Automatic driving ability optimization method and system based on artificial driving data classification model

Abstract

The invention discloses an automatic driving ability optimization method and system based on a manual driving data classification model. The automatic driving ability optimization method based on the artificial driving data classification model comprises the following steps: acquiring and marking artificial driving data; preprocessing the manual driving data to construct a data set; building an automatic driving scene evaluation optimization model, and importing the data set for training; deploying the trained automatic driving scene evaluation optimization model to an automatic driving vehicle; and the automatic driving scene evaluation optimization model evaluates and optimizes driving data automatically acquired by the automatic driving vehicle at regular intervals. According to the method, the driving style of the automatic driving vehicle can be comprehensively evaluated and optimized through a data driving mode, manual intervention is reduced, development cost and time investment are reduced, meanwhile, diversity and accuracy of training data are ensured, safety, comfort and high efficiency of the automatic driving vehicle are remarkably improved, and rapid iteration and wide application are achieved.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and more specifically, to a method and system for optimizing autonomous driving capabilities based on a manual driving data classification model. Background Art

[0002] Current improvements in autonomous driving technology, such as vehicle safety, comfort, and efficiency, typically rely on feedback from testing teams, with R&D personnel subsequently fine-tuning relevant parameters based on this feedback. Testers conduct specialized tests based on the latest version, a process that requires significant manpower and funding. Against this backdrop, this proposal proposes a method for identifying certain abnormal driving behaviors of autonomous vehicles using a pre-trained scenario evaluation model. For example, aggressive driving can lead to sudden acceleration or deceleration, while conservative driving can cause the vehicle to hesitate during lane changes, miss the optimal opportunity to change lanes, or even come to a standstill. By optimizing specific driving behaviors and combining them with version iterations, the driving style of autonomous vehicles can be gradually improved, approaching the robust performance of experienced drivers.

[0003] However, existing technologies rely heavily on manual intervention and subjective judgment to optimize autonomous driving performance, lacking a systematic, data-driven approach. Especially in complex scenarios, autonomous vehicles often struggle to achieve the comprehensive performance of human drivers. Furthermore, existing technologies often focus on a single dimension when classifying and evaluating driving styles, failing to fully consider the balance between safety, comfort, and efficiency. Furthermore, optimizing autonomous driving systems typically requires long development cycles and high costs, which, to a certain extent, limits the rapid iteration and widespread application of the technology. Therefore, there is an urgent need for a data-driven, automated optimization method to improve the overall performance of autonomous driving systems. Summary of the Invention

[0004] In order to solve the problems in the existing technology, the present invention provides an autonomous driving capability optimization method and system based on a manual driving data classification model, which can comprehensively evaluate and optimize the driving style of autonomous driving vehicles in a data-driven manner, reduce manual intervention, and reduce development costs and time investment. At the same time, it ensures the diversity and accuracy of training data, significantly improves the safety, comfort and efficiency of autonomous driving vehicles, and realizes rapid iteration and wide application.

[0005] The present invention provides a method for optimizing autonomous driving capability based on a manual driving data classification model, comprising the following steps: Step 1: Collect manual driving data and mark it; Step 2: preprocessing the manual driving data to construct a data set; Step 3: Build an autonomous driving scenario evaluation optimization model and import the dataset for training; Step 4: Deploy the trained autonomous driving scenario evaluation optimization model to the autonomous driving vehicle; Step 5. The autonomous driving vehicle automatically obtains driving data based on the start and end time of the scene, and transmits it to the deployed autonomous driving scene evaluation optimization model for evaluation. The deployed autonomous driving scene evaluation optimization model is regularly optimized based on the evaluation results.

[0006] In a preferred embodiment of the method for optimizing the autonomous driving capability based on the manual driving data classification model provided by the present invention, step 1 includes: Different drivers are selected as test subjects, and multiple judges are set up to record all driving-related data during the collection process through on-board sensors; Select scenario evaluation indicator parameters based on the test scenario; The judges score the test performance of the tested subject according to the scenario evaluation index parameters or directly classify it as an aggressive, robust or conservative driving style, thereby marking the manual driving data.

