CN116595453A - Method for determining driving capability recovery time of highway automatic driving takeover process - Google Patents
Method for determining driving capability recovery time of highway automatic driving takeover process Download PDFInfo
- Publication number
- CN116595453A CN116595453A CN202310527781.9A CN202310527781A CN116595453A CN 116595453 A CN116595453 A CN 116595453A CN 202310527781 A CN202310527781 A CN 202310527781A CN 116595453 A CN116595453 A CN 116595453A
- Authority
- CN
- China
- Prior art keywords
- driving
- vehicle
- driving state
- takeover
- time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 84
- 238000011084 recovery Methods 0.000 title claims abstract description 43
- 230000008569 process Effects 0.000 title claims abstract description 28
- 238000004088 simulation Methods 0.000 claims abstract description 33
- 239000000203 mixture Substances 0.000 claims abstract description 21
- 238000009826 distribution Methods 0.000 claims description 68
- 230000000694 effects Effects 0.000 claims description 19
- 238000004422 calculation algorithm Methods 0.000 claims description 10
- 238000010801 machine learning Methods 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 7
- 230000001133 acceleration Effects 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 5
- 230000006870 function Effects 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 3
- 230000006399 behavior Effects 0.000 claims description 2
- 238000013461 design Methods 0.000 abstract description 6
- 238000012216 screening Methods 0.000 abstract description 5
- 238000004458 analytical method Methods 0.000 abstract description 3
- 238000012512 characterization method Methods 0.000 abstract 1
- 238000011160 research Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 4
- 241000208340 Araliaceae Species 0.000 description 3
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 3
- 235000003140 Panax quinquefolius Nutrition 0.000 description 3
- 235000008434 ginseng Nutrition 0.000 description 3
- 230000003993 interaction Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 230000006641 stabilisation Effects 0.000 description 2
- 238000011105 stabilization Methods 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000010224 classification analysis Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000013401 experimental design Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 210000001503 joint Anatomy 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- 230000036962 time dependent Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/211—Selection of the most significant subset of features
- G06F18/2113—Selection of the most significant subset of features by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention discloses a method for determining the recovery time of driving capability in the automatic driving takeover process of a highway, which comprises the following specific steps: constructing an automatic driving takeover experimental platform; designing and constructing a typical highway automatic driving takeover scene; recruiting drivers to take over simulation experiments, and collecting vehicle track data in the experimental process; extracting and screening indexes capable of reflecting driving states; constructing a driving state identification method based on a Gaussian mixture model; carrying out driving state identification analysis by combining the screened driving state characterization index combination, and determining the driving capability recovery time of a driver; the method can accurately determine the driving capability recovery time of each driver in the taking over process, and provides a theoretical basis for the optimal design of an automatic driving system.
Description
Technical Field
The invention relates to the technical field of automatic driving, in particular to a method for determining the recovery time of driving capability in the process of taking over automatic driving on a highway.
Background
The conditional automatic driving vehicle can independently monitor road traffic environment in a designed operation area and perform transverse and longitudinal kinematic control, allows driving ginseng and non-driving related subtasks, and prompts a driver to take over when meeting traffic conditions which cannot be processed by the system, and the driver needs to take over the vehicle in a specified time. The secondary tasks related to the driving of the ginseng and the non-driving during the automatic driving can cause the reduction of the perception, decision and driving capability of the driving ginseng and the non-driving, thereby influencing the control state of the driver on the vehicle. After taking over, the driving capability of the vehicle is continuously restored to a stable normal driving state along with the increase of the perception of the environmental information by the driver and the continuous manual operation. The duration of the driving capability recovery time can influence the takeover performance of the driver, understand how long the driver needs to recover the driving capability to the normal driving state after takeover, be favorable for the optimal design of an automatic driving system and a man-machine interaction interface, and improve the takeover performance of the driver.
The current research mainly focuses on determining the time for a driver to take over the control right of a vehicle in the automatic driving taking over process and the influence condition of different factors on taking over performance, and few researches pay attention to how long after taking over the driver can recover the driving capability to a normal state, and the method for determining the recovery time of the driving capability is less. Therefore, it is necessary to design an automatic driving takeover simulation scene by using a driving simulation experiment method and comprehensively considering the influence of different factors, recruit drivers to participate in the simulator experiment, collect track data of vehicles in the experiment process, and provide a driving capability recovery time determining method for the automatic driving takeover process of a highway to determine the driving capability recovery time of each driver.
Disclosure of Invention
The purpose of the invention is that: the method for determining the recovery time of the driving capability of the highway automatic driving takeover process provides a theoretical basis for the optimal design of a man-machine interaction interface of an automatic driving system, and improves takeover performance of a driver.
