Automatic driving lane-changing style switching method based on driver's riding preference

Abstract

The present application relates to the technical field of automobile driving control, and particularly relates to an automatic driving lane-changing style switching method based on driver's riding preference, which uses a lane-changing style analysis model to analyze the driver's riding preference, judges the type of the driver's lane-changing style intention in a specific driving scene, classifies the driver's historical lane-changing trajectory data according to the lane-changing style, and obtains a target trajectory corresponding to each lane-changing style based on the historical lane-changing trajectory corresponding to each lane-changing style; a loss function between the driver's current driving trajectory and the current target trajectory is calculated, and a minimization operation is performed on the loss function, and the obtained parameters of the current driving trajectory are substituted into a lane-changing trajectory prediction model to obtain predicted lane-changing parameters; and based on the predicted lane-changing parameters, the vehicle is controlled to complete a stylized lane-changing action. The present application can ensure that the personalized needs of the driver are met without violating the safety principle.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of automobile driving control, and particularly relates to an automatic driving lane-changing style switching method based on driver seating preferences. BACKGROUND

[0002] With the continuous development of automatic driving technology, the collaborative control between vehicles and drivers has become one of the research hotspots. Existing automatic driving systems mostly rely on preset control modes, often ignoring the personalized needs and driving habits of drivers. Although these systems can ensure the safe driving of vehicles, they lack flexibility and humanization in driving style and cannot make corresponding adjustments according to the preferences of drivers, resulting in poor driving experience for drivers, especially in complex driving behaviors such as lane changing, where the personalized needs of drivers are difficult to meet.

[0003] In order to improve the interaction effect between drivers and automatic driving systems, how to dynamically adjust the lane-changing style of vehicles according to the voice instructions of drivers to improve the personalization and comfort of driving is still a problem to be solved. SUMMARY

[0004] Therefore, the present application aims to provide an automatic driving lane-changing style switching method based on driver seating preferences to establish a large language model driven driver intent analysis model, analyze the voice input of the driver, and determine the type of lane-changing style intent of the driver in a specific driving scenario; based on the analysis result, the automatic driving system dynamically adjusts the lane-changing behavior of the vehicle to ensure that the personalized lane-changing needs of the driver are met without violating safety principles.

[0005] To achieve the above purpose, the technical solution of the present application is as follows:

[0006] An automatic driving lane-changing style switching method based on driver seating preferences, comprising:

[0007] S1: determining the lane-changing style category of the driver based on the seating preferences of the driver using a lane-changing style analysis model;

[0008] S2: classifying the lane-changing style according to the lane-changing style obtained in step S1 based on the historical lane-changing trajectory data of the driver; obtaining the target trajectory corresponding to each lane-changing style based on the historical lane-changing trajectory corresponding to each lane-changing style;

[0009] S3: predicting the future driving state of the vehicle based on the current driving state of the vehicle; determining the lane-changing strategy that meets the future driving state based on the future driving state and the target trajectory corresponding to each lane-changing style obtained in step S2 using reinforcement learning;

[0010] S4: controlling the vehicle to perform a lane-changing action according to the lane-changing strategy obtained in step S3, so as to conform to the current lane-changing style of the driver.

[0011] Further, in step S1: input the voice input instruction and the dialogue history of the driver into the lane-changing style analysis model, and determine the lane-changing style category corresponding to the voice input instruction; the lane-changing style category includes aggressive, conservative and negative.

[0012] Further, in step S2, the process of classifying the historical lane-changing trajectory data of the driver according to the lane-changing style includes: extracting features from the historical lane-changing trajectory data, and using the interquartile range method to identify and remove outliers in the extracted features; integrating the processed features to obtain a comprehensive aggressiveness index reflecting the lane-changing style; using K-means clustering to perform unsupervised clustering analysis on the comprehensive aggressiveness index, and classifying the historical lane-changing trajectory data with similar lane-changing style into the same class.

[0013] Further, the process of integrating the processed features to obtain a comprehensive aggressiveness index reflecting the lane-changing style includes: standardizing each processed feature; and adding the weights of the standardized features to obtain the comprehensive aggressiveness index.

