阅读说明：本技术 自动驾驶人工接管请求时机调节方法及系统 (Method and system for adjusting automatic driving manual takeover request time ) 是由 王文军 李清坤 成波 袁泉 李升波 森大树 于 2020-03-30 设计创作，主要内容包括：本发明涉及自动驾驶人工接管请求时机调节方法及系统,能实现高的用户体验和接管质量。自动驾驶人工接管请求时机调节方法,在驾驶状态达到从自动驾驶模式转换为人工驾驶模式的自动驾驶系统性能边界之前向驾驶员发出接管请求,包括：构建个体驾驶员数据库步骤,将驾驶员完成一次接管记为一个接管事件并将和各接管事件所对应的数据组储存于该驾驶员的个体驾驶员数据库；接管请求提示步骤,计算驾驶员对于驾驶控制权的接管准备就绪程度R,并设定向驾驶员发出所述接管请求的提前时间T；接管质量评估步骤,根据驾驶员的实际操作数据计算接管事件的接管质量P；以及个体驾驶员数据库更新步骤,更新所述个体驾驶员数据库中的所述作用系数α′、β′。(The invention relates to a method and a system for adjusting the time of an automatic driving manual takeover request, which can realize high user experience and takeover quality. A method for adjusting an automated driving manual takeover request opportunity to a driver before a driving state reaches an automated driving system performance boundary that transitions from an automated driving mode to a manual driving mode, comprising: constructing an individual driver database, recording the completion of one-time taking over of the driver as a taking over event, and storing a data group corresponding to each taking over event in the individual driver database of the driver; a takeover request prompting step, namely calculating the takeover readiness degree R of the driver for the driving control right, and setting the advance time T for sending the takeover request to the driver; a takeover quality evaluation step, namely calculating takeover quality P of a takeover event according to actual operation data of a driver; and an individual driver database updating step of updating the action coefficients α ', β' in the individual driver database.)

1. An automated driving manual takeover request timing adjustment method for issuing a takeover request to a driver before a driving state reaches an automated driving system performance boundary where a driving state is switched from an automated driving mode to a manual driving mode due to a driving environment change, thereby reminding the driver of readiness to take over driving control, comprising:

constructing an individual driver database, recording one taking over event of the driver, and storing data sets corresponding to the taking over events in the individual driver database of the driver, wherein each data set comprises: the individual driver database also stores an action coefficient alpha 'related to the take-over quality P and an action coefficient beta' related to the take-over readiness degree R;

a takeover request prompting step, namely calculating a takeover readiness degree R of the driver for the driving control right according to the state data of the driver, and setting a lead time T for sending the takeover request to the driver according to the takeover readiness degree R and the action coefficients alpha 'and beta' stored in the individual driver database;

a takeover quality evaluation step, namely calculating the takeover quality P of the takeover event according to the actual operation data of the driver in the takeover event after the driver completes the takeover corresponding to the takeover request; and

an individual driver database updating step of updating the action coefficients α ', β' in the individual driver database in accordance with the takeover quality P of the driver and the takeover readiness degree R stored in the data group of the individual driver database.

2. The automated driving human takeover request timing adjustment method according to claim 1, wherein,

the method further comprises the step of constructing a driver big data cloud database, wherein the driver big data cloud database is constructed through a driving simulator experiment and a real vehicle experiment based on a takeover scene, and the takeover quality P, the action coefficients alpha 'and beta', the advance time T of a takeover request and the initial value of the takeover readiness degree R are stored in the parameter P stored in the driver big data cloud database₀、α、β、T₀、R₀Each parameter satisfies the formula:

P₀＝αT₀+βR₀。

3. the automated driving human takeover request timing adjustment method according to claim 2, wherein,

in the step of constructing the driver big data cloud database, updating the parameter P stored in the driver big data cloud database by uploading data of all the individual driver databases constructed in the step of constructing the individual driver databases to the driver big data cloud database₀、α、β、T₀、R₀。

4. The automated driving human takeover request timing adjustment method according to claim 1, wherein,

the individual driver database also stores a driver take-over time adjustment item TⁱIn the individual driver database updating step, the action coefficients α ', β' are updated by a multiple linear regression method by the following formulas,

P＝α′(T+Ti)+β′R。

5. the automated driving human takeover request timing adjustment method according to claim 4, wherein,

each of the data sets further comprises a target takeover quality P'₀In the step of prompting the takeover request, the takeover request is sent to the serverThe advance time T is calculated according to the following formula,

6. the automated driving human takeover request timing adjustment method according to claim 5, wherein,

each of said data sets further comprising a posterior takeover advance time T_pAnd a take over cue time error Δ T that satisfies:

ΔT＝T_p-T。

7. the automated driving human takeover request timing adjustment method according to claim 6, wherein,

further comprising updating the takeover time adjustment term T in the individual driver database updating stepⁱIn which the term T is adjusted for the take-over timeⁱTake the average of all deltats in the current individual driver database.

8. The automated driving human takeover request timing adjustment method according to claim 7, wherein,

updating the takeover time adjustment term T when the total number n of data sets in the individual driver database reaches an integer multiple of a first prescribed valueⁱThe step (2).

