A method, device, equipment and medium for unmanned vehicle lane change decision

Abstract

The present application discloses a lane-changing decision method, device, equipment and medium for unmanned vehicles, and relates to the field of unmanned vehicle driving decision planning and control technology. The method comprises: inputting the vehicle time feature sequence of a set historical time period into a trained social vehicle intention inference model to obtain the intention recognition result of the future time period, and inputting the intention recognition result of the future time period and the historical trajectory of the vehicle in the set historical time period into a trained driver FNN lane-changing decision model to obtain the lane-changing decision instruction of the current vehicle. The trained social vehicle intention inference model integrates the attention mechanism to infer the intention of the driver during the lane-changing process, and makes lane-changing decisions through the trained driver FNN lane-changing decision model, which improves the interpretability and realizes lane-changing decisions for complex driving scenarios and styles.

Description

Technical Field

[0001] The present application relates to the technical field of unmanned vehicle driving decision-making planning and control, and in particular to an unmanned vehicle lane change decision-making method, device, equipment and medium. Background Art

[0002] With the development of the economy, the improvement of people's living standards, the continuous advancement of autonomous driving technology and the gradual saturation of the vehicle market, driverless technology has become a hot topic for the global automotive industry and technology companies to compete for research, and driverless vehicles have gradually moved from the laboratory to practical applications. Driverless vehicles perceive the surrounding environment through a variety of sensors, combined with high-precision maps and complex decision-making, planning and control algorithms to achieve autonomous navigation and driving. Among them, lane change decision-making is an important part of driverless driving. It requires the driverless vehicle to accurately judge the driving conditions of the current lane and perform lane change operations safely and effectively according to the driving task, traffic rules and the dynamics of surrounding vehicles. Existing autonomous driving systems have achieved certain results in lane change decisions in single driving scenarios and single driving styles. However, in actual applications, driverless vehicles often need to face a variety of driving scenarios and styles, which puts higher requirements on their lane change decision-making capabilities.

[0003] At present, the mainstream lane-changing decision methods for autonomous driving are mainly based on rule-driven and traditional machine learning algorithms, which often face several key technical problems. First, most existing methods perform well in a single scenario, but their performance in multiple complex scenarios (such as urban roads, highways, rural roads, etc.) is often unsatisfactory. Existing methods lack the ability to effectively handle multi-scenario switching. Secondly, different driving styles (such as conservative and aggressive) have different lane-changing decision requirements. Existing methods are usually difficult to adapt to multiple driving styles and cannot be flexibly adjusted according to actual needs. Finally, the deep learning method with the best results currently lacks interpretability, and its decision-making process is often a black box operation and lacks transparency. This brings safety risks and regulatory challenges in practical applications. Summary of the invention

[0004] The purpose of this application is to provide a lane change decision method, device, equipment and medium for an unmanned vehicle, which can adapt to a variety of driving scenarios and styles, and make lane change decisions through a trained driver FNN lane change decision model, thereby improving interpretability.

[0005] To achieve the above objectives, this application provides the following solutions.

[0006] In a first aspect, the present application provides a method for making lane change decisions for an unmanned vehicle, comprising the following steps.

[0007] In response to a lane-changing demand, when there are leading and trailing vehicles in the target lane and the current lane, a vehicle time feature sequence of a set historical time period is obtained; the vehicle time feature sequence of the set historical time period includes the historical trajectory of the ego vehicle and the behavioral interaction data of surrounding vehicles in the set historical time period; the historical trajectory of the ego vehicle includes the longitudinal speed of the current vehicle, the lateral speed of the current vehicle, the longitudinal acceleration of the current vehicle, and the lateral acceleration of the current vehicle; the behavioral interaction data of surrounding vehicles includes the relative distance between the current vehicle and the vehicle behind the target lane, the relative distance between the current vehicle and the vehicle ahead of the target lane, the relative distance between the current vehicle and the vehicle ahead of the current lane, and the relative distance between the vehicle ahead of the target lane and the vehicle behind the target lane.

[0008] The vehicle time feature sequence of a set historical time period is input into the trained social vehicle intention reasoning model to obtain the intention recognition result of the future time period; the intention recognition result includes the yielding intention of the front and rear vehicles in the target lane and the current lane; the trained social vehicle intention reasoning model is constructed based on the attention mechanism.

[0009] The intention recognition results of the future time period and the historical trajectory of the vehicle in the set historical time period are input into the trained driver FNN lane-changing decision model to obtain the lane-changing decision instruction of the current vehicle.

[0010] In a second aspect, the present application provides an unmanned vehicle lane change decision device, comprising the following modules.