In order to achieve the above functions, the present invention designs a method for determining a recovery time of driving capability in a highway autopilot takeover process, wherein for a vehicle performing autopilot on a highway, a takeover process from an autopilot state to a driver driving state is performed by executing the following steps S1 to S6 to complete the determination of the recovery time of driving capability of the driver:
step S1: constructing an automatic driving takeover simulation platform based on a driving simulator, and constructing a takeover simulation test scene by considering the influence of different preset experimental factors;
step S2: based on the built takeover simulation test scene, carrying out takeover simulation experiments aiming at different drivers, wherein the takeover simulation experiments comprise preset takeover trigger events, when the preset takeover trigger events occur, the vehicles driven by the drivers are converted from an automatic driving state to a driver driving state, and vehicle track data of the vehicles driven by the drivers in the takeover simulation experiments are collected;
step S3: extracting each index reflecting the driving state according to the acquired vehicle track data in the takeover simulation experiment;
step S4: combining the indexes reflecting the driving state according to a preset rule, evaluating and quantifying the effect of reflecting the driving state of each index combination, sequencing the effect of reflecting the driving state of each index combination, and selecting the index combination with the best effect of reflecting the driving state;
step S5: constructing a driving state identification method based on a machine learning algorithm, and judging whether the driving state of a driver taking over the vehicle is stable or unstable based on the index combination with the best effect on reflecting the driving state, which is obtained in the step S4;
step S6: and (5) analyzing and determining the driving capability recovery time of each driver by using a driving state identification method.
As a preferred technical scheme of the invention: the influences of different preset experimental factors considered in the step S1 include automatic driving duration time, takeover request time, automatic driving speed and front vehicle speed;
the automatic driving duration time is the time for the automatic driving system to independently control the vehicle to run or the time for the driver to execute the subtask during automatic driving, wherein the subtask is the behavior for distracting the driver, the take-over request time is the early warning time for the vehicle to send out the take-over request, the automatic driving speed is the running speed of the vehicle during automatic driving, and the front vehicle speed is the running speed of the vehicle in front of the vehicle when the vehicle sends out the take-over request.
As a preferred technical scheme of the invention: in step S2, the preset take-over triggering event includes a lane change and deceleration event of the front vehicle, specifically, the automatically driven vehicle runs on the middle lane at a preset speed, and the front vehicle running in front of the left lane changes to the middle lane and decelerates.
As a preferred technical scheme of the invention: the vehicle track data acquired in the step S2 comprise the running track data of the vehicle and the front vehicle after the driver takes over the vehicle, the segment length of the running track data is 60S, and the acquisition frequency is 100Hz.
As a preferred technical scheme of the invention: the index reflecting the driving state extracted in step S3 includes the average longitudinal speed x 1 Standard deviation x of longitudinal velocity 2 Average longitudinal acceleration x 3 Average heel-to-heel spacing x 4 Standard deviation x of heel-to-heel spacing 5 Mean value x of speed difference between front and rear vehicles 6 Standard deviation x of speed difference between front and rear vehicle 7 Average headway x 8 Time x after take over 9 Average lateral velocity x 10 Standard deviation of transverse velocity x 11 Average lateral acceleration x 12 Average lane offset x 13 Standard deviation x of lane offset 14 The method comprises the steps of carrying out a first treatment on the surface of the Wherein time x after take-over 9 Time 0 counted by the time when the driver takes over the vehicle.
As a preferred technical scheme of the invention: in step S3, the indexes reflecting the driving state are combined by adopting a combination optimizing method, and the effect of reflecting the driving state of each combination is evaluated by adopting a bulldozing distance, wherein the bulldozing distance is the minimum cost required for converting one distribution into another distribution, and the calculation method is as follows:
assume thatFor a first distribution having m classes, where p i Is the ith class of P, +.>Is p i Corresponding weights, i e {1,2, …, m }; let->For another distribution having n classes, where q j J-th category of Q, +.>Is q j Corresponding weights, j e {1,2, …, n }; d= [ D ] ij ]Is a distance matrix, d ij For category p i And category q j The distance between them, the total cost l required to convert distribution P into distribution Q cost The formula is as follows:
wherein f i,j For category p i And category q j The amount of earth in between, the minimum cost required to convert the distribution P into the distribution Q is obtained according to the following constraint:
f i,j ≥0,1≤i≤m,1≤j≤n
solving the linearization problemObtain the optimumThe bulldozing distance is as follows:
where EMD (P, Q) is the bulldozing distance converted from distribution P to distribution Q.
As a preferred technical scheme of the invention: the combined optimizing method comprises the following steps:
step S31: taking over the vehicle track data in the range of 60s after the driver takes over the vehicle in the collected take-over simulation experiment, extracting each index reflecting the driving state from the vehicle track data, and carrying out normalized conversion;
step S32: selecting K indexes of indexes reflecting driving states, wherein the K indexes do not contain the time x after taking over 9 Combining the two components by adopting a permutation and combination principle to obtainGroup index combination and corresponding vehicle track data thereof, and time x after taking over 9 Respectively adding the data into each group of index combinations, numbering the vehicle track data, and taking K indexes and the time x after taking over 9 As reference data;
step S33: reducing the dimension of the high-dimension vehicle track data corresponding to each group of index combinations obtained in the step S32 to two-dimensional data by adopting a t-SNE algorithm, respectively calculating bulldozing distances between the two-dimensional data corresponding to each group of index combinations and the two-dimensional data corresponding to the reference data, and carrying out descending order arrangement on the numerical values of each bulldozing distance;
step S34: the index combination corresponding to the minimum bulldozing distance is used as the index combination with the best effect for reflecting the driving state.