[0014] Further, in step S2, based on the historical lane-changing trajectory corresponding to each lane-changing style, the process of obtaining the target trajectory corresponding to each lane-changing style includes: calculating the average value of the historical lane-changing trajectory included in each lane-changing style to obtain the corresponding target trajectory.

[0015] Further, step S3 includes:

[0016] using a planar kinematics model to model the motion state of the autonomous vehicle, to determine a vehicle state vector and a control input vector of the vehicle;

[0017] wherein the vehicle state vector is represented as:

[0018] X(t)=[x(t),y(t),ψ(t),v(t)] T ;

[0019] wherein X(t) represents the vehicle state vector of the autonomous vehicle at time t, x(t) and y(t) represent the longitudinal and lateral coordinate positions of the autonomous vehicle at time t, respectively, ψ(t) represents the heading angle of the autonomous vehicle at time t, and v(t) represents the speed of the autonomous vehicle at time t;

[0020] the control input vector is represented as:

[0021] u(t)=[a(t),ψ’(t)];

[0022] wherein u(t) represents the control input vector of the autonomous vehicle at time t, a(t) represents the acceleration of the autonomous vehicle at time t, and ψ'(t) represents the yaw rate of the autonomous vehicle at time t;

[0023] Based on the vehicle state vector and the control input vector, the future driving state of the autonomous vehicle at the next time is determined by the following formula:

[0024] ;

[0025] wherein Δt represents the sampling period;

[0026] Based on the vehicle state vector and the control input vector, a lane changing control model is modeled by using a Markov decision process, and the lane changing control model is obtained:

[0027] M = (S, A, P, R, γ) ;

[0028] wherein M represents the lane changing control model, S represents the state space, including the vehicle state vector and the surrounding environment information of the autonomous vehicle, A represents the action space, corresponding to the control input vector, P represents the state transition probability of the vehicle state vector, R represents the reward function, and γ represents the discount factor;

[0029] According to the reward function R, the optimal lane changing strategy that meets the future driving state is determined by the intensity learning of the following formula:

[0030] ;

[0031] wherein, represents the optimal lane changing strategy under the kth lane changing style, π represents the lane changing strategy, E represents the expected calculation operation, γ(t) represents the discount factor at time t, and R k (t) represents the reward function under the kth lane changing style at time t.

[0032] Further, the reward function R is obtained by the following formula:

[0033] R = λ1 ∙ R safety + λ2 ∙ R smooth + λ3 ∙ R style ;

[0034] wherein R safety represents the safety reward, R smooth represents the smoothness reward, and R style represents the individualization reward, and λ1, λ2 and λ3 represent the reward weights;

[0035] The safety reward R safety is obtained by the following formula:

[0036] ;

[0037] wherein, d min represents the minimum Euclidean distance between the predicted trajectory and all obstacle vehicles, and epsilon represents a constant to prevent the denominator from being zero;

[0038] smoothness reward R smooth is obtained by the following formula:

[0039] ;

[0040] wherein, H represents the length of the prediction time domain, and h represents the time in the length of the prediction time domain;

[0041] personalized reward R style is obtained by the following formula:

[0042] ;

[0043] wherein, represents the predicted trajectory coordinates of the autonomous vehicle at the time t+h, represents the predicted trajectory coordinates of the autonomous vehicle in the kth lane-changing style at the time h.

[0044] Further, the minimum Euclidean distance d min is obtained by the following formula:

[0045] ;

[0046] wherein, N t represents the set of surrounding obstacle vehicles detected at the current time, represents the position coordinates of the jth vehicle in the set of surrounding obstacle vehicles of the autonomous vehicle.

