CN118238847A - Autonomous lane change decision planning method and system adaptive to different driving styles and road surface environments - Google Patents

Abstract

The invention belongs to the technical field of automatic driving, and particularly relates to an autonomous lane change decision planning method and system adaptive to different driving styles and road surface environments, comprising the steps of acquiring driving tracks and road surface environment data, identifying to obtain different driving styles, cascading multidimensional decision vectors of the driving styles and the driving environments of a self-vehicle and a side vehicle, inputting the multi-dimensional decision vectors into a fully-connected neural network, and outputting lane change decision instructions; determining safety constraint, comfort constraint, self-adaptive acceleration constraint and self-adaptive jerk constraint according to road surface adhesion coefficients in road surface environment information, and determining a lane change track through a seven-order polynomial function in combination with a decision instruction; the invention designs a personalized lane change decision model based on a multi-neural network by combining the current traffic situation and the driving experience of a human driver, adopts self-adaptive constraint of road adhesion coefficient and speed to replace fixed constraint, thereby solving the problem of safe and reliable lane change decision planning under different road conditions.

Description

Autonomous lane change decision planning method and system adaptive to different driving styles and road surface environments

Technical Field

The invention belongs to the technical field of automatic driving, and particularly relates to an autonomous lane change decision planning method and system adaptive to different driving styles and road surface environments.

Background

Autopilot technology is becoming a new variable in the development of the automotive industry. The decision-making planning is one of three key technologies of the automatic driving vehicle, and plays an important role of a bridge connecting the upper sensing unit and the lower control unit. Statistics data show that improper lane changing behavior of vehicles and abrupt changes in road surface environment are main factors causing road traffic accidents and congestion.

The lane change decision method of the automatic driving vehicle in the prior art for coping with the unknown pavement environment also has the following problems:

1. the decision method based on the rules is established on the basis of the prior test rules, but the problems of low coverage of traffic scenes, difficult rule design, low expansibility and the like are faced; the decision method based on the algorithm does not fully consider the influence of the style change of the driver on the lane change decision accuracy in the complex traffic environment at present. In a similar driving scenario, some drivers choose to change lanes, while some choose to keep the current lane.

2. The influence of the existing decision-making planning strategy on the driving safety of the road surface environment is not fully studied, and the road surface environment, particularly the change of the road surface adhesion coefficient, can bring great challenges to the safe driving of the automatic driving vehicle.

3. Currently, most research reports use road adhesion coefficients as a friction ellipse-based system constraint. However, the resulting friction-based constraints (e.g., acceleration limits) are too large compared to comfort requirements. In addition to this, the upper limit defined by the comfort requirement is always set to a fixed value, and this approach is clearly not feasible when the road adhesion coefficient changes. In addition, only one or two different road friction conditions are often considered in the existing method, and the influence of the change of the road adhesion coefficient on the generation track is not fully studied yet.

Disclosure of Invention

The invention aims to provide an autonomous lane change decision planning method and system adaptive to different driving styles and road surface environments, so as to solve the problems in the background technology.

The invention realizes the above purpose through the following technical scheme:

In a first aspect, the present invention provides an autonomous lane change decision planning method adaptive to different driving styles and road surface environments, the method comprising:

acquiring driving track data of a self vehicle and a side vehicle and road surface environment data in a driving environment;

The driving track data of the self-vehicle and the side-vehicle are subjected to dimension reduction and clustering, then are identified to obtain different driving styles, multi-dimensional decision vectors of the driving styles and the driving environments of the self-vehicle and the side-vehicle are cascaded and then are input into a fully-connected neural network, and lane change decision instructions corresponding to the different driving styles are output by combining the driving track data of the self-vehicle and the side-vehicle;

determining a lane change track target constraint condition including safety constraint, comfort constraint, self-adaptive acceleration constraint and self-adaptive jerk constraint according to the road adhesion coefficient in the road environment information, and determining a lane change track through a seven-order polynomial function by combining the decision instruction and the target constraint condition;

and receiving the lane change track, responding to control signals for generating an accelerator and a steering, and feeding back to the automatic driving vehicle for lane change.

As a further optimization scheme of the invention, the steps of performing dimension reduction and clustering on the driving track data of the own vehicle and the side vehicle and then identifying to obtain different driving styles include:

The main component analysis method is adopted to carry out dimension reduction on the driving track data of the self-vehicle and the side-vehicle to obtain dimension reduction data, the density-based spatial noise application clustering method is adopted to carry out clustering on the dimension reduction data to obtain driving style clustering data, and the self-adaptive attention and residual refinement neural network-based neural network is adopted to identify the clustering data to obtain the driving styles of the self-vehicle and the side-vehicle,

The driving style and the corresponding vector thereof are set as follows: the conservative type is denoted as 1, the general type is denoted as 2, and the aggressive type is denoted as 3.

