CN110597245A - Automatic driving track-changing planning method based on quadratic planning and neural network - Google Patents

Abstract

The invention belongs to the field of intelligent transportation, and provides an automatic driving track-changing planning method based on quadratic programming and a neural network. The invention utilizes the information perception module to obtain the driving information of the lane changing vehicle and the vehicles around the lane changing vehicle; based on the vehicle running information, a transverse deviation recommending module gives a transverse deviation value so as to determine starting and ending point position information of the lane changing vehicle; based on the vehicle running information and the transverse deviation as input, the path planning module calculates a plane coordinate system static lane changing path of the lane changing vehicle; and finally, the track planning module gives out the change rule of the track changing vehicles and the motion tracks of the surrounding vehicles along with time, and the goal of planning the track changing process is completed. The invention can comprehensively consider the movement tracks of the lane changing vehicle and the vehicles around the lane changing vehicle on the premise of ensuring the safety of the lane changing vehicle, utilizes quadratic programming to quickly solve, and improves the speed and the comfort degree of the lane changing process, thereby meeting the requirements of automatic driving on the optimization of the lane changing subtask and the rapidity of solving.

Description

Automatic driving track-changing planning method based on quadratic planning and neural network

Technical Field

The invention relates to the field of intelligent transportation, in particular to an automatic driving track changing planning method based on quadratic programming and a neural network.

Background

With the progress of urbanization in China, traffic accidents and traffic jam caused by the rapid increase of urban vehicles are further aggravated. The automatic driving shows great potential in the aspects of reducing traffic accidents, relieving traffic jam and the like, and becomes a research hotspot of various automobile manufacturers. The lane changing behavior is one of basic behaviors in vehicle running, and research on the lane changing behavior is of great significance for improving the intelligent driving level of vehicles and increasing the road traffic capacity.

The lane change trajectory planning refers to that the vehicle calculates a space-time driving trajectory in a given future time period by comprehensively considering factors such as the position, the speed and the acceleration of the vehicle and surrounding vehicles in order to acquire the speed advantage or due to driving requirements and the like so as to ensure that the lane change behavior is smoothly and safely carried out. In general, the lane change process is divided into three phases: firstly, information perception is achieved, namely position, speed, acceleration and other information of a vehicle and surrounding vehicles are obtained through hardware facilities such as a sensor or V2I, and bottom layer input is conducted on a track planning process; secondly, planning a track, and calculating space-time track information in a given time period in the future based on the sensed information, wherein the space-time track information can be further represented as space position information and speed information changing along with time; and finally, for executing lane change, the vehicle completes a lane change task along the planned space-time trajectory based on the control module at the bottom layer. The track changing process is comprehensively considered, and the track changing track planning can be clearly characterized by the following remarkable characteristics: the control process has the disadvantages of complex consideration, large calculation amount and high real-time requirement.

The existing method for planning the track change track mainly comprises the following steps: a combination method and a mathematical programming method. The combination method can realize the track planning of the automatic driving vehicle in the real world by combining the grid division and the fast random search tree method. The combination method has been successfully applied to a plurality of actual projects, and stable space-time trajectories can be calculated in the automatic driving lane changing process. However, the combined method cannot realize the optimal search of the planned track, and simultaneously, comfort and safety constraints in the lane changing track are not considered in the method, so that certain defects exist in the real-time safety interaction of the lane changing vehicle and the surrounding vehicle information.

At present, scholars at home and abroad use a mathematical programming method to research the track changing planning problem of automatic vehicles. A mathematical programming method is provided, constraint conditions are established according to the comfort and safety of passengers, and an optimization model of an objective function is established according to the speed, the comfort and the like so as to plan a space-time trajectory in the lane changing process. In the process of solving the track changing track by using mathematical programming, in order to ensure the safety of the track changing process, a plurality of researches introduce a rectangle to describe the vehicle outline or a circle to describe the vehicle outline, and construct a series of non-convex constrained optimization problems. However, for the non-convex optimization problem, a set of standard and rapid solving method is not available at present, a multi-purpose heuristic algorithm is used for solving, and the requirements on absolute safety and real-time performance of lane change tracks in automatic driving cannot be met.

Disclosure of Invention

Aiming at the technical problems, the invention provides an automatic driving lane change trajectory planning method based on quadratic programming and a neural network, which can provide space-time trajectory support for realizing a lane change subtask in automatic driving by constructing an information perception module, a transverse deviation recommendation module, a path planning module and a trajectory planning module, and further ensure the stability, safety and comfort in the lane change process.

The method of the invention assumes that all vehicles run on a straight road, wherein the lane-changing vehicle target is changed to the target lane through the current lane, thereby completing the lane-changing behavior. In the invention, the lane where the lane changing vehicle is located is defined as a current lane, and the lane where the lane changing vehicle is to change is named as a target lane; the lane-changing vehicle is named as a lane-changing vehicle 0 (vehicle 0), a vehicle which is on the same current lane as the lane-changing vehicle 0 but is before the lane-changing vehicle is named as a vehicle 1 (vehicle 1) in front of the current lane, a vehicle which is on the target lane after the lane-changing vehicle 0 is named as a vehicle 2 (vehicle 2) behind the target lane, and a vehicle on the target lane before the lane-changing vehicle is named as a vehicle 3 (vehicle 3) in front of the target lane.

In the invention, a lane changing vehicle 0 is assumed, and the acceleration and deceleration of the lane changing vehicle is influenced by a vehicle 1 in the current lane in the lane changing process; during the lane changing process, the vehicle 0 is also affected by the vehicles 2 and 3 of the target lane, and the vehicle 2 can actively decelerate if necessary to ensure the safety of the vehicle 0 during the lane changing process.

The invention is realized by the following technical scheme:

an automatic driving track-changing planning method based on quadratic programming and a neural network comprises the following steps:

the method comprises the steps that driving information of lane changing vehicles and surrounding vehicles about positions, speeds, accelerations and jerks is obtained through an information sensing module;

the method comprises the steps that transverse deviation of a horizontal coordinate of a lane changing vehicle at the end of a lane changing path is obtained through a transverse deviation recommending module, so that the position of the lane changing vehicle at the end of the lane changing path is obtained, and basic input is provided for a path planning module;

based on the vehicle running information acquired by the information perception module and the transverse offset given by the transverse offset recommendation module, a path planning module is used for establishing a path model of a quintic polynomial for the lane change vehicle, a constraint condition of a start point and an end point of a lane change path is established, a Gaussian elimination method is adopted for solving a simultaneous equation set, parameters of the quintic polynomial are obtained, and a path with the given optimal transverse offset is generated;

discretizing the track planning time period through a track planning module to obtain a plurality of sub time periods, establishing driving distance, speed, acceleration and jerk variables in different sub time periods, taking kinematic constraint, anti-collision constraint, target lane rear vehicle jerk constraint and following model constraint as constraint conditions, taking the minimum jerk and the driving distance as objective functions, and solving by using a quadratic programming method to obtain track information of lane changing vehicles.

Further, the information sensing module obtains the position, speed and acceleration driving information of the lane-changing vehicle 0, the vehicle 1 in front of the current lane, the vehicle 2 behind the target lane and the vehicle 3 in front of the target lane, namely the vehicle 0, the vehicle 1, the vehicle 2 and the vehicle 3:

the travel information of the vehicle k at time t equal to 0 is represented as:

veh_k，t＝0，k＝{0，1，2，3}

veh_k，t＝0＝[x_k，t＝0，y_k，t＝0，v_k，t＝0，a_k，t＝0，j_k，t＝0]；

wherein x_k，t＝0，y_k，t＝0，v_k，t＝0，a_k，t＝0，j_k，t＝0An abscissa value, an ordinate value, a velocity, an acceleration, and a jerk representing the vehicle k at time t ═ 0, respectively;

when the time t is equal to 0, the vehicle 0 and the vehicle 1 are in the current lane, the vehicles 2 and 3 are in the target lane, a coordinate system is established by taking the current lane as a reference, D is the width of a single lane, and the position relation of the vehicles 0, 1, 2 and 3 is as follows:

further, the transverse deviation recommending module gives the transverse deviation delta x of the lane changing vehicle 0 at the end of the lane changing path_fFurther, the absolute coordinates of the lane change vehicle 0 at the time when the lane change path is completed, i.e., when t ═ f, are obtained:

x_0，t＝f＝x_0，t＝0+Δx_f，y0_，t＝f＝D

wherein the lateral offset recommending module gives a lateral offset Deltax_fThe method comprises the following steps:

(1) generating a certain number of random scenes, wherein the random scenes comprise driving information veh of the lane-changing vehicle 0, the vehicle 1 in front of the current lane, the vehicle 2 behind the target lane and the vehicle 3 in front of the target lane at the initial moment_k，t＝0，k＝{0，1，2，3}；

(2) Setting a lateral offset Deltax_fHas a lower bound of av_0，t＝0The upper bound is bv_3，t＝0Wherein a and b are time-dependent variables;

will interval [ av_0，t＝0，bv_3，t＝0]Evenly equally spaced into N_xPortions for each portion thereof Inputting a path planning module and a track planning module, and solving an objective function value by adopting a quadratic form planning model established by an objective function moduleWherein the content of the first and second substances,traversing i ∈ [0, N ]_x]To obtain a minimum objective function z_iCorresponding toLet the minimum objective function z_iCorresponding toFor optimum lateral offset deltax_fNamely:

construction of vehicle-integrated driving information veh_0，t＝0，veh_1，t＝0，veh_2，t＝0，veh_3，t＝0]For input, the optimum lateral offset Δ x_fIs the output sample;

(3) collecting data including a certain number of random scenes and optimal lateral deviation to construct a training set and a testing set; the neural network is adopted for training, so that a certain special random scene is given, and a target of transverse deviation is given quickly through neural network prediction.

