the feature amount is corrected from the driving state attribute for which the driving characteristic feature amount is calculated, and higher driver characteristic estimation is aimed.
The vehicle speed sensor 430 detects the vehicle speed of the host vehicle by measuring the number of rotations of the wheels and the number of rotations on the output side of the transmission, and outputs the detected host vehicle speed to the information presentation controller 2.

The calculation process of the driving characteristic feature amount of the present embodiment will be described with reference to FIG. The difference is that the processing of step S415 and step S425 is added.
First, in step S410, the size of PE within a predetermined range (hereinafter referred to as PEdr) is obtained.
Next, in step S415, the running state attribute is recorded.
The attributes are recorded, for example: • Whether the vehicle is traveling on a curve or traveling in a straight line • The type of traveling road is recorded. The attribute of the traveling state is an attribute related to the magnitude of the steering amount that is normally required by the driver.

  The type of running state may be determined from navigation information or the like. The navigation system includes a GPS receiver, a map database, a display monitor, and the like, and performs route search, route guidance, and the like. The navigation system can acquire information such as the type of road on which the host vehicle is running and the road width based on the current position of the host vehicle obtained from the GPS receiver and the road information stored in the map database.

  Next, in step S420, it is determined whether the data amount for estimating the driver characteristics is satisfied. The amount of data refers to the amount of data for which driver characteristics can be estimated within a short time. If the amount of data is not satisfied, it is not necessary to identify the parameter, so the driver characteristic estimation process is terminated and the process returns. On the other hand, when the data amount is satisfied, the process proceeds to step S430 in order to identify the parameter.

In step S430, PEdr calculated in step S410 is corrected based on the following equation.
PEdr_modify = PEdr + ΣPE_mod
Also, ΣPE_mod = Curve_mod + LoadIDX_mod
Here, Curve_mod indicates the ratio of the curve, and is obtained using a map as shown in FIG. 14 based on the ratio of the curve at the target learning period. That is, the larger the curve ratio, the smaller. RoadIDX_mod indicates the ratio of the expressway, and is obtained using a map as shown in FIG. 15 based on the ratio of the expressway in the target learning period. That is, the larger the ratio of highways, the larger the model. Next, in step S430, a model is created.
Next, in step S440, the model is adapted as an initial value. The prediction error distribution created as a model is substituted as an initial value, and parameter identification is completed.

(Operation / Action)
Steering error is affected by road alignment and driver load conditions. The prediction error distribution for specifying the model is corrected by the amount affected by this influence.
Here, the vehicle speed sensor 4 constitutes a travel information acquisition unit. Steps S415 and S425 constitute driving characteristic feature amount correcting means.

(Effect of this embodiment)
(1) The characteristic feature amount correcting unit corrects the characteristic amount of the driving characteristic based on the driving state attribute. The attribute of the driving state is an attribute related to the magnitude of the operation amount of the driver that is normally requested.
This improves the accuracy of the model.
(2) The attribute of the traveling state is a road type or a road shape.
As a result, attributes that affect the steering state can be used.

(3) The attribute of the traveling state is whether or not the road shape is a curve. The driving characteristic feature amount correcting unit corrects the driving characteristic feature amount according to a time ratio in which the attribute is a curve road having a predetermined value or more.
As a result, the influence of the curve running on the model can be reduced.
(4) The attribute of the traveling state is a road type. The driving characteristic feature amount correcting unit corrects the driving characteristic feature amount using a time ratio for each road type.
As a result, the influence of the road type on the model can be reduced.

(Modification)
(1) The travel state to be acquired is not limited to the travel state. Any traveling state related to the requested steering amount can be used.
(2) In the above description, a highway is exemplified as the road type, but other road types such as a gravel road may be used. What is necessary is just to obtain | require a correction amount according to the load amount with respect to the steering by driving | running | working of the object road.

