Automatic driving decision optimization system and method based on data driving

Abstract

The invention discloses an automatic driving decision optimization system and method based on data driving, the system comprises a data production module, a data screening module, a model packaging module and a parameter adjusting and optimizing module, the data production module takes human driving data as input, after labeling, preprocessing and format conversion, key features are extracted, standardized, normalized and coded, and the data screening module is used for screening the human driving data; generating original data meeting the algorithm input requirement in the data driving process; the data screening module screens out corresponding data from the training data according to the scenes divided by the decision maker and performs effective classification; the model packaging module is used for packaging decision codes of C + + and constructing a track pair evaluation cost graph required by training by adopting a truth value evaluation method based on track pairs; and the parameter tuning module obtains decision parameters under the current decision algorithm under the corresponding scene by using a black box optimization method according to the screened data under the different scenes, the packaged decision algorithm model and the track pair evaluation cost graph. According to the method, a rule-based method and a learning-based method are combined, and a large amount of data generated in the running process of the automatic driving vehicle is utilized, so that the upper limit of a decision-making system is improved, the expression of a decision-making device in an actual scene is facilitated, and a data-driven decision-making parameter adjusting technology belonging to a white-box property is realized.

Description

Technical Field

[0001] The present invention relates to a technology in the field of autonomous driving, specifically a data-driven autonomous driving decision optimization system and method. Background Art

[0002] Decision-making technology is the core of the intelligent driving system, affecting safety and comfort. Existing technologies include rule-based and learning-based methods. Rule-based methods rely on experience to design objective functions and seek optimal solutions, but the results are far from those of skilled drivers. Learning-based methods can improve decision-making output effects by training models, but they face the problem of insufficient explainability. Summary of the invention

[0003] The present invention aims at the defect that the prior art is difficult to cope with dynamic and complex environmental changes through training for specific scenarios or tasks. The present invention proposes a data-driven autonomous driving decision optimization system and method, which combines rule-based and learning-based methods and utilizes the large amount of data generated by autonomous driving vehicles during operation, thereby improving the upper limit of the decision system, which is more conducive to the performance of the decision maker in actual scenarios and is a data-driven decision parameter adjustment technology of white box nature.

[0004] The present invention is achieved through the following technical solutions:

[0005] The present invention relates to an automatic driving decision optimization system based on data-driven, comprising: a data production module, a data screening module, a model encapsulation module and a parameter tuning module, wherein: the data production module takes human driving data as input, extracts key features and performs standardization, normalization and encoding after labeling, preprocessing and format conversion, and generates original data in the data-driven process that meets the algorithm input requirements; the data screening module screens out corresponding data from the training data according to each scene divided by the decision maker and performs effective classification; the model encapsulation module encapsulates the C++ decision code and adopts a trajectory pair-based true value evaluation method to construct a trajectory pair evaluation cost map required for training; the parameter tuning module obtains the decision parameters under the current decision algorithm under the corresponding scene according to the screened data under different scenes, the encapsulated decision algorithm model and the trajectory pair evaluation cost map by using a black box optimization method.

[0006] The human driving data is data generated by experienced drivers driving vehicles and has been initially labeled. Each small data set consists of a ten-minute driving clip, and each frame of data needs to contain comprehensive vehicle status information and surrounding environment information, including vehicle position, speed, acceleration, current lane ID, lane line information, lane association information, vehicle current behavior information such as left turn, straight driving, passing intersections, surrounding vehicle information such as vehicle ID, vehicle type, vehicle size, position, speed, and future trajectory of the vehicle predicted by the prediction module.

[0007] The raw data in the data-driven process include: sampling trajectory information of the own vehicle, sampling trajectory information of other vehicles, a list of vehicle IDs required for game, a list of status information of the own vehicle, a list of status information of other vehicles, conflict point information, a posteriori trajectory of the own vehicle within the next 6 seconds, a posteriori trajectory of other vehicles within the next 6 seconds, and an environmental information list.

[0008] The decision-making scenarios include straight driving, intersection, merging in / out, and lane changing scenarios.

[0009] The corresponding data are divided according to the corresponding decision-making scenarios and key data are obtained through data mining for training.

