A lane-changing decision behavior method and system based on lane-changing primitives

Abstract

The present invention belongs to the field of autonomous driving technology, and specifically relates to a lane changing decision behavior method and system based on lane changing primitives. The method includes: collecting lane changing behavior data, training long short-term memory networks, extracting lateral lane changing features, and obtaining lateral human-like lane changing primitive models and primitive libraries; training dynamic motion primitives, extracting longitudinal lane changing features, and obtaining longitudinal human-like lane changing primitive models and primitive libraries; based on new driving scene information and driving behavior information, using lateral and longitudinal human-like lane changing primitive models and lateral and longitudinal lane changing decision primitive libraries to determine lateral lane changing decision and longitudinal lane changing decision behavior trajectories. The present invention determines the current lateral lane changing decision and longitudinal lane changing decision behavior trajectories of the vehicle by constructing lateral and longitudinal human-like lane changing primitive models and lateral and longitudinal lane changing decision primitive libraries, so as to assist the autonomous driving vehicle in making lane changing decisions in the lane changing scene and improve the adaptability to random dynamic environments and the human-likeness and intelligence during driving.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and in particular relates to a lane changing decision behavior method and system based on lane changing primitives. Background Art

[0002] The current self-driving vehicles do not fully understand human driving behavior and driving experience, resulting in poor adaptability of self-driving vehicles to the environment, lack of human-likeness, and low degree of simulation of human driving behavior. Faced with complex and highly dynamic driving scenarios, human drivers first generate lateral lane change decision intentions, and then perform lateral and longitudinal lane change operations. However, during the lane change process, due to the random dynamics of the driving environment and high traffic density, self-driving vehicles are sometimes unable to make highly intelligent lane change decisions like human drivers. Most of the current lane change decision methods do not fully consider the highly intelligent lane change decision process of human drivers. It is necessary to study how to understand driving scenarios and learn human driving behaviors more clearly, and improve the adaptability to random dynamic environments and the intelligence of self-driving vehicles. Summary of the invention

[0003] 1. Technical issues to be resolved

[0004] The technical problem to be solved by the present invention is: how to provide a lane-changing decision behavior method and system based on lane-changing primitives.

[0005] (II) Technical solution

[0006] In order to solve the above technical problems, the present invention provides a lane-changing decision behavior method based on lane-changing primitives, which comprises the following steps:

[0007] Step 101: Collect lane-changing behavior data;

[0008] The lane-changing behavior data includes driving scene information and human driving information;

[0009] The driving scene information includes: the current lane number, the lane in which the vehicle is located, lane lines, and traffic signs; the human driving information includes: the lateral and longitudinal position information, lateral and longitudinal speeds, lateral and longitudinal accelerations of the lane-changing vehicle and surrounding vehicles, the relative position information between vehicles, the relative speed information, and whether the lane is changed;

[0010] The driving scene information and human driving information are collected by on-board sensors; the on-board sensors include: cameras, millimeter-wave radars, integrated navigation systems, and electronic control systems; the scene is a multi-lane lane-changing scene, and the lane-changing vehicle, i.e., the target vehicle, makes a lane-changing decision of changing left lanes, changing right lanes, or keeping lanes; in addition, overtaking behavior is a combined derivative form of changing left lanes, changing right lanes, or keeping lanes in this method;

[0011] Step 102: classifying and labeling the human driving information according to the driving scene information, and extracting corresponding human-like lane-changing behavior features and lane-changing scene labels from the driving scene information and the classified and labeled human driving information; the human-like lane-changing behavior features include lane-changing interaction information between vehicles and vehicle driving behavior information;

[0012] Preprocessing the human-like lane-changing behavior characteristics to obtain preprocessed human-like lane-changing behavior characteristics;

[0013] Step 103: training a long short-term memory network using the human-like lane-changing behavior characteristics and lane-changing scene labels to obtain a lateral human-like lane-changing primitive model;

[0014] According to the lane-changing scene labels, the lane-changing interaction information between vehicles, and the vehicle driving behavior information preprocessed in step 102, the lateral lane-changing decision primitives of the lane-changing vehicles are extracted and a lateral lane-changing decision primitive library is constructed;

[0015] Step 104: training a dynamic motion primitive using the lateral lane-changing decision primitive and the lane-changing behavior data to obtain a longitudinal human-like lane-changing primitive model;

[0016] Based on the lateral lane-changing decision primitives extracted in step 103, the longitudinal lane-changing decision primitives of the lane-changing vehicle are extracted and a longitudinal lane-changing decision primitive library is constructed;

[0017] Step 105: Determine a lane-changing decision behavior trajectory according to the lateral humanoid lane-changing primitive model, the lateral lane-changing decision primitive library, the longitudinal humanoid lane-changing primitive model, and the longitudinal lane-changing decision primitive library.

[0018] Wherein, in the step 102, the human-like lane-changing behavior characteristics are preprocessed, wherein the preprocessing method includes normalization, filtering, and smoothing processing methods to improve the performance and effect of the lane-changing primitive model and the lane-changing decision.

[0019] Wherein, in the step 102, the human driving information is classified and labeled according to the driving scene information captured by the camera, and the collected human driving information is classified and labeled into three lane change scene labels, namely left lane change L1, right lane change L2, and lane keeping L3;

[0020] Define the lane-changing scene label set O scene as follows:

[0021] O scene =[L1,L2,L3]

[0022] And obtain the lane-changing interaction information between vehicles in the human-like lane-changing behavior characteristics O R :

[0023] O R(t)=[Δx(t),Δy(t),Δv x (t),Δv y (t),Δa x (t),Δa y (t)]

[0024] And the vehicle driving behavior information O in the human-like lane-changing behavior characteristics c :

[0025]

[0026] Where R represents the surrounding vehicles, C represents the lane-changing vehicle; Δx, Δy represent the relative longitudinal and lateral position information between the surrounding vehicles and the lane-changing vehicle, Δv x , Δv y and Δa x , Δa y Respectively represent the relative speed and relative speed component of the surrounding vehicles and the lane-changing vehicle in the x and y directions; x and y are the longitudinal and lateral position information of the lane-changing vehicle, respectively; v x , v y and a x , a y are the speed and acceleration components of the lane-changing vehicle in the x and y directions respectively, and t represents the time.

[0027] Wherein, in step 103, the data input format of the lateral humanoid lane-changing primitive model is as follows:

[0028] O=[O scene ,O R ,O C ]

[0029] The data input O of the lateral humanoid lane-changing primitive model is driving data O;

[0030] The long short-term memory network (LSTM) is used to learn, train and extract features from the data input O of the lateral humanoid lane-changing primitive model under different lane-changing scene labels, and then the lateral humanoid lane-changing primitive model corresponding to the lane-changing vehicle is established.

[0031] During the model training process, based on the preprocessed data input O, the network’s forget gate f is first used tThe activation function determines and controls which driving features are forgotten or discarded by the current cell; subsequently, the network input gate uses an activation function to control the preservation and update of the driving features for lane change decisions, and obtains new candidate driving state features through a tanh function, thereby obtaining new internal driving state features; finally, the input state information and the hidden state of the previous sequence pass through the activation function of an output gate, and the internal driving state features are transmitted to the external state through the combined output gate, which is the extracted lane change decision primitive;

[0032] Finally, based on the obtained long short-term memory network model, three types of lane change decision feature parameters and a lane change decision primitive library for left lane change, right lane change, and lane keeping are extracted from the driving features; and corresponding human-like lane change primitive models are established respectively;

[0033] Among them, the three types of lane change decision primitives are: the left lane change decision primitive P LC =1, the right lane change decision primitive P LC =2, and the lane keeping decision primitive P LC =3; furthermore, for different lane change decisions, the model outputs the corresponding lane change decision primitive P LC =1, 2, 3.

[0034] Among them, in step 104, based on the extraction of the lane change decision primitive P LC =1, 2, 3 in step 103, the longitudinal lane change decision primitive of the lane-changing vehicle is extracted and a longitudinal lane change decision primitive library is constructed;

[0035] The data input of the longitudinal human-like lane change primitive model is the lane change decision primitive P LC and the driving data O;

[0036] Using dynamic motion primitives to construct a longitudinal human-like lane change primitive model, the behavior of the vehicle is represented by the following transition system:

[0037]

[0038]

[0039] Among them, the vehicle position d = [x, y], the vehicle speed v = [v x , v y , the vehicle acceleration d0 = [x0, y0] and g = g T = [x T , y T] represent the starting position of the vehicle at t=0 and the target position at t=T respectively; k and b represent the spring coefficient and damping coefficient, and τ represents the time scale factor. The convergence of the model to the target position is ensured by selecting appropriate k and b. In order to simulate human driving behavior as much as possible, a new control item f is introduced, which can generate arbitrarily complex longitudinal lane change decision primitives through learning;

[0040]

[0041] Among them, ψ i (s) = exp(-h i (s-μ i ) 2 ) is the Gaussian basis function, μ i and h i is the position function, ω i is an adjustable weight that can be obtained through learning and training, N represents the number of basis functions; the function f depends on the phase variable s, which is controlled by the predefined k, and s converges monotonically from the initial value 1 to 0 during the vehicle's driving process; s is defined as follows:

[0042]

[0043] It is called a regular system. When the model is trained, the objective function is as follows:

[0044]

[0045] Among them, the initial position d0 and the target position g are given respectively T ; Solve J = ∑ by minimizing s ω(f target (s)-f(s)) 2 , and then obtain the weight ω=[ω x ,ω y ] and longitudinal humanoid lane-changing primitive model.

[0046] Among them, in the step 105, based on the current driving scene information and human driving information, the lateral lane changing decision primitive output by the lateral humanoid lane changing primitive model and the longitudinal lane changing decision primitive output by the longitudinal humanoid lane changing primitive model are used to obtain the lateral lane changing decision and longitudinal lane changing decision behavior trajectories of the lane changing vehicle.

[0047] In addition, the present invention also provides a lane-changing decision behavior method system based on lane-changing primitives, which is used to implement the lane-changing decision behavior method based on lane-changing primitives, and the behavior system includes:

[0048] A data acquisition module, which is used to collect lane-changing behavior data; the lane-changing behavior data includes driving scene information and human driving information;

[0049] a labeling and driving feature extraction module, which is used to classify and label the human driving information through the driving scene information, and extract corresponding human-like lane-changing behavior features and lane-changing scene labels from the driving scene information and the classified and labeled human driving information; the human-like lane-changing behavior features include lane-changing interaction information between vehicles and vehicle driving behavior information;

[0050] A preprocessing module, which is used to preprocess the human-like lane-changing behavior characteristics to obtain the preprocessed human-like lane-changing behavior characteristics;

[0051] A training module, which is used to construct a lateral human-like lane-changing primitive model and a longitudinal human-like lane-changing primitive model according to the human-like lane-changing behavior characteristics and the lane-changing scene label;

[0052] A lane-changing decision primitive library construction module, which is used to construct a lateral lane-changing decision primitive library and a longitudinal lane-changing decision primitive library of the lane-changing vehicle according to the lateral lane-changing decision primitive output by the lateral human-like lane-changing primitive model and the longitudinal lane-changing decision primitive output by the longitudinal human-like lane-changing primitive model;

[0053] The lane-changing decision module is used to determine the current lateral lane-changing decision and longitudinal lane-changing decision behavior trajectories according to the lateral humanoid lane-changing primitive model, the longitudinal humanoid lane-changing primitive model, the lateral lane-changing decision primitive library, and the longitudinal lane-changing decision primitive library.

[0054] Wherein, the training module includes:

[0055] A first training module, which is used to train a long short-term memory network through the human-like lane-changing behavior characteristics and lane-changing scene labels to obtain a lateral human-like lane-changing primitive model and a lateral lane-changing decision primitive;

[0056] The second training module is used to train the dynamic motion primitives through the lateral lane-changing decision primitives and the lane-changing behavior data to obtain a longitudinal humanoid lane-changing primitive model and a longitudinal lane-changing decision primitive.

[0057] The driving scene information includes: the current lane number, the lane in which the vehicle is located, lane lines, and traffic signs.

[0058] The human driving information includes: the lateral and longitudinal position information, lateral and longitudinal speeds, lateral and longitudinal accelerations of the lane-changing vehicle and surrounding vehicles, the relative position information, relative speed information between vehicles, and whether to change lanes.

[0059] (III) Beneficial effects

[0060] The present invention provides a lane changing decision behavior method and system based on lane changing primitives. The method first constructs a lateral humanoid lane changing primitive model and a lateral primitive library based on current driving scene information and human driving information, and then constructs a longitudinal humanoid lane changing primitive model and a longitudinal primitive library to make lateral discrete lane changing decisions and generate longitudinal continuous lane changing decision behavior trajectories, thereby assisting autonomous driving vehicles in making humanoid lane changing decisions.

[0061] In summary, the present invention first constructs a lateral human-like lane-changing primitive model of a lane-changing vehicle and extracts a lateral lane-changing decision primitive. On this basis, a longitudinal human-like lane-changing primitive model is further constructed to obtain a lateral and longitudinal lane-changing decision primitive library. The present invention realizes the human-like lane-changing decision of an autonomous driving vehicle and makes a lane-changing decision of the vehicle in real time according to the current driving environment. The present invention is more in line with the actual lane-changing situation of a human driver in the process of driving a vehicle and has great novelty and practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 The flowchart of the lane changing decision behavior method based on the lane changing primitive of the present invention. DETAILED DESCRIPTION

[0063] In order to make the purpose, content, and advantages of the present invention more clear, the specific implementation methods of the present invention are further described in detail below in conjunction with the accompanying drawings and examples.

[0064] Example 1

[0065] The purpose of this embodiment is to provide a lane changing decision behavior method and system based on lane changing primitives. First, the vehicle constructs a lateral human-like lane changing primitive model and a lateral primitive library based on the current driving scene information and human driving information, and then constructs a longitudinal human-like lane changing primitive model and a longitudinal primitive library to make lateral discrete lane changing decisions and generate longitudinal continuous lane changing decision behavior trajectories, thereby assisting the autonomous driving vehicle in making human-like lane changing decisions.

[0066] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0067] like Figure 1 As shown, the lane-changing decision behavior method based on lane-changing primitives disclosed in the present invention comprises the following steps:

[0068] Step 101: Collect lane-changing behavior data;

[0069] The lane-changing behavior data includes driving scene information and human driving information;

[0070] The driving scene information includes: the current lane number, the lane in which the vehicle is located, lane lines, and traffic signs; the human driving information includes: the lateral and longitudinal position information, lateral and longitudinal speeds, lateral and longitudinal accelerations of the lane-changing vehicle and surrounding vehicles, the relative position information between vehicles, the relative speed information, and whether the lane is changed;

[0071] The driving scene information and human driving information are collected by on-board sensors; the on-board sensors include: cameras, millimeter-wave radars, integrated navigation systems, and electronic control systems; the scene is a multi-lane lane-changing scene, and the lane-changing vehicle, i.e., the target vehicle, makes a lane-changing decision of changing left lanes, changing right lanes, or keeping lanes; in addition, overtaking behavior is a combined derivative form of changing left lanes, changing right lanes, or keeping lanes in this method;

[0072] Step 102: classifying and labeling the human driving information according to the driving scene information, and extracting corresponding human-like lane-changing behavior features and lane-changing scene labels from the driving scene information and the classified and labeled human driving information; the human-like lane-changing behavior features include lane-changing interaction information between vehicles and vehicle driving behavior information;

[0073] Preprocessing the human-like lane-changing behavior characteristics to obtain preprocessed human-like lane-changing behavior characteristics;

[0074] Step 103: training a long short-term memory network using the human-like lane-changing behavior characteristics and lane-changing scene labels to obtain a lateral human-like lane-changing primitive model;

[0075] According to the lane-changing scene labels, the lane-changing interaction information between vehicles, and the vehicle driving behavior information preprocessed in step 102, the lateral lane-changing decision primitives of the lane-changing vehicles are extracted and a lateral lane-changing decision primitive library is constructed;

[0076] Step 104: training a dynamic motion primitive using the lateral lane-changing decision primitive and the lane-changing behavior data to obtain a longitudinal human-like lane-changing primitive model;

[0077] Based on the lateral lane-changing decision primitives extracted in step 103, the longitudinal lane-changing decision primitives of the lane-changing vehicle are extracted and a longitudinal lane-changing decision primitive library is constructed;

[0078] Step 105: Determine a lane-changing decision behavior trajectory according to the lateral humanoid lane-changing primitive model, the lateral lane-changing decision primitive library, the longitudinal humanoid lane-changing primitive model, and the longitudinal lane-changing decision primitive library.

[0079] Wherein, in the step 102, the human-like lane-changing behavior characteristics are preprocessed, wherein the preprocessing method includes normalization, filtering, and smoothing processing methods to improve the performance and effect of the lane-changing primitive model and the lane-changing decision.

[0080] Wherein, in the step 102, the human driving information is classified and labeled according to the driving scene information captured by the camera, and the collected human driving information is classified and labeled into three lane change scene labels, namely left lane change L1, right lane change L2, and lane keeping L3;

[0081] Define the lane-changing scene label set O scene as follows:

[0082] O scene =[L1,L2,L3]

[0083] And obtain the lane-changing interaction information between vehicles in the human-like lane-changing behavior characteristics O R :

[0084] O R (t)=[Δx(t),Δy(t),Δv x (t),Δv y (t),Δa x (t),Δa y (t)]

[0085] And the vehicle driving behavior information O in the human-like lane-changing behavior characteristics c :

[0086]

[0087] Where R represents the surrounding vehicles, C represents the lane-changing vehicle; Δx, Δy represent the relative longitudinal and lateral position information between the surrounding vehicles and the lane-changing vehicle, Δv x , Δv y and Δa x , Δa y Respectively represent the relative speed and relative speed component of the surrounding vehicles and the lane-changing vehicle in the x and y directions; x and y are the longitudinal and lateral position information of the lane-changing vehicle, respectively; v x , v y and a x , a y are the speed and acceleration components of the lane-changing vehicle in the x and y directions respectively, and t represents the time.

[0088] Wherein, in step 103, the data input format of the lateral humanoid lane-changing primitive model is as follows:

[0089] O=[O scene ,O R ,O C ]

[0090] The data input O of the lateral humanoid lane-changing primitive model is driving data O;

[0091] The long short-term memory network (LSTM) is used to learn, train and extract features from the data input O of the lateral humanoid lane-changing primitive model under different lane-changing scene labels, and then the lateral humanoid lane-changing primitive model corresponding to the lane-changing vehicle is established.

[0092] During the model training process, based on the preprocessed data input O, the network’s forget gate f is first used t The activation function determines and controls which driving features are forgotten or discarded by the current unit; then, the network input gate uses an activation function to control the preservation and update of the lateral lane change decision driving features, and obtains new candidate driving state features through a tanh function, thereby obtaining new internal driving state features; finally, the input state information and the hidden state of the previous sequence pass through the activation function of an output gate, and the internal driving state features are passed to the external state by combining the output gate, which is the extracted lateral lane change decision primitive;

[0093] Finally, based on the obtained long short-term memory network model, three types of lateral lane-changing decision feature parameters and lateral lane-changing decision primitive libraries, including left lane-changing, right lane-changing and lane-keeping, are extracted from the driving characteristics; and the corresponding lateral human-like lane-changing primitive models are established respectively;

[0094] Among them, the three lateral lane change decision primitives are: left lateral lane change decision primitive P LC =1, right lateral lane change decision primitive P LC =2, and lane keeping decision primitive P LC =3; Then, for different lateral lane-changing decisions, the model outputs the corresponding lateral lane-changing decision primitive P LC =1,2,3.

[0095] Wherein, in step 104, the lateral lane change decision primitive P is extracted in step 103 LC =1,2,3, extract the longitudinal lane-changing decision primitives of the lane-changing vehicle and construct a longitudinal lane-changing decision primitive library;

[0096] The data input of the longitudinal human-like lane-changing primitive model is the lateral lane-changing decision primitive P LC and driving data O;

[0097] The longitudinal humanoid lane-changing primitive model is constructed using dynamic motion primitives, and the vehicle behavior is represented by the following transformation system:

[0098]

[0099]

[0100] Wherein, the vehicle position d = [x, y], the vehicle speed v = [v x ,v y ], vehicle acceleration d0 = [x0, y0] and g = g T = [x T , y T represent the starting position of the vehicle at t = 0 and the target position at t = T respectively; k and b represent the spring coefficient and damping coefficient, τ represents the time scale factor, and by selecting appropriate k and b, the convergence of the model to the target position is ensured. To simulate human driving behavior as much as possible, a new control term f is introduced, which can generate arbitrary complex longitudinal lane-changing decision primitives through learning;

[0101]

[0102] where, ψ i (s) = exp(-h i (s - μ i ) 2 ) is a Gaussian basis function, μ i and h i are position functions, ω i is an adjustable weight, which can be obtained through learning and training, and N represents the number of basis functions; the function f depends on the phase variable s, which is controlled by a predefined k, and s monotonically converges from the initial value 1 to 0 during the vehicle driving process; s is defined as follows:

[0103]

[0104] is called a regular system; when the model is learning and training, the objective function is as follows:

[0105]

[0106] where, the initial position d0 and the target position g are given respectively T ; by minimizing and solving J = ∑ s ω(f target (s) - f(s)) 2 , and then the weight ω = [ω x , ω y and the longitudinal human-like lane-changing primitive model are obtained.

[0107] Among them, in step 105, based on the current driving scene information and human driving information, using the lateral lane-changing decision primitive output by the lateral human-like lane-changing primitive model and the longitudinal lane-changing decision primitive output by the longitudinal human-like lane-changing primitive model, the lateral lane-changing decision and the longitudinal lane-changing decision behavior trajectory of the lane-changing vehicle are obtained.

[0108] In addition, the present invention also provides a lane-changing decision behavior method system based on lane-changing primitives, which is used to implement the lane-changing decision behavior method based on lane-changing primitives, and the behavior system includes:

[0109] A data acquisition module, which is used to collect lane-changing behavior data; the lane-changing behavior data includes driving scene information and human driving information;

[0110] a labeling and driving feature extraction module, which is used to classify and label the human driving information through the driving scene information, and extract corresponding human-like lane-changing behavior features and lane-changing scene labels from the driving scene information and the classified and labeled human driving information; the human-like lane-changing behavior features include lane-changing interaction information between vehicles and vehicle driving behavior information;

[0111] A preprocessing module, which is used to preprocess the human-like lane-changing behavior characteristics to obtain the preprocessed human-like lane-changing behavior characteristics;

[0112] A training module, which is used to construct a lateral human-like lane-changing primitive model and a longitudinal human-like lane-changing primitive model according to the human-like lane-changing behavior characteristics and the lane-changing scene label;

[0113] A lane-changing decision primitive library construction module, which is used to construct a lateral lane-changing decision primitive library and a longitudinal lane-changing decision primitive library of the lane-changing vehicle according to the lateral lane-changing decision primitive output by the lateral human-like lane-changing primitive model and the longitudinal lane-changing decision primitive output by the longitudinal human-like lane-changing primitive model;

[0114] The lane-changing decision module is used to determine the current lateral lane-changing decision and longitudinal lane-changing decision behavior trajectories according to the lateral humanoid lane-changing primitive model, the longitudinal humanoid lane-changing primitive model, the lateral lane-changing decision primitive library, and the longitudinal lane-changing decision primitive library.

[0115] Wherein, the training module includes:

[0116] A first training module, which is used to train a long short-term memory network through the human-like lane-changing behavior characteristics and lane-changing scene labels to obtain a lateral human-like lane-changing primitive model and a lateral lane-changing decision primitive;

[0117] The second training module is used to train the dynamic motion primitives through the lateral lane-changing decision primitives and the lane-changing behavior data to obtain a longitudinal humanoid lane-changing primitive model and a longitudinal lane-changing decision primitive.

[0118] The driving scene information includes: the current lane number, the lane in which the vehicle is located, lane lines, and traffic signs.

[0119] The human driving information includes: the lateral and longitudinal position information, lateral and longitudinal speeds, lateral and longitudinal accelerations of the lane-changing vehicle and surrounding vehicles, the relative position information, relative speed information between vehicles, and whether to change lanes.

[0120] In summary, the present invention first constructs a lateral humanoid lane-changing primitive model of a lane-changing vehicle, and extracts a lateral lane-changing decision primitive. On this basis, a longitudinal humanoid lane-changing primitive model is further constructed, and then a lateral and longitudinal lane-changing decision primitive library is obtained. The present invention realizes the humanoid lane-changing decision of an autonomous driving vehicle, and makes a lane-changing decision of the vehicle in real time according to the current driving environment. The present invention is more in line with the actual lane-changing situation of a humanoid driver driving a vehicle, and has great novelty and practicality.

[0121] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0122] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A lane-changing decision behavior method based on lane-changing primitives, characterized in that: It includes the following steps: Step 101: Collect lane-changing behavior data; The lane-changing behavior data includes driving scene information and human driving information; The driving scene information includes: the current lane number, the lane in which the vehicle is located, lane lines, and traffic signs; the human driving information includes: the lateral and longitudinal position information, lateral and longitudinal speeds, lateral and longitudinal accelerations of the lane-changing vehicle and surrounding vehicles, the relative position information between vehicles, the relative speed information, and whether the lane is changed; The driving scene information and human driving information are collected by on-board sensors; the on-board sensors include: cameras, millimeter-wave radars, integrated navigation systems, and electronic control systems; the scene is a multi-lane lane-changing scene, and the lane-changing vehicle, i.e., the target vehicle, makes a lane-changing decision of changing left lanes, changing right lanes, or keeping lanes; in addition, overtaking behavior is a combined derivative form of changing left lanes, changing right lanes, or keeping lanes in this method; Step 102: classifying and labeling the human driving information according to the driving scene information, and extracting corresponding human-like lane-changing behavior features and lane-changing scene labels from the driving scene information and the classified and labeled human driving information; the human-like lane-changing behavior features include lane-changing interaction information between vehicles and vehicle driving behavior information; Preprocessing the human-like lane-changing behavior characteristics to obtain preprocessed human-like lane-changing behavior characteristics; Step 103: training a long short-term memory network using the human-like lane-changing behavior characteristics and lane-changing scene labels to obtain a lateral human-like lane-changing primitive model; According to the lane-changing scene labels, the lane-changing interaction information between vehicles, and the vehicle driving behavior information preprocessed in step 102, the lateral lane-changing decision primitives of the lane-changing vehicles are extracted and a lateral lane-changing decision primitive library is constructed; Step 104: training a dynamic motion primitive using the lateral lane-changing decision primitive and the lane-changing behavior data to obtain a longitudinal human-like lane-changing primitive model; Based on the lateral lane-changing decision primitives extracted in step 103, the longitudinal lane-changing decision primitives of the lane-changing vehicle are extracted and a longitudinal lane-changing decision primitive library is constructed; Step 105: determining a lane-changing decision behavior trajectory according to the lateral human-like lane-changing primitive model, the lateral lane-changing decision primitive library, the longitudinal human-like lane-changing primitive model, and the longitudinal lane-changing decision primitive library; Wherein, in the step 102, the human-like lane-changing behavior characteristics are preprocessed, wherein the preprocessing method includes normalization, filtering, and smoothing processing methods to improve the performance and effect of the lane-changing primitive model and the lane-changing decision; Wherein, in the step 102, the human driving information is classified and labeled according to the driving scene information captured by the camera, and the collected human driving information is classified and labeled into three lane change scene labels, namely left lane change L1, right lane change L2, and lane keeping L3; Define the lane-changing scene label set O scene as follows: <h2 style=";text-align:left;direction:ltr">O<h2 style=";text-align:left;direction:ltr"> scene <h2 style=";text-align:left;direction:ltr"> (L1,L2,L3) And obtain the lane-changing interaction information between vehicles in the human-like lane-changing behavior characteristics O R : O R (t)=[Δx(t),Δy(t),Δv x (t),Δv y (t),Δa x (t),Δa y (t)] And the vehicle driving behavior information O in the human-like lane-changing behavior characteristics c : Where R represents the surrounding vehicles, C represents the lane-changing vehicle; Δx, Δy represent the relative longitudinal and lateral position information between the surrounding vehicles and the lane-changing vehicle, Δv x , Δv y and Δa x , Δa y Respectively represent the relative speed and relative speed component of the surrounding vehicles and the lane-changing vehicle in the x and y directions; x and y are the longitudinal and lateral position information of the lane-changing vehicle, respectively; v x , v y and a x , a y are the speed and acceleration components of the lane-changing vehicle in the x and y directions, respectively, and t represents time; Wherein, in step 103, the data input format of the lateral humanoid lane-changing primitive model is as follows: O=[O scene ,O R ,O C ] The data input O of the lateral humanoid lane-changing primitive model is driving data O; The long short-term memory network (LSTM) is used to learn, train and extract features from the data input O of the lateral humanoid lane-changing primitive model under different lane-changing scene labels, and then the lateral humanoid lane-changing primitive model corresponding to the lane-changing vehicle is established. During the model training process, based on the preprocessed data input O, the network’s forget gate f is first used t The activation function determines and controls which driving features are forgotten or discarded by the current unit; then, the network input gate uses an activation function to control the preservation and update of the lateral lane change decision driving features, and obtains new candidate driving state features through a tanh function, thereby obtaining new internal driving state features; finally, the input state information and the hidden state of the previous sequence pass through the activation function of an output gate, and the internal driving state features are passed to the external state by combining the output gate, which is the extracted lateral lane change decision primitive; Finally, based on the obtained long short-term memory network model, three types of lateral lane-changing decision feature parameters and lateral lane-changing decision primitive libraries, including left lane-changing, right lane-changing and lane-keeping, are extracted from the driving characteristics; and the corresponding lateral human-like lane-changing primitive models are established respectively; Among them, the three lateral lane change decision primitives are: left lateral lane change decision primitive P LC =1, right lateral lane change decision primitive P LC =2, and lane keeping decision primitive P LC =3; Then, for different lateral lane-changing decisions, the model outputs the corresponding lateral lane-changing decision primitive P LC =1,2,3; Wherein, in step 104, the lateral lane change decision primitive P is extracted in step 103 LC =1,2,3, extract the longitudinal lane-changing decision primitives of the lane-changing vehicle and construct a longitudinal lane-changing decision primitive library; The data input of the longitudinal human-like lane-changing primitive model is the lateral lane-changing decision primitive P LC and driving data O; The longitudinal humanoid lane-changing primitive model is constructed using dynamic motion primitives, and the vehicle behavior is represented by the following transformation system: where the vehicle position \(d = [x, y]\), the vehicle velocity \(v=[v x ,v y \), and the vehicle acceleration \(d_0 = [x_0, y_0]\) and \(g = g T =[x T ,y T \) represent the starting position of the vehicle at \(t = 0\) and the target position at \(t = T\), respectively; \(k\) and \(b\) represent the spring coefficient and the damping coefficient, and \(\tau\) represents the time scale factor. By selecting appropriate \(k\) and \(b\), the convergence of the model to the target position is ensured. To simulate human driving behavior, a new control term \(f\) is introduced, and any arbitrarily complex longitudinal lane-changing decision primitive is generated through learning. Among them, ψ i (s) = exp(-h i (s-μ i ) 2 ) is the Gaussian basis function, μ i and h i is the position function, ω i is an adjustable weight that can be obtained through learning and training, N represents the number of basis functions; the function f depends on the phase variable s, which is controlled by the predefined k, and s converges monotonically from the initial value 1 to 0 during the vehicle's driving process; s is defined as follows: It is called a regular system. When the model is trained, the objective function is as follows: Among them, the initial position d0 and the target position g are given respectively T ; Solve J = Σ by minimizing s ω(f target (s)-f(s)) 2 , and then obtain the weight ω=[ω x ,ω y ] and longitudinal humanoid lane-changing primitive model; Wherein, in the step 105, based on the current driving scene information and human driving information, the lateral lane changing decision primitive output by the lateral human-like lane changing primitive model and the longitudinal lane changing decision primitive output by the longitudinal human-like lane changing primitive model are used to obtain the lateral lane changing decision and longitudinal lane changing decision behavior trajectories of the lane changing vehicle; The method first constructs a lateral human-like lane-changing primitive model of the lane-changing vehicle and extracts the lateral lane-changing decision primitive. On this basis, a longitudinal human-like lane-changing primitive model is further constructed, and then a lateral lane-changing decision primitive library and a longitudinal lane-changing decision primitive library are obtained, so as to determine the vehicle's current lateral lane-changing decision and longitudinal lane-changing decision behavior trajectories, so as to assist the autonomous driving vehicle in making lane-changing decisions in the lane-changing scenario and improve the adaptability to the random dynamic environment as well as the human-likeness and intelligence during the driving process. Therefore, the method realizes the human-like lane-changing decision of the autonomous driving vehicle, and makes the vehicle's lane-changing decision in real time according to the current driving environment, which is more in line with the actual lane-changing situation of human drivers in the process of driving the vehicle.

2. A lane-changing decision behavior system based on lane-changing primitives, which is used to implement the lane-changing decision behavior method based on lane-changing primitives as claimed in claim 1, characterized in that: The behavior system includes: A data acquisition module, which is used to collect lane-changing behavior data; the lane-changing behavior data includes driving scene information and human driving information; a labeling and driving feature extraction module, which is used to classify and label the human driving information through the driving scene information, and extract corresponding human-like lane-changing behavior features and lane-changing scene labels from the driving scene information and the classified and labeled human driving information; the human-like lane-changing behavior features include lane-changing interaction information between vehicles and vehicle driving behavior information; A preprocessing module, which is used to preprocess the human-like lane-changing behavior characteristics to obtain the preprocessed human-like lane-changing behavior characteristics; A training module, which is used to construct a lateral human-like lane-changing primitive model and a longitudinal human-like lane-changing primitive model according to the human-like lane-changing behavior characteristics and the lane-changing scene label; A lane-changing decision primitive library construction module, which is used to construct a lateral lane-changing decision primitive library and a longitudinal lane-changing decision primitive library of the lane-changing vehicle according to the lateral lane-changing decision primitive output by the lateral human-like lane-changing primitive model and the longitudinal lane-changing decision primitive output by the longitudinal human-like lane-changing primitive model; The lane-changing decision module is used to determine the current lateral lane-changing decision and longitudinal lane-changing decision behavior trajectories according to the lateral humanoid lane-changing primitive model, the longitudinal humanoid lane-changing primitive model, the lateral lane-changing decision primitive library, and the longitudinal lane-changing decision primitive library.

3. The lane-changing decision behavior system based on lane-changing primitives as claimed in claim 2, characterized in that: The training module includes: A first training module, which is used to train a long short-term memory network through the human-like lane-changing behavior characteristics and lane-changing scene labels to obtain a lateral human-like lane-changing primitive model and a lateral lane-changing decision primitive; The second training module is used to train the dynamic motion primitives through the lateral lane-changing decision primitives and the lane-changing behavior data to obtain a longitudinal humanoid lane-changing primitive model and a longitudinal lane-changing decision primitive.

4. The lane-changing decision behavior system based on lane-changing primitives as claimed in claim 3, characterized in that: The driving scene information includes: current lane number, lane, lane lines, and traffic signs.

5. The lane-changing decision behavior system based on lane-changing primitives as claimed in claim 4, characterized in that: The human driving information includes: the lateral and longitudinal position information, lateral and longitudinal speeds, lateral and longitudinal accelerations of the lane-changing vehicle and surrounding vehicles, relative position information, relative speed information between vehicles, and whether to change lanes.

Images