As a preferred technical scheme of the invention: the driving state identification method in step S5 is based on a gaussian mixture model, and the expression thereof is as follows:
λ＝{w i ,μ i ,Σ i },i＝1,2,…,N
wherein p (x|lambda) is the probability distribution of the Gaussian mixture model, lambda is the parameter of the Gaussian mixture model, w i The weight term of the ith Gaussian distribution, x is a vector formed by D-dimensional continuous observation data, g (x|mu) i ,Σ i ) Probability density function, μ, as an ith Gaussian distribution i Is the mean value of the ith Gaussian distribution, Σ i Covariance matrix of ith Gaussian distribution, N is number of Gaussian distribution;
the posterior probability P (i|x of the ith Gaussian distribution t Lambda) is as follows:
wherein x is t For a given T-dimensional training vector, x t ∈{x 1 ,x 2 ,...,x T }；
And (3) presetting a posterior probability threshold, judging that the Gaussian distribution which is larger than or equal to the preset posterior probability threshold is stable, and judging that the Gaussian distribution which is smaller than the preset posterior probability threshold is unstable.
As a preferred technical scheme of the invention: in step S6, the hyper-parameters in the gaussian mixture model are set as follows: the number of components is 2, the maximum iteration number is 1000, the covariance type is complete covariance, and the weight, the mean value and the precision are initialized by using a k-means method.
As a preferred technical scheme of the invention: in step S6, a Gaussian mixture model is applied, the index combinations related to the transverse direction and the longitudinal direction are selected respectively, the driving state of the driver after taking over the vehicle is judged in the transverse direction and the longitudinal direction, the time for the driver to reach the stable state in the transverse direction and the longitudinal direction after taking over the vehicle is determined respectively, and the maximum value of the two is taken as the driving capability recovery time of the driver.
The beneficial effects are that: the advantages of the present invention over the prior art include:
1. the invention constructs an automatic driving take-over experimental platform based on the driving simulator, processes and analyzes track data in the process of butting pipes, extracts and calculates a plurality of indexes reflecting driving states, and provides a combined optimizing method for screening the indexes to determine a multi-index combination which can most reflect the driving states, thereby making up for the unilateralness and vulnerability of the driving states represented by a single index, and improving the objectivity and comprehensiveness of evaluation index selection by adopting the combined optimizing method;
2. the invention utilizes the Gaussian mixture model in the unsupervised machine learning algorithm to identify and classify the driving state after taking over, thereby improving the objectivity and reliability of the identification result;
3. the driving capability recovery time determining method provided by the invention can accurately determine the driving capability recovery time of each driver under different takeover scenes, and improves the calculation accuracy of the driving capability recovery time;
4. the invention reduces the risk and cost of the real road experiment through the simulation experiment, ensures the safety of the driver, improves the accuracy of the experimental result and has good economic benefit.
Drawings
Fig. 1 is a flowchart of a driving ability restoration time determining method of an expressway automatic driving takeover process according to an embodiment of the invention;
fig. 2 is a schematic diagram of an autopilot take-over experimental platform provided according to an embodiment of the present invention;
FIG. 3 is a flow chart of a take over simulation experiment provided in accordance with an embodiment of the present invention;
FIG. 4 is a schematic diagram of a scenario for taking over a triggering event provided according to an embodiment of the present invention;
FIGS. 5 (a) -5 (b) illustrate bulldozing distances between data and reference data for different longitudinal index combinations provided in accordance with embodiments of the present invention;
FIGS. 6 (a) -6 (l) are longitudinal driving state classification results at different time intervals provided according to embodiments of the present invention;
fig. 7 (a) -7 (l) are lateral driving state classification results at different time intervals provided according to an embodiment of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings. The following examples are only for more clearly illustrating the technical aspects of the present invention, and are not intended to limit the scope of the present invention.