[0007] In a preferred embodiment of the method for optimizing autonomous driving capabilities based on a manual driving data classification model provided by the present invention, drivers of different age groups, different driving experiences, and different genders are selected as test subjects, and three testers are arranged as scorers to score the test performance of the test subjects in each experiment from the passenger seat and the back seat respectively, or directly classify them as having an aggressive, steady, or conservative driving style, thereby labeling the manual driving data; the scoring results are subject to a consistency check. If the score difference exceeds a preset threshold, the scores are re-scored until a consensus is reached.

[0008] In a preferred embodiment of the method for optimizing the autonomous driving capability based on the manual driving data classification model provided by the present invention, step 2 includes: All manual driving data collected by on-board sensors is cleaned, and then resampled according to the different sampling frequencies of different sensors to align the timestamps. Then, data processing is performed through normalization, standardization, feature scaling, and feature dimensionality reduction. Finally, the judges' marks are spliced ​​together with the manual driving data after data processing to construct a scenario-based dataset.

[0009] In a preferred embodiment of the method for optimizing the autonomous driving capability based on the manual driving data classification model provided by the present invention, step three includes: A long short-term memory network algorithm is used to build the autonomous driving scenario evaluation optimization model; the input of the autonomous driving scenario evaluation optimization model is the data set constructed in step 2, and the output is the driving style classification result; during the model training process, the data set is divided into a training set, a validation set, and a test set in a ratio of 7:2:1; the model performance is optimized by adjusting the learning rate, batch size, and number of iterations; finally, 10% of the test set is used to test the performance of the model. If the model has overfitting and causes poor performance, regularization technology is introduced or data enhancement is added to improve the generalization ability of the model.

[0010] In a preferred embodiment of the method for optimizing the autonomous driving capability based on the manual driving data classification model provided by the present invention, step five includes: When an autonomous vehicle enters a specific scenario, automatic data recording is triggered, and vehicle operation data is collected in real time through on-board sensors. At the end of the scenario, an end recording instruction is triggered and the collected vehicle operation data is transmitted to the trained autonomous driving scenario evaluation optimization model deployed on the autonomous vehicle. The autonomous driving scenario evaluation optimization model loads vehicle operation data obtained by the autonomous driving vehicle and inputs it into the model for evaluation; If the evaluation results show that the current driving style is aggressive, driving style optimization is initiated to reduce the probability of aggressive driving behavior by adjusting relevant control parameters; If the proportion of aggressive driving behaviors of autonomous vehicles in specific scenarios exceeds 30% within a month, the control parameters will be optimized regularly until the driving style assessed by the autonomous driving scenario evaluation optimization model tends to be robust; then, the updated autonomous driving scenario evaluation optimization model will be pushed to a small batch of test vehicles for verification via remote OTA push; if the verification passes, the updated autonomous driving scenario evaluation optimization model will be promoted to the remaining vehicles.

[0011] The present invention also provides an automatic driving capability optimization system based on a manual driving data classification model, including a manual driving data acquisition module, a data preprocessing module, a scene evaluation model module, an automatic driving data automatic identification and extraction module, and a driving style optimization module, wherein: The manual driving data collection module is used to collect the driver's manual driving data in a specific scenario and mark each set of manual driving data; The data preprocessing module is used to preprocess the collected manual driving data and combine the markers with the preprocessed manual driving data to construct a scenario-based data set; The scenario evaluation model module is used to build an autonomous driving scenario evaluation optimization model and import the data set into the autonomous driving scenario evaluation optimization model for training; The automatic driving data identification and extraction module is set on the autonomous driving vehicle and is used to automatically obtain the driving data of the autonomous driving vehicle according to the start and end time of the scene, and transmit the driving data of the autonomous driving vehicle to the driving style optimization module The driving style optimization module is set on the autonomous driving vehicle and is used to deploy a trained autonomous driving scene evaluation optimization model to evaluate the driving data of the autonomous driving vehicle transmitted by the autonomous driving data automatic identification and extraction module. The deployed autonomous driving scene evaluation optimization model is regularly optimized based on the evaluation results.