As a further optimization scheme of the present invention, the step of cascading the driving style and the multidimensional decision vectors of the driving environments of the own vehicle and the side vehicle and inputting the multi-dimensional decision vectors into the fully connected neural network, and the step of outputting the lane change decision instructions corresponding to different driving styles by combining the driving track data of the own vehicle and the side vehicle comprises:

Combining the driving style vectors of the own vehicle and the side vehicle with the multidimensional decision vector, inputting the multi-dimensional decision vector into a lane change probability function of a selected fully-connected neural network, and determining a decision instruction by combining a safety coefficient F _safe, a comfort coefficient F _comfort, an efficiency coefficient F _efficiency, a gain coefficient F _gain and the driving style F _style, wherein the method comprises the following steps:

In the above formula, y is a lane change decision vector of a three-element, corresponds to the probabilities of lane keeping, left lane changing and right lane changing, f _LC (·) is a lane changing probability function of a fully connected neural network using the above input, θ is a set of parameters of the model, d _LP represents the longitudinal distance between the host vehicle and the left front vehicle, d _RP represents the longitudinal distance between the host vehicle and the right front vehicle, d _CP represents the longitudinal distance between the host vehicle and the front vehicle on the current lane, d _th represents the minimum safe distance, x (t) and y (t) represent the longitudinal coordinates and the transverse coordinates, respectively, corresponding to the current driving time t, s (t) and v (t) represent the current vehicle driving distance and driving speed at the time t, v _CP represents the driving speed of the host vehicle, v _LP and v _RP represent the driving speeds of the front vehicle in the left lane and the right lane, d _LP represents the time distance between the host vehicle and the front vehicle on the current lane and the front vehicle, d _LP and d 76 of the front vehicle in the left lane and the right lane, respectively;

the decision instruction includes a lane keeping, a left lane change, or a right lane change.

As a further optimization of the present invention, in the security constraint:

The minimum safe distance d _th at the beginning of the lane change action is selected as:

Wherein, Is used for predicting the duration time of the positions of the own vehicle H and the front vehicle P in the course of lane change operation, and is setThe sizes of the own vehicle H and the front vehicle P are set to be the same, and l _v is the vehicle length of the own vehicle H and the front vehicle P,/>Is the maximum deceleration of the preceding vehicle,/>Wherein mu and g are road adhesion coefficient and gravitational acceleration respectively, and t _stop is the time interval when the preceding vehicle is completely stopped: t _stop＝v_P/μg,v_H is the running speed of the own vehicle H, v _P is the running speed of the preceding vehicle P, l _v is the vehicle length, and t _slc is the lane change time.

As a further optimization of the invention, the comfort constraint is: the maximum lateral acceleration a _ymax, the maximum lateral jerk j _ymax, and the lane change time t _slc are calculated from the vehicle maximum forward speed v _ymax and the road surface adhesion coefficient, and the comfort limit is determined therefrom.

As a further optimization scheme of the present invention, in the adaptive acceleration constraint: design self-adaptive acceleration limit changing along with road adhesion coefficient and road self-adaptive lateral acceleration limit The formula is as follows:

Wherein b ₀、b₁ and b ₂ are three undetermined coefficients, Mu _o and mu _p are the maximum and minimum values, respectively, of the road adhesion coefficient calculated based on the friction ellipses, as the upper limit of the lateral acceleration.

As a further optimization scheme of the invention, in the adaptive jerk constraint: the maximum lateral jerk j _ymax is set to the maximum allowable value to keep the shortest track-change duration, and the speed adaptive jerk limit is designed as a piecewise function:

Wherein, V _lim and v _th are the forward speed, the limit speed and the threshold speed of the own vehicle, respectively, in km/h, v _lim is selected as 120km/h, and c ₁～c₅ is a constant.

As a further optimization scheme of the present invention, the determining the lane-changing track by using the seventh order polynomial function in combination with the decision instruction and the target constraint condition includes:

the lane change time t _slc is approximately L _x/v_H, where L _x is the lane change starting point, and thus the lane change optimal trajectory selection problem is further translated into: determining the shortest lane change duration by the minimum value of t _slc under satisfaction of the safety and comfort constraints Thereby selecting the optimal lane change track adapting to the current road surface condition:

Selecting After that, then/>

The lane change track calculation formula based on the seven-order polynomial function is as follows:

Where l _w is the lane width, x is the longitudinal position of the own vehicle H relative to the lane change starting point S, Y (x) represents the lateral position of the own vehicle H and a _i, and a _i (i=1..4) is the four coefficients to be determined.

In a second aspect, the present invention provides an autonomous lane change decision planning system adaptive to different driving styles and road surface environments, which is applied to execute the autonomous lane change decision planning method described in any one of the above, and the system includes:

the environment sensing module is used for acquiring driving track data of a self vehicle and a side vehicle and road surface environment data in a driving environment;

The lane change decision module is used for carrying out dimension reduction and clustering on the driving track data of the own vehicle and the side vehicle, then identifying the driving track data to obtain different driving styles, cascading the driving styles and multidimensional decision vectors of the driving environment of the own vehicle and the side vehicle, inputting the multi-dimensional decision vectors into the fully-connected neural network, and outputting lane change decision instructions corresponding to different driving styles by combining the driving track data of the own vehicle and the side vehicle;

The track planning module is used for determining track changing track target constraint conditions including safety constraint, comfort constraint, self-adaptive acceleration constraint and self-adaptive jerk constraint according to road adhesion coefficients in road environment information, and determining track changing tracks adapting to different road environments through seven-order polynomial functions by combining the decision instruction and the target constraint conditions;

And the lane change control module is used for receiving the lane change track, responding to control signals for generating an accelerator and a steering, and feeding back to the automatic driving vehicle for lane change.