[0011] A feature acquisition module is used to respond to lane changing requirements. When there are leading and trailing vehicles in the target lane and the current lane, a vehicle time feature sequence of a set historical time period is acquired. The vehicle time feature sequence of the set historical time period includes the historical trajectory of the vehicle and the behavioral interaction data of surrounding vehicles in the set historical time period. The historical trajectory of the vehicle includes the longitudinal speed of the current vehicle, the lateral speed of the current vehicle, the longitudinal acceleration of the current vehicle, and the lateral acceleration of the current vehicle. The behavioral interaction data of surrounding vehicles includes the relative distance between the current vehicle and the trailing vehicle in the target lane, the relative distance between the current vehicle and the leading vehicle in the target lane, the relative distance between the current vehicle and the leading vehicle in the current lane, and the relative distance between the leading vehicle in the target lane and the trailing vehicle in the target lane.

[0012] The intention recognition module is used to input the vehicle time feature sequence of a set historical time period into the trained social vehicle intention reasoning model to obtain the intention recognition result of the future time period; the intention recognition result includes the yielding intention of the front and rear vehicles in the target lane and the current lane; the trained social vehicle intention reasoning model is constructed based on the attention mechanism.

[0013] The lane-changing decision module is used to input the intention recognition results of the future time period and the historical trajectory of the vehicle in the set historical time period into the trained driver FNN lane-changing decision model to obtain the lane-changing decision instruction of the current vehicle.

[0014] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned unmanned vehicle lane change decision method.

[0015] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned unmanned vehicle lane change decision method.

[0016] According to the specific embodiments provided in this application, this application discloses the following technical effects: This application provides a method, device, equipment and medium for unmanned vehicle lane changing decision-making. The trained social vehicle intention reasoning model integrates the attention mechanism to infer the driver's intention during the lane changing process, and makes lane changing decisions through the trained driver FNN lane changing decision model, which improves interpretability, solves the problem that the existing lane changing decision method cannot adapt to a variety of driving scenarios and styles, and realizes lane changing decisions in complex driving scenarios and styles. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 This is a diagram of the application environment of an unmanned vehicle lane change decision method in one embodiment of the present application.

[0019] Figure 2 A flowchart of a lane-changing decision method for an unmanned vehicle provided in one embodiment of the present application.

[0020] Figure 3 A schematic diagram of the modeling process of each module of the lane change decision framework provided in one embodiment of the present application.

[0021] Figure 4 A schematic diagram of general dynamic environment modeling under multiple driving scenarios and multiple driving styles provided in one embodiment of the present application.

[0022] Figure 5 A schematic diagram of the structure of a social vehicle intention reasoning model provided for one embodiment of the present application.

[0023] Figure 6 A schematic diagram of the structure of a driver's FNN lane-changing decision model provided in one embodiment of the present application.

[0024] Figure 7 A schematic diagram of the structure of a driver's lane-changing decision model integrating intention reasoning provided in one embodiment of the present application.

[0025] Figure 8 A schematic diagram of the structure of a lane-changing decision model for driver spatiotemporal interaction provided in one embodiment of the present application.

[0026] Fig. 9 A schematic diagram of the functional modules of a lane-changing decision device for an unmanned vehicle provided in another embodiment of the present application.

[0027] Fig.10 A schematic diagram of the structure of a computer device provided in one embodiment of the present application.

[0028] Reference numerals: server—102, terminal—104. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0030] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0031] The unmanned vehicle lane change decision method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the vehicle time feature sequence of the set historical time period to the server 104. After the server 104 receives the vehicle time feature sequence of the set historical time period, for the vehicle time feature sequence of the set historical time period, the server 104 inputs the vehicle time feature sequence of the set historical time period into the trained social vehicle intention reasoning model to obtain the intention recognition result of the future time period, and inputs the intention recognition result of the future time period and the historical trajectory of the self-vehicle of the set historical time period into the trained driver FNN lane change decision model to obtain the lane change decision instruction of the current vehicle. The server 104 can feedback the lane change decision instruction of the current vehicle for the vehicle time feature sequence of the set historical time period to the terminal 102. In addition, in some embodiments, the unmanned vehicle lane change decision method can also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 can directly perform lane change decision processing on the vehicle time feature sequence of a set historical time period, or the server 104 can obtain the vehicle time feature sequence of a set historical time period from the data storage system, and perform lane change decision processing on the historical time series data to be processed.

[0032] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, and IoT devices, and the IoT devices may be smart speakers, smart TVs, smart air conditioners, smart car devices, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0033] In an exemplary embodiment, Figure 2 As shown, a method for lane changing decision of an unmanned vehicle is provided. The method is executed by a computer device, and specifically can be executed by a computer device such as a terminal or a server alone, or can be executed by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used for explanation, and the steps include the following steps 201 to 203.

[0034] Step 201: In response to a lane change demand, when there are leading and trailing vehicles in the target lane and the current lane, a vehicle time feature sequence of a set historical time period is obtained; the vehicle time feature sequence of the set historical time period includes the historical trajectory of the vehicle and the behavioral interaction data of surrounding vehicles in the set historical time period; the historical trajectory of the vehicle includes the longitudinal speed of the current vehicle, the lateral speed of the current vehicle, the longitudinal acceleration of the current vehicle, and the lateral acceleration of the current vehicle; the behavioral interaction data of surrounding vehicles includes the relative distance between the current vehicle and the trailing vehicle in the target lane, the relative distance between the current vehicle and the leading vehicle in the target lane, the relative distance between the current vehicle and the leading vehicle in the current lane, and the relative distance between the leading vehicle in the target lane and the trailing vehicle in the target lane.

[0035] Step 202: Input the vehicle time feature sequence of the set historical time period into the trained social vehicle intention reasoning model to obtain the intention recognition result of the future time period; the intention recognition result includes the yielding intention of the front and rear vehicles in the target lane and the current lane; the trained social vehicle intention reasoning model is constructed based on the attention mechanism.

[0036] Step 203: Input the intention recognition result of the future time period and the historical trajectory of the vehicle in the set historical time period into the trained driver FNN lane change decision model to obtain the lane change decision instruction of the current vehicle.

[0037] Implement the above steps 201 to 203, the trained social vehicle intention reasoning model integrates the attention mechanism, infers the intention of the driver in the lane changing process, and makes lane changing decisions through the trained driver FNN lane changing decision model, which improves the interpretability, solves the problem that the existing lane changing decision method cannot adapt to a variety of driving scenarios and styles, and realizes the lane changing decision of complex driving scenarios and styles. This application proposes a multi-scene and multi-style interpretable deep learning lane changing decision framework for unmanned vehicles that considers the spatiotemporal interaction and intention reasoning of social vehicle driving and integrates the attention mechanism fuzzy neural network (I-FNN). The lane changing decision framework has an adaptive learning mechanism, which enables the model to quickly adapt to different traffic scenarios and different driving styles, and facilitates the deployment of the algorithm to different scenarios and adapts to different driver styles. In addition, an interpretable decision mechanism is designed based on the fuzzy neural network. By automatically learning and extracting model rules through the model, the deep learning network can output the basis and reason for the decision when making a lane changing decision. This not only helps to quickly locate and solve problems when problems arise, but also improves the user's trust in the unmanned vehicle system. Finally, the attention mechanism and spatiotemporal interaction mechanism are introduced to improve the behavioral interaction and intention reasoning capabilities of autonomous driving vehicles in complex interactive driving scenarios with multiple drivers, ensure the safety of lane changing decisions, reduce potential risks, and generate safe and reliable lane changing decision instructions that adapt to multiple scenarios and styles.

[0038] This application provides a multi-scenario and multi-style interpretable deep learning lane-changing decision framework for unmanned vehicles that considers the spatiotemporal interaction and intention reasoning of social vehicles. The multi-scenario and multi-style interpretable deep learning lane-changing decision framework for unmanned vehicles that considers the spatiotemporal interaction and intention reasoning of social vehicles is abstracted into a general dynamic driving environment modeling module, a driving data collection and feature extraction module, an interaction intention acquisition module based on an attention mechanism, a single-vehicle lane-changing decision module based on an FNN model, and a multi-vehicle spatiotemporal interaction lane-changing decision module based on a Social IFNN algorithm. The modeling process of each module is as follows: Figure 3 shown.

[0039] First, a general dynamic environment model under multiple driving scenarios and multiple driving styles is established, so that the controlled autonomous vehicle can model the dynamic environment of any road scenario. The lane change process usually involves two lanes, namely the current lane before the lane change (Original Lane) and the target lane after the lane change (Target Lane), such as Figure 4 As shown in the figure, the dynamic environment mainly includes the vehicle OP in front of the current lane, the vehicle OF behind the current lane, the vehicle TP in front of the target lane, and the vehicle TF behind the target lane.

[0040] After establishing a general dynamic driving environment model for multiple driving scenarios and multiple driving style scenarios, based on the general dynamic driving environment, the autonomous driving vehicle collects its own driving data and dynamically changing surrounding vehicle information in real time from the perception system or vehicle networking system composed of various sensors.

[0041] In step 201, the vehicle time feature sequence of the historical time period is set ,in express The vehicle feature space at each moment, gap Indicates the length of observation data time. Define the feature space of the vehicle ,in, is the relative distance between the current vehicle and the vehicle behind TF in the target lane, is the relative distance between the current vehicle and the vehicle in front of the target lane TP, is the relative distance between the current vehicle and the vehicle in front of the current lane, is the relative distance between the front vehicle TP and the rear vehicle TF in the target lane, is the current longitudinal velocity of the vehicle, is the current lateral velocity of the vehicle, is the current vehicle's longitudinal acceleration, is the current lateral acceleration of the vehicle.

[0042] It should be noted that according to the survey, drivers generally make decisions and actions at the current moment based on the situation 0.4 seconds before the prediction moment. Therefore, the set time period can be the period between the predicted moment and 0.4 seconds before the predicted moment.

[0043] Before inputting the vehicle time feature sequence of a set historical time period into the trained social vehicle intention reasoning model, the unmanned vehicle lane change decision method of the present application also includes: training the social vehicle intention reasoning model, and the training process is as follows.

[0044] Obtain a multi-scenario and multi-style driving data set; the multi-scenario and multi-style driving data set includes data sets of several different traffic scenarios and driving styles; each data set includes vehicle time feature sequences of several sample historical time periods and labels corresponding to the vehicle time feature sequences of each sample historical time period;

[0045] The social vehicle intention reasoning model is trained using a multi-scenario and multi-style driving dataset to obtain a trained social vehicle intention reasoning model.

[0046] Among them, obtaining a multi-scenario multi-style driving dataset specifically includes: obtaining datasets of several different traffic scenarios; performing multi-style classification of drivers for the datasets of all traffic scenarios to obtain a multi-scenario multi-style driving dataset, specifically including: using an unsupervised clustering algorithm to perform multi-style classification of drivers for the datasets of all traffic scenarios to obtain a multi-scenario multi-style driving dataset.

[0047] This application can handle different types of traffic road environments, such as urban roads, highways, rural roads, etc., different traffic conditions, such as congestion, smooth traffic, etc., and different driving styles of surrounding drivers, such as conservative, aggressive or defensive driving, etc. The dynamic environment model can consistently describe the environment in which the vehicle is located and provide accurate information for lane change decisions. This consistency ensures the versatility and reliability of the social vehicle intention reasoning model in different scenarios.

[0048] Based on the Peachtree and Lankershim Street dataset (street scene), I-80 Highway (highway scene) and HighD (highway scene), multi-scene and multi-style data feature extraction is performed. The extracted data features include . According to the extracted features, multi-style classification of drivers is performed based on the unsupervised clustering algorithm Kmeans. The classification process is as follows: (1) Assignment step: For each data point, the distance (such as Euclidean distance) from all cluster centers is calculated based on its features, and it is assigned to the nearest cluster center. (2) Update step: Recalculate the center of each cluster, that is, calculate the mean of all points in the cluster as the new cluster center. (3) Repeat the assignment and update steps until the cluster center no longer changes significantly or reaches the predetermined number of iterations. After obtaining the driving style classification sets with different numbers of styles, the clustering effect under different numbers of driving styles is evaluated according to the silhouette coefficient method. The calculation method is shown in formula (1).

[0049] (1).

[0050] in, S is the clustering evaluation result, a It represents the average distance from the sample to other samples in the same cluster, and b represents the average distance from the sample to all samples in the nearest cluster. According to the evaluation results, the two styles with the best effect (aggressive driving style and mild driving style) are selected to generate a multi-scene multi-style driving dataset. The multi-scene multi-style driving dataset includes street scene (aggressive style) dataset, street scene (mild style) dataset, highway scene (aggressive style) dataset, highway scene (mild style) dataset, high-speed scene (aggressive style) dataset, and high-speed scene (mild style) dataset.

[0051] like Figure 5 As shown, the trained social vehicle intention inference model includes a projection module, a scene encoder and an intention decoder connected in sequence; wherein the projection module includes a first projection unit and a second projection unit.

[0052] The first projection unit is used to set the historical trajectory of the vehicle in a historical time period to perform data projection processing to obtain first projection data.

[0053] The second projection unit is used to perform data projection processing on the behavior interaction data of surrounding vehicles in a set historical time period to obtain second projection data.

[0054] The scene encoder includes a first attention mechanism unit, a second attention mechanism unit, a serial connection unit, and a third attention mechanism unit.

[0055] The first attention mechanism unit is used to extract features from the first projection data to obtain a first attention feature.

[0056] The second attention mechanism unit is used to extract features from the second projection data to obtain second attention features.

[0057] A concatenation unit is used to concatenate the first attention feature and the second attention feature to obtain a concatenated feature.

[0058] The third attention mechanism unit is used to connect features in series for feature extraction to obtain the third attention feature.

[0059] The intent decoder is used to decode the third attention feature to obtain the intent recognition result for the future time period.

[0060] In the process of driving social vehicles, the driver's intention is crucial to predicting their behavior and avoiding potential conflicts. The Attention model is a deep learning model that simulates the human attention mechanism. It can focus on key information when processing a large amount of information, thereby improving the efficiency and accuracy of the model.

[0061] The core of the Attention mechanism is to generate a weighted average output by calculating the correlation between Query, Key and Value. The specific steps may include steps 301 to 304.

[0062] Step 301: Calculate the Query matrix, Key matrix, and Value matrix according to formula (2).

[0063] (2).

[0064] in, X is the input of the Attention unit, is a trainable parameter matrix.

[0065] Step 302: Calculate the Attention score according to formula (3).

[0066] (3).

[0067] in, Q Represents the query matrix, which indicates the context or location that currently needs attention or query; K Represents the key matrix, which is used to compare with the query vector to decide which value vectors are important; Represents the scaling factor, which is used to scale the dot product result; V Represents the value matrix, which contains the actual information content.

[0068] Step 303: Calculate the Attention weight according to formula (4).

[0069] (4).

[0070] in, Indicates i Query vector With j Key vector The attention weight, n is the number of Key vectors.

[0071] Step 304: weighted output according to formula (5).

[0072] (5).

[0073] in It is the weighted summation result of the result of step 303 on the Value matrix.

[0074] The input of the social vehicle intention reasoning model is the history trajectory of the vehicle and the behavioral interaction data with surrounding vehicles (Agent interaction), that is, the vehicle time feature sequence of the set historical time period. After the projection module performs data projection processing, it is processed by the scene encoder module composed of the attention mechanism unit (Attention Encoder) and then input into the intention decoder module to output the time series intention recognition result (that is, the intention sequence) , ... ), Intention Sequence. Sequential intention recognition is the willingness of the vehicles ahead and behind in the target lane and the current lane to yield in a future time period. The willingness to yield is represented by a number in [0, 1]. The closer the number is to 0, the weaker the yield intention of the surrounding vehicles; the closer the number is to 1, the stronger the yield intention of the surrounding vehicles.

[0075] The definition of the Projection module is shown in equation (6).

[0076] (6).

[0077] in, is the input vector of the Projection module, is the projection matrix, is the bias vector, It is the output vector of the Projection module.

[0078] The intent decoder consists of multiple Attention units and a softmax layer. The softmax layer is defined as shown in the following formula (7).

[0079] (7).

[0080] in, is the input vector of the softmax layer, is the output vector of the softmax layer, is the dimension of the input vector, is the base of natural logarithms.

[0081] The input of each Attention unit in the intent decoder is the same, but the is different, that is to say, the parameter matrix of each Attention unit is different, and the physical meaning is that they focus on different things.

[0082] The trained driver FNN lane-changing decision model includes a membership function layer, a fuzzy rule layer, a normalization layer, and an output layer. Fuzzy neural network (FNN) is a neural network that can integrate expert prior knowledge. Figure 6 As shown in Figure 1, the FNN framework simulates the human decision-making process through three stages: fuzzification, reasoning, and defuzzification. The fuzzification stage (membership function layer) converts the crisp input into a fuzzy set by assigning membership to each input based on its relevance to the fuzzy set; in the reasoning stage (fuzzy rule layer and normalization layer, i.e., standardization layer), fuzzy sets are combined using fuzzy operators to create fuzzy rules, usually expressed in IF-THEN format. These rules are then used to produce fuzzy outputs; finally, the defuzzification stage (output layer) converts the fuzzy output into a crisp output by determining a single value or a set of values ​​that best represents the fuzzy output.

[0083] This application proposes a driver lane-changing decision model that integrates intention reasoning, such as Figure 7 shown.

[0084] Among them, the historical trajectory of the ego vehicle and the behavioral interaction with the surrounding vehicles, that is, the vehicle time feature sequence of the set time period is input into the intention reasoning module (the trained social vehicle intention reasoning model). After generating the intention recognition results of the surrounding vehicles in the future time period, the historical trajectory of the ego vehicle in the set historical time period is combined and input into the trained driver FNN lane changing decision model. After the fusion reasoning of the trained driver FNN lane changing decision model, the instruction on whether the current vehicle should change lanes is finally obtained, that is, the lane changing decision instruction of the current vehicle is obtained.

[0085] Initialize the parameters of the lane-changing decision model based on expert prior knowledge: In a fuzzy neural network, the parameters of the network can be initialized by expert knowledge, such as the membership function of the fuzzy set, the weight of the fuzzy rule, etc. These parameters can be set based on the experience of experts, thus providing a reasonable starting point for learning.

[0086] 1) Use expert knowledge to initialize the membership function of FNN, as shown in equation (8).

[0087] (8).

[0088] in, is the output of the membership layer, express The membership function of is the language label associated with the node function {such as near, middle, far, etc.}, For the j Inputs; and is the parameter set of the Gaussian membership function, which can be initialized and assigned according to expert knowledge and prior experience. Its change leads to the corresponding change of the Gaussian membership function, which in turn leads to the language label The membership functions of are various.

[0089] 2) Use expert knowledge to initialize the rule layer function of FNN, as shown in equation (9).

[0090] (9).

[0091] Among them, this layer is mainly aimed at achieving transparent and accurate expression of the interpretability of fuzzy rules. The node connection mode of this layer can also be defined by expert knowledge and prior experience. In the formula, They are The output of the feature after passing through the membership function layer; V is the vehicle speed, I The output of the intention inference module is the willingness of surrounding vehicles to give way.

[0092] 3) The FNN normalization layer function is defined based on formula (10).

[0093] (10).

[0094] in, is the input vector, i.e., the output of the regularization layer, is the output vector of the normalization layer, is the dimension of the input vector, N is the number of features, M is the number of morphemes in the language set, is the firstk dimensions.

[0095] 4) The FNN output layer function is defined based on formula (11).

[0096] (11).

[0097] in, is the output vector of the output layer, Enter the corresponding parameters for each feature, which are The weights of the connections mapped to the output layer.

[0098] In this application, the initialization of the normalization layer and output layer node parameters is initialized using normal distribution and is defined using the following formula (12).

[0099] (12).

[0100] in, is the initialization function of the normalization layer or output layer node parameters, is the mean of the normal distribution, is the standard deviation of the normal distribution.

[0101] In another exemplary embodiment of the present application, the unmanned vehicle lane change decision method may also include the following steps 401 to 403.

[0102] Step 401: Construct a driver's spatiotemporal interactive lane-changing decision model (Social-IFNN).

[0103] Step 402: Based on the driver's time-space interaction lane-changing decision model, the time-space information of surrounding vehicles is interacted to obtain shared parameters; surrounding vehicles are vehicles whose distance from the current vehicle is less than the set vehicle social distance.

[0104] Step 403: Obtain a real-time lane change decision instruction according to the shared parameters.

[0105] The trained driver Social FNN lane-changing decision model includes a social pool layer, a membership function layer, a fuzzy rule layer, a normalization layer, and an output layer. The social pool layer in the driver's spatiotemporal interactive lane-changing decision model interacts with the spatiotemporal information of surrounding vehicles to obtain shared parameters. The shared parameters include And other parameters.

[0106] This application provides a fuzzy neural network for a single driver that integrates expert prior knowledge and intention reasoning results, defined as I-FNN. Figure 8 The fusion attention mechanism neural network in the Figure 7The driver's lane-changing decision model based on fusion intention reasoning is shown in the figure (the driver's lane-changing decision model based on fusion intention reasoning consists of a social vehicle intention reasoning model and a driver's FNN lane-changing decision model). On this basis, the spatiotemporal information of surrounding vehicles is integrated to propose a lane-changing decision model based on driver spatiotemporal interaction. This lane-changing decision model based on driver spatiotemporal interaction adjusts the parameters of a single vehicle I-FNN in real time by sharing parameters between multiple vehicles, making the lane-changing decision safer and more reliable. The structure of the lane-changing decision model based on driver spatiotemporal interaction is shown in the figure. Figure 8 As shown. Each environment vehicle (the current vehicle and the current vehicle’s OF, OP, TF, TP) deploys an I-FNN ( Figure 7 ), and then fuse the parameters of each I-FNN through social pool (Formula 14).

[0107] The time series processing capability of Social-IFNN can be expressed by formula (13).

[0108] (13).

[0109] In the formula, for T Time series processing results for time periods, For the The timing processing result of the moment, For the T- The timing processing result at time 1, is the timing processing result at the initial moment, for T- The sample distribution corresponding to time 1 is T In the time series data, each I-FNN model Sample distribution for the corresponding stage Optimize instead of the initial distribution at the initial moment This method can gradually improve the entire model as the time series is iterated, rather than focusing only on the initial moment. The " symbol represents an implicit relationship and is a mathematical operation symbol.

[0110] The spatial processing capability of Social-IFNN can be expressed by formula (14).

[0111] (14).

[0112] in, is the spatial processing result, is t- 1 moment corresponds to vehicle The hidden state of the I-FNN, It is a vehicle The corresponding neighbor set. For inspection Is it in the vehicle of In the neighboring cells, for Time Vehicle The coordinates of for Time Vehicle The coordinates of .

[0113] (15).

[0114] in, = , indicating vehicle With vehicle lateral distance; = ,vehicle With vehicle The longitudinal distance; If the vehicle With vehicle The lateral distance is less than 120 and the vehicle With vehicle The longitudinal distance is less than , Indicates if; if the vehicle With vehicle The lateral distance is less than 120 and the vehicle With vehicle The longitudinal distance is less than ,but =1, indicating In vehicle of In neighbor cells; =0, indicating Not in vehicle of Neighboring cells. Horizontal boundaries is the lane width, with a longitudinal limit of 120 meters for vehicle social distance. If the distance between vehicles exceeds the limit, that is, the vehicle With vehicle The lateral distance is not less than 120 and the vehicle With vehicle The longitudinal distance is not less than , then the vehicles are considered to be no longer socially influenced by each other.

[0115] The present application also provides an application scenario, which applies the above-mentioned unmanned vehicle lane change decision method. Specifically: the unmanned vehicle lane change decision method provided in this embodiment can be applied in unmanned vehicle driving scenarios. The unmanned vehicle driving scenario includes a parameter acquisition link, a lane change decision link and a driving link; the vehicle time feature sequence of the set historical time period enters the lane change decision link from the parameter acquisition link, obtains the corresponding lane change decision instruction, and enters the downstream driving link. The unmanned vehicle lane change decision method provided in this embodiment belongs to the lane change decision link. Specifically, in the lane change decision link process for the vehicle time feature sequence of the set historical time period, the vehicle time feature sequence of the set historical time period can be input into the trained social vehicle intention inference model to obtain the intention recognition result of the future time period, and the intention recognition result of the future time period and the historical trajectory of the self-vehicle of the set historical time period are input into the trained driver FNN lane change decision model to obtain the lane change decision instruction of the current vehicle.

[0116] This application is based on FNN that can utilize expert experience. For a large-scale natural driving dataset with multiple scenarios and styles, it integrates the Attention mechanism to infer the driver's intention during the lane changing process, and constructs a multi-level interpretable deep learning network structure (I-FNN), so as to make safe and effective lane changing decisions in a multi-lane environment.

[0117] In order to solve the prediction efficiency problem encountered by a single I-FNN model when predicting continuous decision nodes, this application further proposes a multi-I-FNN model fusion technology that integrates a shared weight adaptive adjustment algorithm with spatiotemporal information. Through this technology, the framework can combine multiple models of different spatiotemporal scales to form a social cascade intention inference fuzzy neural network (Social-IFNN), which not only improves the accuracy of lane change decisions, but also enhances the decision robustness of the entire framework when facing different driving environments and driving styles.

[0118] Based on the same inventive concept, the embodiment of the present application also provides an unmanned vehicle lane change decision device for implementing the unmanned vehicle lane change decision method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more unmanned vehicle lane change decision device embodiments provided below can refer to the limitations of the unmanned vehicle lane change decision method above, and will not be repeated here.

[0119] In an exemplary embodiment, Fig. 9 As shown, a lane-changing decision device for an unmanned vehicle is provided, which includes the following modules.

[0120] The feature acquisition module T1 is used to respond to the lane changing demand. When there are leading vehicles and following vehicles in the target lane and the current lane, the vehicle time feature sequence of the set historical time period is obtained; the vehicle time feature sequence of the set historical time period includes the historical trajectory of the vehicle and the behavioral interaction data of surrounding vehicles in the set historical time period; the historical trajectory of the vehicle includes the longitudinal speed of the current vehicle, the lateral speed of the current vehicle, the longitudinal acceleration of the current vehicle and the lateral acceleration of the current vehicle; the behavioral interaction data of surrounding vehicles includes the relative distance between the current vehicle and the vehicle behind the target lane, the relative distance between the current vehicle and the vehicle ahead of the target lane, the relative distance between the current vehicle and the vehicle ahead of the current lane, and the relative distance between the vehicle ahead of the target lane and the vehicle behind the target lane.

[0121] The intention recognition module T2 is used to input the vehicle time feature sequence of the set historical time period into the trained social vehicle intention reasoning model to obtain the intention recognition result of the future time period; the intention recognition result includes the yielding intention of the front and rear vehicles in the target lane and the current lane; the trained social vehicle intention reasoning model is constructed based on the attention mechanism.

[0122] The lane-changing decision module T3 is used to input the intention recognition results of the future time period and the historical trajectory of the vehicle in the set historical time period into the trained driver FNN lane-changing decision model to obtain the lane-changing decision instruction of the current vehicle.

[0123] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Fig.10 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store lane change decision data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for making a lane change decision for an unmanned vehicle is implemented.

[0124] Those skilled in the art will understand that Fig.10The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0125] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0126] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0127] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0128] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0129] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0130] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0131] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A lane-changing decision method for an unmanned vehicle, characterized in that: The unmanned vehicle lane change decision method comprises: In response to a lane change demand, when there are leading and trailing vehicles in the target lane and the current lane, a vehicle time feature sequence of a set historical time period is obtained; the vehicle time feature sequence of the set historical time period includes the historical trajectory of the ego vehicle and the behavioral interaction data of surrounding vehicles in the set historical time period; the historical trajectory of the ego vehicle includes the longitudinal speed of the current vehicle, the lateral speed of the current vehicle, the longitudinal acceleration of the current vehicle, and the lateral acceleration of the current vehicle; the behavioral interaction data of surrounding vehicles includes the relative distance between the current vehicle and the trailing vehicle in the target lane, the relative distance between the current vehicle and the leading vehicle in the target lane, the relative distance between the current vehicle and the leading vehicle in the current lane, and the relative distance between the leading vehicle in the target lane and the trailing vehicle in the target lane; Inputting the vehicle time feature sequence of a set historical time period into the trained social vehicle intention inference model to obtain the intention recognition result of the future time period; the intention recognition result includes the yielding intention of the front and rear vehicles in the target lane and the current lane; the trained social vehicle intention inference model is constructed based on the attention mechanism; The trained social vehicle intention inference model includes a projection module, a scene encoder and an intention decoder connected in sequence; wherein the projection module includes a first projection unit and a second projection unit; the first projection unit is used to set the historical trajectory of the vehicle in the historical time period for data projection processing to obtain the first projection data; the second projection unit is used to set the behavior interaction data of the surrounding vehicles in the historical time period for data projection processing to obtain the second projection data; the scene encoder includes a first attention mechanism unit, a second attention mechanism unit, a series connection unit and a third attention mechanism unit; the first attention mechanism unit is used to extract features from the first projection data to obtain the first attention feature; the second attention mechanism unit is used to extract features from the second projection data to obtain the second attention feature; the series connection unit is used to connect the first attention feature and the second attention feature in series to obtain the series connection feature; the third attention mechanism unit is used to extract features from the series connection feature to obtain the third attention feature; the intention decoder is used to decode the third attention feature to obtain the intention recognition result of the future time period; Before inputting the vehicle time feature sequence of the set historical time period into the trained social vehicle intention inference model, the unmanned vehicle lane change decision method also includes: training the social vehicle intention inference model, and the training process is as follows: obtaining a multi-scenario multi-style driving data set; the multi-scenario multi-style driving data set includes a number of data sets of different traffic scenes and driving styles; each data set includes a number of vehicle time feature sequences of sample historical time periods and a label corresponding to the vehicle time feature sequence of each sample historical time period; using the multi-scenario multi-style driving data set to train the social vehicle intention inference model, to obtain a trained social vehicle intention inference model; Input the intention recognition result of the future time period and the historical trajectory of the vehicle in the set historical time period into the trained driver FNN lane change decision model to obtain the lane change decision instruction of the current vehicle; the trained driver FNN lane change decision model includes a membership function layer, a fuzzy rule layer, a normalization layer and an output layer; Construct a lane-changing decision model for drivers’ time-space interaction; The lane-changing decision model based on the driver's spatiotemporal interaction interacts with the spatiotemporal information of surrounding vehicles to obtain shared parameters; surrounding vehicles are vehicles whose distance from the current vehicle is less than the set vehicle social distance; Get real-time lane-changing decision instructions based on shared parameters.

2. The unmanned vehicle lane change decision method according to claim 1, characterized in that: Get multi-scenario and multi-style driving datasets, including: Obtain data sets of several different traffic scenarios; The data sets of all traffic scenes are subjected to multi-style classification of drivers to obtain a multi-scene and multi-style driving data set.

3. The lane-changing decision method for an unmanned vehicle according to claim 2, characterized in that: The data sets of all traffic scenes are classified into multiple driver styles to obtain a multi-scene and multi-style driving data set, which includes: Using an unsupervised clustering algorithm, the data sets of all traffic scenarios are classified into multiple driver styles to obtain a multi-scenario and multi-style driving data set.

4. An unmanned vehicle lane-changing decision device based on the unmanned vehicle lane-changing decision method according to claim 1, characterized in that: The unmanned vehicle lane change decision device comprises: A feature acquisition module is used to respond to lane-changing requirements, and when there are leading and trailing vehicles in the target lane and the current lane, acquire a vehicle time feature sequence for a set historical time period; the vehicle time feature sequence for the set historical time period includes the historical trajectory of the vehicle and the behavioral interaction data of surrounding vehicles for the set historical time period; the historical trajectory of the vehicle includes the longitudinal speed of the current vehicle, the lateral speed of the current vehicle, the longitudinal acceleration of the current vehicle, and the lateral acceleration of the current vehicle; the behavioral interaction data of surrounding vehicles includes the relative distance between the current vehicle and the trailing vehicle in the target lane, the relative distance between the current vehicle and the leading vehicle in the target lane, the relative distance between the current vehicle and the leading vehicle in the current lane, and the relative distance between the leading vehicle in the target lane and the trailing vehicle in the target lane; An intention recognition module is used to input the vehicle time feature sequence of a set historical time period into a trained social vehicle intention inference model to obtain the intention recognition result of the future time period; the intention recognition result includes the yielding intention of the preceding and following vehicles in the target lane and the current lane; the trained social vehicle intention inference model is constructed based on the attention mechanism; The lane-changing decision module is used to input the intention recognition results of the future time period and the historical trajectory of the vehicle in the set historical time period into the trained driver FNN lane-changing decision model to obtain the lane-changing decision instruction of the current vehicle.

5. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the unmanned vehicle lane change decision method according to any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the unmanned vehicle lane change decision method described in any one of claims 1 to 3 is implemented.

Images