Further, the path planning module formulates a path model of a fifth-order polynomial established for the path of the lane-changing vehicle, wherein the path model is expressed as:

in the formula, x_0，t，y_0，tRespectively indicates that the lane-change vehicle 0 is in the time period t ═ 0, f]T ═ 0 represents the start time of road changing path, t ═ f represents the end time of road changing path; alpha is alpha_iIs a parameter of a fifth order polynomial i ∈ [0, 1, 2, 3, 4, 5 ∈ [ ]]；

α_iThe method can solve corresponding values by a Gaussian elimination method through vehicle information of a starting time t-0 and an ending time t-f of a lane changing path of a lane changing vehicle 0, and specifically comprises the following steps:

the state equation of lane-changing vehicle 0 at start time t-0 and end time t-f is as follows:

wherein tau is_0，t，κ_0，tRespectively representing the derivative and curvature of the lane-change vehicle 0 at different times t, according to the requirement of the stability of the lane changing process, the path derivative and the curvature at the initial time t ═ 0 and the end time t ═ f of the lane changing are as follows: tau is_0，t＝0＝0，κ_0，t＝0＝0，τ_0，t＝f＝0，κ_0，t＝f＝0；

Solving simultaneous equations by adopting a Gaussian elimination method to obtain a parameter alpha of a fifth-order polynomial_iGenerating a given optimal lateral offset Deltax_fThe path of (2).

Further, the trajectory planning module is used for giving the running information of the lane changing vehicle 0 and the running information of the rear vehicle 2 of the target lane, which change along with time;

the track planning module plans the track for a time period [0, t_max]Discretized into I sub-periods, the ith time being denoted t_i，Wherein I belongs to {0, 1, 2.., I }; wherein the length of the neutron time interval is t ═ t_max/I；

Suppose that the time period t is planned_maxKnowing, the lane change time f is unknown; by s_dIndicated during the planned time period t_maxInner farthest possible distance, s_fRepresents the lane change path distance in the lane change time f, denoted by s_fTo s_dThe driving process of (2) is called as changing the road path extension area, then:

the trajectory planning module comprises a variable determination module, a constraint construction module, an objective function construction module and a quadratic programming solving module; the constraint construction module comprises a kinematics constraint module, an anti-collision constraint module, a target lane rear vehicle 2 jerk constraint module and a following model constraint module;

constructing, by the variable determination module, the following variables: travel distance s of lane-change vehicle 0_0，iVelocity v_0，iAcceleration a_0，iImpact j_0，iJerk j of rear vehicle 2 of target lane₂Lane changing vehicle 0 acceleration penalty term delta a_0，I(ii) a Respectively constructing a kinematic constraint condition, an anti-collision constraint condition, a target lane rear vehicle 2 impact degree constraint condition and a following model constraint condition through a kinematic constraint module, an anti-collision module, a target lane rear vehicle 2 impact degree constraint module and a following model constraint module in the constraint construction module(ii) a And constructing an objective function through the objective function construction module, and solving the track information of the lane changing vehicle through the quadratic programming solving module.

Further, the kinematic constraint module is used for establishing a driving distance s of the lane changing vehicle 0_0，iVelocity v_0，iAcceleration a_0，iAnd acceleration a_0，iThe kinematic constraint is constructed as:

Δt＝t_maxi is the length of the sub-period;

during a lane change of the vehicle, the speed v of the lane change vehicle 0_0，iAcceleration a_0，iJerk j_0，iMust not be out of a certain range, and is constrained by the following inequality:

0＜v_0，i＜vub，i＝1，2，...，I

alb＜a_0，i＜aub，i＝1，2，...，I

jlb＜j_0，i＜jub，i＝1，2，...，I

wherein vub, aub and jub respectively represent the running speed v of the lane-changing vehicle 0_0，iAcceleration a_0，iJerk j_0，iRespectively, alb, jlb represent the acceleration a of the vehicle_0，iJerk j_0，iThe lower limit of (d);

the target lane rear vehicle 2 jerk constraint module is used for taking jerk of the target lane rear vehicle 2 in the lane changing process along with the lane changing vehicle 0 as a variable, further obtaining a motion track of the target lane rear vehicle 2, and reflecting the interaction behavior of the lane changing vehicle 0 and the target lane rear vehicle 2 in the lane changing process;

the kinematic constraint of the rear vehicle 2 of the target lane is constructed as follows:

jlb＜j₂＜jub

alb＜a_2，I＜aub

v_2，I＞0

in the formula, a_2，IPlanning the end time t for the trajectory_IAcceleration of the vehicle 2 behind the target lane, a_2，I＝a_2，0+j₂t_I；v_2，IPlanning the end time t for the trajectory_IThe speed of the vehicle 2 behind the target lane,

further, the anti-collision constraint module is used for preventing the absolute distance between the lane changing vehicle 0 and the surrounding vehicles from being too small so as to avoid the occurrence of dangerous conditions; the vehicles around the lane changing vehicle 0 comprise a front vehicle 1 in the current lane, a rear vehicle 2 in the target lane and a front vehicle 3 in the target lane;

(1) the collision avoidance constraint construction is represented as:

s_f＜s_0，I＜s_d

x_1，i-xb_1，i-0.5vehl₁＞0，i＝1，2，...，I

xb_2，i-x_2，i-0.5vehl₂＞0，i＝1，2，...，I

x_3，i-xb_3，i-0.5vehl₃＞0，i＝1，2，...，I

in the formula, xb_1，i、xb_2，iAnd xb_3，iRespectively at time t_iCollision points of the lane-changing vehicle 0 with a front vehicle 1 of a current lane, a rear vehicle 2 of a target lane and a front vehicle 3 of the target lane; vehl₁、vehl₂And vehl₃The lengths of a front vehicle 1 of a current lane, a rear vehicle 2 of a target lane and a front vehicle 3 of the target lane are respectively set; x is the number of_1，i、x_2，iAnd x_3，iRespectively showing the vehicle 1 in front of the current lane, the vehicle 2 behind the target lane and the vehicle 3 in front of the target lane at the time t_iThe center of (a); wherein the front vehicle 1 of the current lane, the rear vehicle 2 of the target lane and the front vehicle 3 of the target lane are at the moment t_iThe center of (1) is:

the front vehicle 1 of the current lane and the front vehicle 3 of the target lane are ahead of the lane-changing vehicle 0 in the track planning time period [0, t ]_max]In the method, the law of motion of the vehicle 1 in front of the current lane and the vehicle 3 in front of the target lane can obtain information veh with t equal to 0 from the initial time_1，t＝0，veh_3，t＝0Calculating to obtain; the target lane rear vehicle 2 is behind the lane change vehicle 0, and based on the interactive consideration of the lane change process, the motion rule of the target lane rear vehicle 2 is the information veh obtained by the initial time t being 0_2，t＝0Sum variable vehicle 2 jerk j₂Jointly calculating to obtain;

(2) collision point xb in collision avoidance constraints_1，i、xb_2，iAnd xb_3，iCan be driven by the lane-changing vehicle 0 for a distance s_0，iSpecifically, the method comprises the following steps:

the collision point of the lane-change vehicle 0 is at time t_iIn the discontinuous change of the lane changing process, uniformly and discretely taking L samples, wherein in each sample L, the driving distance of a lane changing vehicle 0 is s_0，lAnd for each subsample L epsilon L through the path model in the path planning module, obtaining the driving distance s_0，lAnd corresponding collision point xb_1，l、xb_2，lAnd xb_3，l；

For each s_0，lDetermine its upper bound slb_0，iAnd the lower boundary sub_0，iBy using the upper bound slb_0，iAnd the lower boundary sub_0，iScreening of the subsample set L among L samples_iFor each subsample L ∈ L_iThe running distance satisfies sub_0，i＜s_0，l＜slb_0，i(ii) a Wherein, the upper boundary slb_0，iAnd lower boundary sub_0，iThe calculation formula of (2) is as follows:

slb_0，i＝v_0，0t_i

sub_0，i＝0.8v_3，0t_i

for each time period t_iBased on the subsample set L_iOn the abscissa xb of the impact point k_1，i、xb_2，iAnd xb_3，iEstablishing a linear fit:

xb_k，i＝α_k，i+β_k，is_0，i+ε_k，k＝1，2，3

in the formula, the collision point k considers three outer end boundary points of the vehicle, and the abscissa of the collision point is xb_k，iThe estimated value isα_k，iRepresents truncation,. beta._k，iRepresents the slope, ε_kRepresenting the random error term when linearly fitting the collision point k; therefore, the abscissa xb of the collision point k_k，iThe driving distance s of the lane-changing vehicle 0 can be used_0，iEstimating:

introducing maximum absolute error term delta e_k，iSo as to ensure the safety:

Δe_k，i＝max_i(|xb_k，i-α_k，i+β_k，is_0，i|)l_i∈L_i；

using s_0，iLinear fitting xb_k，iAnd k is 1, 2, 3, establishing collision avoidance constraint:

x_1，i-0.5vehl₁-(α_1，i+β_1，is_0，i+Δe_1，i)＞0，i＝1，2，...，I

α_2，i+β_2，is_0，i-Δe_2，i-(x_2，i+0.5vehl₂)＞0，i＝1，2，...，I

x_3，i-0.5vehl₃-(α_3，i+β_3，is_0，i+Δe_3，i)＞0，i＝1，2，...，I

wherein the distance s is traveled_0，lAnd corresponding collision point value xb_1，l、xb_2，lAnd xb_3，lThe calculating method of (2):

will be interval [ x_0，0，x_d]Evenly divided into L parts to form a sample set L, and for each sample L_lE.g. L, obtaining x through a path quintic polynomial function_0，l，y_0，l，s_0，l，By passingCalculating to obtain the course angle of the lane-changing vehicle at the current momentBy changing the position of the vehicle (x)_0，l，y_0，l) And the course angle of the lane changing vehicle at the current momentAnd vehicle length vehl₀Simplifying the lane-changing vehicle 0 into a rectangle to obtain the coordinates cx of four corners₁，cx₂，cx₃，cx₄And intersection cx of four sides and lane boundary₅，cx₆，cx₇，cx₈；

Let point cx₁，cx₂，cx₃，cx₄，cx₅，cx₆，cx₇，cx₈Forming a point set omega, and respectively calculating a point cx_qE.g. ordinate of ΩAnd the abscissaLet the set of points in the target lane be omega^UPoint set omega^UIn The set of points in the current lane is omega^LPoint set omega^LIn

Collision point xb_1，l、xb_2，lAnd xb_3，lFrom cx_qCalculating by epsilon to obtain:

further, the following model constraint module is used for completing the trajectory planning at the moment t_IEnabling the rear vehicle 2 of the target lane and the lane changing vehicle 0 as well as the lane changing vehicle 0 and the front vehicle 3 of the target lane to meet a following model, and avoiding collision after the track planning is finished;

using the linear car following model of Pariotata, assuming car n is behind car n-1, the acceleration of car n needs to satisfy:

a_n＝ω₁(Δx_n-Δx^*)+ω₂Δv_n

in the formula,. DELTA.x_nRepresents the distance, Δ x, between vehicle n and vehicle n-1_n＝x_n-1-x_n；Δv_nRepresents the speed difference between vehicle n and vehicle n-1, Δ v_n＝v_n-1-v_n；ω₁，ω₂，Δx^*Parameters calibrated for the linear following model;

at the end time t of trajectory planning_IThe lane change vehicle 0 and the target lane front vehicle 3 need to satisfy the following constraints on acceleration, distance, and speed difference:

a_0，I≤ω₁(x_3，I-x_0，I-0.5vehl₃-0.5vehl₀-Δx^*)+ω₂(v_3，I-v_0，I)

in the formula, v_3，IRepresenting the front vehicle 3 of the target lane at the track planning end time t_IVelocity v of_3，I＝v_3，0+a_3，0t_I；

x_3，IRepresenting the front vehicle 3 of the target lane at the track planning end time t_IThe abscissa of the (c) axis of the (c),

x_0，Irepresenting the lane-changing vehicle 0 at the end of the trajectory planning time t_IAbscissa of (a), x_0，I＝x_0，f+s_0，I-s_f；

At the end time t of trajectory planning_IThe target lane rear vehicle 2 and the lane change vehicle 0 need to satisfy the following constraints on acceleration, distance, and speed difference:

a_2，I≤ω₁(x_0，I-x_2，I-0.5vehl₀-0.5vehl₂-Δx^*)+ω₂(v_0，I-v_2，I)

in the formula: x is the number of_0，I-x_2，IFor the vehicle 2 and the lane-changing vehicle 0 at the track planning end time t_IThe distance of (d); position x of rear vehicle 2 of target lane_2，IAnd velocity v_2，IFrom jerk j of rear vehicle 2 of the target lane₂And calculating to obtain:

further, the target function building module takes the lane changing vehicle 0 and the target lane rear vehicle 2 affected by lane changing as research objects, comprehensively considers safety, comfort and rapidness as targets, and builds a corresponding quadratic programming model:

in the formula, mu and lambda are respectively set parameters;

an objective function representing a spatio-temporal trajectory plan;

Δa_0，I＝a_0，I-[ω₁(x_3，I-x_0，I-0.5vehl₃-0.5vehl₀-Δx^*)+ω₂(v_3，I-v_0，I)]。

the invention has the beneficial effects that:

(1) the invention provides a method for rapidly solving a lane change space-time trajectory based on quadratic programming, and meanwhile, linear anti-collision constraint is established based on the driving information of lane change vehicles and surrounding vehicles. Therefore, the method can effectively guarantee the safety in the lane changing vehicle process, simultaneously establishes the anti-collision constraint as the linear constraint, further can utilize quadratic programming to solve, and effectively reduces the solving complexity and the calculated amount, thereby meeting the requirement of real-time rapidity of the solving process in automatic driving and providing effective guarantee for the engineering practice of the automatic driving lane changing subtask.

(2) In the traditional lane change track, the given lane change time and the speed of a vehicle near a lane change vehicle are generally assumed to meet certain conditions, but the time of the lane change process and the speed and the acceleration of the lane change vehicle and the surrounding vehicles are not mandatory when the lane change space-time track is calculated by the method. Therefore, the method is more suitable for real complex time-varying traffic conditions.

(3) The traditional combination method generally adopts the steps of generating a transverse offset set, generating a track changing track for each transverse offset, constructing a plurality of track changing tracks, optimizing a target function and determining an optimal track changing track. In addition, the traditional combination method adopts a traversal method, so that the optimal lane changing track can be determined, and the method is relatively stable; however, the combination method needs to provide multiple tracks for a certain amount of lateral deviation, and the calculation amount is large, so that the requirement of real-time performance of track changing track planning in engineering practice is not necessarily met. The method adopts a method of recommending the transverse deviation by the neural network, can quickly predict the optimal transverse deviation, and further directly provides the optimal lane changing track. The method for recommending the lateral deviation by the neural network can quickly predict the optimized lateral deviation so as to directly calculate the optimal track changing track, can remarkably reduce the calculated amount in the track changing track while ensuring the optimization performance of the track changing track, and provides convenience for the planning of the track changing track in engineering practice.

(4) The research of the traditional scheme focuses on the time period in the lane changing process, but the research of the state of the lane changing vehicle at the end of the trajectory planning is lacked. In a common lane change trajectory planning, during the transition from a lane change process to a following process of a vehicle, the acceleration of the lane change vehicle at the end of the trajectory planning may be too low, so that the distance between the lane change vehicle and a front vehicle is too large in the following process, thereby causing the waste of space-time efficiency, and the distance between the lane change vehicle and a rear vehicle is sharply reduced, so that danger is easily caused. The method of the invention comprehensively considers the following model and the lane changing process, establishes the restriction of the running state of the lane changing vehicle at the end of the track planning time, ensures the safe and stable running of the vehicle after the track planning is finished, ensures that some extreme situations can not occur after the track planning, and further provides the guarantee for the stable transition from the lane changing process to the following process of the lane changing vehicle.

(5) The method of the invention takes into account the interactivity of the lane-change vehicle with the following vehicle (vehicle 2). The method takes the jerk of a vehicle (vehicle 2) behind a target lane of a lane changing vehicle as a variable, and constructs a corresponding objective function and a constraint condition. Therefore, the method of the invention can consider the track changing track of the jerk variable of the vehicle 2, and further can check whether the current track changing track is within the acceptable capability of the vehicle 2, if the current track changing track is within the acceptable range, the jerk of the vehicle 2 is adjusted to provide convenience for the track changing process of the track changing vehicle, and the track changing process is ensured to be carried out safely and smoothly.

Therefore, the invention provides a feasible method, which expresses the vehicle safety constraint by using a linear function, so that a mature standard method for solving a linear convex function, namely quadratic programming can be utilized for solving, and the method can simultaneously meet the requirements of real-time performance and absolute safety in the track change planning in the automatic driving process.

Drawings

Fig. 1 is a schematic structural diagram of an automatic driving track-changing planning method based on quadratic programming and a neural network in an embodiment of the present invention;

FIG. 2 is a schematic diagram of a lane change process scenario in an embodiment of the present invention;

FIG. 3 is a schematic diagram of a track change path and a track change track according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of a collision point in the lane change process in the embodiment of the present invention;

FIG. 5 is a schematic diagram illustrating a solution of collision points during a lane change process according to an embodiment of the present invention;

FIG. 6 is a schematic diagram of a collision-resistant neutron sample construction during a lane change process in an embodiment of the present invention;

fig. 7 is a schematic diagram of a neural network structure according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

On the contrary, the invention is intended to cover alternatives, modifications, equivalents and alternatives which may be included within the spirit and scope of the invention as defined by the appended claims. Furthermore, in the following detailed description of the present invention, certain specific details are set forth in order to provide a better understanding of the present invention. It will be apparent to one skilled in the art that the present invention may be practiced without these specific details.

The embodiment of the invention provides an automatic driving track-changing planning method based on quadratic programming and a neural network, which is characterized by comprising the following steps of:

the method comprises the steps that driving information of lane changing vehicles and surrounding vehicles about positions, speeds, accelerations and jerks is obtained through an information sensing module;

the method comprises the steps that transverse deviation of a horizontal coordinate of a lane changing vehicle at the end of a lane changing path is obtained through a transverse deviation recommending module, so that the position of the lane changing vehicle at the end of the lane changing path is obtained, and basic input is provided for a path planning module;

based on the vehicle running information acquired by the information perception module and the transverse offset given by the transverse offset recommendation module, a path planning module is used for establishing a path model of a quintic polynomial for the lane change vehicle, a constraint condition of a start point and an end point of a lane change path is established, a Gaussian elimination method is adopted for solving a simultaneous equation set, parameters of the quintic polynomial are obtained, and a path with the given optimal transverse offset is generated;

discretizing the track planning time period through a track planning module to obtain a plurality of sub time periods, establishing driving distance, speed, acceleration and jerk variables in different sub time periods, taking kinematic constraint, anti-collision constraint, target lane rear vehicle jerk constraint and following model constraint as constraint conditions, taking the minimum jerk and the driving distance as objective functions, and solving by using a quadratic programming method to obtain track information of lane changing vehicles.

As shown in fig. 3, the present invention distinguishes two concepts of a lane change path and a lane change track. The transverse deviation recommending module and the path generating module mainly research a lane changing path and aim to provide lane changing spatial path information for a lane changing vehicle, and the motion change condition of the lane changing vehicle along with time is not considered at the moment. And the track generation module is mainly used for researching the track changing track and comprises the motion information of the track changing vehicle on the space and time. The lane change trajectory may be further divided into a lane change route and a lane change route extension area (straight travel area) in terms of space. Through the trajectory generation module, the lane-change vehicle can obtain motion information such as position, speed, acceleration and jerk over time during the entire lane-change process.

In the present embodiment, the lane-change vehicle 0 is in a traffic situation as shown in fig. 2, assuming that all vehicles travel on a straight road, wherein the lane-change vehicle target is changed to an expressway through a current low-speed road to increase the travel speed and reduce the travel time. The lane where the lane changing vehicle is located is defined as a current lane, and the lane where the lane changing vehicle is to change is named as a target lane; the lane-changing vehicle is named as a lane-changing vehicle 0 (vehicle 0), a vehicle which is on the same current lane as the lane-changing vehicle 0 but is before the lane-changing vehicle is named as a vehicle 1 (vehicle 1) in front of the current lane, a vehicle which is on the target lane after the lane-changing vehicle 0 is named as a vehicle 2 (vehicle 2) behind the target lane, and a vehicle on the target lane before the lane-changing vehicle is named as a vehicle 3 (vehicle 3) in front of the target lane.

In this embodiment, the information sensing module obtains the position, speed and acceleration driving information of the lane-changing vehicle 0, the vehicle 1 in front of the current lane, the vehicle 2 behind the target lane, and the vehicle 3 in front of the target lane, namely, the vehicle 0, the vehicle 1, the vehicle 2, and the vehicle 3:

the travel information of the vehicle k at time t equal to 0 is represented as:

veh_k，t＝0，k＝{0，1，2，3}

veh_k，t＝0＝[x_k，t＝0，y_k，t＝0，v_k，t＝0，a_k，t＝0，j_k，t＝0]；

wherein x_k，t＝0，y_k，t＝0，v_k，t＝0，a_k，t＝0，j_k，t＝0Respectively representAn abscissa value, an ordinate value, a speed, an acceleration, and a jerk of the vehicle k at time t ═ 0;

when the time t is equal to 0, as shown in fig. 2, the vehicle 0 and the vehicle 1 are in the current lane, the vehicles 2 and 3 are in the target lane, a coordinate system is established with the current lane as a reference, D is the width of a single lane, and the positional relationship of the vehicles 0, 1, 2 and 3 is as follows:

preferably, the information sensing module obtains the driving information through a hardware device such as a sensor or V2I (vehicle-to-infrastructure communication).

In this embodiment, the lateral deviation recommending module may give the lateral deviation Δ x of the lane-changing vehicle 0 at the end of the lane-changing process_fAnd further, the absolute coordinates of the lane changing vehicle at the end of the lane changing process can be obtained to provide basic input for the path planning module. The absolute coordinates at the time when the road change path ends t ═ f are expressed as follows:

x_0，t＝f＝x_0，t＝0+Δx_f，y_0，t＝f＝D

the lateral offset recommendation module may accomplish the following tasks: generating a certain number of random scenes, wherein the generated random scenes represent the driving information of lane changing vehicles and surrounding vehicles; dividing the horizontal deviation of each random scene at equal intervals, inputting the horizontal deviation of the equal interval division into the input of a path planning module and a track planning module to calculate an objective function value, traversing the horizontal deviation of the equal interval division, and minimizing the objective function to determine the optimal horizontal deviation of the current random scene; collecting data including a certain number of random scenes and optimal lateral deviation to construct a training set and a testing set; the data set is trained by adopting a neural network, so that a certain special random scene is given, and a target of transverse deviation is given quickly through neural network prediction.

The method specifically comprises the following steps:

step 201, generating a certain number of random scenes including lane-changing vehicle 0 and surroundingDriving information veh of vehicles 1, 2 and 3_k，t＝0k＝{0，1，2，3}；

Step 202, setting a lateral offset Δ x_fHas a lower bound of av_0，t＝0The upper bound is bv_3，t＝0Where a and b are time-dependent variables, typically set to a-2 and b-4;

will interval [ av_0，t＝0，bv_3，t＝0]Evenly equally spaced into N_xPortions for each portion thereof Inputting a path planning module and a track planning module, and solving an objective function value by adopting a quadratic form planning model established by an objective function moduleWherein the content of the first and second substances,where N is_xCorresponding values may be set according to actual requirements, accuracy and computing power. Further, traverse i ∈ [0, N_x]To obtain a minimum objective function z_iCorresponding toLet the minimum objective function z_iCorresponding toFor optimum lateral offset deltax_fNamely:

constructing driving information [ veh ] integrated with vehicles_0，t＝0，veh_1，t＝0，veh_2，t＝0，veh_3，t＝0]To input,. DELTA.x_fAre samples of the output.

Step 203, the transverse deviation recommending module generates a series of random scenes, and determines the optimal transverse deviation through steps 202 and 203, so as to construct a training set and a testing set; the data set is trained by adopting a neural network, so that a certain special random scene is given, and a target of transverse deviation is given quickly through neural network prediction.

In this embodiment, as shown in fig. 3, the path planning module establishes a path model of a quintic polynomial for the path of the lane change vehicle based on the vehicle driving information acquired by the information sensing module and the lateral offset given by the lateral offset recommending module, establishes a constraint condition of starting and ending points of the lane change vehicle, solves a simultaneous equation set by using a gaussian elimination method, obtains parameters of the quintic polynomial, and further generates a path with the given optimal lateral offset.

The path planning module is used for establishing a path model of a fifth-order polynomial on the path of the lane changing vehicle, and the formula expression is as follows:

in the formula, x_0，t，y_0，tRespectively indicates that the lane-change vehicle 0 is in the time period t ═ 0, f]T ═ 0 represents the start time of road changing path, t ═ f represents the end time of road changing path; alpha is alpha_iIs a parameter of a fifth order polynomial i ∈ [0, 1, 2, 3, 4, 5 ∈ [ ]]；

α_iThe method can solve corresponding values by a Gaussian elimination method through vehicle information of a starting time t-0 and an ending time t-f of a lane changing path of a lane changing vehicle 0, and specifically comprises the following steps:

the state equation of lane-changing vehicle 0 at start time t-0 and end time t-f is as follows:

wherein tau is_0，t，κ_0，tRespectively representing the derivative and curvature of the lane-change vehicle 0 at different times t, according to the requirement of the stability of the lane changing process, the path derivative and the curvature at the initial time t ═ 0 and the end time t ═ f of the lane changing are as follows: tau is_0，t＝0＝0，κ_0，t＝0＝0，τ_0，t＝f＝0，κ_0，t＝f＝0；

Solving simultaneous equations by adopting a Gaussian elimination method to obtain a parameter alpha of a fifth-order polynomial_iGenerating a given optimal lateral offset Deltax_fThe path of (2).

In the present embodiment, the trajectory planning module is configured to provide the running information of the lane-changing vehicle 0 and the running information of the target lane following vehicle 2, which vary with time. The path planning module gives space static information of horizontal and vertical coordinate changes of the lane changing vehicles in the lane changing process, but the track planning module aims to give related driving information of the lane changing vehicles changing with time in the lane changing process. In order to quickly and effectively solve the space-time trajectory, the trajectory planning module adopts a time discretization method to discretize a given time period into a certain number of sub-time periods, the driving information of the sub-time periods is used as a variable, a comprehensive objective function of the lane changing process is optimized, and therefore the driving information of the sub-time periods is obtained, and the driving information of the lane changing vehicle changing along with the time in the trajectory planning process can be further obtained through interpolation.

The track planning module plans the track for a time period [0, t_max]Discretized into I sub-periods, the ith time being denoted t_i，Wherein I belongs to {0, 1, 2.., I }; wherein the length of the neutron time interval is t ═ t_max/I；

Suppose that the time period t is planned_maxKnowing, the lane change time f is unknown; by s_dIndicated during the planned time period t_maxInner farthest possible distance, s_fRepresents the lane change path distance in the lane change time f, denoted by s_fTo s_dThe driving process of (2) is called a lane change path extension area, as shown in fig. 3, then:

the trajectory planning module comprises a constraint construction module, an objective function construction module and a quadratic programming solving module; the constraint construction module comprises a kinematics constraint module, an anti-collision constraint module, a target lane rear vehicle 2 jerk constraint module and a following model constraint module;

constructing, by the variable determination module, the following variables: travel distance s of lane-change vehicle 0_0，iVelocity v_0，iAcceleration a_0，iImpact j_0，iJerk j of rear vehicle 2 of target lane₂Lane changing vehicle 0 acceleration penalty term delta a_0，I(ii) a Respectively constructing a kinematic constraint condition, an anti-collision constraint condition, a target lane rear vehicle 2 impact degree constraint condition and a following model constraint condition through a kinematic constraint module, an anti-collision module, a target lane rear vehicle 2 impact degree constraint module and a following model constraint module in the constraint construction module; and constructing an objective function through the objective function construction module, and solving the track information of the lane changing vehicle through the quadratic programming solving module.

The trajectory planning module comprises the following variables: travel distance s of lane-change vehicle 0_0，iVelocity v_0，iAcceleration a_0，iJerk j_0，iJerk j of rear vehicle 2 of target lane₂Lane changing vehicle 0 acceleration penalty term delta a_0，I. Wherein the distance s is traveled_0，iVelocity v_0，iAcceleration a_0，iJerk j_0，iMeaning represented is that the lane-change vehicle 0 is at time t_iThe running information of (2). Wherein the acceleration j is added₂Representing the vehicle 2 during the time period [0, t_max]The jerk of (1). Acceleration penalty term Δ a_0，IExpressing the difference in acceleration between vehicle 0 and vehicle 3 at the end of the trajectory planning to avoid | Δ a_0，IIf the | is too large, the vehicle 0 needs to accelerate immediately when the trajectory planning is finished, so that the speed change of the vehicle 0 is too large, and great discomfort is brought to passengers; the item adopts a Lagrange relaxation method, and an item is added to the target function to facilitate solving; wherein, Δ a_0，IThe calculation of (a) is defined as:

Δa_0，I＝a_0，I-[ω₁(x_3，I-x_0，I-0.5vehl₃-0.5vehl₀-Δx^*)+ω₂(v_3，I-v_0，I)]minimization of an objective functionAiming at reducing the track planning end time t_IAcceleration difference between car 0 and car 3 to avoid | Δ a_0，IIf | is too large, the vehicle 0 needs to accelerate immediately when the trajectory planning is finished, which causes too large speed change of the vehicle 0 and brings great discomfort to passengers.

Respectively constructing a kinematic constraint condition, an anti-collision constraint condition, a target lane rear vehicle 2 jerk constraint condition and a following model constraint condition through a kinematic constraint module, an anti-collision module, a target lane rear vehicle 2 jerk constraint module and a following model constraint module in the constraint construction module;

the kinematic constraint module is used for establishing a driving distance s of a lane changing vehicle 0_0，iVelocity v_0，iAcceleration a_0，iAnd acceleration a_0，iThe link between them and some constraints in the course of the travel of the vehicle 2.

(1) Kinematic constraint of car 0:

Δt＝t_maxi is the length of the sub-period;

during a lane change of the vehicle, the speed v of the lane change vehicle 0_0，iAcceleration a_0，iJerk j_0，iMust not be out of a certain range, and is constrained by the following inequality:

0＜v_0，i＜vub，i＝1，2，...，I

alb＜a_0，i＜aub，i＝1，2，...，I

jlb＜j_0，i＜jub，i＝1，2，...，I

wherein vub, aub and jub respectively represent the running speed v of the lane-changing vehicle 0_0，iAcceleration a_0，iJerk j_0，iRespectively, alb, jlb represent the acceleration a of the vehicle_0，iJerk j_0，iThe lower limit of (3).

(2) Kinematic constraint of vehicle 2

The target lane rear vehicle 2 jerk constraint module is used for taking jerk of the vehicle 2 in the lane changing process along with the lane changing vehicle 0 as a variable, further obtaining a motion track of the vehicle 2 and reflecting the interaction behavior of the lane changing vehicle 0 and the vehicle 2 in the lane changing process;

the kinematic constraint of the rear vehicle 2 of the target lane is constructed as follows:

jlb＜j₂＜jub

alb＜a_2，I＜aub

v_2，I＞0

in the formula, a_2，IPlanning the end time t for the trajectory_IAcceleration of the vehicle 2, a_2，I＝a_2，0+j₂t_I；v_2，IPlanning the end time t for the trajectory_IVehicle 2The speed of the motor vehicle is set to be,

the anti-collision constraint module is used for preventing the absolute distance between the lane changing vehicle 0 and the surrounding vehicles from being too small so as to avoid the occurrence of dangerous conditions; the vehicles around the lane changing vehicle 0 comprise a front vehicle 1 in the current lane, a rear vehicle 2 in the target lane and a front vehicle 3 in the target lane;

(1) the collision avoidance constraint construction is represented as:

s_f＜s_0，I＜s_d

x_1，i-xb_1，i-0.5vehl₁＞0，i＝1，2，...，I

xb_2，i-x_2，i-0.5vehl₂＞0，i＝1，2，...，I

x_3，i-xb_3，i-0.5vehl₃＞0，i＝1，2，...，I

in the formula, xb_1，i、xb_2，iAnd xb_3，iRespectively at time t_iThe collision points of the lane-change vehicle 0 with the vehicles 1, 2, and 3, as shown in fig. 4 and 5; vehl₁、vehl₂And vehl₃The lengths of a front vehicle 1 of a current lane, a rear vehicle 2 of a target lane and a front vehicle 3 of the target lane are respectively set; x is the number of_1，i、x_2，iAnd x_3，iRespectively, of cars 1, 2 and 3 at time t_iThe center of (a); wherein the vehicles 1, 2 and 3 are at time t_iThe center of (1) is:

in this embodiment, cars 1 and 3 are in front of lane-change car 0 and therefore on the track gaugeTime interval [0, t_max]The law of motion of vehicles 1 and 3 may be derived from the initial time by obtaining the information veh that t is 0_1，t＝0，veh_3，t＝0Calculating to obtain; the vehicle 2 is behind the lane change vehicle 0, and based on the interactive consideration of the lane change process, the motion rule of the vehicle 2 is the information veh obtained by the initial time t being 0_2，t＝0Sum variable vehicle 2 jerk j₂Jointly calculating to obtain;

(2) collision point xb in collision avoidance constraints_1，i、xb_2，iAnd xb_3，iThe distance s can be driven by the lane-changing vehicle 0_0，iIt was found that this is shown in FIG. 4. Wherein the distance s is travelled_0，iAnd the collision point xb_1，i、xb_2，iAnd xb_3，iThe functional relationship of (a) is a discontinuous nonlinear function, which is not beneficial to constructing a quadratic programming with linear conditions as constraints for fast solving. In order to express the anti-collision constraint by using a linear function, the embodiment of the invention adopts the driving distance s of the lane changing vehicle 0_0，iLinear fitting collision point xb_1，i、xb_2，iAnd xb_3，iThe method of (1).

Due to the collision point of the lane-changing vehicle being at time t_iIn the discontinuous change, the embodiment of the invention provides that L samples are uniformly and discretely taken in the lane changing process, and the driving distance s can be obtained for each sample L belonging to the path function in the path planning module of L_0，lAnd corresponding collision point xb_1，l、xb_2，lAnd xb_3，l。

For each s_0，iDetermine its possible upper and lower bounds slb_0，i，sub_0，iUsing an upper bound of slb_0，iLower boundary sub_0，iScreening of the subsample set L among L samples_iFor each subsample L ∈ L_iThe running distance of which satisfies sub_0，i＜s_0，l＜slb_0，i. Wherein, the upper boundary slb_0，iLower boundary sub_0，iThe calculation formula of (2) is as follows:

slb_0，i＝v_0，0t_i

sub_0，i＝0.8v_3，0t_i

further, for each time period t_iBased on the subsample set L_iTo collision point xb_1，i、xb_2，iAnd xb_3，iA linear fit can be established.

xb_k，i＝α_k，i+β_k，is_0，i+ε_k，k＝1，2，3

In the formula, the collision point k considers three outer end boundary points of the vehicle, and the abscissa of the collision point is xb_k，iThe estimated value isα_k，iRepresents truncation,. beta._k，iRepresents the slope, ε_kRepresenting the random error term when linearly fitting the collision point k; therefore, the abscissa xb of the collision point k_k，iThe driving distance s of the lane-changing vehicle 0 can be used_0，iEstimating: meanwhile, in order to ensure safety, a maximum absolute error term delta e is introduced_k，iSo as to ensure the safety: Δ e_k，i＝max_i(|xb_k，i-α_k，i+β_k，is_0，i|)l_i∈L_i。

Using s_0，iLinear fitting xb_k，ik ═ 1, 2, 3 can establish collision avoidance constraints:

x_1，i-0.5vehl₁-(α_1，i+β_1，is_0，i+Δe_1，i)＞0i＝1，2，...，I

α_2，i+β_2，is_0，i-Δe_2，i-(x_2，i+0.5vehl₂)＞0i＝1，2，...，I

x_3，i-0.5vehl₃-(α_3，i+β_3，is_0，i+Δe_3，i)＞0i＝1，2，...，I

further, wherein the distance s travelled_0，lAnd corresponding collision point value xb_1，l、xb_2，lAnd xb_3，lThe calculating method of (2):

will be interval [ x_0，0，x_d]Evenly divided into L parts to form a sample set L, and for each sample L_lE.g. L, obtaining x through a path quintic polynomial function_0，l，y_0，l，s_0，l，By passingCalculating to obtain the course angle of the lane-changing vehicle at the current momentBy changing the position of the vehicle (x)_0，l，y_0，l) And the course angle of the lane changing vehicle at the current momentAnd vehicle length vehl₀Simplifying the lane-changing vehicle 0 into a rectangle to obtain the coordinates cx of four corners₁，cx₂，cx₃，cx₄And intersection cx of four sides and lane boundary₅，cx₆，cx₇，cx₈；

Let point cx₁，cx₂，cx₃，cx₄，cx₅，cx₆，cx₇，cx₈Forming a point set omega, and respectively calculating a point cx_qE.g. ordinate of ΩAnd the abscissaLet the set of points in the target lane be omega^UPoint set omega^UIn The set of points in the current lane is omega^LPoint set omega^LIn

Collision point xb_1，l、xb_2，lAnd xb_3，lFrom cx_qCalculating by epsilon to obtain:

therefore, the travel distance s of L samples can be obtained using a uniform dispersion method_0，lAnd corresponding collision point value xb_1，l、xb_2，lAnd xb_3，l。

The following model constraint module is used for completing the track planning at the moment t_IAnd the rear vehicle 2 of the target lane and the lane changing vehicle 0 as well as the lane changing vehicle 0 and the front vehicle 3 of the target lane need to meet the following model, so that collision is avoided after the track planning is finished.

In the embodiment, a linear following model of Pariota is adopted, and it is assumed that the acceleration of the car n needs to satisfy after the car n-1:

a_n＝ω₁(Δx_n-Δx^*)+ω₂Δv_n

in the formula,. DELTA.x_nRepresents the distance, Δ x, between vehicle n and vehicle n-1_n＝x_n-₁-x_n。Δv_nRepresents the speed difference between vehicle n and vehicle n-1, Δ v_n＝v_n-1-v_n。ω₁，ω₂，Δx^*The parameters calibrated for the linear following model preferably take values of 0.0343, 0.948 and 30 respectively.

At the end time t of trajectory planning_IThe lane change vehicle 0 and the target lane front vehicle 3 need to satisfy the following constraints on acceleration, distance, and speed difference:

a_0，I≤ω₁(x_3，I-x_0，I-0.5vehl₃-0.5vehl₀-Δx^*)+ω₂(v_3，I-v_0，I)

in the formula, v_3，IRepresenting the front vehicle 3 of the target lane at the track planning end time t_IVelocity v of_3，I＝v_3，0+a_3，0t_I；

x_3，IRepresenting the front vehicle 3 of the target lane at the track planning end time t_IThe abscissa of the (c) axis of the (c),

x_0，Irepresenting the lane-changing vehicle 0 at the end of the trajectory planning time t_IAbscissa of (a), x_0，I＝x_0，f+s_0，I-s_f；

At the end time t of trajectory planning_IThe target lane rear vehicle 2 and the lane change vehicle 0 need to satisfy the following constraints on acceleration, distance, and speed difference:

a_2，I≤ω₁(x_0，I-x_2，I-0.5vehl₀-0.5vehl₂-Δx^*)+ω₂(v_0，I-v_2，I)

in the formula: x is the number of_0，I-x_2，IFor the vehicle 2 and the lane-changing vehicle 0 at the track planning end time t_IThe distance of (d); position x of rear vehicle 2 of target lane_2，IAnd velocity v_2，IFrom jerk j of rear vehicle 2 of the target lane₂And calculating to obtain:

in this embodiment, the objective function construction module takes the lane change vehicle 0 and the target lane rear vehicle 2 affected by lane change as research objects, and establishes a corresponding quadratic programming model by taking safety, comfort and rapidness as targets:

wherein mu and lambda are respectively set parameters, and the values are-0.04 and 1000;

an objective function representing a spatio-temporal trajectory plan;

the invention aims to improve the comfort of the lane-changing vehicle, thereby minimizing the acceleration of the lane-changing vehicle during the lane-changing processAnd acceleration

In the lane changing process, the lane changing vehicle 0 inevitably influences the normal running of the vehicle 2, and the vehicle 2 is used for acceleration j in the invention₂To reflect the change, however, the ideal state of the vehicle 2 is not influenced by the lane-changing vehicle 0, the driving rule at the initial moment can be kept, and the invention uses minimizationSo as to reduce the influence of the lane changing process on the subsequent vehicles;

the invention plans the track of the lane changing vehicle based on the given time length, wherein the lower the time of the lane changing process, the more the safety is guaranteed. Given a planned time length t_maxA faster lane-change procedure means that the distance traveled by the lane-change vehicle is greater during the planned time period, and the present invention therefore aims to maximizeThe parameter mu therefore assumes a negative value of-0.04 and is expressed in terms of a minimization target.

Δa_0，I＝a_0，I-[ω₁(x_3，I-x_0，I-0.5vehl₃-0.5vehl₀-Δx^*)+ω₂(v_3，I-v_0，I)]Minimization of an objective functionAiming at reducing the track planning end time t_IAcceleration difference between car 0 and car 3 to avoid | Δ a_0，IIf | is too large, the vehicle 0 needs to accelerate immediately when the trajectory planning is finished, which causes too large speed change of the vehicle 0 and brings great discomfort to passengers.

In this embodiment, the quadratic programming model is composed of an objective function and a constraint condition, and is solved by using a quadprog function of matlab.

In summary, the automatic driving lane change trajectory planning method based on quadratic programming and neural network obtains the driving information of the lane change vehicle and the vehicles around the lane change vehicle by using the information sensing module, gives the transverse offset for finishing the lane change process by using the transverse offset recommending module, and further gives the start and end point position information of the lane change vehicle in the lane change process. And obtaining a static plane coordinate system lane change path by using a path planning module based on the driving information of the lane change vehicle in the lane change process as input. And obtaining the motion rule of the lane changing vehicle along with the change of time based on the given static path and the running information of the surrounding lane changing vehicles through a track planning module. The invention can comprehensively consider the movement tracks of the lane changing vehicles and the vehicles around the lane changing vehicles on the premise of ensuring the safety of the lane changing vehicles, and utilizes quadratic programming to quickly solve, thereby improving the speed and the comfort degree of the lane changing process. Therefore, the optimization of the automatic driving to the lane changing subtask is met, and the requirement of rapidity is met.

Claims (9)

1. An automatic driving track-changing planning method based on quadratic programming and a neural network is characterized by comprising the following steps:

the method comprises the steps that driving information of lane changing vehicles and surrounding vehicles about positions, speeds, accelerations and jerks is obtained through an information sensing module;

the method comprises the steps that transverse deviation of a horizontal coordinate of a lane changing vehicle at the end of a lane changing path is obtained through a transverse deviation recommending module, so that the position of the lane changing vehicle at the end of the lane changing path is obtained, and basic input is provided for a path planning module;

based on the vehicle running information acquired by the information perception module and the transverse offset given by the transverse offset recommendation module, a path planning module is used for establishing a path model of a quintic polynomial for the lane change vehicle, a constraint condition of a start point and an end point of a lane change path is established, a Gaussian elimination method is adopted for solving a simultaneous equation set, parameters of the quintic polynomial are obtained, and a path with the given optimal transverse offset is generated;

discretizing the track planning time period through a track planning module to obtain a plurality of sub time periods, establishing driving distance, speed, acceleration and jerk variables in different sub time periods, taking kinematic constraint, anti-collision constraint, target lane rear vehicle jerk constraint and following model constraint as constraint conditions, taking the minimum jerk and the driving distance as objective functions, and solving by using a quadratic programming method to obtain track information of lane changing vehicles.

2. The quadratic programming and neural network-based automatic driving lane change trajectory planning method according to claim 1, wherein the information sensing module obtains driving information of lane change vehicle 0, vehicle 1 ahead of the current lane, vehicle 2 behind the target lane, and vehicle 3 ahead of the target lane, namely, position, speed and acceleration of vehicle 0, vehicle 1, vehicle 2, and vehicle 3:

the travel information of the vehicle k at time t equal to 0 is represented as:

veh_k，t＝0，k＝{0，1，2，3}

veh_k，t＝0＝[x_k，t＝0，y_k，t＝0，v_k，t＝0，a_k，t＝0，j_k，t＝0]；

wherein x_k，t＝0，y_k，t＝0，v_k，t＝0，a_k，t＝0，j_k，t＝0An abscissa value, an ordinate value, a velocity, an acceleration, and a jerk representing the vehicle k at time t ═ 0, respectively;

when the time t is equal to 0, the vehicle 0 and the vehicle 1 are in the current lane, the vehicles 2 and 3 are in the target lane, a coordinate system is established by taking the current lane as a reference, D is the width of a single lane, and the position relation of the vehicles 0, 1, 2 and 3 is as follows:

3. the quadratic programming and neural network-based automatic driving lane change trajectory planning method according to claim 1, wherein the transverse deviation recommending module gives the transverse deviation Δ x of the lane change vehicle 0 at the end of the lane change path_fFurther, the absolute coordinates of the lane change vehicle 0 at the time when the lane change path is completed, i.e., when t ═ f, are obtained:

x_0，t＝f＝x_0，t＝0+Δx_f，y_0，t＝f＝D

wherein the lateral offset recommending module gives a lateral offset Deltax_fThe method comprises the following steps:

(1) generating a certain number of random scenes, wherein the random scenes comprise driving information veh of the lane-changing vehicle 0, the vehicle 1 in front of the current lane, the vehicle 2 behind the target lane and the vehicle 3 in front of the target lane at the initial moment_k，t＝0，k＝{0，1，2，3}；

(2) Setting a lateral offset Deltax_fHas a lower bound of av_0，t＝0The upper bound is bv_3，t＝0Wherein a and b are time-dependent variables;

will interval [ av_0，t＝0，bv_3，t＝0]Evenly equally spaced into N_xPortions for each portion thereof Inputting a path planning module and a track planning module, and solving an objective function value by adopting a quadratic form planning model established by an objective function moduleTraversing i ∈ [0, N ]_x]To obtain a minimum objective function z_iCorresponding toLet the minimum objective function z_iCorresponding toFor optimum lateral offset deltax_fNamely:

construction of vehicle-integrated driving information veh_0，t＝0，veh_1，t＝0，veh_2，t＝0，veh_3，t＝0]For input, the optimum lateral offset Δ x_fIs the output sample;

(3) collecting data including a certain number of random scenes and optimal lateral deviation to construct a training set and a testing set; the neural network is adopted for training, so that a certain special random scene is given, and a target of transverse deviation is given quickly through neural network prediction.

4. The automated driving lane-changing trajectory planning method based on quadratic programming and neural network as claimed in claim 1, wherein the path planning module builds a path model of a fifth-order polynomial on the path of the lane-changing vehicle, and the path model is formulated as:

in the formula, x_0，t，y_0，tRespectively indicates that the lane-change vehicle 0 is in the time period t ═ 0, f]T ═ 0 represents the start time of road changing path, t ═ f represents the end time of road changing path; alpha is alpha_iIs a parameter of a fifth order polynomial i ∈ [0, 1, 2, 3, 4, 5 ∈ [ ]]；

α_iThe method can solve corresponding values by a Gaussian elimination method through vehicle information of a starting time t-0 and an ending time t-f of a lane changing path of a lane changing vehicle 0, and specifically comprises the following steps:

the state equation of lane-changing vehicle 0 at start time t-0 and end time t-f is as follows:

wherein tau is_0，t，κ_0，tRespectively representing the derivative and curvature of the lane-change vehicle 0 at different times t, according to the requirement of the stability of the lane changing process, the path derivative and the curvature at the initial time t ═ 0 and the end time t ═ f of the lane changing are as follows: tau is_0，t＝0＝0，κ_0，t＝0＝0，τ_0，t＝f＝0，κ_0，t＝f＝0；

Solving simultaneous equations by adopting a Gaussian elimination method to obtain a parameter alpha of a fifth-order polynomial_iGenerating a given optimal lateral offset Deltax_fThe path of (2).

5. The automated driving lane change trajectory planning method based on quadratic programming and neural network according to claim 1, wherein the trajectory planning module is used for giving the running information of the lane change vehicle 0 and the running information of the target lane rear vehicle 2 which change along with time;

the track planning module plans the track for a time period [0, t_max]Discretized into I sub-periods, the ith time being denoted t_i，Wherein I belongs to {0, 1, 2.., I }; wherein the length of the neutron time interval is t ═ t_max/I；

Suppose that the time period t is planned_maxKnowing, the lane change time f is unknown; by s_dIndicated during the planned time period t_maxInner farthest possible distance, s_fRepresents the lane change path distance in the lane change time f, denoted by s_fTo s_dThe driving process of (2) is called as changing the road path extension area, then:

the trajectory planning module comprises a variable determination module, a constraint construction module, an objective function construction module and a quadratic programming solving module; the constraint construction module comprises a kinematics constraint module, an anti-collision constraint module, a target lane rear vehicle 2 jerk constraint module and a following model constraint module;

constructing, by the variable determination module, the following variables: travel distance s of lane-change vehicle 0_0，iVelocity v_0，iAcceleration a_0，iImpact j_0，iJerk j of rear vehicle 2 of target lane₂Lane changing vehicle 0 acceleration penalty term delta a_0，I(ii) a Through a kinematic constraint module, an anti-collision module and a target lane rear vehicle 2-stroke module in the constraint construction moduleThe impact degree constraint module and the following model constraint module respectively construct a kinematic constraint condition, an anti-collision constraint condition, an impact degree constraint condition of a rear vehicle 2 of the target lane and a following model constraint condition; and constructing an objective function through the objective function construction module, and solving the track information of the lane changing vehicle through the quadratic programming solving module.

6. The quadratic programming and neural network-based automatic driving lane-changing trajectory planning method according to claim 5, wherein the kinematic constraint module is used for establishing a 0-driving distance s of a lane-changing vehicle_0，iVelocity v_0，iAcceleration a_0，iAnd acceleration a_0，iThe kinematic constraint is constructed as:

Δt＝t_maxi is the length of the sub-period;

during a lane change of the vehicle, the speed v of the lane change vehicle 0_0，iAcceleration a_0，iJerk j_0，iMust not be out of a certain range, and is constrained by the following inequality:

0＜v_0，i＜vub，i＝1，2，...，I

alb＜a_0，i＜aub，i＝1，2，...，I

jlb＜j_0，i＜jub，i＝1，2，...，I

wherein vub, aub and jub respectively represent the running speed v of the lane-changing vehicle 0_0，iAcceleration a_0，iJerk j_0，iUpper limit of (a), alb, jlb representAcceleration a of the vehicle_0，iJerk j_0，iThe lower limit of (d);

the target lane rear vehicle 2 jerk constraint module is used for taking jerk of the target lane rear vehicle 2 in the lane changing process along with the lane changing vehicle 0 as a variable, further obtaining a motion track of the target lane rear vehicle 2, and reflecting the interaction behavior of the lane changing vehicle 0 and the target lane rear vehicle 2 in the lane changing process;

the kinematic constraint of the rear vehicle 2 of the target lane is constructed as follows:

jlb＜j₂＜jub

alb＜a_2，I＜aub

v_2，I＞0

in the formula, a_2，IPlanning the end time t for the trajectory_IAcceleration of the vehicle 2 behind the target lane, a_2，I＝a_2，0+j₂t_I；v_2，IPlanning the end time t for the trajectory_IThe speed of the vehicle 2 behind the target lane,

7. the quadratic programming and neural network-based automatic driving lane change trajectory planning method according to claim 5, wherein the anti-collision constraint module is used for preventing an absolute distance between a lane change vehicle 0 and surrounding vehicles from being too small so as to avoid a dangerous condition; the vehicles around the lane changing vehicle 0 comprise a front vehicle 1 in the current lane, a rear vehicle 2 in the target lane and a front vehicle 3 in the target lane;

(1) the collision avoidance constraint construction is represented as:

s_f＜s_0，I＜s_d

x_1，i-xb_1，i-0.5vehl₁＞0，i＝1，2，...，I

xb_2，i-x_2，i-0.5vehl₂＞0，i＝1，2，...，I

x_3，i-xb_3，i-0.5vehl₃＞0，i＝1，2，...，I

in the formula, xb_1，i、xb_2，iAnd xb_3，iRespectively at time t_iCollision points of the lane-changing vehicle 0 with a front vehicle 1 of a current lane, a rear vehicle 2 of a target lane and a front vehicle 3 of the target lane; vehl₁、vehl₂And vehl₃The lengths of a front vehicle 1 of a current lane, a rear vehicle 2 of a target lane and a front vehicle 3 of the target lane are respectively set; x is the number of_1，i、x_2，iAnd x_3，iRespectively showing the vehicle 1 in front of the current lane, the vehicle 2 behind the target lane and the vehicle 3 in front of the target lane at the time t_iThe center of (a); wherein the front vehicle 1 of the current lane, the rear vehicle 2 of the target lane and the front vehicle 3 of the target lane are at the moment t_iThe center of (1) is:

the front vehicle 1 of the current lane and the front vehicle 3 of the target lane are ahead of the lane-changing vehicle 0 in the track planning time period [0, t ]_max]In the method, the law of motion of the vehicle 1 in front of the current lane and the vehicle 3 in front of the target lane can obtain information veh with t equal to 0 from the initial time_1，t＝0，veh_3，t＝0Calculating to obtain; the target lane rear vehicle 2 is behind the lane change vehicle 0, and based on the interactive consideration of the lane change process, the motion rule of the target lane rear vehicle 2 is the information veh obtained by the initial time t being 0_2，t＝0Sum variable vehicle 2 jerk j₂Jointly calculating to obtain;

(2) collision point xb in collision avoidance constraints_1，i、xb_2，iAnd xb_3，iCan be driven by the lane-changing vehicle 0 for a distance s_0，iSpecifically, the method comprises the following steps:

the collision point of the lane-change vehicle 0 is at time t_iIn the discontinuous change of the lane changing process, uniformly and discretely taking L samples, wherein in each sample L, the driving distance of a lane changing vehicle 0 is s_0，lAnd for each subsample L epsilon L through the path model in the path planning module, obtaining the driving distance s_0，lAnd corresponding collision point xb_1，l、xb_2，lAnd xb_3，l；

For each s_0，iDetermine its upper bound slb_0，iAnd the lower boundary sub_0，iUsing an upper bound of slb_0，iAnd the lower boundary sub_0，iScreening of the subsample set L among L samples_iFor each subsample L ∈ L_iThe running distance satisfies sub_0，i＜s_0，l＜slb_0，i(ii) a Wherein, the upper boundary slb_0，iAnd lower boundary sub_0，iThe calculation formula of (2) is as follows:

slb_0，i＝v_0，0t_i

sub_0，i＝0.8v_3，0t_i

for each time period t_iBased on the subsample set L_iOn the abscissa xb of the impact point k_1，i、xb_2，iAnd xb_3，iEstablishing a linear fit:

xb_k，i＝α_k，i+β_k，is_0，i+ε_k，k＝1，2，3

in the formula, the collision point k considers three outer end boundary points of the vehicle, and the abscissa of the collision point is xb_k，iThe estimated value isα_k，iRepresents truncation,. beta._k，iRepresents the slope, ε_kRepresenting the random error term when linearly fitting the collision point k; therefore, the abscissa xb of the collision point k_k，iThe driving distance s of the lane-changing vehicle 0 can be used_0，iEstimating:

introducing maximum absolute error term delta e_k，iSo as to ensure the safety:

Δe_k，i＝max_i(|xb_k，i-α_k，i+β_k，is_0，i|)l_i∈L_i；

using s_0，iLinear fitting xb_k，iAnd k is 1, 2, 3, establishing collision avoidance constraint:

x_1，i-0.5vehl₁-(α_1，i+β_1，is_0，i+Δe_1，i)＞0，i＝1，2，...，I

α_2，i+β_2，is_0，i-Δe_2，i-(x_2，i+0.5vehl₂)＞0，i＝1，2，...，I

x_3，i-0.5vehl₃-(α_3，i+β_3，is_0，i+Δe_3，i)＞0，i＝1，2，...，I

wherein the distance s is traveled_0，lAnd corresponding collision point value xb_1，l、xb_2，lAnd xb_3，lThe calculating method of (2):

will be interval [ x_0，0，x_d]Evenly divided into L parts to form a sample set L, and for each sample L_lE.g. L, obtaining x through a path quintic polynomial function_0，l，y_0，l，s_0，l，By passingCalculating to obtain the course angle of the lane-changing vehicle at the current momentBy changing the position of the vehicle (x)_0，l，y_0，l) And the course angle of the lane changing vehicle at the current momentAnd vehicle length vehl₀Simplifying the lane-changing vehicle 0 into a rectangle to obtain the coordinates cx of four corners₁，cx₂，cx₃，cx₄And intersection cx of four sides and lane boundary₅，cx₆，cx₇，cx₈；

Let point cx₁，cx₂，cx₃，cx₄，cx₅，cx₆，cx₇，cx₈Forming a point set omega, and respectively calculating a point cx_qE.g. ordinate of ΩAnd the abscissaq is 1, 2, 3, 4, 5, 6, 7, 8; let the set of points in the target lane be omega^UPoint set omega^UIn The set of points in the current lane is omega^LPoint set omega^LIn

Collision point xb_1，l、xb_2，lAnd xb_3，lFrom cx_qCalculating by epsilon to obtain:

8. the quadratic programming and neural network-based automatic driving lane-changing trajectory planning method according to claim 5, wherein the following model constraint module is used for the trajectory planning ending time t_IEnabling the rear vehicle 2 of the target lane and the lane changing vehicle 0 as well as the lane changing vehicle 0 and the front vehicle 3 of the target lane to meet a following model, and avoiding collision after the track planning is finished;

using the linear car following model of Pariotata, assuming car n is behind car n-1, the acceleration of car n needs to satisfy:

a_n＝ω₁(Δx_n-Δx^*)+ω₂Δv_n

in the formula,. DELTA.x_nRepresents the distance, Δ x, between vehicle n and vehicle n-1_n＝x_n-1-x_n；Δv_nRepresents the speed difference between vehicle n and vehicle n-1, Δ v_n＝v_n-1-v_n；ω₁，ω₂，Δx^*Parameters calibrated for the linear following model;

at the end time t of trajectory planning_IThe lane change vehicle 0 and the target lane front vehicle 3 need to satisfy the following constraints on acceleration, distance, and speed difference:

a_0，I≤ω₁(x_3，I-x_0，I-0.5vehl₃-0.5vehl₀-Δx^*)+ω₂(v_3，I-v_0，I)

in the formula, v_3，IRepresenting the front vehicle 3 of the target lane at the track planning end time t_IVelocity v of_3，I＝v_3，0+a_3，0t_I；

x_3，IRepresenting the front vehicle 3 of the target lane at the track planning end time t_IThe abscissa of the (c) axis of the (c),

x_0，Irepresenting the lane-changing vehicle 0 at the end of the trajectory planning time t_IAbscissa of (a), x_0，I＝x_0，f+s_0，I-s_f；

At the end time t of trajectory planning_IThe target lane rear vehicle 2 and the lane change vehicle 0 need to satisfy the following constraints on acceleration, distance, and speed difference:

a_2，I≤ω₁(x_0，I-x_2，I-0.5vehl₀-0.5vehl₂-Δx^*)+ω₂(v_0，I-v_2，I)

in the formula: x is the number of_0，I-x_2，IFor the vehicle 2 and the lane-changing vehicle 0 at the track planning end time t_IThe distance of (d); position x of rear vehicle 2 of target lane_2，IAnd velocity v_2，IFrom jerk j of rear vehicle 2 of the target lane₂And calculating to obtain:

9. the automatic driving lane change trajectory planning method based on the quadratic programming and the neural network according to claim 5, wherein the objective function building module takes a lane change vehicle 0 and a target lane rear vehicle 2 affected by lane change as research objects, comprehensively considers safety, comfort and rapidness as targets, and builds a corresponding quadratic programming model:

in the formula, mu and lambda are respectively set parameters;

an objective function representing a spatio-temporal trajectory plan;

Δa_0，I＝a_0，I-[ω₁(x_3，I-x_0，I-0.5vehl₃-0.5vehl₀-Δx^*)+ω₂(v_3，I-v_0，I)]。