The invention has the beneficial effects that:

1. The invention designs a personalized lane change decision model based on a multi-neural network by combining the current traffic situation and the driving experience of a human driver, and redefines traffic factors for analyzing lane change decisions. The model solves the problem that the traditional lane change decision model is limited in a complex traffic environment by learning the lane change decision planning characteristics of a human driver. Realizing more comprehensive and flexible personalized channel switching decision.

2. The invention establishes a new standard for the comfort variant of an automatic driving vehicle by analyzing the driving comfort threshold value and the comfort driving characteristics of a human driver in the past research report. The standard adopts the self-adaptive constraint of road adhesion coefficient and speed to replace the fixed constraint, thereby solving the problem of safe and reliable lane change decision planning under different conditions.

3. The invention provides a road self-adaptive track planning scheme which creatively integrates road adhesion coefficient, safety, comfort and dynamic constraint defined by a human driving mode and considers individual differences of drivers. The scheme solves the problem that the existing model cannot adapt to road condition changes (especially road adhesion coefficient changes) in time, thereby improving the safety and instantaneity of lane change decision planning.

Drawings

FIG. 1 is a schematic flow chart of an autonomous channel change decision planning method according to an embodiment of the present application;

FIG. 2 is an overall frame diagram of an autonomous lane change decision making system in accordance with an embodiment of the present application;

FIG. 3 is a frame diagram of a lane change decision making method according to an embodiment of the present application;

FIG. 4 is a schematic diagram of a multi-lane-change trajectory planning in an embodiment of the present application;

FIG. 5 is a schematic diagram of a road surface adaptive track trajectory planning framework in accordance with an embodiment of the present application;

fig. 6 is a schematic diagram showing a comparison of lane change behavior of drivers in different driving styles in a case portion according to an embodiment of the present application;

Fig. 7 is a further schematic illustration of a comparison of driver lane change behavior for a portion of different driving styles for a case in an embodiment of the present application;

Fig. 8 is a schematic diagram of a result of a case-portion joint simulation experiment in an embodiment of the present application;

fig. 9 is a schematic diagram of a case part Quanser platform autopilot (QCar) experimental setup in an embodiment of the application;

Fig. 10 is a schematic diagram of experimental results of the case part QCar in the example of the present application.

Detailed Description

The present application will be described in further detail with reference to the accompanying drawings, wherein it is to be understood that the following detailed description is for the purpose of further illustrating the application only and is not to be construed as limiting the scope of the application, as various insubstantial modifications and adaptations of the application to those skilled in the art can be made in light of the foregoing disclosure.

Example 1

As shown in fig. 1, the present embodiment provides an autonomous lane change decision planning method and system adaptive to different driving styles and road surface environments, where the method includes:

S1, acquiring driving track data of a self vehicle and a side vehicle and road surface environment data in a driving environment;

s2, dimension reduction and clustering are carried out on the driving track data of the own vehicle and the driving track data of the side vehicle, then different driving styles are obtained through recognition, multi-dimensional decision vectors of the driving styles and the driving environments of the own vehicle and the driving environments of the side vehicle are input into a fully-connected neural network after cascade connection, and lane change decision instructions corresponding to the different driving styles are output by combining the driving style data of the own vehicle and the driving wind lattice data of the side vehicle;

S3, determining a lane change track target constraint condition including safety constraint, comfort constraint, self-adaptive acceleration constraint and self-adaptive jerk constraint according to the road adhesion coefficient in the road environment information, and determining a lane change track through a seven-order polynomial function by combining a decision instruction and the target constraint condition;

S4, receiving the lane changing track, responding to control signals for generating the accelerator and the steering, and feeding back to the automatic driving vehicle for lane changing.

More specifically, step S1 is specifically a data set selection and data preprocessing process, which specifically includes the following steps:

NGSIM the dataset is a public traffic safety dataset as part of a data collection project. The method is widely used by students in the traffic field and is used for researching driving behaviors such as vehicle tracking and lane changing, traffic flow analysis, micro-traffic model construction, vehicle motion trail prediction, driver intention recognition, automatic driving decision planning and the like. First, a CSV format data set is derived for subsequent use, and a lane change decision framework proposed by the present invention is shown in fig. 3.

Since there are a large number of data items of the actual driving data employed, filtering and extraction of data related to the behavior of the lane change are required. Specifically, it is first necessary to select vehicles classified as sedans and reject irrelevant data. In order to solve the noise caused by the system error and the measurement error in the sample data, the invention adopts a median filtering algorithm with the time window length of 5 to carry out smoothing. Then, the index corresponding to 100 frames before and after the lane change is determined, so that the key time in the lane change process is focused. Finally, the processed data set is derived for a subsequent driving style classification operation.