1 Steering angle sensor (Driving operation information acquisition means)
2 Information presentation controller 2A Driving state determination device 2Aa Driving characteristic model 2Ab Driving characteristic feature amount calculation means 2Ac Model acquisition means 2Ad Unstable driving state determination means 2B Information transmission means 3 Information presentation means 4 Vehicle speed sensor (running information acquisition means)
PE prediction error _{p i} distribution ratio pl _i distribution ratio ps _i distribution ratio R_Hpls relative entropy θ steering angle θp prediction steering angle

Claims (10)

Driving operation information acquisition means for acquiring driving operation information by the driver;
Based on the driving operation information acquired by the driving operation information acquiring means, driving characteristic feature amount calculating means for calculating the characteristic amount of the driving characteristic of the driver as information of the prediction error distribution of the driving operation;
A model for acquiring a learned driving characteristic model based on the driving characteristic feature amount calculated by the driving characteristic feature amount calculating means and the driving characteristic model information including the prediction error distribution information of the driving operation as a reference set in advance. Acquisition means;
Whether the current driving state is an unstable driving state different from the normal driving characteristic based on the driving characteristic model after learning acquired by the model acquiring unit and the current characteristic amount calculated by the driving characteristic feature amount calculating unit. An unstable operation state determination means for determining whether or not
The driving | running state determination apparatus characterized by the above-mentioned.   The driving state determination device according to claim 1, wherein the characteristic amount of the driving characteristic is information related to a steering angle prediction error distribution of the driving operation.   The said model acquisition means acquires the driving | running characteristic model after learning by using the prediction error used as predetermined | prescribed accumulation frequency in the prediction error distribution of driving | operation operation as the feature-value of driving | running characteristic. The operating state determination apparatus described in 2.   The said model acquisition means acquires the driving characteristic model after learning by using the cumulative frequency in the value of the predetermined prediction error in the prediction error distribution of the driving operation as the characteristic amount of the driving characteristic. Item 3. The operating state determination device according to Item 2. Comprising travel information acquisition means for acquiring travel information of the vehicle;
The model acquisition means is a characteristic value of the driving characteristic when the driving characteristic feature quantity calculated by the driving characteristic feature quantity calculation means is determined to be a predetermined driving state based on the vehicle driving information acquired by the driving information acquisition means. The driving | running state determination apparatus described in any one of Claims 1-4 which acquires the driving | running characteristic model after learning using. Based on the vehicle travel information acquired by the travel information acquisition means, comprising a travel state attribute acquisition means for acquiring the attribute of the travel state,
The model acquisition unit includes a driving characteristic feature amount correction unit that corrects the characteristic amount of the driving characteristic based on the attribute of the driving state acquired by the driving state acquisition unit,
The driving state determination device according to any one of claims 1 to 5, wherein the attribute of the traveling state is an attribute relating to a magnitude of a driver's operation amount that is normally requested.   The driving state determination device according to claim 6, wherein the attribute of the driving state is a road type or a road shape. The attribute of the driving state is whether or not the road shape is a curve,
8. The driving state determination apparatus according to claim 7, wherein the driving characteristic feature amount correcting unit corrects the driving characteristic feature amount according to a time ratio of an attribute having a curve road having a predetermined value or more. The attribute of the above running state is a road type,
The driving state determination device according to claim 7 or 8, wherein the driving characteristic feature amount correcting unit corrects the driving characteristic feature amount using a time ratio for each road type. Based on the driving operation information of the driver, the characteristic amount of the driving characteristic of the driver is calculated as information of the prediction error distribution of the driving operation, and the calculated characteristic value of the driving characteristic and the prediction of the driving operation as a preset reference are calculated. Based on the driving characteristic model consisting of error distribution information, obtain the driving characteristic model after learning,
A driving state characterized by determining whether or not the current driving state is an unstable driving state different from the normal driving characteristic based on the acquired driving characteristic model after learning and the characteristic amount of the current driving characteristic. Judgment method.

Images