9. The automated driving human takeover request timing adjustment method according to claim 1, wherein,

updating the action coefficients α ', β' in the individual driver database when the total number n of groups of the data groups in the individual driver database reaches an integer multiple of a second prescribed value.

10. The automated driving human takeover request timing adjustment method according to claim 1, wherein,

further comprising a driver state acquisition step of acquiring a yaw angle pitch and a pitch angle yaw of the face of the driver as state data of the driver,

in the takeover request prompting step, the takeover readiness degree R is calculated using the driver state data collected in the driver state collecting step.

11. The automated driving human takeover request timing adjustment method according to claim 10,

the degree of readiness for take-over is calculated according to the following formula,

wherein f is_tAcquiring frequency of the driver state data, namely acquiring frequency of data of a yaw angle pitch and a pitch angle yaw of the face of the driver;

T_Wis the time window length;

T_cis the current time;

t is the time corresponding to each facial data point collected in the current driving process;

S_drepresenting a value obtained by normalizing an actually collected face data point, a normalization function S_dThe expression of (a) is as follows:

12. an automated driving manual takeover request timing adjustment system that issues a takeover request to a driver before reaching an automated driving system performance boundary that transitions from an automated driving mode to a manual driving mode due to a driving environment change, thereby alerting the driver to be ready to take over driving control, comprising:

a storage module, which stores an individual driver database corresponding to each driver, wherein the individual driver database stores data sets corresponding to each taking-over event of the driver, and each data set comprises: the individual driver database also stores an action coefficient alpha 'related to the takeover quality P and an action coefficient beta' related to the takeover readiness degree R;

the takeover request prompting module is used for calculating the takeover readiness degree R of the driver for the driving control right according to the state data of the driver and setting the advance time T for sending the takeover request to the driver according to the takeover readiness degree R and the action coefficients alpha 'and beta' stored in the individual driver database;

the takeover quality evaluation module is used for calculating the takeover quality P of the takeover event according to the actual operation data of the driver in the takeover event when the driver completes the takeover driving control right corresponding to the takeover request; and

and the database updating module is used for updating the action coefficients alpha 'and beta' in the individual driver database according to the take-over quality P of the driver and the take-over readiness degree R stored in the data group of the individual driver database.

13. The automated driving human takeover request opportunity adjustment system of claim 12, wherein,

the storage module further stores a driver big data cloud database, the driver big data cloud database is constructed through a driving simulator experiment and a real vehicle experiment based on a takeover scene, and the takeover quality P, the action coefficients alpha 'and beta', the advance time T of a takeover request and the initial value of the takeover readiness degree R are stored in the driver big data cloud databaseStored parameter P₀、α、β、T₀、R₀Each parameter satisfies the formula:

P₀＝αT₀+βR₀。

14. the automated driving human takeover request opportunity adjustment system of claim 13, wherein,

updating the parameter P stored in the driver big data cloud database by uploading data of all individual driver databases to the driver big data cloud database₀、α、β、T₀、R₀。

15. The automated driving human takeover request opportunity adjustment system of claim 12, wherein,

the individual driver database also stores a driver take-over time adjustment item TⁱThe database updating module adopts a multiple linear regression method and updates the action coefficients alpha 'and beta' through the following formulas,

P＝α′(T+Tⁱ)+β′R。

16. the automated driving human takeover request opportunity adjustment system of claim 15, wherein,

each of the data sets further comprises a target takeover quality P'₀The takeover request prompt module calculates the takeover request advance time T by the following formula:

17. the automated driving human takeover request opportunity adjustment system of claim 16, wherein,

each of said data sets further comprising a posterior takeover advance time T_pAnd a take over cue time error Δ T that satisfies:

ΔT＝T_p-T。

18. the automated driving human takeover request opportunity adjustment system of claim 17, wherein,

the database updating module further adjusts the takeover time item TⁱUpdating to make the said takeover time adjustment term TⁱThe average of all deltats within the individual driver database is taken.

19. The automated driving human takeover request opportunity adjustment system of claim 18, wherein,

the database update module updates the takeover time adjustment term T when the total number n of the data groups in the individual driver database reaches an integer multiple of a first prescribed valueⁱ。

20. The automated driving human takeover request opportunity adjustment system of claim 12, wherein,

the database update module updates the action coefficients α ', β' in the individual driver database when the total number n of sets of the data sets in the individual driver database reaches an integer multiple of a second prescribed value.

21. The automated driving human takeover request opportunity adjustment system of claim 12, wherein,

further comprises a monitoring module for collecting the yaw angle pitch and the pitch angle yaw of the face of the driver as the state data of the driver,

the takeover request prompting module calculates the takeover readiness degree R using the driver state data collected in the driver state collection step.