As a further preferable scheme, performing dimension reduction and clustering on the driving track data of the own vehicle and the side vehicle, and then identifying to obtain different driving styles comprises the following steps:

performing dimension reduction on driving track data of the vehicle and the side vehicle by adopting a Principal Component Analysis (PCA) to obtain dimension reduction data, clustering the dimension reduction data by adopting a density-based spatial noise application clustering method (DBSCAN) to obtain driving style clustered data, and identifying the clustered data by adopting a neural network based on self-adaptive attention and residual refinement to obtain driving styles of the vehicle and the side vehicle,

The driving style and its corresponding vector are set as: the conservative type is denoted as 1, the general type is denoted as 2, and the aggressive type is denoted as 3.

In order to learn the lane-changing decision strategy of the human driver, the invention designs a new learning-based decision model (i.e. corresponding to the step S2). The model not only considers the traffic environment information around the vehicle, but also considers the driving style of the vehicle and the surrounding obstacle vehicles. The design of the invention considers the complex lane changing scene of the multi-lane shown in fig. 4 and is used for redefining and analyzing traffic factors influencing lane changing decisions.

As a further preferable scheme, the method for outputting the lane change decision instruction corresponding to different driving styles by combining the driving track data of the own vehicle and the side vehicle comprises the following steps of:

Combining the self-vehicle and side-vehicle driving style vectors and the multidimensional decision vector cascade, inputting the multi-dimensional decision vector cascade into a lane change probability function of a selected fully-connected neural network, and determining a decision instruction by combining a safety coefficient F _safe, a comfort coefficient F _comfort, an efficiency coefficient F _efficiency, a gain coefficient F _gain and a driving style F _style, wherein the method comprises the following steps:

In the above formula, y is a lane change decision vector of a three-element, corresponds to the probabilities of lane keeping, left lane changing and right lane changing, f _LC (·) is a lane changing probability function of a fully connected neural network using the above input, θ is a set of parameters of the model, d _LP represents the longitudinal distance between the host vehicle and the left front vehicle, d _RP represents the longitudinal distance between the host vehicle and the right front vehicle, d _CP represents the longitudinal distance between the host vehicle and the front vehicle on the current lane, d _th represents the minimum safe distance, x (t) and y (t) represent the longitudinal coordinates and the transverse coordinates, respectively, corresponding to the current driving time t, s (t) and v (t) represent the current vehicle driving distance and driving speed at the time t, v _CP represents the driving speed of the host vehicle, v _LP and v _RP represent the driving speeds of the front vehicle in the left lane and the right lane, d _LP represents the time distance between the host vehicle and the front vehicle on the current lane and the front vehicle, d _LP and d 76 of the front vehicle in the left lane and the right lane, respectively;

Decision instructions include lane keeping, left lane change, or right lane change.

It will be appreciated that the present embodiment provides for the input to the Fully Connected Neural Network (FCNN) by concatenating the self-propelled cabin and surrounding vehicle driving style vector S with the other 13-dimensional vectors mentioned above. Next, using the previously processed NGSIM lane change trajectory dataset, the present invention extracts the input variables and decisions in the dataset from frame to frame and models them as a supervised learning problem. The goal of the learning is to find a model of f _θ (·) to minimize the long-term average loss. Finally, the model may output three types of results, namely lane keeping, left lane change, or right lane change.

The establishment of comfort standard for self-adaptive strain channel operation is also included in this embodiment:

After the lane change instruction is obtained from the decision module, the selection of the comfort criterion for achieving a comfortable automatic driving trajectory planning is a first challenge, which may vary greatly due to the different methods used or the different targets involved. Therefore, the invention combines the driving characteristics of human drivers and establishes more suitable standard for comfortable lane-changing driving operation.

After considering the acceleration and jerk limits suggested in the past studies in combination, it can be inferred that comfort thresholds of 2m/s ² and 3.6m/s ², respectively, would be sufficient to meet the requirements of lane-change operation. The lane-change trajectory planned according to the acceleration and jerk thresholds may be different from the trajectory assumed by the human driver. Therefore, it is necessary to consider the characteristics of the human driver in the track planning process, and it is found through literature arrangement that the maximum transverse speed and acceleration of more than 80% of the human driver are respectively in the ranges of [0.8,2.4] m/s, [0.5,4.0] m/s ², the total lane change time is in the ranges of [2.3,6.3] s, and the average duration of a single lane change operation of the vehicle is 4.6s.

The embodiment also comprises an automatic lane change planning scheme:

Seven-order polynomial lane change track:

The fifth-order polynomial in the prior art has the advantages of smooth curvature, closed expression, simple calculation and the like, and has been widely used for lane change track planning. However, the fifth order polynomial trajectory can only guarantee the continuity of acceleration. The derivative of acceleration (jerk) may be abrupt at the start point and not converge to zero at the end point of the lane change operation. Which is not acceptable for lane change trajectory planning. In order to achieve a more comfortable driving experience, the present invention contemplates achieving a zero rate of change trajectory across the course of the lane change. Based on the concise form of the fifth order polynomial curve, the calculation formula for obtaining the seven-order polynomial lane change track is as follows:

Where l _w is the lane width, x is the longitudinal position of the own vehicle H relative to the lane change starting point S, Y (x) represents the lateral position of the own vehicle H and a _i, and a _i (i=1..4) is the four coefficients to be determined.