[0012] Compared with the existing technology, the automatic driving capability optimization method based on the manual driving data classification model provided by the present invention has the following beneficial effects: 1. This invention deploys the trained autonomous driving scenario evaluation optimization model directly on the vehicle side. By evaluating the daily driving data of the autonomous driving vehicle, it optimizes the parameters related to driving style, improves passenger comfort and efficiency, and enhances the acceptance of autonomous driving as a public transportation mode.

[0013] 2. The present invention optimizes parameters and iterations of the autonomous driving scenario evaluation optimization model version through regular updates, and the optimal autonomous driving scenario evaluation optimization model is obtained, which can be pushed to other autonomous driving vehicles for updating through remote OTA. For example, if more than 30% or even 50% of the vehicle-following scenarios in the autonomous driving process within a month have aggressive driving, then the performance of the autonomous driving vehicle in the following process can be optimized by regularly updating parameters. Specifically, the iteration of the autonomous driving scenario evaluation optimization model version is updated regularly until there is no significant reduction in aggressive driving and conservative driving and maintains a low percentage every month or every quarter after a small range of parameter optimization. Then, it can be used as the optimal autonomous driving scenario evaluation optimization model version for small-batch OTA push testing, and when the test passes, it can be pushed for large-scale updates. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive work, among which: Figure 1 This is a flow chart of the method for optimizing the automatic driving capability based on the manual driving data classification model provided by the present invention. DETAILED DESCRIPTION

[0015] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0016] See also Figure 1 , Figure 1 This is a flow chart of the method for optimizing the autonomous driving capability based on the manual driving data classification model provided by the present invention. The method for optimizing the autonomous driving capability based on the manual driving data classification model includes the following steps: Step 1: Collect manual driving data and mark it; Specifically, drivers of different age groups, driving experience, and gender were selected as test subjects, and three testers were assigned as scorers. All driving-related data during the collection process was recorded through on-board sensors; Select scenario evaluation indicator parameters based on the test scenario; The judges score the test performance of the subject based on the scenario evaluation index parameters or directly classify it as an aggressive, steady or conservative driving style, thereby labeling the manual driving data; the scoring results are subject to consistency testing. If the score difference exceeds a preset threshold, the scores are re-scored until a consensus is reached.

[0017] Specifically, first, ensure that the data collection vehicle is equipped with and supports autonomous driving capabilities and can record all driving-related data during the collection process for subsequent extraction and processing. Second, select scenario evaluation metrics based on the test scenario. For example, for the following vehicle test scenario, safety, comfort, and stability must be comprehensively considered throughout the following process, and metrics related to safety, comfort, and stability must be selected. Finally, determine the plan for collecting manual driving data. Specifically, the data collection process requires selecting drivers of different ages, driving experience, and genders to eliminate data flattening caused by a single sample. Similarly, the test requires three judges—one in the passenger seat and two in the rear seat—to score the test performance of each scenario or directly categorize the test subject as having an aggressive, steady, or conservative driving style. Because subsequent neural network training requires a large dataset, we tentatively plan to collect 2,000 sets of following vehicle data. The following vehicle process can be divided into three phases based on actual conditions: starting acceleration, stable following, and deceleration and braking.

[0018] Step 2: preprocessing the manual driving data to construct a data set; Specifically, all manual driving data collected by on-board sensors is cleaned, and then the manual driving data is resampled according to the different sampling frequencies of different sensors to align the timestamps. Then, data processing such as normalization, standardization, feature scaling, and feature dimensionality reduction is performed. Finally, the judges' markings are spliced ​​together with the manual driving data after data processing to construct a scenario-based dataset.

[0019] Step 3: Build an autonomous driving scenario evaluation optimization model and import the dataset for training; Specifically, a long short-term memory network (LSTM) algorithm is used to build the autonomous driving scenario evaluation optimization model; the input of the autonomous driving scenario evaluation optimization model is the data set constructed in the step 2, and the output is the driving style classification result; during the model training process, the data set is divided into a training set, a validation set, and a test set in a ratio of 7:2:1; the model performance is optimized by adjusting the learning rate, batch size, and number of iterations; finally, 10% of the test set is used to test the performance of the model. If the model has overfitting and causes poor performance, regularization technology is introduced or data enhancement is added to improve the generalization ability of the model.