22. The automated driving human takeover request opportunity adjustment system of claim 21,

the takeover request prompting module calculates the takeover readiness degree R according to the following formula,

wherein f is_tAcquiring frequency of the driver state data, namely acquiring frequency of data of a yaw angle pitch and a pitch angle yaw of the face of the driver;

T_Wis the time window length;

T_cis the current time;

t is the time corresponding to each facial data point collected in the current driving process;

S_drepresenting a value obtained by normalizing an actually collected face data point, a normalization function S_dThe expression of (a) is as follows:

23. an automated driving manual takeover request timing adjustment method for issuing a takeover request to a driver before reaching an automated driving system performance boundary at which a driver switches from an automated driving mode to a manual driving mode due to a change in driving environment, thereby reminding the driver of readiness to take over driving control, comprising:

a driver state acquisition step, which is used for acquiring driving state data of a driver;

taking over request prompting step, setting time window length T according to driving taking over scene_WAnd calculating the takeover readiness degree R of the driver for the driving control right according to the state data of the driver collected by the driver state collection step in the time window length of the current time Tc.

24. The automated driving human takeover request timing adjustment method according to claim 23, wherein,

in the driver state collection step, a yaw angle pitch and a pitch angle yaw of the face of the driver are collected as the state data of the driver,

in the takeover request prompting step, the takeover readiness degree R is calculated using the state data of the driver collected in the driver state collecting step.

25. The automated driving human takeover request timing adjustment method according to claim 24, wherein,

in the takeover request prompting step, the takeover readiness degree R is calculated according to the following formula,

wherein f is_tAcquiring frequency of the driver state data, namely acquiring frequency of data of a yaw angle pitch and a pitch angle yaw of the face of the driver;

T_Wis the time window length;

T_cis the current time;

t is the time corresponding to each facial data point collected in the current driving process;

S_drepresenting a value obtained by normalizing an actually collected face data point, a normalization function S_dThe expression of (a) is as follows:

26. an automated driving manual takeover request timing adjustment system that issues a takeover request to a driver before reaching an automated driving system performance boundary that transitions from an automated driving mode to a manual driving mode due to a driving environment change, thereby alerting the driver to be ready to take over driving control, comprising:

the monitoring module is used for collecting driving state data of a driver;

and the takeover request prompting module is used for setting a time window length TW according to a driving takeover scene and calculating the takeover readiness degree R of the driver for the driving control right according to the driving state data acquired by the monitoring module in the time window length of the current time Tc.

27. The automated driving human takeover request opportunity adjustment system of claim 26, wherein,

the monitoring module collects the yaw angle pitch and the pitch angle yaw of the driver's face as the driver's status data,

the takeover request prompting module calculates the takeover readiness degree R using the driver's state data collected by the monitoring module.

28. The automated driving human takeover request opportunity adjustment system of claim 27, wherein,

the takeover request prompting module calculates the takeover readiness degree R according to the following formula,

wherein f is_tFor the frequency of acquisition of the driver status data,acquiring the frequency of the data of the yaw angle pitch and the pitch angle yaw of the face of the driver;

T_Wis the time window length;

T_cis the current time;

t is the time corresponding to each facial data point collected in the current driving process;

S_drepresenting a value obtained by normalizing an actually collected face data point, a normalization function S_dThe expression of (a) is as follows:

Technical Field

The invention relates to the field of automatic driving technology and human factors engineering, in particular to a method and a system for adjusting the time of taking over request of automatic driving and manual operation.

Background

Autodrive technology is classified into six classes (L0-L5) according to the american Society of Automotive Engineers (SAE), and autodrive systems of the L3 class are capable of performing certain driving tasks and monitoring driving environment in certain situations, but the driver must be ready to regain driving control when the autodrive system makes a request.

That is, under the automatic driving condition of the level L3, when the system reaches the performance boundary (non-emergency), such as the road construction area, it is necessary to issue a manual take-over request and to give the driver the control right of the vehicle. Automatic driving at the level of L4 also requires the driver to take over in conditions that the system cannot handle.

In the takeover process, the sending time of the takeover request of the system is critical, and the takeover prompt time is too early, so that the user experience of a driver is influenced, the driver can also be caused to excessively trust (or over-dependence) the automatic driving system, the attention degree to the surrounding traffic environment is reduced, and further when the system cannot send the takeover request in time due to a fault, the serious consequence that the driver cannot safely complete the takeover can be caused; and the too late time for taking over prompt can cause the quality of taking over of the driver to be poor, and even the task of taking over can not be completed. Therefore, the appropriate takeover time is of great significance to the quality of manual takeover and the safety of vehicle driving.

The existing takeover request timing is usually set by setting a fixed advance, and a takeover request is sent to the driver at the time when the autopilot system is away from the boundary of the system reaches the advance. Studies have shown that driver distraction can negatively impact driver ride quality and take-over quality.

Under the condition that the takeover request advance is a fixed value, the takeover quality can be different due to different takeover readiness degrees (such as the degree of distraction) of drivers: under the conditions that a driver has better cognition on the traffic environment and the takeover readiness degree is higher, the takeover can be safely completed; if the driver is distracted to a high degree and does not fully know the traffic environment, the quality of the take-over is reduced, and even the take-over cannot be completed safely.

Patent document 1 proposes a control device that monitors a driving state such as a driver's sight-line direction and driving posture and predicts a required take-over time based on the current driving state of the driver.

[ Prior art documents ]

[ patent document ]

Patent document 1: japanese patent laid-open No. 6342856

However, in the automatic driving system disclosed in patent document 1, the required take-over time is predicted only based on the current driving state of the driver, which does not allow accurate evaluation of the required take-over time, resulting in a reduction in the accuracy of the prediction and consequently a unsatisfactory driver user experience and take-over quality.