[0047] Compared with the prior art, the application can achieve the following beneficial effects:

[0048] The automatic driving lane-changing style switching method based on the driver's riding preference provided by the application establishes a driver's riding preference analysis model using a large language model, takes the voice command made by the driver to the automatic driving system in real time as the input of the model, uses the powerful logical reasoning and intention understanding ability of the large language model, designs efficient prompt words and calling logic, analyzes the riding preference implied by the driver's voice command, and outputs the lane-changing style identifier corresponding to the riding preference as the analysis result, thereby providing a basis for switching the automatic driving lane-changing style, and improving the individualization and comfort of the automatic driving lane-changing process. BRIEF DESCRIPTION OF DRAWINGS

[0049] The accompanying drawings, which form a part of this specification, are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification. The embodiments of the application, and their

[0050] Figure 1 Flowchart of the automatic driving lane changing style switching method based on driver riding preference according to an embodiment of the application. DETAILED DESCRIPTION

[0051] In order to make the objectives, technical solutions and advantages of the application clearer, the application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application and do not constitute limitations on the application.

[0052] It should be noted that the embodiments in the application and the features in the embodiments can be combined with each other without conflict.

[0053] In the description of the application, it should be understood that the terms "center", "longitudinal", "transverse", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only used to facilitate the description of the application and simplify the description, and therefore cannot be understood as indicating or implying that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the application. In addition, the terms "first", "second" and the like are only used for description purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features limited by "first", "second" and the like can explicitly or implicitly include one or more features. In the description of the application, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0054] In the description of the application, it should be noted that unless otherwise specified and limited, the terms "mounting", "connection", "connection" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the communication between two elements. For those skilled in the art, the specific meaning of the above terms in the application can be understood according to the specific circumstances.

[0055] The application will be described in detail below with reference to the drawings and embodiments.

[0056] As Figure 1As shown, the automatic driving lane-changing style switching method based on the driving preference of the driver according to the embodiments of the application comprises the following steps:

[0057] S1: determining the lane-changing style category of the driver based on the driving preference of the driver by using a lane-changing style analysis model.

[0058] In step S1 of the embodiments of the application: inputting the voice input instruction and the dialogue history of the driver into the lane-changing style analysis model, and determining the lane-changing style category corresponding to the voice input instruction; the lane-changing style category includes aggressive, conservative and negative.

[0059] Specifically, in order to accurately analyze the driving preference of the driver, the driving preference of the driver is defined as aggressive, normal and conservative in the embodiments of the application, and for the convenience of expression and processing, these driving preferences can be regarded as a discrete set D={aggressive, normal, conservative}, wherein D represents the driving preference set. When the vehicle is started or no explicit instruction of the driver is received, the system initializes the driving preference as normal, and the corresponding lane-changing style category is conservative. In addition, the large dialogue language model "moonshot-v1-8k" provided by the Moonshot platform is used as the lane-changing style analysis model in the embodiments of the application, and the analysis process of the driving preference of the driver is realized by using this model. The corresponding lane-changing style is obtained by using the large language model through the method of calling the program programming interface.

[0060] Since the purpose of the voice instruction issued by the driver at a certain moment is to change the lane-changing style at the current moment, i.e., the more aggressive and conservative driving preference of the driver compared with the current automatic driving lane-changing style. In order to realize the dynamic identification of the driving preference according to the current automatic driving lane-changing style, the levels of the three defined driving preferences aggressive, normal and conservative are highest, middle and lowest in turn, and the analysis model dynamically adjusts the level of the driving preference according to the following task rules:

[0061] judging the lane-changing style intention bias of the driver based on the input voice text;

[0062] if the input voice text expresses the lane-changing style that the driver hopes to be more conservative, the current lane-changing style level is lowered by one level;

[0063] if the input voice text expresses the lane-changing style that the driver hopes to be more aggressive, the current lane-changing style level is raised by one level.

[0064] It should be noted that each adjustment only allows switching between adjacent levels, and cross-level jumps of lane-changing style levels are prohibited.

[0065] It can be seen that the application converts the process of analyzing the driver's ride preference into a mapping from natural language input to the lane-changing style identifier corresponding to the ride preference, that is:

[0066] (I, E) = f(U, H, S current );

[0067] Where I represents the lane-changing style identifier, representing the target lane-changing style corresponding to the driver's ride preference, including three values of aggressive, conservative and negative, E represents the explanation text corresponding to the lane-changing style I, used for verification analysis of the analysis result, U represents the driver's voice instruction, H represents the dialogue history between the input lane-changing style analysis model and the driver's voice instruction, S current represents the numerical coding of the current lane-changing style, and f represents the lane-changing style analysis model, wherein the voice instruction U and the dialogue history H are preprocessed text sequences.

[0068] The application outputs the lane-changing style identifier corresponding to the current ride preference obtained by analyzing the driver's voice instruction through the lane-changing style analysis model, realizes real-time analysis of the driver's ride preference, and ensures that the style switching process has history and gradualness. The application can adapt to various natural language expressions by introducing a large language model, has universality and robustness, and meets the human-machine semantic understanding needs in actual automatic driving interaction systems.

[0069] Based on the target lane-changing style identifier obtained in step S1, in order to enable the automatic driving control system to realize the style lane-changing action according to the lane-changing style identifier, the application needs to provide the automatic driving control system with a mimicked style lane-changing trajectory, so a driver lane-changing trajectory data classification method is designed for the driver lane-changing trajectory data, which is used to extract feature parameters of the lane-changing process and classify the lane-changing behavior according to the driving style.

[0070] S2: According to the historical lane-changing trajectory data of the driver, the lane-changing style obtained in step S1 is classified; based on the historical lane-changing trajectory corresponding to each lane-changing style, the target trajectory corresponding to each lane-changing style is obtained.

[0071] In step S2 of the embodiment of the application, the process of classifying the historical lane-changing trajectory data of the driver according to the lane-changing style includes:

[0072] S21: Feature extraction is performed on the historical lane-changing trajectory data, and the quartile interval method is used to identify and remove outliers in the extracted features.

[0073] Specifically, in this embodiment of the invention, historical lane-changing trajectory data during the lane-changing process is obtained from vehicle sensors or test data, with each lane change considered as a sample. By analyzing the lateral and longitudinal motion states of the vehicle during the lane change, key feature parameters characterizing the lane-changing style are extracted. Specifically, this embodiment of the invention extracts five features from the historical lane-changing trajectory data, including the average lateral velocity of the vehicle during the lane change. Mean rate of change of heading angle Mean longitudinal acceleration Mean lateral acceleration and initial distance of lane change Specifically:

[0074] Mean lateral velocity Defined as the mean rate of change of the lateral position coordinate y (in meters) of a vehicle during lane changing with time t (in seconds), expressed as:

[0075] ;

[0076] in, This represents the lateral position coordinates at the i-th lane change data sampling point. Indicates the sampling period, and N represents the total number of sampling points for track change data;

[0077] mean rate of change of heading angle Defined as the average rate of change of the vehicle's heading angle θ (unit: rad) over time, expressed as:

[0078] ;

[0079] in, The heading angle θ represents the vehicle's trajectory at the i-th lane change data sampling point;

[0080] mean longitudinal acceleration Defined as the average longitudinal acceleration during lane changing, expressed as:

[0081] ;

[0082] in, It is represented as the lateral velocity at the i-th lane change data sampling point;

[0083] Mean lateral acceleration Lateral velocity The mean of the derivative over time is expressed as:

[0084] ;

[0085] in, Vx(i) represents the lateral velocity at the i-th lane-changing data sampling point derivative of time t;

[0086] lane-changing initial distance defined as the Euclidean distance between the lane-changing vehicle and the adjacent vehicle at the lane-changing start time, denoted as:

[0087] ;

[0088] wherein, represents the initial coordinate position of the lane-changing vehicle, represents the initial coordinate position of the adjacent vehicle.

[0089] Since there are noises or abnormal situations in the lane-changing data, in order to prevent abnormal data from interfering with the analysis results, the present application adopts the interquartile range method to identify and remove outliers. Specifically,

[0090] The lower quartile Q1 and the upper quartile Q3 of the sample data set of each feature are calculated Then the interquartile range IQR of the feature is obtained: The threshold range of outliers is determined by the interquartile range IQR: sample values lower than or higher than are regarded as outliers. For any lane-changing sample, if any of its feature values falls outside the above abnormal range, it is determined that the sample is an abnormal sample, which is removed from the data set. Through the outlier removal step, the reliability of the data used in subsequent analysis is ensured.

[0091] S22: The processed features are integrated to obtain a comprehensive aggressiveness index reflecting the lane-changing style.

[0092] Since the value range and unit of each feature are different, directly using the unprocessed features for clustering will lead to the fact that the features with large dimensions have an undue dominant effect on distance measurement. Therefore, in the process of integrating the processed features to obtain a comprehensive aggressiveness index reflecting the lane-changing style, in order to eliminate the influence of the dimension difference of different features on classification, the feature data obtained in step S21 is first standardized. The Z-score method is used to convert the feature values into dimensionless standard values. Specifically, the lateral velocity mean , the heading angle change rate mean , the longitudinal acceleration mean , the lateral acceleration mean and the lane-changing initial distance are standardized as follows:

[0093] ; ;​ ; ; ;

[0094] wherein, , , , and respectively represent the mean of the lateral velocity, the mean of the heading angle rate, the mean of the longitudinal acceleration, the mean of the lateral acceleration and the initial distance of lane changing after standardization, ε represents the mean of the corresponding feature, and σ represents the standard deviation of the corresponding feature;

[0095] Based on the standardized features, a comprehensive aggressiveness index A is constructed to measure the overall aggressiveness of the lane changing behavior of the driver, and the calculation formula of the comprehensive aggressiveness index A is:

[0096] ;

[0097] wherein, w1-w5 respectively represent the weight of the corresponding feature. In the comprehensive aggressiveness index A, the greater the mean of the lateral velocity, the mean of the heading angle rate, the mean of the longitudinal acceleration and the mean of the lateral acceleration, the more intense the lane changing action is, and the greater the initial distance of lane changing represents the greater the initial interval with the surrounding vehicles, and the lower the intensity of lane changing, so it appears in the form of reciprocal addition in the index. By adjusting the value of the weight, the contribution of different features to the aggressiveness index can be balanced, so that the index can comprehensively reflect the intensity of the lane changing behavior. Since each index has been standardized, the weight can be 1, and the calculation is a dimensionless value. The greater the value, the more intense the lane changing behavior is.

[0098] S23: unsupervised clustering analysis is performed on the comprehensive aggressiveness index by using K-means clustering, and historical lane changing trajectory data with similar lane changing styles are classified into the same class. In the embodiment of the present application, it is assumed that there are M lane changing samples in total, and the aggressiveness index of each sample is denoted as A j , j=1, 2,..., M, first, K cluster centers o1, o2,..., o k ,..., o K For each aggressiveness index A j , the distance |A j -o k | between it and each cluster center is calculated, and the aggressiveness index A j is assigned to the nearest cluster o k ; subsequently, according to the samples assigned to the same cluster, the center of the cluster is updated to be the average of all sample indexes, that is,

[0099] ;

[0100] wherein, C kdenotes the sample subset contained in the kth cluster, |C k denotes the sample subset contained in the kth cluster. By constantly iterating the minimum of the sum of squares of the distance between the sample and the cluster center to which it belongs, the optimal clustering division is realized, and the optimization objective can be expressed as:

[0101] ;

[0102] In the next two iterations, if the position change amplitude of all cluster centers is less than a preset threshold, it is determined that the clustering has converged, and the iteration process is terminated. In the embodiment of the present application, the threshold is set to 10 -4 .

[0103] In the embodiment of the present application, the number of clusters is taken as 3, that is, K = 3, and the lane changing trajectory data is automatically divided into three clusters according to the difference of the aggressiveness index, corresponding to aggressive, normal and conservative lane changing styles respectively. The value of the cluster center o k can be used to judge the aggressive behavior level of each type of lane changing style. The type with the maximum aggressiveness index is defined as the aggressive style, the type with the minimum aggressiveness index is defined as the conservative style, and the intermediate value corresponds to the normal style.

[0104] Finally, the corresponding lane changing style label is determined based on the clustering result. The value corresponding to the cluster center of each cluster obtained by clustering represents the overall style characteristics of the cluster. The clusters are divided into different style types: the cluster with the lowest average aggressiveness is labeled as the "conservative" lane changing style, the cluster with the highest aggressiveness is labeled as the "aggressive" lane changing style, and the cluster with the intermediate level is labeled as the "normal" lane changing style.

[0105] At this point, the design of the driver lane changing trajectory data classification method is completed. Through the above steps, the lane changing trajectory data of different drivers can be effectively classified according to the style type. The classification method finally outputs three sets of driver lane changing trajectory data labeled with lane changing styles, providing imitable stylized lane changing trajectories for the automatic driving control system.

[0106] In the embodiment of the present application, the target trajectory corresponding to each type of lane changing style is obtained based on the historical lane changing trajectory corresponding to each type of lane changing style in step S2. Specifically, for each type of style, the number of lane changing trajectory samples belonging to the style in the clustering result is denoted as S, and the corresponding trajectory set is denoted as:

[0107] ;

[0108] wherein, denotes the trajectory set, denotes the s-th lane changing trajectory sample. For each type of style, the average trajectory is calculated from the classified vehicle lane changing trajectory data to obtain the corresponding target trajectory, that is, the corresponding target lane changing trajectory is obtained by the following formula:

[0109] ;

[0110] wherein, denotes the target trajectory corresponding to the s-th trajectory sample, denotes the position at time t in the s-th trajectory sample.

[0111] S3: predicting a future driving state of the vehicle based on the current driving state of the vehicle; determining a trajectory switching strategy in line with the future driving state based on the future driving state and the target trajectory corresponding to each trajectory switching style obtained in step S2 by using reinforcement learning.

[0112] The application proposes a personalized trajectory switching decision method. The personalized trajectory switching decision system does not rely on only experience-based trial-and-error reinforcement learning, but also combines the idea of model predictive control: the trajectory switching decision system evaluates the evolution of the interaction between the trajectory and the environment in the future prediction time domain by using a vehicle kinematic prediction model, so as to introduce the consideration of future dynamics in the decision-making process, and thus to balance the safety-smoothness-personalization triple objective in the planning stage. Specifically, in some embodiments, step S3 comprises:

[0113] A planar kinematic model is used to model the motion state of the autonomous vehicle, and a vehicle state vector and a vehicle control input vector are determined. The vehicle state vector is expressed as:

[0114] X(t) = [x(t), y(t), ψ(t), v(t)] T ;

[0115] wherein X(t) denotes the vehicle state vector of the autonomous vehicle at time t, x(t) and y(t) respectively denote the longitudinal and lateral coordinate positions of the autonomous vehicle at time t, ψ(t) denotes the heading angle of the autonomous vehicle at time t, and v(t) denotes the speed of the autonomous vehicle at time t.

[0116] The control input vector is expressed as:

[0117] u(t) = [a(t), ψ'(t)];

[0118] wherein u(t) denotes the control input vector of the autonomous vehicle at time t, a(t) denotes the acceleration of the autonomous vehicle at time t, and ψ'(t) denotes the yaw rate of the autonomous vehicle at time t;

[0119] Based on the vehicle state vector and the control input vector, the future driving state of the autonomous vehicle at the next time is determined by the following formula:

[0120] ;

[0121] Wherein, Δt represents a sampling period.

[0122] Based on the vehicle state vector and the control input vector, a lane changing control model is obtained by using a Markov decision process for lane changing control modeling.

[0123] M=(S,A,P,R,γ);

[0124] Wherein, M represents the lane changing control model, S represents the state space, including the vehicle state vector, and the surrounding environment information of the autonomous vehicle, A represents the action space, corresponding to the control input vector, P represents the state transition probability of the vehicle state vector, R represents the reward function, and γ represents the discount factor. In the embodiment of the application, the state transition probability P is 1.

[0125] The application adopts reinforcement learning to find the optimal lane changing strategy, and maximizes the expected cumulative return. Specifically, according to the reward function R, the optimal lane changing strategy that meets the future driving state is determined by the intensity learning of the following formula:

[0126] ;

[0127] Wherein, represents the optimal lane changing strategy under the kth lane changing style, π represents the lane changing strategy, E represents the expected calculation operation, γ(t) represents the discount factor at time t, and R k (t) represents the reward function under the kth lane changing style at time t. In the embodiment of the application, k=1 represents that the lane changing style is aggressive, k=2 represents that the lane changing style is conservative, and k=3 represents that the lane changing style is negative.

[0128] In some embodiments, the reward function R is obtained by the following formula:

[0129] R=λ1∙R safety +λ2∙R smooth +λ3∙R style ;

[0130] Wherein, R safety represents the safety reward, R smooth represents the smoothness reward, and R style represents the individualized reward, and λ1, λ2 and λ3 represent the reward weights.

[0131] The safety reward R safety is obtained by the following formula:

[0132] ;

[0133] Wherein, d minrepresents the minimum Euclidean distance between the predicted trajectory and all obstacle vehicles, and ε represents a constant to prevent the denominator from being zero, where ε>0, the smaller the minimum Euclidean distance d, the greater the penalty, reflecting the risk of collision. The minimum Euclidean distance d min is obtained by the following formula:

[0134] ;

[0135] where N t represents the set of surrounding obstacle vehicles detected at the current time, represents the position coordinates of the jth vehicle in the set of surrounding obstacle vehicles of the autonomous vehicle.

[0136] Smoothness reward R smooth To make the vehicle control process smooth, limit the sharp acceleration and steering, which is obtained by the following formula:

[0137] ;

[0138] where H represents the prediction time domain length, and h represents the time in the prediction time domain length.

[0139] Personalized reward R style is obtained by the following formula:

[0140] ;

[0141] where, represents the predicted trajectory coordinates of the autonomous vehicle at t+h, represents the predicted trajectory coordinates of the autonomous vehicle at h in the kth lane-changing style.

[0142] In the training process, the personalized lane-changing decision system interacts with the environment, and the above rewards are used as feedback signals to learn three sets of optimal strategies respectively;

[0143] S4: According to the lane-changing strategy obtained in step S3, the vehicle is controlled to perform lane-changing action according to the lane-changing strategy consistent with the current lane-changing style of the driver.

[0144] In the running phase, the method provided by the present application comprises:

[0145] Firstly, the voice instruction of the driver is received, which is converted into text via a voice recognition module and appended to a dialogue history, a large language model is used to give an analysis result, and an optimal strategy corresponding to the analysis result is selected. At each control time, a control input u(t) is generated based on a current vehicle state and an environmental observation, a future predicted trajectory is calculated within a prediction time domain, and closed-loop tracking of underlying actions such as acceleration and steering is completed by an actuator, so as to realize lane changing behavior consistent with the target style. For different style selections, the application presents differentiated control characteristics under the premise of ensuring safety and feasibility constraints: when the lane changing is more rapid and decisive, when the lane changing is more stable and cautious, and when the lane changing is between the two, a compromise is made. The system synchronously updates the analysis result as a new one, which is used as a reference state for subsequent voice analysis and strategy calling. Through the above mechanism, the system can realize online and switchable control of the three styles in a closed loop of semantic intention-strategy selection-prediction control-safety check during the running stage, so as to meet the personalized needs of the driver without violating the safety principle.

[0146] It should be understood that the steps shown above can be reordered, added, or deleted. For example, the steps described in the present disclosure can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions of the present disclosure can be achieved, and the present disclosure is not limited herein.

[0147] The above specific embodiments do not constitute a limitation on the scope of protection of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. An automatic driving lane-changing style switching method based on a driver's ride preference, characterized by, Comprise: S1: using lane changing style analysis model, based on the driving of the driver's ride preference, determine the driver's lane changing style category; In step S1: the driver's voice input instruction and conversation history input into the lane changing style analysis model, determine the voice input instruction corresponding to the lane changing style category; the lane changing style category includes aggressive, conservative and negative; S2: according to the driving of the driver's historical lane changing trajectory data, according to the lane changing style obtained in step S1; based on the historical lane changing trajectory corresponding to each lane changing style, the target trajectory corresponding to each lane changing style is obtained; In step S2, the process of classifying the historical lane changing trajectory data according to the lane changing style from the driver includes: extracting features from the historical lane changing trajectory data, and using the interquartile range method to identify and remove outliers in the extracted features; the processed features are integrated to obtain a comprehensive aggressiveness index reflecting the lane changing style; using K-means clustering to perform unsupervised clustering analysis on the comprehensive aggressiveness index, and classifying the historical lane changing trajectory data with similar lane changing style into the same class; In step S2, based on the historical lane changing trajectory corresponding to each lane changing style, the process of obtaining the target trajectory corresponding to each lane changing style includes: calculate the average value of the historical lane changing trajectory included in each lane changing style, and obtain the corresponding target trajectory; S3: predict the future driving state of the vehicle with the current driving state of the vehicle; using reinforcement learning, based on the future driving state and the target trajectory corresponding to each lane changing style obtained in step S2, determine the lane changing strategy conforming to the future driving state; S4: according to the lane changing strategy obtained in step S3, control the vehicle to change lane according to the lane changing strategy conforming to the current lane changing style of the driver.

2. The automatic driving lane-changing style switching method based on a driver's seating preference according to claim 1, characterized in that, The process of integrating the processed features to obtain a comprehensive aggressiveness index reflecting the lane changing style includes: standardize each processed feature; the standardized features are added up with weights to obtain the comprehensive aggressiveness index. 3.The automatic driving lane-changing style switching method based on a driver's seating preference according to claim 1, wherein Step S3 includes: using a planar kinematics model to model the motion state of the autonomous vehicle, to determine the vehicle state vector and the control input vector of the vehicle; wherein the vehicle state vector is represented as: X(t) = [x(t), y(t), ψ(t), v(t)] T ; wherein X(t) represents the vehicle state vector of the autonomous vehicle at time t, x(t) and y(t) represent the longitudinal and lateral coordinate positions of the autonomous vehicle at time t, respectively, ψ(t) represents the heading angle of the autonomous vehicle at time t, and v(t) represents the speed of the autonomous vehicle at time t; the control input vector is represented as: u(t)=[a(t),ψ’(t)]; wherein u(t) represents the control input vector of the autonomous vehicle at time t, a(t) represents the acceleration of the autonomous vehicle at time t, and ψ’(t) represents the yaw rate of the autonomous vehicle at time t; based on the vehicle state vector and the control input vector, the future driving state of the autonomous vehicle at the next time is determined by the following formula: ； wherein Δt represents the sampling period; Based on the vehicle state vector and the control input vector, the lane change control model is modeled by using Markov decision process, and the lane change control model is obtained: M=(S,A,P,R,γ); Wherein, M represents the lane change control model, S represents the state space, including the vehicle state vector, and the surrounding environment information of the autonomous vehicle, A represents the action space, corresponding to the control input vector, P represents the state transition probability of the vehicle state vector, R represents the reward function, and γ represents the discount factor; According to the reward function R, the optimal lane change strategy meeting the future driving state is determined by the intensity learning of the following formula: ； wherein, represents the optimal lane-changing strategy under the kth lane-changing style, represents the lane-changing strategy, E represents the expectation calculation operation, represents the discount factor at time t, R k represents the reward function under the kth lane-changing style at time t.

4. The automatic driving lane-changing style switching method based on a driver's seating preference according to claim 3, characterized in that, The reward function R is obtained by the following formula: R = λ1 R safety + λ2 R smooth + λ3 R style ; wherein R safety represents a security reward, R smooth represents a smoothness reward, R style represents a personalization reward, and λ1, λ2, and λ3 represent reward weights; Security reward R safety is obtained from the equation: ； where d min represents the minimum Euclidean distance between the predicted trajectory and all obstacle vehicles, and ε represents a constant to prevent the denominator from being zero. Smoothness reward R smooth is obtained from the equation: ； Wherein, H represents the prediction time domain length, and h represents the time in the prediction time domain length; Personalized reward R style is obtained from the equation: ； wherein, represents the predicted trajectory coordinates of the autonomous vehicle at time t + h, represents the predicted trajectory coordinates of the autonomous vehicle at time h in the kth lane-changing style.

5. The automatic driving lane-changing style switching method based on a driver's seating preference according to claim 4, characterized in that, Minimum Euclidean distance d min is obtained from the equation: ； wherein N t represents a set of surrounding obstacle vehicles detected at the current time, represents the position coordinates of the jth vehicle in the set of surrounding obstacle vehicles of the autonomous vehicle.

Images