The method for determining the recovery time of the driving capability of the highway automatic driving takeover process provided by the embodiment of the invention is used for determining the recovery time of the driving capability of the driver by executing the following steps S1-S6 with reference to FIG. 1 in order to finish the determination of the recovery time of the driving capability of the driver for the vehicle executing automatic driving on the highway and for the takeover process of the automatic driving state to the driving state of the driver:
step S1: constructing an automatic driving takeover simulation platform based on a driving simulator, referring to fig. 2, wherein the automatic driving takeover simulation platform comprises a hardware input system, a data acquisition system and a simulation scene rendering system, taking the influence of different preset experimental factors into consideration, constructing a takeover simulation test scene by using an orthogonal experimental design principle and using simulation software, and generating a vehicle controlled by a driver in the takeover simulation test scene by driving simulation software, such as UC-win/Road simulation software;
the influence of different preset experimental factors comprises automatic driving duration time, takeover request time, automatic driving speed and front vehicle speed;
wherein the autopilot duration is the time that the autopilot system independently operates the vehicle, or the time that the driver performs a subtask (60 s, 180s, 300 s) while autopilot, wherein the subtask is a distracting action for the driver, such as playing a game; the take-over request time is early warning time (taking 4s, 5s and 6 s) when the vehicle sends a take-over request, the automatic driving speed is the running speed (taking 80km/h, 85km/h and 90 km/h) when the vehicle is automatically driven, and the front vehicle speed is the running speed (taking 55km/h, 60km/h and 65 km/h) of the front vehicle when the vehicle sends a take-over request.
Step S2: based on the built takeover simulation test scene, carrying out takeover simulation experiments aiming at different drivers, referring to FIG. 3, the takeover simulation experiments comprise preset takeover triggering events, when the preset takeover triggering events occur, the vehicles driven by the drivers are converted from an automatic driving state to a driver driving state, and vehicle track data of the vehicles driven by the drivers in the takeover simulation experiments are collected;
referring to fig. 4, the preset takeover triggering event includes a lane change and deceleration event of a front vehicle, specifically, an automatically driven vehicle traveling at a preset speed on a center lane, and a front vehicle traveling in front of left lane change to the center lane and deceleration traveling.
The acquired vehicle track data comprise the running track data of the vehicles and the front vehicles after the drivers take over the vehicles, the segment length of the running track data is 60s, and the acquisition frequency is 100Hz.
Step S3: extracting each index reflecting the driving state according to the acquired vehicle track data in the takeover simulation experiment; the calculation period of each extracted index reflecting the driving state is 1s.
The extracted index reflecting the driving state includes the average longitudinal speed x 1 (m/s), standard deviation of longitudinal velocity x 2 Average longitudinal acceleration x 3 (m/s 2 ) Average heel-to-heel spacing x 4 (m) Standard deviation of the heel-to-heel distance x 5 Mean value x of speed difference between front and rear vehicles 6 (m/s), standard deviation x of front and rear vehicle speed difference 7 Average headway x 8 (s), time after takeover x 9 (s) average lateral velocity x 10 (m/s), transverse velocity standard deviation x 11 Average lateral acceleration x 12 (m/s 2 ) Average lane offset x 13 (m), criterion of lane offsetDifference x 14 The method comprises the steps of carrying out a first treatment on the surface of the Wherein time x after take-over 9 Time 0 counted by the time when the driver takes over the vehicle.
In step S3, the indexes reflecting the driving state are combined by adopting a combination optimizing method, and the effect of reflecting the driving state by adopting each combination is evaluated by adopting a bulldozing distance, wherein the bulldozing distance is the minimum cost required for converting one distribution into another distribution, and the cost is the product of the direct moving 'soil' quantity and the moving distance. The calculation method is as follows:
assume thatFor a first distribution having m classes, where p i Is the ith class of P, +.>Is p i Corresponding weights, i e {1,2, …, m }; let->For another distribution having n classes, where q j J-th category of Q, +.>Is q j Corresponding weights, j e {1,2, …, n }; d= [ D ] ij ]Is a distance matrix, d ij For category p i And category q j The distance between them, the total cost l required to convert distribution P into distribution Q cost The formula is as follows:
wherein f i,j For category p i And category q j The amount of earth in between, the minimum cost required to convert the distribution P into the distribution Q is obtained according to the following constraint:
f i,j ≥0,1≤i≤m,1≤j≤n
solving the linearization problem to obtain the optimal valueThe bulldozing distance is as follows:
where EMD (P, Q) is the bulldozing distance converted from distribution P to distribution Q.
The combined optimizing method comprises the following steps:
step S31: taking over the vehicle track data in the range of 60s after the driver takes over the vehicle in the collected take-over simulation experiment, extracting each index reflecting the driving state from the vehicle track data, and carrying out normalized conversion;
step S32: selecting K indexes of indexes reflecting driving states, wherein the K indexes do not contain the time x after taking over 9 Combining the two components by adopting a permutation and combination principle to obtainGroup index combination and corresponding vehicle track data thereof, and time x after taking over 9 Respectively adding the data into each group of index combinations, numbering the vehicle track data, and taking K indexes and the time x after taking over 9 As reference data;
in one embodiment, an index x reflecting driving status 1 、x 2 、x 3 、x 4 、x 5 、x 6 、x 7 、x 8 Is used to determine the state of motion in the longitudinal direction of the vehicle, the combination of index points in 254 can be obtained And vehicle track data corresponding to different index combinations. Since the drivability recovery time is time-dependent, the time x after index take over 9 Are added to the above 254 sets of index combinations, respectively, and then the vehicle track data are numbered. The index included in the reference data is x 1 ,x 2 ,x 3 ,x 4 ,x 5 ,x 6 ,x 7 ,x 8 ,x 9 。
Step S33: reducing the dimension of the high-dimensional vehicle track data corresponding to the 255 groups of index combinations obtained in the step S32 to two-dimensional data by adopting a t-SNE (t-distributed stochastic neighbor embedding) algorithm, respectively calculating the bulldozing distances between the two-dimensional data corresponding to the 254 groups of index combinations and the two-dimensional data corresponding to the reference data, and arranging the numerical values of the bulldozing distances in a descending order, wherein the bulldozing distances between the data of different longitudinal index combinations and the reference data are shown in the figures 5 (a) -5 (b);
step S34: the combination of the indexes corresponding to the minimum bulldozing distance is used as the combination with the best effect for reflecting the driving state.