As a further preferred solution, the safety constraint is first designed, the invention proposes a simple way of designing the safety distance, i.e. taking into account a worst case possible to be encountered by the vehicle H: the front vehicle P suddenly applies the maximum brake until it is completely stopped. In this case, the maximum deceleration of PExpressed in μg, where μ and g are road adhesion coefficient and gravitational acceleration, respectively. The time interval for the front vehicle to reach a complete stop can be obtained by: t _stop＝v_P/μg. Considering that the lateral position of the midpoint (Y _m) is 0.5l _w, H requires 0.5t _slc to cross the boundary between the current and target lanes, safety constraints are:

The minimum safe distance d _th at the beginning of the lane change action is selected as:

Wherein, Is used for predicting the duration time of the positions of the own vehicle H and the front vehicle P in the course of lane change operation, and is setIn order to avoid collision, assuming that the dimensions of H and P are the same, the longitudinal distance d _c between the CG of H and P should be greater than the vehicle length (l _v), and assuming that the dimensions of the own vehicle H and the preceding vehicle P are the same, l _v is the vehicle length of the own vehicle H and the preceding vehicle P,/>Is the maximum deceleration of the preceding vehicle,/>Wherein mu and g are road adhesion coefficient and gravitational acceleration respectively, and t _stop is the time interval when the preceding vehicle is completely stopped: t _stop＝v_P/μg,v_H is the running speed of the own vehicle H, v _P is the running speed of the preceding vehicle P, l _v is the vehicle length, and t _slc is the lane change time.

Correspondingly, more than 80% of human drivers have lane change durations between 2.3s and 6.3 s. Thus it is considered thatIt is sufficient to estimate d _th.

In terms of comfort constraints, previous studies on lane change trajectory planning generally only consider fixed comfort constraints, and not speed and road adhesion coefficients. However, variations in the vehicle forward speed and road adhesion coefficient may have an effect on acceleration and jerk limits during vehicle travel. Considering comfort limits that accommodate vehicle forward speeds and road attachment coefficients, more efficient trajectory planning may be achieved. A _ymax、j_ymax and t _slc are calculated from a given v _ymax. Taking the characteristics into consideration comprehensively, the comfort limit of the main characteristics of lane change operation is obtained. The values of all constraint parameters are shown in table 1:

TABLE 1 constraint parameter settings

Comfort constraints: the maximum lateral acceleration a _ymax, the maximum lateral jerk j _ymax, and the lane change time t _slc are calculated from the vehicle maximum forward speed v _ymax and the road surface adhesion coefficient, and the comfort limit is determined therefrom.

For adaptive acceleration constraints, the maximum longitudinal and lateral acceleration will change positively as the road attachment coefficient increases. For roads with very low friction (e.g. μ=0.2), the upper longitudinal and lateral acceleration limits are designed in table 1And/>While both within the comfort range, they are physically impractical. To solve this problem, a method is proposed: and designing the self-adaptive acceleration limit changing along with the road adhesion coefficient. Based on the friction ellipses, two minimum friction values can be calculated, so that the linear road adhesion coefficient self-adaptive lateral acceleration limit is obtained. On this basis, a suitable margin is introduced between the allowable a _ymax and the available acceleration μg to ensure that the vehicle can run normally also well below the road adhesion limit.

As a further preferred aspect, in the adaptive acceleration constraint: design self-adaptive acceleration limit changing along with road adhesion coefficient, and secondary road self-adaptive transverse acceleration limitThe formula is as follows:

Wherein b ₀、b₁ and b ₂ are three undetermined coefficients, Mu _o and mu _p are the maximum and minimum values, respectively, of the road adhesion coefficient calculated based on the friction ellipses, as the upper limit of the lateral acceleration.

For adaptive jerk constraints, the maximum lateral jerk j _ymax should be limited in order to ensure comfort and safety, although the seven-order polynomial trajectory exhibits continuity before the acceleration is abrupt. In general, j _ymax can be designed to decrease gradually with forward speed. When the vehicle speed increases from 0 to the speed limit (v _lim), the designed j _ymax will change fromReduced toThis is because faster lane-change operation corresponding to a larger jerk is more acceptable when the vehicle is traveling at a low speed; while in high speed driving, a small jerk should be ensured for safety. The threshold speed (v _th) was chosen to be 80km/h in combination with the usual driving experience. When the vehicle speed exceeds v _th, the designed j _ymax should be rapidly reduced to ensure driving safety and comfort. Instead, the low and medium speed designs j _ymax should be kept within a range around the limit value in order to perform lane-changing operations in time. To reduce the impact on traffic, the shortest lane change duration is maintained by setting the designed j _ymax to the maximum allowable value.

As a further preferred solution, in the adaptive jerk constraint: the maximum lateral jerk j _ymax is set to the maximum allowable value to keep the shortest track-change duration, and the speed adaptive jerk limit is designed as a piecewise function:

Wherein, V _lim and v _th are the forward speed, the limit speed and the threshold speed of the own vehicle, respectively, in km/h, v _lim is selected as 120km/h, and c ₁～c₅ is a constant.