Disclosure of Invention

The invention aims to provide a method and a system for adjusting the time of an automatic driving manual takeover request, which can realize high user experience and takeover quality.

Means for solving the problems

According to the method for adjusting the timing of the automatic driving manual takeover request, the takeover request is sent to the driver before reaching the boundary of the automatic driving system which is switched from the automatic driving mode to the manual driving mode due to the change of the driving environment, so that the driver is reminded to prepare to take over the driving control right, wherein the method comprises the following steps:

constructing an individual driver database, recording one taking over event of the driver, and storing data sets corresponding to the taking over events in the individual driver database of the driver, wherein each data set comprises: the individual driver database also stores an action coefficient alpha 'related to the take-over quality P and an action coefficient beta' related to the take-over readiness degree R;

a takeover request prompting step, namely calculating a takeover readiness degree R of the driver for the driving control right according to the state data of the driver, and setting a lead time T for sending the takeover request to the driver according to the takeover readiness degree R and the action coefficients alpha 'and beta' stored in the individual driver database;

a takeover quality evaluation step, namely calculating the takeover quality P of the takeover event according to the actual operation data of the driver in the takeover event after the driver completes the takeover driving control right corresponding to the takeover request; and an individual driver database updating step of updating the action coefficients α ', β' in the individual driver database in accordance with the takeover quality P of the driver and the takeover readiness degree R recorded in the data group of the individual driver database.

According to the present invention, an automatic driving manual takeover request timing adjustment system that issues a takeover request to a driver before reaching an automatic driving system boundary where an automatic driving mode is switched to a manual driving mode due to a change in driving environment, thereby reminding the driver of readiness to take over driving control, comprises:

a storage module, which stores an individual driver database corresponding to each driver, wherein the individual driver database stores data sets corresponding to each taking over event of the driver, and each data set comprises: the individual driver database also stores an action coefficient alpha 'related to the take-over quality P and an action coefficient beta' related to the take-over readiness degree R;

the takeover request prompting module is used for calculating the takeover readiness degree R of the driver for the driving control right according to the state data of the driver and setting the advance time T for sending the takeover request to the driver according to the takeover readiness degree R and the action coefficients alpha 'and beta' stored in the individual driver database;

the takeover quality evaluation module is used for calculating the takeover quality P of the takeover event according to the actual operation data of the driver in the takeover event when the driver completes the takeover driving control right corresponding to the takeover request; and

and the database updating module is used for updating the action coefficients alpha 'and beta' in the individual driver database according to the take-over quality P of the driver and the take-over readiness degree R recorded in the data group of the individual driver database.

According to the method and the system for adjusting the automatic driving manual takeover request time, when the takeover readiness degree R of each takeover event and the advance time T for sending the takeover request are calculated, the action coefficient alpha 'related to the takeover quality P and the action coefficient beta' related to the takeover readiness degree R are added, an individual driver database is built to store historical driving data of individual drivers, namely the takeover quality P of each takeover event and the takeover readiness degree R of the drivers, and the action coefficients alpha 'and beta' are updated according to the historical data. Through such a procedure, the coefficients of action α ', β' reflecting the operation habits of the individual drivers can be updated in time by learning the historical driving data of the individual drivers, so that a value conforming to the driving habits of each driver can be obtained when calculating the advance time T at which the takeover request is issued. A high user experience and take-over quality can be achieved.

According to another method for adjusting the timing of an autopilot manual takeover request according to the present invention, a takeover request is issued to a driver before reaching an autopilot system performance boundary where an autopilot mode is switched from an autopilot mode to a manual driving mode due to a change in driving environment, thereby reminding the driver of the readiness to take over driving control, wherein the method comprises:

a driver state acquisition step, which is used for acquiring driving state data of a driver;

and a takeover request prompting step, wherein a time window length TW is set according to a driving takeover scene, and the takeover readiness degree R of the driver for the driving control right is calculated according to the driving state data collected by the driver state collecting step in the time window length of the current time Tc.

According to another automated driving manual takeover request timing adjustment system related to the present invention, a takeover request is issued to a driver before reaching an automated driving system performance boundary at which an automated driving mode is switched from an automated driving mode to a manual driving mode due to a change in driving environment, thereby reminding the driver of readiness to take over driving control, comprising:

the monitoring module is used for collecting driving state data of a driver;

and the takeover request prompting module is used for setting a time window length TW according to a driving takeover scene and calculating the takeover readiness degree R of the driver for the driving control right according to the driving state data acquired by the monitoring module in the time window length of the current time Tc.

In the method and the system for adjusting the automatic driving manual takeover request opportunity, the time window length T is set according to the driving takeover scene_WAnd the takeover readiness degree R is calculated according to the driver state data collected within the time window length of the current time Tc, and compared with the situation that the takeover readiness degree R is calculated only according to the driving state data of the current time, the driving state of the driver can be judged more comprehensively and accurately according to the data of a period of time before the takeover event, so that high user experience and takeover quality are realized. In addition, by setting the time window length T appropriately_WThe early driver state data before the taking over event can be eliminated, so that the interference generated by the data with low relevance to the taking over event can be reduced when the taking over readiness degree R is calculated. The user experience and take-over quality can be further improved.