Step S4: combining the indexes reflecting the driving state according to a preset rule, evaluating and quantifying the effect of reflecting the driving state of each index combination, sequencing the effect of reflecting the driving state of each index combination, and selecting the index combination with the best effect of reflecting the driving state;
as can be seen from fig. 4, the bulldozing distances between the data numbered 253 and 254 and the reference data in this example are respectively 0.056 and 0.054, which are very small. And the data corresponding to the number 254 contains 7 variables, x respectively 1 ,x 2 ,x 3 ,x 5 ,x 6 ,x 7 ,x 9 Data corresponding to the number 253Containing 5 variables, in turn x 1 ,x 2 ,x 3 ,x 6 ,x 9 Less data than the data corresponding to number 254, so the data corresponding to number 253 is used for driving state identification studies.
Step S5: constructing a driving state identification method based on a machine learning algorithm, and judging whether the driving state of a driver taking over the vehicle is stable or unstable based on the index combination with the best effect on reflecting the driving state, which is obtained in the step S4;
the driving state recognition method in step S5 is based on a gaussian mixture model, which is a probabilistic model belonging to an unsupervised machine learning classification algorithm, and is usually solved by using a maximum expectation algorithm. The gaussian mixture model assumes that all data points are mixed by a finite number of gaussian distributions containing unknown parameters, and can be expressed as a weighted function of the finite number of gaussian distributions, with the expression:
λ＝{w i ,μ i ,Σ i },i＝1,2,…,N
wherein p (x|lambda) is the probability distribution of the Gaussian mixture model, lambda is the parameter of the Gaussian mixture model, w i The weight term of the ith Gaussian distribution, x is a vector formed by D-dimensional continuous observation data, g (x|mu) i ,Σ i ) Probability density function, μ, as an ith Gaussian distribution i Is the mean value of the ith Gaussian distribution, Σ i For the covariance matrix of the ith Gaussian distribution, N is the number of Gaussian distributions, the invention divides driving states into stable and unstableTwo types are defined, so N is equal to 2;
the posterior probability P (i|x of the ith Gaussian distribution t Lambda) is as follows:
wherein x is t For a given T-dimensional training vector, x t ∈{x 1 ,x 2 ,...,x T }。
The posterior probability threshold is set to 0.5, and a gaussian distribution of 0.5 or more is determined to be stable, and a gaussian distribution of less than 0.5 is determined to be unstable.
Step S6: and (5) analyzing and determining the driving capability recovery time of each driver by using a driving state identification method.
When the driving state identification method is applied, setting the super parameters in the Gaussian mixture model as follows: the number of components is 2, the maximum iteration number is 1000, the covariance type is complete covariance, and the weight, the mean value and the precision are initialized by using a k-means method.
And respectively selecting the index combinations related to the transverse direction and the longitudinal direction by using a Gaussian mixture model, judging the driving state of the driver after taking over the vehicle in the transverse direction and the longitudinal direction, respectively determining the time for the driver to reach the stable state in the transverse direction and the longitudinal direction after taking over the vehicle, and taking the maximum value of the two as the driving capability recovery time of the driver.
The specific process is as follows:
step S61: the time length of each piece of vehicle track data is 60s, and the time interval is 1s, namely, each piece of vehicle track data has 60 track points;
step S62: extracting indexes reflecting the transverse and longitudinal driving states of the vehicle from each track point based on the method in the step S3, and screening index combinations with the best effect on reflecting the transverse and longitudinal driving states based on the method in the step S4;
step S63: respectively identifying whether the driving state of the vehicle in the transverse direction and the longitudinal direction is stable or unstable by utilizing the vehicle track data corresponding to the screened index combination and combining with a Gaussian mixture model;
step S64: in combination with the driving state recognition result of step S63 and the vehicle trajectory data used for driving state recognition, it is determined whether the driving state of the vehicle is stable or unstable every second.
Step S65: after the driver takes over the vehicle, the driving state is unstable and then evolves from unstable to stable, and the moment when the unstable state ends is the moment when the stable state starts, which is the driving state stabilization time.