[0010] The present invention relates to an automatic driving decision optimization method based on the above system, which generates final data by reading raw data, calling a decision algorithm and setting rules; performing data cleaning and data mining processing according to the obtained data; adopting a sampling-objective function-evaluation-screening decision algorithm framework to encapsulate the decision algorithm and construct a true value; and using a black box optimization method to iteratively find the optimal parameter combination value that satisfies the objective function satisfied by the training data in the corresponding scenario, so as to obtain an optimized automatic driving decision.

[0011] The data mining process is based on the difference in TTC (Time To Collision) between the vehicle and the other vehicle when they reach the conflict point, specifically: DTTC = TTC ego -TTC agent , where: DTTC represents the time difference between the vehicle and other vehicles to the conflict point, TTC ego and TTC agent Respectively represent the time it takes for the vehicle and other vehicles to reach the conflict point.

[0012] The conflict point refers to the intersection of the posterior trajectory of the own vehicle and the posterior trajectory of the other vehicle in space, which is obtained by calculating the collision detection between the own vehicle and the other vehicle.

[0013] The decision algorithm encapsulation means: first sampling the future driving trajectories of the own vehicle and the other vehicle, and then comprehensively evaluating the sampling results of the own vehicle and the other vehicle according to the pre-objective function, so as to obtain the trajectory pair that satisfies the minimum objective function, thereby obtaining the grabbing and yielding strategy between the own vehicle and the other vehicle.

[0014] The objective function func=w s ·F safe +w c ·F comfort +w p ·F pass +w r ·F right , where: F safe F comfort F pass F right They represent the objective function expressions of safety, comfort, trafficability, and road right, respectively, and w s w c w p w r Indicates the weight of each in the objective function.

[0015] The constructed truth value, i.e., the decision truth value required in the process of constructing training data, determines more similar trajectory pairs and driver's decision results by calculating the similarity of trajectory pairs and the similarity of modes.

[0016] The similarity of the trajectory pair includes: the end point error (fde), average error (ade), and lateral error (lde) between the sampled trajectory and the posterior trajectory, specifically: Where: Δd is the Euclidean distance, and Δl is the lateral distance in the Frenet coordinate system.

[0017] The similarity of the modes is determined by obtaining the a posteriori yielding relationship based on the DTTC of the vehicle and the other vehicle, and then punishing the solution that is inconsistent with the a posteriori yielding relationship with the sampling result, and obtaining a true value evaluation calculation method based on the data and decision algorithm framework, specifically: label__cost=w1×ade+w2×fde+w3×lde+w4×bde, where: the difference value of the mode is bde, w1, w2, w3, w4 are hyperparameters and can be designed manually.

[0018] The parameter tuning described above uses the KL divergence loss function in its iterative process, specifically: Where: p(x i ) is a true value, q(x i ) is the value output by the decision algorithm. To improve efficiency, the entire training process can adopt a multi-threaded training method. Technical Effects

[0019] The present invention utilizes the driver style characteristics implicit in a large amount of human driving data, and extracts parameters that reflect the laws of driving behavior through offline calculation and analysis to drive the dynamic parameter adjustment in the autonomous driving decision-making system. This method combines the characteristics of white box technology to internalize the driver's operating style into an explainable algorithm logic. A method based on TTC (collision time) difference and conflict point analysis is used to accurately screen the data at critical moments for decision-making, and at the same time generate trajectory predictions for the vehicle and other vehicles, and complete high-quality truth value construction through a comprehensive evaluation of trajectory modal similarity and trajectory error. On the basis of data-driven, combined with a regularized framework, a multi-objective function of safety, comfort, passability and road rights is established. Compared with the existing technology, the present invention combines the white-box data-driven method with the rule framework to effectively integrate data and logic, making parameter decisions more human-like and explainable, rather than relying solely on black-box model optimization; it utilizes multi-threaded acceleration, conflict point analysis and modal-based truth value construction methods to improve the efficiency of data processing and model training, which is a significant improvement on the data cleaning and screening process in the existing technology; it adopts the method of offline parameter generation to avoid instability in online real-time optimization, and further ensure the security and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Schematic diagram of the system of the present invention. DETAILED DESCRIPTION

[0021] like Figure 1 As shown, this embodiment relates to a data-driven autonomous driving decision parameter adjustment system, including a data production module, a data screening module, a model encapsulation module and a parameter tuning module.

[0022] The data production module includes: a human driving data unit, a main line sampling code encapsulation unit, a sampling trajectory feature generation unit and an optimal trajectory calibration unit, wherein: the human driving data unit collects and integrates a large amount of data generated by human drivers based on human driving data, including the vehicle status (position, speed, acceleration, lane information) and surrounding vehicles and environmental information (such as vehicle ID, type, trajectory, conflict point, etc.), to reflect driving style and behavior rules; the main line sampling code encapsulation unit encapsulates the main line sampling code according to the data sampling rules and objective function constraints and automatically generates trajectory sampling that meets the rules. The sampling scheme is used to obtain a preliminary set of sampling trajectories constrained by rules; the sampling trajectory feature generation unit extracts trajectory sampling features (such as speed, acceleration, TTC difference, etc.) according to the trajectory sampling results, including the trajectory points and interaction information of the own vehicle and other vehicles, generates a trajectory feature vector for analysis, and obtains the sampling trajectory features for subsequent model packaging and evaluation; the optimal trajectory calibration unit performs a multi-objective comprehensive evaluation on the sampling trajectory according to the trajectory features and the objective function value (such as safety, comfort, etc.), calibrates the optimal trajectory that meets the objective function conditions, and obtains the calibrated optimal trajectory for subsequent training and verification.

[0023] The data screening module includes: a scene screening unit, a valid frame screening unit, a key obstacle screening unit and a problem category screening unit, wherein: the scene screening unit classifies the trajectory data according to the trajectory data and the scene classification rules, and roughly classifies the scenes such as straight, intersection, inlet / outlet, etc., and provides labels for specific scenes to obtain a classified scene data set, which provides a basis for detailed analysis; the valid frame screening unit screens out key frames (such as frames with DTTC in the range of [-1.0, 3.0]) according to the time-related indicators (such as TTC, DTTC) according to the effective time window ), remove redundant information, and obtain a set of valid frames in high-risk scenarios; the key obstacle screening unit screens out key obstacle information that has an important impact on decision-making (such as objects that are close or about to interact with the vehicle) based on trajectory data and environmental information (such as obstacle position and movement trajectory), and obtains a subset of scene data containing key obstacle information; the problem category screening unit screens scene data by problem category based on scene features and potential problem categories (such as collision risk and overtaking behavior) in the data, focusing on solving specific driving problems, and obtaining a high-quality training data set optimized for problem categories.

[0024] The model encapsulation module includes: a model input unit, a trajectory pair white box evaluation unit and a sampling space evaluation distribution unit, wherein: the model input unit inputs trajectory pair features according to the trajectory feature vector after data screening and the relevant parameter recommendation rules, generates a recommended parameter set adapted to the model, and obtains a recommended parameter set to provide support for the accuracy and applicability of the model; the trajectory pair white box evaluation parameter unit uses a white box model to calculate the evaluation parameters of the trajectory pair according to the trajectory pair features and the objective function (such as trajectory similarity, modal similarity, etc.), including FDE, ADE, LDE and other indicators, to obtain the trajectory pair white box evaluation parameters for accurate analysis of decision effects; the model output unit outputs a comprehensive evaluation distribution map of the sampling trajectory space according to the white box evaluation parameters and the sampling trajectory distribution, which is used to describe the pros and cons of the trajectory under the objective function, obtain the evaluation distribution of the sampling space, and provide a basis for parameter tuning.

[0025] The parameter tuning module includes: a parameter black box optimization input unit, a black box optimization unit and an iterative parameter recommendation unit, wherein: the parameter black box optimization input unit provides an initial parameter black box optimization input according to the model input parameters and the training data feature vector as the starting condition for optimization, and obtains an initial parameter configuration for starting the optimization process; the black box optimization unit uses a black box optimization algorithm (such as a genetic algorithm, Bayesian optimization, etc.) to adjust the parameters according to the parameter configuration and the optimization objective function (KL divergence), gradually approaches the optimal value, and obtains the optimized parameter configuration result; the iterative parameter recommendation unit iteratively recommends parameter configuration according to the intermediate results and evaluation indicators output by the black box optimization, adjusts the parameter combination to improve the optimization efficiency, and obtains a set of recommended optimization parameters.

[0026] The automatic driving decision optimization method of this embodiment based on the above system includes:

[0027] Step 1: Data production: read the original data, call the decision algorithm, set rules to generate the final data, including:

[0028] 1.1 Generated by experienced drivers, each data segment is divided into 10-minute segments, and each frame of data contains the following information: vehicle status (position, speed, acceleration, lane information, current behavior, etc.); surrounding vehicle information (vehicle ID, type, size, speed, predicted trajectory, etc.); conflict point, trajectory information of the vehicle and other vehicles in the next 6 seconds, environmental information, etc.

[0029] 1.2 Organize and output the data for training, including: sampling trajectory, key vehicle ID, status information of the vehicle and other vehicles, conflict point information, etc.

[0030] 1.3 Improve data production efficiency through multi-threaded operations, and call decision algorithms and rules to generate final data.

[0031] Step 2: Data screening: data cleaning and data mining process, including:

[0032] 2.1 Scenario division: Data is preliminarily classified according to scenarios such as straight driving, intersections, merging / exiting, and further subdivided, such as large and small car scenarios for other vehicles cutting in.

[0033] 2.2 Key frame extraction: Use TTC (collision time) difference to filter key frames, and filter frames with DTTC between [-1.0, 3.0].

[0034] 2.3 Conflict point identification: Calculate the intersection of the posterior trajectories of the self-vehicle and the other vehicle.

[0035] Step 3: Model encapsulation: The decision algorithm is encapsulated and the truth value is constructed using the decision algorithm framework of sampling-objective function-evaluation-screening, which includes:

[0036] 3.1 Core framework: The decision algorithm is based on the sampling → objective function → evaluation → screening process. The objective function comprehensively considers factors such as safety, comfort, passability, and road rights, evaluates trajectory pairs, and obtains the optimal grabbing and yielding strategy.

[0037] 3.2 Truth value construction method: measure the similarity of trajectory pairs (end point error FDE, average error ADE, lateral error LDE). Modal similarity: penalize the deviation (BDE) between the sampling result and the posterior grab-and-leave relationship. Construct the final truth value evaluation function for the training process.

[0038] Step 4: Use the black box optimization method to iteratively find the optimal parameter combination value that satisfies the objective function satisfied by the training data in the corresponding scenario, including:

[0039] 4.1 The filtered training data and the calibrated true value are used as input, and the optimal parameter combination is found through black-box optimization iteration.

[0040] 4.2 The loss function used in the training process is the KL divergence loss function Where: p(x i ) is a true value, q(x i ) is the value output by the decision algorithm. The entire training process uses a multi-threaded training method to improve efficiency.

[0041] After specific practical experiments, using the Simulink-Carsim-Prescan joint simulation platform environment, Prescan was used to build the experimental scene, Carsim provided the vehicle dynamics model, and Simulink deployed the algorithm. Multi-lane, multi-vehicle high-speed dynamic scenes, other vehicle cut-in scenes, and self-vehicle lane-changing scenes were selected for simulation. The PHP algorithm using this method was compared with the existing fixed strategy modeling uncertainty method (PHP*) that did not use this method in the above three types of simulation scenarios.

[0042] In high-speed dynamic scenarios with multiple lanes and multiple vehicles, the vehicle speed of the PHP method is significantly higher than that of PHP*, indicating that the PHP algorithm using the decision optimization method of the present invention has higher traffic efficiency. For scenarios where other vehicles cut in, the PHP method performs better in terms of comfort, with the absolute value of the maximum deceleration increasing by 86.5% compared to PHP*; for scenarios where the vehicle changes lanes, the PHP method performs better in terms of passability, with the average vehicle speed increasing by 28.5% compared to PHP*. This shows that the PHP method exhibits higher comfort and passability in actual vehicle experiments, especially when dealing with dynamic vehicle cut-in and lane-changing scenarios, which reflects the effectiveness of the decision optimization method of the present invention.

[0043] Compared with the existing technology, this method comprehensively considers the information of the vehicle, other vehicles, and the environment to construct more realistic and detailed data input, and the data dimension is more comprehensive; through conflict point analysis and posterior trajectory generation, the pertinence and scenario applicability of data generation are improved; based on TTC, DTTC and scene classification, key frames are accurately extracted, invalid data is removed, and cleaning and screening are more efficient; comprehensive trajectory and modal similarity, the true value data is more in line with actual needs; using black box optimization and KL divergence, the efficiency of parameter tuning is improved, and the optimal configuration is quickly obtained. This method has significant improvements in efficiency, accuracy and applicability, and can achieve higher performance in the fields of training data generation and algorithm optimization.

[0044] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principle and purpose of the present invention. The protection scope of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. Each implementation scheme within its scope shall be subject to the constraints of the present invention.

Claims

1. A data-driven autonomous driving decision optimization system, characterized in that: include: Data production module, data screening module, model encapsulation module and parameter tuning module, among which: the data production module takes human driving data as input, extracts key features and performs standardization, normalization and encoding after labeling, preprocessing and format conversion to generate raw data in the data-driven process that meets the algorithm input requirements; the data screening module filters out corresponding data from the training data according to the various scenarios divided by the decision maker and performs effective classification; the model encapsulation module encapsulates the C++ decision code and uses the trajectory pair-based true value evaluation method to construct the trajectory pair evaluation cost map required for training; the parameter tuning module uses the black box optimization method to obtain the decision parameters under the current decision algorithm in the corresponding scenario based on the screened data in different scenarios, the encapsulated decision algorithm model and the trajectory pair evaluation cost map.

2. The data-driven autonomous driving decision optimization system according to claim 1, characterized in that: The human driving data is data generated by experienced drivers driving vehicles, and after initial annotation, each small data set consists of a ten-minute driving segment, and each frame of data needs to contain comprehensive vehicle status information and surrounding environment information, including vehicle position, speed, acceleration, current lane ID, lane line information, lane association information, vehicle current behavior information such as left turn, straight, passing intersection, surrounding vehicle information such as vehicle ID, vehicle type, vehicle size, position, speed, and future trajectory of the vehicle predicted by the prediction module; The raw data in the data-driven process include: sampling trajectory information of the own vehicle, sampling trajectory information of other vehicles, vehicle ID list required for game, status information list of the own vehicle, status information list of other vehicles, conflict point information, posterior trajectory of the own vehicle within the next 6 seconds, posterior trajectory of other vehicles within the next 6 seconds, and environmental information list; The decision-making scenarios include straight driving, intersection, merging in / out, and lane changing scenarios.

3. The data-driven autonomous driving decision optimization system according to claim 1, characterized in that: The data production module includes: a human driving data unit, a main line sampling code encapsulation unit, a sampling trajectory feature generation unit and an optimal trajectory calibration unit, wherein: the human driving data unit collects and integrates a large amount of data generated by human drivers based on human driving data, including the state of the vehicle and surrounding vehicles and environmental information, to reflect driving style and behavior rules; the main line sampling code encapsulation unit encapsulates the main line sampling code and automatically generates a trajectory sampling scheme that meets the rules according to the data sampling rules and the objective function constraints, and obtains a preliminary sampling trajectory set constrained by the rules; the sampling trajectory feature generation unit extracts trajectory sampling features based on the trajectory sampling results, including the trajectory points and interaction information of the vehicle and other vehicles, generates a trajectory feature vector for analysis, and obtains the sampling trajectory features for subsequent model encapsulation and evaluation; the optimal trajectory calibration unit performs a multi-objective comprehensive evaluation of the sampling trajectory based on the trajectory features and the objective function value, calibrates the optimal trajectory that meets the objective function conditions, and obtains the calibrated optimal trajectory for subsequent training and verification.

4. The data-driven autonomous driving decision optimization system according to claim 1, characterized in that: The data screening module includes: a scene screening unit, a valid frame screening unit, a key obstacle screening unit and a problem category screening unit, wherein: the scene screening unit classifies the trajectory data according to the trajectory data and scene classification rules, and roughly classifies the scenes into straight-ahead, intersection, inflow / outflow, etc., and provides labels for specific scenes to obtain a classified scene data set, providing a basis for detailed analysis; the valid frame screening unit screens out key frames according to the effective time window based on time-related indicators, removes redundant information, and obtains a set of valid frames in high-risk scenes; the key obstacle screening unit screens out key obstacle information that has an important impact on decision-making based on trajectory data and environmental information, and obtains a scene data subset containing key obstacle information; the problem category screening unit screens scene data according to problem categories based on scene features and potential problem categories in the data, focuses on solving specific driving problems, and obtains a high-quality training data set optimized for problem categories.

5. The data-driven autonomous driving decision optimization system according to claim 1, characterized in that: The model encapsulation module includes: a model input unit, a trajectory pair white box evaluation unit and a sampling space evaluation distribution unit, wherein: the model input unit inputs trajectory pair features according to the trajectory feature vector after data screening and the relevant parameter recommendation rules, generates a recommended parameter set adapted to the model, and obtains a recommended parameter set to provide support for the accuracy and applicability of the model; the trajectory pair white box evaluation parameter unit calculates the evaluation parameters of the trajectory pair according to the trajectory pair features and the objective function using a white box model to obtain the trajectory pair white box evaluation parameters for accurate analysis of the decision effect; the model output unit outputs a comprehensive evaluation distribution map of the sampling trajectory space according to the white box evaluation parameters and the sampling trajectory distribution, which is used to describe the pros and cons of the trajectory under the objective function, obtains the evaluation distribution of the sampling space, and provides a basis for parameter tuning.

6. The data-driven autonomous driving decision optimization system according to claim 1, characterized in that: The parameter tuning module includes: a parameter black box optimization input unit, a black box optimization unit and an iterative parameter recommendation unit, wherein: the parameter black box optimization input unit provides an initial parameter black box optimization input as a starting condition for optimization based on the model input parameters and the training data feature vector, and obtains an initial parameter configuration for starting the optimization process; the black box optimization unit uses a black box optimization algorithm to adjust the parameters according to the parameter configuration and the optimization objective function, gradually approaches the optimal value, and obtains the optimized parameter configuration result; the iterative parameter recommendation unit iteratively recommends parameter configuration according to the intermediate results and evaluation indicators output by the black box optimization, adjusts the parameter combination to improve the optimization efficiency, and obtains a set of recommended optimization parameters.

7. An automatic driving decision optimization method based on the system described in any one of claims 1 to 6, characterized in that: By reading the original data, calling the decision algorithm and setting rules to generate the final data; performing data cleaning and data mining based on the obtained data; The sampling-objective function-evaluation-screening decision algorithm framework is used to encapsulate the decision algorithm and construct the true value. The black box optimization method is used to iteratively find the optimal parameter combination value that satisfies the objective function satisfied by the training data in the corresponding scenario to obtain the optimized autonomous driving decision.

8. The automatic driving decision optimization method according to claim 7, characterized in that: The data mining process is performed based on the difference in TTC between the vehicle and the other vehicle when they reach the conflict point, specifically: DTTC = TTC ego -TTC agent , where: DTTC represents the time difference between the vehicle and other vehicles to the conflict point, DTTC ego and TTC agent Respectively represent the time it takes for the vehicle and other vehicles to reach the conflict point; The conflict point refers to the intersection of the posterior trajectory of the own vehicle and the posterior trajectory of the other vehicle in space, which is obtained by calculating the collision detection between the own vehicle and the other vehicle.

9. The automatic driving decision optimization method according to claim 7, characterized in that: The decision algorithm is encapsulated as follows: firstly, sampling the future driving trajectories of the self-vehicle and the other-vehicle is performed, and then the sampling results of the self-vehicle and the other-vehicle are comprehensively evaluated according to the pre-target function, so as to obtain the trajectory pair that satisfies the minimum target function, thereby obtaining the grabbing and yielding strategy between the self-vehicle and the other-vehicle; The objective function func=w s ·F safe +w c ·F ocmfort +w p ·F pass +w r ·F right , where: F safe F comfort F pass F right Respectively represent safety, comfort, trafficability, and road rights, w s w c w p w r Indicates the weight of each in the objective function.

10. The automatic driving decision optimization method according to claim 7, characterized in that: The construction of the truth value, i.e., the decision truth value required in the process of constructing the training data, determines the more similar trajectory pairs and the driver's decision results by calculating the similarity of the trajectory pairs and the similarity of the modes; The similarity of the trajectory pair includes: the end point error (fde), average error (ade), and lateral error (lde) between the sampled trajectory and the posterior trajectory, specifically: Where: Δd is the Euclidean distance, Δl is the lateral distance in the Frenet coordinate system; The modal similarity is obtained by penalizing the solution that is inconsistent with the posterior giving way relationship according to the DTTC of the vehicle and the other vehicle, and obtaining a true value evaluation calculation method based on the data and decision algorithm framework, specifically: label__cost=w1×ade+w2×fde+w3×lde+w4×bde, where: the difference value of the modal is bde, w1, w2, w3, w4 are hyperparameters; The parameter tuning described above uses the KL divergence loss function in its iterative process, specifically: Where: p(x i ) is true value, q(x i ) is the value output by the decision algorithm.

Images