As a further preferred solution, combining the decision instruction and the target constraint condition, determining the lane-change trajectory by a seventh order polynomial function includes:

the lane change time t _slc is approximately L _x/v_H, where L _x is the lane change starting point, and thus the lane change optimal trajectory selection problem is further translated into: determining the shortest lane change duration by the minimum value of t _slc under satisfaction of the safety and comfort constraints Thereby selecting the optimal lane change track adapting to the current road surface condition:

Selecting After that, then/>

The lane change track calculation formula based on the seven-order polynomial function is as follows:

Where l _w is the lane width, x is the longitudinal position of the own vehicle H relative to the lane change starting point S, Y (x) represents the lateral position of the own vehicle H and a _i, and a _i (i=1..4) is the four coefficients to be determined.

Based on the same inventive concept, the embodiment also provides an autonomous channel switching decision system corresponding to the autonomous channel switching decision method, and since the principle of solving the problem of the system in the embodiment of the disclosure is similar to that of the autonomous channel switching decision method in the embodiment of the disclosure, the implementation of the system can refer to the implementation of the method, and the repetition is omitted.

The invention combines the current traffic situation and the driving experience of the human driver, designs a personalized lane change decision model based on a multi-neural network, and solves the problem of limitation of the traditional lane change decision model in a complex traffic environment by learning lane change decision planning data of the human driver. Realizing more comprehensive and flexible personalized lane change decision.

With reference to fig. 2, this embodiment proposes an autonomous lane change decision planning system adaptive to different driving styles and road environments, which is applied to execute the above autonomous lane change decision planning method, and the system includes:

the environment sensing module is used for acquiring driving track data of a self vehicle and a side vehicle and road surface environment data in a driving environment;

The lane change decision module is used for carrying out dimension reduction and clustering on the driving track data of the own vehicle and the side vehicle, then identifying to obtain different driving styles, cascading the driving styles and the multidimensional decision vectors of the driving environment of the own vehicle and the side vehicle, inputting the multi-dimensional decision vectors into the fully-connected neural network, and outputting lane change decision instructions corresponding to the different driving styles by combining the driving track data of the own vehicle and the side vehicle;

The track planning module is used for determining track changing track target constraint conditions including safety constraint, comfort constraint, self-adaptive acceleration constraint and self-adaptive jerk constraint according to road adhesion coefficients in road surface environment information, and determining a track changing track through a seven-order polynomial function by combining decision instructions and the target constraint conditions;

The lane change control module is used for receiving the lane change track, responding to control signals for generating the throttle and steering, and feeding back to the automatic driving vehicle for lane change.

According to an embodiment of the invention, the invention establishes new standards for the comfort of an autonomous vehicle by analyzing driving comfort thresholds and comfort driving characteristics of human drivers in past research reports. And the self-adaptive constraint of road adhesion coefficient and speed is adopted to replace the fixed constraint, so that the method is used for intelligent and reliable lane change decision planning under different conditions. The proposed planning scheme creatively integrates dynamic constraints defined by road adhesion coefficient, safety, comfort and human driving mode, and takes the personalized difference of drivers into consideration. The scheme solves the problem that the existing model cannot adapt to road condition changes in time, thereby improving the safety and instantaneity of lane change decision planning.

The autonomous channel change decision method provided by the application is exemplified below with reference to a specific example.

In order to study the performance of the proposed decision planning scheme, firstly, the decision module is used for obtaining lane changing decisions of drivers in different styles under the same scene, and then the track planning module is used for planning a self-adaptive lane changing path according to the current road surface condition. And then, simulating the lane change tracks of different styles by using MATLAB/Simulink software. The feasibility of the personalized autonomous lane change decision and the pavement self-adaptive track planning method is evaluated by observing the vehicle response obtained in the CarSim platform. And finally, carrying out experimental verification by automatically driving the automobile (QCar) through a Quanser platform.

In the present invention, a scenario is designed in which the host vehicle H is holding on a dry high adhesion coefficient road surface (μ=0.8)Wherein a stationary obstacle vehicle P is located 50m in front of the vehicle, and at an initial time t ₀, CG having an origin of a global coordinate system XOY (fig. 4) of H is taken and an initial distance d ₀ of 50m. The key parameters are selected as follows, the vehicle length l _v =3.35 m; vehicle width t=1.74 m; lane width i _w =3.5 m and v _lim =120 km/h. Under the driving scene, the decision module obtains the conclusion that three groups of driving style drivers all select to change lanes leftwards, and then the trajectory planning module obtains the lane changing trajectory adapting to the friction coefficient. The lane change path, speed, acceleration and jerk of the driver with different driving styles are shown in fig. 6, in which μ=0.8,/>

It can be seen that for the same driving scenario, drivers of different driving styles will choose different lane-changing paths, in general, conservative lane-changing choices for drivers will pay more attention to safety and comfort, while aggressive drivers will choose to meet lane-changing demands in a faster way.

To verify the adaptability of our decision planning scheme to various road adhesion coefficient changes and speed changes, a road surface with low adhesion coefficient (mu=0.3) is designed, and the vehicle runs at medium speed Is a driving scene of (a). Wherein there is a stationary barrier vehicle P in front of the vehicle 50 m. The decision module also obtains the conclusion that three groups of drivers of driving style select to change lanes leftwards in the driving scene, and then the trajectory planning module obtains the lane changing trajectory which is suitable for the friction coefficient, the corresponding lane changing path, speed, acceleration and jerk changes of the drivers of different driving styles are shown as figure 7, in the figure, mu=0.3,

Since the driving behavior of a general driver in various driving styles is more general and representative, in order to verify the effectiveness of the decision-making strategy, the present invention uses a high-fidelity all-vehicle model for μ=0.8 in CarSim and MATLAB/Simulink,In the driving scene, the lane change behavior of a general driver is simulated.