In this embodiment, first, longitudinal driving state recognition analysis is performed, and as shown in fig. 6 (a) -6 (l), dark black dots in fig. 6 (a) -6 (l) represent unstable states, and light gray dots represent stable states. It is obvious that the data distribution (distribution 1) corresponding to fig. 6 (a) and the data distribution (distribution 2) corresponding to fig. 6 (j) are two completely different distributions, and the distribution 1 gradually transits and evolves to be the distribution 2 along with the time change, which indicates that the driving state identification research by using the gaussian mixture model is completely reasonable. Furthermore, the driving state is unstable for the front 20s behind the driver taking over the vehicle (see fig. 6 (a) to 6 (d)); 46 to 60s after taking over, the driving state is stable; 21 to 45s after taking over, the driving state transitions from unstable to stable state. According to the driving state identification result, the longitudinal driving state stability time of each driver can be calculated by combining the original vehicle track data. The longitudinal settling time of the driving state obeys a normal distribution with a mean value of 27.23s and a standard deviation of 3.67.
Subsequently, a lateral driving state recognition study is performed, the variables considered including x 10 ,x 11 ,x 12 ,x 13 ,x 14 And x 9 According to the above procedure, the study was developed, and found to contain the variable x 9 ,x 11 ,x 13 ,x 14 Data and reference data (variables contain x 9 ,x 10 ,x 11 ,x 12 ,x 13 ,x 14 ) The minimum distance between the two is used as a lateral driving state recognition analysis, and the recognition results are shown in fig. 7 (a) -7 (l), and it is known that the driving state of 1 to 10s after the driver takes over the vehicle is unstable, and the driving state of 31 to 60s isStable, the driving state transitions from unstable to stable between 11s and 30 s. According to the driving state identification result, the transverse driving state stable time of each driver can be calculated by combining the original vehicle track data, and the longitudinal stable time of the driving state obeys the normal distribution with the mean value of 17.37s and the standard deviation of 3.13.
And finally, aiming at the same driver, selecting the maximum value in the horizontal and vertical driving state stabilization time as the driving capability recovery time of the driver. The comparison can be used to obtain the driving capability recovery time of all drivers in the example, and the driving capability recovery time of all drivers is subjected to normal distribution with the average value of 27.25s and the standard deviation of 3.67.
According to the method, an automatic driving takeover experimental platform is built by using a driving simulator, an automatic driving takeover scene is designed and built by considering the influence of different experimental factors, a driver is recruited to carry out simulation takeover experiments, a plurality of indexes capable of representing driving states are extracted and calculated by combining vehicle track data after takeover is completed, a method for screening index combinations capable of reflecting the driving states is provided, the indexes are screened, a driving state identification method based on a machine learning algorithm is built, lateral driving state stability identification research and longitudinal driving state stability identification research are respectively carried out based on the screened multi-index combination data capable of reflecting the driving states, and lateral driving state stability time and longitudinal driving state stability time of each driver are respectively determined by combining driving state identification results and original data. Finally, the driving ability recovery time of all drivers is determined through comparison.
The invention provides a combination optimizing method for screening indexes representing driving states, and determining multi-index combinations which can reflect the driving states most, so that the unilaterality and the vulnerability of the single index representing the driving states are compensated, and the objectivity and the comprehensiveness of evaluation index selection are improved by the combination optimizing method; according to the invention, a Gaussian mixture model based on a machine learning algorithm is constructed, so that the driving state identification and classification analysis after pipe butt joint are realized, and the objectivity and reliability of an identification result are improved; the driving capability recovery time determining method provided by the invention can accurately calculate the driving capability recovery time of each driver under different takeover scenes, and improves the calculation accuracy of the driving capability recovery time. The research result can provide theoretical basis for the optimal design of the man-machine interaction interface of the automatic driving system, improves the taking-over performance of the driver, and has strong reference value.
The embodiments of the present invention have been described in detail with reference to the drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the spirit of the present invention.
Claims (10)
1. The method for determining the recovery time of the driving capability of the automatic driving takeover process of the expressway is characterized in that the takeover process of switching from the automatic driving state to the driving state of the driver is performed for the vehicle executing the automatic driving on the expressway, and the following steps S1-S6 are executed to complete the determination of the recovery time of the driving capability of the driver:
step S1: constructing an automatic driving takeover simulation platform based on a driving simulator, and constructing a takeover simulation test scene by considering the influence of different preset experimental factors;
step S2: based on the built takeover simulation test scene, carrying out takeover simulation experiments aiming at different drivers, wherein the takeover simulation experiments comprise preset takeover trigger events, when the preset takeover trigger events occur, the vehicles driven by the drivers are converted from an automatic driving state to a driver driving state, and vehicle track data of the vehicles driven by the drivers in the takeover simulation experiments are collected;
step S3: extracting each index reflecting the driving state according to the acquired vehicle track data in the takeover simulation experiment;
step S4: combining the indexes reflecting the driving state according to a preset rule, evaluating and quantifying the effect of reflecting the driving state of each index combination, sequencing the effect of reflecting the driving state of each index combination, and selecting the index combination with the best effect of reflecting the driving state;
step S5: constructing a driving state identification method based on a machine learning algorithm, and judging whether the driving state of a driver taking over the vehicle is stable or unstable based on the index combination with the best effect on reflecting the driving state, which is obtained in the step S4;
step S6: and (5) analyzing and determining the driving capability recovery time of each driver by using a driving state identification method.