Drawings

Fig. 1 is a flowchart illustrating a control method of an automatic driving system according to the present invention.

Fig. 2 is a schematic diagram showing constituent modules of the automatic driving system according to the present invention.

Fig. 3 is a schematic diagram showing a monitoring module.

FIG. 4 is a schematic diagram showing a calculation and adjustment module.

FIG. 5 is a schematic diagram illustrating a takeover request prompt module.

Fig. 6 is a schematic diagram showing the takeover quality evaluation module.

FIG. 7 is a schematic diagram illustrating an embodiment takeover scenario.

Fig. 8 is a graph showing the results of correlation analysis of the driver takeover readiness degree and takeover quality.

Fig. 9 is a diagram showing the results of correlation analysis of the driver takeover readiness level and the subjective score of the driving distraction level.

Fig. 10 is a diagram showing the results of correlation analysis of takeover request issuance advance time with takeover quality.

Detailed Description

The first embodiment of the present invention is a method for adjusting an automated driving manual takeover request timing, and the overall flow is shown in fig. 1. The method comprises the following steps.

Step S1: step of constructing driver big data cloud database

The method comprises the steps of constructing a driver big data cloud database through a driving simulator experiment and a real vehicle experiment based on a takeover scene, wherein parameters stored in the database comprise P₀、α、β、T₀、R₀Each parameter satisfies the formula: p₀＝αT₀+βR₀。

Wherein P is₀The method is based on the target takeover quality of experimental big data, and can take 6 to 12 seconds based on experimental data and different takeover scenes to ensure the takeover quality of the safe takeover of a driver; alpha is the action coefficient of the advance time sent by the takeover request obtained through the driving simulation experiment on the takeover quality of the driver, beta is the action coefficient of the takeover readiness degree of the driver obtained through the driving simulation experiment on the takeover quality, and both alpha and beta are obtained by performing multiple linear regression analysis on the data pair obtained through the driving simulation experiment; t is₀Is the initial takeover request issueThe method comprises the following steps of (1) advancing time, namely the lowest takeover request advancing time which can be safely taken over and is counted based on driving simulation big data, namely the takeover advancing time required when a driver is completely concentrated on observing a traffic environment; r₀Is the readiness for take-over when the driver is fully attentive to the observation of the traffic environment, which has a value of 100%.

Step S2: step of constructing an individual driver database

An individual driver database is constructed through real vehicle takeover data of a driven vehicle (hereinafter referred to as 'the vehicle'), obviously, the individual driver databases of different vehicles are different, one-time takeover of the driver is marked as a takeover event, and parameters stored in the database comprise alpha ', beta' and Tⁱ、T₀And a data group { P 'corresponding to each takeover event'₀，ΔT，T_p，P，R}。

Alpha 'is the coefficient of action of the advance time sent by the vehicle's takeover request on the driver's takeover quality, the initial value is alpha, beta' is the coefficient of action of the driver's takeover readiness on the vehicle's takeover quality, the initial value is beta; t isⁱIs a time adjustment item for taking over the time of an individual driver, and the initial value is selected from 5s to 10s to ensure the safety. Data group { P 'corresponding to each takeover event'₀，ΔT，T_pOf P, R'₀The target takeover quality is obtained, and P in a driver big data cloud database is taken as an initial value₀Δ T is the take over time error, T_pIs the posterior takeover advance time, R is the takeover readiness of the driver, P is the actual takeover mass in seconds.

Step S3: step of judging performance boundary of automatic driving system

Whether the performance boundary of the automatic driving system exists in the automatic driving (namely, the working condition which cannot be processed when the automatic driving systems of the L3 and L4 levels are in the automatic driving mode) or not is judged. If so, go to step 4. If not, the process is exited.

Step S4: driver state acquisition step

According to different detection means, the state of the driver is mainly divided into two types, one is a bioelectric signal measured by a contact device, such as electrocardio and electroencephalogram, and the other is information measured by a non-contact device, such as facial information and voice information of the driver.

According to the invention, through non-contact measurement, the pitch angle and the yaw angle of the face of the driver are selected as the state of the driver, the face orientation detection device simultaneously acquires the yaw angle pitch and the pitch angle yaw of the face of the driver at a set frequency, the simultaneously acquired pitch and yaw are used as a face data point, and the yaw angle and the pitch angle of the face of the driver during manual driving are used as references.

Step S5: taking over request prompting step

The driver's readiness to take over R is first calculated. By over-normalizing function S_dStandardizing the actually collected face data points, and if the face orientation of the driver is within the manual driving operation range, considering the driver not to be distracted, otherwise, considering the driver to be distracted, and standardizing a function S_dThe expression of (a) is as follows:

taking into account the timeliness of the degree of driver distraction, i.e. premature distraction does not have an effect on the current driving behavior, the time window weighting function W is passed_tWeighting the readiness of the nozzle R, the readiness of the nozzle R and W_tThe expression of (a) is as follows:

in the formula (f)_tDetecting a frequency for the face orientation, i.e. a frequency at which data is collected by the face orientation detection device; t is_WAccording to different pipe-connecting fields for the length of time windowSetting the scene, the more complex the scene is taken over, the time window length T_WThe longer, the preferred T_WIn the range of 8s-15 s; t is_cIs the current time, t is the time corresponding to each facial data point collected in the current driving process, sigma S_dRepresenting a normalization function S corresponding to all facial data points collected over a time window length_dThe summation is performed.

Then, the current takeover request advance time T is calculated according to the current takeover readiness degree R of the driver, and the calculation formula is as follows:

a distance boundary time TTB is calculated from the vehicle speed and the distance from the driving system boundary (step S5A).

Whether or not to send a takeover request to the driver is determined based on the time TTB required for the host vehicle to travel to the performance boundary of the automated driving system (i.e., the condition that cannot be handled when the automated driving system of the L3, L4 class is in the automated driving mode) and the current takeover request advance time T (step S5B).

Comparing the T with the TTB, if the T is larger than or equal to the TTB, sending a takeover request to the driver, and executing the step S6; and if T is less than TTB, continuing to wait until T is more than or equal to TTB, sending a take-over request to the driver, and executing step S6.

Step S6: quality evaluation step of connecting pipe

The driver completes the current takeover according to the current takeover request (step S6A). Then, the current actual takeover quality P is calculated based on the current takeover operation data of the driver (step S6B). There are various indexes for taking over the quality, such as statistical value of input data of a steering wheel of the driver, statistical value of input data of an accelerator pedal of the driver, statistical value of input data of a decelerator pedal of the driver, reaction time of the driver, and the like. The difference between the time distance TTBT of the boundary between the self-vehicle and the automatic driving system and the reaction time TOT of the driver is adopted as the quality of taking over, namely:

P＝TTBT-TOT

in particular, when the take-over failsThat is, when a collision or the like occurs, P is 0. At the same time, the posterior joint tube advance time T is calculated from the following formula_p：

Namely T_pIs the takeover advance time calculated from the actual takeover quality.

Step S7: recording the takeover time data set

After the current takeover event occurs, calculating a takeover prompt time error delta T corresponding to the current takeover event, wherein the delta T is T_p-T, i.e. the takeover prompt time error Δ T, is the difference between the a posteriori takeover advance time and the takeover request advance time. Since for safety, TⁱIs selected from 5s to 10 s. Recording data { P'₀，ΔT，T_p，P，R}。

In addition, the data recorded in the individual driver database are uploaded to a driver big data cloud database, and the data stored in the driver big data cloud database and used as the parameter P 'in the individual driver database are updated periodically according to the uploaded data'₀、α′、β′、Tⁱ、T₀Parameter P of the initial value of₀、α、β、T₀、R₀。

Step S8: individual driver database updating step

Determine { P 'in the Current Individual driver database'₀，ΔT，T_pAnd whether the total number N of P, R reaches an integral multiple of N1 (step S8A). If so, updating T in the current individual driver databaseⁱ，TⁱTaking all the statistical values of delta T in the current individual driver database, such as the mean value or the median of the delta T and the like, and then entering the following step 8B; otherwise, T in the current individual driver database is not updatedⁱThe process returns to step S2.

If the value of N1 is too large, the data update will be untimely, and the driver will be over-trusted (because the value of Δ T is negative for safety); if the value of N1 is too small, the fluctuation of data is large and unstable, so N1 should be a value in the range of 8 to 12, preferably 10, based on the data obtained by the experiment.

9) Determine { P 'in the Current Individual driver database'₀，ΔT，T_pAnd whether the total number N of P, R reaches an integral multiple of N2 (step S8B). If yes, a method of multiple linear regression is adopted, and the alpha 'and the beta' in the current individual driver database are updated through the following formulas: p ═ α' (T + T)ⁱ) + β' R, and then returns to step S2. Otherwise, α 'and β' in the current individual driver database are not updated, and the process returns directly to step S2.

It should be noted that if the value of N2 is too large, it will result in a data update that is not timely, and thus in excessive trust (since the value of Δ T is typically negative for security); if the value of N2 is too small, the fluctuation of data will be large and unstable, so based on the data obtained by experiment, N2 should be 500, preferably 450 and 550.

Further, in step S8A, if { P 'is in the current individual driver database'₀，ΔT，T_pAnd when the total number N of the P, R reaches integer multiple of N1, firstly judging the delta T calculated in the step S6, and rejecting the corresponding data group { P'₀，ΔT，T_pP, R } and then updates T in the current individual driver databaseⁱ，TⁱAll Δ T statistics, such as the mean, standard deviation, median, etc., of the Δ ts in the current individual driver database are taken and returned to step S2.

According to the above embodiment, the following advantageous effects can be obtained.

(1) In the method and the system for adjusting the timing of the automatic driving manual takeover request, the action coefficient alpha 'related to the takeover quality P and the action coefficient beta' related to the takeover readiness degree R are added when the takeover readiness degree R of each takeover event and the advance time T for sending the takeover request are calculated, historical driving data of an individual driver, namely the takeover quality P of each takeover event and the takeover readiness degree R of the driver, of the individual driver are constructed in an individual driver database, and the action coefficients alpha 'and beta' are updated according to the historical data. Through such steps, the coefficients of action α ', β' reflecting the operating habits of the individual drivers can be updated in time by learning the historical driving data of the individual drivers, so that a value conforming to the driving habits of each driver can be obtained when calculating the advance time T at which the takeover request is issued. A high user experience and take-over quality can be achieved.

(2) As the data recorded in the individual driver database can be uploaded to the driver big data cloud database, and the parameter P 'stored in the driver big data cloud database and taken as the individual driver database is updated according to the uploaded data at regular intervals'₀、α′、β′、Tⁱ、T₀Parameter P of the initial value of₀、α、β、T₀、R₀Therefore, the initial value of the cloud database can be updated based on more takeover data performed by more drivers, and a better initial value is provided for vehicles produced later.

(3) In the step of prompting the takeover request, the time window length T is set according to the driving takeover scene_WAnd the takeover readiness degree R is calculated according to the driver state data collected within the time window length of the current time Tc, and compared with the situation that the takeover readiness degree R is calculated only according to the driving state data of the current time, the driving state of the driver can be judged more comprehensively and accurately according to the data of a period of time before the takeover event, so that high user experience and takeover quality are realized. In addition, by setting the time window length T appropriately_WThe early driver state data before the taking over event can be eliminated, so that the interference generated by the data with low relevance to the taking over event can be reduced when the taking over readiness degree R is calculated. The user experience and take-over quality can be further improved.

(4) In the individual driver database updating step, the action coefficients α ', β ', P ═ α ' (T + T) are updated by a formula using a multiple linear regression methodⁱ) + beta' R, can be easily obtained that can accurately reflect the individual driverDriving habits and state, and calculating a proper takeover readiness degree R and a lead time T for sending a takeover request. In addition, the adjustment term T of the take-over time of the individual driver is introduced into the updating formulaⁱBy adjusting the parameters according to the driver, the safety of the take-over can be further ensured.

(5) Butt-joint time adjustment T in individual driver database update stepⁱUpdating, in particular, the take-over time adjustment term TⁱTake the average of all deltats in the current individual driver database. Thus, the driving history data of the driver can be effectively utilized to adjust the pipe-taking time TⁱTimely updating is carried out, so that the driving habits of the driver are better reflected in the calculation of the taking over readiness degree R and the advance time T for sending the taking over request.

(6) Updating the takeover time adjustment term T when the total number N of data groups reaches an integer multiple of N1 or N2ⁱAnd the action coefficients alpha 'and beta' are set, and N1 and N2 are respectively set to be a value in the range of 8-12 and a value in the range of 450-550, so that the data can be ensured to be updated in time, and the data is prevented from being large in fluctuation and unstable.

(7) By collecting the face pitch angle yaw and yaw angle pitch of the driver as the state data of the driver, the degree of distraction of the driver can be accurately detected, thereby more reasonably calculating the readiness degree R.

Other variants

In the above embodiment, a method of multiple linear regression is adopted and the formula P ═ α' (T + T)ⁱ) The action coefficients α ', β ' are updated by + β ' R, but they may be updated by other formulas as long as the driving history of the driver can be reflected.

In the above embodiment, the individual driver database also stores the driver's takeover time adjustment term TⁱHowever, on the premise of ensuring the take-over safety, the take-over time adjustment item T may not be storedⁱ。

In the above embodiment, the period is regularUpdating parameters P 'stored in the driver big data cloud database and used as individual driver database'₀、α′、β′、Tⁱ、T₀Parameter P of the initial value of₀、α、β、T₀、R₀However, if the driving history of the driver can be accurately reflected in the update of the action coefficients α 'and β' only by using the individual driver database, the initial values of the respective parameters may be set by other methods without using the cloud database of the large data.

In the above embodiment, the face pitch angle yaw and yaw angle pitch of the driver are collected as the state data of the driver, but other parameters of the driver such as the positions of the hands and feet of the driver, the driving posture, and the like may be used as the state data.

A second embodiment of the present invention is an automatic driving system. The system mainly comprises a storage module, a monitoring module, a calculation and adjustment module, a takeover request prompting module, a takeover quality evaluation module and a system parameter online learning module, and is shown in figure 2.

1) A storage module: the driver data management system comprises a driver big data cloud database and an individual driver database, wherein data stored in the two databases are described in the control method, and are not described herein again.

2) A monitoring module: the camera collects the pitch angle and the yaw angle of the face of the driver, and the face of the driver is processed and calculated to output the takeover readiness (R) of the driver in real time. Mainly including face recognition, face orientation recognition, and driver's take over readiness (R) calculation, as shown in fig. 3.

3) A calculation and adjustment module: including takeover request lead time (T) calculation, time to system boundary (TTB) calculation, takeover request decision function, as shown in fig. 4. The formula for calculating is as described above, and is not described herein again.

4) Take over request prompt module: and sending a take-over request to the driver in modes of prompt tone, head-up display, instrument board graphical prompt and the like according to the take-over request decision of the calculation and adjustment module until the take-over of the driver is finished, as shown in fig. 5.