[0020] Step 4: Deploy the trained autonomous driving scenario evaluation optimization model to the autonomous driving vehicle; Step 5. The autonomous driving vehicle automatically obtains driving data based on the start and end time of the scene, and transmits it to the deployed autonomous driving scene evaluation optimization model for evaluation. The deployed autonomous driving scene evaluation optimization model is regularly optimized based on the evaluation results.

[0021] Specifically, when an autonomous vehicle enters a specific scenario, automatic data recording is triggered, and vehicle operation data is collected in real time through on-board sensors. When the scenario ends, an end recording instruction is triggered and the collected vehicle operation data is transmitted to a trained autonomous driving scenario evaluation optimization model deployed on the autonomous vehicle. The autonomous driving scenario evaluation optimization model loads vehicle operation data obtained by the autonomous driving vehicle and inputs it into the model for evaluation; If the evaluation results show that the current driving style is aggressive, driving style optimization is initiated to reduce the probability of aggressive driving behavior by adjusting relevant control parameters; If the proportion of aggressive driving behaviors of autonomous vehicles in specific scenarios exceeds 30% within a month, the control parameters will be optimized regularly until the driving style assessed by the autonomous driving scenario evaluation optimization model tends to be robust; then, the updated autonomous driving scenario evaluation optimization model will be pushed to a small batch of test vehicles for verification via remote OTA push; if the verification passes, the updated autonomous driving scenario evaluation optimization model will be promoted to the remaining vehicles.

[0022] The present invention also discloses a system for optimizing the automatic driving capability based on a manual driving data classification model, comprising a manual driving data acquisition module, a data preprocessing module, a scene evaluation model module, an automatic driving data automatic identification and extraction module, and a driving style optimization module, wherein: The manual driving data collection module is used to collect the driver's manual driving data in a specific scenario and mark each set of manual driving data; The data preprocessing module is used to preprocess the collected manual driving data and combine the markers with the preprocessed manual driving data to construct a scenario-based data set; The scenario evaluation model module is used to build an autonomous driving scenario evaluation optimization model and import the data set into the autonomous driving scenario evaluation optimization model for training; The automatic driving data identification and extraction module is set on the autonomous driving vehicle and is used to automatically obtain the driving data of the autonomous driving vehicle according to the start and end time of the scene, and transmit the driving data of the autonomous driving vehicle to the driving style optimization module The driving style optimization module is set on the autonomous driving vehicle and is used to deploy a trained autonomous driving scene evaluation optimization model to evaluate the driving data of the autonomous driving vehicle transmitted by the autonomous driving data automatic identification and extraction module. The deployed autonomous driving scene evaluation optimization model is regularly optimized based on the evaluation results.

[0023] Take the following vehicle scenario as an example. First, in the manual driving data collection module, one or more data collection vehicles are configured to perform the task of collecting data on the driver's manual driving operation in a specific scenario. These vehicles are equipped with a variety of sensors and data recording devices, which can record the driver's operating behavior in real time. Specifically, drivers of different age groups, different driving experience and different genders are selected as test subjects to eliminate the data flattening caused by a single sample, and parameters related to safety and comfort such as starting acceleration, following distance, braking deceleration, and braking jerk are obtained through on-board sensors. In this process, three testers are arranged as judges at the same time, one in the co-pilot seat and two in the back seat. They score the driver's performance in each experiment from the co-pilot and back seat positions respectively, or directly classify them as aggressive, steady or conservative driving styles, to achieve labeling of manual driving data; the scoring results are subject to consistency test. If the score difference exceeds the preset threshold, the score is re-scored until a consensus is reached.

[0024] Next, in the data preprocessing module, the data collected by the manual driving data acquisition module is further processed. The data preprocessing module receives the labeled data from the manual driving data acquisition module. These data will inevitably have some missing values ​​and outliers, so data cleaning is required at this time. At the same time, considering that different sensors have different sampling frequencies, the data also needs to be resampled to align the timestamps. Whether upsampling or downsampling is used depends on the situation. In addition, data processing methods such as normalization, standardization, feature scaling, and feature dimensionality reduction are also included. Finally, the judges' scoring labels and each set of data need to be spliced ​​together to construct a data set for the car-following scenario.

[0025] In the scenario evaluation model module, a long short-term memory (LSTM) algorithm is used to build an autonomous driving scenario evaluation optimization model for classifying and evaluating driving styles. The model's input is a dataset processed by the data preprocessing module, and its output is the driving style classification results. During training in the data preprocessing module, the dataset is divided into training, validation, and test sets in a ratio of 7:2:1. Model performance is optimized by adjusting hyperparameters such as the learning rate, batch size, and number of iterations. If the model's performance on the test set shows overfitting, regularization techniques are introduced or a data augmentation module is added to improve the model's generalization capabilities. The output of the scenario evaluation model module is used not only to analyze manual driving data but also to provide a reference for subsequent driving style optimization of autonomous vehicles.

[0026] The trained autonomous driving scenario evaluation and optimization model is then deployed to the autonomous vehicle's driving style optimization module. When the autonomous vehicle enters a specific scenario, the automatic data recording function in the autonomous driving data automatic identification and extraction module is triggered. This module uses onboard sensors to collect real-time operating data from the autonomous vehicle, including but not limited to parameters such as vehicle speed, acceleration, steering wheel angle, and brake pedal depth. At the end of the scenario, a recording stop instruction is triggered, and the collected data is transmitted to the driving style optimization module for evaluation to determine the current driving style of the autonomous vehicle.

[0027] The driving style optimization module optimizes the autonomous vehicle's driving style based on the evaluation results of the autonomous driving scenario evaluation optimization model. If the evaluation results indicate that the current driving style is aggressive, the driving style optimization module is activated. This module adjusts relevant control parameters to reduce the probability of aggressive driving behavior. For example, if a vehicle's aggressive driving behavior in a specific scenario exceeds 30% within a month, the control parameters are regularly updated until the driving style becomes more conservative. Subsequently, the updated autonomous driving scenario evaluation optimization model is pushed to a small batch of test vehicles via remote OTA (over-the-air) push for verification. If verification passes, the updated autonomous driving scenario evaluation optimization model is rolled out to the remaining vehicles. When adjusting control parameters, priority is given to those that have a greater impact on the passenger experience. For example, in a following vehicle scenario, starting acceleration and braking deceleration are prioritized; in a lane change scenario, steering wheel angle change rate and lateral acceleration are prioritized. The optimized parameters are then subjected to simulation testing to verify their safety and comfort. If the simulation test results meet preset standards, the parameters are updated to the vehicle control system.

[0028] It can be seen from the above embodiments that the present invention can comprehensively evaluate the driving style of autonomous driving vehicles in specific scenarios by constructing an autonomous driving capability optimization method and system based on an artificial driving data classification model; through diversified data collection schemes and a strict scoring mechanism, the representativeness and accuracy of training data are ensured; the autonomous driving scenario evaluation optimization model adopts a long short-term memory network algorithm, which can effectively process time series data and capture dynamic characteristics in driving behavior; the driving style optimization module gradually improves the comprehensive performance of the autonomous driving vehicle through targeted adjustment of key control parameters, making it close to the robust driving style possessed by experienced drivers; finally, through remote OTA push of the updated autonomous driving scenario evaluation optimization model, rapid iteration and wide application of the autonomous driving system are achieved.

[0029] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for optimizing autonomous driving capabilities based on a manual driving data classification model, characterized in that: The following steps are involved: Step 1: Collect manual driving data and mark it; Step 2: preprocessing the manual driving data to construct a data set; Step 3: Build an autonomous driving scenario evaluation optimization model and import the dataset for training; Step 4: Deploy the trained autonomous driving scenario evaluation optimization model to the autonomous driving vehicle; Step 5. The autonomous driving vehicle automatically obtains driving data based on the start and end time of the scene, and transmits it to the deployed autonomous driving scene evaluation optimization model for evaluation. The deployed autonomous driving scene evaluation optimization model is regularly optimized based on the evaluation results.

2. The method for optimizing the autonomous driving capability based on the manual driving data classification model according to claim 1, characterized in that: The step one comprises: Different drivers are selected as test subjects, and multiple judges are set up to record all driving-related data during the collection process through on-board sensors; Select scenario evaluation indicator parameters based on the test scenario; The judges score the test performance of the tested subject according to the scenario evaluation index parameters or directly classify it as an aggressive, robust or conservative driving style, thereby marking the manual driving data.

3. The method for optimizing the autonomous driving capability based on the manual driving data classification model according to claim 2, characterized in that: Drivers of different age groups, different driving experience and different genders are selected as test subjects, and three testers are arranged as scorers. They score the test performance of the test subjects in each experiment from the front passenger seat and the back seat respectively, or directly classify them as aggressive, steady or conservative driving style, so as to achieve the labeling of manual driving data; the scoring results are subject to consistency test. If the score difference exceeds the preset threshold, the score will be re-scored until consensus is reached.

4. The method for optimizing the autonomous driving capability based on the manual driving data classification model according to claim 1, characterized in that: The second step includes: All manual driving data collected by on-board sensors is cleaned, and then resampled according to the different sampling frequencies of different sensors to align the timestamps. Then, data processing is performed through normalization, standardization, feature scaling, and feature dimensionality reduction. Finally, the judges' marks are spliced ​​together with the manual driving data after data processing to construct a scenario-based dataset.

5. The method for optimizing the autonomous driving capability based on the manual driving data classification model according to claim 1, characterized in that: The step three includes: A long short-term memory network algorithm is used to build the autonomous driving scenario evaluation optimization model; the input of the autonomous driving scenario evaluation optimization model is the data set constructed in step 2, and the output is the driving style classification result; during the model training process, the data set is divided into a training set, a validation set, and a test set in a ratio of 7:2:1; the model performance is optimized by adjusting the learning rate, batch size, and number of iterations; finally, 10% of the test set is used to test the performance of the model. If the model has overfitting and causes poor performance, regularization technology is introduced or data enhancement is added to improve the generalization ability of the model.

6. The method for optimizing the autonomous driving capability based on the manual driving data classification model according to claim 1, characterized in that: The step five includes: When an autonomous vehicle enters a specific scenario, automatic data recording is triggered, and vehicle operation data is collected in real time through on-board sensors. At the end of the scenario, an end recording instruction is triggered and the collected vehicle operation data is transmitted to the trained autonomous driving scenario evaluation optimization model deployed on the autonomous vehicle. The autonomous driving scenario evaluation optimization model loads vehicle operation data obtained by the autonomous driving vehicle and inputs it into the model for evaluation; If the evaluation results show that the current driving style is aggressive, driving style optimization is initiated to reduce the probability of aggressive driving behavior by adjusting relevant control parameters; If the proportion of aggressive driving behaviors of autonomous vehicles in specific scenarios exceeds 30% within a month, the control parameters will be optimized regularly until the driving style assessed by the autonomous driving scenario evaluation optimization model tends to be robust; then, the updated autonomous driving scenario evaluation optimization model will be pushed to a small batch of test vehicles for verification via remote OTA push; if the verification passes, the updated autonomous driving scenario evaluation optimization model will be promoted to the remaining vehicles.

7. A system based on the method for optimizing autonomous driving capability based on a manual driving data classification model according to any one of claims 1 to 6, characterized in that: It includes manual driving data acquisition module, data preprocessing module, scene evaluation model module, automatic driving data automatic identification and extraction module and driving style optimization module, among which, The manual driving data collection module is used to collect the driver's manual driving data in a specific scenario and mark each set of manual driving data; The data preprocessing module is used to preprocess the collected manual driving data and combine the markers with the preprocessed manual driving data to construct a scenario-based data set; The scenario evaluation model module is used to build an autonomous driving scenario evaluation optimization model and import the data set into the autonomous driving scenario evaluation optimization model for training; The autonomous driving data automatic identification and extraction module is provided on the autonomous driving vehicle and is used to automatically obtain driving data of the autonomous driving vehicle according to the start and end times of the scene, and transmit the driving data of the autonomous driving vehicle to the driving style optimization module; The driving style optimization module is set on the autonomous driving vehicle and is used to deploy a trained autonomous driving scene evaluation optimization model to evaluate the driving data of the autonomous driving vehicle transmitted by the autonomous driving data automatic identification and extraction module. The deployed autonomous driving scene evaluation optimization model is regularly optimized based on the evaluation results.

Images