2. The method for determining the driving ability recovery time of the highway autopilot take-over process according to claim 1, wherein the influences of the different experimental factors preset in step S1 include autopilot duration, take-over request time, autopilot speed, and preceding vehicle speed;
the automatic driving duration time is the time for the automatic driving system to independently control the vehicle to run or the time for the driver to execute the subtask during automatic driving, wherein the subtask is the behavior for distracting the driver, the take-over request time is the early warning time for the vehicle to send out the take-over request, the automatic driving speed is the running speed of the vehicle during automatic driving, and the front vehicle speed is the running speed of the vehicle in front of the vehicle when the vehicle sends out the take-over request.
3. The method for determining the driving ability recovery time of a highway autopilot process according to claim 1, wherein the preset takeover triggering event in step S2 includes a lane change and deceleration event, specifically, an autopilot vehicle traveling at a preset speed on a middle lane, and a front vehicle traveling in front of left is changed to the middle lane and decelerated.
4. The method for determining the driving ability recovery time of the highway automatic driving takeover process according to claim 1, wherein the vehicle track data collected in the step S2 includes the running track data of the vehicle and the preceding vehicle after the driver takes over the vehicle, the segment length of the running track data is 60S, and the collection frequency is 100Hz.
5. The highway automatic driving according to claim 4Method for determining the recovery time of the drivability of a vehicle, characterized in that the indicators reflecting the driving state extracted in step S3 comprise an average longitudinal speed x 1 Standard deviation x of longitudinal velocity 2 Average longitudinal acceleration x 3 Average heel-to-heel spacing x 4 Standard deviation x of heel-to-heel spacing 5 Mean value x of speed difference between front and rear vehicles 6 Standard deviation x of speed difference between front and rear vehicle 7 Average headway x 8 Time x after take over 9 Average lateral velocity x 10 Standard deviation of transverse velocity x 11 Average lateral acceleration x 12 Average lane offset x 13 Standard deviation x of lane offset 14 The method comprises the steps of carrying out a first treatment on the surface of the Wherein time x after take-over 9 Time 0 counted by the time when the driver takes over the vehicle.
6. The method for determining the recovery time of driving ability of an expressway automatic driving take over process according to claim 5, wherein in step S3, the indexes reflecting the driving state are combined by using a combined optimizing method, and the effect of reflecting the driving state is evaluated by using a bulldozing distance, which is the minimum cost required for converting one distribution into another, is calculated as follows:
assume thatFor a first distribution having m classes, where p i Is the ith class of P, +.>Is p i Corresponding weights, i e {1,2, …, m }; let->For another distribution having n classes, where q j J-th category of Q, +.>Is q j Corresponding weights, j e {1,2, …, n }; d= [ D ] ij ]Is a distance matrix, d ij For category p i And category q j The distance between them, the total cost l required to convert distribution P into distribution Q cost The formula is as follows:
wherein f i,j For category p i And category q j The amount of earth in between, the minimum cost required to convert the distribution P into the distribution Q is obtained according to the following constraint:
f i,j ≥0,1≤i≤m,1≤j≤n
solving the linearization problem to obtain the optimal valueThe bulldozing distance is as follows:
where EMD (P, Q) is the bulldozing distance converted from distribution P to distribution Q.
7. The method for determining the driving ability recovery time of a highway automatic driving takeover process according to claim 6, wherein the combined optimizing method comprises the steps of:
step S31: taking over the vehicle track data in the range of 60s after the driver takes over the vehicle in the collected take-over simulation experiment, extracting each index reflecting the driving state from the vehicle track data, and carrying out normalized conversion;
step S32: selecting K indexes of indexes reflecting driving states, wherein the K indexes do not contain the time x after taking over 9 Combining the two components by adopting a permutation and combination principle to obtainGroup index combination and corresponding vehicle track data thereof, and time x after taking over 9 Respectively adding the data into each group of index combinations, numbering the vehicle track data, and taking K indexes and the time x after taking over 9 As reference data;
step S33: reducing the dimension of the high-dimension vehicle track data corresponding to each group of index combinations obtained in the step S32 to two-dimensional data by adopting a t-SNE algorithm, respectively calculating bulldozing distances between the two-dimensional data corresponding to each group of index combinations and the two-dimensional data corresponding to the reference data, and carrying out descending order arrangement on the numerical values of each bulldozing distance;
step S34: the index combination corresponding to the minimum bulldozing distance is used as the index combination with the best effect for reflecting the driving state.