5) A take-over quality evaluation module: from the moment when the takeover prompt is sent to the moment when the takeover is completed, the takeover quality is calculated according to the vehicle-mounted sensor information, the original information such as the acceleration, the angular velocity, the steering wheel angle, the pedal stroke and the like, as shown in fig. 6.

6) The system parameter online learning module: and updating the system parameter values of the individual driver database in the storage module according to the rule by the obtained takeover quality result and the takeover readiness degree result.

This embodiment can obtain advantageous effects corresponding to those of the first embodiment.

The effectiveness of the invention is verified with reference to the following examples:

1) takeover scenario

One of the typical boundaries of the current L3-level autopilot system, the highway construction area, is selected as the boundary to take over. The traffic scene of the takeover is set as a three-lane highway, and in the daytime of sunny days, the highway only has one lane (the leftmost lane or the rightmost lane) for traffic due to construction and closure of the two lanes. The driver needs to change the lane after receiving the taking-over prompt, and before receiving the taking-over request, the vehicle is in an automatic driving mode and automatically keeps the speed of 100 kilometers per hour to drive in the middle lane. Random traffic flow occurs during automatic driving, traffic flow occurs in a target lane during lane changing, two vehicles are respectively positioned in front of and behind the own vehicle, and lane changing is performed randomly from left to right during taking over every time so as to avoid a learning effect, as shown in fig. 7.

2) Subjective assessment of readiness of driver take-over

The driver needs to score the distraction (an integer of 0-10) of the driver after taking over each time, with 0 being the least distraction and 10 being the most distraction. And standardizing the data of each driver, namely obtaining the scoring average value of the test taken over by a certain driver for many times, and subtracting the average value from the original data to obtain the final used driver distraction degree score.

3) Design of experiments

A driving simulator is selected as a test platform, the data of 16 Chinese drivers are tested, and the taking over prompting time is divided into three types: 6 seconds before the boundary, 8 seconds before the boundary, and 10 seconds before the boundary. Each driver performed 3 groups of 18 take-over experiments. And the driver is distracted visually by playing the video, and the driver can decide to watch the video or the surrounding environment.

The method comprises the steps of collecting a yaw angle and a pitch angle (frequency of 20 Hz) of the face orientation of a driver through a monocular camera in a vehicle, standardizing the actually collected yaw angle pitch and pitch angle yaw by taking the yaw angle and pitch angle of the driver during manual driving as references, namely, if the face orientation of the driver is in the range of the driver during manual driving, considering the driver not to be distracted, otherwise, considering the driver to be distracted, and performing normalization through a normalization function S_dAnd (4) carrying out standardization:

considering the timeliness of the driver distraction, the time window weighting function W is used_tWeighted, time window length W_tThe sampling time is 12 seconds:

i.e. the normalized function (S) of the yaw angle and pitch angle of the driver' S original face orientation_d) And a time window weighting function (W)_t) The degree of readiness for the driver to take over is obtained after weighting, as shown in fig. 8. Wherein, the calculation formula of the pipe taking degree R is as follows:

the judgment conditions for the occurrence of the take-over time are as follows: the absolute value of the steering wheel angle is larger than 2 degrees, the travel of an accelerator pedal is larger than 5 percent, and the travel of a decelerator pedal is larger than 5 percent. When any judgment condition is reached, the system judges that the takeover event occurs.

4) Results and analysis of the experiments

After the original data are processed, the taking-over quality P of the driver is obtained through calculation, and the taking-over readiness degree R and the distraction degree of the driver are subjectively scored.

4.1 driver takeover readiness and driver subjective distraction scoring

As shown in FIG. 9, the driver takeover level is inversely related to the driver's subjective distraction score, with a significance level p<0.001, coefficient of determination r²The method proposed in the present invention for evaluating the readiness of the driver to take over based on visual distraction is reasonable and effective as 0.436.

4.2 degree of readiness for driver to take over and quality of driver take over (proving that R has an effect on P)

Based on experimental data, the relevance analysis of the takeover readiness of the driver and the takeover quality P of the driver is carried out, and the obtained result is the significance level P<0.001, coefficient of determination r²0.309, as shown in fig. 8, wherein,

P＝TTBT-TOT

the results indicate that the driver's readiness to take over is related to the quality of the driver's take over.

4.3 takeover request lead time and driver takeover quality (proving that T has an effect on P)

Based on experimental data, the takeover request advance time (T) is analyzed in variance with the driver takeover quality (P). Data at the same takeover readiness level were selected from the experimental results, and the takeover quality at the time of sending advance of different takeover requests was examined, and the result was that the significance level p was <0.001, and the F value was 41.535, as shown in fig. 10. It is demonstrated that the effect of adjusting the pipe quality by sending an advance time for a take-over request is effective. Under the condition of determining the takeover scene, the takeover quality of the driver is related to the takeover readiness of the driver and the takeover request sending advance time, and the earlier the takeover prompt is, the better the takeover can be finished with high quality.

In conjunction with the foregoing, when the driver is low in the readiness to take over, the quality of take over can be improved by increasing the take over request advance time. Further, different takeover request lead times can be provided according to the takeover readiness degree of the driver, and further similar or same target takeover quality can be achieved by adjusting the takeover request lead times under the condition of any takeover readiness degree of the driver.