Fig. 8 (a) -8 (c) show position, lateral velocity and lateral acceleration information during lane changes for a typical driving style driver. As shown in fig. 8 (d), the whole lane change process was continued for 1.84s, and the distance between the two vehicles was 10.3m when the lane change was completed. The lane changing process is safely realized.

To further verify the utility of the proposed lane-change trajectory planning scheme, experimental tests were performed on Quanser platform 1/10 model car QCar. The validity of the proposed decision planning scheme is verified by comparing the QCar response with the expected characteristics of the trajectory based on the scheme. The architecture of the experimental set-up is shown in figure 9.

In the experiment, the road friction coefficient μ=0.8. Considering QCar is a 1/10 scale model, and the actual lane width is typically 3.75m, we set the lane width toQCar has a width of 0.192m and a length of 0.425m. In selecting QCar speeds, we introduce a scaling factor ρ:

Wherein, Representing QCar a maximum speed of 2m/s,/>Corresponding to an actual speed of 33 m/s. We set the actual vehicle speed to/>The velocity corresponding to QCar is about/>=1.33 M/s. Distance between vehicle and obstacle/>/>

Fig. 10 illustrates QCar planning and tracking results based on the proposed lane change decision planning scheme. It can be seen that QCar starts the lane change operation at x=0.75m and completes the lane change around x=3.5m. However, significant tracking errors are observed near the start and end of the track change. The initial tracking error results from QCar rapid acceleration from its initial position to the desired speed, while the tracking error at the end point is due to the spatial constraints imposed by the laboratory environment in the room.

The above formulas are all formulas with dimensionality removed and numerical calculation, the formulas are formulas with the latest real situation obtained by software simulation through collecting a large amount of data, and preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

In addition, each functional module in the embodiments of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.

The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a random access memory (random access memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The above embodiments are only for illustrating the technical solution of the present application, and are not limiting; although the application has been described in detail with reference to the foregoing embodiments, it will be appreciated by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not drive the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims (9)

1. An autonomous lane change decision planning method adaptive to different driving styles and road surface environments is characterized by comprising the following steps:

acquiring driving track data of a self vehicle and a side vehicle and road surface environment data in a driving environment;

the driving track data of the self-vehicle and the side-vehicle are subjected to dimension reduction and clustering, then are identified to obtain different driving styles, multi-dimensional decision vectors of the driving styles and the driving environments of the self-vehicle and the side-vehicle are input into a fully-connected neural network after being cascaded, and lane change decision instructions corresponding to the different driving styles are output by combining the driving track data of the self-vehicle and the side-vehicle;

determining a lane change track target constraint condition including safety constraint, comfort constraint, self-adaptive acceleration constraint and self-adaptive jerk constraint according to the road adhesion coefficient in the road environment information, and determining a lane change track through a seven-order polynomial function by combining the decision instruction and the target constraint condition;

and receiving the lane change track, responding to control signals for generating an accelerator and a steering, and feeding back to the automatic driving vehicle for lane change.

2. The autonomous lane change decision making method adaptive to different driving styles and road surface environments according to claim 1, wherein the method comprises the following steps: the step of carrying out dimension reduction and clustering on the driving track data of the own vehicle and the side vehicle and then identifying to obtain different driving styles comprises the following steps:

The main component analysis method is adopted to carry out dimension reduction on the driving track data of the self-vehicle and the side-vehicle to obtain dimension reduction data, the density-based spatial noise application clustering method is adopted to carry out clustering on the dimension reduction data to obtain driving style clustering data, and the self-adaptive attention and residual refinement neural network-based neural network is adopted to identify the clustering data to obtain the driving styles of the self-vehicle and the side-vehicle,

The driving style and the corresponding vector thereof are set as follows: the conservative type is denoted as 1, the general type is denoted as 2, and the aggressive type is denoted as 3.

3. The autonomous lane change decision making method adaptive to different driving styles and road surface environments according to claim 2, wherein the method comprises the following steps: the step of cascading the driving style and the multidimensional decision vectors of the driving environments of the own vehicle and the side vehicle and inputting the multi-dimensional decision vectors into the fully-connected neural network, and the step of outputting the lane change decision instructions corresponding to different driving styles by combining the driving track data of the own vehicle and the side vehicle comprises the following steps:

Combining the driving style vectors of the own vehicle and the side vehicle with the multidimensional decision vector, inputting the multi-dimensional decision vector into a lane change probability function of a selected fully-connected neural network, and determining a decision instruction by combining a safety coefficient F _safe, a comfort coefficient F _comfort, an efficiency coefficient F _efficiency, a gain coefficient F _gain and the driving style F _style, wherein the method comprises the following steps:

In the above formula, y is a lane change decision vector of a three-element, corresponds to the probabilities of lane keeping, left lane changing and right lane changing, f _LC (·) is a lane changing probability function of a fully connected neural network using the above input, θ is a set of parameters of the model, d _LP represents the longitudinal distance between the host vehicle and the left front vehicle, d _RP represents the longitudinal distance between the host vehicle and the right front vehicle, d _CP represents the longitudinal distance between the host vehicle and the front vehicle on the current lane, d _th represents the minimum safe distance, x (t) and y (t) represent the longitudinal coordinates and the transverse coordinates, respectively, corresponding to the current driving time t, s (t) and v (t) represent the current vehicle driving distance and driving speed at the time t, v _CP represents the driving speed of the host vehicle, v _LP and v _RP represent the driving speeds of the front vehicle in the left lane and the right lane, d _LP represents the time distance between the host vehicle and the front vehicle on the current lane and the front vehicle, d _LP and d 76 of the front vehicle in the left lane and the right lane, respectively;

the decision instruction includes a lane keeping, a left lane change, or a right lane change.

4. An autonomous lane change decision making method adaptive to different driving styles and road surface environments according to claim 3, wherein: among the security constraints:

The minimum safe distance d _th at the beginning of the lane change action is selected as:

Wherein, Is used for predicting the duration time of the positions of the own vehicle H and the front vehicle P in the course of lane change operation, and is setThe sizes of the own vehicle H and the front vehicle P are set to be the same, and l _v is the vehicle length of the own vehicle H and the front vehicle P,/>Is the maximum deceleration of the preceding vehicle,/>Wherein mu and g are road adhesion coefficient and gravitational acceleration respectively, and t _stop is the time interval when the preceding vehicle is completely stopped: t _stop＝v_P/μg,v_H is the running speed of the own vehicle H, v _P is the running speed of the preceding vehicle P, l _v is the vehicle length, and t _slc is the lane change time.

5. The autonomous lane change decision making method adaptive to different driving styles and road surface environments according to claim 4, wherein the method comprises the following steps: among the comfort constraints are: the maximum lateral acceleration a _ymax, the maximum lateral jerk j _ymax, and the lane change time t _slc are calculated from the vehicle maximum forward speed v _ymax and the road surface adhesion coefficient, and the comfort limit is determined therefrom.

6. The autonomous lane change decision making method adaptive to different driving styles and road surface environments according to claim 5, wherein the method comprises the following steps: in the adaptive acceleration constraint: design self-adaptive acceleration limit changing along with road adhesion coefficient and road self-adaptive lateral acceleration limitThe formula is as follows:

Wherein b ₀、b₁ and b ₂ are three undetermined coefficients, Mu _o and mu _p are the maximum and minimum values, respectively, of the road adhesion coefficient calculated based on the friction ellipses, as the upper limit of the lateral acceleration.

7. The autonomous lane change decision making method adaptive to different driving styles and road surface environments according to claim 6, wherein the method comprises the following steps: in the adaptive jerk constraint: the maximum lateral jerk j _ymax is set to the maximum allowable value to keep the shortest track-change duration, and the speed adaptive jerk limit is designed as a piecewise function:

Wherein, V _lim and v _th are the forward speed, the limit speed and the threshold speed of the own vehicle, respectively, in km/h, v _lim is selected as 120km/h, and c ₁～c₅ is a constant.

8. The autonomous lane change decision making method for adapting to different driving styles and road surface environments according to claim 7, wherein: the step of determining the lane change track through a seven-order polynomial function by combining the decision instruction and the target constraint condition comprises the following steps:

the lane change time t _slc is approximately L _x/v_H, where L _x is the lane change starting point, and thus the lane change optimal trajectory selection problem is further translated into: determining the shortest lane change duration by the minimum value of t _slc under satisfaction of the safety and comfort constraints Thereby selecting the optimal lane change track adapting to the current road surface condition:

Selecting After that, then/>

The lane change track calculation formula based on the seven-order polynomial function is as follows:

Where l _w is the lane width, x is the longitudinal position of the own vehicle H relative to the lane change starting point S, Y (x) represents the lateral position of the own vehicle H and a _i, and a _i (i=1..4) is the four coefficients to be determined.

9. An autonomous lane change decision making system adapted to different driving styles and road surface environments, characterized by being applied to execute the autonomous lane change decision making method according to any one of claims 1-8, the system comprising:

the environment sensing module is used for acquiring driving track data of a self vehicle and a side vehicle and road surface environment data in a driving environment;

The lane change decision module is used for carrying out dimension reduction and clustering on the driving track data of the own vehicle and the side vehicle, then identifying the driving track data to obtain different driving styles, cascading the driving styles and multidimensional decision vectors of the driving environment of the own vehicle and the side vehicle, inputting the multi-dimensional decision vectors into the fully-connected neural network, and outputting lane change decision instructions corresponding to different driving styles by combining the driving track data of the own vehicle and the side vehicle;

The track planning module is used for determining track changing track target constraint conditions including safety constraint, comfort constraint, self-adaptive acceleration constraint and self-adaptive jerk constraint according to road adhesion coefficients in road environment information, and determining track changing tracks adapting to different road environments through seven-order polynomial functions by combining the decision instruction and the target constraint conditions;

And the lane change control module is used for receiving the lane change track, responding to control signals for generating an accelerator and a steering, and feeding back to the automatic driving vehicle for lane change.