8. The driving ability restoration time determination method for an expressway automatic driving take over process according to claim 7, wherein the driving state identification method in step S5 is based on a gaussian mixture model, and the expression thereof is as follows:
λ＝{w i ,μ i ,Σ i },i＝1,2,…,N
wherein p (x|lambda) is the probability distribution of the Gaussian mixture model, lambda is the parameter of the Gaussian mixture model, w i The weight term of the ith Gaussian distribution, x is a vector formed by D-dimensional continuous observation data, g (x|mu) i ,Σ i ) Probability density function, μ, as an ith Gaussian distribution i Is the mean value of the ith Gaussian distribution, Σ i Covariance matrix of ith Gaussian distribution, N is number of Gaussian distribution;
the posterior probability P (i|x of the ith Gaussian distribution t Lambda) is as follows:
wherein x is t For a given T-dimensional training vector, x t ∈{x 1 ,x 2 ,...,x T }；
And (3) presetting a posterior probability threshold, judging that the Gaussian distribution which is larger than or equal to the preset posterior probability threshold is stable, and judging that the Gaussian distribution which is smaller than the preset posterior probability threshold is unstable.
9. The method for determining the driving ability recovery time of the highway automatic driving take over process according to claim 8, wherein the super parameters in the gaussian mixture model are set as follows in step S6: the number of components is 2, the maximum iteration number is 1000, the covariance type is complete covariance, and the weight, the mean value and the precision are initialized by using a k-means method.
10. The method for determining the recovery time of the driving ability of the highway automatic driving takeover process according to claim 9, wherein in step S6, a gaussian mixture model is applied, a combination of indicators related to the lateral direction and the longitudinal direction is selected, the driving state of the driver after taking over the vehicle is determined in the lateral direction and the longitudinal direction, the time for the driver to reach the stable state in the lateral direction and the longitudinal direction after taking over the vehicle is determined, and the maximum value of the two is taken as the recovery time of the driving ability of the driver.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310527781.9A CN116595453A (en) | 2023-05-11 | 2023-05-11 | Method for determining driving capability recovery time of highway automatic driving takeover process |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310527781.9A CN116595453A (en) | 2023-05-11 | 2023-05-11 | Method for determining driving capability recovery time of highway automatic driving takeover process |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116595453A true CN116595453A (en) | 2023-08-15 |
Family
ID=87594839
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310527781.9A Pending CN116595453A (en) | 2023-05-11 | 2023-05-11 | Method for determining driving capability recovery time of highway automatic driving takeover process |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116595453A (en) |
-
2023
- 2023-05-11 CN CN202310527781.9A patent/CN116595453A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107492251B (en) | Driver identity recognition and driving state monitoring method based on machine learning and deep learning | |
| CN107169567B (en) | Method and device for generating decision network model for automatic vehicle driving | |
| Hou et al. | Modeling mandatory lane changing using Bayes classifier and decision trees | |
| CN111738337B (en) | Driver distraction state detection and identification method in mixed traffic environment | |
| CN110386144B (en) | GHMM/GGAP-RBF hybrid model for identifying driver braking intention and identification method | |
| CN112863245B (en) | Vehicle track change real-time prediction method based on deep neural network | |
| Li et al. | Modeling vehicle merging position selection behaviors based on a finite mixture of linear regression models | |
| CN110619340B (en) | Method for generating lane change rule of automatic driving automobile | |
| CN111695737A (en) | Group target advancing trend prediction method based on LSTM neural network | |
| CN111907523B (en) | Vehicle following optimizing control method based on fuzzy reasoning | |
| CN114299607A (en) | Human-vehicle collision risk degree analysis method based on automatic driving of vehicle | |
| Toledo et al. | State dependence in lane-changing models | |
| CN115923833A (en) | Personifying decision control method and device for vehicle, vehicle and storage medium | |
| CN114169444B (en) | Driving style classification method considering risk potential field distribution under vehicle following working condition | |
| CN115422747A (en) | Method and device for calculating discharge amount of pollutants in tail gas of motor vehicle | |
| Yan et al. | A game-theoretical approach to driving decision making in highway scenarios | |
| CN116946183A (en) | Commercial vehicle driving behavior prediction method considering driving capability and vehicle equipment | |
| CN116894395A (en) | Automatic driving test scene generation method, system and storage medium | |
| CN116595453A (en) | Method for determining driving capability recovery time of highway automatic driving takeover process | |
| CN116588123A (en) | Risk perception early warning strategy method based on safety potential field model | |
| Chen et al. | Platoon separation strategy optimization method based on deep cognition of a driver’s behavior at signalized intersections | |
| Yi et al. | Data-driven situation awareness algorithm for vehicle lane change | |
| Deng et al. | Modeling and prediction of human behaviors based on driving data using multi-layer HMMs | |
| CN114701517A (en) | Multi-target complex traffic scene automatic driving solution based on reinforcement learning | |
| Ma et al. | Lane change analysis and prediction using mean impact value method and logistic regression model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |